hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

early-access version 1291

Browse Source
This commit is contained in:
pineappleEA
2021-01-06 04:42:38 +01:00
parent 92380ca74d
commit 210a322d74
24 changed files with 621 additions and 199 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/video_core/renderer_vulkan/renderer_vulkan.cpp
View File

@@ -23,7 +23,6 @@
#include "video_core/renderer_vulkan/renderer_vulkan.h"
#include "video_core/renderer_vulkan/vk_blit_screen.h"
#include "video_core/renderer_vulkan/vk_master_semaphore.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_state_tracker.h"
@@ -32,6 +31,7 @@
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_instance.h"
#include "video_core/vulkan_common/vulkan_library.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_surface.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@@ -137,7 +137,7 @@ bool RendererVulkan::Init() try {
InitializeDevice();
Report();
memory_manager = std::make_unique<VKMemoryManager>(*device);
memory_allocator = std::make_unique<MemoryAllocator>(*device);
state_tracker = std::make_unique<StateTracker>(gpu);
@@ -149,11 +149,11 @@ bool RendererVulkan::Init() try {
rasterizer = std::make_unique<RasterizerVulkan>(render_window, gpu, gpu.MemoryManager(),
cpu_memory, screen_info, *device,
*memory_manager, *state_tracker, *scheduler);
*memory_allocator, *state_tracker, *scheduler);
blit_screen =
std::make_unique<VKBlitScreen>(cpu_memory, render_window, *rasterizer, *device,
*memory_manager, *swapchain, *scheduler, screen_info);
*memory_allocator, *swapchain, *scheduler, screen_info);
return true;
} catch (const vk::Exception& exception) {
@@ -172,7 +172,7 @@ void RendererVulkan::ShutDown() {
blit_screen.reset();
scheduler.reset();
swapchain.reset();
memory_manager.reset();
memory_allocator.reset();
device.reset();
}

4
src/video_core/renderer_vulkan/renderer_vulkan.h
View File

@@ -29,8 +29,8 @@ namespace Vulkan {
class Device;
class StateTracker;
class MemoryAllocator;
class VKBlitScreen;
class VKMemoryManager;
class VKSwapchain;
class VKScheduler;
@@ -75,7 +75,7 @@ private:
vk::DebugUtilsMessenger debug_callback;
std::unique_ptr<Device> device;
std::unique_ptr<VKMemoryManager> memory_manager;
std::unique_ptr<MemoryAllocator> memory_allocator;
std::unique_ptr<StateTracker> state_tracker;
std::unique_ptr<VKScheduler> scheduler;
std::unique_ptr<VKSwapchain> swapchain;

22
src/video_core/renderer_vulkan/vk_blit_screen.cpp
View File

@@ -22,13 +22,13 @@
#include "video_core/renderer_vulkan/renderer_vulkan.h"
#include "video_core/renderer_vulkan/vk_blit_screen.h"
#include "video_core/renderer_vulkan/vk_master_semaphore.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_util.h"
#include "video_core/renderer_vulkan/vk_swapchain.h"
#include "video_core/surface.h"
#include "video_core/textures/decoders.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
@@ -115,10 +115,10 @@ struct VKBlitScreen::BufferData {
VKBlitScreen::VKBlitScreen(Core::Memory::Memory& cpu_memory_,
Core::Frontend::EmuWindow& render_window_,
VideoCore::RasterizerInterface& rasterizer_, const Device& device_,
VKMemoryManager& memory_manager_, VKSwapchain& swapchain_,
MemoryAllocator& memory_allocator_, VKSwapchain& swapchain_,
VKScheduler& scheduler_, const VKScreenInfo& screen_info_)
: cpu_memory{cpu_memory_}, render_window{render_window_}, rasterizer{rasterizer_},
device{device_}, memory_manager{memory_manager_}, swapchain{swapchain_},
device{device_}, memory_allocator{memory_allocator_}, swapchain{swapchain_},
scheduler{scheduler_}, image_count{swapchain.GetImageCount()}, screen_info{screen_info_} {
resource_ticks.resize(image_count);
@@ -150,8 +150,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
SetUniformData(data, framebuffer);
SetVertexData(data, framebuffer);
auto map = buffer_commit->Map();
std::memcpy(map.Address(), &data, sizeof(data));
const std::span<u8> map = buffer_commit.Map();
std::memcpy(map.data(), &data, sizeof(data));
if (!use_accelerated) {
const u64 image_offset = GetRawImageOffset(framebuffer, image_index);
@@ -165,8 +165,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
constexpr u32 block_height_log2 = 4;
const u32 bytes_per_pixel = GetBytesPerPixel(framebuffer);
Tegra::Texture::UnswizzleTexture(
std::span(map.Address() + image_offset, size_bytes), std::span(host_ptr, size_bytes),
bytes_per_pixel, framebuffer.width, framebuffer.height, 1, block_height_log2, 0);
map.subspan(image_offset, size_bytes), std::span(host_ptr, size_bytes), bytes_per_pixel,
framebuffer.width, framebuffer.height, 1, block_height_log2, 0);
const VkBufferImageCopy copy{
.bufferOffset = image_offset,
@@ -224,8 +224,6 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier);
});
}
map.Release();
scheduler.Record([renderpass = *renderpass, framebuffer = *framebuffers[image_index],
descriptor_set = descriptor_sets[image_index], buffer = *buffer,
size = swapchain.GetSize(), pipeline = *pipeline,
@@ -642,7 +640,7 @@ void VKBlitScreen::ReleaseRawImages() {
raw_images.clear();
raw_buffer_commits.clear();
buffer.reset();
buffer_commit.reset();
buffer_commit = MemoryCommit{};
}
void VKBlitScreen::CreateStagingBuffer(const Tegra::FramebufferConfig& framebuffer) {
@@ -659,7 +657,7 @@ void VKBlitScreen::CreateStagingBuffer(const Tegra::FramebufferConfig& framebuff
};
buffer = device.GetLogical().CreateBuffer(ci);
buffer_commit = memory_manager.Commit(buffer, true);
buffer_commit = memory_allocator.Commit(buffer, MemoryUsage::Upload);
}
void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) {
@@ -690,7 +688,7 @@ void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer)
.pQueueFamilyIndices = nullptr,
.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
});
raw_buffer_commits[i] = memory_manager.Commit(raw_images[i], false);
raw_buffer_commits[i] = memory_allocator.Commit(raw_images[i], MemoryUsage::DeviceLocal);
raw_image_views[i] = device.GetLogical().CreateImageView(VkImageViewCreateInfo{
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.pNext = nullptr,

10
src/video_core/renderer_vulkan/vk_blit_screen.h
View File

@@ -6,7 +6,7 @@
#include <memory>
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Core {
@@ -43,7 +43,7 @@ public:
explicit VKBlitScreen(Core::Memory::Memory& cpu_memory,
Core::Frontend::EmuWindow& render_window,
VideoCore::RasterizerInterface& rasterizer, const Device& device,
VKMemoryManager& memory_manager, VKSwapchain& swapchain,
MemoryAllocator& memory_allocator, VKSwapchain& swapchain,
VKScheduler& scheduler, const VKScreenInfo& screen_info);
~VKBlitScreen();
@@ -86,7 +86,7 @@ private:
Core::Frontend::EmuWindow& render_window;
VideoCore::RasterizerInterface& rasterizer;
const Device& device;
VKMemoryManager& memory_manager;
MemoryAllocator& memory_allocator;
VKSwapchain& swapchain;
VKScheduler& scheduler;
const std::size_t image_count;
@@ -104,14 +104,14 @@ private:
vk::Sampler sampler;
vk::Buffer buffer;
VKMemoryCommit buffer_commit;
MemoryCommit buffer_commit;
std::vector<u64> resource_ticks;
std::vector<vk::Semaphore> semaphores;
std::vector<vk::Image> raw_images;
std::vector<vk::ImageView> raw_image_views;
std::vector<VKMemoryCommit> raw_buffer_commits;
std::vector<MemoryCommit> raw_buffer_commits;
u32 raw_width = 0;
u32 raw_height = 0;
};

40
src/video_core/renderer_vulkan/vk_buffer_cache.cpp
View File

@@ -40,11 +40,11 @@ std::unique_ptr<VKStreamBuffer> CreateStreamBuffer(const Device& device, VKSched
} // Anonymous namespace
Buffer::Buffer(const Device& device_, VKMemoryManager& memory_manager, VKScheduler& scheduler_,
VKStagingBufferPool& staging_pool_, VAddr cpu_addr_, std::size_t size_)
Buffer::Buffer(const Device& device_, MemoryAllocator& memory_allocator, VKScheduler& scheduler_,
StagingBufferPool& staging_pool_, VAddr cpu_addr_, std::size_t size_)
: BufferBlock{cpu_addr_, size_}, device{device_}, scheduler{scheduler_}, staging_pool{
staging_pool_} {
const VkBufferCreateInfo ci{
buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
@@ -53,22 +53,20 @@ Buffer::Buffer(const Device& device_, VKMemoryManager& memory_manager, VKSchedul
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
};
buffer.handle = device.GetLogical().CreateBuffer(ci);
buffer.commit = memory_manager.Commit(buffer.handle, false);
});
commit = memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
}
Buffer::~Buffer() = default;
void Buffer::Upload(std::size_t offset, std::size_t data_size, const u8* data) {
const auto& staging = staging_pool.GetUnusedBuffer(data_size, true);
std::memcpy(staging.commit->Map(data_size), data, data_size);
const auto& staging = staging_pool.Request(data_size, MemoryUsage::Upload);
std::memcpy(staging.mapped_span.data(), data, data_size);
scheduler.RequestOutsideRenderPassOperationContext();
const VkBuffer handle = Handle();
scheduler.Record([staging = *staging.handle, handle, offset, data_size,
scheduler.Record([staging = staging.buffer, handle, offset, data_size,
&device = device](vk::CommandBuffer cmdbuf) {
const VkBufferMemoryBarrier read_barrier{
.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
@@ -104,12 +102,12 @@ void Buffer::Upload(std::size_t offset, std::size_t data_size, const u8* data) {
}
void Buffer::Download(std::size_t offset, std::size_t data_size, u8* data) {
const auto& staging = staging_pool.GetUnusedBuffer(data_size, true);
auto staging = staging_pool.Request(data_size, MemoryUsage::Download);
scheduler.RequestOutsideRenderPassOperationContext();
const VkBuffer handle = Handle();
scheduler.Record(
[staging = *staging.handle, handle, offset, data_size](vk::CommandBuffer cmdbuf) {
[staging = staging.buffer, handle, offset, data_size](vk::CommandBuffer cmdbuf) {
const VkBufferMemoryBarrier barrier{
.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
.pNext = nullptr,
@@ -130,7 +128,7 @@ void Buffer::Download(std::size_t offset, std::size_t data_size, u8* data) {
});
scheduler.Finish();
std::memcpy(data, staging.commit->Map(data_size), data_size);
std::memcpy(data, staging.mapped_span.data(), data_size);
}
void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset,
@@ -168,29 +166,29 @@ void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst
VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer_,
Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
const Device& device_, VKMemoryManager& memory_manager_,
const Device& device_, MemoryAllocator& memory_allocator_,
VKScheduler& scheduler_, VKStreamBuffer& stream_buffer_,
VKStagingBufferPool& staging_pool_)
StagingBufferPool& staging_pool_)
: VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer_, gpu_memory_,
cpu_memory_, stream_buffer_},
device{device_}, memory_manager{memory_manager_}, scheduler{scheduler_}, staging_pool{
staging_pool_} {}
device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
staging_pool{staging_pool_} {}
VKBufferCache::~VKBufferCache() = default;
std::shared_ptr<Buffer> VKBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
return std::make_shared<Buffer>(device, memory_manager, scheduler, staging_pool, cpu_addr,
return std::make_shared<Buffer>(device, memory_allocator, scheduler, staging_pool, cpu_addr,
size);
}
VKBufferCache::BufferInfo VKBufferCache::GetEmptyBuffer(std::size_t size) {
size = std::max(size, std::size_t(4));
const auto& empty = staging_pool.GetUnusedBuffer(size, false);
const auto& empty = staging_pool.Request(size, MemoryUsage::DeviceLocal);
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([size, buffer = *empty.handle](vk::CommandBuffer cmdbuf) {
scheduler.Record([size, buffer = empty.buffer](vk::CommandBuffer cmdbuf) {
cmdbuf.FillBuffer(buffer, 0, size, 0);
});
return {*empty.handle, 0, 0};
return {empty.buffer, 0, 0};
}
} // namespace Vulkan

22
src/video_core/renderer_vulkan/vk_buffer_cache.h
View File

@@ -8,21 +8,20 @@
#include "common/common_types.h"
#include "video_core/buffer_cache/buffer_cache.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_stream_buffer.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
class Device;
class VKMemoryManager;
class VKScheduler;
class Buffer final : public VideoCommon::BufferBlock {
public:
explicit Buffer(const Device& device, VKMemoryManager& memory_manager, VKScheduler& scheduler,
VKStagingBufferPool& staging_pool, VAddr cpu_addr_, std::size_t size_);
explicit Buffer(const Device& device, MemoryAllocator& memory_allocator, VKScheduler& scheduler,
StagingBufferPool& staging_pool, VAddr cpu_addr_, std::size_t size_);
~Buffer();
void Upload(std::size_t offset, std::size_t data_size, const u8* data);
@@ -33,7 +32,7 @@ public:
std::size_t copy_size);
VkBuffer Handle() const {
return *buffer.handle;
return *buffer;
}
u64 Address() const {
@@ -43,18 +42,19 @@ public:
private:
const Device& device;
VKScheduler& scheduler;
VKStagingBufferPool& staging_pool;
StagingBufferPool& staging_pool;
VKBuffer buffer;
vk::Buffer buffer;
MemoryCommit commit;
};
class VKBufferCache final : public VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer> {
public:
explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer,
Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory,
const Device& device, VKMemoryManager& memory_manager,
const Device& device, MemoryAllocator& memory_allocator,
VKScheduler& scheduler, VKStreamBuffer& stream_buffer,
VKStagingBufferPool& staging_pool);
StagingBufferPool& staging_pool);
~VKBufferCache();
BufferInfo GetEmptyBuffer(std::size_t size) override;
@@ -64,9 +64,9 @@ protected:
private:
const Device& device;
VKMemoryManager& memory_manager;
MemoryAllocator& memory_allocator;
VKScheduler& scheduler;
VKStagingBufferPool& staging_pool;
StagingBufferPool& staging_pool;
};
} // namespace Vulkan

32
src/video_core/renderer_vulkan/vk_compute_pass.cpp
View File

@@ -164,7 +164,7 @@ VkDescriptorSet VKComputePass::CommitDescriptorSet(
QuadArrayPass::QuadArrayPass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKStagingBufferPool& staging_buffer_pool_,
StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
: VKComputePass(device_, descriptor_pool_, BuildQuadArrayPassDescriptorSetLayoutBinding(),
BuildQuadArrayPassDescriptorUpdateTemplateEntry(),
@@ -177,18 +177,18 @@ QuadArrayPass::~QuadArrayPass() = default;
std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) {
const u32 num_triangle_vertices = (num_vertices / 4) * 6;
const std::size_t staging_size = num_triangle_vertices * sizeof(u32);
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
const auto staging_ref = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size);
update_descriptor_queue.AddBuffer(staging_ref.buffer, 0, staging_size);
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
scheduler.RequestOutsideRenderPassOperationContext();
ASSERT(num_vertices % 4 == 0);
const u32 num_quads = num_vertices / 4;
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, num_quads,
first, set](vk::CommandBuffer cmdbuf) {
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging_ref.buffer,
num_quads, first, set](vk::CommandBuffer cmdbuf) {
constexpr u32 dispatch_size = 1024;
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
@@ -208,11 +208,11 @@ std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, {barrier}, {});
});
return {*buffer.handle, 0};
return {staging_ref.buffer, 0};
}
Uint8Pass::Uint8Pass(const Device& device, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool, VKStagingBufferPool& staging_buffer_pool_,
VKDescriptorPool& descriptor_pool, StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
: VKComputePass(device, descriptor_pool, BuildInputOutputDescriptorSetBindings(),
BuildInputOutputDescriptorUpdateTemplate(), {}, VULKAN_UINT8_COMP_SPV),
@@ -224,15 +224,15 @@ Uint8Pass::~Uint8Pass() = default;
std::pair<VkBuffer, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer,
u64 src_offset) {
const u32 staging_size = static_cast<u32>(num_vertices * sizeof(u16));
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
const auto staging_ref = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices);
update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size);
update_descriptor_queue.AddBuffer(staging_ref.buffer, 0, staging_size);
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, set,
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging_ref.buffer, set,
num_vertices](vk::CommandBuffer cmdbuf) {
constexpr u32 dispatch_size = 1024;
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
@@ -252,12 +252,12 @@ std::pair<VkBuffer, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buff
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {});
});
return {*buffer.handle, 0};
return {staging_ref.buffer, 0};
}
QuadIndexedPass::QuadIndexedPass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKStagingBufferPool& staging_buffer_pool_,
StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
: VKComputePass(device_, descriptor_pool_, BuildInputOutputDescriptorSetBindings(),
BuildInputOutputDescriptorUpdateTemplate(),
@@ -286,15 +286,15 @@ std::pair<VkBuffer, u64> QuadIndexedPass::Assemble(
const u32 num_tri_vertices = (num_vertices / 4) * 6;
const std::size_t staging_size = num_tri_vertices * sizeof(u32);
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
const auto staging_ref = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size);
update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size);
update_descriptor_queue.AddBuffer(staging_ref.buffer, 0, staging_size);
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, set,
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging_ref.buffer, set,
num_tri_vertices, base_vertex, index_shift](vk::CommandBuffer cmdbuf) {
static constexpr u32 dispatch_size = 1024;
const std::array push_constants = {base_vertex, index_shift};
@@ -317,7 +317,7 @@ std::pair<VkBuffer, u64> QuadIndexedPass::Assemble(
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {});
});
return {*buffer.handle, 0};
return {staging_ref.buffer, 0};
}
} // namespace Vulkan

15
src/video_core/renderer_vulkan/vk_compute_pass.h
View File

@@ -16,8 +16,8 @@
namespace Vulkan {
class Device;
class StagingBufferPool;
class VKScheduler;
class VKStagingBufferPool;
class VKUpdateDescriptorQueue;
class VKComputePass {
@@ -45,7 +45,7 @@ class QuadArrayPass final : public VKComputePass {
public:
explicit QuadArrayPass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKStagingBufferPool& staging_buffer_pool_,
StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_);
~QuadArrayPass();
@@ -53,15 +53,14 @@ public:
private:
VKScheduler& scheduler;
VKStagingBufferPool& staging_buffer_pool;
StagingBufferPool& staging_buffer_pool;
VKUpdateDescriptorQueue& update_descriptor_queue;
};
class Uint8Pass final : public VKComputePass {
public:
explicit Uint8Pass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKStagingBufferPool& staging_buffer_pool_,
VKDescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_);
~Uint8Pass();
@@ -69,7 +68,7 @@ public:
private:
VKScheduler& scheduler;
VKStagingBufferPool& staging_buffer_pool;
StagingBufferPool& staging_buffer_pool;
VKUpdateDescriptorQueue& update_descriptor_queue;
};
@@ -77,7 +76,7 @@ class QuadIndexedPass final : public VKComputePass {
public:
explicit QuadIndexedPass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKStagingBufferPool& staging_buffer_pool_,
StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_);
~QuadIndexedPass();
@@ -87,7 +86,7 @@ public:
private:
VKScheduler& scheduler;
VKStagingBufferPool& staging_buffer_pool;
StagingBufferPool& staging_buffer_pool;
VKUpdateDescriptorQueue& update_descriptor_queue;
};

14
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View File

@@ -409,24 +409,24 @@ void RasterizerVulkan::DrawParameters::Draw(vk::CommandBuffer cmdbuf) const {
RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
Tegra::MemoryManager& gpu_memory_,
Core::Memory::Memory& cpu_memory_, VKScreenInfo& screen_info_,
const Device& device_, VKMemoryManager& memory_manager_,
const Device& device_, MemoryAllocator& memory_allocator_,
StateTracker& state_tracker_, VKScheduler& scheduler_)
: RasterizerAccelerated{cpu_memory_}, gpu{gpu_},
gpu_memory{gpu_memory_}, maxwell3d{gpu.Maxwell3D()}, kepler_compute{gpu.KeplerCompute()},
screen_info{screen_info_}, device{device_}, memory_manager{memory_manager_},
screen_info{screen_info_}, device{device_}, memory_allocator{memory_allocator_},
state_tracker{state_tracker_}, scheduler{scheduler_}, stream_buffer(device, scheduler),
staging_pool(device, memory_manager, scheduler), descriptor_pool(device, scheduler),
staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler),
update_descriptor_queue(device, scheduler),
blit_image(device, scheduler, state_tracker, descriptor_pool),
quad_array_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
quad_indexed_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
texture_cache_runtime{device, scheduler, memory_manager, staging_pool, blit_image},
texture_cache_runtime{device, scheduler, memory_allocator, staging_pool, blit_image},
texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory),
pipeline_cache(*this, gpu, maxwell3d, kepler_compute, gpu_memory, device, scheduler,
descriptor_pool, update_descriptor_queue),
buffer_cache(*this, gpu_memory, cpu_memory_, device, memory_manager, scheduler, stream_buffer,
staging_pool),
buffer_cache(*this, gpu_memory, cpu_memory_, device, memory_allocator, scheduler,
stream_buffer, staging_pool),
query_cache{*this, maxwell3d, gpu_memory, device, scheduler},
fence_manager(*this, gpu, gpu_memory, texture_cache, buffer_cache, query_cache, device,
scheduler),
@@ -1463,7 +1463,7 @@ VkBuffer RasterizerVulkan::DefaultBuffer() {
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
});
default_buffer_commit = memory_manager.Commit(default_buffer, false);
default_buffer_commit = memory_allocator.Commit(default_buffer, MemoryUsage::DeviceLocal);
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([buffer = *default_buffer](vk::CommandBuffer cmdbuf) {

10
src/video_core/renderer_vulkan/vk_rasterizer.h
View File

@@ -21,7 +21,6 @@
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_fence_manager.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_query_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
@@ -30,6 +29,7 @@
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/shader/async_shaders.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Core {
@@ -56,7 +56,7 @@ public:
explicit RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
VKScreenInfo& screen_info_, const Device& device_,
VKMemoryManager& memory_manager_, StateTracker& state_tracker_,
MemoryAllocator& memory_allocator_, StateTracker& state_tracker_,
VKScheduler& scheduler_);
~RasterizerVulkan() override;
@@ -215,12 +215,12 @@ private:
VKScreenInfo& screen_info;
const Device& device;
VKMemoryManager& memory_manager;
MemoryAllocator& memory_allocator;
StateTracker& state_tracker;
VKScheduler& scheduler;
VKStreamBuffer stream_buffer;
VKStagingBufferPool staging_pool;
StagingBufferPool staging_pool;
VKDescriptorPool descriptor_pool;
VKUpdateDescriptorQueue update_descriptor_queue;
BlitImageHelper blit_image;
@@ -236,7 +236,7 @@ private:
VKFenceManager fence_manager;
vk::Buffer default_buffer;
VKMemoryCommit default_buffer_commit;
MemoryCommit default_buffer_commit;
vk::Event wfi_event;
VideoCommon::Shader::AsyncShaders async_shaders;

124
src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
View File

@@ -3,10 +3,12 @@
// Refer to the license.txt file included.
#include <algorithm>
#include <unordered_map>
#include <utility>
#include <vector>
#include <fmt/format.h>
#include "common/assert.h"
#include "common/bit_util.h"
#include "common/common_types.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
@@ -16,45 +18,51 @@
namespace Vulkan {
VKStagingBufferPool::StagingBuffer::StagingBuffer(std::unique_ptr<VKBuffer> buffer_)
: buffer{std::move(buffer_)} {}
StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& memory_allocator_,
VKScheduler& scheduler_)
: device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_} {}
VKStagingBufferPool::VKStagingBufferPool(const Device& device_, VKMemoryManager& memory_manager_,
VKScheduler& scheduler_)
: device{device_}, memory_manager{memory_manager_}, scheduler{scheduler_} {}
StagingBufferPool::~StagingBufferPool() = default;
VKStagingBufferPool::~VKStagingBufferPool() = default;
VKBuffer& VKStagingBufferPool::GetUnusedBuffer(std::size_t size, bool host_visible) {
if (const auto buffer = TryGetReservedBuffer(size, host_visible)) {
return *buffer;
StagingBufferRef StagingBufferPool::Request(size_t size, MemoryUsage usage) {
if (const std::optional<StagingBufferRef> ref = TryGetReservedBuffer(size, usage)) {
return *ref;
}
return CreateStagingBuffer(size, host_visible);
return CreateStagingBuffer(size, usage);
}
void VKStagingBufferPool::TickFrame() {
current_delete_level = (current_delete_level + 1) % NumLevels;
void StagingBufferPool::TickFrame() {
current_delete_level = (current_delete_level + 1) % NUM_LEVELS;
ReleaseCache(true);
ReleaseCache(false);
ReleaseCache(MemoryUsage::DeviceLocal);
ReleaseCache(MemoryUsage::Upload);
ReleaseCache(MemoryUsage::Download);
}
VKBuffer* VKStagingBufferPool::TryGetReservedBuffer(std::size_t size, bool host_visible) {
for (StagingBuffer& entry : GetCache(host_visible)[Common::Log2Ceil64(size)].entries) {
if (!scheduler.IsFree(entry.tick)) {
continue;
std::optional<StagingBufferRef> StagingBufferPool::TryGetReservedBuffer(size_t size,
MemoryUsage usage) {
StagingBuffers& cache_level = GetCache(usage)[Common::Log2Ceil64(size)];
const auto is_free = [this](const StagingBuffer& entry) {
return scheduler.IsFree(entry.tick);
};
auto& entries = cache_level.entries;
const auto hint_it = entries.begin() + cache_level.iterate_index;
auto it = std::find_if(entries.begin() + cache_level.iterate_index, entries.end(), is_free);
if (it == entries.end()) {
it = std::find_if(entries.begin(), hint_it, is_free);
if (it == hint_it) {
return std::nullopt;
}
entry.tick = scheduler.CurrentTick();
return &*entry.buffer;
}
return nullptr;
cache_level.iterate_index = std::distance(entries.begin(), it) + 1;
it->tick = scheduler.CurrentTick();
return it->Ref();
}
VKBuffer& VKStagingBufferPool::CreateStagingBuffer(std::size_t size, bool host_visible) {
StagingBufferRef StagingBufferPool::CreateStagingBuffer(size_t size, MemoryUsage usage) {
const u32 log2 = Common::Log2Ceil64(size);
auto buffer = std::make_unique<VKBuffer>();
buffer->handle = device.GetLogical().CreateBuffer({
vk::Buffer buffer = device.GetLogical().CreateBuffer({
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
@@ -66,49 +74,63 @@ VKBuffer& VKStagingBufferPool::CreateStagingBuffer(std::size_t size, bool host_v
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
});
buffer->commit = memory_manager.Commit(buffer->handle, host_visible);
if (device.HasDebuggingToolAttached()) {
++buffer_index;
buffer.SetObjectNameEXT(fmt::format("Staging Buffer {}", buffer_index).c_str());
}
MemoryCommit commit = memory_allocator.Commit(buffer, usage);
const std::span<u8> mapped_span = IsHostVisible(usage) ? commit.Map() : std::span<u8>{};
std::vector<StagingBuffer>& entries = GetCache(host_visible)[log2].entries;
StagingBuffer& entry = entries.emplace_back(std::move(buffer));
entry.tick = scheduler.CurrentTick();
return *entry.buffer;
StagingBuffer& entry = GetCache(usage)[log2].entries.emplace_back(StagingBuffer{
.buffer = std::move(buffer),
.commit = std::move(commit),
.mapped_span = mapped_span,
.tick = scheduler.CurrentTick(),
});
return entry.Ref();
}
VKStagingBufferPool::StagingBuffersCache& VKStagingBufferPool::GetCache(bool host_visible) {
return host_visible ? host_staging_buffers : device_staging_buffers;
}
void VKStagingBufferPool::ReleaseCache(bool host_visible) {
auto& cache = GetCache(host_visible);
const u64 size = ReleaseLevel(cache, current_delete_level);
if (size == 0) {
return;
StagingBufferPool::StagingBuffersCache& StagingBufferPool::GetCache(MemoryUsage usage) {
switch (usage) {
case MemoryUsage::DeviceLocal:
return device_local_cache;
case MemoryUsage::Upload:
return upload_cache;
case MemoryUsage::Download:
return download_cache;
default:
UNREACHABLE_MSG("Invalid memory usage={}", usage);
return upload_cache;
}
}
u64 VKStagingBufferPool::ReleaseLevel(StagingBuffersCache& cache, std::size_t log2) {
static constexpr std::size_t deletions_per_tick = 16;
void StagingBufferPool::ReleaseCache(MemoryUsage usage) {
ReleaseLevel(GetCache(usage), current_delete_level);
}
void StagingBufferPool::ReleaseLevel(StagingBuffersCache& cache, size_t log2) {
constexpr size_t deletions_per_tick = 16;
auto& staging = cache[log2];
auto& entries = staging.entries;
const std::size_t old_size = entries.size();
const size_t old_size = entries.size();
const auto is_deleteable = [this](const StagingBuffer& entry) {
return scheduler.IsFree(entry.tick);
};
const std::size_t begin_offset = staging.delete_index;
const std::size_t end_offset = std::min(begin_offset + deletions_per_tick, old_size);
const auto begin = std::begin(entries) + begin_offset;
const auto end = std::begin(entries) + end_offset;
const size_t begin_offset = staging.delete_index;
const size_t end_offset = std::min(begin_offset + deletions_per_tick, old_size);
const auto begin = entries.begin() + begin_offset;
const auto end = entries.begin() + end_offset;
entries.erase(std::remove_if(begin, end, is_deleteable), end);
const std::size_t new_size = entries.size();
const size_t new_size = entries.size();
staging.delete_index += deletions_per_tick;
if (staging.delete_index >= new_size) {
staging.delete_index = 0;
}
return (1ULL << log2) * (old_size - new_size);
if (staging.iterate_index > new_size) {
staging.iterate_index = 0;
}
}
} // namespace Vulkan

62
src/video_core/renderer_vulkan/vk_staging_buffer_pool.h
View File

@@ -9,7 +9,7 @@
#include "common/common_types.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
@@ -17,55 +17,65 @@ namespace Vulkan {
class Device;
class VKScheduler;
struct VKBuffer final {
vk::Buffer handle;
VKMemoryCommit commit;
struct StagingBufferRef {
VkBuffer buffer;
std::span<u8> mapped_span;
};
class VKStagingBufferPool final {
class StagingBufferPool {
public:
explicit VKStagingBufferPool(const Device& device, VKMemoryManager& memory_manager,
VKScheduler& scheduler);
~VKStagingBufferPool();
explicit StagingBufferPool(const Device& device, MemoryAllocator& memory_allocator,
VKScheduler& scheduler);
~StagingBufferPool();
VKBuffer& GetUnusedBuffer(std::size_t size, bool host_visible);
StagingBufferRef Request(size_t size, MemoryUsage usage);
void TickFrame();
private:
struct StagingBuffer final {
explicit StagingBuffer(std::unique_ptr<VKBuffer> buffer);
std::unique_ptr<VKBuffer> buffer;
struct StagingBuffer {
vk::Buffer buffer;
MemoryCommit commit;
std::span<u8> mapped_span;
u64 tick = 0;
StagingBufferRef Ref() const noexcept {
return StagingBufferRef{
.buffer = *buffer,
.mapped_span = mapped_span,
};
}
};
struct StagingBuffers final {
struct StagingBuffers {
std::vector<StagingBuffer> entries;
std::size_t delete_index = 0;
size_t delete_index = 0;
size_t iterate_index = 0;
};
static constexpr std::size_t NumLevels = sizeof(std::size_t) * CHAR_BIT;
using StagingBuffersCache = std::array<StagingBuffers, NumLevels>;
static constexpr size_t NUM_LEVELS = sizeof(size_t) * CHAR_BIT;
using StagingBuffersCache = std::array<StagingBuffers, NUM_LEVELS>;
VKBuffer* TryGetReservedBuffer(std::size_t size, bool host_visible);
std::optional<StagingBufferRef> TryGetReservedBuffer(size_t size, MemoryUsage usage);
VKBuffer& CreateStagingBuffer(std::size_t size, bool host_visible);
StagingBufferRef CreateStagingBuffer(size_t size, MemoryUsage usage);
StagingBuffersCache& GetCache(bool host_visible);
StagingBuffersCache& GetCache(MemoryUsage usage);
void ReleaseCache(bool host_visible);
void ReleaseCache(MemoryUsage usage);
u64 ReleaseLevel(StagingBuffersCache& cache, std::size_t log2);
void ReleaseLevel(StagingBuffersCache& cache, size_t log2);
const Device& device;
VKMemoryManager& memory_manager;
MemoryAllocator& memory_allocator;
VKScheduler& scheduler;
StagingBuffersCache host_staging_buffers;
StagingBuffersCache device_staging_buffers;
StagingBuffersCache device_local_cache;
StagingBuffersCache upload_cache;
StagingBuffersCache download_cache;
std::size_t current_delete_level = 0;
size_t current_delete_level = 0;
u64 buffer_index = 0;
};
} // namespace Vulkan

20
src/video_core/renderer_vulkan/vk_texture_cache.cpp
View File

@@ -10,12 +10,12 @@
#include "video_core/engines/fermi_2d.h"
#include "video_core/renderer_vulkan/blit_image.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
@@ -554,10 +554,18 @@ void TextureCacheRuntime::Finish() {
}
ImageBufferMap TextureCacheRuntime::MapUploadBuffer(size_t size) {
const auto& buffer = staging_buffer_pool.GetUnusedBuffer(size, true);
const auto staging_ref = staging_buffer_pool.Request(size, MemoryUsage::Upload);
return ImageBufferMap{
.handle = *buffer.handle,
.map = buffer.commit->Map(size),
.handle = staging_ref.buffer,
.span = staging_ref.mapped_span,
};
}
ImageBufferMap TextureCacheRuntime::MapDownloadBuffer(size_t size) {
const auto staging_ref = staging_buffer_pool.Request(size, MemoryUsage::Download);
return ImageBufferMap{
.handle = staging_ref.buffer,
.span = staging_ref.mapped_span,
};
}
@@ -788,9 +796,9 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_
image(MakeImage(runtime.device, info)), buffer(MakeBuffer(runtime.device, info)),
aspect_mask(ImageAspectMask(info.format)) {
if (image) {
commit = runtime.memory_manager.Commit(image, false);
commit = runtime.memory_allocator.Commit(image, MemoryUsage::DeviceLocal);
} else {
commit = runtime.memory_manager.Commit(buffer, false);
commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
}
if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) {
flags |= VideoCommon::ImageFlagBits::Converted;

22
src/video_core/renderer_vulkan/vk_texture_cache.h
View File

@@ -7,8 +7,8 @@
#include <compare>
#include <span>
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/texture_cache/texture_cache.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
@@ -19,14 +19,13 @@ using VideoCommon::Offset2D;
using VideoCommon::RenderTargets;
using VideoCore::Surface::PixelFormat;
class VKScheduler;
class VKStagingBufferPool;
class BlitImageHelper;
class Device;
class Image;
class ImageView;
class Framebuffer;
class StagingBufferPool;
class VKScheduler;
struct RenderPassKey {
constexpr auto operator<=>(const RenderPassKey&) const noexcept = default;
@@ -60,18 +59,18 @@ struct ImageBufferMap {
}
[[nodiscard]] std::span<u8> Span() const noexcept {
return map.Span();
return span;
}
VkBuffer handle;
MemoryMap map;
std::span<u8> span;
};
struct TextureCacheRuntime {
const Device& device;
VKScheduler& scheduler;
VKMemoryManager& memory_manager;
VKStagingBufferPool& staging_buffer_pool;
MemoryAllocator& memory_allocator;
StagingBufferPool& staging_buffer_pool;
BlitImageHelper& blit_image_helper;
std::unordered_map<RenderPassKey, vk::RenderPass> renderpass_cache;
@@ -79,10 +78,7 @@ struct TextureCacheRuntime {
[[nodiscard]] ImageBufferMap MapUploadBuffer(size_t size);
[[nodiscard]] ImageBufferMap MapDownloadBuffer(size_t size) {
// TODO: Have a special function for this
return MapUploadBuffer(size);
}
[[nodiscard]] ImageBufferMap MapDownloadBuffer(size_t size);
void BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src,
const std::array<Offset2D, 2>& dst_region,
@@ -141,7 +137,7 @@ private:
VKScheduler* scheduler;
vk::Image image;
vk::Buffer buffer;
VKMemoryCommit commit;
MemoryCommit commit;
VkImageAspectFlags aspect_mask = 0;
bool initialized = false;
};