adding async chunk generation

This commit is contained in:
2025-05-17 23:10:31 +02:00
parent e6982f99e5
commit be29d9a3be
19 changed files with 167 additions and 75 deletions

View File

@@ -45,7 +45,7 @@ _GRAY := $(shell $(TPUT) setaf 8)
_PURPLE := $(shell $(TPUT) setaf 5)
ifeq ($(DEBUG), true)
CXXFLAGS += -g -D DEBUG -D IMGUI_IMPL_VULKAN_NO_PROTOTYPES -I ThirdParty/imgui
CXXFLAGS += -g3 -D DEBUG -D IMGUI_IMPL_VULKAN_NO_PROTOTYPES -I ThirdParty/imgui
SRCS += $(wildcard $(addsuffix /*.cpp, ./Runtime/Sources/Debug))
SRCS += $(wildcard $(addsuffix /*.cpp, ./ThirdParty/imgui))
SRCS += $(wildcard $(addsuffix /*.cpp, ./ThirdParty/imgui/backends))

View File

@@ -27,7 +27,7 @@ namespace Scop
std::memcpy(data.GetData() + vertices.size() * sizeof(Vertex), indices.data(), indices.size() * sizeof(std::uint32_t));
buffer.Init(vertices.size() * sizeof(Vertex), indices.size() * sizeof(std::uint32_t), 0, std::move(data));
this->index_size = index_size == 0 ? indices.size() : index_size;
triangle_count = index_size / 3;
triangle_count = this->index_size / 3;
}
inline void SetData(const std::vector<Vertex>& vertices, const std::vector<std::uint32_t>& indices, std::size_t index_size = 0)
@@ -41,7 +41,7 @@ namespace Scop
buffer.SetVertexData(std::move(index_data));
this->index_size = index_size == 0 ? indices.size() : index_size;
triangle_count = index_size / 3;
triangle_count = this->index_size / 3;
}
};

View File

@@ -34,6 +34,8 @@ namespace Scop
~Model() = default;
private:
inline static std::shared_ptr<Material> s_default_material = nullptr;
Vec3f m_center = { 0.0f, 0.0f, 0.0f };
std::vector<std::shared_ptr<Material>> m_materials;
std::shared_ptr<Mesh> p_mesh;

View File

@@ -15,7 +15,7 @@ namespace Scop
public:
GPUBuffer() = default;
void Init(BufferType type, VkDeviceSize size, VkBufferUsageFlags usage, CPUBuffer data, std::string_view name = {});
void Init(BufferType type, VkDeviceSize size, VkBufferUsageFlags usage, CPUBuffer data, std::string_view name = {}, bool dedicated_alloc = false);
void Destroy() noexcept;
bool CopyFrom(const GPUBuffer& buffer, std::size_t src_offset = 0, std::size_t dst_offset = 0) noexcept;
@@ -43,7 +43,7 @@ namespace Scop
MemoryBlock m_memory = NULL_MEMORY_BLOCK;
private:
void CreateBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, std::string_view name);
void CreateBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, std::string_view name, bool dedicated_alloc);
private:
inline static std::size_t s_buffer_count = 0;
@@ -52,6 +52,7 @@ namespace Scop
VkBufferUsageFlags m_usage = 0;
VkMemoryPropertyFlags m_flags = 0;
bool m_is_dedicated_alloc = false;
};
class VertexBuffer : public GPUBuffer
@@ -77,7 +78,7 @@ namespace Scop
{
m_vertex_offset = 0;
m_index_offset = vertex_size;
GPUBuffer::Init(BufferType::LowDynamic, vertex_size + index_size, VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT | additional_flags, std::move(data), std::move(name));
GPUBuffer::Init(BufferType::LowDynamic, vertex_size + index_size, VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT | additional_flags, std::move(data), std::move(name), true);
}
void SetVertexData(CPUBuffer data);
void SetIndexData(CPUBuffer data);

View File

@@ -12,12 +12,13 @@ namespace Scop
class MemoryChunk
{
public:
MemoryChunk(VkDevice device, VkPhysicalDevice physical, VkDeviceSize size, std::int32_t memory_type_index);
MemoryChunk(VkDevice device, VkPhysicalDevice physical, VkDeviceSize size, std::int32_t memory_type_index, bool is_dedicated);
[[nodiscard]] std::optional<MemoryBlock> Allocate(VkDeviceSize size, VkDeviceSize alignment);
void Deallocate(const MemoryBlock& block);
[[nodiscard]] inline bool Has(const MemoryBlock& block) const noexcept { return block.memory == m_memory; }
[[nodiscard]] inline std::int32_t GetMemoryTypeIndex() const noexcept { return m_memory_type_index; }
[[nodiscard]] inline bool IsDedicated() const noexcept { return m_is_dedicated; }
~MemoryChunk();
@@ -29,6 +30,7 @@ namespace Scop
void* p_map = nullptr;
VkDeviceSize m_size = 0;
std::int32_t m_memory_type_index;
bool m_is_dedicated;
};
}

View File

@@ -1,6 +1,7 @@
#ifndef __SCOP_VULKAN_MEMORY_DEVICE_ALLOCATOR__
#define __SCOP_VULKAN_MEMORY_DEVICE_ALLOCATOR__
#include <mutex>
#include <vector>
#include <memory>
#include <cstdint>
@@ -29,6 +30,8 @@ namespace Scop
VkDevice m_device = VK_NULL_HANDLE;
VkPhysicalDevice m_physical = VK_NULL_HANDLE;
std::size_t m_allocations_count = 0;
std::mutex m_alloc_mutex;
std::mutex m_dealloc_mutex;
};
}

View File

@@ -6,9 +6,6 @@
#include <cstdint>
#include <optional>
#ifdef DEBUG
#define KVF_ENABLE_VALIDATION_LAYERS
#endif
#include <kvf.h>
#include <Renderer/Memory/DeviceAllocator.h>
@@ -36,17 +33,21 @@ namespace Scop
[[nodiscard]] inline VkInstance& GetInstanceRef() noexcept { return m_instance; }
[[nodiscard]] inline VkDevice GetDevice() const noexcept { return m_device; }
[[nodiscard]] inline VkPhysicalDevice GetPhysicalDevice() const noexcept { return m_physical_device; }
[[nodiscard]] inline DeviceAllocator& GetAllocator() noexcept { return m_allocator; }
[[nodiscard]] inline DeviceAllocator& GetAllocator() noexcept { return m_allocator; }
[[nodiscard]] inline bool StackSubmits() const noexcept { return m_stack_submits; }
[[nodiscard]] inline std::shared_ptr<class Shader> GetDefaultVertexShader() const { return m_internal_shaders[DEFAULT_VERTEX_SHADER_ID]; }
[[nodiscard]] inline std::shared_ptr<class Shader> GetBasicFragmentShader() const { return m_internal_shaders[BASIC_FRAGMENT_SHADER_ID]; }
[[nodiscard]] inline std::shared_ptr<class Shader> GetDefaultFragmentShader() const { return m_internal_shaders[DEFAULT_FRAGMENT_SHADER_ID]; }
inline void WaitDeviceIdle() const noexcept { vkDeviceWaitIdle(m_device); }
inline void WaitQueueIdle(KvfQueueType queue) const noexcept { vkQueueWaitIdle(kvfGetDeviceQueue(m_device, queue)); }
inline static bool IsInit() noexcept { return s_instance != nullptr; }
inline static RenderCore& Get() noexcept { return *s_instance; }
inline void ShouldStackSubmits(bool should) noexcept { m_stack_submits = should; }
#define SCOP_VULKAN_GLOBAL_FUNCTION(fn) PFN_##fn fn = nullptr;
#define SCOP_VULKAN_INSTANCE_FUNCTION(fn) PFN_##fn fn = nullptr;
#define SCOP_VULKAN_DEVICE_FUNCTION(fn) PFN_##fn fn = nullptr;
@@ -70,6 +71,7 @@ namespace Scop
VkInstance m_instance = VK_NULL_HANDLE;
VkDevice m_device = VK_NULL_HANDLE;
VkPhysicalDevice m_physical_device = VK_NULL_HANDLE;
bool m_stack_submits = false;
};
}

View File

@@ -141,7 +141,7 @@ namespace Scop
ImGui::Text("Swapchain images count %ld", p_renderer->GetSwapchain().GetSwapchainImages().size());
ImGui::Text("Drawcalls %ld", p_renderer->GetDrawCallsCounterRef());
ImGui::Text("Polygon drawn %ld", p_renderer->GetPolygonDrawnCounterRef());
ImGui::Text("Allocations count %ld", RenderCore::Get().GetAllocator().GetAllocationsCount());
ImGui::Text("Allocations count %ld / %u", RenderCore::Get().GetAllocator().GetAllocationsCount(), props.limits.maxMemoryAllocationCount);
ImGui::Text("Buffer count %ld", GPUBuffer::GetBufferCount());
ImGui::Text("Image count %ld", Image::GetImageCount());
ImGui::Text("Window dimensions: %ux%u", p_renderer->GetWindow()->GetWidth(), p_renderer->GetWindow()->GetHeight());

View File

@@ -12,13 +12,17 @@ namespace Scop
if(p_mesh)
m_materials.resize(p_mesh->GetSubMeshCount() + 1);
CPUBuffer default_pixels{ kvfFormatSize(VK_FORMAT_R8G8B8A8_SRGB) };
default_pixels.GetDataAs<std::uint32_t>()[0] = 0xFFFFFFFF;
std::shared_ptr<Texture> default_texture = std::make_shared<Texture>(std::move(default_pixels), 1, 1, VK_FORMAT_R8G8B8A8_SRGB);
if(!s_default_material)
{
CPUBuffer default_pixels{ kvfFormatSize(VK_FORMAT_R8G8B8A8_SRGB) };
default_pixels.GetDataAs<std::uint32_t>()[0] = 0xFFFFFFFF;
std::shared_ptr<Texture> default_texture = std::make_shared<Texture>(std::move(default_pixels), 1, 1, VK_FORMAT_R8G8B8A8_SRGB);
MaterialTextures textures;
textures.albedo = default_texture;
m_materials.back() = std::make_shared<Material>(textures);
MaterialTextures textures;
textures.albedo = default_texture;
s_default_material = std::make_shared<Material>(textures);
}
m_materials.back() = s_default_material;
}
void Model::Draw(VkCommandBuffer cmd, const DescriptorSet& matrices_set, const GraphicPipeline& pipeline, DescriptorSet& set, std::size_t& drawcalls, std::size_t& polygondrawn, std::size_t frame_index) const

View File

@@ -4,7 +4,7 @@
namespace Scop
{
void GPUBuffer::Init(BufferType type, VkDeviceSize size, VkBufferUsageFlags usage, CPUBuffer data, std::string_view name)
void GPUBuffer::Init(BufferType type, VkDeviceSize size, VkBufferUsageFlags usage, CPUBuffer data, std::string_view name, bool dedicated_alloc)
{
if(type == BufferType::Constant)
{
@@ -30,7 +30,7 @@ namespace Scop
if(type == BufferType::Staging && data.Empty())
Warning("Vulkan: trying to create staging buffer without data (wtf?)");
CreateBuffer(size, m_usage, m_flags, std::move(name));
CreateBuffer(size, m_usage, m_flags, std::move(name), dedicated_alloc);
if(!data.Empty())
{
@@ -41,7 +41,7 @@ namespace Scop
PushToGPU();
}
void GPUBuffer::CreateBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, std::string_view name)
void GPUBuffer::CreateBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, std::string_view name, bool dedicated_alloc)
{
auto device = RenderCore::Get().GetDevice();
m_buffer = kvfCreateBuffer(device, usage, size);
@@ -49,7 +49,7 @@ namespace Scop
VkMemoryRequirements mem_requirements;
RenderCore::Get().vkGetBufferMemoryRequirements(device, m_buffer, &mem_requirements);
m_memory = RenderCore::Get().GetAllocator().Allocate(size, mem_requirements.alignment, *FindMemoryType(mem_requirements.memoryTypeBits, properties));
m_memory = RenderCore::Get().GetAllocator().Allocate(size, mem_requirements.alignment, *FindMemoryType(mem_requirements.memoryTypeBits, properties), dedicated_alloc);
//m_memory = RenderCore::Get().GetAllocator().Allocate(mem_requirements.size, mem_requirements.alignment, *FindMemoryType(mem_requirements.memoryTypeBits, properties));
RenderCore::Get().vkBindBufferMemory(device, m_buffer, m_memory.memory, m_memory.offset);
@@ -73,12 +73,10 @@ namespace Scop
name_info.objectHandle = reinterpret_cast<std::uint64_t>(m_buffer);
name_info.pObjectName = m_name.c_str();
RenderCore::Get().vkSetDebugUtilsObjectNameEXT(RenderCore::Get().GetDevice(), &name_info);
Message("Vulkan: % buffer created", m_name);
#else
Message("Vulkan: buffer created");
#endif
m_is_dedicated_alloc = dedicated_alloc;
s_buffer_count++;
}
@@ -99,10 +97,16 @@ namespace Scop
kvfBeginCommandBuffer(cmd, VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT);
kvfCopyBufferToBuffer(cmd, m_buffer, buffer.Get(), buffer.GetSize(), src_offset, dst_offset);
kvfEndCommandBuffer(cmd);
VkFence fence = kvfCreateFence(RenderCore::Get().GetDevice());
kvfSubmitSingleTimeCommandBuffer(RenderCore::Get().GetDevice(), cmd, KVF_GRAPHICS_QUEUE, fence);
kvfWaitForFence(RenderCore::Get().GetDevice(), fence);
kvfDestroyFence(RenderCore::Get().GetDevice(), fence);
if(!RenderCore::Get().StackSubmits())
{
VkFence fence = kvfCreateFence(RenderCore::Get().GetDevice());
kvfSubmitSingleTimeCommandBuffer(RenderCore::Get().GetDevice(), cmd, KVF_GRAPHICS_QUEUE, fence);
kvfWaitForFence(RenderCore::Get().GetDevice(), fence);
kvfDestroyFence(RenderCore::Get().GetDevice(), fence);
kvfDestroyCommandBuffer(RenderCore::Get().GetDevice(), cmd);
}
else
kvfSubmitSingleTimeCommandBuffer(RenderCore::Get().GetDevice(), cmd, KVF_GRAPHICS_QUEUE, VK_NULL_HANDLE);
return true;
}
@@ -111,12 +115,11 @@ namespace Scop
GPUBuffer new_buffer;
new_buffer.m_usage = (this->m_usage & 0xFFFFFFFC) | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
new_buffer.m_flags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
new_buffer.CreateBuffer(m_memory.size, new_buffer.m_usage, new_buffer.m_flags, m_name);
new_buffer.CreateBuffer(m_memory.size, new_buffer.m_usage, new_buffer.m_flags, m_name, m_is_dedicated_alloc);
if(new_buffer.CopyFrom(*this))
Swap(new_buffer);
new_buffer.Destroy();
Message("Vulkan: pushed buffer to GPU memory");
}
void GPUBuffer::Destroy() noexcept
@@ -128,7 +131,6 @@ namespace Scop
RenderCore::Get().GetAllocator().Deallocate(m_memory);
m_buffer = VK_NULL_HANDLE;
m_memory = NULL_MEMORY_BLOCK;
Message("Vulkan: destroyed buffer");
s_buffer_count--;
}

View File

@@ -57,9 +57,6 @@ namespace Scop
name_info.objectHandle = reinterpret_cast<std::uint64_t>(m_image);
name_info.pObjectName = name.data();
RenderCore::Get().vkSetDebugUtilsObjectNameEXT(RenderCore::Get().GetDevice(), &name_info);
Message("Vulkan: % image created", name);
#else
Message("Vulkan: image created");
#endif
s_image_count++;
@@ -151,7 +148,6 @@ namespace Scop
m_memory = NULL_MEMORY_BLOCK;
kvfDestroyImage(RenderCore::Get().GetDevice(), m_image);
}
Message("Vulkan: image destroyed");
m_image = VK_NULL_HANDLE;
m_memory = NULL_MEMORY_BLOCK;
m_image = VK_NULL_HANDLE;

View File

@@ -6,8 +6,8 @@
namespace Scop
{
MemoryChunk::MemoryChunk(VkDevice device, VkPhysicalDevice physical, VkDeviceSize size, std::int32_t memory_type_index)
: m_device(device), m_physical(physical), m_size(size), m_memory_type_index(memory_type_index)
MemoryChunk::MemoryChunk(VkDevice device, VkPhysicalDevice physical, VkDeviceSize size, std::int32_t memory_type_index, bool is_dedicated)
: m_device(device), m_physical(physical), m_size(size), m_memory_type_index(memory_type_index), m_is_dedicated(is_dedicated)
{
Verify(device != VK_NULL_HANDLE, "Memory Chunk : invalid device");
VkMemoryAllocateInfo alloc_info{};

View File

@@ -12,6 +12,7 @@ namespace Scop
{
Verify(m_device != VK_NULL_HANDLE, "invalid device");
Verify(m_physical != VK_NULL_HANDLE, "invalid physical device");
const std::lock_guard<std::mutex> guard(m_alloc_mutex);
if(!dedicated_chunk)
{
for(auto& chunk : m_chunks)
@@ -24,7 +25,7 @@ namespace Scop
}
}
}
m_chunks.emplace_back(std::make_unique<MemoryChunk>(m_device, m_physical, (CHUNK_SIZE < size + alignment ? size + alignment : CHUNK_SIZE), memory_type_index));
m_chunks.emplace_back(std::make_unique<MemoryChunk>(m_device, m_physical, (CHUNK_SIZE < size + alignment ? size + alignment : CHUNK_SIZE), memory_type_index, dedicated_chunk));
std::optional<MemoryBlock> block = m_chunks.back()->Allocate(size, alignment);
m_allocations_count++;
if(block.has_value())
@@ -37,11 +38,17 @@ namespace Scop
{
Verify(m_device != VK_NULL_HANDLE, "invalid device");
Verify(m_physical != VK_NULL_HANDLE, "invalid physical device");
for(auto& chunk : m_chunks)
const std::lock_guard<std::mutex> guard(m_dealloc_mutex);
for(auto it = m_chunks.begin(); it != m_chunks.end(); ++it)
{
if(chunk->Has(block))
if((*it)->Has(block))
{
chunk->Deallocate(block);
(*it)->Deallocate(block);
if((*it)->IsDedicated())
{
m_chunks.erase(it);
m_allocations_count--;
}
return;
}
}