sneedmca/suyu
1
0
Fork 0
mirror of https://git.suyu.dev/suyu/suyu synced 2024-12-24 10:23:01 -06:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

texture_cache: Split texture cache into different files

Browse source
This commit is contained in:
ReinUsesLisp 2019-04-24 16:35:54 -03:00
parent 5f3aacdc37
commit 1b4503c571
12 changed files with 965 additions and 876 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
src/video_core/CMakeLists.txt
View file

@ -109,6 +109,13 @@ add_library(video_core STATIC
shader/track.cpp
surface.cpp
surface.h
texture_cache/surface_base.cpp
texture_cache/surface_base.h
texture_cache/surface_params.cpp
texture_cache/surface_params.h
texture_cache/surface_view.cpp
texture_cache/surface_view.h
texture_cache/texture_cache.h
textures/astc.cpp
textures/astc.h
textures/convert.cpp
@ -116,8 +123,6 @@ add_library(video_core STATIC
textures/decoders.cpp
textures/decoders.h
textures/texture.h
texture_cache.cpp
texture_cache.h
video_core.cpp
video_core.h
)

2
src/video_core/renderer_opengl/gl_texture_cache.cpp
View file

@ -9,7 +9,7 @@
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_texture_cache.h"
#include "video_core/renderer_opengl/utils.h"
#include "video_core/texture_cache.h"
#include "video_core/texture_cache/texture_cache_contextless.h"
#include "video_core/textures/convert.h"
#include "video_core/textures/texture.h"

2
src/video_core/renderer_opengl/gl_texture_cache.h
View file

@ -13,7 +13,7 @@
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/texture_cache.h"
#include "video_core/texture_cache/texture_cache_contextless.h"
namespace OpenGL {

750
src/video_core/texture_cache.h
View file

@ -1,750 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <list>
#include <memory>
#include <set>
#include <tuple>
#include <type_traits>
#include <unordered_map>
#include <boost/icl/interval_map.hpp>
#include <boost/range/iterator_range.hpp>
#include "common/assert.h"
#include "common/common_types.h"
#include "core/memory.h"
#include "video_core/engines/fermi_2d.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/gpu.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/surface.h"
namespace Core {
class System;
}
namespace Tegra::Texture {
struct FullTextureInfo;
}
namespace VideoCore {
class RasterizerInterface;
}
namespace VideoCommon {
class HasheableSurfaceParams {
public:
std::size_t Hash() const;
bool operator==(const HasheableSurfaceParams& rhs) const;
bool operator!=(const HasheableSurfaceParams& rhs) const {
return !operator==(rhs);
}
protected:
// Avoid creation outside of a managed environment.
HasheableSurfaceParams() = default;
bool is_tiled;
bool srgb_conversion;
u32 block_width;
u32 block_height;
u32 block_depth;
u32 tile_width_spacing;
u32 width;
u32 height;
u32 depth;
u32 pitch;
u32 unaligned_height;
u32 num_levels;
VideoCore::Surface::PixelFormat pixel_format;
VideoCore::Surface::ComponentType component_type;
VideoCore::Surface::SurfaceType type;
VideoCore::Surface::SurfaceTarget target;
};
class SurfaceParams final : public HasheableSurfaceParams {
public:
/// Creates SurfaceCachedParams from a texture configuration.
static SurfaceParams CreateForTexture(Core::System& system,
const Tegra::Texture::FullTextureInfo& config);
/// Creates SurfaceCachedParams for a depth buffer configuration.
static SurfaceParams CreateForDepthBuffer(
Core::System& system, u32 zeta_width, u32 zeta_height, Tegra::DepthFormat format,
u32 block_width, u32 block_height, u32 block_depth,
Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type);
/// Creates SurfaceCachedParams from a framebuffer configuration.
static SurfaceParams CreateForFramebuffer(Core::System& system, std::size_t index);
/// Creates SurfaceCachedParams from a Fermi2D surface configuration.
static SurfaceParams CreateForFermiCopySurface(
const Tegra::Engines::Fermi2D::Regs::Surface& config);
bool IsTiled() const {
return is_tiled;
}
bool GetSrgbConversion() const {
return srgb_conversion;
}
u32 GetBlockWidth() const {
return block_width;
}
u32 GetTileWidthSpacing() const {
return tile_width_spacing;
}
u32 GetWidth() const {
return width;
}
u32 GetHeight() const {
return height;
}
u32 GetDepth() const {
return depth;
}
u32 GetPitch() const {
return pitch;
}
u32 GetNumLevels() const {
return num_levels;
}
VideoCore::Surface::PixelFormat GetPixelFormat() const {
return pixel_format;
}
VideoCore::Surface::ComponentType GetComponentType() const {
return component_type;
}
VideoCore::Surface::SurfaceTarget GetTarget() const {
return target;
}
VideoCore::Surface::SurfaceType GetType() const {
return type;
}
std::size_t GetGuestSizeInBytes() const {
return guest_size_in_bytes;
}
std::size_t GetHostSizeInBytes() const {
return host_size_in_bytes;
}
u32 GetNumLayers() const {
return num_layers;
}
/// Returns the width of a given mipmap level.
u32 GetMipWidth(u32 level) const;
/// Returns the height of a given mipmap level.
u32 GetMipHeight(u32 level) const;
/// Returns the depth of a given mipmap level.
u32 GetMipDepth(u32 level) const;
/// Returns true if these parameters are from a layered surface.
bool IsLayered() const;
/// Returns the block height of a given mipmap level.
u32 GetMipBlockHeight(u32 level) const;
/// Returns the block depth of a given mipmap level.
u32 GetMipBlockDepth(u32 level) const;
/// Returns the offset in bytes in guest memory of a given mipmap level.
std::size_t GetGuestMipmapLevelOffset(u32 level) const;
/// Returns the offset in bytes in host memory (linear) of a given mipmap level.
std::size_t GetHostMipmapLevelOffset(u32 level) const;
/// Returns the size in bytes in host memory (linear) of a given mipmap level.
std::size_t GetHostMipmapSize(u32 level) const;
/// Returns the size of a layer in bytes in guest memory.
std::size_t GetGuestLayerSize() const;
/// Returns the size of a layer in bytes in host memory for a given mipmap level.
std::size_t GetHostLayerSize(u32 level) const;
/// Returns the default block width.
u32 GetDefaultBlockWidth() const;
/// Returns the default block height.
u32 GetDefaultBlockHeight() const;
/// Returns the bits per pixel.
u32 GetBitsPerPixel() const;
/// Returns the bytes per pixel.
u32 GetBytesPerPixel() const;
/// Returns true if another surface can be familiar with this. This is a loosely defined term
/// that reflects the possibility of these two surface parameters potentially being part of a
/// bigger superset.
bool IsFamiliar(const SurfaceParams& view_params) const;
/// Returns true if the pixel format is a depth and/or stencil format.
bool IsPixelFormatZeta() const;
/// Creates a map that redirects an address difference to a layer and mipmap level.
std::map<u64, std::pair<u32, u32>> CreateViewOffsetMap() const;
/// Returns true if the passed surface view parameters is equal or a valid subset of this.
bool IsViewValid(const SurfaceParams& view_params, u32 layer, u32 level) const;
private:
/// Calculates values that can be deduced from HasheableSurfaceParams.
void CalculateCachedValues();
/// Returns the size of a given mipmap level inside a layer.
std::size_t GetInnerMipmapMemorySize(u32 level, bool as_host_size, bool uncompressed) const;
/// Returns the size of all mipmap levels and aligns as needed.
std::size_t GetInnerMemorySize(bool as_host_size, bool layer_only, bool uncompressed) const;
/// Returns the size of a layer
std::size_t GetLayerSize(bool as_host_size, bool uncompressed) const;
/// Returns true if the passed view width and height match the size of this params in a given
/// mipmap level.
bool IsDimensionValid(const SurfaceParams& view_params, u32 level) const;
/// Returns true if the passed view depth match the size of this params in a given mipmap level.
bool IsDepthValid(const SurfaceParams& view_params, u32 level) const;
/// Returns true if the passed view layers and mipmap levels are in bounds.
bool IsInBounds(const SurfaceParams& view_params, u32 layer, u32 level) const;
std::size_t guest_size_in_bytes;
std::size_t host_size_in_bytes;
u32 num_layers;
};
struct ViewKey {
std::size_t Hash() const;
bool operator==(const ViewKey& rhs) const;
u32 base_layer{};
u32 num_layers{};
u32 base_level{};
u32 num_levels{};
};
} // namespace VideoCommon
namespace std {
template <>
struct hash<VideoCommon::SurfaceParams> {
std::size_t operator()(const VideoCommon::SurfaceParams& k) const noexcept {
return k.Hash();
}
};
template <>
struct hash<VideoCommon::ViewKey> {
std::size_t operator()(const VideoCommon::ViewKey& k) const noexcept {
return k.Hash();
}
};
} // namespace std
namespace VideoCommon {
class SurfaceBaseImpl {
public:
void LoadBuffer();
void FlushBuffer();
GPUVAddr GetGpuAddr() const {
ASSERT(is_registered);
return gpu_addr;
}
VAddr GetCpuAddr() const {
ASSERT(is_registered);
return cpu_addr;
}
u8* GetHostPtr() const {
ASSERT(is_registered);
return host_ptr;
}
CacheAddr GetCacheAddr() const {
ASSERT(is_registered);
return cache_addr;
}
const SurfaceParams& GetSurfaceParams() const {
return params;
}
void Register(GPUVAddr gpu_addr_, VAddr cpu_addr_, u8* host_ptr_) {
ASSERT(!is_registered);
is_registered = true;
gpu_addr = gpu_addr_;
cpu_addr = cpu_addr_;
host_ptr = host_ptr_;
cache_addr = ToCacheAddr(host_ptr_);
DecorateSurfaceName();
}
void Unregister() {
ASSERT(is_registered);
is_registered = false;
}
bool IsRegistered() const {
return is_registered;
}
std::size_t GetSizeInBytes() const {
return params.GetGuestSizeInBytes();
}
u8* GetStagingBufferLevelData(u32 level) {
return staging_buffer.data() + params.GetHostMipmapLevelOffset(level);
}
protected:
explicit SurfaceBaseImpl(const SurfaceParams& params);
~SurfaceBaseImpl(); // non-virtual is intended
virtual void DecorateSurfaceName() = 0;
const SurfaceParams params;
private:
GPUVAddr gpu_addr{};
VAddr cpu_addr{};
u8* host_ptr{};
CacheAddr cache_addr{};
bool is_registered{};
std::vector<u8> staging_buffer;
};
template <typename TTextureCache, typename TView, typename TExecutionContext>
class SurfaceBase : public SurfaceBaseImpl {
static_assert(std::is_trivially_copyable_v<TExecutionContext>);
public:
virtual TExecutionContext UploadTexture(TExecutionContext exctx) = 0;
virtual TExecutionContext DownloadTexture(TExecutionContext exctx) = 0;
TView* TryGetView(GPUVAddr view_addr, const SurfaceParams& view_params) {
if (view_addr < GetGpuAddr() || !params.IsFamiliar(view_params)) {
// It can't be a view if it's in a prior address.
return {};
}
const auto relative_offset{static_cast<u64>(view_addr - GetGpuAddr())};
const auto it{view_offset_map.find(relative_offset)};
if (it == view_offset_map.end()) {
// Couldn't find an aligned view.
return {};
}
const auto [layer, level] = it->second;
if (!params.IsViewValid(view_params, layer, level)) {
return {};
}
return GetView(layer, view_params.GetNumLayers(), level, view_params.GetNumLevels());
}
void MarkAsModified(bool is_modified_) {
is_modified = is_modified_;
if (is_modified_) {
modification_tick = texture_cache.Tick();
}
}
TView* GetView(GPUVAddr view_addr, const SurfaceParams& view_params) {
TView* view{TryGetView(view_addr, view_params)};
ASSERT(view != nullptr);
return view;
}
bool IsModified() const {
return is_modified;
}
u64 GetModificationTick() const {
return modification_tick;
}
protected:
explicit SurfaceBase(TTextureCache& texture_cache, const SurfaceParams& params)
: SurfaceBaseImpl{params}, texture_cache{texture_cache},
view_offset_map{params.CreateViewOffsetMap()} {}
~SurfaceBase() = default;
virtual std::unique_ptr<TView> CreateView(const ViewKey& view_key) = 0;
private:
TView* GetView(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels) {
const ViewKey key{base_layer, num_layers, base_level, num_levels};
const auto [entry, is_cache_miss] = views.try_emplace(key);
auto& view{entry->second};
if (is_cache_miss) {
view = CreateView(key);
}
return view.get();
}
TTextureCache& texture_cache;
const std::map<u64, std::pair<u32, u32>> view_offset_map;
bool is_modified{};
u64 modification_tick{};
std::unordered_map<ViewKey, std::unique_ptr<TView>> views;
};
template <typename TSurface, typename TView, typename TExecutionContext>
class TextureCache {
static_assert(std::is_trivially_copyable_v<TExecutionContext>);
using ResultType = std::tuple<TView*, TExecutionContext>;
using IntervalMap = boost::icl::interval_map<CacheAddr, std::set<std::shared_ptr<TSurface>>>;
using IntervalType = typename IntervalMap::interval_type;
public:
void InvalidateRegion(CacheAddr addr, std::size_t size) {
for (const auto& surface : GetSurfacesInRegion(addr, size)) {
if (!surface->IsRegistered()) {
// Skip duplicates
continue;
}
Unregister(surface);
}
}
ResultType GetTextureSurface(TExecutionContext exctx,
const Tegra::Texture::FullTextureInfo& config) {
const auto gpu_addr{config.tic.Address()};
if (!gpu_addr) {
return {{}, exctx};
}
const auto params{SurfaceParams::CreateForTexture(system, config)};
return GetSurfaceView(exctx, gpu_addr, params, true);
}
ResultType GetDepthBufferSurface(TExecutionContext exctx, bool preserve_contents) {
const auto& regs{system.GPU().Maxwell3D().regs};
const auto gpu_addr{regs.zeta.Address()};
if (!gpu_addr || !regs.zeta_enable) {
return {{}, exctx};
}
const auto depth_params{SurfaceParams::CreateForDepthBuffer(
system, regs.zeta_width, regs.zeta_height, regs.zeta.format,
regs.zeta.memory_layout.block_width, regs.zeta.memory_layout.block_height,
regs.zeta.memory_layout.block_depth, regs.zeta.memory_layout.type)};
return GetSurfaceView(exctx, gpu_addr, depth_params, preserve_contents);
}
ResultType GetColorBufferSurface(TExecutionContext exctx, std::size_t index,
bool preserve_contents) {
ASSERT(index < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets);
const auto& regs{system.GPU().Maxwell3D().regs};
if (index >= regs.rt_control.count || regs.rt[index].Address() == 0 ||
regs.rt[index].format == Tegra::RenderTargetFormat::NONE) {
return {{}, exctx};
}
auto& memory_manager{system.GPU().MemoryManager()};
const auto& config{system.GPU().Maxwell3D().regs.rt[index]};
const auto gpu_addr{config.Address() +
config.base_layer * config.layer_stride * sizeof(u32)};
if (!gpu_addr) {
return {{}, exctx};
}
return GetSurfaceView(exctx, gpu_addr, SurfaceParams::CreateForFramebuffer(system, index),
preserve_contents);
}
ResultType GetFermiSurface(TExecutionContext exctx,
const Tegra::Engines::Fermi2D::Regs::Surface& config) {
return GetSurfaceView(exctx, config.Address(),
SurfaceParams::CreateForFermiCopySurface(config), true);
}
std::shared_ptr<TSurface> TryFindFramebufferSurface(const u8* host_ptr) const {
const auto it{registered_surfaces.find(ToCacheAddr(host_ptr))};
return it != registered_surfaces.end() ? *it->second.begin() : nullptr;
}
u64 Tick() {
return ++ticks;
}
protected:
TextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer)
: system{system}, rasterizer{rasterizer} {}
~TextureCache() = default;
virtual ResultType TryFastGetSurfaceView(
TExecutionContext exctx, GPUVAddr gpu_addr, VAddr cpu_addr, u8* host_ptr,
const SurfaceParams& params, bool preserve_contents,
const std::vector<std::shared_ptr<TSurface>>& overlaps) = 0;
virtual std::shared_ptr<TSurface> CreateSurface(const SurfaceParams& params) = 0;
void Register(std::shared_ptr<TSurface> surface, GPUVAddr gpu_addr, VAddr cpu_addr,
u8* host_ptr) {
surface->Register(gpu_addr, cpu_addr, host_ptr);
registered_surfaces.add({GetSurfaceInterval(surface), {surface}});
rasterizer.UpdatePagesCachedCount(surface->GetCpuAddr(), surface->GetSizeInBytes(), 1);
}
void Unregister(std::shared_ptr<TSurface> surface) {
registered_surfaces.subtract({GetSurfaceInterval(surface), {surface}});
rasterizer.UpdatePagesCachedCount(surface->GetCpuAddr(), surface->GetSizeInBytes(), -1);
surface->Unregister();
}
std::shared_ptr<TSurface> GetUncachedSurface(const SurfaceParams& params) {
if (const auto surface = TryGetReservedSurface(params); surface)
return surface;
// No reserved surface available, create a new one and reserve it
auto new_surface{CreateSurface(params)};
ReserveSurface(params, new_surface);
return new_surface;
}
Core::System& system;
private:
ResultType GetSurfaceView(TExecutionContext exctx, GPUVAddr gpu_addr,
const SurfaceParams& params, bool preserve_contents) {
auto& memory_manager{system.GPU().MemoryManager()};
const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
DEBUG_ASSERT(cpu_addr);
const auto host_ptr{memory_manager.GetPointer(gpu_addr)};
const auto cache_addr{ToCacheAddr(host_ptr)};
auto overlaps{GetSurfacesInRegion(cache_addr, params.GetGuestSizeInBytes())};
if (overlaps.empty()) {
return LoadSurfaceView(exctx, gpu_addr, *cpu_addr, host_ptr, params, preserve_contents);
}
if (overlaps.size() == 1) {
if (TView* view = overlaps[0]->TryGetView(gpu_addr, params); view) {
return {view, exctx};
}
}
TView* fast_view;
std::tie(fast_view, exctx) = TryFastGetSurfaceView(exctx, gpu_addr, *cpu_addr, host_ptr,
params, preserve_contents, overlaps);
if (!fast_view) {
std::sort(overlaps.begin(), overlaps.end(), [](const auto& lhs, const auto& rhs) {
return lhs->GetModificationTick() < rhs->GetModificationTick();
});
}
for (const auto& surface : overlaps) {
if (!fast_view) {
// Flush even when we don't care about the contents, to preserve memory not
// written by the new surface.
exctx = FlushSurface(exctx, surface);
}
Unregister(surface);
}
if (fast_view) {
return {fast_view, exctx};
}
return LoadSurfaceView(exctx, gpu_addr, *cpu_addr, host_ptr, params, preserve_contents);
}
ResultType LoadSurfaceView(TExecutionContext exctx, GPUVAddr gpu_addr, VAddr cpu_addr,
u8* host_ptr, const SurfaceParams& params, bool preserve_contents) {
const auto new_surface{GetUncachedSurface(params)};
Register(new_surface, gpu_addr, cpu_addr, host_ptr);
if (preserve_contents) {
exctx = LoadSurface(exctx, new_surface);
}
return {new_surface->GetView(gpu_addr, params), exctx};
}
TExecutionContext LoadSurface(TExecutionContext exctx,
const std::shared_ptr<TSurface>& surface) {
surface->LoadBuffer();
exctx = surface->UploadTexture(exctx);
surface->MarkAsModified(false);
return exctx;
}
TExecutionContext FlushSurface(TExecutionContext exctx,
const std::shared_ptr<TSurface>& surface) {
if (!surface->IsModified()) {
return exctx;
}
exctx = surface->DownloadTexture(exctx);
surface->FlushBuffer();
return exctx;
}
std::vector<std::shared_ptr<TSurface>> GetSurfacesInRegion(CacheAddr cache_addr,
std::size_t size) const {
if (size == 0) {
return {};
}
const IntervalType interval{cache_addr, cache_addr + size};
std::vector<std::shared_ptr<TSurface>> surfaces;
for (auto& pair : boost::make_iterator_range(registered_surfaces.equal_range(interval))) {
surfaces.push_back(*pair.second.begin());
}
return surfaces;
}
void ReserveSurface(const SurfaceParams& params, std::shared_ptr<TSurface> surface) {
surface_reserve[params].push_back(std::move(surface));
}
std::shared_ptr<TSurface> TryGetReservedSurface(const SurfaceParams& params) {
auto search{surface_reserve.find(params)};
if (search == surface_reserve.end()) {
return {};
}
for (auto& surface : search->second) {
if (!surface->IsRegistered()) {
return surface;
}
}
return {};
}
IntervalType GetSurfaceInterval(std::shared_ptr<TSurface> surface) const {
return IntervalType::right_open(surface->GetCacheAddr(),
surface->GetCacheAddr() + surface->GetSizeInBytes());
}
VideoCore::RasterizerInterface& rasterizer;
u64 ticks{};
IntervalMap registered_surfaces;
/// The surface reserve is a "backup" cache, this is where we put unique surfaces that have
/// previously been used. This is to prevent surfaces from being constantly created and
/// destroyed when used with different surface parameters.
std::unordered_map<SurfaceParams, std::list<std::shared_ptr<TSurface>>> surface_reserve;
};
struct DummyExecutionContext {};
template <typename TSurface, typename TView>
class TextureCacheContextless : protected TextureCache<TSurface, TView, DummyExecutionContext> {
using Base = TextureCache<TSurface, TView, DummyExecutionContext>;
public:
void InvalidateRegion(CacheAddr addr, std::size_t size) {
Base::InvalidateRegion(addr, size);
}
TView* GetTextureSurface(const Tegra::Texture::FullTextureInfo& config) {
return RemoveContext(Base::GetTextureSurface({}, config));
}
TView* GetDepthBufferSurface(bool preserve_contents) {
return RemoveContext(Base::GetDepthBufferSurface({}, preserve_contents));
}
TView* GetColorBufferSurface(std::size_t index, bool preserve_contents) {
return RemoveContext(Base::GetColorBufferSurface({}, index, preserve_contents));
}
TView* GetFermiSurface(const Tegra::Engines::Fermi2D::Regs::Surface& config) {
return RemoveContext(Base::GetFermiSurface({}, config));
}
std::shared_ptr<TSurface> TryFindFramebufferSurface(const u8* host_ptr) const {
return Base::TryFindFramebufferSurface(host_ptr);
}
u64 Tick() {
return Base::Tick();
}
protected:
explicit TextureCacheContextless(Core::System& system,
VideoCore::RasterizerInterface& rasterizer)
: TextureCache<TSurface, TView, DummyExecutionContext>{system, rasterizer} {}
virtual TView* TryFastGetSurfaceView(
GPUVAddr gpu_addr, VAddr cpu_addr, u8* host_ptr, const SurfaceParams& params,
bool preserve_contents, const std::vector<std::shared_ptr<TSurface>>& overlaps) = 0;
private:
std::tuple<TView*, DummyExecutionContext> TryFastGetSurfaceView(
DummyExecutionContext, GPUVAddr gpu_addr, VAddr cpu_addr, u8* host_ptr,
const SurfaceParams& params, bool preserve_contents,
const std::vector<std::shared_ptr<TSurface>>& overlaps) {
return {TryFastGetSurfaceView(gpu_addr, cpu_addr, host_ptr, params, preserve_contents,
overlaps),
{}};
}
TView* RemoveContext(std::tuple<TView*, DummyExecutionContext> return_value) {
const auto [view, exctx] = return_value;
return view;
}
};
template <typename TTextureCache, typename TView>
class SurfaceBaseContextless : public SurfaceBase<TTextureCache, TView, DummyExecutionContext> {
public:
DummyExecutionContext DownloadTexture(DummyExecutionContext) {
DownloadTextureImpl();
return {};
}
DummyExecutionContext UploadTexture(DummyExecutionContext) {
UploadTextureImpl();
return {};
}
protected:
explicit SurfaceBaseContextless(TTextureCache& texture_cache, const SurfaceParams& params)
: SurfaceBase<TTextureCache, TView, DummyExecutionContext>{texture_cache, params} {}
virtual void DownloadTextureImpl() = 0;
virtual void UploadTextureImpl() = 0;
};
} // namespace VideoCommon

118
src/video_core/texture_cache/surface_base.cpp Normal file
View file

@ -0,0 +1,118 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/morton.h"
#include "video_core/texture_cache/surface_base.h"
#include "video_core/texture_cache/surface_params.h"
#include "video_core/textures/convert.h"
namespace VideoCommon {
using Tegra::Texture::ConvertFromGuestToHost;
using VideoCore::MortonSwizzleMode;
namespace {
void SwizzleFunc(MortonSwizzleMode mode, u8* memory, const SurfaceParams& params, u8* buffer,
u32 level) {
const u32 width{params.GetMipWidth(level)};
const u32 height{params.GetMipHeight(level)};
const u32 block_height{params.GetMipBlockHeight(level)};
const u32 block_depth{params.GetMipBlockDepth(level)};
std::size_t guest_offset{params.GetGuestMipmapLevelOffset(level)};
if (params.IsLayered()) {
std::size_t host_offset{0};
const std::size_t guest_stride = params.GetGuestLayerSize();
const std::size_t host_stride = params.GetHostLayerSize(level);
for (u32 layer = 0; layer < params.GetNumLayers(); layer++) {
MortonSwizzle(mode, params.GetPixelFormat(), width, block_height, height, block_depth,
1, params.GetTileWidthSpacing(), buffer + host_offset,
memory + guest_offset);
guest_offset += guest_stride;
host_offset += host_stride;
}
} else {
MortonSwizzle(mode, params.GetPixelFormat(), width, block_height, height, block_depth,
params.GetMipDepth(level), params.GetTileWidthSpacing(), buffer,
memory + guest_offset);
}
}
} // Anonymous namespace
SurfaceBaseImpl::SurfaceBaseImpl(const SurfaceParams& params) : params{params} {
staging_buffer.resize(params.GetHostSizeInBytes());
}
SurfaceBaseImpl::~SurfaceBaseImpl() = default;
void SurfaceBaseImpl::LoadBuffer() {
if (params.IsTiled()) {
ASSERT_MSG(params.GetBlockWidth() == 1, "Block width is defined as {} on texture target {}",
params.GetBlockWidth(), static_cast<u32>(params.GetTarget()));
for (u32 level = 0; level < params.GetNumLevels(); ++level) {
u8* const buffer{GetStagingBufferLevelData(level)};
SwizzleFunc(MortonSwizzleMode::MortonToLinear, host_ptr, params, buffer, level);
}
} else {
ASSERT_MSG(params.GetNumLevels() == 1, "Linear mipmap loading is not implemented");
const u32 bpp{GetFormatBpp(params.GetPixelFormat()) / CHAR_BIT};
const u32 block_width{params.GetDefaultBlockWidth()};
const u32 block_height{params.GetDefaultBlockHeight()};
const u32 width{(params.GetWidth() + block_width - 1) / block_width};
const u32 height{(params.GetHeight() + block_height - 1) / block_height};
const u32 copy_size{width * bpp};
if (params.GetPitch() == copy_size) {
std::memcpy(staging_buffer.data(), host_ptr, params.GetHostSizeInBytes());
} else {
const u8* start{host_ptr};
u8* write_to{staging_buffer.data()};
for (u32 h = height; h > 0; --h) {
std::memcpy(write_to, start, copy_size);
start += params.GetPitch();
write_to += copy_size;
}
}
}
for (u32 level = 0; level < params.GetNumLevels(); ++level) {
ConvertFromGuestToHost(GetStagingBufferLevelData(level), params.GetPixelFormat(),
params.GetMipWidth(level), params.GetMipHeight(level),
params.GetMipDepth(level), true, true);
}
}
void SurfaceBaseImpl::FlushBuffer() {
if (params.IsTiled()) {
ASSERT_MSG(params.GetBlockWidth() == 1, "Block width is defined as {}",
params.GetBlockWidth());
for (u32 level = 0; level < params.GetNumLevels(); ++level) {
u8* const buffer = GetStagingBufferLevelData(level);
SwizzleFunc(MortonSwizzleMode::LinearToMorton, GetHostPtr(), params, buffer, level);
}
} else {
UNIMPLEMENTED();
/*
ASSERT(params.GetTarget() == SurfaceTarget::Texture2D);
ASSERT(params.GetNumLevels() == 1);
const u32 bpp{params.GetFormatBpp() / 8};
const u32 copy_size{params.GetWidth() * bpp};
if (params.GetPitch() == copy_size) {
std::memcpy(host_ptr, staging_buffer.data(), GetSizeInBytes());
} else {
u8* start{host_ptr};
const u8* read_to{staging_buffer.data()};
for (u32 h = params.GetHeight(); h > 0; --h) {
std::memcpy(start, read_to, copy_size);
start += params.GetPitch();
read_to += copy_size;
}
}
*/
}
}
} // namespace VideoCommon

172
src/video_core/texture_cache/surface_base.h Normal file
View file

@ -0,0 +1,172 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <unordered_map>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/gpu.h"
#include "video_core/texture_cache/surface_params.h"
#include "video_core/texture_cache/surface_view.h"
namespace VideoCommon {
class SurfaceBaseImpl {
public:
void LoadBuffer();
void FlushBuffer();
GPUVAddr GetGpuAddr() const {
ASSERT(is_registered);
return gpu_addr;
}
VAddr GetCpuAddr() const {
ASSERT(is_registered);
return cpu_addr;
}
u8* GetHostPtr() const {
ASSERT(is_registered);
return host_ptr;
}
CacheAddr GetCacheAddr() const {
ASSERT(is_registered);
return cache_addr;
}
const SurfaceParams& GetSurfaceParams() const {
return params;
}
void Register(GPUVAddr gpu_addr_, VAddr cpu_addr_, u8* host_ptr_) {
ASSERT(!is_registered);
is_registered = true;
gpu_addr = gpu_addr_;
cpu_addr = cpu_addr_;
host_ptr = host_ptr_;
cache_addr = ToCacheAddr(host_ptr_);
DecorateSurfaceName();
}
void Unregister() {
ASSERT(is_registered);
is_registered = false;
}
bool IsRegistered() const {
return is_registered;
}
std::size_t GetSizeInBytes() const {
return params.GetGuestSizeInBytes();
}
u8* GetStagingBufferLevelData(u32 level) {
return staging_buffer.data() + params.GetHostMipmapLevelOffset(level);
}
protected:
explicit SurfaceBaseImpl(const SurfaceParams& params);
~SurfaceBaseImpl(); // non-virtual is intended
virtual void DecorateSurfaceName() = 0;
const SurfaceParams params;
private:
GPUVAddr gpu_addr{};
VAddr cpu_addr{};
u8* host_ptr{};
CacheAddr cache_addr{};
bool is_registered{};
std::vector<u8> staging_buffer;
};
template <typename TTextureCache, typename TView, typename TExecutionContext>
class SurfaceBase : public SurfaceBaseImpl {
static_assert(std::is_trivially_copyable_v<TExecutionContext>);
public:
virtual TExecutionContext UploadTexture(TExecutionContext exctx) = 0;
virtual TExecutionContext DownloadTexture(TExecutionContext exctx) = 0;
TView* TryGetView(GPUVAddr view_addr, const SurfaceParams& view_params) {
if (view_addr < GetGpuAddr() || !params.IsFamiliar(view_params)) {
// It can't be a view if it's in a prior address.
return {};
}
const auto relative_offset{static_cast<u64>(view_addr - GetGpuAddr())};
const auto it{view_offset_map.find(relative_offset)};
if (it == view_offset_map.end()) {
// Couldn't find an aligned view.
return {};
}
const auto [layer, level] = it->second;
if (!params.IsViewValid(view_params, layer, level)) {
return {};
}
return GetView(layer, view_params.GetNumLayers(), level, view_params.GetNumLevels());
}
void MarkAsModified(bool is_modified_) {
is_modified = is_modified_;
if (is_modified_) {
modification_tick = texture_cache.Tick();
}
}
TView* GetView(GPUVAddr view_addr, const SurfaceParams& view_params) {
TView* view{TryGetView(view_addr, view_params)};
ASSERT(view != nullptr);
return view;
}
bool IsModified() const {
return is_modified;
}
u64 GetModificationTick() const {
return modification_tick;
}
protected:
explicit SurfaceBase(TTextureCache& texture_cache, const SurfaceParams& params)
: SurfaceBaseImpl{params}, texture_cache{texture_cache},
view_offset_map{params.CreateViewOffsetMap()} {}
~SurfaceBase() = default;
virtual std::unique_ptr<TView> CreateView(const ViewKey& view_key) = 0;
private:
TView* GetView(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels) {
const ViewKey key{base_layer, num_layers, base_level, num_levels};
const auto [entry, is_cache_miss] = views.try_emplace(key);
auto& view{entry->second};
if (is_cache_miss) {
view = CreateView(key);
}
return view.get();
}
TTextureCache& texture_cache;
const std::map<u64, std::pair<u32, u32>> view_offset_map;
std::unordered_map<ViewKey, std::unique_ptr<TView>> views;
bool is_modified{};
u64 modification_tick{};
};
} // namespace VideoCommon

126
src/video_core/texture_cache.cpp → src/video_core/texture_cache/surface_params.cpp
View file

@ -2,22 +2,17 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/alignment.h"
#include "common/assert.h"
#include <map>
#include "common/cityhash.h"
#include "common/common_types.h"
#include "common/alignment.h"
#include "core/core.h"
#include "video_core/morton.h"
#include "video_core/surface.h"
#include "video_core/texture_cache.h"
#include "video_core/textures/convert.h"
#include "video_core/texture_cache/surface_params.h"
#include "video_core/textures/decoders.h"
#include "video_core/textures/texture.h"
namespace VideoCommon {
using VideoCore::MortonSwizzleMode;
using VideoCore::Surface::ComponentTypeFromDepthFormat;
using VideoCore::Surface::ComponentTypeFromRenderTarget;
using VideoCore::Surface::ComponentTypeFromTexture;
@ -27,115 +22,12 @@ using VideoCore::Surface::PixelFormatFromTextureFormat;
using VideoCore::Surface::SurfaceTarget;
using VideoCore::Surface::SurfaceTargetFromTextureType;
using Tegra::Texture::ConvertFromGuestToHost;
namespace {
constexpr u32 GetMipmapSize(bool uncompressed, u32 mip_size, u32 tile) {
return uncompressed ? mip_size : std::max(1U, (mip_size + tile - 1) / tile);
}
void SwizzleFunc(MortonSwizzleMode mode, u8* memory, const SurfaceParams& params, u8* buffer,
u32 level) {
const u32 width{params.GetMipWidth(level)};
const u32 height{params.GetMipHeight(level)};
const u32 block_height{params.GetMipBlockHeight(level)};
const u32 block_depth{params.GetMipBlockDepth(level)};
std::size_t guest_offset{params.GetGuestMipmapLevelOffset(level)};
if (params.IsLayered()) {
std::size_t host_offset{0};
const std::size_t guest_stride = params.GetGuestLayerSize();
const std::size_t host_stride = params.GetHostLayerSize(level);
for (u32 layer = 0; layer < params.GetNumLayers(); layer++) {
MortonSwizzle(mode, params.GetPixelFormat(), width, block_height, height, block_depth,
1, params.GetTileWidthSpacing(), buffer + host_offset,
memory + guest_offset);
guest_offset += guest_stride;
host_offset += host_stride;
}
} else {
MortonSwizzle(mode, params.GetPixelFormat(), width, block_height, height, block_depth,
params.GetMipDepth(level), params.GetTileWidthSpacing(), buffer,
memory + guest_offset);
}
}
} // Anonymous namespace
SurfaceBaseImpl::SurfaceBaseImpl(const SurfaceParams& params) : params{params} {
staging_buffer.resize(params.GetHostSizeInBytes());
}
SurfaceBaseImpl::~SurfaceBaseImpl() = default;
void SurfaceBaseImpl::LoadBuffer() {
if (params.IsTiled()) {
ASSERT_MSG(params.GetBlockWidth() == 1, "Block width is defined as {} on texture target {}",
params.GetBlockWidth(), static_cast<u32>(params.GetTarget()));
for (u32 level = 0; level < params.GetNumLevels(); ++level) {
u8* const buffer{GetStagingBufferLevelData(level)};
SwizzleFunc(MortonSwizzleMode::MortonToLinear, host_ptr, params, buffer, level);
}
} else {
ASSERT_MSG(params.GetNumLevels() == 1, "Linear mipmap loading is not implemented");
const u32 bpp{GetFormatBpp(params.GetPixelFormat()) / CHAR_BIT};
const u32 block_width{params.GetDefaultBlockWidth()};
const u32 block_height{params.GetDefaultBlockHeight()};
const u32 width{(params.GetWidth() + block_width - 1) / block_width};
const u32 height{(params.GetHeight() + block_height - 1) / block_height};
const u32 copy_size{width * bpp};
if (params.GetPitch() == copy_size) {
std::memcpy(staging_buffer.data(), host_ptr, params.GetHostSizeInBytes());
} else {
const u8* start{host_ptr};
u8* write_to{staging_buffer.data()};
for (u32 h = height; h > 0; --h) {
std::memcpy(write_to, start, copy_size);
start += params.GetPitch();
write_to += copy_size;
}
}
}
for (u32 level = 0; level < params.GetNumLevels(); ++level) {
ConvertFromGuestToHost(GetStagingBufferLevelData(level), params.GetPixelFormat(),
params.GetMipWidth(level), params.GetMipHeight(level),
params.GetMipDepth(level), true, true);
}
}
void SurfaceBaseImpl::FlushBuffer() {
if (params.IsTiled()) {
ASSERT_MSG(params.GetBlockWidth() == 1, "Block width is defined as {}",
params.GetBlockWidth());
for (u32 level = 0; level < params.GetNumLevels(); ++level) {
u8* const buffer = GetStagingBufferLevelData(level);
SwizzleFunc(MortonSwizzleMode::LinearToMorton, GetHostPtr(), params, buffer, level);
}
} else {
UNIMPLEMENTED();
/*
ASSERT(params.GetTarget() == SurfaceTarget::Texture2D);
ASSERT(params.GetNumLevels() == 1);
const u32 bpp{params.GetFormatBpp() / 8};
const u32 copy_size{params.GetWidth() * bpp};
if (params.GetPitch() == copy_size) {
std::memcpy(host_ptr, staging_buffer.data(), GetSizeInBytes());
} else {
u8* start{host_ptr};
const u8* read_to{staging_buffer.data()};
for (u32 h = params.GetHeight(); h > 0; --h) {
std::memcpy(start, read_to, copy_size);
start += params.GetPitch();
read_to += copy_size;
}
}
*/
}
}
SurfaceParams SurfaceParams::CreateForTexture(Core::System& system,
const Tegra::Texture::FullTextureInfo& config) {
SurfaceParams params;
@ -517,14 +409,4 @@ bool HasheableSurfaceParams::operator==(const HasheableSurfaceParams& rhs) const
rhs.type, rhs.target);
}
std::size_t ViewKey::Hash() const {
return static_cast<std::size_t>(
Common::CityHash64(reinterpret_cast<const char*>(this), sizeof(*this)));
}
bool ViewKey::operator==(const ViewKey& rhs) const {
return std::tie(base_layer, num_layers, base_level, num_levels) ==
std::tie(rhs.base_layer, rhs.num_layers, rhs.base_level, rhs.num_levels);
}
} // namespace VideoCommon

229
src/video_core/texture_cache/surface_params.h Normal file
View file

@ -0,0 +1,229 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <map>
#include "common/common_types.h"
#include "video_core/engines/fermi_2d.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/surface.h"
namespace VideoCommon {
class HasheableSurfaceParams {
public:
std::size_t Hash() const;
bool operator==(const HasheableSurfaceParams& rhs) const;
bool operator!=(const HasheableSurfaceParams& rhs) const {
return !operator==(rhs);
}
protected:
// Avoid creation outside of a managed environment.
HasheableSurfaceParams() = default;
bool is_tiled;
bool srgb_conversion;
u32 block_width;
u32 block_height;
u32 block_depth;
u32 tile_width_spacing;
u32 width;
u32 height;
u32 depth;
u32 pitch;
u32 unaligned_height;
u32 num_levels;
VideoCore::Surface::PixelFormat pixel_format;
VideoCore::Surface::ComponentType component_type;
VideoCore::Surface::SurfaceType type;
VideoCore::Surface::SurfaceTarget target;
};
class SurfaceParams final : public HasheableSurfaceParams {
public:
/// Creates SurfaceCachedParams from a texture configuration.
static SurfaceParams CreateForTexture(Core::System& system,
const Tegra::Texture::FullTextureInfo& config);
/// Creates SurfaceCachedParams for a depth buffer configuration.
static SurfaceParams CreateForDepthBuffer(
Core::System& system, u32 zeta_width, u32 zeta_height, Tegra::DepthFormat format,
u32 block_width, u32 block_height, u32 block_depth,
Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type);
/// Creates SurfaceCachedParams from a framebuffer configuration.
static SurfaceParams CreateForFramebuffer(Core::System& system, std::size_t index);
/// Creates SurfaceCachedParams from a Fermi2D surface configuration.
static SurfaceParams CreateForFermiCopySurface(
const Tegra::Engines::Fermi2D::Regs::Surface& config);
bool IsTiled() const {
return is_tiled;
}
bool GetSrgbConversion() const {
return srgb_conversion;
}
u32 GetBlockWidth() const {
return block_width;
}
u32 GetTileWidthSpacing() const {
return tile_width_spacing;
}
u32 GetWidth() const {
return width;
}
u32 GetHeight() const {
return height;
}
u32 GetDepth() const {
return depth;
}
u32 GetPitch() const {
return pitch;
}
u32 GetNumLevels() const {
return num_levels;
}
VideoCore::Surface::PixelFormat GetPixelFormat() const {
return pixel_format;
}
VideoCore::Surface::ComponentType GetComponentType() const {
return component_type;
}
VideoCore::Surface::SurfaceTarget GetTarget() const {
return target;
}
VideoCore::Surface::SurfaceType GetType() const {
return type;
}
std::size_t GetGuestSizeInBytes() const {
return guest_size_in_bytes;
}
std::size_t GetHostSizeInBytes() const {
return host_size_in_bytes;
}
u32 GetNumLayers() const {
return num_layers;
}
/// Returns the width of a given mipmap level.
u32 GetMipWidth(u32 level) const;
/// Returns the height of a given mipmap level.
u32 GetMipHeight(u32 level) const;
/// Returns the depth of a given mipmap level.
u32 GetMipDepth(u32 level) const;
/// Returns true if these parameters are from a layered surface.
bool IsLayered() const;
/// Returns the block height of a given mipmap level.
u32 GetMipBlockHeight(u32 level) const;
/// Returns the block depth of a given mipmap level.
u32 GetMipBlockDepth(u32 level) const;
/// Returns the offset in bytes in guest memory of a given mipmap level.
std::size_t GetGuestMipmapLevelOffset(u32 level) const;
/// Returns the offset in bytes in host memory (linear) of a given mipmap level.
std::size_t GetHostMipmapLevelOffset(u32 level) const;
/// Returns the size in bytes in host memory (linear) of a given mipmap level.
std::size_t GetHostMipmapSize(u32 level) const;
/// Returns the size of a layer in bytes in guest memory.
std::size_t GetGuestLayerSize() const;
/// Returns the size of a layer in bytes in host memory for a given mipmap level.
std::size_t GetHostLayerSize(u32 level) const;
/// Returns the default block width.
u32 GetDefaultBlockWidth() const;
/// Returns the default block height.
u32 GetDefaultBlockHeight() const;
/// Returns the bits per pixel.
u32 GetBitsPerPixel() const;
/// Returns the bytes per pixel.
u32 GetBytesPerPixel() const;
/// Returns true if another surface can be familiar with this. This is a loosely defined term
/// that reflects the possibility of these two surface parameters potentially being part of a
/// bigger superset.
bool IsFamiliar(const SurfaceParams& view_params) const;
/// Returns true if the pixel format is a depth and/or stencil format.
bool IsPixelFormatZeta() const;
/// Creates a map that redirects an address difference to a layer and mipmap level.
std::map<u64, std::pair<u32, u32>> CreateViewOffsetMap() const;
/// Returns true if the passed surface view parameters is equal or a valid subset of this.
bool IsViewValid(const SurfaceParams& view_params, u32 layer, u32 level) const;
private:
/// Calculates values that can be deduced from HasheableSurfaceParams.
void CalculateCachedValues();
/// Returns the size of a given mipmap level inside a layer.
std::size_t GetInnerMipmapMemorySize(u32 level, bool as_host_size, bool uncompressed) const;
/// Returns the size of all mipmap levels and aligns as needed.
std::size_t GetInnerMemorySize(bool as_host_size, bool layer_only, bool uncompressed) const;
/// Returns the size of a layer
std::size_t GetLayerSize(bool as_host_size, bool uncompressed) const;
/// Returns true if the passed view width and height match the size of this params in a given
/// mipmap level.
bool IsDimensionValid(const SurfaceParams& view_params, u32 level) const;
/// Returns true if the passed view depth match the size of this params in a given mipmap level.
bool IsDepthValid(const SurfaceParams& view_params, u32 level) const;
/// Returns true if the passed view layers and mipmap levels are in bounds.
bool IsInBounds(const SurfaceParams& view_params, u32 layer, u32 level) const;
std::size_t guest_size_in_bytes;
std::size_t host_size_in_bytes;
u32 num_layers;
};
} // namespace VideoCommon
namespace std {
template <>
struct hash<VideoCommon::SurfaceParams> {
std::size_t operator()(const VideoCommon::SurfaceParams& k) const noexcept {
return k.Hash();
}
};
} // namespace std

23
src/video_core/texture_cache/surface_view.cpp Normal file
View file

@ -0,0 +1,23 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <tuple>
#include "common/common_types.h"
#include "video_core/texture_cache/surface_view.h"
namespace VideoCommon {
std::size_t ViewKey::Hash() const {
return static_cast<std::size_t>(base_layer) ^ static_cast<std::size_t>(num_layers << 16) ^
(static_cast<std::size_t>(base_level) << 32) ^
(static_cast<std::size_t>(num_levels) << 48);
}
bool ViewKey::operator==(const ViewKey& rhs) const {
return std::tie(base_layer, num_layers, base_level, num_levels) ==
std::tie(rhs.base_layer, rhs.num_layers, rhs.base_level, rhs.num_levels);
}
} // namespace VideoCommon

35
src/video_core/texture_cache/surface_view.h Normal file
View file

@ -0,0 +1,35 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <functional>
#include "common/common_types.h"
namespace VideoCommon {
struct ViewKey {
std::size_t Hash() const;
bool operator==(const ViewKey& rhs) const;
u32 base_layer{};
u32 num_layers{};
u32 base_level{};
u32 num_levels{};
};
} // namespace VideoCommon
namespace std {
template <>
struct hash<VideoCommon::ViewKey> {
std::size_t operator()(const VideoCommon::ViewKey& k) const noexcept {
return k.Hash();
}
};
} // namespace std

282
src/video_core/texture_cache/texture_cache.h Normal file
View file

@ -0,0 +1,282 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <list>
#include <memory>
#include <set>
#include <tuple>
#include <type_traits>
#include <unordered_map>
#include <boost/icl/interval_map.hpp>
#include <boost/range/iterator_range.hpp>
#include "common/assert.h"
#include "common/common_types.h"
#include "core/memory.h"
#include "video_core/engines/fermi_2d.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/gpu.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/surface.h"
#include "video_core/texture_cache/surface_base.h"
#include "video_core/texture_cache/surface_params.h"
#include "video_core/texture_cache/surface_view.h"
namespace Core {
class System;
}
namespace Tegra::Texture {
struct FullTextureInfo;
}
namespace VideoCore {
class RasterizerInterface;
}
namespace VideoCommon {
template <typename TSurface, typename TView, typename TExecutionContext>
class TextureCache {
static_assert(std::is_trivially_copyable_v<TExecutionContext>);
using ResultType = std::tuple<TView*, TExecutionContext>;
using IntervalMap = boost::icl::interval_map<CacheAddr, std::set<std::shared_ptr<TSurface>>>;
using IntervalType = typename IntervalMap::interval_type;
public:
void InvalidateRegion(CacheAddr addr, std::size_t size) {
for (const auto& surface : GetSurfacesInRegion(addr, size)) {
if (!surface->IsRegistered()) {
// Skip duplicates
continue;
}
Unregister(surface);
}
}
ResultType GetTextureSurface(TExecutionContext exctx,
const Tegra::Texture::FullTextureInfo& config) {
const auto gpu_addr{config.tic.Address()};
if (!gpu_addr) {
return {{}, exctx};
}
const auto params{SurfaceParams::CreateForTexture(system, config)};
return GetSurfaceView(exctx, gpu_addr, params, true);
}
ResultType GetDepthBufferSurface(TExecutionContext exctx, bool preserve_contents) {
const auto& regs{system.GPU().Maxwell3D().regs};
const auto gpu_addr{regs.zeta.Address()};
if (!gpu_addr || !regs.zeta_enable) {
return {{}, exctx};
}
const auto depth_params{SurfaceParams::CreateForDepthBuffer(
system, regs.zeta_width, regs.zeta_height, regs.zeta.format,
regs.zeta.memory_layout.block_width, regs.zeta.memory_layout.block_height,
regs.zeta.memory_layout.block_depth, regs.zeta.memory_layout.type)};
return GetSurfaceView(exctx, gpu_addr, depth_params, preserve_contents);
}
ResultType GetColorBufferSurface(TExecutionContext exctx, std::size_t index,
bool preserve_contents) {
ASSERT(index < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets);
const auto& regs{system.GPU().Maxwell3D().regs};
if (index >= regs.rt_control.count || regs.rt[index].Address() == 0 ||
regs.rt[index].format == Tegra::RenderTargetFormat::NONE) {
return {{}, exctx};
}
auto& memory_manager{system.GPU().MemoryManager()};
const auto& config{system.GPU().Maxwell3D().regs.rt[index]};
const auto gpu_addr{config.Address() +
config.base_layer * config.layer_stride * sizeof(u32)};
if (!gpu_addr) {
return {{}, exctx};
}
return GetSurfaceView(exctx, gpu_addr, SurfaceParams::CreateForFramebuffer(system, index),
preserve_contents);
}
ResultType GetFermiSurface(TExecutionContext exctx,
const Tegra::Engines::Fermi2D::Regs::Surface& config) {
return GetSurfaceView(exctx, config.Address(),
SurfaceParams::CreateForFermiCopySurface(config), true);
}
std::shared_ptr<TSurface> TryFindFramebufferSurface(const u8* host_ptr) const {
const auto it{registered_surfaces.find(ToCacheAddr(host_ptr))};
return it != registered_surfaces.end() ? *it->second.begin() : nullptr;
}
u64 Tick() {
return ++ticks;
}
protected:
TextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer)
: system{system}, rasterizer{rasterizer} {}
~TextureCache() = default;
virtual ResultType TryFastGetSurfaceView(
TExecutionContext exctx, GPUVAddr gpu_addr, VAddr cpu_addr, u8* host_ptr,
const SurfaceParams& params, bool preserve_contents,
const std::vector<std::shared_ptr<TSurface>>& overlaps) = 0;
virtual std::shared_ptr<TSurface> CreateSurface(const SurfaceParams& params) = 0;
void Register(std::shared_ptr<TSurface> surface, GPUVAddr gpu_addr, VAddr cpu_addr,
u8* host_ptr) {
surface->Register(gpu_addr, cpu_addr, host_ptr);
registered_surfaces.add({GetSurfaceInterval(surface), {surface}});
rasterizer.UpdatePagesCachedCount(surface->GetCpuAddr(), surface->GetSizeInBytes(), 1);
}
void Unregister(std::shared_ptr<TSurface> surface) {
registered_surfaces.subtract({GetSurfaceInterval(surface), {surface}});
rasterizer.UpdatePagesCachedCount(surface->GetCpuAddr(), surface->GetSizeInBytes(), -1);
surface->Unregister();
}
std::shared_ptr<TSurface> GetUncachedSurface(const SurfaceParams& params) {
if (const auto surface = TryGetReservedSurface(params); surface)
return surface;
// No reserved surface available, create a new one and reserve it
auto new_surface{CreateSurface(params)};
ReserveSurface(params, new_surface);
return new_surface;
}
Core::System& system;
private:
ResultType GetSurfaceView(TExecutionContext exctx, GPUVAddr gpu_addr,
const SurfaceParams& params, bool preserve_contents) {
auto& memory_manager{system.GPU().MemoryManager()};
const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
DEBUG_ASSERT(cpu_addr);
const auto host_ptr{memory_manager.GetPointer(gpu_addr)};
const auto cache_addr{ToCacheAddr(host_ptr)};
auto overlaps{GetSurfacesInRegion(cache_addr, params.GetGuestSizeInBytes())};
if (overlaps.empty()) {
return LoadSurfaceView(exctx, gpu_addr, *cpu_addr, host_ptr, params, preserve_contents);
}
if (overlaps.size() == 1) {
if (TView* view = overlaps[0]->TryGetView(gpu_addr, params); view) {
return {view, exctx};
}
}
TView* fast_view;
std::tie(fast_view, exctx) = TryFastGetSurfaceView(exctx, gpu_addr, *cpu_addr, host_ptr,
params, preserve_contents, overlaps);
if (!fast_view) {
std::sort(overlaps.begin(), overlaps.end(), [](const auto& lhs, const auto& rhs) {
return lhs->GetModificationTick() < rhs->GetModificationTick();
});
}
for (const auto& surface : overlaps) {
if (!fast_view) {
// Flush even when we don't care about the contents, to preserve memory not
// written by the new surface.
exctx = FlushSurface(exctx, surface);
}
Unregister(surface);
}
if (fast_view) {
return {fast_view, exctx};
}
return LoadSurfaceView(exctx, gpu_addr, *cpu_addr, host_ptr, params, preserve_contents);
}
ResultType LoadSurfaceView(TExecutionContext exctx, GPUVAddr gpu_addr, VAddr cpu_addr,
u8* host_ptr, const SurfaceParams& params, bool preserve_contents) {
const auto new_surface{GetUncachedSurface(params)};
Register(new_surface, gpu_addr, cpu_addr, host_ptr);
if (preserve_contents) {
exctx = LoadSurface(exctx, new_surface);
}
return {new_surface->GetView(gpu_addr, params), exctx};
}
TExecutionContext LoadSurface(TExecutionContext exctx,
const std::shared_ptr<TSurface>& surface) {
surface->LoadBuffer();
exctx = surface->UploadTexture(exctx);
surface->MarkAsModified(false);
return exctx;
}
TExecutionContext FlushSurface(TExecutionContext exctx,
const std::shared_ptr<TSurface>& surface) {
if (!surface->IsModified()) {
return exctx;
}
exctx = surface->DownloadTexture(exctx);
surface->FlushBuffer();
return exctx;
}
std::vector<std::shared_ptr<TSurface>> GetSurfacesInRegion(CacheAddr cache_addr,
std::size_t size) const {
if (size == 0) {
return {};
}
const IntervalType interval{cache_addr, cache_addr + size};
std::vector<std::shared_ptr<TSurface>> surfaces;
for (auto& pair : boost::make_iterator_range(registered_surfaces.equal_range(interval))) {
surfaces.push_back(*pair.second.begin());
}
return surfaces;
}
void ReserveSurface(const SurfaceParams& params, std::shared_ptr<TSurface> surface) {
surface_reserve[params].push_back(std::move(surface));
}
std::shared_ptr<TSurface> TryGetReservedSurface(const SurfaceParams& params) {
auto search{surface_reserve.find(params)};
if (search == surface_reserve.end()) {
return {};
}
for (auto& surface : search->second) {
if (!surface->IsRegistered()) {
return surface;
}
}
return {};
}
IntervalType GetSurfaceInterval(std::shared_ptr<TSurface> surface) const {
return IntervalType::right_open(surface->GetCacheAddr(),
surface->GetCacheAddr() + surface->GetSizeInBytes());
}
VideoCore::RasterizerInterface& rasterizer;
u64 ticks{};
IntervalMap registered_surfaces;
/// The surface reserve is a "backup" cache, this is where we put unique surfaces that have
/// previously been used. This is to prevent surfaces from being constantly created and
/// destroyed when used with different surface parameters.
std::unordered_map<SurfaceParams, std::list<std::shared_ptr<TSurface>>> surface_reserve;
};
} // namespace VideoCommon

93
src/video_core/texture_cache/texture_cache_contextless.h Normal file
View file

@ -0,0 +1,93 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "video_core/texture_cache/texture_cache.h"
namespace VideoCommon {
struct DummyExecutionContext {};
template <typename TSurface, typename TView>
class TextureCacheContextless : protected TextureCache<TSurface, TView, DummyExecutionContext> {
using Base = TextureCache<TSurface, TView, DummyExecutionContext>;
public:
void InvalidateRegion(CacheAddr addr, std::size_t size) {
Base::InvalidateRegion(addr, size);
}
TView* GetTextureSurface(const Tegra::Texture::FullTextureInfo& config) {
return RemoveContext(Base::GetTextureSurface({}, config));
}
TView* GetDepthBufferSurface(bool preserve_contents) {
return RemoveContext(Base::GetDepthBufferSurface({}, preserve_contents));
}
TView* GetColorBufferSurface(std::size_t index, bool preserve_contents) {
return RemoveContext(Base::GetColorBufferSurface({}, index, preserve_contents));
}
TView* GetFermiSurface(const Tegra::Engines::Fermi2D::Regs::Surface& config) {
return RemoveContext(Base::GetFermiSurface({}, config));
}
std::shared_ptr<TSurface> TryFindFramebufferSurface(const u8* host_ptr) const {
return Base::TryFindFramebufferSurface(host_ptr);
}
u64 Tick() {
return Base::Tick();
}
protected:
explicit TextureCacheContextless(Core::System& system,
VideoCore::RasterizerInterface& rasterizer)
: TextureCache<TSurface, TView, DummyExecutionContext>{system, rasterizer} {}
virtual TView* TryFastGetSurfaceView(
GPUVAddr gpu_addr, VAddr cpu_addr, u8* host_ptr, const SurfaceParams& params,
bool preserve_contents, const std::vector<std::shared_ptr<TSurface>>& overlaps) = 0;
private:
std::tuple<TView*, DummyExecutionContext> TryFastGetSurfaceView(
DummyExecutionContext, GPUVAddr gpu_addr, VAddr cpu_addr, u8* host_ptr,
const SurfaceParams& params, bool preserve_contents,
const std::vector<std::shared_ptr<TSurface>>& overlaps) {
return {TryFastGetSurfaceView(gpu_addr, cpu_addr, host_ptr, params, preserve_contents,
overlaps),
{}};
}
TView* RemoveContext(std::tuple<TView*, DummyExecutionContext> return_value) {
const auto [view, exctx] = return_value;
return view;
}
};
template <typename TTextureCache, typename TView>
class SurfaceBaseContextless : public SurfaceBase<TTextureCache, TView, DummyExecutionContext> {
public:
DummyExecutionContext DownloadTexture(DummyExecutionContext) {
DownloadTextureImpl();
return {};
}
DummyExecutionContext UploadTexture(DummyExecutionContext) {
UploadTextureImpl();
return {};
}
protected:
explicit SurfaceBaseContextless(TTextureCache& texture_cache, const SurfaceParams& params)
: SurfaceBase<TTextureCache, TView, DummyExecutionContext>{texture_cache, params} {}
virtual void DownloadTextureImpl() = 0;
virtual void UploadTextureImpl() = 0;
};
} // namespace VideoCommon