diff options
Diffstat (limited to 'src/video_core/renderer_vulkan/vk_device.cpp')
-rw-r--r-- | src/video_core/renderer_vulkan/vk_device.cpp | 302 |
1 files changed, 200 insertions, 102 deletions
diff --git a/src/video_core/renderer_vulkan/vk_device.cpp b/src/video_core/renderer_vulkan/vk_device.cpp index 897cbb4e8..92854a4b3 100644 --- a/src/video_core/renderer_vulkan/vk_device.cpp +++ b/src/video_core/renderer_vulkan/vk_device.cpp @@ -3,6 +3,7 @@ // Refer to the license.txt file included. #include <bitset> +#include <cstdlib> #include <optional> #include <set> #include <string_view> @@ -15,6 +16,15 @@ namespace Vulkan { namespace { +namespace Alternatives { + +constexpr std::array Depth24UnormS8Uint = {vk::Format::eD32SfloatS8Uint, + vk::Format::eD16UnormS8Uint, vk::Format{}}; +constexpr std::array Depth16UnormS8Uint = {vk::Format::eD24UnormS8Uint, + vk::Format::eD32SfloatS8Uint, vk::Format{}}; + +} // namespace Alternatives + template <typename T> void SetNext(void**& next, T& data) { *next = &data; @@ -22,7 +32,7 @@ void SetNext(void**& next, T& data) { } template <typename T> -T GetFeatures(vk::PhysicalDevice physical, vk::DispatchLoaderDynamic dldi) { +T GetFeatures(vk::PhysicalDevice physical, const vk::DispatchLoaderDynamic& dldi) { vk::PhysicalDeviceFeatures2 features; T extension_features; features.pNext = &extension_features; @@ -30,17 +40,14 @@ T GetFeatures(vk::PhysicalDevice physical, vk::DispatchLoaderDynamic dldi) { return extension_features; } -} // Anonymous namespace - -namespace Alternatives { - -constexpr std::array Depth24UnormS8Uint = {vk::Format::eD32SfloatS8Uint, - vk::Format::eD16UnormS8Uint, vk::Format{}}; -constexpr std::array Depth16UnormS8Uint = {vk::Format::eD24UnormS8Uint, - vk::Format::eD32SfloatS8Uint, vk::Format{}}; -constexpr std::array Astc = {vk::Format::eA8B8G8R8UnormPack32, vk::Format{}}; - -} // namespace Alternatives +template <typename T> +T GetProperties(vk::PhysicalDevice physical, const vk::DispatchLoaderDynamic& dldi) { + vk::PhysicalDeviceProperties2 properties; + T extension_properties; + properties.pNext = &extension_properties; + physical.getProperties2(&properties, dldi); + return extension_properties; +} constexpr const vk::Format* GetFormatAlternatives(vk::Format format) { switch (format) { @@ -53,8 +60,7 @@ constexpr const vk::Format* GetFormatAlternatives(vk::Format format) { } } -constexpr vk::FormatFeatureFlags GetFormatFeatures(vk::FormatProperties properties, - FormatType format_type) { +vk::FormatFeatureFlags GetFormatFeatures(vk::FormatProperties properties, FormatType format_type) { switch (format_type) { case FormatType::Linear: return properties.linearTilingFeatures; @@ -67,11 +73,13 @@ constexpr vk::FormatFeatureFlags GetFormatFeatures(vk::FormatProperties properti } } +} // Anonymous namespace + VKDevice::VKDevice(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical, vk::SurfaceKHR surface) - : physical{physical}, format_properties{GetFormatProperties(dldi, physical)} { + : physical{physical}, properties{physical.getProperties(dldi)}, + format_properties{GetFormatProperties(dldi, physical)} { SetupFamilies(dldi, surface); - SetupProperties(dldi); SetupFeatures(dldi); } @@ -89,12 +97,22 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan features.depthClamp = true; features.samplerAnisotropy = true; features.largePoints = true; + features.multiViewport = true; + features.depthBiasClamp = true; + features.geometryShader = true; + features.tessellationShader = true; + features.fragmentStoresAndAtomics = true; + features.shaderImageGatherExtended = true; + features.shaderStorageImageWriteWithoutFormat = true; features.textureCompressionASTC_LDR = is_optimal_astc_supported; - vk::PhysicalDeviceVertexAttributeDivisorFeaturesEXT vertex_divisor; - vertex_divisor.vertexAttributeInstanceRateDivisor = true; - vertex_divisor.vertexAttributeInstanceRateZeroDivisor = true; - SetNext(next, vertex_divisor); + vk::PhysicalDevice16BitStorageFeaturesKHR bit16_storage; + bit16_storage.uniformAndStorageBuffer16BitAccess = true; + SetNext(next, bit16_storage); + + vk::PhysicalDevice8BitStorageFeaturesKHR bit8_storage; + bit8_storage.uniformAndStorageBuffer8BitAccess = true; + SetNext(next, bit8_storage); vk::PhysicalDeviceFloat16Int8FeaturesKHR float16_int8; if (is_float16_supported) { @@ -120,6 +138,10 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan LOG_INFO(Render_Vulkan, "Device doesn't support uint8 indexes"); } + if (!ext_depth_range_unrestricted) { + LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted"); + } + vk::DeviceCreateInfo device_ci({}, static_cast<u32>(queue_cis.size()), queue_cis.data(), 0, nullptr, static_cast<u32>(extensions.size()), extensions.data(), nullptr); @@ -135,16 +157,7 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan logical = UniqueDevice( dummy_logical, vk::ObjectDestroy<vk::NoParent, vk::DispatchLoaderDynamic>(nullptr, dld)); - if (khr_driver_properties) { - vk::PhysicalDeviceDriverPropertiesKHR driver; - vk::PhysicalDeviceProperties2 properties; - properties.pNext = &driver; - physical.getProperties2(&properties, dld); - driver_id = driver.driverID; - LOG_INFO(Render_Vulkan, "Driver: {} {}", driver.driverName, driver.driverInfo); - } else { - LOG_INFO(Render_Vulkan, "Driver: Unknown"); - } + CollectTelemetryParameters(); graphics_queue = logical->getQueue(graphics_family, 0, dld); present_queue = logical->getQueue(present_family, 0, dld); @@ -190,6 +203,18 @@ vk::Format VKDevice::GetSupportedFormat(vk::Format wanted_format, bool VKDevice::IsOptimalAstcSupported(const vk::PhysicalDeviceFeatures& features, const vk::DispatchLoaderDynamic& dldi) const { + // Disable for now to avoid converting ASTC twice. + return false; + static constexpr std::array astc_formats = { + vk::Format::eAstc4x4SrgbBlock, vk::Format::eAstc8x8SrgbBlock, + vk::Format::eAstc8x5SrgbBlock, vk::Format::eAstc5x4SrgbBlock, + vk::Format::eAstc5x5UnormBlock, vk::Format::eAstc5x5SrgbBlock, + vk::Format::eAstc10x8UnormBlock, vk::Format::eAstc10x8SrgbBlock, + vk::Format::eAstc6x6UnormBlock, vk::Format::eAstc6x6SrgbBlock, + vk::Format::eAstc10x10UnormBlock, vk::Format::eAstc10x10SrgbBlock, + vk::Format::eAstc12x12UnormBlock, vk::Format::eAstc12x12SrgbBlock, + vk::Format::eAstc8x6UnormBlock, vk::Format::eAstc8x6SrgbBlock, + vk::Format::eAstc6x5UnormBlock, vk::Format::eAstc6x5SrgbBlock}; if (!features.textureCompressionASTC_LDR) { return false; } @@ -197,12 +222,6 @@ bool VKDevice::IsOptimalAstcSupported(const vk::PhysicalDeviceFeatures& features vk::FormatFeatureFlagBits::eSampledImage | vk::FormatFeatureFlagBits::eBlitSrc | vk::FormatFeatureFlagBits::eBlitDst | vk::FormatFeatureFlagBits::eTransferSrc | vk::FormatFeatureFlagBits::eTransferDst}; - constexpr std::array astc_formats = { - vk::Format::eAstc4x4UnormBlock, vk::Format::eAstc4x4SrgbBlock, - vk::Format::eAstc8x8SrgbBlock, vk::Format::eAstc8x6SrgbBlock, - vk::Format::eAstc5x4SrgbBlock, vk::Format::eAstc5x5UnormBlock, - vk::Format::eAstc5x5SrgbBlock, vk::Format::eAstc10x8UnormBlock, - vk::Format::eAstc10x8SrgbBlock}; for (const auto format : astc_formats) { const auto format_properties{physical.getFormatProperties(format, dldi)}; if (!(format_properties.optimalTilingFeatures & format_feature_usage)) { @@ -225,11 +244,17 @@ bool VKDevice::IsFormatSupported(vk::Format wanted_format, vk::FormatFeatureFlag bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical, vk::SurfaceKHR surface) { - LOG_INFO(Render_Vulkan, "{}", physical.getProperties(dldi).deviceName); bool is_suitable = true; - constexpr std::array required_extensions = {VK_KHR_SWAPCHAIN_EXTENSION_NAME, - VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME}; + constexpr std::array required_extensions = { + VK_KHR_SWAPCHAIN_EXTENSION_NAME, + VK_KHR_16BIT_STORAGE_EXTENSION_NAME, + VK_KHR_8BIT_STORAGE_EXTENSION_NAME, + VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME, + VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME, + VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME, + VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME, + }; std::bitset<required_extensions.size()> available_extensions{}; for (const auto& prop : physical.enumerateDeviceExtensionProperties(nullptr, dldi)) { @@ -246,7 +271,7 @@ bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDev if (available_extensions[i]) { continue; } - LOG_INFO(Render_Vulkan, "Missing required extension: {}", required_extensions[i]); + LOG_ERROR(Render_Vulkan, "Missing required extension: {}", required_extensions[i]); is_suitable = false; } } @@ -263,7 +288,7 @@ bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDev has_present |= physical.getSurfaceSupportKHR(i, surface, dldi) != 0; } if (!has_graphics || !has_present) { - LOG_INFO(Render_Vulkan, "Device lacks a graphics and present queue"); + LOG_ERROR(Render_Vulkan, "Device lacks a graphics and present queue"); is_suitable = false; } @@ -273,8 +298,15 @@ bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDev constexpr u32 required_ubo_size = 65536; if (limits.maxUniformBufferRange < required_ubo_size) { - LOG_INFO(Render_Vulkan, "Device UBO size {} is too small, {} is required)", - limits.maxUniformBufferRange, required_ubo_size); + LOG_ERROR(Render_Vulkan, "Device UBO size {} is too small, {} is required", + limits.maxUniformBufferRange, required_ubo_size); + is_suitable = false; + } + + constexpr u32 required_num_viewports = 16; + if (limits.maxViewports < required_num_viewports) { + LOG_INFO(Render_Vulkan, "Device number of viewports {} is too small, {} is required", + limits.maxViewports, required_num_viewports); is_suitable = false; } @@ -285,24 +317,32 @@ bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDev std::make_pair(features.depthClamp, "depthClamp"), std::make_pair(features.samplerAnisotropy, "samplerAnisotropy"), std::make_pair(features.largePoints, "largePoints"), + std::make_pair(features.multiViewport, "multiViewport"), + std::make_pair(features.depthBiasClamp, "depthBiasClamp"), + std::make_pair(features.geometryShader, "geometryShader"), + std::make_pair(features.tessellationShader, "tessellationShader"), + std::make_pair(features.fragmentStoresAndAtomics, "fragmentStoresAndAtomics"), + std::make_pair(features.shaderImageGatherExtended, "shaderImageGatherExtended"), + std::make_pair(features.shaderStorageImageWriteWithoutFormat, + "shaderStorageImageWriteWithoutFormat"), }; for (const auto& [supported, name] : feature_report) { if (supported) { continue; } - LOG_INFO(Render_Vulkan, "Missing required feature: {}", name); + LOG_ERROR(Render_Vulkan, "Missing required feature: {}", name); is_suitable = false; } + if (!is_suitable) { + LOG_ERROR(Render_Vulkan, "{} is not suitable", properties.deviceName); + } + return is_suitable; } std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynamic& dldi) { std::vector<const char*> extensions; - extensions.reserve(7); - extensions.push_back(VK_KHR_SWAPCHAIN_EXTENSION_NAME); - extensions.push_back(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME); - const auto Test = [&](const vk::ExtensionProperties& extension, std::optional<std::reference_wrapper<bool>> status, const char* name, bool push) { @@ -317,13 +357,30 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami } }; + extensions.reserve(13); + extensions.push_back(VK_KHR_SWAPCHAIN_EXTENSION_NAME); + extensions.push_back(VK_KHR_16BIT_STORAGE_EXTENSION_NAME); + extensions.push_back(VK_KHR_8BIT_STORAGE_EXTENSION_NAME); + extensions.push_back(VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME); + extensions.push_back(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME); + extensions.push_back(VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME); + extensions.push_back(VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME); + + [[maybe_unused]] const bool nsight = + std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED"); bool khr_shader_float16_int8{}; + bool ext_subgroup_size_control{}; for (const auto& extension : physical.enumerateDeviceExtensionProperties(nullptr, dldi)) { Test(extension, khr_uniform_buffer_standard_layout, VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_EXTENSION_NAME, true); - Test(extension, ext_index_type_uint8, VK_EXT_INDEX_TYPE_UINT8_EXTENSION_NAME, true); - Test(extension, khr_driver_properties, VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME, true); Test(extension, khr_shader_float16_int8, VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME, false); + Test(extension, ext_depth_range_unrestricted, + VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME, true); + Test(extension, ext_index_type_uint8, VK_EXT_INDEX_TYPE_UINT8_EXTENSION_NAME, true); + Test(extension, ext_shader_viewport_index_layer, + VK_EXT_SHADER_VIEWPORT_INDEX_LAYER_EXTENSION_NAME, true); + Test(extension, ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME, + false); } if (khr_shader_float16_int8) { @@ -332,6 +389,23 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami extensions.push_back(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME); } + if (ext_subgroup_size_control) { + const auto features = + GetFeatures<vk::PhysicalDeviceSubgroupSizeControlFeaturesEXT>(physical, dldi); + const auto properties = + GetProperties<vk::PhysicalDeviceSubgroupSizeControlPropertiesEXT>(physical, dldi); + + is_warp_potentially_bigger = properties.maxSubgroupSize > GuestWarpSize; + + if (features.subgroupSizeControl && properties.minSubgroupSize <= GuestWarpSize && + properties.maxSubgroupSize >= GuestWarpSize) { + extensions.push_back(VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME); + guest_warp_stages = properties.requiredSubgroupSizeStages; + } + } else { + is_warp_potentially_bigger = true; + } + return extensions; } @@ -358,19 +432,23 @@ void VKDevice::SetupFamilies(const vk::DispatchLoaderDynamic& dldi, vk::SurfaceK present_family = *present_family_; } -void VKDevice::SetupProperties(const vk::DispatchLoaderDynamic& dldi) { - const auto props = physical.getProperties(dldi); - device_type = props.deviceType; - uniform_buffer_alignment = static_cast<u64>(props.limits.minUniformBufferOffsetAlignment); - storage_buffer_alignment = static_cast<u64>(props.limits.minStorageBufferOffsetAlignment); - max_storage_buffer_range = static_cast<u64>(props.limits.maxStorageBufferRange); -} - void VKDevice::SetupFeatures(const vk::DispatchLoaderDynamic& dldi) { const auto supported_features{physical.getFeatures(dldi)}; is_optimal_astc_supported = IsOptimalAstcSupported(supported_features, dldi); } +void VKDevice::CollectTelemetryParameters() { + const auto driver = GetProperties<vk::PhysicalDeviceDriverPropertiesKHR>(physical, dld); + driver_id = driver.driverID; + vendor_name = driver.driverName; + + const auto extensions = physical.enumerateDeviceExtensionProperties(nullptr, dld); + reported_extensions.reserve(std::size(extensions)); + for (const auto& extension : extensions) { + reported_extensions.push_back(extension.extensionName); + } +} + std::vector<vk::DeviceQueueCreateInfo> VKDevice::GetDeviceQueueCreateInfos() const { static const float QUEUE_PRIORITY = 1.0f; @@ -385,50 +463,70 @@ std::vector<vk::DeviceQueueCreateInfo> VKDevice::GetDeviceQueueCreateInfos() con std::unordered_map<vk::Format, vk::FormatProperties> VKDevice::GetFormatProperties( const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical) { - constexpr std::array formats{vk::Format::eA8B8G8R8UnormPack32, - vk::Format::eA8B8G8R8SnormPack32, - vk::Format::eA8B8G8R8SrgbPack32, - vk::Format::eB5G6R5UnormPack16, - vk::Format::eA2B10G10R10UnormPack32, - vk::Format::eR32G32B32A32Sfloat, - vk::Format::eR16G16B16A16Uint, - vk::Format::eR16G16Unorm, - vk::Format::eR16G16Snorm, - vk::Format::eR16G16Sfloat, - vk::Format::eR16Unorm, - vk::Format::eR8G8B8A8Srgb, - vk::Format::eR8G8Unorm, - vk::Format::eR8G8Snorm, - vk::Format::eR8Unorm, - vk::Format::eB10G11R11UfloatPack32, - vk::Format::eR32Sfloat, - vk::Format::eR16Sfloat, - vk::Format::eR16G16B16A16Sfloat, - vk::Format::eB8G8R8A8Unorm, - vk::Format::eD32Sfloat, - vk::Format::eD16Unorm, - vk::Format::eD16UnormS8Uint, - vk::Format::eD24UnormS8Uint, - vk::Format::eD32SfloatS8Uint, - vk::Format::eBc1RgbaUnormBlock, - vk::Format::eBc2UnormBlock, - vk::Format::eBc3UnormBlock, - vk::Format::eBc4UnormBlock, - vk::Format::eBc5UnormBlock, - vk::Format::eBc5SnormBlock, - vk::Format::eBc7UnormBlock, - vk::Format::eBc1RgbaSrgbBlock, - vk::Format::eBc3SrgbBlock, - vk::Format::eBc7SrgbBlock, - vk::Format::eAstc4x4UnormBlock, - vk::Format::eAstc4x4SrgbBlock, - vk::Format::eAstc8x8SrgbBlock, - vk::Format::eAstc8x6SrgbBlock, - vk::Format::eAstc5x4SrgbBlock, - vk::Format::eAstc5x5UnormBlock, - vk::Format::eAstc5x5SrgbBlock, - vk::Format::eAstc10x8UnormBlock, - vk::Format::eAstc10x8SrgbBlock}; + static constexpr std::array formats{vk::Format::eA8B8G8R8UnormPack32, + vk::Format::eA8B8G8R8SnormPack32, + vk::Format::eA8B8G8R8SrgbPack32, + vk::Format::eB5G6R5UnormPack16, + vk::Format::eA2B10G10R10UnormPack32, + vk::Format::eA1R5G5B5UnormPack16, + vk::Format::eR32G32B32A32Sfloat, + vk::Format::eR32G32B32A32Uint, + vk::Format::eR32G32Sfloat, + vk::Format::eR32G32Uint, + vk::Format::eR16G16B16A16Uint, + vk::Format::eR16G16B16A16Unorm, + vk::Format::eR16G16Unorm, + vk::Format::eR16G16Snorm, + vk::Format::eR16G16Sfloat, + vk::Format::eR16Unorm, + vk::Format::eR8G8B8A8Srgb, + vk::Format::eR8G8Unorm, + vk::Format::eR8G8Snorm, + vk::Format::eR8Unorm, + vk::Format::eR8Uint, + vk::Format::eB10G11R11UfloatPack32, + vk::Format::eR32Sfloat, + vk::Format::eR32Uint, + vk::Format::eR16Sfloat, + vk::Format::eR16G16B16A16Sfloat, + vk::Format::eB8G8R8A8Unorm, + vk::Format::eR4G4B4A4UnormPack16, + vk::Format::eD32Sfloat, + vk::Format::eD16Unorm, + vk::Format::eD16UnormS8Uint, + vk::Format::eD24UnormS8Uint, + vk::Format::eD32SfloatS8Uint, + vk::Format::eBc1RgbaUnormBlock, + vk::Format::eBc2UnormBlock, + vk::Format::eBc3UnormBlock, + vk::Format::eBc4UnormBlock, + vk::Format::eBc5UnormBlock, + vk::Format::eBc5SnormBlock, + vk::Format::eBc7UnormBlock, + vk::Format::eBc6HUfloatBlock, + vk::Format::eBc6HSfloatBlock, + vk::Format::eBc1RgbaSrgbBlock, + vk::Format::eBc3SrgbBlock, + vk::Format::eBc7SrgbBlock, + vk::Format::eAstc4x4SrgbBlock, + vk::Format::eAstc8x8SrgbBlock, + vk::Format::eAstc8x5SrgbBlock, + vk::Format::eAstc5x4SrgbBlock, + vk::Format::eAstc5x5UnormBlock, + vk::Format::eAstc5x5SrgbBlock, + vk::Format::eAstc10x8UnormBlock, + vk::Format::eAstc10x8SrgbBlock, + vk::Format::eAstc6x6UnormBlock, + vk::Format::eAstc6x6SrgbBlock, + vk::Format::eAstc10x10UnormBlock, + vk::Format::eAstc10x10SrgbBlock, + vk::Format::eAstc12x12UnormBlock, + vk::Format::eAstc12x12SrgbBlock, + vk::Format::eAstc8x6UnormBlock, + vk::Format::eAstc8x6SrgbBlock, + vk::Format::eAstc6x5UnormBlock, + vk::Format::eAstc6x5SrgbBlock, + vk::Format::eE5B9G9R9UfloatPack32}; std::unordered_map<vk::Format, vk::FormatProperties> format_properties; for (const auto format : formats) { format_properties.emplace(format, physical.getFormatProperties(format, dldi)); |