KytyPS5/src/graphics/host_gpu/vma.cpp

#include "graphics/host_gpu/vulkanCommon.h"

#if defined(__clang__)
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wnullability-completeness"
#pragma clang diagnostic ignored "-Wunused-private-field"
#pragma clang diagnostic ignored "-Wunused-variable"
#endif

#define VMA_IMPLEMENTATION
#include <vk_mem_alloc.h>

#if defined(__clang__)
#pragma clang diagnostic pop
#endif

#include "common/assert.h"
#include "common/logging/log.h"
#include "common/profiler.h"
#include "graphics/host_gpu/graphicContext.h"
#include "graphics/host_gpu/vma.h"

#include <algorithm>
#include <atomic>
#include <cinttypes>

namespace Libs::Graphics {

namespace {

struct MemoryStats {
	std::atomic_uint64_t allocated[VK_MAX_MEMORY_TYPES] {};
	std::atomic_uint64_t count[VK_MAX_MEMORY_TYPES] {};
};

MemoryStats g_memory_stats;

} // namespace

void VulkanTrackAllocation(const VulkanMemory& memory) {
	g_memory_stats.allocated[memory.type] += memory.requirements.size;
	g_memory_stats.count[memory.type]++;
}

void VulkanUntrackAllocation(const VulkanMemory& memory) {
	EXIT_IF(g_memory_stats.allocated[memory.type] < memory.requirements.size);
	EXIT_IF(g_memory_stats.count[memory.type] == 0);
	g_memory_stats.allocated[memory.type] -= memory.requirements.size;
	g_memory_stats.count[memory.type]--;
}

bool GraphicContext::CreateAllocator() {
	KYTY_PROFILER_FUNCTION();
	EXIT_IF(instance == nullptr || physical_device == nullptr || device == nullptr ||
	        allocator != nullptr);

	VmaVulkanFunctions functions {};
	functions.vkGetInstanceProcAddr = VULKAN_HPP_DEFAULT_DISPATCHER.vkGetInstanceProcAddr;
	functions.vkGetDeviceProcAddr   = VULKAN_HPP_DEFAULT_DISPATCHER.vkGetDeviceProcAddr;

	VmaAllocatorCreateInfo info {};
	info.instance         = instance;
	info.physicalDevice   = physical_device;
	info.device           = device;
	info.pVulkanFunctions = &functions;
	info.vulkanApiVersion = VULKAN_TARGET_API_VERSION;
	info.flags = VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT;
	if (memory_budget_ext_enabled) {
		info.flags |= VMA_ALLOCATOR_CREATE_EXT_MEMORY_BUDGET_BIT;
	}

	const auto result = static_cast<vk::Result>(vmaCreateAllocator(&info, &allocator));
	if (result != vk::Result::eSuccess) {
		LOGF("vmaCreateAllocator failed: %s\n", vk::to_string(result).c_str());
		return false;
	}
	return true;
}

void GraphicContext::DestroyAllocator() {
	if (allocator == nullptr) {
		return;
	}
	vmaDestroyAllocator(allocator);
	allocator = nullptr;
}

void GraphicContext::LogMemoryBudget() const {
	if (allocator == nullptr || physical_device == nullptr) {
		return;
	}

	const auto& properties = GetPhysicalDeviceMemoryProperties();
	VmaBudget   budgets[VK_MAX_MEMORY_HEAPS] {};
	vmaGetHeapBudgets(allocator, budgets);
	for (uint32_t i = 0; i < properties.memoryHeapCount; i++) {
		LOGF("VMA heap %u: usage=%" PRIu64 ", budget=%" PRIu64 ", allocation=%" PRIu64
		     ", blocks=%" PRIu64 "\n",
		     i, static_cast<uint64_t>(budgets[i].usage), static_cast<uint64_t>(budgets[i].budget),
		     static_cast<uint64_t>(budgets[i].statistics.allocationBytes),
		     static_cast<uint64_t>(budgets[i].statistics.blockBytes));
	}
}

uint64_t GraphicContext::GetDeviceMemoryUsage() const {
	if (!CanReportMemoryUsage() || allocator == nullptr) {
		return 0;
	}
	VmaBudget budgets[VK_MAX_MEMORY_HEAPS] {};
	vmaGetHeapBudgets(allocator, budgets);
	const bool discrete =
	    physical_device_properties.deviceType == vk::PhysicalDeviceType::eDiscreteGpu;
	uint64_t usage = 0;
	for (uint32_t heap = 0; heap < physical_device_memory_properties.memoryHeapCount; heap++) {
		const bool device_local =
		    static_cast<bool>(physical_device_memory_properties.memoryHeaps[heap].flags &
		                      vk::MemoryHeapFlagBits::eDeviceLocal);
		if (!discrete || device_local) {
			usage += budgets[heap].usage;
		}
	}
	return usage;
}

uint64_t GraphicContext::GetTotalMemoryBudget() const {
	if (allocator == nullptr) {
		return 0;
	}
	VmaBudget budgets[VK_MAX_MEMORY_HEAPS] {};
	vmaGetHeapBudgets(allocator, budgets);
	const bool discrete =
	    physical_device_properties.deviceType == vk::PhysicalDeviceType::eDiscreteGpu;
	uint64_t budget = 0;
	uint64_t local  = 0;
	uint64_t usage  = 0;
	for (uint32_t heap = 0; heap < physical_device_memory_properties.memoryHeapCount; heap++) {
		const auto& properties = physical_device_memory_properties.memoryHeaps[heap];
		const bool  device_local =
		    static_cast<bool>(properties.flags & vk::MemoryHeapFlagBits::eDeviceLocal);
		if (device_local) {
			local += properties.size;
		}
		if (!discrete || device_local) {
			budget += CanReportMemoryUsage() ? budgets[heap].budget : properties.size;
			usage += CanReportMemoryUsage() ? budgets[heap].usage : 0;
		}
	}
	if (discrete) {
		return budget - std::min<uint64_t>(budget / 8, 1024ull * 1024 * 1024);
	}
	constexpr uint64_t system_reserve = 8ull * 1024 * 1024 * 1024;
	const auto         available      = budget > usage ? budget - usage : uint64_t {0};
	return std::max(local, available > system_reserve ? available - system_reserve : uint64_t {0});
}

bool GraphicContext::CreateImage(const vk::ImageCreateInfo& image_info, VulkanImage& image) {
	KYTY_PROFILER_FUNCTION();
	EXIT_IF(allocator == nullptr || image.image != nullptr || image.memory.allocation != nullptr);

	auto&                   memory = image.memory;
	VmaAllocationCreateInfo alloc_info {};
	alloc_info.requiredFlags = static_cast<vk::MemoryPropertyFlags::MaskType>(memory.property);
	alloc_info.preferredFlags =
	    static_cast<vk::MemoryPropertyFlags::MaskType>(memory.preferred_property);

	VmaAllocationInfo allocation_result {};
	vk::Image::CType native_image = VK_NULL_HANDLE;
	const auto        result       = static_cast<vk::Result>(
	    vmaCreateImage(allocator, static_cast<const vk::ImageCreateInfo::NativeType*>(image_info),
	                   &alloc_info, &native_image, &memory.allocation, &allocation_result));
	image.image = native_image;
	if (result != vk::Result::eSuccess) {
		LogMemoryBudget();
		return false;
	}

	device.getImageMemoryRequirements(image.image, &memory.requirements);
	memory.type = allocation_result.memoryType;
	VulkanTrackAllocation(memory);
	return true;
}

void GraphicContext::DeleteImage(VulkanImage& image) {
	KYTY_PROFILER_FUNCTION();
	EXIT_IF(allocator == nullptr || image.image == nullptr || image.memory.allocation == nullptr);

	auto& memory = image.memory;
	VulkanUntrackAllocation(memory);
	vmaDestroyImage(allocator, image.image, memory.allocation);
	image.image            = nullptr;
	memory.allocation      = nullptr;
}

} // namespace Libs::Graphics