Merge pull request #2675 from ReinUsesLisp/opengl-buffer-cache

buffer_cache: Implement a generic buffer cache and its OpenGL backend
This commit is contained in:
bunnei 2019-07-14 19:03:43 -04:00 committed by GitHub
commit 3477b92289
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
16 changed files with 538 additions and 408 deletions

View file

@ -1,4 +1,5 @@
add_library(video_core STATIC
buffer_cache.h
dma_pusher.cpp
dma_pusher.h
debug_utils/debug_utils.cpp
@ -43,8 +44,6 @@ add_library(video_core STATIC
renderer_opengl/gl_device.h
renderer_opengl/gl_framebuffer_cache.cpp
renderer_opengl/gl_framebuffer_cache.h
renderer_opengl/gl_global_cache.cpp
renderer_opengl/gl_global_cache.h
renderer_opengl/gl_rasterizer.cpp
renderer_opengl/gl_rasterizer.h
renderer_opengl/gl_resource_manager.cpp

View file

@ -0,0 +1,299 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <memory>
#include <mutex>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
#include "common/alignment.h"
#include "common/common_types.h"
#include "core/core.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_cache.h"
namespace VideoCore {
class RasterizerInterface;
}
namespace VideoCommon {
template <typename BufferStorageType>
class CachedBuffer final : public RasterizerCacheObject {
public:
explicit CachedBuffer(VAddr cpu_addr, u8* host_ptr)
: RasterizerCacheObject{host_ptr}, host_ptr{host_ptr}, cpu_addr{cpu_addr} {}
~CachedBuffer() override = default;
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;
}
const BufferStorageType& GetBuffer() const {
return buffer;
}
void SetSize(std::size_t new_size) {
size = new_size;
}
void SetInternalState(bool is_internal_) {
is_internal = is_internal_;
}
BufferStorageType ExchangeBuffer(BufferStorageType buffer_, std::size_t new_capacity) {
capacity = new_capacity;
std::swap(buffer, buffer_);
return buffer_;
}
private:
u8* host_ptr{};
VAddr cpu_addr{};
std::size_t size{};
std::size_t capacity{};
bool is_internal{};
BufferStorageType buffer;
};
template <typename BufferStorageType, typename BufferType, typename StreamBuffer>
class BufferCache : public RasterizerCache<std::shared_ptr<CachedBuffer<BufferStorageType>>> {
public:
using Buffer = std::shared_ptr<CachedBuffer<BufferStorageType>>;
using BufferInfo = std::pair<const BufferType*, u64>;
explicit BufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
std::unique_ptr<StreamBuffer> stream_buffer)
: RasterizerCache<Buffer>{rasterizer}, system{system},
stream_buffer{std::move(stream_buffer)}, stream_buffer_handle{
this->stream_buffer->GetHandle()} {}
~BufferCache() = default;
void Unregister(const Buffer& entry) override {
std::lock_guard lock{RasterizerCache<Buffer>::mutex};
if (entry->IsInternalized()) {
internalized_entries.erase(entry->GetCacheAddr());
}
ReserveBuffer(entry);
RasterizerCache<Buffer>::Unregister(entry);
}
void TickFrame() {
marked_for_destruction_index =
(marked_for_destruction_index + 1) % marked_for_destruction_ring_buffer.size();
MarkedForDestruction().clear();
}
BufferInfo UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
bool internalize = false, bool is_written = false) {
std::lock_guard lock{RasterizerCache<Buffer>::mutex};
auto& memory_manager = system.GPU().MemoryManager();
const auto host_ptr = memory_manager.GetPointer(gpu_addr);
if (!host_ptr) {
return {GetEmptyBuffer(size), 0};
}
const auto cache_addr = ToCacheAddr(host_ptr);
// 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.
constexpr std::size_t max_stream_size = 0x800;
if (!internalize && size < max_stream_size &&
internalized_entries.find(cache_addr) == internalized_entries.end()) {
return StreamBufferUpload(host_ptr, size, alignment);
}
auto entry = RasterizerCache<Buffer>::TryGet(cache_addr);
if (!entry) {
return FixedBufferUpload(gpu_addr, host_ptr, size, internalize, is_written);
}
if (entry->GetSize() < size) {
IncreaseBufferSize(entry, size);
}
if (is_written) {
entry->MarkAsModified(true, *this);
}
return {ToHandle(entry->GetBuffer()), 0};
}
/// 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) {
std::lock_guard lock{RasterizerCache<Buffer>::mutex};
return StreamBufferUpload(raw_pointer, size, alignment);
}
void Map(std::size_t max_size) {
std::tie(buffer_ptr, buffer_offset_base, invalidated) = stream_buffer->Map(max_size, 4);
buffer_offset = buffer_offset_base;
}
/// Finishes the upload stream, returns true on bindings invalidation.
bool Unmap() {
stream_buffer->Unmap(buffer_offset - buffer_offset_base);
return std::exchange(invalidated, false);
}
virtual const BufferType* GetEmptyBuffer(std::size_t size) = 0;
protected:
void FlushObjectInner(const Buffer& entry) override {
DownloadBufferData(entry->GetBuffer(), 0, entry->GetSize(), entry->GetWritableHostPtr());
}
virtual BufferStorageType CreateBuffer(std::size_t size) = 0;
virtual const BufferType* ToHandle(const BufferStorageType& storage) = 0;
virtual void UploadBufferData(const BufferStorageType& buffer, std::size_t offset,
std::size_t size, const u8* data) = 0;
virtual void DownloadBufferData(const BufferStorageType& buffer, std::size_t offset,
std::size_t size, u8* data) = 0;
virtual void CopyBufferData(const BufferStorageType& src, const BufferStorageType& dst,
std::size_t src_offset, std::size_t dst_offset,
std::size_t size) = 0;
private:
BufferInfo StreamBufferUpload(const void* raw_pointer, std::size_t size,
std::size_t alignment) {
AlignBuffer(alignment);
const std::size_t uploaded_offset = buffer_offset;
std::memcpy(buffer_ptr, raw_pointer, size);
buffer_ptr += size;
buffer_offset += size;
return {&stream_buffer_handle, uploaded_offset};
}
BufferInfo FixedBufferUpload(GPUVAddr gpu_addr, u8* host_ptr, std::size_t size,
bool internalize, bool is_written) {
auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
const auto cpu_addr = memory_manager.GpuToCpuAddress(gpu_addr);
ASSERT(cpu_addr);
auto entry = GetUncachedBuffer(*cpu_addr, host_ptr);
entry->SetSize(size);
entry->SetInternalState(internalize);
RasterizerCache<Buffer>::Register(entry);
if (internalize) {
internalized_entries.emplace(ToCacheAddr(host_ptr));
}
if (is_written) {
entry->MarkAsModified(true, *this);
}
if (entry->GetCapacity() < size) {
MarkedForDestruction().push_back(entry->ExchangeBuffer(CreateBuffer(size), size));
}
UploadBufferData(entry->GetBuffer(), 0, size, host_ptr);
return {ToHandle(entry->GetBuffer()), 0};
}
void IncreaseBufferSize(Buffer& entry, std::size_t new_size) {
const std::size_t old_size = entry->GetSize();
if (entry->GetCapacity() < new_size) {
const auto& old_buffer = entry->GetBuffer();
auto new_buffer = CreateBuffer(new_size);
// Copy bits from the old buffer to the new buffer.
CopyBufferData(old_buffer, new_buffer, 0, 0, old_size);
MarkedForDestruction().push_back(
entry->ExchangeBuffer(std::move(new_buffer), new_size));
// This buffer could have been used
invalidated = true;
}
// Upload the new bits.
const std::size_t size_diff = new_size - old_size;
UploadBufferData(entry->GetBuffer(), old_size, size_diff, entry->GetHostPtr() + old_size);
// Update entry's size in the object and in the cache.
Unregister(entry);
entry->SetSize(new_size);
RasterizerCache<Buffer>::Register(entry);
}
Buffer GetUncachedBuffer(VAddr cpu_addr, u8* host_ptr) {
if (auto entry = TryGetReservedBuffer(host_ptr)) {
return entry;
}
return std::make_shared<CachedBuffer<BufferStorageType>>(cpu_addr, host_ptr);
}
Buffer 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 ReserveBuffer(Buffer entry) {
buffer_reserve[entry->GetCacheAddr()].push_back(std::move(entry));
}
void AlignBuffer(std::size_t alignment) {
// Align the offset, not the mapped pointer
const std::size_t offset_aligned = Common::AlignUp(buffer_offset, alignment);
buffer_ptr += offset_aligned - buffer_offset;
buffer_offset = offset_aligned;
}
std::vector<BufferStorageType>& MarkedForDestruction() {
return marked_for_destruction_ring_buffer[marked_for_destruction_index];
}
Core::System& system;
std::unique_ptr<StreamBuffer> stream_buffer;
BufferType stream_buffer_handle{};
bool invalidated = false;
u8* buffer_ptr = nullptr;
u64 buffer_offset = 0;
u64 buffer_offset_base = 0;
std::size_t marked_for_destruction_index = 0;
std::array<std::vector<BufferStorageType>, 4> marked_for_destruction_ring_buffer;
std::unordered_set<CacheAddr> internalized_entries;
std::unordered_map<CacheAddr, std::vector<Buffer>> buffer_reserve;
};
} // namespace VideoCommon

View file

@ -67,6 +67,7 @@ public:
static constexpr std::size_t MaxShaderStage = 5;
// Maximum number of const buffers per shader stage.
static constexpr std::size_t MaxConstBuffers = 18;
static constexpr std::size_t MaxConstBufferSize = 0x10000;
enum class QueryMode : u32 {
Write = 0,

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,

View file

@ -2,103 +2,57 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <memory>
#include "common/alignment.h"
#include "core/core.h"
#include "video_core/memory_manager.h"
#include <glad/glad.h>
#include "common/assert.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 {
CachedBufferEntry::CachedBufferEntry(VAddr cpu_addr, std::size_t size, GLintptr offset,
std::size_t alignment, u8* host_ptr)
: RasterizerCacheObject{host_ptr}, cpu_addr{cpu_addr}, size{size}, offset{offset},
alignment{alignment} {}
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(RasterizerOpenGL& rasterizer, std::size_t size)
: RasterizerCache{rasterizer}, stream_buffer(size, true) {}
OGLBufferCache::~OGLBufferCache() = default;
GLintptr OGLBufferCache::UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment,
bool cache) {
std::lock_guard lock{mutex};
auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
// 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.
cache &= size >= 2048;
const auto& host_ptr{memory_manager.GetPointer(gpu_addr)};
if (cache) {
auto entry = TryGet(host_ptr);
if (entry) {
if (entry->GetSize() >= size && entry->GetAlignment() == alignment) {
return entry->GetOffset();
}
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;
}
AlignBuffer(alignment);
const GLintptr uploaded_offset = buffer_offset;
if (!host_ptr) {
return uploaded_offset;
const GLuint* OGLBufferCache::ToHandle(const OGLBuffer& buffer) {
return &buffer.handle;
}
std::memcpy(buffer_ptr, host_ptr, size);
buffer_ptr += size;
buffer_offset += size;
if (cache) {
auto entry = std::make_shared<CachedBufferEntry>(
*memory_manager.GpuToCpuAddress(gpu_addr), size, uploaded_offset, alignment, host_ptr);
Register(entry);
const GLuint* OGLBufferCache::GetEmptyBuffer(std::size_t) {
static const GLuint null_buffer = 0;
return &null_buffer;
}
return uploaded_offset;
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);
}
GLintptr OGLBufferCache::UploadHostMemory(const void* raw_pointer, std::size_t size,
std::size_t alignment) {
std::lock_guard lock{mutex};
AlignBuffer(alignment);
std::memcpy(buffer_ptr, raw_pointer, size);
const GLintptr uploaded_offset = buffer_offset;
buffer_ptr += size;
buffer_offset += size;
return uploaded_offset;
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);
}
bool OGLBufferCache::Map(std::size_t max_size) {
bool invalidate;
std::tie(buffer_ptr, buffer_offset_base, invalidate) =
stream_buffer.Map(static_cast<GLsizeiptr>(max_size), 4);
buffer_offset = buffer_offset_base;
if (invalidate) {
InvalidateAll();
}
return invalidate;
}
void OGLBufferCache::Unmap() {
stream_buffer.Unmap(buffer_offset - buffer_offset_base);
}
GLuint OGLBufferCache::GetHandle() const {
return stream_buffer.GetHandle();
}
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

View file

@ -4,80 +4,44 @@
#pragma once
#include <cstddef>
#include <memory>
#include <tuple>
#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"
namespace Core {
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, std::size_t size, GLintptr offset,
std::size_t alignment, u8* host_ptr);
explicit OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
std::size_t stream_size);
~OGLBufferCache();
VAddr GetCpuAddr() const override {
return cpu_addr;
}
std::size_t GetSizeInBytes() const override {
return size;
}
std::size_t GetSize() const {
return size;
}
GLintptr GetOffset() const {
return offset;
}
std::size_t GetAlignment() const {
return alignment;
}
private:
VAddr cpu_addr{};
std::size_t size{};
GLintptr offset{};
std::size_t alignment{};
};
class OGLBufferCache final : public RasterizerCache<std::shared_ptr<CachedBufferEntry>> {
public:
explicit OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size);
/// Uploads data from a guest GPU address. Returns host's buffer offset where it's been
/// allocated.
GLintptr UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
bool cache = true);
/// Uploads from a host memory. Returns host's buffer offset where it's been allocated.
GLintptr UploadHostMemory(const void* raw_pointer, std::size_t size, std::size_t alignment = 4);
bool Map(std::size_t max_size);
void Unmap();
GLuint GetHandle() const;
const GLuint* GetEmptyBuffer(std::size_t) override;
protected:
void AlignBuffer(std::size_t alignment);
OGLBuffer CreateBuffer(std::size_t size) override;
// We do not have to flush this cache as things in it are never modified by us.
void FlushObjectInner(const std::shared_ptr<CachedBufferEntry>& object) override {}
const GLuint* ToHandle(const OGLBuffer& buffer) override;
private:
OGLStreamBuffer stream_buffer;
void UploadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
const u8* data) override;
u8* buffer_ptr = nullptr;
GLintptr buffer_offset = 0;
GLintptr buffer_offset_base = 0;
void DownloadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
u8* data) override;
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

View file

@ -24,6 +24,7 @@ T GetInteger(GLenum pname) {
Device::Device() {
uniform_buffer_alignment = GetInteger<std::size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT);
shader_storage_alignment = GetInteger<std::size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT);
max_vertex_attributes = GetInteger<u32>(GL_MAX_VERTEX_ATTRIBS);
max_varyings = GetInteger<u32>(GL_MAX_VARYING_VECTORS);
has_variable_aoffi = TestVariableAoffi();

View file

@ -18,6 +18,10 @@ public:
return uniform_buffer_alignment;
}
std::size_t GetShaderStorageBufferAlignment() const {
return shader_storage_alignment;
}
u32 GetMaxVertexAttributes() const {
return max_vertex_attributes;
}
@ -39,6 +43,7 @@ private:
static bool TestComponentIndexingBug();
std::size_t uniform_buffer_alignment{};
std::size_t shader_storage_alignment{};
u32 max_vertex_attributes{};
u32 max_varyings{};
bool has_variable_aoffi{};

View file

@ -1,102 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <glad/glad.h>
#include "common/logging/log.h"
#include "core/core.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_opengl/gl_global_cache.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_opengl/utils.h"
namespace OpenGL {
CachedGlobalRegion::CachedGlobalRegion(VAddr cpu_addr, u8* host_ptr, u32 size, u32 max_size)
: RasterizerCacheObject{host_ptr}, cpu_addr{cpu_addr}, host_ptr{host_ptr}, size{size},
max_size{max_size} {
buffer.Create();
LabelGLObject(GL_BUFFER, buffer.handle, cpu_addr, "GlobalMemory");
}
CachedGlobalRegion::~CachedGlobalRegion() = default;
void CachedGlobalRegion::Reload(u32 size_) {
size = size_;
if (size > max_size) {
size = max_size;
LOG_CRITICAL(HW_GPU, "Global region size {} exceeded the supported size {}!", size_,
max_size);
}
glNamedBufferData(buffer.handle, size, host_ptr, GL_STREAM_DRAW);
}
void CachedGlobalRegion::Flush() {
LOG_DEBUG(Render_OpenGL, "Flushing {} bytes to CPU memory address 0x{:16}", size, cpu_addr);
glGetNamedBufferSubData(buffer.handle, 0, static_cast<GLsizeiptr>(size), host_ptr);
}
GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const {
const auto search{reserve.find(addr)};
if (search == reserve.end()) {
return {};
}
return search->second;
}
GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(GPUVAddr addr, u8* host_ptr,
u32 size) {
GlobalRegion region{TryGetReservedGlobalRegion(ToCacheAddr(host_ptr), size)};
if (!region) {
// No reserved surface available, create a new one and reserve it
auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
const auto cpu_addr{memory_manager.GpuToCpuAddress(addr)};
ASSERT(cpu_addr);
region = std::make_shared<CachedGlobalRegion>(*cpu_addr, host_ptr, size, max_ssbo_size);
ReserveGlobalRegion(region);
}
region->Reload(size);
return region;
}
void GlobalRegionCacheOpenGL::ReserveGlobalRegion(GlobalRegion region) {
reserve.insert_or_assign(region->GetCacheAddr(), std::move(region));
}
GlobalRegionCacheOpenGL::GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer)
: RasterizerCache{rasterizer} {
GLint max_ssbo_size_;
glGetIntegerv(GL_MAX_SHADER_STORAGE_BLOCK_SIZE, &max_ssbo_size_);
max_ssbo_size = static_cast<u32>(max_ssbo_size_);
}
GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion(
const GLShader::GlobalMemoryEntry& global_region,
Tegra::Engines::Maxwell3D::Regs::ShaderStage stage) {
std::lock_guard lock{mutex};
auto& gpu{Core::System::GetInstance().GPU()};
auto& memory_manager{gpu.MemoryManager()};
const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<std::size_t>(stage)]};
const auto addr{cbufs.const_buffers[global_region.GetCbufIndex()].address +
global_region.GetCbufOffset()};
const auto actual_addr{memory_manager.Read<u64>(addr)};
const auto size{memory_manager.Read<u32>(addr + 8)};
// Look up global region in the cache based on address
const auto& host_ptr{memory_manager.GetPointer(actual_addr)};
GlobalRegion region{TryGet(host_ptr)};
if (!region) {
// No global region found - create a new one
region = GetUncachedGlobalRegion(actual_addr, host_ptr, size);
Register(region);
}
return region;
}
} // namespace OpenGL

View file

@ -1,82 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <memory>
#include <unordered_map>
#include <glad/glad.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/rasterizer_cache.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
namespace OpenGL {
namespace GLShader {
class GlobalMemoryEntry;
}
class RasterizerOpenGL;
class CachedGlobalRegion;
using GlobalRegion = std::shared_ptr<CachedGlobalRegion>;
class CachedGlobalRegion final : public RasterizerCacheObject {
public:
explicit CachedGlobalRegion(VAddr cpu_addr, u8* host_ptr, u32 size, u32 max_size);
~CachedGlobalRegion();
VAddr GetCpuAddr() const override {
return cpu_addr;
}
std::size_t GetSizeInBytes() const override {
return size;
}
/// Gets the GL program handle for the buffer
GLuint GetBufferHandle() const {
return buffer.handle;
}
/// Reloads the global region from guest memory
void Reload(u32 size_);
void Flush();
private:
VAddr cpu_addr{};
u8* host_ptr{};
u32 size{};
u32 max_size{};
OGLBuffer buffer;
};
class GlobalRegionCacheOpenGL final : public RasterizerCache<GlobalRegion> {
public:
explicit GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer);
/// Gets the current specified shader stage program
GlobalRegion GetGlobalRegion(const GLShader::GlobalMemoryEntry& descriptor,
Tegra::Engines::Maxwell3D::Regs::ShaderStage stage);
protected:
void FlushObjectInner(const GlobalRegion& object) override {
object->Flush();
}
private:
GlobalRegion TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const;
GlobalRegion GetUncachedGlobalRegion(GPUVAddr addr, u8* host_ptr, u32 size);
void ReserveGlobalRegion(GlobalRegion region);
std::unordered_map<CacheAddr, GlobalRegion> reserve;
u32 max_ssbo_size{};
};
} // namespace OpenGL

View file

@ -20,6 +20,7 @@
#include "core/hle/kernel/process.h"
#include "core/settings.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/renderer_opengl/gl_shader_gen.h"
@ -80,11 +81,25 @@ struct DrawParameters {
}
};
static std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer,
const GLShader::ConstBufferEntry& entry) {
if (!entry.IsIndirect()) {
return entry.GetSize();
}
if (buffer.size > Maxwell::MaxConstBufferSize) {
LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", buffer.size,
Maxwell::MaxConstBufferSize);
return Maxwell::MaxConstBufferSize;
}
return buffer.size;
}
RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWindow& emu_window,
ScreenInfo& info)
: texture_cache{system, *this, device}, shader_cache{*this, system, emu_window, device},
global_cache{*this}, system{system}, screen_info{info},
buffer_cache(*this, STREAM_BUFFER_SIZE) {
system{system}, screen_info{info}, buffer_cache{*this, system, STREAM_BUFFER_SIZE} {
OpenGLState::ApplyDefaultState();
shader_program_manager = std::make_unique<GLShader::ProgramManager>();
@ -129,8 +144,6 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
state.draw.vertex_array = vao;
state.ApplyVertexArrayState();
glVertexArrayElementBuffer(vao, buffer_cache.GetHandle());
// Use the vertex array as-is, assumes that the data is formatted correctly for OpenGL.
// Enables the first 16 vertex attributes always, as we don't know which ones are actually
// used until shader time. Note, Tegra technically supports 32, but we're capping this to 16
@ -197,10 +210,10 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
ASSERT(end > start);
const u64 size = end - start + 1;
const GLintptr vertex_buffer_offset = buffer_cache.UploadMemory(start, size);
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, buffer_cache.GetHandle(), vertex_buffer_offset,
vertex_array_pushbuffer.SetVertexBuffer(index, vertex_buffer, vertex_buffer_offset,
vertex_array.stride);
if (regs.instanced_arrays.IsInstancingEnabled(index) && vertex_array.divisor != 0) {
@ -215,7 +228,19 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
gpu.dirty_flags.vertex_array.reset();
}
DrawParameters RasterizerOpenGL::SetupDraw() {
GLintptr RasterizerOpenGL::SetupIndexBuffer() {
if (accelerate_draw != AccelDraw::Indexed) {
return 0;
}
MICROPROFILE_SCOPE(OpenGL_Index);
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);
vertex_array_pushbuffer.SetIndexBuffer(buffer);
return offset;
}
DrawParameters RasterizerOpenGL::SetupDraw(GLintptr index_buffer_offset) {
const auto& gpu = system.GPU().Maxwell3D();
const auto& regs = gpu.regs;
const bool is_indexed = accelerate_draw == AccelDraw::Indexed;
@ -227,11 +252,9 @@ DrawParameters RasterizerOpenGL::SetupDraw() {
params.primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology);
if (is_indexed) {
MICROPROFILE_SCOPE(OpenGL_Index);
params.index_format = MaxwellToGL::IndexFormat(regs.index_array.format);
params.count = regs.index_array.count;
params.index_buffer_offset =
buffer_cache.UploadMemory(regs.index_array.IndexStart(), CalculateIndexBufferSize());
params.index_buffer_offset = index_buffer_offset;
params.base_vertex = static_cast<GLint>(regs.vb_element_base);
} else {
params.count = regs.vertex_buffer.count;
@ -247,10 +270,6 @@ 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)};
@ -271,12 +290,11 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
GLShader::MaxwellUniformData ubo{};
ubo.SetFromRegs(gpu, stage);
const GLintptr offset =
const auto [buffer, offset] =
buffer_cache.UploadHostMemory(&ubo, sizeof(ubo), device.GetUniformBufferAlignment());
// Bind the emulation info buffer
bind_ubo_pushbuffer.Push(buffer_cache.GetHandle(), offset,
static_cast<GLsizeiptr>(sizeof(ubo)));
bind_ubo_pushbuffer.Push(buffer, offset, static_cast<GLsizeiptr>(sizeof(ubo)));
Shader shader{shader_cache.GetStageProgram(program)};
@ -321,9 +339,6 @@ 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;
@ -634,26 +649,46 @@ void RasterizerOpenGL::DrawArrays() {
Maxwell::MaxShaderStage;
// Add space for at least 18 constant buffers
buffer_size +=
Maxwell::MaxConstBuffers * (MaxConstbufferSize + device.GetUniformBufferAlignment());
buffer_size += Maxwell::MaxConstBuffers *
(Maxwell::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();
SetupVertexBuffer(vao);
vertex_array_pushbuffer.Setup(vao);
DrawParameters params = SetupDraw();
// Upload vertex and index data.
SetupVertexBuffer(vao);
const GLintptr index_buffer_offset = SetupIndexBuffer();
// 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);
texture_cache.GuardSamplers(false);
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.
vertex_array_pushbuffer.Bind();
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();
@ -675,7 +710,7 @@ void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) {
return;
}
texture_cache.FlushRegion(addr, size);
global_cache.FlushRegion(addr, size);
buffer_cache.FlushRegion(addr, size);
}
void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
@ -685,7 +720,6 @@ void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
}
texture_cache.InvalidateRegion(addr, size);
shader_cache.InvalidateRegion(addr, size);
global_cache.InvalidateRegion(addr, size);
buffer_cache.InvalidateRegion(addr, size);
}
@ -696,6 +730,10 @@ 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) {
@ -739,11 +777,9 @@ void RasterizerOpenGL::SetupDrawConstBuffers(Tegra::Engines::Maxwell3D::Regs::Sh
MICROPROFILE_SCOPE(OpenGL_UBO);
const auto stage_index = static_cast<std::size_t>(stage);
const auto& shader_stage = system.GPU().Maxwell3D().state.shader_stages[stage_index];
const auto& entries = shader->GetShaderEntries().const_buffers;
// Upload only the enabled buffers from the 16 constbuffers of each shader stage
for (u32 bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
const auto& entry = entries[bindpoint];
for (const auto& entry : shader->GetShaderEntries().const_buffers) {
SetupConstBuffer(shader_stage.const_buffers[entry.GetIndex()], entry);
}
}
@ -752,46 +788,34 @@ void RasterizerOpenGL::SetupConstBuffer(const Tegra::Engines::ConstBufferInfo& b
const GLShader::ConstBufferEntry& entry) {
if (!buffer.enabled) {
// Set values to zero to unbind buffers
bind_ubo_pushbuffer.Push(0, 0, 0);
bind_ubo_pushbuffer.Push(buffer_cache.GetEmptyBuffer(sizeof(float)), 0, sizeof(float));
return;
}
std::size_t size;
if (entry.IsIndirect()) {
// Buffer is accessed indirectly, so upload the entire thing
size = buffer.size;
if (size > MaxConstbufferSize) {
LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", size,
MaxConstbufferSize);
size = MaxConstbufferSize;
}
} else {
// Buffer is accessed directly, upload just what we use
size = entry.GetSize();
}
// Align the actual size so it ends up being a multiple of vec4 to meet the OpenGL std140
// UBO alignment requirements.
size = Common::AlignUp(size, sizeof(GLvec4));
ASSERT_MSG(size <= MaxConstbufferSize, "Constant buffer is too big");
const std::size_t size = Common::AlignUp(GetConstBufferSize(buffer, entry), sizeof(GLvec4));
const std::size_t alignment = device.GetUniformBufferAlignment();
const GLintptr offset = buffer_cache.UploadMemory(buffer.address, size, alignment);
bind_ubo_pushbuffer.Push(buffer_cache.GetHandle(), offset, size);
const auto alignment = device.GetUniformBufferAlignment();
const auto [cbuf, offset] = buffer_cache.UploadMemory(buffer.address, size, alignment);
bind_ubo_pushbuffer.Push(cbuf, offset, size);
}
void RasterizerOpenGL::SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
const Shader& shader) {
const auto& entries = shader->GetShaderEntries().global_memory_entries;
for (std::size_t bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
const auto& entry{entries[bindpoint]};
const auto& region{global_cache.GetGlobalRegion(entry, stage)};
if (entry.IsWritten()) {
region->MarkAsModified(true, global_cache);
}
bind_ssbo_pushbuffer.Push(region->GetBufferHandle(), 0,
static_cast<GLsizeiptr>(region->GetSizeInBytes()));
auto& gpu{system.GPU()};
auto& memory_manager{gpu.MemoryManager()};
const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<std::size_t>(stage)]};
const auto alignment{device.GetShaderStorageBufferAlignment()};
for (const auto& entry : shader->GetShaderEntries().global_memory_entries) {
const auto addr{cbufs.const_buffers[entry.GetCbufIndex()].address + entry.GetCbufOffset()};
const auto actual_addr{memory_manager.Read<u64>(addr)};
const auto size{memory_manager.Read<u32>(addr + 8)};
const auto [ssbo, buffer_offset] =
buffer_cache.UploadMemory(actual_addr, size, alignment, true, entry.IsWritten());
bind_ssbo_pushbuffer.Push(ssbo, buffer_offset, static_cast<GLsizeiptr>(size));
}
}

View file

@ -24,7 +24,6 @@
#include "video_core/renderer_opengl/gl_buffer_cache.h"
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_framebuffer_cache.h"
#include "video_core/renderer_opengl/gl_global_cache.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_sampler_cache.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
@ -63,6 +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;
@ -73,11 +73,6 @@ public:
void LoadDiskResources(const std::atomic_bool& stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) override;
/// Maximum supported size that a constbuffer can have in bytes.
static constexpr std::size_t MaxConstbufferSize = 0x10000;
static_assert(MaxConstbufferSize % sizeof(GLvec4) == 0,
"The maximum size of a constbuffer must be a multiple of the size of GLvec4");
private:
struct FramebufferConfigState {
bool using_color_fb{};
@ -191,7 +186,6 @@ private:
TextureCacheOpenGL texture_cache;
ShaderCacheOpenGL shader_cache;
GlobalRegionCacheOpenGL global_cache;
SamplerCacheOpenGL sampler_cache;
FramebufferCacheOpenGL framebuffer_cache;
@ -210,6 +204,7 @@ private:
static constexpr std::size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024;
OGLBufferCache buffer_cache;
VertexArrayPushBuffer vertex_array_pushbuffer;
BindBuffersRangePushBuffer bind_ubo_pushbuffer{GL_UNIFORM_BUFFER};
BindBuffersRangePushBuffer bind_ssbo_pushbuffer{GL_SHADER_STORAGE_BUFFER};
@ -222,7 +217,9 @@ private:
void SetupVertexBuffer(GLuint vao);
DrawParameters SetupDraw();
GLintptr SetupIndexBuffer();
DrawParameters SetupDraw(GLintptr index_buffer_offset);
void SetupShaders(GLenum primitive_mode);

View file

@ -46,7 +46,7 @@ using TextureArgument = std::pair<Type, Node>;
using TextureIR = std::variant<TextureAoffi, TextureArgument>;
constexpr u32 MAX_CONSTBUFFER_ELEMENTS =
static_cast<u32>(RasterizerOpenGL::MaxConstbufferSize) / (4 * sizeof(float));
static_cast<u32>(Maxwell::MaxConstBufferSize) / (4 * sizeof(float));
class ShaderWriter {
public:

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);
}

View file

@ -13,29 +13,67 @@
namespace OpenGL {
VertexArrayPushBuffer::VertexArrayPushBuffer() = default;
VertexArrayPushBuffer::~VertexArrayPushBuffer() = default;
void VertexArrayPushBuffer::Setup(GLuint vao_) {
vao = vao_;
index_buffer = nullptr;
vertex_buffers.clear();
}
void VertexArrayPushBuffer::SetIndexBuffer(const GLuint* buffer) {
index_buffer = buffer;
}
void VertexArrayPushBuffer::SetVertexBuffer(GLuint binding_index, const GLuint* buffer,
GLintptr offset, GLsizei stride) {
vertex_buffers.push_back(Entry{binding_index, buffer, offset, stride});
}
void VertexArrayPushBuffer::Bind() {
if (index_buffer) {
glVertexArrayElementBuffer(vao, *index_buffer);
}
// TODO(Rodrigo): Find a way to ARB_multi_bind this
for (const auto& entry : vertex_buffers) {
glVertexArrayVertexBuffer(vao, entry.binding_index, *entry.buffer, entry.offset,
entry.stride);
}
}
BindBuffersRangePushBuffer::BindBuffersRangePushBuffer(GLenum target) : target{target} {}
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());
}

View file

@ -11,20 +11,49 @@
namespace OpenGL {
class BindBuffersRangePushBuffer {
class VertexArrayPushBuffer final {
public:
BindBuffersRangePushBuffer(GLenum target);
explicit VertexArrayPushBuffer();
~VertexArrayPushBuffer();
void Setup(GLuint vao_);
void SetIndexBuffer(const GLuint* buffer);
void SetVertexBuffer(GLuint binding_index, const GLuint* buffer, GLintptr offset,
GLsizei stride);
void Bind();
private:
struct Entry {
GLuint binding_index{};
const GLuint* buffer{};
GLintptr offset{};
GLsizei stride{};
};
GLuint vao{};
const GLuint* index_buffer{};
std::vector<Entry> vertex_buffers;
};
class BindBuffersRangePushBuffer final {
public:
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;