Merge pull request #13050 from TellowKrinkle/HKIsNotMVK

Vulkan: Don't do MoltenVK things on Asahi Linux
This commit is contained in:
JMC47 2024-10-05 00:51:27 -04:00 committed by GitHub
commit 2d9f789940
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 269 additions and 199 deletions

View File

@ -627,24 +627,24 @@ bool ObjectCache::ValidatePipelineCache(const u8* data, size_t data_length)
return false; return false;
} }
if (header.vendor_id != g_vulkan_context->GetDeviceProperties().vendorID) if (header.vendor_id != g_vulkan_context->GetDeviceInfo().vendorID)
{ {
ERROR_LOG_FMT( ERROR_LOG_FMT(
VIDEO, "Pipeline cache failed validation: Incorrect vendor ID (file: {:#X}, device: {:#X})", VIDEO, "Pipeline cache failed validation: Incorrect vendor ID (file: {:#X}, device: {:#X})",
header.vendor_id, g_vulkan_context->GetDeviceProperties().vendorID); header.vendor_id, g_vulkan_context->GetDeviceInfo().vendorID);
return false; return false;
} }
if (header.device_id != g_vulkan_context->GetDeviceProperties().deviceID) if (header.device_id != g_vulkan_context->GetDeviceInfo().deviceID)
{ {
ERROR_LOG_FMT( ERROR_LOG_FMT(
VIDEO, "Pipeline cache failed validation: Incorrect device ID (file: {:#X}, device: {:#X})", VIDEO, "Pipeline cache failed validation: Incorrect device ID (file: {:#X}, device: {:#X})",
header.device_id, g_vulkan_context->GetDeviceProperties().deviceID); header.device_id, g_vulkan_context->GetDeviceInfo().deviceID);
return false; return false;
} }
if (std::memcmp(header.uuid, g_vulkan_context->GetDeviceProperties().pipelineCacheUUID, if (std::memcmp(header.uuid, g_vulkan_context->GetDeviceInfo().pipelineCacheUUID, VK_UUID_SIZE) !=
VK_UUID_SIZE) != 0) 0)
{ {
ERROR_LOG_FMT(VIDEO, "Pipeline cache failed validation: Incorrect UUID"); ERROR_LOG_FMT(VIDEO, "Pipeline cache failed validation: Incorrect UUID");
return false; return false;

View File

@ -54,11 +54,8 @@ void VideoBackend::InitBackendInfo(const WindowSystemInfo& wsi)
device_index = 0; device_index = 0;
VkPhysicalDevice gpu = gpu_list[device_index]; VkPhysicalDevice gpu = gpu_list[device_index];
VkPhysicalDeviceProperties properties; VulkanContext::PhysicalDeviceInfo properties(gpu);
vkGetPhysicalDeviceProperties(gpu, &properties); VulkanContext::PopulateBackendInfoFeatures(&g_Config, gpu, properties);
VkPhysicalDeviceFeatures features;
vkGetPhysicalDeviceFeatures(gpu, &features);
VulkanContext::PopulateBackendInfoFeatures(&g_Config, gpu, properties, features);
VulkanContext::PopulateBackendInfoMultisampleModes(&g_Config, gpu, properties); VulkanContext::PopulateBackendInfoMultisampleModes(&g_Config, gpu, properties);
} }
} }
@ -187,10 +184,9 @@ bool VideoBackend::Initialize(const WindowSystemInfo& wsi)
// Since VulkanContext maintains a copy of the device features and properties, we can use this // Since VulkanContext maintains a copy of the device features and properties, we can use this
// to initialize the backend information, so that we don't need to enumerate everything again. // to initialize the backend information, so that we don't need to enumerate everything again.
VulkanContext::PopulateBackendInfoFeatures(&g_Config, g_vulkan_context->GetPhysicalDevice(), VulkanContext::PopulateBackendInfoFeatures(&g_Config, g_vulkan_context->GetPhysicalDevice(),
g_vulkan_context->GetDeviceProperties(), g_vulkan_context->GetDeviceInfo());
g_vulkan_context->GetDeviceFeatures());
VulkanContext::PopulateBackendInfoMultisampleModes( VulkanContext::PopulateBackendInfoMultisampleModes(
&g_Config, g_vulkan_context->GetPhysicalDevice(), g_vulkan_context->GetDeviceProperties()); &g_Config, g_vulkan_context->GetPhysicalDevice(), g_vulkan_context->GetDeviceInfo());
g_Config.backend_info.bSupportsExclusiveFullscreen = g_Config.backend_info.bSupportsExclusiveFullscreen =
enable_surface && g_vulkan_context->SupportsExclusiveFullscreen(wsi, surface); enable_surface && g_vulkan_context->SupportsExclusiveFullscreen(wsi, surface);

View File

@ -92,8 +92,8 @@ bool VertexManager::Initialize()
// Prefer an 8MB buffer if possible, but use less if the device doesn't support this. // Prefer an 8MB buffer if possible, but use less if the device doesn't support this.
// This buffer is potentially going to be addressed as R8s in the future, so we assume // This buffer is potentially going to be addressed as R8s in the future, so we assume
// that one element is one byte. // that one element is one byte.
const u32 texel_buffer_size = std::min( const u32 texel_buffer_size =
TEXEL_STREAM_BUFFER_SIZE, g_vulkan_context->GetDeviceLimits().maxTexelBufferElements); std::min(TEXEL_STREAM_BUFFER_SIZE, g_vulkan_context->GetDeviceInfo().maxTexelBufferElements);
m_texel_stream_buffer = m_texel_stream_buffer =
StreamBuffer::Create(VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT, texel_buffer_size); StreamBuffer::Create(VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT, texel_buffer_size);
if (!m_texel_stream_buffer) if (!m_texel_stream_buffer)

View File

@ -22,22 +22,113 @@ static constexpr const char* VALIDATION_LAYER_NAME = "VK_LAYER_KHRONOS_validatio
std::unique_ptr<VulkanContext> g_vulkan_context; std::unique_ptr<VulkanContext> g_vulkan_context;
VulkanContext::VulkanContext(VkInstance instance, VkPhysicalDevice physical_device) template <typename Chain, typename Element>
: m_instance(instance), m_physical_device(physical_device) static void InsertIntoChain(Chain* chain, Element* element)
{ {
// Read device physical memory properties, we need it for allocating buffers element->pNext = chain->pNext;
vkGetPhysicalDeviceProperties(physical_device, &m_device_properties); chain->pNext = element;
vkGetPhysicalDeviceMemoryProperties(physical_device, &m_device_memory_properties); }
// Would any drivers be this silly? I hope not... VulkanContext::PhysicalDeviceInfo::PhysicalDeviceInfo(VkPhysicalDevice device)
m_device_properties.limits.minUniformBufferOffsetAlignment = std::max( {
m_device_properties.limits.minUniformBufferOffsetAlignment, static_cast<VkDeviceSize>(1)); VkPhysicalDeviceFeatures features;
m_device_properties.limits.minTexelBufferOffsetAlignment = std::max( VkPhysicalDeviceProperties2 properties2;
m_device_properties.limits.minTexelBufferOffsetAlignment, static_cast<VkDeviceSize>(1)); VkPhysicalDeviceProperties& properties = properties2.properties;
m_device_properties.limits.optimalBufferCopyOffsetAlignment = std::max(
m_device_properties.limits.optimalBufferCopyOffsetAlignment, static_cast<VkDeviceSize>(1)); vkGetPhysicalDeviceProperties(device, &properties);
m_device_properties.limits.optimalBufferCopyRowPitchAlignment = std::max( vkGetPhysicalDeviceFeatures(device, &features);
m_device_properties.limits.optimalBufferCopyRowPitchAlignment, static_cast<VkDeviceSize>(1)); apiVersion = vkGetPhysicalDeviceProperties2 ? properties.apiVersion : VK_API_VERSION_1_0;
if (apiVersion >= VK_API_VERSION_1_1)
{
VkPhysicalDeviceSubgroupProperties properties_subgroup = {};
VkPhysicalDeviceVulkan12Properties properties_vk12 = {};
properties2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2;
properties2.pNext = nullptr;
properties_subgroup.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES;
InsertIntoChain(&properties2, &properties_subgroup);
if (apiVersion >= VK_API_VERSION_1_2)
{
properties_vk12.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES;
InsertIntoChain(&properties2, &properties_vk12);
}
vkGetPhysicalDeviceProperties2(device, &properties2);
if (apiVersion >= VK_API_VERSION_1_2)
{
driverID = properties_vk12.driverID;
}
subgroupSize = properties_subgroup.subgroupSize;
// We require basic ops (for gl_SubgroupInvocationID), ballot (for subgroupBallot,
// subgroupBallotFindLSB), and arithmetic (for subgroupMin/subgroupMax).
// Shuffle is enabled as a workaround until SPIR-V >= 1.5 is enabled with broadcast(uniform)
// support.
constexpr VkSubgroupFeatureFlags required_operations =
VK_SUBGROUP_FEATURE_BASIC_BIT | VK_SUBGROUP_FEATURE_ARITHMETIC_BIT |
VK_SUBGROUP_FEATURE_BALLOT_BIT | VK_SUBGROUP_FEATURE_SHUFFLE_BIT;
shaderSubgroupOperations =
(properties_subgroup.supportedOperations & required_operations) == required_operations &&
properties_subgroup.supportedStages & VK_SHADER_STAGE_FRAGMENT_BIT;
}
memcpy(deviceName, properties.deviceName, sizeof(deviceName));
memcpy(pipelineCacheUUID, properties.pipelineCacheUUID, sizeof(pipelineCacheUUID));
vendorID = properties.vendorID;
deviceID = properties.deviceID;
minUniformBufferOffsetAlignment =
std::max<VkDeviceSize>(properties.limits.minUniformBufferOffsetAlignment, 1);
bufferImageGranularity = std::max<VkDeviceSize>(properties.limits.bufferImageGranularity, 1);
maxTexelBufferElements = properties.limits.maxTexelBufferElements;
maxImageDimension2D = properties.limits.maxImageDimension2D;
framebufferColorSampleCounts = properties.limits.framebufferColorSampleCounts;
framebufferDepthSampleCounts = properties.limits.framebufferDepthSampleCounts;
memcpy(pointSizeRange, properties.limits.pointSizeRange, sizeof(pointSizeRange));
maxSamplerAnisotropy = properties.limits.maxSamplerAnisotropy;
dualSrcBlend = features.dualSrcBlend != VK_FALSE;
geometryShader = features.geometryShader != VK_FALSE;
samplerAnisotropy = features.samplerAnisotropy != VK_FALSE;
logicOp = features.logicOp != VK_FALSE;
fragmentStoresAndAtomics = features.fragmentStoresAndAtomics != VK_FALSE;
sampleRateShading = features.sampleRateShading != VK_FALSE;
largePoints = features.largePoints != VK_FALSE;
shaderStorageImageMultisample = features.shaderStorageImageMultisample != VK_FALSE;
shaderTessellationAndGeometryPointSize =
features.shaderTessellationAndGeometryPointSize != VK_FALSE;
occlusionQueryPrecise = features.occlusionQueryPrecise != VK_FALSE;
shaderClipDistance = features.shaderClipDistance != VK_FALSE;
depthClamp = features.depthClamp != VK_FALSE;
textureCompressionBC = features.textureCompressionBC != VK_FALSE;
}
VkPhysicalDeviceFeatures VulkanContext::PhysicalDeviceInfo::features() const
{
VkPhysicalDeviceFeatures features;
memset(&features, 0, sizeof(features));
features.dualSrcBlend = dualSrcBlend ? VK_TRUE : VK_FALSE;
features.geometryShader = geometryShader ? VK_TRUE : VK_FALSE;
features.samplerAnisotropy = samplerAnisotropy ? VK_TRUE : VK_FALSE;
features.logicOp = logicOp ? VK_TRUE : VK_FALSE;
features.fragmentStoresAndAtomics = fragmentStoresAndAtomics ? VK_TRUE : VK_FALSE;
features.sampleRateShading = sampleRateShading ? VK_TRUE : VK_FALSE;
features.largePoints = largePoints ? VK_TRUE : VK_FALSE;
features.shaderStorageImageMultisample = shaderStorageImageMultisample ? VK_TRUE : VK_FALSE;
features.shaderTessellationAndGeometryPointSize =
shaderTessellationAndGeometryPointSize ? VK_TRUE : VK_FALSE;
features.occlusionQueryPrecise = occlusionQueryPrecise ? VK_TRUE : VK_FALSE;
features.shaderClipDistance = shaderClipDistance ? VK_TRUE : VK_FALSE;
features.depthClamp = depthClamp ? VK_TRUE : VK_FALSE;
features.textureCompressionBC = textureCompressionBC ? VK_TRUE : VK_FALSE;
return features;
}
VulkanContext::VulkanContext(VkInstance instance, VkPhysicalDevice physical_device)
: m_instance(instance), m_physical_device(physical_device), m_device_info(physical_device)
{
} }
VulkanContext::~VulkanContext() VulkanContext::~VulkanContext()
@ -115,6 +206,27 @@ bool VulkanContext::CheckValidationLayerAvailablility()
return supports_debug_utils && supports_validation_layers; return supports_debug_utils && supports_validation_layers;
} }
static u32 getAPIVersion()
{
u32 supported_version;
u32 used_version = VK_API_VERSION_1_0;
if (vkEnumerateInstanceVersion && vkEnumerateInstanceVersion(&supported_version) == VK_SUCCESS)
{
// The device itself may not support 1.1, so we check that before using any 1.1 functionality.
if (supported_version >= VK_API_VERSION_1_2)
used_version = VK_API_VERSION_1_2;
else if (supported_version >= VK_API_VERSION_1_1)
used_version = VK_API_VERSION_1_1;
WARN_LOG_FMT(HOST_GPU, "Using Vulkan 1.{}, supported: {}.{}", VK_VERSION_MINOR(used_version),
VK_VERSION_MAJOR(supported_version), VK_VERSION_MINOR(supported_version));
}
else
{
WARN_LOG_FMT(HOST_GPU, "Using Vulkan 1.0");
}
return used_version;
}
VkInstance VulkanContext::CreateVulkanInstance(WindowSystemType wstype, bool enable_debug_utils, VkInstance VulkanContext::CreateVulkanInstance(WindowSystemType wstype, bool enable_debug_utils,
bool enable_validation_layer, bool enable_validation_layer,
u32* out_vk_api_version) u32* out_vk_api_version)
@ -131,31 +243,7 @@ VkInstance VulkanContext::CreateVulkanInstance(WindowSystemType wstype, bool ena
app_info.applicationVersion = VK_MAKE_VERSION(5, 0, 0); app_info.applicationVersion = VK_MAKE_VERSION(5, 0, 0);
app_info.pEngineName = "Dolphin Emulator"; app_info.pEngineName = "Dolphin Emulator";
app_info.engineVersion = VK_MAKE_VERSION(5, 0, 0); app_info.engineVersion = VK_MAKE_VERSION(5, 0, 0);
app_info.apiVersion = VK_MAKE_VERSION(1, 0, 0); app_info.apiVersion = getAPIVersion();
// Try for Vulkan 1.1 if the loader supports it.
if (vkEnumerateInstanceVersion)
{
u32 supported_api_version = 0;
VkResult res = vkEnumerateInstanceVersion(&supported_api_version);
if (res == VK_SUCCESS && (VK_VERSION_MAJOR(supported_api_version) > 1 ||
VK_VERSION_MINOR(supported_api_version) >= 1))
{
// The device itself may not support 1.1, so we check that before using any 1.1 functionality.
app_info.apiVersion = VK_MAKE_VERSION(1, 1, 0);
WARN_LOG_FMT(HOST_GPU, "Using Vulkan 1.1, supported: {}.{}",
VK_VERSION_MAJOR(supported_api_version),
VK_VERSION_MINOR(supported_api_version));
}
else
{
WARN_LOG_FMT(HOST_GPU, "Using Vulkan 1.0");
}
}
else
{
WARN_LOG_FMT(HOST_GPU, "Using Vulkan 1.0");
}
*out_vk_api_version = app_info.apiVersion; *out_vk_api_version = app_info.apiVersion;
@ -397,55 +485,49 @@ void VulkanContext::PopulateBackendInfoAdapters(VideoConfig* config, const GPULi
} }
void VulkanContext::PopulateBackendInfoFeatures(VideoConfig* config, VkPhysicalDevice gpu, void VulkanContext::PopulateBackendInfoFeatures(VideoConfig* config, VkPhysicalDevice gpu,
const VkPhysicalDeviceProperties& properties, const PhysicalDeviceInfo& info)
const VkPhysicalDeviceFeatures& features)
{ {
config->backend_info.MaxTextureSize = properties.limits.maxImageDimension2D; config->backend_info.MaxTextureSize = info.maxImageDimension2D;
config->backend_info.bUsesLowerLeftOrigin = false; config->backend_info.bUsesLowerLeftOrigin = false;
config->backend_info.bSupportsDualSourceBlend = (features.dualSrcBlend == VK_TRUE); config->backend_info.bSupportsDualSourceBlend = info.dualSrcBlend;
config->backend_info.bSupportsGeometryShaders = (features.geometryShader == VK_TRUE); config->backend_info.bSupportsGeometryShaders = info.geometryShader;
config->backend_info.bSupportsGSInstancing = (features.geometryShader == VK_TRUE); config->backend_info.bSupportsGSInstancing = info.geometryShader;
config->backend_info.bSupportsBBox = config->backend_info.bSupportsFragmentStoresAndAtomics = config->backend_info.bSupportsBBox = config->backend_info.bSupportsFragmentStoresAndAtomics =
(features.fragmentStoresAndAtomics == VK_TRUE); info.fragmentStoresAndAtomics;
config->backend_info.bSupportsSSAA = (features.sampleRateShading == VK_TRUE); config->backend_info.bSupportsSSAA = info.sampleRateShading;
config->backend_info.bSupportsLogicOp = (features.logicOp == VK_TRUE); config->backend_info.bSupportsLogicOp = info.logicOp;
#ifdef __APPLE__
// Metal doesn't support this. // Metal doesn't support this.
config->backend_info.bSupportsLodBiasInSampler = false; config->backend_info.bSupportsLodBiasInSampler = info.driverID != VK_DRIVER_ID_MOLTENVK;
#else
config->backend_info.bSupportsLodBiasInSampler = true;
#endif
// Disable geometry shader when shaderTessellationAndGeometryPointSize is not supported. // Disable geometry shader when shaderTessellationAndGeometryPointSize is not supported.
// Seems this is needed for gl_Layer. // Seems this is needed for gl_Layer.
if (!features.shaderTessellationAndGeometryPointSize) if (!info.shaderTessellationAndGeometryPointSize)
{ {
config->backend_info.bSupportsGeometryShaders = VK_FALSE; config->backend_info.bSupportsGeometryShaders = VK_FALSE;
config->backend_info.bSupportsGSInstancing = VK_FALSE; config->backend_info.bSupportsGSInstancing = VK_FALSE;
} }
// Depth clamping implies shaderClipDistance and depthClamp // Depth clamping implies shaderClipDistance and depthClamp
config->backend_info.bSupportsDepthClamp = config->backend_info.bSupportsDepthClamp = info.depthClamp && info.shaderClipDistance;
(features.depthClamp == VK_TRUE && features.shaderClipDistance == VK_TRUE);
// textureCompressionBC implies BC1 through BC7, which is a superset of DXT1/3/5, which we need. // textureCompressionBC implies BC1 through BC7, which is a superset of DXT1/3/5, which we need.
const bool supports_bc = features.textureCompressionBC == VK_TRUE; config->backend_info.bSupportsST3CTextures = info.textureCompressionBC;
config->backend_info.bSupportsST3CTextures = supports_bc; config->backend_info.bSupportsBPTCTextures = info.textureCompressionBC;
config->backend_info.bSupportsBPTCTextures = supports_bc;
// Some devices don't support point sizes >1 (e.g. Adreno). // Some devices don't support point sizes >1 (e.g. Adreno).
// If we can't use a point size above our maximum IR, use triangles instead for EFB pokes. // If we can't use a point size above our maximum IR, use triangles instead for EFB pokes.
// This means a 6x increase in the size of the vertices, though. // This means a 6x increase in the size of the vertices, though.
config->backend_info.bSupportsLargePoints = features.largePoints && config->backend_info.bSupportsLargePoints =
properties.limits.pointSizeRange[0] <= 1.0f && info.largePoints && info.pointSizeRange[0] <= 1.0f && info.pointSizeRange[1] >= 16;
properties.limits.pointSizeRange[1] >= 16;
std::string device_name = properties.deviceName; std::string device_name = info.deviceName;
u32 vendor_id = properties.vendorID; u32 vendor_id = info.vendorID;
bool is_moltenvk = info.driverID == VK_DRIVER_ID_MOLTENVK;
// Only Apple family GPUs support framebuffer fetch. // Only Apple family GPUs support framebuffer fetch.
if (vendor_id == 0x106B || device_name.find("Apple") != std::string::npos) // We currently use a hacked MoltenVK to implement this, so don't attempt outside of MVK
if (is_moltenvk && (vendor_id == 0x106B || device_name.find("Apple") != std::string::npos))
{ {
config->backend_info.bSupportsFramebufferFetch = true; config->backend_info.bSupportsFramebufferFetch = true;
} }
@ -465,8 +547,8 @@ void VulkanContext::PopulateBackendInfoFeatures(VideoConfig* config, VkPhysicalD
config->backend_info.bSupportsDynamicSamplerIndexing = false; config->backend_info.bSupportsDynamicSamplerIndexing = false;
} }
void VulkanContext::PopulateBackendInfoMultisampleModes( void VulkanContext::PopulateBackendInfoMultisampleModes(VideoConfig* config, VkPhysicalDevice gpu,
VideoConfig* config, VkPhysicalDevice gpu, const VkPhysicalDeviceProperties& properties) const PhysicalDeviceInfo& info)
{ {
// Query image support for the EFB texture formats. // Query image support for the EFB texture formats.
VkImageFormatProperties efb_color_properties = {}; VkImageFormatProperties efb_color_properties = {};
@ -479,10 +561,9 @@ void VulkanContext::PopulateBackendInfoMultisampleModes(
VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT, 0, &efb_depth_properties); VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT, 0, &efb_depth_properties);
// We can only support MSAA if it's supported on our render target formats. // We can only support MSAA if it's supported on our render target formats.
VkSampleCountFlags supported_sample_counts = properties.limits.framebufferColorSampleCounts & VkSampleCountFlags supported_sample_counts =
properties.limits.framebufferDepthSampleCounts & info.framebufferColorSampleCounts & info.framebufferDepthSampleCounts &
efb_color_properties.sampleCounts & efb_color_properties.sampleCounts & efb_depth_properties.sampleCounts;
efb_depth_properties.sampleCounts;
// No AA // No AA
config->backend_info.AAModes.clear(); config->backend_info.AAModes.clear();
@ -522,7 +603,6 @@ std::unique_ptr<VulkanContext> VulkanContext::Create(VkInstance instance, VkPhys
// Initialize DriverDetails so that we can check for bugs to disable features if needed. // Initialize DriverDetails so that we can check for bugs to disable features if needed.
context->InitDriverDetails(); context->InitDriverDetails();
context->PopulateShaderSubgroupSupport();
// Enable debug messages if the "Host GPU" log category is enabled. // Enable debug messages if the "Host GPU" log category is enabled.
if (enable_debug_utils) if (enable_debug_utils)
@ -599,41 +679,15 @@ bool VulkanContext::SelectDeviceExtensions(bool enable_surface)
return true; return true;
} }
bool VulkanContext::SelectDeviceFeatures() void VulkanContext::WarnMissingDeviceFeatures()
{ {
VkPhysicalDeviceProperties properties; if (!m_device_info.largePoints)
vkGetPhysicalDeviceProperties(m_physical_device, &properties);
VkPhysicalDeviceFeatures available_features;
vkGetPhysicalDeviceFeatures(m_physical_device, &available_features);
// Not having geometry shaders or wide lines will cause issues with rendering.
if (!available_features.geometryShader && !available_features.wideLines)
WARN_LOG_FMT(VIDEO, "Vulkan: Missing both geometryShader and wideLines features.");
if (!available_features.largePoints)
WARN_LOG_FMT(VIDEO, "Vulkan: Missing large points feature. CPU EFB writes will be slower."); WARN_LOG_FMT(VIDEO, "Vulkan: Missing large points feature. CPU EFB writes will be slower.");
if (!available_features.occlusionQueryPrecise) if (!m_device_info.occlusionQueryPrecise)
{ {
WARN_LOG_FMT(VIDEO, WARN_LOG_FMT(VIDEO,
"Vulkan: Missing precise occlusion queries. Perf queries will be inaccurate."); "Vulkan: Missing precise occlusion queries. Perf queries will be inaccurate.");
} }
// Enable the features we use.
m_device_features.dualSrcBlend = available_features.dualSrcBlend;
m_device_features.geometryShader = available_features.geometryShader;
m_device_features.samplerAnisotropy = available_features.samplerAnisotropy;
m_device_features.logicOp = available_features.logicOp;
m_device_features.fragmentStoresAndAtomics = available_features.fragmentStoresAndAtomics;
m_device_features.sampleRateShading = available_features.sampleRateShading;
m_device_features.largePoints = available_features.largePoints;
m_device_features.shaderStorageImageMultisample =
available_features.shaderStorageImageMultisample;
m_device_features.shaderTessellationAndGeometryPointSize =
available_features.shaderTessellationAndGeometryPointSize;
m_device_features.occlusionQueryPrecise = available_features.occlusionQueryPrecise;
m_device_features.shaderClipDistance = available_features.shaderClipDistance;
m_device_features.depthClamp = available_features.depthClamp;
m_device_features.textureCompressionBC = available_features.textureCompressionBC;
return true;
} }
bool VulkanContext::CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer) bool VulkanContext::CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer)
@ -749,11 +803,10 @@ bool VulkanContext::CreateDevice(VkSurfaceKHR surface, bool enable_validation_la
device_info.enabledExtensionCount = static_cast<uint32_t>(extension_name_pointers.size()); device_info.enabledExtensionCount = static_cast<uint32_t>(extension_name_pointers.size());
device_info.ppEnabledExtensionNames = extension_name_pointers.data(); device_info.ppEnabledExtensionNames = extension_name_pointers.data();
// Check for required features before creating. WarnMissingDeviceFeatures();
if (!SelectDeviceFeatures())
return false;
device_info.pEnabledFeatures = &m_device_features; VkPhysicalDeviceFeatures device_features = m_device_info.features();
device_info.pEnabledFeatures = &device_features;
// Enable debug layer on debug builds // Enable debug layer on debug builds
if (enable_validation_layer) if (enable_validation_layer)
@ -880,6 +933,41 @@ bool VulkanContext::SupportsDeviceExtension(const char* name) const
[name](const std::string& extension) { return extension == name; }); [name](const std::string& extension) { return extension == name; });
} }
static bool DriverIsMesa(VkDriverId driver_id)
{
switch (driver_id)
{
case VK_DRIVER_ID_MESA_RADV:
case VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA:
case VK_DRIVER_ID_MESA_LLVMPIPE:
case VK_DRIVER_ID_MESA_TURNIP:
case VK_DRIVER_ID_MESA_V3DV:
case VK_DRIVER_ID_MESA_PANVK:
case VK_DRIVER_ID_MESA_VENUS:
case VK_DRIVER_ID_MESA_DOZEN:
case VK_DRIVER_ID_MESA_NVK:
case VK_DRIVER_ID_IMAGINATION_OPEN_SOURCE_MESA:
case VK_DRIVER_ID_MESA_HONEYKRISP:
return true;
default:
return false;
}
}
static DriverDetails::Driver GetMesaDriver(VkDriverId driver_id)
{
switch (driver_id)
{
// clang-format off
case VK_DRIVER_ID_MESA_RADV: return DriverDetails::DRIVER_R600;
case VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA: return DriverDetails::DRIVER_I965;
case VK_DRIVER_ID_MESA_NVK: return DriverDetails::DRIVER_NOUVEAU;
case VK_DRIVER_ID_MESA_TURNIP: return DriverDetails::DRIVER_FREEDRENO;
default: return DriverDetails::DRIVER_UNKNOWN;
// clang-format on
}
}
void VulkanContext::InitDriverDetails() void VulkanContext::InitDriverDetails()
{ {
DriverDetails::Vendor vendor; DriverDetails::Vendor vendor;
@ -888,9 +976,17 @@ void VulkanContext::InitDriverDetails()
// String comparisons aren't ideal, but there doesn't seem to be any other way to tell // String comparisons aren't ideal, but there doesn't seem to be any other way to tell
// which vendor a driver is for. These names are based on the reports submitted to // which vendor a driver is for. These names are based on the reports submitted to
// vulkan.gpuinfo.org, as of 19/09/2017. // vulkan.gpuinfo.org, as of 19/09/2017.
std::string device_name = m_device_properties.deviceName; std::string device_name = m_device_info.deviceName;
u32 vendor_id = m_device_properties.vendorID; u32 vendor_id = m_device_info.vendorID;
if (vendor_id == 0x10DE) // Note: driver_id may be 0 on vulkan < 1.2
VkDriverId driver_id = m_device_info.driverID;
if (DriverIsMesa(driver_id))
{
vendor = DriverDetails::VENDOR_MESA;
driver = GetMesaDriver(driver_id);
}
else if (vendor_id == 0x10DE)
{ {
// Currently, there is only the official NV binary driver. // Currently, there is only the official NV binary driver.
// "NVIDIA" does not appear in the device name. // "NVIDIA" does not appear in the device name.
@ -958,51 +1054,17 @@ void VulkanContext::InitDriverDetails()
driver = DriverDetails::DRIVER_UNKNOWN; driver = DriverDetails::DRIVER_UNKNOWN;
} }
#ifdef __APPLE__
// Vulkan on macOS goes through Metal, and is not susceptible to the same bugs // Vulkan on macOS goes through Metal, and is not susceptible to the same bugs
// as the vendor's native Vulkan drivers. We use a different driver fields to // as the vendor's native Vulkan drivers. We use a different driver fields to
// differentiate MoltenVK. // differentiate MoltenVK.
if (driver_id == VK_DRIVER_ID_MOLTENVK)
driver = DriverDetails::DRIVER_PORTABILITY; driver = DriverDetails::DRIVER_PORTABILITY;
#endif
DriverDetails::Init(DriverDetails::API_VULKAN, vendor, driver, DriverDetails::Init(DriverDetails::API_VULKAN, vendor, driver,
static_cast<double>(m_device_properties.driverVersion), static_cast<double>(m_device_info.driverVersion),
DriverDetails::Family::UNKNOWN, std::move(device_name)); DriverDetails::Family::UNKNOWN, std::move(device_name));
} }
void VulkanContext::PopulateShaderSubgroupSupport()
{
// Vulkan 1.1 support is required for vkGetPhysicalDeviceProperties2(), but we can't rely on the
// function pointer alone.
if (!vkGetPhysicalDeviceProperties2 || (VK_VERSION_MAJOR(m_device_properties.apiVersion) == 1 &&
VK_VERSION_MINOR(m_device_properties.apiVersion) < 1))
{
return;
}
VkPhysicalDeviceProperties2 device_properties_2 = {};
device_properties_2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2;
VkPhysicalDeviceSubgroupProperties subgroup_properties = {};
subgroup_properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES;
device_properties_2.pNext = &subgroup_properties;
vkGetPhysicalDeviceProperties2(m_physical_device, &device_properties_2);
m_shader_subgroup_size = subgroup_properties.subgroupSize;
// We require basic ops (for gl_SubgroupInvocationID), ballot (for subgroupBallot,
// subgroupBallotFindLSB), and arithmetic (for subgroupMin/subgroupMax).
// Shuffle is enabled as a workaround until SPIR-V >= 1.5 is enabled with broadcast(uniform)
// support.
constexpr VkSubgroupFeatureFlags required_operations =
VK_SUBGROUP_FEATURE_BASIC_BIT | VK_SUBGROUP_FEATURE_ARITHMETIC_BIT |
VK_SUBGROUP_FEATURE_BALLOT_BIT | VK_SUBGROUP_FEATURE_SHUFFLE_BIT;
m_supports_shader_subgroup_operations =
(subgroup_properties.supportedOperations & required_operations) == required_operations &&
subgroup_properties.supportedStages & VK_SHADER_STAGE_FRAGMENT_BIT;
}
bool VulkanContext::SupportsExclusiveFullscreen(const WindowSystemInfo& wsi, VkSurfaceKHR surface) bool VulkanContext::SupportsExclusiveFullscreen(const WindowSystemInfo& wsi, VkSurfaceKHR surface)
{ {
#ifdef SUPPORTS_VULKAN_EXCLUSIVE_FULLSCREEN #ifdef SUPPORTS_VULKAN_EXCLUSIVE_FULLSCREEN

View File

@ -18,6 +18,44 @@ namespace Vulkan
class VulkanContext class VulkanContext
{ {
public: public:
struct PhysicalDeviceInfo
{
PhysicalDeviceInfo(const PhysicalDeviceInfo&) = default;
explicit PhysicalDeviceInfo(VkPhysicalDevice device);
VkPhysicalDeviceFeatures features() const;
char deviceName[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE];
u8 pipelineCacheUUID[VK_UUID_SIZE];
u32 apiVersion;
u32 driverVersion;
u32 vendorID;
u32 deviceID;
VkDeviceSize minUniformBufferOffsetAlignment;
VkDeviceSize bufferImageGranularity;
u32 maxTexelBufferElements;
u32 maxImageDimension2D;
VkSampleCountFlags framebufferColorSampleCounts;
VkSampleCountFlags framebufferDepthSampleCounts;
float pointSizeRange[2];
float maxSamplerAnisotropy;
u32 subgroupSize = 1;
VkDriverId driverID = static_cast<VkDriverId>(0);
bool dualSrcBlend;
bool geometryShader;
bool samplerAnisotropy;
bool logicOp;
bool fragmentStoresAndAtomics;
bool sampleRateShading;
bool largePoints;
bool shaderStorageImageMultisample;
bool shaderTessellationAndGeometryPointSize;
bool occlusionQueryPrecise;
bool shaderClipDistance;
bool depthClamp;
bool textureCompressionBC;
bool shaderSubgroupOperations = false;
};
VulkanContext(VkInstance instance, VkPhysicalDevice physical_device); VulkanContext(VkInstance instance, VkPhysicalDevice physical_device);
~VulkanContext(); ~VulkanContext();
@ -37,10 +75,9 @@ public:
static void PopulateBackendInfo(VideoConfig* config); static void PopulateBackendInfo(VideoConfig* config);
static void PopulateBackendInfoAdapters(VideoConfig* config, const GPUList& gpu_list); static void PopulateBackendInfoAdapters(VideoConfig* config, const GPUList& gpu_list);
static void PopulateBackendInfoFeatures(VideoConfig* config, VkPhysicalDevice gpu, static void PopulateBackendInfoFeatures(VideoConfig* config, VkPhysicalDevice gpu,
const VkPhysicalDeviceProperties& properties, const PhysicalDeviceInfo& info);
const VkPhysicalDeviceFeatures& features);
static void PopulateBackendInfoMultisampleModes(VideoConfig* config, VkPhysicalDevice gpu, static void PopulateBackendInfoMultisampleModes(VideoConfig* config, VkPhysicalDevice gpu,
const VkPhysicalDeviceProperties& properties); const PhysicalDeviceInfo& info);
// Creates a Vulkan device context. // Creates a Vulkan device context.
// This assumes that PopulateBackendInfo and PopulateBackendInfoAdapters has already // This assumes that PopulateBackendInfo and PopulateBackendInfoAdapters has already
@ -65,39 +102,20 @@ public:
{ {
return m_graphics_queue_properties; return m_graphics_queue_properties;
} }
const VkPhysicalDeviceMemoryProperties& GetDeviceMemoryProperties() const const PhysicalDeviceInfo& GetDeviceInfo() const { return m_device_info; }
{
return m_device_memory_properties;
}
const VkPhysicalDeviceProperties& GetDeviceProperties() const { return m_device_properties; }
const VkPhysicalDeviceFeatures& GetDeviceFeatures() const { return m_device_features; }
const VkPhysicalDeviceLimits& GetDeviceLimits() const { return m_device_properties.limits; }
// Support bits // Support bits
bool SupportsAnisotropicFiltering() const bool SupportsAnisotropicFiltering() const { return m_device_info.samplerAnisotropy; }
{ bool SupportsPreciseOcclusionQueries() const { return m_device_info.occlusionQueryPrecise; }
return m_device_features.samplerAnisotropy == VK_TRUE; u32 GetShaderSubgroupSize() const { return m_device_info.subgroupSize; }
} bool SupportsShaderSubgroupOperations() const { return m_device_info.shaderSubgroupOperations; }
bool SupportsPreciseOcclusionQueries() const
{
return m_device_features.occlusionQueryPrecise == VK_TRUE;
}
u32 GetShaderSubgroupSize() const { return m_shader_subgroup_size; }
bool SupportsShaderSubgroupOperations() const { return m_supports_shader_subgroup_operations; }
// Helpers for getting constants // Helpers for getting constants
VkDeviceSize GetUniformBufferAlignment() const VkDeviceSize GetUniformBufferAlignment() const
{ {
return m_device_properties.limits.minUniformBufferOffsetAlignment; return m_device_info.minUniformBufferOffsetAlignment;
} }
VkDeviceSize GetTexelBufferAlignment() const VkDeviceSize GetBufferImageGranularity() const { return m_device_info.bufferImageGranularity; }
{ float GetMaxSamplerAnisotropy() const { return m_device_info.maxSamplerAnisotropy; }
return m_device_properties.limits.minUniformBufferOffsetAlignment;
}
VkDeviceSize GetBufferImageGranularity() const
{
return m_device_properties.limits.bufferImageGranularity;
}
float GetMaxSamplerAnisotropy() const { return m_device_properties.limits.maxSamplerAnisotropy; }
// Returns true if the specified extension is supported and enabled. // Returns true if the specified extension is supported and enabled.
bool SupportsDeviceExtension(const char* name) const; bool SupportsDeviceExtension(const char* name) const;
@ -118,10 +136,9 @@ private:
WindowSystemType wstype, bool enable_debug_utils, WindowSystemType wstype, bool enable_debug_utils,
bool validation_layer_enabled); bool validation_layer_enabled);
bool SelectDeviceExtensions(bool enable_surface); bool SelectDeviceExtensions(bool enable_surface);
bool SelectDeviceFeatures(); void WarnMissingDeviceFeatures();
bool CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer); bool CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer);
void InitDriverDetails(); void InitDriverDetails();
void PopulateShaderSubgroupSupport();
bool CreateAllocator(u32 vk_api_version); bool CreateAllocator(u32 vk_api_version);
VkInstance m_instance = VK_NULL_HANDLE; VkInstance m_instance = VK_NULL_HANDLE;
@ -137,12 +154,7 @@ private:
VkDebugUtilsMessengerEXT m_debug_utils_messenger = VK_NULL_HANDLE; VkDebugUtilsMessengerEXT m_debug_utils_messenger = VK_NULL_HANDLE;
VkPhysicalDeviceFeatures m_device_features = {}; PhysicalDeviceInfo m_device_info;
VkPhysicalDeviceProperties m_device_properties = {};
VkPhysicalDeviceMemoryProperties m_device_memory_properties = {};
u32 m_shader_subgroup_size = 1;
bool m_supports_shader_subgroup_operations = false;
std::vector<std::string> m_device_extensions; std::vector<std::string> m_device_extensions;
}; };