mirror of
https://github.com/RPCS3/rpcs3.git
synced 2025-07-03 05:21:25 +12:00
580bd2b25e
* Update asmjit dependency (aarch64 branch) * Disable USE_DISCORD_RPC by default * Dump some JIT objects in rpcs3 cache dir * Add SIGILL handler for all platforms * Fix resetting zeroing denormals in thread pool * Refactor most v128:: utils into global gv_** functions * Refactor PPU interpreter (incomplete), remove "precise" * - Instruction specializations with multiple accuracy flags * - Adjust calling convention for speed * - Removed precise/fast setting, replaced with static * - Started refactoring interpreters for building at runtime JIT * (I got tired of poor compiler optimizations) * - Expose some accuracy settings (SAT, NJ, VNAN, FPCC) * - Add exec_bytes PPU thread variable (akin to cycle count) * PPU LLVM: fix VCTUXS+VCTSXS instruction NaN results * SPU interpreter: remove "precise" for now (extremely non-portable) * - As with PPU, settings changed to static/dynamic for interpreters. * - Precise options will be implemented later * Fix termination after fatal error dialog
977 lines
33 KiB
C++
977 lines
33 KiB
C++
#include "device.h"
|
|
#include "instance.hpp"
|
|
#include "util/logs.hpp"
|
|
#include "Emu/system_config.h"
|
|
|
|
namespace vk
|
|
{
|
|
// Global shared render device
|
|
const render_device* g_render_device = nullptr;
|
|
|
|
void physical_device::get_physical_device_features(bool allow_extensions)
|
|
{
|
|
if (!allow_extensions)
|
|
{
|
|
vkGetPhysicalDeviceFeatures(dev, &features);
|
|
return;
|
|
}
|
|
|
|
supported_extensions instance_extensions(supported_extensions::instance);
|
|
supported_extensions device_extensions(supported_extensions::device, nullptr, dev);
|
|
|
|
if (!instance_extensions.is_supported(VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME))
|
|
{
|
|
vkGetPhysicalDeviceFeatures(dev, &features);
|
|
}
|
|
else
|
|
{
|
|
VkPhysicalDeviceFeatures2KHR features2;
|
|
features2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
|
|
features2.pNext = nullptr;
|
|
|
|
VkPhysicalDeviceFloat16Int8FeaturesKHR shader_support_info{};
|
|
VkPhysicalDeviceDescriptorIndexingFeatures descriptor_indexing_info{};
|
|
|
|
if (device_extensions.is_supported(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME))
|
|
{
|
|
shader_support_info.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR;
|
|
features2.pNext = &shader_support_info;
|
|
}
|
|
|
|
if (device_extensions.is_supported(VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME))
|
|
{
|
|
driver_properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR;
|
|
driver_properties.pNext = features2.pNext;
|
|
features2.pNext = &driver_properties;
|
|
}
|
|
|
|
if (device_extensions.is_supported(VK_EXT_DESCRIPTOR_INDEXING_EXTENSION_NAME))
|
|
{
|
|
descriptor_indexing_info.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT;
|
|
descriptor_indexing_info.pNext = features2.pNext;
|
|
features2.pNext = &descriptor_indexing_info;
|
|
descriptor_indexing_support = true;
|
|
}
|
|
|
|
auto _vkGetPhysicalDeviceFeatures2KHR = reinterpret_cast<PFN_vkGetPhysicalDeviceFeatures2KHR>(vkGetInstanceProcAddr(parent, "vkGetPhysicalDeviceFeatures2KHR"));
|
|
ensure(_vkGetPhysicalDeviceFeatures2KHR); // "vkGetInstanceProcAddress failed to find entry point!"
|
|
_vkGetPhysicalDeviceFeatures2KHR(dev, &features2);
|
|
|
|
shader_types_support.allow_float64 = !!features2.features.shaderFloat64;
|
|
shader_types_support.allow_float16 = !!shader_support_info.shaderFloat16;
|
|
shader_types_support.allow_int8 = !!shader_support_info.shaderInt8;
|
|
features = features2.features;
|
|
|
|
if (descriptor_indexing_support)
|
|
{
|
|
#define SET_DESCRIPTOR_BITFLAG(field, bit) if (descriptor_indexing_info.field) descriptor_update_after_bind_mask |= (1ull << bit)
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingUniformBufferUpdateAfterBind, VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingSampledImageUpdateAfterBind, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingSampledImageUpdateAfterBind, VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingStorageImageUpdateAfterBind, VK_DESCRIPTOR_TYPE_STORAGE_IMAGE);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingStorageBufferUpdateAfterBind, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingUniformTexelBufferUpdateAfterBind, VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingStorageTexelBufferUpdateAfterBind, VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER);
|
|
#undef SET_DESCRIPTOR_BITFLAG
|
|
}
|
|
}
|
|
|
|
stencil_export_support = device_extensions.is_supported(VK_EXT_SHADER_STENCIL_EXPORT_EXTENSION_NAME);
|
|
conditional_render_support = device_extensions.is_supported(VK_EXT_CONDITIONAL_RENDERING_EXTENSION_NAME);
|
|
external_memory_host_support = device_extensions.is_supported(VK_EXT_EXTERNAL_MEMORY_HOST_EXTENSION_NAME);
|
|
sampler_mirror_clamped_support = device_extensions.is_supported(VK_KHR_SAMPLER_MIRROR_CLAMP_TO_EDGE_EXTENSION_NAME);
|
|
unrestricted_depth_range_support = device_extensions.is_supported(VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME);
|
|
debug_utils_support = instance_extensions.is_supported(VK_EXT_DEBUG_UTILS_EXTENSION_NAME);
|
|
surface_capabilities_2_support = instance_extensions.is_supported(VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME);
|
|
}
|
|
|
|
void physical_device::get_physical_device_properties(bool allow_extensions)
|
|
{
|
|
vkGetPhysicalDeviceMemoryProperties(dev, &memory_properties);
|
|
|
|
if (!allow_extensions)
|
|
{
|
|
vkGetPhysicalDeviceProperties(dev, &props);
|
|
return;
|
|
}
|
|
|
|
supported_extensions instance_extensions(supported_extensions::instance);
|
|
if (!instance_extensions.is_supported(VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME))
|
|
{
|
|
vkGetPhysicalDeviceProperties(dev, &props);
|
|
}
|
|
else
|
|
{
|
|
VkPhysicalDeviceProperties2KHR properties2;
|
|
properties2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2_KHR;
|
|
properties2.pNext = nullptr;
|
|
|
|
VkPhysicalDeviceDescriptorIndexingPropertiesEXT descriptor_indexing_props{};
|
|
|
|
if (descriptor_indexing_support)
|
|
{
|
|
descriptor_indexing_props.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_PROPERTIES_EXT;
|
|
descriptor_indexing_props.pNext = properties2.pNext;
|
|
properties2.pNext = &descriptor_indexing_props;
|
|
}
|
|
|
|
auto _vkGetPhysicalDeviceProperties2KHR = reinterpret_cast<PFN_vkGetPhysicalDeviceProperties2KHR>(vkGetInstanceProcAddr(parent, "vkGetPhysicalDeviceProperties2KHR"));
|
|
ensure(_vkGetPhysicalDeviceProperties2KHR);
|
|
|
|
_vkGetPhysicalDeviceProperties2KHR(dev, &properties2);
|
|
props = properties2.properties;
|
|
|
|
if (descriptor_indexing_support)
|
|
{
|
|
if (descriptor_indexing_props.maxUpdateAfterBindDescriptorsInAllPools < 800'000)
|
|
{
|
|
rsx_log.error("Physical device does not support enough descriptors for deferred updates to work effectively. Deferred updates are disabled.");
|
|
descriptor_update_after_bind_mask = 0;
|
|
}
|
|
else if (descriptor_indexing_props.maxUpdateAfterBindDescriptorsInAllPools < 2'000'000)
|
|
{
|
|
rsx_log.warning("Physical device reports a low amount of allowed deferred descriptor updates. Draw call threshold will be lowered accordingly.");
|
|
descriptor_max_draw_calls = 8192;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void physical_device::create(VkInstance context, VkPhysicalDevice pdev, bool allow_extensions)
|
|
{
|
|
dev = pdev;
|
|
parent = context;
|
|
|
|
get_physical_device_features(allow_extensions);
|
|
get_physical_device_properties(allow_extensions);
|
|
|
|
rsx_log.always()("Found vulkan-compatible GPU: '%s' running on driver %s", get_name(), get_driver_version());
|
|
|
|
if (get_driver_vendor() == driver_vendor::RADV && get_name().find("LLVM 8.0.0") != umax)
|
|
{
|
|
// Serious driver bug causing black screens
|
|
// See https://bugs.freedesktop.org/show_bug.cgi?id=110970
|
|
rsx_log.fatal("RADV drivers have a major driver bug with LLVM 8.0.0 resulting in no visual output. Upgrade to LLVM version 8.0.1 or greater to avoid this issue.");
|
|
}
|
|
else if (get_driver_vendor() == driver_vendor::NVIDIA)
|
|
{
|
|
#ifdef _WIN32
|
|
// SPIRV bugs were fixed in 452.28 for windows
|
|
const u32 threshold_version = (452u << 22) | (28 << 14);
|
|
#else
|
|
// SPIRV bugs were fixed in 450.56 for linux/BSD
|
|
const u32 threshold_version = (450u << 22) | (56 << 14);
|
|
#endif
|
|
const auto current_version = props.driverVersion & ~0x3fffu; // Clear patch and revision fields
|
|
if (current_version < threshold_version)
|
|
{
|
|
rsx_log.error("Your current NVIDIA graphics driver version %s has known issues and is unsupported. Update to the latest NVIDIA driver.", get_driver_version());
|
|
}
|
|
}
|
|
|
|
if (get_chip_class() == chip_class::AMD_vega)
|
|
{
|
|
// Disable fp16 if driver uses LLVM emitter. It does fine with AMD proprietary drivers though.
|
|
shader_types_support.allow_float16 = (driver_properties.driverID == VK_DRIVER_ID_AMD_PROPRIETARY_KHR);
|
|
}
|
|
}
|
|
|
|
std::string physical_device::get_name() const
|
|
{
|
|
return props.deviceName;
|
|
}
|
|
|
|
driver_vendor physical_device::get_driver_vendor() const
|
|
{
|
|
if (!driver_properties.driverID)
|
|
{
|
|
const auto gpu_name = get_name();
|
|
if (gpu_name.find("Radeon") != umax)
|
|
{
|
|
return driver_vendor::AMD;
|
|
}
|
|
|
|
if (gpu_name.find("NVIDIA") != umax || gpu_name.find("GeForce") != umax || gpu_name.find("Quadro") != umax)
|
|
{
|
|
return driver_vendor::NVIDIA;
|
|
}
|
|
|
|
if (gpu_name.find("RADV") != umax)
|
|
{
|
|
return driver_vendor::RADV;
|
|
}
|
|
|
|
if (gpu_name.find("Intel") != umax)
|
|
{
|
|
#ifdef _WIN32
|
|
return driver_vendor::INTEL;
|
|
#else
|
|
return driver_vendor::ANV;
|
|
#endif
|
|
}
|
|
|
|
return driver_vendor::unknown;
|
|
}
|
|
else
|
|
{
|
|
switch (driver_properties.driverID)
|
|
{
|
|
case VK_DRIVER_ID_AMD_PROPRIETARY_KHR:
|
|
case VK_DRIVER_ID_AMD_OPEN_SOURCE_KHR:
|
|
return driver_vendor::AMD;
|
|
case VK_DRIVER_ID_MESA_RADV_KHR:
|
|
return driver_vendor::RADV;
|
|
case VK_DRIVER_ID_NVIDIA_PROPRIETARY_KHR:
|
|
return driver_vendor::NVIDIA;
|
|
case VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS_KHR:
|
|
return driver_vendor::INTEL;
|
|
case VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA_KHR:
|
|
return driver_vendor::ANV;
|
|
default:
|
|
// Mobile?
|
|
return driver_vendor::unknown;
|
|
}
|
|
}
|
|
}
|
|
|
|
std::string physical_device::get_driver_version() const
|
|
{
|
|
switch (get_driver_vendor())
|
|
{
|
|
case driver_vendor::NVIDIA:
|
|
{
|
|
// 10 + 8 + 8 + 6
|
|
const auto major_version = props.driverVersion >> 22;
|
|
const auto minor_version = (props.driverVersion >> 14) & 0xff;
|
|
const auto patch = (props.driverVersion >> 6) & 0xff;
|
|
const auto revision = (props.driverVersion & 0x3f);
|
|
|
|
return fmt::format("%u.%u.%u.%u", major_version, minor_version, patch, revision);
|
|
}
|
|
default:
|
|
{
|
|
// 10 + 10 + 12 (standard vulkan encoding created with VK_MAKE_VERSION)
|
|
return fmt::format("%u.%u.%u", (props.driverVersion >> 22), (props.driverVersion >> 12) & 0x3ff, (props.driverVersion) & 0x3ff);
|
|
}
|
|
}
|
|
}
|
|
|
|
chip_class physical_device::get_chip_class() const
|
|
{
|
|
return get_chip_family(props.vendorID, props.deviceID);
|
|
}
|
|
|
|
u32 physical_device::get_queue_count() const
|
|
{
|
|
if (!queue_props.empty())
|
|
return ::size32(queue_props);
|
|
|
|
u32 count = 0;
|
|
vkGetPhysicalDeviceQueueFamilyProperties(dev, &count, nullptr);
|
|
|
|
return count;
|
|
}
|
|
|
|
const VkQueueFamilyProperties& physical_device::get_queue_properties(u32 queue)
|
|
{
|
|
if (queue_props.empty())
|
|
{
|
|
u32 count = 0;
|
|
vkGetPhysicalDeviceQueueFamilyProperties(dev, &count, nullptr);
|
|
|
|
queue_props.resize(count);
|
|
vkGetPhysicalDeviceQueueFamilyProperties(dev, &count, queue_props.data());
|
|
}
|
|
|
|
if (queue >= queue_props.size())
|
|
fmt::throw_exception("Bad queue index passed to get_queue_properties (%u)", queue);
|
|
return queue_props[queue];
|
|
}
|
|
|
|
const VkPhysicalDeviceMemoryProperties& physical_device::get_memory_properties() const
|
|
{
|
|
return memory_properties;
|
|
}
|
|
|
|
const VkPhysicalDeviceLimits& physical_device::get_limits() const
|
|
{
|
|
return props.limits;
|
|
}
|
|
|
|
physical_device::operator VkPhysicalDevice() const
|
|
{
|
|
return dev;
|
|
}
|
|
|
|
physical_device::operator VkInstance() const
|
|
{
|
|
return parent;
|
|
}
|
|
|
|
// Render Device - The actual usable device
|
|
void render_device::create(vk::physical_device& pdev, u32 graphics_queue_idx, u32 present_queue_idx, u32 transfer_queue_idx)
|
|
{
|
|
std::string message_on_error;
|
|
float queue_priorities[1] = { 0.f };
|
|
pgpu = &pdev;
|
|
|
|
ensure(graphics_queue_idx == present_queue_idx || present_queue_idx == umax); // TODO
|
|
std::vector<VkDeviceQueueCreateInfo> device_queues;
|
|
|
|
auto& graphics_queue = device_queues.emplace_back();
|
|
graphics_queue.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO;
|
|
graphics_queue.pNext = NULL;
|
|
graphics_queue.flags = 0;
|
|
graphics_queue.queueFamilyIndex = graphics_queue_idx;
|
|
graphics_queue.queueCount = 1;
|
|
graphics_queue.pQueuePriorities = queue_priorities;
|
|
|
|
u32 transfer_queue_sub_index = 0;
|
|
if (transfer_queue_idx == umax)
|
|
{
|
|
// Transfer queue must be a valid device queue
|
|
rsx_log.warning("Dedicated transfer+compute queue was not found on this GPU. Will use graphics queue instead.");
|
|
transfer_queue_idx = graphics_queue_idx;
|
|
|
|
// Check if we can at least get a second graphics queue
|
|
if (pdev.get_queue_properties(graphics_queue_idx).queueCount > 1)
|
|
{
|
|
rsx_log.notice("Will use a spare graphics queue to push transfer operations.");
|
|
graphics_queue.queueCount++;
|
|
transfer_queue_sub_index = 1;
|
|
}
|
|
}
|
|
|
|
m_graphics_queue_family = graphics_queue_idx;
|
|
m_present_queue_family = present_queue_idx;
|
|
m_transfer_queue_family = transfer_queue_idx;
|
|
|
|
if (graphics_queue_idx != transfer_queue_idx)
|
|
{
|
|
auto& transfer_queue = device_queues.emplace_back();
|
|
transfer_queue.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO;
|
|
transfer_queue.pNext = NULL;
|
|
transfer_queue.flags = 0;
|
|
transfer_queue.queueFamilyIndex = transfer_queue_idx;
|
|
transfer_queue.queueCount = 1;
|
|
transfer_queue.pQueuePriorities = queue_priorities;
|
|
}
|
|
|
|
// Set up instance information
|
|
std::vector<const char*> requested_extensions = { VK_KHR_SWAPCHAIN_EXTENSION_NAME };
|
|
|
|
// Enable hardware features manually
|
|
// Currently we require:
|
|
// 1. Anisotropic sampling
|
|
// 2. DXT support
|
|
// 3. Indexable storage buffers
|
|
VkPhysicalDeviceFeatures enabled_features{};
|
|
if (pgpu->shader_types_support.allow_float16)
|
|
{
|
|
requested_extensions.push_back(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME);
|
|
}
|
|
|
|
if (pgpu->conditional_render_support)
|
|
{
|
|
requested_extensions.push_back(VK_EXT_CONDITIONAL_RENDERING_EXTENSION_NAME);
|
|
}
|
|
|
|
if (pgpu->unrestricted_depth_range_support)
|
|
{
|
|
requested_extensions.push_back(VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME);
|
|
}
|
|
|
|
if (pgpu->external_memory_host_support)
|
|
{
|
|
requested_extensions.push_back(VK_KHR_EXTERNAL_MEMORY_EXTENSION_NAME);
|
|
requested_extensions.push_back(VK_EXT_EXTERNAL_MEMORY_HOST_EXTENSION_NAME);
|
|
}
|
|
|
|
if (pgpu->stencil_export_support)
|
|
{
|
|
requested_extensions.push_back(VK_EXT_SHADER_STENCIL_EXPORT_EXTENSION_NAME);
|
|
}
|
|
|
|
if (pgpu->sampler_mirror_clamped_support)
|
|
{
|
|
requested_extensions.push_back(VK_KHR_SAMPLER_MIRROR_CLAMP_TO_EDGE_EXTENSION_NAME);
|
|
}
|
|
|
|
if (pgpu->descriptor_indexing_support)
|
|
{
|
|
requested_extensions.push_back(VK_KHR_MAINTENANCE3_EXTENSION_NAME);
|
|
requested_extensions.push_back(VK_EXT_DESCRIPTOR_INDEXING_EXTENSION_NAME);
|
|
}
|
|
|
|
enabled_features.robustBufferAccess = VK_TRUE;
|
|
enabled_features.fullDrawIndexUint32 = VK_TRUE;
|
|
enabled_features.independentBlend = VK_TRUE;
|
|
enabled_features.logicOp = VK_TRUE;
|
|
enabled_features.depthClamp = VK_TRUE;
|
|
enabled_features.depthBounds = VK_TRUE;
|
|
enabled_features.wideLines = VK_TRUE;
|
|
enabled_features.largePoints = VK_TRUE;
|
|
enabled_features.shaderFloat64 = VK_TRUE;
|
|
|
|
if (g_cfg.video.antialiasing_level != msaa_level::none)
|
|
{
|
|
// MSAA features
|
|
enabled_features.sampleRateShading = VK_TRUE;
|
|
enabled_features.alphaToOne = VK_TRUE;
|
|
enabled_features.shaderStorageImageMultisample = VK_TRUE;
|
|
// enabled_features.shaderStorageImageReadWithoutFormat = VK_TRUE; // Unused currently, may be needed soon
|
|
enabled_features.shaderStorageImageWriteWithoutFormat = VK_TRUE;
|
|
}
|
|
|
|
if (g_cfg.video.precise_zpass_count)
|
|
{
|
|
enabled_features.occlusionQueryPrecise = VK_TRUE;
|
|
}
|
|
|
|
// enabled_features.shaderSampledImageArrayDynamicIndexing = TRUE; // Unused currently but will be needed soon
|
|
enabled_features.shaderClipDistance = VK_TRUE;
|
|
// enabled_features.shaderCullDistance = VK_TRUE; // Alt notation of clip distance
|
|
|
|
enabled_features.samplerAnisotropy = VK_TRUE;
|
|
enabled_features.textureCompressionBC = VK_TRUE;
|
|
enabled_features.shaderStorageBufferArrayDynamicIndexing = VK_TRUE;
|
|
|
|
// Optionally disable unsupported stuff
|
|
if (!pgpu->features.shaderStorageImageMultisample || !pgpu->features.shaderStorageImageWriteWithoutFormat)
|
|
{
|
|
// Disable MSAA if any of these two features are unsupported
|
|
if (g_cfg.video.antialiasing_level != msaa_level::none)
|
|
{
|
|
rsx_log.error("Your GPU driver does not support some required MSAA features. MSAA will be disabled.");
|
|
g_cfg.video.antialiasing_level.set(msaa_level::none);
|
|
}
|
|
|
|
enabled_features.sampleRateShading = VK_FALSE;
|
|
enabled_features.alphaToOne = VK_FALSE;
|
|
enabled_features.shaderStorageImageMultisample = VK_FALSE;
|
|
enabled_features.shaderStorageImageWriteWithoutFormat = VK_FALSE;
|
|
}
|
|
|
|
if (!pgpu->features.shaderClipDistance)
|
|
{
|
|
rsx_log.error("Your GPU does not support shader clip distance. Graphics will not render correctly.");
|
|
enabled_features.shaderClipDistance = VK_FALSE;
|
|
}
|
|
|
|
if (!pgpu->features.shaderStorageBufferArrayDynamicIndexing)
|
|
{
|
|
rsx_log.error("Your GPU does not support shader storage buffer array dynamic indexing. Graphics will not render correctly.");
|
|
enabled_features.shaderStorageBufferArrayDynamicIndexing = VK_FALSE;
|
|
}
|
|
|
|
if (!pgpu->features.samplerAnisotropy)
|
|
{
|
|
rsx_log.error("Your GPU does not support anisotropic filtering. Graphics may not render correctly.");
|
|
enabled_features.samplerAnisotropy = VK_FALSE;
|
|
}
|
|
|
|
if (!pgpu->features.shaderFloat64)
|
|
{
|
|
rsx_log.error("Your GPU does not support double precision floats in shaders. Graphics may not render correctly.");
|
|
enabled_features.shaderFloat64 = VK_FALSE;
|
|
}
|
|
|
|
if (!pgpu->features.depthBounds)
|
|
{
|
|
rsx_log.error("Your GPU does not support depth bounds testing. Graphics may not render correctly.");
|
|
enabled_features.depthBounds = VK_FALSE;
|
|
}
|
|
|
|
if (!pgpu->features.wideLines)
|
|
{
|
|
rsx_log.error("Your GPU does not support wide lines. Graphics may not render correctly.");
|
|
enabled_features.wideLines = VK_FALSE;
|
|
}
|
|
|
|
if (!pgpu->features.sampleRateShading && enabled_features.sampleRateShading)
|
|
{
|
|
rsx_log.error("Your GPU does not support sample rate shading for multisampling. Graphics may be inaccurate when MSAA is enabled.");
|
|
enabled_features.sampleRateShading = VK_FALSE;
|
|
}
|
|
|
|
if (!pgpu->features.alphaToOne && enabled_features.alphaToOne)
|
|
{
|
|
// AMD proprietary drivers do not expose alphaToOne support
|
|
rsx_log.error("Your GPU does not support alpha-to-one for multisampling. Graphics may be inaccurate when MSAA is enabled.");
|
|
enabled_features.alphaToOne = VK_FALSE;
|
|
}
|
|
|
|
if (!pgpu->features.occlusionQueryPrecise && enabled_features.occlusionQueryPrecise)
|
|
{
|
|
rsx_log.error("Your GPU does not support precise occlusion queries. Graphics may not render correctly.");
|
|
enabled_features.occlusionQueryPrecise = VK_FALSE;
|
|
}
|
|
|
|
#ifdef __APPLE__
|
|
if (!pgpu->features.logicOp)
|
|
{
|
|
rsx_log.error("Your GPU does not support framebuffer logical operations. Graphics may not render correctly.");
|
|
enabled_features.logicOp = VK_FALSE;
|
|
}
|
|
#endif
|
|
|
|
VkDeviceCreateInfo device = {};
|
|
device.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO;
|
|
device.pNext = nullptr;
|
|
device.queueCreateInfoCount = ::size32(device_queues);
|
|
device.pQueueCreateInfos = device_queues.data();
|
|
device.enabledLayerCount = 0;
|
|
device.ppEnabledLayerNames = nullptr; // Deprecated
|
|
device.enabledExtensionCount = ::size32(requested_extensions);
|
|
device.ppEnabledExtensionNames = requested_extensions.data();
|
|
device.pEnabledFeatures = &enabled_features;
|
|
|
|
VkPhysicalDeviceFloat16Int8FeaturesKHR shader_support_info{};
|
|
if (pgpu->shader_types_support.allow_float16)
|
|
{
|
|
// Allow use of f16 type in shaders if possible
|
|
shader_support_info.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR;
|
|
shader_support_info.shaderFloat16 = VK_TRUE;
|
|
device.pNext = &shader_support_info;
|
|
|
|
rsx_log.notice("GPU/driver supports float16 data types natively. Using native float16_t variables if possible.");
|
|
}
|
|
else
|
|
{
|
|
rsx_log.notice("GPU/driver lacks support for float16 data types. All float16_t arithmetic will be emulated with float32_t.");
|
|
}
|
|
|
|
VkPhysicalDeviceDescriptorIndexingFeatures indexing_features{};
|
|
if (pgpu->descriptor_indexing_support)
|
|
{
|
|
#define SET_DESCRIPTOR_BITFLAG(field, bit) if (pgpu->descriptor_update_after_bind_mask & (1ull << bit)) indexing_features.field = VK_TRUE
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingUniformBufferUpdateAfterBind, VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingSampledImageUpdateAfterBind, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingSampledImageUpdateAfterBind, VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingStorageImageUpdateAfterBind, VK_DESCRIPTOR_TYPE_STORAGE_IMAGE);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingStorageBufferUpdateAfterBind, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingUniformTexelBufferUpdateAfterBind, VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER);
|
|
SET_DESCRIPTOR_BITFLAG(descriptorBindingStorageTexelBufferUpdateAfterBind, VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER);
|
|
#undef SET_DESCRIPTOR_BITFLAG
|
|
|
|
indexing_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT;
|
|
device.pNext = &indexing_features;
|
|
}
|
|
|
|
CHECK_RESULT_EX(vkCreateDevice(*pgpu, &device, nullptr, &dev), message_on_error);
|
|
|
|
// Initialize queues
|
|
vkGetDeviceQueue(dev, graphics_queue_idx, 0, &m_graphics_queue);
|
|
vkGetDeviceQueue(dev, transfer_queue_idx, transfer_queue_sub_index, &m_transfer_queue);
|
|
|
|
if (present_queue_idx != umax)
|
|
{
|
|
vkGetDeviceQueue(dev, present_queue_idx, 0, &m_present_queue);
|
|
}
|
|
|
|
// Import optional function endpoints
|
|
if (pgpu->conditional_render_support)
|
|
{
|
|
_vkCmdBeginConditionalRenderingEXT = reinterpret_cast<PFN_vkCmdBeginConditionalRenderingEXT>(vkGetDeviceProcAddr(dev, "vkCmdBeginConditionalRenderingEXT"));
|
|
_vkCmdEndConditionalRenderingEXT = reinterpret_cast<PFN_vkCmdEndConditionalRenderingEXT>(vkGetDeviceProcAddr(dev, "vkCmdEndConditionalRenderingEXT"));
|
|
}
|
|
|
|
if (pgpu->debug_utils_support)
|
|
{
|
|
_vkSetDebugUtilsObjectNameEXT = reinterpret_cast<PFN_vkSetDebugUtilsObjectNameEXT>(vkGetDeviceProcAddr(dev, "vkSetDebugUtilsObjectNameEXT"));
|
|
_vkQueueInsertDebugUtilsLabelEXT = reinterpret_cast<PFN_vkQueueInsertDebugUtilsLabelEXT>(vkGetDeviceProcAddr(dev, "vkQueueInsertDebugUtilsLabelEXT"));
|
|
_vkCmdInsertDebugUtilsLabelEXT = reinterpret_cast<PFN_vkCmdInsertDebugUtilsLabelEXT>(vkGetDeviceProcAddr(dev, "vkCmdInsertDebugUtilsLabelEXT"));
|
|
}
|
|
|
|
memory_map = vk::get_memory_mapping(pdev);
|
|
m_formats_support = vk::get_optimal_tiling_supported_formats(pdev);
|
|
m_pipeline_binding_table = vk::get_pipeline_binding_table(pdev);
|
|
|
|
if (pgpu->external_memory_host_support)
|
|
{
|
|
memory_map._vkGetMemoryHostPointerPropertiesEXT = reinterpret_cast<PFN_vkGetMemoryHostPointerPropertiesEXT>(vkGetDeviceProcAddr(dev, "vkGetMemoryHostPointerPropertiesEXT"));
|
|
}
|
|
|
|
if (g_cfg.video.disable_vulkan_mem_allocator)
|
|
m_allocator = std::make_unique<vk::mem_allocator_vk>(dev, pdev);
|
|
else
|
|
m_allocator = std::make_unique<vk::mem_allocator_vma>(dev, pdev);
|
|
}
|
|
|
|
void render_device::destroy()
|
|
{
|
|
if (g_render_device == this)
|
|
{
|
|
g_render_device = nullptr;
|
|
}
|
|
|
|
if (dev && pgpu)
|
|
{
|
|
if (m_allocator)
|
|
{
|
|
m_allocator->destroy();
|
|
m_allocator.reset();
|
|
}
|
|
|
|
vkDestroyDevice(dev, nullptr);
|
|
dev = nullptr;
|
|
memory_map = {};
|
|
m_formats_support = {};
|
|
}
|
|
}
|
|
|
|
VkQueue render_device::get_present_queue() const
|
|
{
|
|
return m_present_queue;
|
|
}
|
|
|
|
VkQueue render_device::get_graphics_queue() const
|
|
{
|
|
return m_graphics_queue;
|
|
}
|
|
|
|
VkQueue render_device::get_transfer_queue() const
|
|
{
|
|
return m_transfer_queue;
|
|
}
|
|
|
|
u32 render_device::get_graphics_queue_family() const
|
|
{
|
|
return m_graphics_queue_family;
|
|
}
|
|
|
|
u32 render_device::get_present_queue_family() const
|
|
{
|
|
return m_graphics_queue_family;
|
|
}
|
|
|
|
u32 render_device::get_transfer_queue_family() const
|
|
{
|
|
return m_transfer_queue_family;
|
|
}
|
|
|
|
const VkFormatProperties render_device::get_format_properties(VkFormat format) const
|
|
{
|
|
auto found = pgpu->format_properties.find(format);
|
|
if (found != pgpu->format_properties.end())
|
|
{
|
|
return found->second;
|
|
}
|
|
|
|
auto& props = pgpu->format_properties[format];
|
|
vkGetPhysicalDeviceFormatProperties(*pgpu, format, &props);
|
|
return props;
|
|
}
|
|
|
|
bool render_device::get_compatible_memory_type(u32 typeBits, u32 desired_mask, u32* type_index) const
|
|
{
|
|
VkPhysicalDeviceMemoryProperties mem_infos = pgpu->get_memory_properties();
|
|
|
|
for (u32 i = 0; i < 32; i++)
|
|
{
|
|
if ((typeBits & 1) == 1)
|
|
{
|
|
if ((mem_infos.memoryTypes[i].propertyFlags & desired_mask) == desired_mask)
|
|
{
|
|
if (type_index)
|
|
{
|
|
*type_index = i;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
}
|
|
|
|
typeBits >>= 1;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
const physical_device& render_device::gpu() const
|
|
{
|
|
return *pgpu;
|
|
}
|
|
|
|
const memory_type_mapping& render_device::get_memory_mapping() const
|
|
{
|
|
return memory_map;
|
|
}
|
|
|
|
const gpu_formats_support& render_device::get_formats_support() const
|
|
{
|
|
return m_formats_support;
|
|
}
|
|
|
|
const pipeline_binding_table& render_device::get_pipeline_binding_table() const
|
|
{
|
|
return m_pipeline_binding_table;
|
|
}
|
|
|
|
const gpu_shader_types_support& render_device::get_shader_types_support() const
|
|
{
|
|
return pgpu->shader_types_support;
|
|
}
|
|
|
|
bool render_device::get_shader_stencil_export_support() const
|
|
{
|
|
return pgpu->stencil_export_support;
|
|
}
|
|
|
|
bool render_device::get_depth_bounds_support() const
|
|
{
|
|
return pgpu->features.depthBounds != VK_FALSE;
|
|
}
|
|
|
|
bool render_device::get_alpha_to_one_support() const
|
|
{
|
|
return pgpu->features.alphaToOne != VK_FALSE;
|
|
}
|
|
|
|
bool render_device::get_anisotropic_filtering_support() const
|
|
{
|
|
return pgpu->features.samplerAnisotropy != VK_FALSE;
|
|
}
|
|
|
|
bool render_device::get_wide_lines_support() const
|
|
{
|
|
return pgpu->features.wideLines != VK_FALSE;
|
|
}
|
|
|
|
bool render_device::get_conditional_render_support() const
|
|
{
|
|
return pgpu->conditional_render_support;
|
|
}
|
|
|
|
bool render_device::get_unrestricted_depth_range_support() const
|
|
{
|
|
return pgpu->unrestricted_depth_range_support;
|
|
}
|
|
|
|
bool render_device::get_external_memory_host_support() const
|
|
{
|
|
return pgpu->external_memory_host_support;
|
|
}
|
|
|
|
bool render_device::get_surface_capabilities_2_support() const
|
|
{
|
|
return pgpu->surface_capabilities_2_support;
|
|
}
|
|
|
|
bool render_device::get_debug_utils_support() const
|
|
{
|
|
return g_cfg.video.renderdoc_compatiblity && pgpu->debug_utils_support;
|
|
}
|
|
|
|
bool render_device::get_descriptor_indexing_support() const
|
|
{
|
|
return pgpu->descriptor_indexing_support;
|
|
}
|
|
|
|
u64 render_device::get_descriptor_update_after_bind_support() const
|
|
{
|
|
return pgpu->descriptor_update_after_bind_mask;
|
|
}
|
|
|
|
u32 render_device::get_descriptor_max_draw_calls() const
|
|
{
|
|
return pgpu->descriptor_max_draw_calls;
|
|
}
|
|
|
|
mem_allocator_base* render_device::get_allocator() const
|
|
{
|
|
return m_allocator.get();
|
|
}
|
|
|
|
render_device::operator VkDevice() const
|
|
{
|
|
return dev;
|
|
}
|
|
|
|
void render_device::rebalance_memory_type_usage()
|
|
{
|
|
// Rebalance device local memory types
|
|
memory_map.device_local.rebalance();
|
|
}
|
|
|
|
// Shared Util
|
|
memory_type_mapping get_memory_mapping(const vk::physical_device& dev)
|
|
{
|
|
VkPhysicalDevice pdev = dev;
|
|
VkPhysicalDeviceMemoryProperties memory_properties;
|
|
vkGetPhysicalDeviceMemoryProperties(pdev, &memory_properties);
|
|
|
|
memory_type_mapping result;
|
|
result.device_local_total_bytes = 0;
|
|
result.host_visible_total_bytes = 0;
|
|
result.device_bar_total_bytes = 0;
|
|
|
|
// Sort the confusingly laid out heap-type map into something easier to scan.
|
|
// Not performance-critical, this method is called once at initialization.
|
|
struct memory_type
|
|
{
|
|
u32 type_index;
|
|
VkFlags flags;
|
|
VkDeviceSize size;
|
|
};
|
|
|
|
struct heap_type_map_entry
|
|
{
|
|
VkMemoryHeap heap;
|
|
std::vector<memory_type> types;
|
|
};
|
|
|
|
std::vector<heap_type_map_entry> memory_heap_map;
|
|
for (u32 i = 0; i < memory_properties.memoryHeapCount; ++i)
|
|
{
|
|
memory_heap_map.push_back(
|
|
{
|
|
.heap = memory_properties.memoryHeaps[i],
|
|
.types = {}
|
|
});
|
|
}
|
|
|
|
for (u32 i = 0; i < memory_properties.memoryTypeCount; i++)
|
|
{
|
|
auto& type_info = memory_properties.memoryTypes[i];
|
|
memory_heap_map[type_info.heapIndex].types.push_back({ i, type_info.propertyFlags, 0 });
|
|
}
|
|
|
|
auto find_memory_type_with_property = [&memory_heap_map](VkFlags desired_flags, VkFlags excluded_flags)
|
|
{
|
|
std::vector<memory_type> results;
|
|
|
|
for (auto& heap : memory_heap_map)
|
|
{
|
|
for (auto &type : heap.types)
|
|
{
|
|
if (((type.flags & desired_flags) == desired_flags) && !(type.flags & excluded_flags))
|
|
{
|
|
// Match, only once allowed per heap!
|
|
results.push_back({ type.type_index, type.flags, heap.heap.size });
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
return results;
|
|
};
|
|
|
|
auto device_local_types = find_memory_type_with_property(
|
|
VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
|
|
(VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD | VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD));
|
|
auto host_coherent_types = find_memory_type_with_property(
|
|
(VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT),
|
|
0);
|
|
auto bar_memory_types = find_memory_type_with_property(
|
|
(VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT),
|
|
0);
|
|
|
|
if (host_coherent_types.empty())
|
|
{
|
|
rsx_log.warning("[Performance Warning] Could not identify a cached upload heap. Will fall back to uncached transport.");
|
|
host_coherent_types = find_memory_type_with_property(
|
|
(VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT),
|
|
0);
|
|
}
|
|
|
|
ensure(!device_local_types.empty());
|
|
ensure(!host_coherent_types.empty());
|
|
|
|
// BAR heap, currently parked for future use, I have some plans for it (kd-11)
|
|
for (auto& type : bar_memory_types)
|
|
{
|
|
result.device_bar.push(type.type_index, type.size);
|
|
result.device_bar_total_bytes += type.size;
|
|
}
|
|
|
|
// Generic VRAM access, requires some minor prioritization based on flags
|
|
// Most devices have a 'PURE' device local type, pin that as the first priority
|
|
// Internally, there will be some reshuffling based on memory load later, but this is rare
|
|
if (device_local_types.size() > 1)
|
|
{
|
|
std::sort(device_local_types.begin(), device_local_types.end(), [](const auto& a, const auto& b)
|
|
{
|
|
if (a.flags == b.flags)
|
|
{
|
|
return a.size > b.size;
|
|
}
|
|
|
|
return (a.flags == VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) || (b.flags != VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT && a.size > b.size);
|
|
});
|
|
}
|
|
|
|
for (auto& type : device_local_types)
|
|
{
|
|
result.device_local.push(type.type_index, type.size);
|
|
result.device_local_total_bytes += type.size;
|
|
}
|
|
|
|
// Sort upload heap entries based on size.
|
|
if (host_coherent_types.size() > 1)
|
|
{
|
|
std::sort(host_coherent_types.begin(), host_coherent_types.end(), [](const auto& a, const auto& b)
|
|
{
|
|
return a.size > b.size;
|
|
});
|
|
}
|
|
|
|
for (auto& type : host_coherent_types)
|
|
{
|
|
result.host_visible_coherent.push(type.type_index, type.size);
|
|
result.host_visible_total_bytes += type.size;
|
|
}
|
|
|
|
rsx_log.notice("Detected %llu MB of device local memory", result.device_local_total_bytes / (0x100000));
|
|
rsx_log.notice("Detected %llu MB of host coherent memory", result.host_visible_total_bytes / (0x100000));
|
|
rsx_log.notice("Detected %llu MB of BAR memory", result.device_bar_total_bytes / (0x100000));
|
|
|
|
return result;
|
|
}
|
|
|
|
gpu_formats_support get_optimal_tiling_supported_formats(const physical_device& dev)
|
|
{
|
|
gpu_formats_support result = {};
|
|
|
|
VkFormatProperties props;
|
|
vkGetPhysicalDeviceFormatProperties(dev, VK_FORMAT_D24_UNORM_S8_UINT, &props);
|
|
|
|
result.d24_unorm_s8 = !!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT) && !!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT) &&
|
|
!!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_BLIT_SRC_BIT) && !!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_LINEAR_BIT);
|
|
|
|
vkGetPhysicalDeviceFormatProperties(dev, VK_FORMAT_D32_SFLOAT_S8_UINT, &props);
|
|
result.d32_sfloat_s8 = !!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT) && !!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT) &&
|
|
!!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_BLIT_SRC_BIT);
|
|
|
|
// Hide d24_s8 if force high precision z buffer is enabled
|
|
if (g_cfg.video.force_high_precision_z_buffer && result.d32_sfloat_s8)
|
|
result.d24_unorm_s8 = false;
|
|
|
|
// Checks if BGRA8 images can be used for blitting
|
|
vkGetPhysicalDeviceFormatProperties(dev, VK_FORMAT_B8G8R8A8_UNORM, &props);
|
|
result.bgra8_linear = !!(props.linearTilingFeatures & VK_FORMAT_FEATURE_BLIT_SRC_BIT);
|
|
|
|
// Check if device supports RGBA8 format
|
|
vkGetPhysicalDeviceFormatProperties(dev, VK_FORMAT_R8G8B8A8_UNORM, &props);
|
|
if (!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT) || !(props.optimalTilingFeatures & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT) ||
|
|
!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_BLIT_SRC_BIT))
|
|
{
|
|
// Non-fatal. Most games use BGRA layout due to legacy reasons as old GPUs typically supported BGRA and RGBA was emulated.
|
|
rsx_log.error("Your GPU and/or driver does not support RGBA8 format. This can cause problems in some rare games that use this memory layout.");
|
|
}
|
|
|
|
result.argb8_linear = !!(props.linearTilingFeatures & VK_FORMAT_FEATURE_BLIT_SRC_BIT);
|
|
return result;
|
|
}
|
|
|
|
pipeline_binding_table get_pipeline_binding_table(const vk::physical_device& dev)
|
|
{
|
|
pipeline_binding_table result{};
|
|
|
|
// Need to check how many samplers are supported by the driver
|
|
const auto usable_samplers = std::min(dev.get_limits().maxPerStageDescriptorSampledImages, 32u);
|
|
result.vertex_textures_first_bind_slot = result.textures_first_bind_slot + usable_samplers;
|
|
result.total_descriptor_bindings = result.vertex_textures_first_bind_slot + 4;
|
|
return result;
|
|
}
|
|
}
|