yuzu-emu
/
yuzu-android
Archived
1
0
Fork 0
Code

gl_buffer_cache: Implement with generic buffer cache

This commit is contained in:
ReinUsesLisp 2019-06-20 03:22:25 -03:00
parent 32c0212b24
commit 1fa21fa192
8 changed files with 94 additions and 293 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/video_core/rasterizer_interface.h
View File

@ -47,6 +47,9 @@ public:
/// and invalidated
virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
/// Notify rasterizer that a frame is about to finish
virtual void TickFrame() = 0;
/// Attempt to use a faster method to perform a surface copy
virtual bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
const Tegra::Engines::Fermi2D::Regs::Surface& dst,

193
src/video_core/renderer_opengl/gl_buffer_cache.cpp
View File

@ -2,192 +2,57 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <memory>
#include <utility>
#include "common/alignment.h"
#include <glad/glad.h>
#include "common/assert.h"
#include "core/core.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_opengl/gl_buffer_cache.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
namespace OpenGL {
namespace {
constexpr GLuint EmptyBuffer = 0;
constexpr GLintptr CachedBufferOffset = 0;
OGLBuffer CreateBuffer(std::size_t size, GLenum usage) {
OGLBuffer buffer;
buffer.Create();
glNamedBufferData(buffer.handle, size, nullptr, usage);
return buffer;
}
} // Anonymous namespace
CachedBufferEntry::CachedBufferEntry(VAddr cpu_addr, u8* host_ptr)
: RasterizerCacheObject{host_ptr}, host_ptr{host_ptr}, cpu_addr{cpu_addr} {}
OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system, std::size_t size)
: RasterizerCache{rasterizer}, system{system}, stream_buffer(size, true) {}
OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
std::size_t stream_size)
: VideoCommon::BufferCache<OGLBuffer, GLuint, OGLStreamBuffer>{
rasterizer, system, std::make_unique<OGLStreamBuffer>(stream_size, true)} {}
OGLBufferCache::~OGLBufferCache() = default;
void OGLBufferCache::Unregister(const std::shared_ptr<CachedBufferEntry>& entry) {
std::lock_guard lock{mutex};
if (entry->IsInternalized()) {
internalized_entries.erase(entry->GetCacheAddr());
}
ReserveBuffer(entry);
RasterizerCache<std::shared_ptr<CachedBufferEntry>>::Unregister(entry);
OGLBuffer OGLBufferCache::CreateBuffer(std::size_t size) {
OGLBuffer buffer;
buffer.Create();
glNamedBufferData(buffer.handle, static_cast<GLsizeiptr>(size), nullptr, GL_DYNAMIC_DRAW);
return buffer;
}
OGLBufferCache::BufferInfo OGLBufferCache::UploadMemory(GPUVAddr gpu_addr, std::size_t size,
std::size_t alignment, bool internalize,
bool is_written) {
std::lock_guard lock{mutex};
auto& memory_manager = system.GPU().MemoryManager();
const auto host_ptr{memory_manager.GetPointer(gpu_addr)};
const auto cache_addr{ToCacheAddr(host_ptr)};
if (!host_ptr) {
return {EmptyBuffer, 0};
const GLuint* OGLBufferCache::ToHandle(const OGLBuffer& buffer) {
return &buffer.handle;
}
// Cache management is a big overhead, so only cache entries with a given size.
// TODO: Figure out which size is the best for given games.
if (!internalize && size < 0x800 &&
internalized_entries.find(cache_addr) == internalized_entries.end()) {
return StreamBufferUpload(host_ptr, size, alignment);
const GLuint* OGLBufferCache::GetEmptyBuffer(std::size_t) {
static const GLuint null_buffer = 0;
return &null_buffer;
}
auto entry = TryGet(host_ptr);
if (!entry) {
return FixedBufferUpload(gpu_addr, host_ptr, size, internalize, is_written);
void OGLBufferCache::UploadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
const u8* data) {
glNamedBufferSubData(buffer.handle, static_cast<GLintptr>(offset),
static_cast<GLsizeiptr>(size), data);
}
if (entry->GetSize() < size) {
GrowBuffer(entry, size);
}
if (is_written) {
entry->MarkAsModified(true, *this);
}
return {entry->GetBuffer(), CachedBufferOffset};
void OGLBufferCache::DownloadBufferData(const OGLBuffer& buffer, std::size_t offset,
std::size_t size, u8* data) {
glGetNamedBufferSubData(buffer.handle, static_cast<GLintptr>(offset),
static_cast<GLsizeiptr>(size), data);
}
OGLBufferCache::BufferInfo OGLBufferCache::UploadHostMemory(const void* raw_pointer,
std::size_t size,
std::size_t alignment) {
std::lock_guard lock{mutex};
return StreamBufferUpload(raw_pointer, size, alignment);
}
bool OGLBufferCache::Map(std::size_t max_size) {
const auto max_size_ = static_cast<GLsizeiptr>(max_size);
bool invalidate;
std::tie(buffer_ptr, buffer_offset_base, invalidate) = stream_buffer.Map(max_size_, 4);
buffer_offset = buffer_offset_base;
return invalidate;
}
void OGLBufferCache::Unmap() {
stream_buffer.Unmap(buffer_offset - buffer_offset_base);
}
void OGLBufferCache::FlushObjectInner(const std::shared_ptr<CachedBufferEntry>& entry) {
glGetNamedBufferSubData(entry->GetBuffer(), 0, entry->GetSize(), entry->GetWritableHostPtr());
}
OGLBufferCache::BufferInfo OGLBufferCache::StreamBufferUpload(const void* raw_pointer,
std::size_t size,
std::size_t alignment) {
AlignBuffer(alignment);
const GLintptr uploaded_offset = buffer_offset;
std::memcpy(buffer_ptr, raw_pointer, size);
buffer_ptr += size;
buffer_offset += size;
return {stream_buffer.GetHandle(), uploaded_offset};
}
OGLBufferCache::BufferInfo OGLBufferCache::FixedBufferUpload(GPUVAddr gpu_addr, u8* host_ptr,
std::size_t size, bool internalize,
bool is_written) {
auto& memory_manager = system.GPU().MemoryManager();
const auto cpu_addr = *memory_manager.GpuToCpuAddress(gpu_addr);
auto entry = GetUncachedBuffer(cpu_addr, host_ptr);
entry->SetSize(size);
entry->SetInternalState(internalize);
Register(entry);
if (internalize) {
internalized_entries.emplace(ToCacheAddr(host_ptr));
}
if (is_written) {
entry->MarkAsModified(true, *this);
}
if (entry->GetCapacity() < size) {
entry->SetCapacity(CreateBuffer(size, GL_STATIC_DRAW), size);
}
glNamedBufferSubData(entry->GetBuffer(), 0, static_cast<GLintptr>(size), host_ptr);
return {entry->GetBuffer(), CachedBufferOffset};
}
void OGLBufferCache::GrowBuffer(std::shared_ptr<CachedBufferEntry>& entry, std::size_t new_size) {
const auto old_size = static_cast<GLintptr>(entry->GetSize());
if (entry->GetCapacity() < new_size) {
const auto old_buffer = entry->GetBuffer();
OGLBuffer new_buffer = CreateBuffer(new_size, GL_STATIC_COPY);
// Copy bits from the old buffer to the new buffer.
glCopyNamedBufferSubData(old_buffer, new_buffer.handle, 0, 0, old_size);
entry->SetCapacity(std::move(new_buffer), new_size);
}
// Upload the new bits.
const auto size_diff = static_cast<GLintptr>(new_size - old_size);
glNamedBufferSubData(entry->GetBuffer(), old_size, size_diff, entry->GetHostPtr() + old_size);
// Update entry's size in the object and in the cache.
entry->SetSize(new_size);
Unregister(entry);
Register(entry);
}
std::shared_ptr<CachedBufferEntry> OGLBufferCache::GetUncachedBuffer(VAddr cpu_addr, u8* host_ptr) {
if (auto entry = TryGetReservedBuffer(host_ptr)) {
return entry;
}
return std::make_shared<CachedBufferEntry>(cpu_addr, host_ptr);
}
std::shared_ptr<CachedBufferEntry> OGLBufferCache::TryGetReservedBuffer(u8* host_ptr) {
const auto it = buffer_reserve.find(ToCacheAddr(host_ptr));
if (it == buffer_reserve.end()) {
return {};
}
auto& reserve = it->second;
auto entry = reserve.back();
reserve.pop_back();
return entry;
}
void OGLBufferCache::ReserveBuffer(std::shared_ptr<CachedBufferEntry> entry) {
buffer_reserve[entry->GetCacheAddr()].push_back(std::move(entry));
}
void OGLBufferCache::AlignBuffer(std::size_t alignment) {
// Align the offset, not the mapped pointer
const GLintptr offset_aligned =
static_cast<GLintptr>(Common::AlignUp(static_cast<std::size_t>(buffer_offset), alignment));
buffer_ptr += offset_aligned - buffer_offset;
buffer_offset = offset_aligned;
void OGLBufferCache::CopyBufferData(const OGLBuffer& src, const OGLBuffer& dst,
std::size_t src_offset, std::size_t dst_offset,
std::size_t size) {
glCopyNamedBufferSubData(src.handle, dst.handle, static_cast<GLintptr>(src_offset),
static_cast<GLintptr>(dst_offset), static_cast<GLsizeiptr>(size));
}
} // namespace OpenGL

115
src/video_core/renderer_opengl/gl_buffer_cache.h
View File

@ -4,15 +4,10 @@
#pragma once
#include <cstddef>
#include <map>
#include <memory>
#include <tuple>
#include <unordered_set>
#include <utility>
#include <vector>
#include "common/common_types.h"
#include "video_core/buffer_cache.h"
#include "video_core/rasterizer_cache.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_stream_buffer.h"
@ -23,112 +18,30 @@ class System;
namespace OpenGL {
class OGLStreamBuffer;
class RasterizerOpenGL;
class CachedBufferEntry final : public RasterizerCacheObject {
class OGLBufferCache final : public VideoCommon::BufferCache<OGLBuffer, GLuint, OGLStreamBuffer> {
public:
explicit CachedBufferEntry(VAddr cpu_addr, u8* host_ptr);
VAddr GetCpuAddr() const override {
return cpu_addr;
}
std::size_t GetSizeInBytes() const override {
return size;
}
u8* GetWritableHostPtr() const {
return host_ptr;
}
std::size_t GetSize() const {
return size;
}
std::size_t GetCapacity() const {
return capacity;
}
bool IsInternalized() const {
return is_internal;
}
GLuint GetBuffer() const {
return buffer.handle;
}
void SetSize(std::size_t new_size) {
size = new_size;
}
void SetInternalState(bool is_internal_) {
is_internal = is_internal_;
}
void SetCapacity(OGLBuffer&& new_buffer, std::size_t new_capacity) {
capacity = new_capacity;
buffer = std::move(new_buffer);
}
private:
u8* host_ptr{};
VAddr cpu_addr{};
std::size_t size{};
std::size_t capacity{};
bool is_internal{};
OGLBuffer buffer;
};
class OGLBufferCache final : public RasterizerCache<std::shared_ptr<CachedBufferEntry>> {
using BufferInfo = std::pair<GLuint, GLintptr>;
public:
explicit OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system, std::size_t size);
explicit OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
std::size_t stream_size);
~OGLBufferCache();
void Unregister(const std::shared_ptr<CachedBufferEntry>& entry) override;
/// Uploads data from a guest GPU address. Returns the OpenGL buffer where it's located and its
/// offset.
BufferInfo UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
bool internalize = false, bool is_written = false);
/// Uploads from a host memory. Returns the OpenGL buffer where it's located and its offset.
BufferInfo UploadHostMemory(const void* raw_pointer, std::size_t size,
std::size_t alignment = 4);
bool Map(std::size_t max_size);
void Unmap();
protected:
// We do not have to flush this cache as things in it are never modified by us.
void FlushObjectInner(const std::shared_ptr<CachedBufferEntry>& entry) override;
OGLBuffer CreateBuffer(std::size_t size) override;
private:
BufferInfo StreamBufferUpload(const void* raw_pointer, std::size_t size, std::size_t alignment);
const GLuint* ToHandle(const OGLBuffer& buffer) override;
BufferInfo FixedBufferUpload(GPUVAddr gpu_addr, u8* host_ptr, std::size_t size,
bool internalize, bool is_written);
const GLuint* GetEmptyBuffer(std::size_t) override;
void GrowBuffer(std::shared_ptr<CachedBufferEntry>& entry, std::size_t new_size);
void UploadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
const u8* data) override;
std::shared_ptr<CachedBufferEntry> GetUncachedBuffer(VAddr cpu_addr, u8* host_ptr);
void DownloadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
u8* data) override;
std::shared_ptr<CachedBufferEntry> TryGetReservedBuffer(u8* host_ptr);
void ReserveBuffer(std::shared_ptr<CachedBufferEntry> entry);
void AlignBuffer(std::size_t alignment);
Core::System& system;
u8* buffer_ptr = nullptr;
GLintptr buffer_offset = 0;
GLintptr buffer_offset_base = 0;
OGLStreamBuffer stream_buffer;
std::unordered_set<CacheAddr> internalized_entries;
std::unordered_map<CacheAddr, std::vector<std::shared_ptr<CachedBufferEntry>>> buffer_reserve;
void CopyBufferData(const OGLBuffer& src, const OGLBuffer& dst, std::size_t src_offset,
std::size_t dst_offset, std::size_t size) override;
};
} // namespace OpenGL

40
src/video_core/renderer_opengl/gl_rasterizer.cpp
View File

@ -198,7 +198,8 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
const auto [vertex_buffer, vertex_buffer_offset] = buffer_cache.UploadMemory(start, size);
// Bind the vertex array to the buffer at the current offset.
glVertexArrayVertexBuffer(vao, index, vertex_buffer, vertex_buffer_offset,
// FIXME(Rodrigo): This dereferenced pointer might be invalidated in future uploads.
glVertexArrayVertexBuffer(vao, index, *vertex_buffer, vertex_buffer_offset,
vertex_array.stride);
if (regs.instanced_arrays.IsInstancingEnabled(index) && vertex_array.divisor != 0) {
@ -221,7 +222,8 @@ GLintptr RasterizerOpenGL::SetupIndexBuffer(GLuint vao) {
const auto& regs = system.GPU().Maxwell3D().regs;
const std::size_t size = CalculateIndexBufferSize();
const auto [buffer, offset] = buffer_cache.UploadMemory(regs.index_array.IndexStart(), size);
glVertexArrayElementBuffer(vao, buffer);
// FIXME(Rodrigo): This dereferenced pointer might be invalidated in future uploads.
glVertexArrayElementBuffer(vao, *buffer);
return offset;
}
@ -255,10 +257,6 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
BaseBindings base_bindings;
std::array<bool, Maxwell::NumClipDistances> clip_distances{};
// Prepare packed bindings
bind_ubo_pushbuffer.Setup(base_bindings.cbuf);
bind_ssbo_pushbuffer.Setup(base_bindings.gmem);
for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
const auto& shader_config = gpu.regs.shader_config[index];
const Maxwell::ShaderProgram program{static_cast<Maxwell::ShaderProgram>(index)};
@ -328,9 +326,6 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
base_bindings = next_bindings;
}
bind_ubo_pushbuffer.Bind();
bind_ssbo_pushbuffer.Bind();
SyncClipEnabled(clip_distances);
gpu.dirty_flags.shaders = false;
@ -644,11 +639,8 @@ void RasterizerOpenGL::DrawArrays() {
buffer_size +=
Maxwell::MaxConstBuffers * (MaxConstbufferSize + device.GetUniformBufferAlignment());
const bool invalidate = buffer_cache.Map(buffer_size);
if (invalidate) {
// As all cached buffers are invalidated, we need to recheck their state.
gpu.dirty_flags.vertex_array.set();
}
// Prepare the vertex array.
buffer_cache.Map(buffer_size);
// Prepare vertex array format.
const GLuint vao = SetupVertexFormat();
@ -660,6 +652,10 @@ void RasterizerOpenGL::DrawArrays() {
// Setup draw parameters. It will automatically choose what glDraw* method to use.
const DrawParameters params = SetupDraw(index_buffer_offset);
// Prepare packed bindings.
bind_ubo_pushbuffer.Setup(0);
bind_ssbo_pushbuffer.Setup(0);
// Setup shaders and their used resources.
texture_cache.GuardSamplers(true);
SetupShaders(params.primitive_mode);
@ -667,7 +663,17 @@ void RasterizerOpenGL::DrawArrays() {
ConfigureFramebuffers(state);
buffer_cache.Unmap();
// Signal the buffer cache that we are not going to upload more things.
const bool invalidate = buffer_cache.Unmap();
// Now that we are no longer uploading data, we can safely bind the buffers to OpenGL.
bind_ubo_pushbuffer.Bind();
bind_ssbo_pushbuffer.Bind();
if (invalidate) {
// As all cached buffers are invalidated, we need to recheck their state.
gpu.dirty_flags.vertex_array.set();
}
shader_program_manager->ApplyTo(state);
state.Apply();
@ -709,6 +715,10 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
InvalidateRegion(addr, size);
}
void RasterizerOpenGL::TickFrame() {
buffer_cache.TickFrame();
}
bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
const Tegra::Engines::Fermi2D::Regs::Surface& dst,
const Tegra::Engines::Fermi2D::Config& copy_config) {

1
src/video_core/renderer_opengl/gl_rasterizer.h
View File

@ -62,6 +62,7 @@ public:
void FlushRegion(CacheAddr addr, u64 size) override;
void InvalidateRegion(CacheAddr addr, u64 size) override;
void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
void TickFrame() override;
bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
const Tegra::Engines::Fermi2D::Regs::Surface& dst,
const Tegra::Engines::Fermi2D::Config& copy_config) override;

4
src/video_core/renderer_opengl/renderer_opengl.cpp
View File

@ -101,7 +101,6 @@ RendererOpenGL::RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::Syst
RendererOpenGL::~RendererOpenGL() = default;
/// Swap buffers (render frame)
void RendererOpenGL::SwapBuffers(
std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) {
@ -130,6 +129,8 @@ void RendererOpenGL::SwapBuffers(
DrawScreen(render_window.GetFramebufferLayout());
rasterizer->TickFrame();
render_window.SwapBuffers();
}
@ -262,7 +263,6 @@ void RendererOpenGL::CreateRasterizer() {
if (rasterizer) {
return;
}
// Initialize sRGB Usage
OpenGLState::ClearsRGBUsed();
rasterizer = std::make_unique<RasterizerOpenGL>(system, emu_window, screen_info);
}

17
src/video_core/renderer_opengl/utils.cpp
View File

@ -19,23 +19,30 @@ BindBuffersRangePushBuffer::~BindBuffersRangePushBuffer() = default;
void BindBuffersRangePushBuffer::Setup(GLuint first_) {
first = first_;
buffers.clear();
buffer_pointers.clear();
offsets.clear();
sizes.clear();
}
void BindBuffersRangePushBuffer::Push(GLuint buffer, GLintptr offset, GLsizeiptr size) {
buffers.push_back(buffer);
void BindBuffersRangePushBuffer::Push(const GLuint* buffer, GLintptr offset, GLsizeiptr size) {
buffer_pointers.push_back(buffer);
offsets.push_back(offset);
sizes.push_back(size);
}
void BindBuffersRangePushBuffer::Bind() const {
const std::size_t count{buffers.size()};
void BindBuffersRangePushBuffer::Bind() {
// Ensure sizes are valid.
const std::size_t count{buffer_pointers.size()};
DEBUG_ASSERT(count == offsets.size() && count == sizes.size());
if (count == 0) {
return;
}
// Dereference buffers.
buffers.resize(count);
std::transform(buffer_pointers.begin(), buffer_pointers.end(), buffers.begin(),
[](const GLuint* pointer) { return *pointer; });
glBindBuffersRange(target, first, static_cast<GLsizei>(count), buffers.data(), offsets.data(),
sizes.data());
}

14
src/video_core/renderer_opengl/utils.h
View File

@ -11,20 +11,22 @@
namespace OpenGL {
class BindBuffersRangePushBuffer {
class BindBuffersRangePushBuffer final {
public:
BindBuffersRangePushBuffer(GLenum target);
explicit BindBuffersRangePushBuffer(GLenum target);
~BindBuffersRangePushBuffer();
void Setup(GLuint first_);
void Push(GLuint buffer, GLintptr offset, GLsizeiptr size);
void Push(const GLuint* buffer, GLintptr offset, GLsizeiptr size);
void Bind() const;
void Bind();
private:
GLenum target;
GLuint first;
GLenum target{};
GLuint first{};
std::vector<const GLuint*> buffer_pointers;
std::vector<GLuint> buffers;
std::vector<GLintptr> offsets;
std::vector<GLsizeiptr> sizes;