N-archive/citra
1
0
Fork 0
Code Issues Pull requests Packages Projects Releases Wiki Activity

Merge pull request #4923 from jroweboy/diskcachelul

Browse source 
Disk Shader Caching
This commit is contained in:
James Rowe 2020-01-17 18:15:50 -07:00 committed by GitHub
parent 5fdcff9d05 e95bc52b3d
commit e74a402c69
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
41 changed files with 1600 additions and 138 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
.gitmodules vendored
View file

@ -46,3 +46,6 @@
[submodule "lodepng"]
path = externals/lodepng/lodepng
url = https://github.com/lvandeve/lodepng.git
[submodule "zstd"]
path = externals/zstd
url = https://github.com/facebook/zstd.git

90
CMakeModules/GenerateSCMRev.cmake Normal file
View file

@ -0,0 +1,90 @@
# Gets a UTC timstamp and sets the provided variable to it
function(get_timestamp _var)
string(TIMESTAMP timestamp UTC)
set(${_var} "${timestamp}" PARENT_SCOPE)
endfunction()
list(APPEND CMAKE_MODULE_PATH "${SRC_DIR}/externals/cmake-modules")
# Find the package here with the known path so that the GetGit commands can find it as well
find_package(Git QUIET PATHS "${GIT_EXECUTABLE}")
# generate git/build information
include(GetGitRevisionDescription)
get_git_head_revision(GIT_REF_SPEC GIT_REV)
git_describe(GIT_DESC --always --long --dirty)
git_branch_name(GIT_BRANCH)
get_timestamp(BUILD_DATE)
# Generate cpp with Git revision from template
# Also if this is a CI build, add the build name (ie: Nightly, Canary) to the scm_rev file as well
set(REPO_NAME "")
set(BUILD_VERSION "0")
if (DEFINED ENV{CI})
if (DEFINED ENV{TRAVIS})
set(BUILD_REPOSITORY $ENV{TRAVIS_REPO_SLUG})
set(BUILD_TAG $ENV{TRAVIS_TAG})
elseif(DEFINED ENV{APPVEYOR})
set(BUILD_REPOSITORY $ENV{APPVEYOR_REPO_NAME})
set(BUILD_TAG $ENV{APPVEYOR_REPO_TAG_NAME})
elseif(DEFINED ENV{BITRISE_IO})
set(BUILD_REPOSITORY "$ENV{BITRISEIO_GIT_REPOSITORY_OWNER}/$ENV{BITRISEIO_GIT_REPOSITORY_SLUG}")
set(BUILD_TAG $ENV{BITRISE_GIT_TAG})
endif()
# regex capture the string nightly or canary into CMAKE_MATCH_1
string(REGEX MATCH "citra-emu/citra-?(.*)" OUTVAR ${BUILD_REPOSITORY})
if ("${CMAKE_MATCH_COUNT}" GREATER 0)
# capitalize the first letter of each word in the repo name.
string(REPLACE "-" ";" REPO_NAME_LIST ${CMAKE_MATCH_1})
foreach(WORD ${REPO_NAME_LIST})
string(SUBSTRING ${WORD} 0 1 FIRST_LETTER)
string(SUBSTRING ${WORD} 1 -1 REMAINDER)
string(TOUPPER ${FIRST_LETTER} FIRST_LETTER)
set(REPO_NAME "${REPO_NAME}${FIRST_LETTER}${REMAINDER}")
endforeach()
if (BUILD_TAG)
string(REGEX MATCH "${CMAKE_MATCH_1}-([0-9]+)" OUTVAR ${BUILD_TAG})
if (${CMAKE_MATCH_COUNT} GREATER 0)
set(BUILD_VERSION ${CMAKE_MATCH_1})
endif()
if (BUILD_VERSION)
# This leaves a trailing space on the last word, but we actually want that
# because of how it's styled in the title bar.
set(BUILD_FULLNAME "${REPO_NAME} ${BUILD_VERSION} ")
else()
set(BUILD_FULLNAME "")
endif()
endif()
endif()
endif()
# The variable SRC_DIR must be passed into the script (since it uses the current build directory for all values of CMAKE_*_DIR)
set(VIDEO_CORE "${SRC_DIR}/src/video_core")
set(HASH_FILES
"${VIDEO_CORE}/renderer_opengl/gl_shader_decompiler.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_decompiler.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_disk_cache.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_disk_cache.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_gen.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_gen.h"
"${VIDEO_CORE}/shader/shader.cpp"
"${VIDEO_CORE}/shader/shader.h"
"${VIDEO_CORE}/pica.cpp"
"${VIDEO_CORE}/pica.h"
"${VIDEO_CORE}/regs_framebuffer.h"
"${VIDEO_CORE}/regs_lighting.h"
"${VIDEO_CORE}/regs_pipeline.h"
"${VIDEO_CORE}/regs_rasterizer.h"
"${VIDEO_CORE}/regs_shader.h"
"${VIDEO_CORE}/regs_texturing.h"
"${VIDEO_CORE}/regs.cpp"
"${VIDEO_CORE}/regs.h"
)
set(COMBINED "")
foreach (F IN LISTS HASH_FILES)
file(READ ${F} TMP)
set(COMBINED "${COMBINED}${TMP}")
endforeach()
string(MD5 SHADER_CACHE_VERSION "${COMBINED}")
configure_file("${SRC_DIR}/src/common/scm_rev.cpp.in" "scm_rev.cpp" @ONLY)

4
externals/CMakeLists.txt vendored
View file

@ -62,6 +62,10 @@ if (ARCHITECTURE_x86_64)
target_compile_definitions(xbyak INTERFACE XBYAK_NO_OP_NAMES)
endif()
# Zstandard
add_subdirectory(zstd/build/cmake EXCLUDE_FROM_ALL)
target_include_directories(libzstd_static INTERFACE ./zstd/lib)
# ENet
add_subdirectory(enet)
target_include_directories(enet INTERFACE ./enet/include)

1
externals/zstd vendored Submodule

@ -0,0 +1 @@
Subproject commit 10f0e6993f9d2f682da6d04aa2385b7d53cbb4ee

10
src/citra/citra.cpp
View file

@ -43,7 +43,7 @@
#include "core/movie.h"
#include "core/settings.h"
#include "network/network.h"
#include "video_core/video_core.h"
#include "video_core/renderer_base.h"
#undef _UNICODE
#include <getopt.h>
@ -413,6 +413,14 @@ int main(int argc, char** argv) {
}
std::thread render_thread([&emu_window] { emu_window->Present(); });
std::atomic_bool stop_run;
Core::System::GetInstance().Renderer().Rasterizer()->LoadDiskResources(
stop_run, [](VideoCore::LoadCallbackStage stage, std::size_t value, std::size_t total) {
LOG_DEBUG(Frontend, "Loading stage {} progress {} {}", static_cast<u32>(stage), value,
total);
});
while (emu_window->IsOpen()) {
system.RunLoop();
}

2
src/citra/config.cpp
View file

@ -122,6 +122,8 @@ void Config::ReadValues() {
Settings::values.resolution_factor =
static_cast<u16>(sdl2_config->GetInteger("Renderer", "resolution_factor", 1));
Settings::values.use_frame_limit = sdl2_config->GetBoolean("Renderer", "use_frame_limit", true);
Settings::values.use_disk_shader_cache =
sdl2_config->GetBoolean("Renderer", "use_disk_shader_cache", true);
Settings::values.frame_limit =
static_cast<u16>(sdl2_config->GetInteger("Renderer", "frame_limit", 100));
Settings::values.use_vsync_new =

4
src/citra/default_ini.h
View file

@ -117,6 +117,10 @@ use_shader_jit =
# 0: Off, 1 (default): On
use_vsync_new =
# Reduce stuttering by storing and loading generated shaders to disk
# 0: Off, 1 (default. On)
use_disk_shader_cache =
# Resolution scale factor
# 0: Auto (scales resolution to window size), 1: Native 3DS screen resolution, Otherwise a scale
# factor for the 3DS resolution

7
src/citra_qt/bootmanager.cpp
View file

@ -45,6 +45,13 @@ static GMainWindow* GetMainWindow() {
void EmuThread::run() {
MicroProfileOnThreadCreate("EmuThread");
Frontend::ScopeAcquireContext scope(core_context);
Core::System::GetInstance().Renderer().Rasterizer()->LoadDiskResources(
stop_run, [this](VideoCore::LoadCallbackStage stage, std::size_t value, std::size_t total) {
LOG_DEBUG(Frontend, "Loading stage {} progress {} {}", static_cast<u32>(stage), value,
total);
});
// Holds whether the cpu was running during the last iteration,
// so that the DebugModeLeft signal can be emitted before the
// next execution step.

2
src/citra_qt/configuration/config.cpp
View file

@ -239,6 +239,7 @@ void Config::ReadUtilityValues() {
Settings::values.dump_textures = ReadSetting("dump_textures", false).toBool();
Settings::values.custom_textures = ReadSetting("custom_textures", false).toBool();
Settings::values.preload_textures = ReadSetting("preload_textures", false).toBool();
Settings::values.use_disk_shader_cache = ReadSetting("use_disk_shader_cache", true).toBool();
qt_config->endGroup();
}
@ -713,6 +714,7 @@ void Config::SaveUtilityValues() {
WriteSetting("dump_textures", Settings::values.dump_textures, false);
WriteSetting("custom_textures", Settings::values.custom_textures, false);
WriteSetting("preload_textures", Settings::values.preload_textures, false);
WriteSetting("use_disk_shader_cache", Settings::values.use_disk_shader_cache, true);
qt_config->endGroup();
}

2
src/citra_qt/configuration/configure_enhancements.cpp
View file

@ -52,6 +52,7 @@ void ConfigureEnhancements::SetConfiguration() {
ui->toggle_linear_filter->setChecked(Settings::values.filter_mode);
ui->layout_combobox->setCurrentIndex(static_cast<int>(Settings::values.layout_option));
ui->swap_screen->setChecked(Settings::values.swap_screen);
ui->toggle_disk_shader_cache->setChecked(Settings::values.use_disk_shader_cache);
ui->toggle_dump_textures->setChecked(Settings::values.dump_textures);
ui->toggle_custom_textures->setChecked(Settings::values.custom_textures);
ui->toggle_preload_textures->setChecked(Settings::values.preload_textures);
@ -99,6 +100,7 @@ void ConfigureEnhancements::ApplyConfiguration() {
Settings::values.layout_option =
static_cast<Settings::LayoutOption>(ui->layout_combobox->currentIndex());
Settings::values.swap_screen = ui->swap_screen->isChecked();
Settings::values.use_disk_shader_cache = ui->toggle_disk_shader_cache->isChecked();
Settings::values.dump_textures = ui->toggle_dump_textures->isChecked();
Settings::values.custom_textures = ui->toggle_custom_textures->isChecked();
Settings::values.preload_textures = ui->toggle_preload_textures->isChecked();

7
src/citra_qt/configuration/configure_enhancements.ui
View file

@ -269,6 +269,13 @@
<string>Utility</string>
</property>
<layout class="QVBoxLayout" name="verticalLayout_8">
<item>
<widget class="QCheckBox" name="toggle_disk_shader_cache">
<property name="text">
<string>Use Disk Shader Cache</string>
</property>
</widget>
</item>
<item>
<widget class="QCheckBox" name="toggle_custom_textures">
<property name="toolTip">

85
src/common/CMakeLists.txt
View file

@ -1,45 +1,55 @@
# Generate cpp with Git revision from template
# Also if this is a CI build, add the build name (ie: Nightly, Canary) to the scm_rev file as well
set(REPO_NAME "")
set(BUILD_VERSION "0")
if ($ENV{CI})
if ($ENV{TRAVIS})
# Add a custom command to generate a new shader_cache_version hash when any of the following files change
# NOTE: This is an approximation of what files affect shader generation, its possible something else
# could affect the result, but much more unlikely than the following files. Keeping a list of files
# like this allows for much better caching since it doesn't force the user to recompile binary shaders every update
set(VIDEO_CORE "${CMAKE_SOURCE_DIR}/src/video_core")
if (DEFINED ENV{CI})
if (DEFINED ENV{TRAVIS})
set(BUILD_REPOSITORY $ENV{TRAVIS_REPO_SLUG})
set(BUILD_TAG $ENV{TRAVIS_TAG})
elseif($ENV{APPVEYOR})
elseif(DEFINED ENV{APPVEYOR})
set(BUILD_REPOSITORY $ENV{APPVEYOR_REPO_NAME})
set(BUILD_TAG $ENV{APPVEYOR_REPO_TAG_NAME})
elseif($ENV{BITRISE_IO})
set(BUILD_REPOSITORY "$ENV{BITRISEIO_GIT_REPOSITORY_OWNER}/$ENV{BITRISEIO_GIT_REPOSITORY_SLUG}")
set(BUILD_TAG $ENV{BITRISE_GIT_TAG})
endif()
# regex capture the string nightly or canary into CMAKE_MATCH_1
string(REGEX MATCH "citra-emu/citra-?(.*)" OUTVAR ${BUILD_REPOSITORY})
if (${CMAKE_MATCH_COUNT} GREATER 0)
# capitalize the first letter of each word in the repo name.
string(REPLACE "-" ";" REPO_NAME_LIST ${CMAKE_MATCH_1})
foreach(WORD ${REPO_NAME_LIST})
string(SUBSTRING ${WORD} 0 1 FIRST_LETTER)
string(SUBSTRING ${WORD} 1 -1 REMAINDER)
string(TOUPPER ${FIRST_LETTER} FIRST_LETTER)
set(REPO_NAME "${REPO_NAME}${FIRST_LETTER}${REMAINDER}")
endforeach()
if (BUILD_TAG)
string(REGEX MATCH "${CMAKE_MATCH_1}-([0-9]+)" OUTVAR ${BUILD_TAG})
if (${CMAKE_MATCH_COUNT} GREATER 0)
set(BUILD_VERSION ${CMAKE_MATCH_1})
endif()
if (BUILD_VERSION)
# This leaves a trailing space on the last word, but we actually want that
# because of how it's styled in the title bar.
set(BUILD_FULLNAME "${REPO_NAME} ${BUILD_VERSION} ")
else()
set(BUILD_FULLNAME "")
endif()
endif()
endif()
endif()
configure_file("${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.cpp.in" "${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.cpp" @ONLY)
# Pass the path to git to the GenerateSCMRev.cmake as well
find_package(Git QUIET)
add_custom_command(OUTPUT scm_rev.cpp
COMMAND ${CMAKE_COMMAND}
-DSRC_DIR="${CMAKE_SOURCE_DIR}"
-DBUILD_REPOSITORY="${BUILD_REPOSITORY}"
-DBUILD_TAG="${BUILD_TAG}"
-DGIT_EXECUTABLE="${GIT_EXECUTABLE}"
-P "${CMAKE_SOURCE_DIR}/CMakeModules/GenerateSCMRev.cmake"
DEPENDS
# WARNING! It was too much work to try and make a common location for this list,
# so if you need to change it, please update CMakeModules/GenerateSCMRev.cmake as well
"${VIDEO_CORE}/renderer_opengl/gl_shader_decompiler.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_decompiler.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_disk_cache.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_disk_cache.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_gen.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_gen.h"
"${VIDEO_CORE}/shader/shader.cpp"
"${VIDEO_CORE}/shader/shader.h"
"${VIDEO_CORE}/pica.cpp"
"${VIDEO_CORE}/pica.h"
"${VIDEO_CORE}/regs_framebuffer.h"
"${VIDEO_CORE}/regs_lighting.h"
"${VIDEO_CORE}/regs_pipeline.h"
"${VIDEO_CORE}/regs_rasterizer.h"
"${VIDEO_CORE}/regs_shader.h"
"${VIDEO_CORE}/regs_texturing.h"
"${VIDEO_CORE}/regs.cpp"
"${VIDEO_CORE}/regs.h"
# and also check that the scm_rev files haven't changed
"${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.cpp.in"
"${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.h"
# technically we should regenerate if the git version changed, but its not worth the effort imo
"${CMAKE_SOURCE_DIR}/CMakeModules/GenerateSCMRev.cmake"
)
add_library(common STATIC
alignment.h
@ -94,6 +104,8 @@ add_library(common STATIC
timer.h
vector_math.h
web_result.h
zstd_compression.cpp
zstd_compression.h
)
if(ARCHITECTURE_x86_64)
@ -110,6 +122,7 @@ endif()
create_target_directory_groups(common)
target_link_libraries(common PUBLIC fmt microprofile)
target_link_libraries(common PRIVATE libzstd_static)
if (ARCHITECTURE_x86_64)
target_link_libraries(common PRIVATE xbyak)
endif()

116
src/common/file_util.cpp
View file

@ -4,6 +4,7 @@
#include <array>
#include <memory>
#include <sstream>
#include <unordered_map>
#include "common/assert.h"
#include "common/common_funcs.h"
@ -355,12 +356,12 @@ u64 GetSize(FILE* f) {
// can't use off_t here because it can be 32-bit
u64 pos = ftello(f);
if (fseeko(f, 0, SEEK_END) != 0) {
LOG_ERROR(Common_Filesystem, "GetSize: seek failed {}: {}", (void*)f, GetLastErrorMsg());
LOG_ERROR(Common_Filesystem, "GetSize: seek failed {}: {}", fmt::ptr(f), GetLastErrorMsg());
return 0;
}
u64 size = ftello(f);
if ((size != pos) && (fseeko(f, pos, SEEK_SET) != 0)) {
LOG_ERROR(Common_Filesystem, "GetSize: seek failed {}: {}", (void*)f, GetLastErrorMsg());
LOG_ERROR(Common_Filesystem, "GetSize: seek failed {}: {}", fmt::ptr(f), GetLastErrorMsg());
return 0;
}
return size;
@ -369,7 +370,7 @@ u64 GetSize(FILE* f) {
bool CreateEmptyFile(const std::string& filename) {
LOG_TRACE(Common_Filesystem, "{}", filename);
if (!FileUtil::IOFile(filename, "wb")) {
if (!FileUtil::IOFile(filename, "wb").IsOpen()) {
LOG_ERROR(Common_Filesystem, "failed {}: {}", filename, GetLastErrorMsg());
return false;
}
@ -541,12 +542,11 @@ std::optional<std::string> GetCurrentDir() {
// Get the current working directory (getcwd uses malloc)
#ifdef _WIN32
wchar_t* dir;
if (!(dir = _wgetcwd(nullptr, 0)))
if (!(dir = _wgetcwd(nullptr, 0))) {
#else
char* dir;
if (!(dir = getcwd(nullptr, 0)))
if (!(dir = getcwd(nullptr, 0))) {
#endif
{
LOG_ERROR(Common_Filesystem, "GetCurrentDirectory failed: {}", GetLastErrorMsg());
return {};
}
@ -557,7 +557,7 @@ std::optional<std::string> GetCurrentDir() {
#endif
free(dir);
return strDir;
}
} // namespace FileUtil
bool SetCurrentDir(const std::string& directory) {
#ifdef _WIN32
@ -733,7 +733,6 @@ const std::string& GetUserPath(UserPath path) {
SetUserPath();
return g_paths[path];
}
std::size_t WriteStringToFile(bool text_file, const std::string& filename, std::string_view str) {
return IOFile(filename, text_file ? "w" : "wb").WriteString(str);
}
@ -741,8 +740,8 @@ std::size_t WriteStringToFile(bool text_file, const std::string& filename, std::
std::size_t ReadFileToString(bool text_file, const std::string& filename, std::string& str) {
IOFile file(filename, text_file ? "r" : "rb");
if (!file)
return false;
if (!file.IsOpen())
return 0;
str.resize(static_cast<u32>(file.GetSize()));
return file.ReadArray(&str[0], str.size());
@ -783,6 +782,103 @@ void SplitFilename83(const std::string& filename, std::array<char, 9>& short_nam
}
}
std::vector<std::string> SplitPathComponents(std::string_view filename) {
std::string copy(filename);
std::replace(copy.begin(), copy.end(), '\\', '/');
std::vector<std::string> out;
std::stringstream stream(copy);
std::string item;
while (std::getline(stream, item, '/')) {
out.push_back(std::move(item));
}
return out;
}
std::string_view GetParentPath(std::string_view path) {
const auto name_bck_index = path.rfind('\\');
const auto name_fwd_index = path.rfind('/');
std::size_t name_index;
if (name_bck_index == std::string_view::npos || name_fwd_index == std::string_view::npos) {
name_index = std::min(name_bck_index, name_fwd_index);
} else {
name_index = std::max(name_bck_index, name_fwd_index);
}
return path.substr(0, name_index);
}
std::string_view GetPathWithoutTop(std::string_view path) {
if (path.empty()) {
return path;
}
while (path[0] == '\\' || path[0] == '/') {
path.remove_prefix(1);
if (path.empty()) {
return path;
}
}
const auto name_bck_index = path.find('\\');
const auto name_fwd_index = path.find('/');
return path.substr(std::min(name_bck_index, name_fwd_index) + 1);
}
std::string_view GetFilename(std::string_view path) {
const auto name_index = path.find_last_of("\\/");
if (name_index == std::string_view::npos) {
return {};
}
return path.substr(name_index + 1);
}
std::string_view GetExtensionFromFilename(std::string_view name) {
const std::size_t index = name.rfind('.');
if (index == std::string_view::npos) {
return {};
}
return name.substr(index + 1);
}
std::string_view RemoveTrailingSlash(std::string_view path) {
if (path.empty()) {
return path;
}
if (path.back() == '\\' || path.back() == '/') {
path.remove_suffix(1);
return path;
}
return path;
}
std::string SanitizePath(std::string_view path_, DirectorySeparator directory_separator) {
std::string path(path_);
char type1 = directory_separator == DirectorySeparator::BackwardSlash ? '/' : '\\';
char type2 = directory_separator == DirectorySeparator::BackwardSlash ? '\\' : '/';
if (directory_separator == DirectorySeparator::PlatformDefault) {
#ifdef _WIN32
type1 = '/';
type2 = '\\';
#endif
}
std::replace(path.begin(), path.end(), type1, type2);
path.erase(std::unique(path.begin(), path.end(),
[type2](char c1, char c2) { return c1 == type2 && c2 == type2; }),
path.end());
return std::string(RemoveTrailingSlash(path));
}
IOFile::IOFile() {}
IOFile::IOFile(const std::string& filename, const char openmode[], int flags) {

36
src/common/file_util.h
View file

@ -11,6 +11,7 @@
#include <limits>
#include <optional>
#include <string>
#include <string_view>
#include <type_traits>
#include <vector>
#include "common/common_types.h"
@ -166,6 +167,41 @@ std::size_t ReadFileToString(bool text_file, const std::string& filename, std::s
void SplitFilename83(const std::string& filename, std::array<char, 9>& short_name,
std::array<char, 4>& extension);
// Splits the path on '/' or '\' and put the components into a vector
// i.e. "C:\Users\Yuzu\Documents\save.bin" becomes {"C:", "Users", "Yuzu", "Documents", "save.bin" }
std::vector<std::string> SplitPathComponents(std::string_view filename);
// Gets all of the text up to the last '/' or '\' in the path.
std::string_view GetParentPath(std::string_view path);
// Gets all of the text after the first '/' or '\' in the path.
std::string_view GetPathWithoutTop(std::string_view path);
// Gets the filename of the path
std::string_view GetFilename(std::string_view path);
// Gets the extension of the filename
std::string_view GetExtensionFromFilename(std::string_view name);
// Removes the final '/' or '\' if one exists
std::string_view RemoveTrailingSlash(std::string_view path);
// Creates a new vector containing indices [first, last) from the original.
template <typename T>
std::vector<T> SliceVector(const std::vector<T>& vector, std::size_t first, std::size_t last) {
if (first >= last)
return {};
last = std::min<std::size_t>(last, vector.size());
return std::vector<T>(vector.begin() + first, vector.begin() + first + last);
}
enum class DirectorySeparator { ForwardSlash, BackwardSlash, PlatformDefault };
// Removes trailing slash, makes all '\\' into '/', and removes duplicate '/'. Makes '/' into '\\'
// depending if directory_separator is BackwardSlash or PlatformDefault and running on windows
std::string SanitizePath(std::string_view path,
DirectorySeparator directory_separator = DirectorySeparator::ForwardSlash);
// simple wrapper for cstdlib file functions to
// hopefully will make error checking easier
// and make forgetting an fclose() harder

2
src/common/scm_rev.cpp.in
View file

@ -11,6 +11,7 @@
#define BUILD_DATE "@BUILD_DATE@"
#define BUILD_VERSION "@BUILD_VERSION@"
#define BUILD_FULLNAME "@BUILD_FULLNAME@"
#define SHADER_CACHE_VERSION "@SHADER_CACHE_VERSION@"
namespace Common {
@ -21,6 +22,7 @@ const char g_build_name[] = BUILD_NAME;
const char g_build_date[] = BUILD_DATE;
const char g_build_fullname[] = BUILD_FULLNAME;
const char g_build_version[] = BUILD_VERSION;
const char g_shader_cache_version[] = SHADER_CACHE_VERSION;
} // namespace

1
src/common/scm_rev.h
View file

@ -13,5 +13,6 @@ extern const char g_build_name[];
extern const char g_build_date[];
extern const char g_build_fullname[];
extern const char g_build_version[];
extern const char g_shader_cache_version[];
} // namespace Common

51
src/common/zstd_compression.cpp Normal file
View file

@ -0,0 +1,51 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <zstd.h>
#include "common/assert.h"
#include "common/zstd_compression.h"
namespace Common::Compression {
std::vector<u8> CompressDataZSTD(const u8* source, std::size_t source_size, s32 compression_level) {
compression_level = std::clamp(compression_level, ZSTD_minCLevel(), ZSTD_maxCLevel());
const std::size_t max_compressed_size = ZSTD_compressBound(source_size);
std::vector<u8> compressed(max_compressed_size);
const std::size_t compressed_size =
ZSTD_compress(compressed.data(), compressed.size(), source, source_size, compression_level);
if (ZSTD_isError(compressed_size)) {
// Compression failed
return {};
}
compressed.resize(compressed_size);
return compressed;
}
std::vector<u8> CompressDataZSTDDefault(const u8* source, std::size_t source_size) {
return CompressDataZSTD(source, source_size, ZSTD_CLEVEL_DEFAULT);
}
std::vector<u8> DecompressDataZSTD(const std::vector<u8>& compressed) {
const std::size_t decompressed_size =
ZSTD_getDecompressedSize(compressed.data(), compressed.size());
std::vector<u8> decompressed(decompressed_size);
const std::size_t uncompressed_result_size = ZSTD_decompress(
decompressed.data(), decompressed.size(), compressed.data(), compressed.size());
if (decompressed_size != uncompressed_result_size || ZSTD_isError(uncompressed_result_size)) {
// Decompression failed
return {};
}
return decompressed;
}
} // namespace Common::Compression

44
src/common/zstd_compression.h Normal file
View file

@ -0,0 +1,44 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <vector>
#include "common/common_types.h"
namespace Common::Compression {
/**
* Compresses a source memory region with Zstandard and returns the compressed data in a vector.
*
* @param source the uncompressed source memory region.
* @param source_size the size in bytes of the uncompressed source memory region.
* @param compression_level the used compression level. Should be between 1 and 22.
*
* @return the compressed data.
*/
std::vector<u8> CompressDataZSTD(const u8* source, std::size_t source_size, s32 compression_level);
/**
* Compresses a source memory region with Zstandard with the default compression level and returns
* the compressed data in a vector.
*
* @param source the uncompressed source memory region.
* @param source_size the size in bytes of the uncompressed source memory region.
*
* @return the compressed data.
*/
std::vector<u8> CompressDataZSTDDefault(const u8* source, std::size_t source_size);
/**
* Decompresses a source memory region with Zstandard and returns the uncompressed data in a vector.
*
* @param compressed the compressed source memory region.
*
* @return the decompressed data.
*/
std::vector<u8> DecompressDataZSTD(const std::vector<u8>& compressed);
} // namespace Common::Compression

17
src/core/core.cpp
View file

@ -237,9 +237,16 @@ System::ResultStatus System::Init(Frontend::EmuWindow& emu_window, u32 system_mo
Service::Init(*this);
GDBStub::Init();
ResultStatus result = VideoCore::Init(emu_window, *memory);
if (result != ResultStatus::Success) {
return result;
VideoCore::ResultStatus result = VideoCore::Init(emu_window, *memory);
if (result != VideoCore::ResultStatus::Success) {
switch (result) {
case VideoCore::ResultStatus::ErrorGenericDrivers:
return ResultStatus::ErrorVideoCore_ErrorGenericDrivers;
case VideoCore::ResultStatus::ErrorBelowGL33:
return ResultStatus::ErrorVideoCore_ErrorBelowGL33;
default:
return ResultStatus::ErrorVideoCore;
}
}
#ifdef ENABLE_FFMPEG_VIDEO_DUMPER
@ -253,6 +260,10 @@ System::ResultStatus System::Init(Frontend::EmuWindow& emu_window, u32 system_mo
return ResultStatus::Success;
}
RendererBase& System::Renderer() {
return *VideoCore::g_renderer;
}
Service::SM::ServiceManager& System::ServiceManager() {
return *service_manager;
}

4
src/core/core.h
View file

@ -55,6 +55,8 @@ namespace VideoDumper {
class Backend;
}
class RendererBase;
namespace Core {
class Timing;
@ -170,6 +172,8 @@ public:
return *dsp_core;
}
RendererBase& Renderer();
/**
* Gets a reference to the service manager.
* @returns A reference to the service manager.

1
src/core/hw/gpu.cpp
View file

@ -11,6 +11,7 @@
#include "common/logging/log.h"
#include "common/microprofile.h"
#include "common/vector_math.h"
#include "core/core.h"
#include "core/core_timing.h"
#include "core/hle/service/gsp/gsp.h"
#include "core/hw/gpu.h"

1
src/core/settings.cpp
View file

@ -28,6 +28,7 @@ void Apply() {
VideoCore::g_shader_jit_enabled = values.use_shader_jit;
VideoCore::g_hw_shader_enabled = values.use_hw_shader;
VideoCore::g_hw_shader_accurate_mul = values.shaders_accurate_mul;
VideoCore::g_use_disk_shader_cache = values.use_disk_shader_cache;
if (VideoCore::g_renderer) {
VideoCore::g_renderer->UpdateCurrentFramebufferLayout();

1
src/core/settings.h
View file

@ -141,6 +141,7 @@ struct Values {
bool use_gles;
bool use_hw_renderer;
bool use_hw_shader;
bool use_disk_shader_cache;
bool shaders_accurate_mul;
bool use_shader_jit;
u16 resolution_factor;

2
src/video_core/CMakeLists.txt
View file

@ -31,6 +31,8 @@ add_library(video_core STATIC
renderer_opengl/gl_resource_manager.h
renderer_opengl/gl_shader_decompiler.cpp
renderer_opengl/gl_shader_decompiler.h
renderer_opengl/gl_shader_disk_cache.cpp
renderer_opengl/gl_shader_disk_cache.h
renderer_opengl/gl_shader_gen.cpp
renderer_opengl/gl_shader_gen.h
renderer_opengl/gl_shader_manager.cpp

13
src/video_core/rasterizer_interface.h
View file

@ -4,6 +4,8 @@
#pragma once
#include <atomic>
#include <functional>
#include "common/common_types.h"
#include "core/hw/gpu.h"
@ -17,6 +19,14 @@ struct OutputVertex;
namespace VideoCore {
enum class LoadCallbackStage {
Prepare,
Decompile,
Build,
Complete,
};
using DiskResourceLoadCallback = std::function<void(LoadCallbackStage, std::size_t, std::size_t)>;
class RasterizerInterface {
public:
virtual ~RasterizerInterface() {}
@ -71,5 +81,8 @@ public:
virtual bool AccelerateDrawBatch(bool is_indexed) {
return false;
}
virtual void LoadDiskResources(const std::atomic_bool& stop_loading,
const DiskResourceLoadCallback& callback) {}
};
} // namespace VideoCore

4
src/video_core/renderer_base.h
View file

@ -6,8 +6,8 @@
#include <memory>
#include "common/common_types.h"
#include "core/core.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/video_core.h"
namespace Frontend {
class EmuWindow;
@ -23,7 +23,7 @@ public:
virtual ~RendererBase();
/// Initialize the renderer
virtual Core::System::ResultStatus Init() = 0;
virtual VideoCore::ResultStatus Init() = 0;
/// Shutdown the renderer
virtual void ShutDown() = 0;

15
src/video_core/renderer_opengl/gl_rasterizer.cpp
View file

@ -171,6 +171,11 @@ RasterizerOpenGL::RasterizerOpenGL(Frontend::EmuWindow& window)
RasterizerOpenGL::~RasterizerOpenGL() {}
void RasterizerOpenGL::LoadDiskResources(const std::atomic_bool& stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) {
shader_program_manager->LoadDiskCache(stop_loading, callback);
}
void RasterizerOpenGL::SyncEntireState() {
// Sync fixed function OpenGL state
SyncClipEnabled();
@ -378,16 +383,15 @@ void RasterizerOpenGL::SetupVertexArray(u8* array_ptr, GLintptr buffer_offset,
bool RasterizerOpenGL::SetupVertexShader() {
MICROPROFILE_SCOPE(OpenGL_VS);
PicaVSConfig vs_config(Pica::g_state.regs, Pica::g_state.vs);
return shader_program_manager->UseProgrammableVertexShader(vs_config, Pica::g_state.vs);
return shader_program_manager->UseProgrammableVertexShader(Pica::g_state.regs,
Pica::g_state.vs);
}
bool RasterizerOpenGL::SetupGeometryShader() {
MICROPROFILE_SCOPE(OpenGL_GS);
const auto& regs = Pica::g_state.regs;
if (regs.pipeline.use_gs == Pica::PipelineRegs::UseGS::No) {
PicaFixedGSConfig gs_config(regs);
shader_program_manager->UseFixedGeometryShader(gs_config);
shader_program_manager->UseFixedGeometryShader(regs);
return true;
} else {
LOG_ERROR(Render_OpenGL, "Accelerate draw doesn't support geometry shader");
@ -1622,8 +1626,7 @@ void RasterizerOpenGL::SamplerInfo::SyncWithConfig(
}
void RasterizerOpenGL::SetShader() {
auto config = PicaFSConfig::BuildFromRegs(Pica::g_state.regs);
shader_program_manager->UseFragmentShader(config);
shader_program_manager->UseFragmentShader(Pica::g_state.regs);
}
void RasterizerOpenGL::SyncClipEnabled() {

3
src/video_core/renderer_opengl/gl_rasterizer.h
View file

@ -42,6 +42,9 @@ public:
explicit RasterizerOpenGL(Frontend::EmuWindow& renderer);
~RasterizerOpenGL() override;
void LoadDiskResources(const std::atomic_bool& stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) override;
void AddTriangle(const Pica::Shader::OutputVertex& v0, const Pica::Shader::OutputVertex& v1,
const Pica::Shader::OutputVertex& v2) override;
void DrawTriangles() override;

24
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View file

@ -56,7 +56,7 @@ struct Subroutine {
/// Analyzes shader code and produces a set of subroutines.
class ControlFlowAnalyzer {
public:
ControlFlowAnalyzer(const ProgramCode& program_code, u32 main_offset)
ControlFlowAnalyzer(const Pica::Shader::ProgramCode& program_code, u32 main_offset)
: program_code(program_code) {
// Recursively finds all subroutines.
@ -70,7 +70,7 @@ public:
}
private:
const ProgramCode& program_code;
const Pica::Shader::ProgramCode& program_code;
std::set<Subroutine> subroutines;
std::map<std::pair<u32, u32>, ExitMethod> exit_method_map;
@ -246,8 +246,9 @@ constexpr auto GetSelectorSrc3 = GetSelectorSrc<&SwizzlePattern::GetSelectorSrc3
class GLSLGenerator {
public:
GLSLGenerator(const std::set<Subroutine>& subroutines, const ProgramCode& program_code,
const SwizzleData& swizzle_data, u32 main_offset,
GLSLGenerator(const std::set<Subroutine>& subroutines,
const Pica::Shader::ProgramCode& program_code,
const Pica::Shader::SwizzleData& swizzle_data, u32 main_offset,
const RegGetter& inputreg_getter, const RegGetter& outputreg_getter,
bool sanitize_mul)
: subroutines(subroutines), program_code(program_code), swizzle_data(swizzle_data),
@ -865,8 +866,8 @@ private:
private:
const std::set<Subroutine>& subroutines;
const ProgramCode& program_code;
const SwizzleData& swizzle_data;
const Pica::Shader::ProgramCode& program_code;
const Pica::Shader::SwizzleData& swizzle_data;
const u32 main_offset;
const RegGetter& inputreg_getter;
const RegGetter& outputreg_getter;
@ -888,16 +889,17 @@ bool exec_shader();
)";
}
std::optional<std::string> DecompileProgram(const ProgramCode& program_code,
const SwizzleData& swizzle_data, u32 main_offset,
const RegGetter& inputreg_getter,
const RegGetter& outputreg_getter, bool sanitize_mul) {
std::optional<ProgramResult> DecompileProgram(const Pica::Shader::ProgramCode& program_code,
const Pica::Shader::SwizzleData& swizzle_data,
u32 main_offset, const RegGetter& inputreg_getter,
const RegGetter& outputreg_getter,
bool sanitize_mul) {
try {
auto subroutines = ControlFlowAnalyzer(program_code, main_offset).MoveSubroutines();
GLSLGenerator generator(subroutines, program_code, swizzle_data, main_offset,
inputreg_getter, outputreg_getter, sanitize_mul);
return generator.MoveShaderCode();
return {ProgramResult{generator.MoveShaderCode()}};
} catch (const DecompileFail& exception) {
LOG_INFO(HW_GPU, "Shader decompilation failed: {}", exception.what());
return {};

14
src/video_core/renderer_opengl/gl_shader_decompiler.h
View file

@ -11,15 +11,17 @@
namespace OpenGL::ShaderDecompiler {
using ProgramCode = std::array<u32, Pica::Shader::MAX_PROGRAM_CODE_LENGTH>;
using SwizzleData = std::array<u32, Pica::Shader::MAX_SWIZZLE_DATA_LENGTH>;
using RegGetter = std::function<std::string(u32)>;
struct ProgramResult {
std::string code;
};
std::string GetCommonDeclarations();
std::optional<std::string> DecompileProgram(const ProgramCode& program_code,
const SwizzleData& swizzle_data, u32 main_offset,
const RegGetter& inputreg_getter,
const RegGetter& outputreg_getter, bool sanitize_mul);
std::optional<ProgramResult> DecompileProgram(const Pica::Shader::ProgramCode& program_code,
const Pica::Shader::SwizzleData& swizzle_data,
u32 main_offset, const RegGetter& inputreg_getter,
const RegGetter& outputreg_getter, bool sanitize_mul);
} // namespace OpenGL::ShaderDecompiler

495
src/video_core/renderer_opengl/gl_shader_disk_cache.cpp Normal file
View file

@ -0,0 +1,495 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <fmt/format.h>
#include "common/assert.h"
#include "common/common_paths.h"
#include "common/common_types.h"
#include "common/file_util.h"
#include "common/logging/log.h"
#include "common/scm_rev.h"
#include "common/zstd_compression.h"
#include "core/core.h"
#include "core/hle/kernel/process.h"
#include "core/settings.h"
#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
namespace OpenGL {
constexpr std::size_t HASH_LENGTH = 64;
using ShaderCacheVersionHash = std::array<u8, HASH_LENGTH>;
enum class TransferableEntryKind : u32 {
Raw,
};
enum class PrecompiledEntryKind : u32 {
Decompiled,
Dump,
};
constexpr u32 NativeVersion = 1;
ShaderCacheVersionHash GetShaderCacheVersionHash() {
ShaderCacheVersionHash hash{};
const std::size_t length = std::min(std::strlen(Common::g_shader_cache_version), hash.size());
std::memcpy(hash.data(), Common::g_shader_cache_version, length);
return hash;
}
ShaderDiskCacheRaw::ShaderDiskCacheRaw(u64 unique_identifier, ProgramType program_type,
RawShaderConfig config, ProgramCode program_code)
: unique_identifier{unique_identifier}, program_type{program_type}, config{config},
program_code{std::move(program_code)} {}
bool ShaderDiskCacheRaw::Load(FileUtil::IOFile& file) {
if (file.ReadBytes(&unique_identifier, sizeof(u64)) != sizeof(u64) ||
file.ReadBytes(&program_type, sizeof(u32)) != sizeof(u32)) {
return false;
}
u64 reg_array_len{};
if (file.ReadBytes(&reg_array_len, sizeof(u64)) != sizeof(u64)) {
return false;
}
if (file.ReadArray(config.reg_array.data(), reg_array_len) != reg_array_len) {
return false;
}
// Read in type specific configuration
if (program_type == ProgramType::VS) {
u64 code_len{};
if (file.ReadBytes(&code_len, sizeof(u64)) != sizeof(u64)) {
return false;
}
program_code.resize(code_len);
if (file.ReadArray(program_code.data(), code_len) != code_len) {
return false;
}
}
return true;
}
bool ShaderDiskCacheRaw::Save(FileUtil::IOFile& file) const {
if (file.WriteObject(unique_identifier) != 1 ||
file.WriteObject(static_cast<u32>(program_type)) != 1) {
return false;
}
// Just for future proofing, save the sizes of the array to the file
const std::size_t reg_array_len = Pica::Regs::NUM_REGS;
if (file.WriteObject(static_cast<u64>(reg_array_len)) != 1) {
return false;
}
if (file.WriteArray(config.reg_array.data(), reg_array_len) != reg_array_len) {
return false;
}
if (program_type == ProgramType::VS) {
const std::size_t code_len = program_code.size();
if (file.WriteObject(static_cast<u64>(code_len)) != 1) {
return false;
}
if (file.WriteArray(program_code.data(), code_len) != code_len) {
return false;
}
}
return true;
}
ShaderDiskCache::ShaderDiskCache(bool separable) : separable{separable} {}
std::optional<std::vector<ShaderDiskCacheRaw>> ShaderDiskCache::LoadTransferable() {
const bool has_title_id = GetProgramID() != 0;
if (!Settings::values.use_disk_shader_cache || !has_title_id)
return {};
tried_to_load = true;
FileUtil::IOFile file(GetTransferablePath(), "rb");
if (!file.IsOpen()) {
LOG_INFO(Render_OpenGL, "No transferable shader cache found for game with title id={}",
GetTitleID());
return {};
}
u32 version{};
if (file.ReadBytes(&version, sizeof(version)) != sizeof(version)) {
LOG_ERROR(Render_OpenGL,
"Failed to get transferable cache version for title id={} - skipping",
GetTitleID());
return {};
}
if (version < NativeVersion) {
LOG_INFO(Render_OpenGL, "Transferable shader cache is old - removing");
file.Close();
InvalidateAll();
return {};
}
if (version > NativeVersion) {
LOG_WARNING(Render_OpenGL, "Transferable shader cache was generated with a newer version "
"of the emulator - skipping");
return {};
}
// Version is valid, load the shaders
std::vector<ShaderDiskCacheRaw> raws;
while (file.Tell() < file.GetSize()) {
TransferableEntryKind kind{};
if (file.ReadBytes(&kind, sizeof(u32)) != sizeof(u32)) {
LOG_ERROR(Render_OpenGL, "Failed to read transferable file - skipping");
return {};
}
switch (kind) {
case TransferableEntryKind::Raw: {
ShaderDiskCacheRaw entry;
if (!entry.Load(file)) {
LOG_ERROR(Render_OpenGL, "Failed to load transferable raw entry - skipping");
return {};
}
transferable.emplace(entry.GetUniqueIdentifier(), ShaderDiskCacheRaw{});
raws.push_back(std::move(entry));
break;
}
default:
LOG_ERROR(Render_OpenGL, "Unknown transferable shader cache entry kind={} - skipping",
static_cast<u32>(kind));
return {};
}
}
LOG_INFO(Render_OpenGL, "Found a transferable disk cache with {} entries", raws.size());
return {raws};
}
std::pair<std::unordered_map<u64, ShaderDiskCacheDecompiled>, ShaderDumpsMap>
ShaderDiskCache::LoadPrecompiled() {
if (!IsUsable())
return {};
FileUtil::IOFile file(GetPrecompiledPath(), "rb");
if (!file.IsOpen()) {
LOG_INFO(Render_OpenGL, "No precompiled shader cache found for game with title id={}",
GetTitleID());
return {};
}
const auto result = LoadPrecompiledFile(file);
if (!result) {
LOG_INFO(Render_OpenGL,
"Failed to load precompiled cache for game with title id={} - removing",
GetTitleID());
file.Close();
InvalidatePrecompiled();
return {};
}
return *result;
}
std::optional<std::pair<std::unordered_map<u64, ShaderDiskCacheDecompiled>, ShaderDumpsMap>>
ShaderDiskCache::LoadPrecompiledFile(FileUtil::IOFile& file) {
// Read compressed file from disk and decompress to virtual precompiled cache file
std::vector<u8> compressed(file.GetSize());
file.ReadBytes(compressed.data(), compressed.size());
const std::vector<u8> decompressed = Common::Compression::DecompressDataZSTD(compressed);
SaveArrayToPrecompiled(decompressed.data(), decompressed.size());
decompressed_precompiled_cache_offset = 0;
ShaderCacheVersionHash file_hash{};
if (!LoadArrayFromPrecompiled(file_hash.data(), file_hash.size())) {
return {};
}
if (GetShaderCacheVersionHash() != file_hash) {
LOG_INFO(Render_OpenGL, "Precompiled cache is from another version of the emulator");
return {};
}
std::unordered_map<u64, ShaderDiskCacheDecompiled> decompiled;
ShaderDumpsMap dumps;
while (decompressed_precompiled_cache_offset < decompressed_precompiled_cache.size()) {
PrecompiledEntryKind kind{};
if (!LoadObjectFromPrecompiled(kind)) {
return {};
}
switch (kind) {
case PrecompiledEntryKind::Decompiled: {
u64 unique_identifier{};
if (!LoadObjectFromPrecompiled(unique_identifier)) {
return {};
}
auto entry = LoadDecompiledEntry();
if (!entry) {
return {};
}
decompiled.insert({unique_identifier, std::move(*entry)});
break;
}
case PrecompiledEntryKind::Dump: {
u64 unique_identifier;
if (!LoadObjectFromPrecompiled(unique_identifier)) {
return {};
}
ShaderDiskCacheDump dump;
if (!LoadObjectFromPrecompiled(dump.binary_format)) {
return {};
}
u32 binary_length{};
if (!LoadObjectFromPrecompiled(binary_length)) {
return {};
}
dump.binary.resize(binary_length);
if (!LoadArrayFromPrecompiled(dump.binary.data(), dump.binary.size())) {
return {};
}
dumps.insert({unique_identifier, dump});
break;
}
default:
return {};
}
}
LOG_INFO(Render_OpenGL,
"Found a precompiled disk cache with {} decompiled entries and {} binary entries",
decompiled.size(), dumps.size());
return {{decompiled, dumps}};
}
std::optional<ShaderDiskCacheDecompiled> ShaderDiskCache::LoadDecompiledEntry() {
bool sanitize_mul;
if (!LoadObjectFromPrecompiled(sanitize_mul)) {
return {};
}
u32 code_size{};
if (!LoadObjectFromPrecompiled(code_size)) {
return {};
}
std::string code(code_size, '\0');
if (!LoadArrayFromPrecompiled(code.data(), code.size())) {
return {};
}
ShaderDiskCacheDecompiled entry;
entry.result.code = std::move(code);
entry.sanitize_mul = sanitize_mul;
return entry;
}
bool ShaderDiskCache::SaveDecompiledFile(u64 unique_identifier,
const ShaderDecompiler::ProgramResult& result,
bool sanitize_mul) {
if (!SaveObjectToPrecompiled(static_cast<u32>(PrecompiledEntryKind::Decompiled)) ||
!SaveObjectToPrecompiled(unique_identifier) || !SaveObjectToPrecompiled(sanitize_mul) ||
!SaveObjectToPrecompiled(static_cast<u32>(result.code.size())) ||
!SaveArrayToPrecompiled(result.code.data(), result.code.size())) {
return false;
}
return true;
}
void ShaderDiskCache::InvalidateAll() {
if (!FileUtil::Delete(GetTransferablePath())) {
LOG_ERROR(Render_OpenGL, "Failed to invalidate transferable file={}",
GetTransferablePath());
}
InvalidatePrecompiled();
}
void ShaderDiskCache::InvalidatePrecompiled() {
// Clear virtaul precompiled cache file
decompressed_precompiled_cache.resize(0);
if (!FileUtil::Delete(GetPrecompiledPath())) {
LOG_ERROR(Render_OpenGL, "Failed to invalidate precompiled file={}", GetPrecompiledPath());
}
}
void ShaderDiskCache::SaveRaw(const ShaderDiskCacheRaw& entry) {
if (!IsUsable())
return;
const u64 id = entry.GetUniqueIdentifier();
if (transferable.find(id) != transferable.end()) {
// The shader already exists
return;
}
FileUtil::IOFile file = AppendTransferableFile();
if (!file.IsOpen())
return;
if (file.WriteObject(TransferableEntryKind::Raw) != 1 || !entry.Save(file)) {
LOG_ERROR(Render_OpenGL, "Failed to save raw transferable cache entry - removing");
file.Close();
InvalidateAll();
return;
}
transferable.insert({id, entry});
}
void ShaderDiskCache::SaveDecompiled(u64 unique_identifier,
const ShaderDecompiler::ProgramResult& code,
bool sanitize_mul) {
if (!IsUsable())
return;
if (decompressed_precompiled_cache.size() == 0) {
SavePrecompiledHeaderToVirtualPrecompiledCache();
}
if (!SaveDecompiledFile(unique_identifier, code, sanitize_mul)) {
LOG_ERROR(Render_OpenGL,
"Failed to save decompiled entry to the precompiled file - removing");
InvalidatePrecompiled();
}
}
void ShaderDiskCache::SaveDump(u64 unique_identifier, GLuint program) {
if (!IsUsable())
return;
GLint binary_length{};
glGetProgramiv(program, GL_PROGRAM_BINARY_LENGTH, &binary_length);
GLenum binary_format{};
std::vector<u8> binary(binary_length);
glGetProgramBinary(program, binary_length, nullptr, &binary_format, binary.data());
if (!SaveObjectToPrecompiled(static_cast<u32>(PrecompiledEntryKind::Dump)) ||
!SaveObjectToPrecompiled(unique_identifier) ||
!SaveObjectToPrecompiled(static_cast<u32>(binary_format)) ||
!SaveObjectToPrecompiled(static_cast<u32>(binary_length)) ||
!SaveArrayToPrecompiled(binary.data(), binary.size())) {
LOG_ERROR(Render_OpenGL, "Failed to save binary program file in shader={:016x} - removing",
unique_identifier);
InvalidatePrecompiled();
return;
}
}
bool ShaderDiskCache::IsUsable() const {
return tried_to_load && Settings::values.use_disk_shader_cache;
}
FileUtil::IOFile ShaderDiskCache::AppendTransferableFile() {
if (!EnsureDirectories())
return {};
const auto transferable_path{GetTransferablePath()};
const bool existed = FileUtil::Exists(transferable_path);
FileUtil::IOFile file(transferable_path, "ab");
if (!file.IsOpen()) {
LOG_ERROR(Render_OpenGL, "Failed to open transferable cache in path={}", transferable_path);
return {};
}
if (!existed || file.GetSize() == 0) {
// If the file didn't exist, write its version
if (file.WriteObject(NativeVersion) != 1) {
LOG_ERROR(Render_OpenGL, "Failed to write transferable cache version in path={}",
transferable_path);
return {};
}
}
return file;
}
void ShaderDiskCache::SavePrecompiledHeaderToVirtualPrecompiledCache() {
const auto hash{GetShaderCacheVersionHash()};
if (!SaveArrayToPrecompiled(hash.data(), hash.size())) {
LOG_ERROR(
Render_OpenGL,
"Failed to write precompiled cache version hash to virtual precompiled cache file");
}
}
void ShaderDiskCache::SaveVirtualPrecompiledFile() {
decompressed_precompiled_cache_offset = 0;
const std::vector<u8>& compressed = Common::Compression::CompressDataZSTDDefault(
decompressed_precompiled_cache.data(), decompressed_precompiled_cache.size());
const auto precompiled_path{GetPrecompiledPath()};
FileUtil::IOFile file(precompiled_path, "wb");
if (!file.IsOpen()) {
LOG_ERROR(Render_OpenGL, "Failed to open precompiled cache in path={}", precompiled_path);
return;
}
if (file.WriteBytes(compressed.data(), compressed.size()) != compressed.size()) {
LOG_ERROR(Render_OpenGL, "Failed to write precompiled cache version in path={}",
precompiled_path);
return;
}
}
bool ShaderDiskCache::EnsureDirectories() const {
const auto CreateDir = [](const std::string& dir) {
if (!FileUtil::CreateDir(dir)) {
LOG_ERROR(Render_OpenGL, "Failed to create directory={}", dir);
return false;
}
return true;
};
return CreateDir(FileUtil::GetUserPath(FileUtil::UserPath::ShaderDir)) &&
CreateDir(GetBaseDir()) && CreateDir(GetTransferableDir()) &&
CreateDir(GetPrecompiledDir());
}
std::string ShaderDiskCache::GetTransferablePath() {
return FileUtil::SanitizePath(GetTransferableDir() + DIR_SEP_CHR + GetTitleID() + ".bin");
}
std::string ShaderDiskCache::GetPrecompiledPath() {
return FileUtil::SanitizePath(GetPrecompiledDir() + DIR_SEP_CHR + GetTitleID() + ".bin");
}
std::string ShaderDiskCache::GetTransferableDir() const {
return GetBaseDir() + DIR_SEP "transferable";
}
std::string ShaderDiskCache::GetPrecompiledDir() const {
return GetBaseDir() + DIR_SEP "precompiled";
}
std::string ShaderDiskCache::GetBaseDir() const {
return FileUtil::GetUserPath(FileUtil::UserPath::ShaderDir) + DIR_SEP "opengl";
}
u64 ShaderDiskCache::GetProgramID() {
// Skip games without title id
if (program_id != 0) {
return program_id;
}
if (Core::System::GetInstance().GetAppLoader().ReadProgramId(program_id) !=
Loader::ResultStatus::Success) {
return 0;
}
return program_id;
}
std::string ShaderDiskCache::GetTitleID() {
if (!title_id.empty()) {
return title_id;
}
title_id = fmt::format("{:016X}", GetProgramID());
return title_id;
}
} // namespace OpenGL

218
src/video_core/renderer_opengl/gl_shader_disk_cache.h Normal file
View file

@ -0,0 +1,218 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <algorithm>
#include <array>
#include <bitset>
#include <optional>
#include <string>
#include <tuple>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
#include <glad/glad.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/regs.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_opengl/gl_shader_gen.h"
namespace Core {
class System;
}
namespace FileUtil {
class IOFile;
}
namespace OpenGL {
struct ShaderDiskCacheDecompiled;
struct ShaderDiskCacheDump;
using RawShaderConfig = Pica::Regs;
using ProgramCode = std::vector<u32>;
using ShaderDecompiledMap = std::unordered_map<u64, ShaderDiskCacheDecompiled>;
using ShaderDumpsMap = std::unordered_map<u64, ShaderDiskCacheDump>;
/// Describes a shader how it's used by the guest GPU
class ShaderDiskCacheRaw {
public:
explicit ShaderDiskCacheRaw(u64 unique_identifier, ProgramType program_type,
RawShaderConfig config, ProgramCode program_code);
ShaderDiskCacheRaw() = default;
~ShaderDiskCacheRaw() = default;
bool Load(FileUtil::IOFile& file);
bool Save(FileUtil::IOFile& file) const;
u64 GetUniqueIdentifier() const {
return unique_identifier;
}
ProgramType GetProgramType() const {
return program_type;
}
const ProgramCode& GetProgramCode() const {
return program_code;
}
const RawShaderConfig& GetRawShaderConfig() const {
return config;
}
private:
u64 unique_identifier{};
ProgramType program_type{};
RawShaderConfig config{};
ProgramCode program_code{};
};
/// Contains decompiled data from a shader
struct ShaderDiskCacheDecompiled {
ShaderDecompiler::ProgramResult result;
bool sanitize_mul;
};
/// Contains an OpenGL dumped binary program
struct ShaderDiskCacheDump {
GLenum binary_format;
std::vector<u8> binary;
};
class ShaderDiskCache {
public:
explicit ShaderDiskCache(bool separable);
~ShaderDiskCache() = default;
/// Loads transferable cache. If file has a old version or on failure, it deletes the file.
std::optional<std::vector<ShaderDiskCacheRaw>> LoadTransferable();
/// Loads current game's precompiled cache. Invalidates on failure.
std::pair<ShaderDecompiledMap, ShaderDumpsMap> LoadPrecompiled();
/// Removes the transferable (and precompiled) cache file.
void InvalidateAll();
/// Removes the precompiled cache file and clears virtual precompiled cache file.
void InvalidatePrecompiled();
/// Saves a raw dump to the transferable file. Checks for collisions.
void SaveRaw(const ShaderDiskCacheRaw& entry);
/// Saves a decompiled entry to the precompiled file. Does not check for collisions.
void SaveDecompiled(u64 unique_identifier, const ShaderDecompiler::ProgramResult& code,
bool sanitize_mul);
/// Saves a dump entry to the precompiled file. Does not check for collisions.
void SaveDump(u64 unique_identifier, GLuint program);
/// Serializes virtual precompiled shader cache file to real file
void SaveVirtualPrecompiledFile();
private:
/// Loads the transferable cache. Returns empty on failure.
std::optional<std::pair<ShaderDecompiledMap, ShaderDumpsMap>> LoadPrecompiledFile(
FileUtil::IOFile& file);
/// Loads a decompiled cache entry from m_precompiled_cache_virtual_file. Returns empty on
/// failure.
std::optional<ShaderDiskCacheDecompiled> LoadDecompiledEntry();
/// Saves a decompiled entry to the passed file. Returns true on success.
bool SaveDecompiledFile(u64 unique_identifier, const ShaderDecompiler::ProgramResult& code,
bool sanitize_mul);
/// Returns if the cache can be used
bool IsUsable() const;
/// Opens current game's transferable file and write it's header if it doesn't exist
FileUtil::IOFile AppendTransferableFile();
/// Save precompiled header to precompiled_cache_in_memory
void SavePrecompiledHeaderToVirtualPrecompiledCache();
/// Create shader disk cache directories. Returns true on success.
bool EnsureDirectories() const;
/// Gets current game's transferable file path
std::string GetTransferablePath();
/// Gets current game's precompiled file path
std::string GetPrecompiledPath();
/// Get user's transferable directory path
std::string GetTransferableDir() const;
/// Get user's precompiled directory path
std::string GetPrecompiledDir() const;
/// Get user's shader directory path
std::string GetBaseDir() const;
/// Get current game's title id as u64
u64 GetProgramID();
/// Get current game's title id
std::string GetTitleID();
template <typename T>
bool SaveArrayToPrecompiled(const T* data, std::size_t length) {
const u8* data_view = reinterpret_cast<const u8*>(data);
decompressed_precompiled_cache.insert(decompressed_precompiled_cache.end(), &data_view[0],
&data_view[length * sizeof(T)]);
decompressed_precompiled_cache_offset += length * sizeof(T);
return true;
}
template <typename T>
bool LoadArrayFromPrecompiled(T* data, std::size_t length) {
u8* data_view = reinterpret_cast<u8*>(data);
std::copy_n(decompressed_precompiled_cache.data() + decompressed_precompiled_cache_offset,
length * sizeof(T), data_view);
decompressed_precompiled_cache_offset += length * sizeof(T);
return true;
}
template <typename T>
bool SaveObjectToPrecompiled(const T& object) {
return SaveArrayToPrecompiled(&object, 1);
}
bool SaveObjectToPrecompiled(bool object) {
const auto value = static_cast<u8>(object);
return SaveArrayToPrecompiled(&value, 1);
}
template <typename T>
bool LoadObjectFromPrecompiled(T& object) {
return LoadArrayFromPrecompiled(&object, 1);
}
// Stores whole precompiled cache which will be read from or saved to the precompiled chache
// file
std::vector<u8> decompressed_precompiled_cache;
// Stores the current offset of the precompiled cache file for IO purposes
std::size_t decompressed_precompiled_cache_offset = 0;
// Stored transferable shaders
std::unordered_map<u64, ShaderDiskCacheRaw> transferable;
// The cache has been loaded at boot
bool tried_to_load{};
bool separable{};
u64 program_id{};
std::string title_id;
};
} // namespace OpenGL

26
src/video_core/renderer_opengl/gl_shader_gen.cpp
View file

@ -1231,7 +1231,8 @@ float ProcTexNoiseCoef(vec2 x) {
}
}
std::string GenerateFragmentShader(const PicaFSConfig& config, bool separable_shader) {
ShaderDecompiler::ProgramResult GenerateFragmentShader(const PicaFSConfig& config,
bool separable_shader) {
const auto& state = config.state;
std::string out = R"(
@ -1482,7 +1483,7 @@ vec4 secondary_fragment_color = vec4(0.0);
// Do not do any sort of processing if it's obvious we're not going to pass the alpha test
if (state.alpha_test_func == FramebufferRegs::CompareFunc::Never) {
out += "discard; }";
return out;
return {out};
}
// Append the scissor test
@ -1546,7 +1547,7 @@ vec4 secondary_fragment_color = vec4(0.0);
"VideoCore_Pica_UseGasMode", true);
LOG_CRITICAL(Render_OpenGL, "Unimplemented gas mode");
out += "discard; }";
return out;
return {out};
}
if (state.shadow_rendering) {
@ -1584,10 +1585,10 @@ do {
out += "}";
return out;
return {out};
}
std::string GenerateTrivialVertexShader(bool separable_shader) {
ShaderDecompiler::ProgramResult GenerateTrivialVertexShader(bool separable_shader) {
std::string out = "";
if (separable_shader) {
out += "#extension GL_ARB_separate_shader_objects : enable\n";
@ -1630,11 +1631,11 @@ void main() {
}
)";
return out;
return {out};
}
std::optional<std::string> GenerateVertexShader(const Pica::Shader::ShaderSetup& setup,
const PicaVSConfig& config, bool separable_shader) {
std::optional<ShaderDecompiler::ProgramResult> GenerateVertexShader(
const Pica::Shader::ShaderSetup& setup, const PicaVSConfig& config, bool separable_shader) {
std::string out = "";
if (separable_shader) {
out += "#extension GL_ARB_separate_shader_objects : enable\n";
@ -1664,7 +1665,7 @@ std::optional<std::string> GenerateVertexShader(const Pica::Shader::ShaderSetup&
if (!program_source_opt)
return {};
std::string& program_source = *program_source_opt;
std::string& program_source = program_source_opt->code;
out += R"(
#define uniforms vs_uniforms
@ -1696,7 +1697,7 @@ layout (std140) uniform vs_config {
out += program_source;
return out;
return {{out}};
}
static std::string GetGSCommonSource(const PicaGSConfigCommonRaw& config, bool separable_shader) {
@ -1784,7 +1785,8 @@ void EmitPrim(Vertex vtx0, Vertex vtx1, Vertex vtx2) {
return out;
};
std::string GenerateFixedGeometryShader(const PicaFixedGSConfig& config, bool separable_shader) {
ShaderDecompiler::ProgramResult GenerateFixedGeometryShader(const PicaFixedGSConfig& config,
bool separable_shader) {
std::string out = "";
if (separable_shader) {
out += "#extension GL_ARB_separate_shader_objects : enable\n\n";
@ -1814,6 +1816,6 @@ void main() {
out += " EmitPrim(prim_buffer[0], prim_buffer[1], prim_buffer[2]);\n";
out += "}\n";
return out;
return {out};
}
} // namespace OpenGL

25
src/video_core/renderer_opengl/gl_shader_gen.h
View file

@ -16,6 +16,12 @@
namespace OpenGL {
namespace ShaderDecompiler {
struct ProgramResult;
}
enum class ProgramType : u32 { VS, GS, FS };
enum Attributes {
ATTRIBUTE_POSITION,
ATTRIBUTE_COLOR,
@ -161,8 +167,11 @@ struct PicaShaderConfigCommon {
* shader.
*/
struct PicaVSConfig : Common::HashableStruct<PicaShaderConfigCommon> {
explicit PicaVSConfig(const Pica::Regs& regs, Pica::Shader::ShaderSetup& setup) {
state.Init(regs.vs, setup);
explicit PicaVSConfig(const Pica::ShaderRegs& regs, Pica::Shader::ShaderSetup& setup) {
state.Init(regs, setup);
}
explicit PicaVSConfig(PicaShaderConfigCommon& conf) {
state = conf;
}
};
@ -197,20 +206,21 @@ struct PicaFixedGSConfig : Common::HashableStruct<PicaGSConfigCommonRaw> {
* @param separable_shader generates shader that can be used for separate shader object
* @returns String of the shader source code
*/
std::string GenerateTrivialVertexShader(bool separable_shader);
ShaderDecompiler::ProgramResult GenerateTrivialVertexShader(bool separable_shader);
/**
* Generates the GLSL vertex shader program source code for the given VS program
* @returns String of the shader source code; boost::none on failure
*/
std::optional<std::string> GenerateVertexShader(const Pica::Shader::ShaderSetup& setup,
const PicaVSConfig& config, bool separable_shader);
std::optional<ShaderDecompiler::ProgramResult> GenerateVertexShader(
const Pica::Shader::ShaderSetup& setup, const PicaVSConfig& config, bool separable_shader);
/*
* Generates the GLSL fixed geometry shader program source code for non-GS PICA pipeline
* @returns String of the shader source code
*/
std::string GenerateFixedGeometryShader(const PicaFixedGSConfig& config, bool separable_shader);
ShaderDecompiler::ProgramResult GenerateFixedGeometryShader(const PicaFixedGSConfig& config,
bool separable_shader);
/**
* Generates the GLSL fragment shader program source code for the current Pica state
@ -219,7 +229,8 @@ std::string GenerateFixedGeometryShader(const PicaFixedGSConfig& config, bool se
* @param separable_shader generates shader that can be used for separate shader object
* @returns String of the shader source code
*/
std::string GenerateFragmentShader(const PicaFSConfig& config, bool separable_shader);
ShaderDecompiler::ProgramResult GenerateFragmentShader(const PicaFSConfig& config,
bool separable_shader);
} // namespace OpenGL

358
src/video_core/renderer_opengl/gl_shader_manager.cpp
View file

@ -3,13 +3,80 @@
// Refer to the license.txt file included.
#include <algorithm>
#include <thread>
#include <unordered_map>
#include <boost/functional/hash.hpp>
#include <boost/variant.hpp>
#include "core/core.h"
#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
#include "video_core/renderer_opengl/gl_shader_manager.h"
#include "video_core/video_core.h"
namespace OpenGL {
static u64 GetUniqueIdentifier(const Pica::Regs& regs, const ProgramCode& code) {
std::size_t hash = 0;
u64 regs_uid = Common::ComputeHash64(regs.reg_array.data(), Pica::Regs::NUM_REGS * sizeof(u32));
boost::hash_combine(hash, regs_uid);
if (code.size() > 0) {
u64 code_uid = Common::ComputeHash64(code.data(), code.size() * sizeof(u32));
boost::hash_combine(hash, code_uid);
}
return static_cast<u64>(hash);
}
static OGLProgram GeneratePrecompiledProgram(const ShaderDiskCacheDump& dump,
const std::set<GLenum>& supported_formats) {
if (supported_formats.find(dump.binary_format) == supported_formats.end()) {
LOG_INFO(Render_OpenGL, "Precompiled cache entry with unsupported format - removing");
return {};
}
auto shader = OGLProgram();
shader.handle = glCreateProgram();
glProgramParameteri(shader.handle, GL_PROGRAM_SEPARABLE, GL_TRUE);
glProgramBinary(shader.handle, dump.binary_format, dump.binary.data(),
static_cast<GLsizei>(dump.binary.size()));
GLint link_status{};
glGetProgramiv(shader.handle, GL_LINK_STATUS, &link_status);
if (link_status == GL_FALSE) {
LOG_INFO(Render_OpenGL, "Precompiled cache rejected by the driver - removing");
return {};
}
return shader;
}
static std::set<GLenum> GetSupportedFormats() {
std::set<GLenum> supported_formats;
GLint num_formats{};
glGetIntegerv(GL_NUM_PROGRAM_BINARY_FORMATS, &num_formats);
std::vector<GLint> formats(num_formats);
glGetIntegerv(GL_PROGRAM_BINARY_FORMATS, formats.data());
for (const GLint format : formats)
supported_formats.insert(static_cast<GLenum>(format));
return supported_formats;
}
static std::tuple<PicaVSConfig, Pica::Shader::ShaderSetup> BuildVSConfigFromRaw(
const ShaderDiskCacheRaw& raw) {
Pica::Shader::ProgramCode program_code{};
Pica::Shader::SwizzleData swizzle_data{};
std::copy_n(raw.GetProgramCode().begin(), Pica::Shader::MAX_PROGRAM_CODE_LENGTH,
program_code.begin());
std::copy_n(raw.GetProgramCode().begin() + Pica::Shader::MAX_PROGRAM_CODE_LENGTH,
Pica::Shader::MAX_SWIZZLE_DATA_LENGTH, swizzle_data.begin());
Pica::Shader::ShaderSetup setup;
setup.program_code = program_code;
setup.swizzle_data = swizzle_data;
return {PicaVSConfig{raw.GetRawShaderConfig().vs, setup}, setup};
}
static void SetShaderUniformBlockBinding(GLuint shader, const char* name, UniformBindings binding,
std::size_t expected_size) {
const GLuint ub_index = glGetUniformBlockIndex(shader, name);
@ -121,6 +188,12 @@ public:
}
}
void Inject(OGLProgram&& program) {
SetShaderUniformBlockBindings(program.handle);
SetShaderSamplerBindings(program.handle);
shader_or_program = std::move(program);
}
private:
boost::variant<OGLShader, OGLProgram> shader_or_program;
};
@ -128,7 +201,7 @@ private:
class TrivialVertexShader {
public:
explicit TrivialVertexShader(bool separable) : program(separable) {
program.Create(GenerateTrivialVertexShader(separable).c_str(), GL_VERTEX_SHADER);
program.Create(GenerateTrivialVertexShader(separable).code.c_str(), GL_VERTEX_SHADER);
}
GLuint Get() const {
return program.GetHandle();
@ -138,18 +211,28 @@ private:
OGLShaderStage program;
};
template <typename KeyConfigType, std::string (*CodeGenerator)(const KeyConfigType&, bool),
template <typename KeyConfigType,
ShaderDecompiler::ProgramResult (*CodeGenerator)(const KeyConfigType&, bool),
GLenum ShaderType>
class ShaderCache {
public:
explicit ShaderCache(bool separable) : separable(separable) {}
GLuint Get(const KeyConfigType& config) {
std::tuple<GLuint, std::optional<ShaderDecompiler::ProgramResult>> Get(
const KeyConfigType& config) {
auto [iter, new_shader] = shaders.emplace(config, OGLShaderStage{separable});
OGLShaderStage& cached_shader = iter->second;
std::optional<ShaderDecompiler::ProgramResult> result{};
if (new_shader) {
cached_shader.Create(CodeGenerator(config, separable).c_str(), ShaderType);
result = CodeGenerator(config, separable);
cached_shader.Create(result->code.c_str(), ShaderType);
}
return cached_shader.GetHandle();
return {cached_shader.GetHandle(), result};
}
void Inject(const KeyConfigType& key, std::string decomp, OGLProgram&& program) {
OGLShaderStage stage{separable};
stage.Inject(std::move(program));
shaders.emplace(key, std::move(stage));
}
private:
@ -163,36 +246,47 @@ private:
// program buffer from the previous shader, which is hashed into the config, resulting several
// different config values from the same shader program.
template <typename KeyConfigType,
std::optional<std::string> (*CodeGenerator)(const Pica::Shader::ShaderSetup&,
const KeyConfigType&, bool),
std::optional<ShaderDecompiler::ProgramResult> (*CodeGenerator)(
const Pica::Shader::ShaderSetup&, const KeyConfigType&, bool),
GLenum ShaderType>
class ShaderDoubleCache {
public:
explicit ShaderDoubleCache(bool separable) : separable(separable) {}
GLuint Get(const KeyConfigType& key, const Pica::Shader::ShaderSetup& setup) {
std::tuple<GLuint, std::optional<ShaderDecompiler::ProgramResult>> Get(
const KeyConfigType& key, const Pica::Shader::ShaderSetup& setup) {
std::optional<ShaderDecompiler::ProgramResult> result{};
auto map_it = shader_map.find(key);
if (map_it == shader_map.end()) {
auto program_opt = CodeGenerator(setup, key, separable);
if (!program_opt) {
shader_map[key] = nullptr;
return 0;
return {0, {}};
}
std::string& program = *program_opt;
std::string& program = program_opt->code;
auto [iter, new_shader] = shader_cache.emplace(program, OGLShaderStage{separable});
OGLShaderStage& cached_shader = iter->second;
if (new_shader) {
result->code = program;
cached_shader.Create(program.c_str(), ShaderType);
}
shader_map[key] = &cached_shader;
return cached_shader.GetHandle();
return {cached_shader.GetHandle(), result};
}
if (map_it->second == nullptr) {
return 0;
return {0, {}};
}
return map_it->second->GetHandle();
return {map_it->second->GetHandle(), {}};
}
void Inject(const KeyConfigType& key, std::string decomp, OGLProgram&& program) {
OGLShaderStage stage{separable};
stage.Inject(std::move(program));
auto [iter, new_shader] = shader_cache.emplace(decomp, std::move(stage));
OGLShaderStage& cached_shader = iter->second;
shader_map[key] = &cached_shader;
}
private:
@ -214,7 +308,7 @@ public:
explicit Impl(bool separable, bool is_amd)
: is_amd(is_amd), separable(separable), programmable_vertex_shaders(separable),
trivial_vertex_shader(separable), fixed_geometry_shaders(separable),
fragment_shaders(separable) {
fragment_shaders(separable), disk_cache(separable) {
if (separable)
pipeline.Create();
}
@ -244,6 +338,7 @@ public:
};
bool is_amd;
bool separable;
ShaderTuple current;
@ -253,10 +348,9 @@ public:
FixedGeometryShaders fixed_geometry_shaders;
FragmentShaders fragment_shaders;
bool separable;
std::unordered_map<ShaderTuple, OGLProgram, ShaderTuple::Hash> program_cache;
OGLPipeline pipeline;
ShaderDiskCache disk_cache;
};
ShaderProgramManager::ShaderProgramManager(bool separable, bool is_amd)
@ -264,12 +358,23 @@ ShaderProgramManager::ShaderProgramManager(bool separable, bool is_amd)
ShaderProgramManager::~ShaderProgramManager() = default;
bool ShaderProgramManager::UseProgrammableVertexShader(const PicaVSConfig& config,
const Pica::Shader::ShaderSetup setup) {
GLuint handle = impl->programmable_vertex_shaders.Get(config, setup);
bool ShaderProgramManager::UseProgrammableVertexShader(const Pica::Regs& regs,
Pica::Shader::ShaderSetup& setup) {
PicaVSConfig config{regs.vs, setup};
auto [handle, result] = impl->programmable_vertex_shaders.Get(config, setup);
if (handle == 0)
return false;
impl->current.vs = handle;
// Save VS to the disk cache if its a new shader
if (result) {
auto& disk_cache = impl->disk_cache;
ProgramCode program_code{setup.program_code.begin(), setup.program_code.end()};
program_code.insert(program_code.end(), setup.swizzle_data.begin(),
setup.swizzle_data.end());
u64 unique_identifier = GetUniqueIdentifier(regs, program_code);
ShaderDiskCacheRaw raw{unique_identifier, ProgramType::VS, regs, program_code};
disk_cache.SaveRaw(raw);
}
return true;
}
@ -277,25 +382,36 @@ void ShaderProgramManager::UseTrivialVertexShader() {
impl->current.vs = impl->trivial_vertex_shader.Get();
}
void ShaderProgramManager::UseFixedGeometryShader(const PicaFixedGSConfig& config) {
impl->current.gs = impl->fixed_geometry_shaders.Get(config);
void ShaderProgramManager::UseFixedGeometryShader(const Pica::Regs& regs) {
PicaFixedGSConfig gs_config(regs);
auto [handle, _] = impl->fixed_geometry_shaders.Get(gs_config);
impl->current.gs = handle;
}
void ShaderProgramManager::UseTrivialGeometryShader() {
impl->current.gs = 0;
}
void ShaderProgramManager::UseFragmentShader(const PicaFSConfig& config) {
impl->current.fs = impl->fragment_shaders.Get(config);
void ShaderProgramManager::UseFragmentShader(const Pica::Regs& regs) {
PicaFSConfig config = PicaFSConfig::BuildFromRegs(regs);
auto [handle, result] = impl->fragment_shaders.Get(config);
impl->current.fs = handle;
// Save FS to the disk cache if its a new shader
if (result) {
auto& disk_cache = impl->disk_cache;
u64 unique_identifier = GetUniqueIdentifier(regs, {});
ShaderDiskCacheRaw raw{unique_identifier, ProgramType::FS, regs, {}};
disk_cache.SaveRaw(raw);
disk_cache.SaveDecompiled(unique_identifier, *result, false);
}
}
void ShaderProgramManager::ApplyTo(OpenGLState& state) {
if (impl->separable) {
if (impl->is_amd) {
// Without this reseting, AMD sometimes freezes when one stage is changed but not for
// the others.
// On the other hand, including this reset seems to introduce memory leak in Intel
// Graphics.
// Without this reseting, AMD sometimes freezes when one stage is changed but not
// for the others. On the other hand, including this reset seems to introduce memory
// leak in Intel Graphics.
glUseProgramStages(
impl->pipeline.handle,
GL_VERTEX_SHADER_BIT | GL_GEOMETRY_SHADER_BIT | GL_FRAGMENT_SHADER_BIT, 0);
@ -316,4 +432,192 @@ void ShaderProgramManager::ApplyTo(OpenGLState& state) {
state.draw.shader_program = cached_program.handle;
}
}
void ShaderProgramManager::LoadDiskCache(const std::atomic_bool& stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) {
if (!impl->separable) {
LOG_ERROR(Render_OpenGL,
"Cannot load disk cache as separate shader programs are unsupported!");
return;
}
auto& disk_cache = impl->disk_cache;
const auto transferable = disk_cache.LoadTransferable();
if (!transferable) {
return;
}
const auto raws = *transferable;
auto [decompiled, dumps] = disk_cache.LoadPrecompiled();
if (stop_loading) {
return;
}
std::set<GLenum> supported_formats = GetSupportedFormats();
// Track if precompiled cache was altered during loading to know if we have to serialize the
// virtual precompiled cache file back to the hard drive
bool precompiled_cache_altered = false;
std::mutex mutex;
std::size_t built_shaders = 0; // It doesn't have be atomic since it's used behind a mutex
std::atomic_bool compilation_failed = false;
if (callback) {
callback(VideoCore::LoadCallbackStage::Decompile, 0, raws.size());
}
std::vector<std::size_t> load_raws_index;
// Loads both decompiled and precompiled shaders from the cache. If either one is missing for
const auto LoadPrecompiledWorker =
[&](std::size_t begin, std::size_t end, const std::vector<ShaderDiskCacheRaw>& raws,
const ShaderDecompiledMap& decompiled, const ShaderDumpsMap& dumps) {
for (std::size_t i = 0; i < end; ++i) {
if (stop_loading || compilation_failed) {
return;
}
const auto& raw{raws[i]};
const u64 unique_identifier{raw.GetUniqueIdentifier()};
const u64 calculated_hash =
GetUniqueIdentifier(raw.GetRawShaderConfig(), raw.GetProgramCode());
if (unique_identifier != calculated_hash) {
LOG_ERROR(Render_OpenGL,
"Invalid hash in entry={:016x} (obtained hash={:016x}) - removing "
"shader cache",
raw.GetUniqueIdentifier(), calculated_hash);
disk_cache.InvalidateAll();
return;
}
const auto dump{dumps.find(unique_identifier)};
const auto decomp{decompiled.find(unique_identifier)};
OGLProgram shader;
if (dump != dumps.end() && decomp != decompiled.end()) {
// Only load this shader if its sanitize_mul setting matches
if (decomp->second.sanitize_mul == VideoCore::g_hw_shader_accurate_mul) {
continue;
}
// If the shader is dumped, attempt to load it
shader = GeneratePrecompiledProgram(dump->second, supported_formats);
if (shader.handle == 0) {
// If any shader failed, stop trying to compile, delete the cache, and start
// loading from raws
compilation_failed = true;
return;
}
// we have both the binary shader and the decompiled, so inject it into the
// cache
if (raw.GetProgramType() == ProgramType::VS) {
auto [conf, setup] = BuildVSConfigFromRaw(raw);
std::scoped_lock lock(mutex);
impl->programmable_vertex_shaders.Inject(conf, decomp->second.result.code,
std::move(shader));
} else if (raw.GetProgramType() == ProgramType::FS) {
PicaFSConfig conf = PicaFSConfig::BuildFromRegs(raw.GetRawShaderConfig());
std::scoped_lock lock(mutex);
impl->fragment_shaders.Inject(conf, decomp->second.result.code,
std::move(shader));
} else {
// Unsupported shader type got stored somehow so nuke the cache
LOG_CRITICAL(Frontend, "failed to load raw programtype {}",
static_cast<u32>(raw.GetProgramType()));
compilation_failed = true;
return;
}
} else {
// Since precompiled didn't have the dump, we'll load them in the next phase
std::scoped_lock lock(mutex);
load_raws_index.push_back(i);
}
if (callback) {
callback(VideoCore::LoadCallbackStage::Decompile, i, raws.size());
}
}
};
LoadPrecompiledWorker(0, raws.size(), raws, decompiled, dumps);
if (compilation_failed) {
// Invalidate the precompiled cache if a shader dumped shader was rejected
disk_cache.InvalidatePrecompiled();
dumps.clear();
precompiled_cache_altered = true;
}
if (callback) {
callback(VideoCore::LoadCallbackStage::Build, 0, raws.size());
}
compilation_failed = false;
const auto LoadTransferable = [&](std::size_t begin, std::size_t end,
const std::vector<ShaderDiskCacheRaw>& raws) {
for (std::size_t i = 0; i < end; ++i) {
if (stop_loading || compilation_failed) {
return;
}
const auto& raw{raws[i]};
const u64 unique_identifier{raw.GetUniqueIdentifier()};
bool sanitize_mul = false;
GLuint handle{0};
std::optional<ShaderDecompiler::ProgramResult> result;
// Otherwise decompile and build the shader at boot and save the result to the
// precompiled file
if (raw.GetProgramType() == ProgramType::VS) {
// TODO: This isn't the ideal place to lock, since we really only want to
// lock access to the shared cache
auto [conf, setup] = BuildVSConfigFromRaw(raw);
std::scoped_lock lock(mutex);
auto [h, r] = impl->programmable_vertex_shaders.Get(conf, setup);
handle = h;
result = r;
sanitize_mul = conf.state.sanitize_mul;
} else if (raw.GetProgramType() == ProgramType::FS) {
PicaFSConfig conf = PicaFSConfig::BuildFromRegs(raw.GetRawShaderConfig());
std::scoped_lock lock(mutex);
auto [h, r] = impl->fragment_shaders.Get(conf);
handle = h;
result = r;
} else {
// Unsupported shader type got stored somehow so nuke the cache
LOG_ERROR(Frontend, "failed to load raw programtype {}",
static_cast<u32>(raw.GetProgramType()));
compilation_failed = true;
return;
}
if (handle == 0) {
LOG_ERROR(Frontend, "compilation from raw failed {:x} {:x}",
raw.GetProgramCode().at(0), raw.GetProgramCode().at(1));
compilation_failed = true;
return;
}
// If this is a new shader, add it the precompiled cache
if (result) {
disk_cache.SaveDecompiled(unique_identifier, *result, sanitize_mul);
disk_cache.SaveDump(unique_identifier, handle);
precompiled_cache_altered = true;
}
if (callback) {
callback(VideoCore::LoadCallbackStage::Build, i, raws.size());
}
}
};
LoadTransferable(0, raws.size(), raws);
if (compilation_failed) {
disk_cache.InvalidateAll();
}
if (precompiled_cache_altered) {
disk_cache.SaveVirtualPrecompiledFile();
}
} // namespace OpenGL
} // namespace OpenGL

15
src/video_core/renderer_opengl/gl_shader_manager.h
View file

@ -6,12 +6,17 @@
#include <memory>
#include <glad/glad.h>
#include "video_core/rasterizer_interface.h"
#include "video_core/regs_lighting.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_gen.h"
#include "video_core/renderer_opengl/gl_state.h"
#include "video_core/renderer_opengl/pica_to_gl.h"
namespace Core {
class System;
}
namespace OpenGL {
enum class UniformBindings : GLuint { Common, VS, GS };
@ -97,16 +102,18 @@ public:
ShaderProgramManager(bool separable, bool is_amd);
~ShaderProgramManager();
bool UseProgrammableVertexShader(const PicaVSConfig& config,
const Pica::Shader::ShaderSetup setup);
void LoadDiskCache(const std::atomic_bool& stop_loading,
const VideoCore::DiskResourceLoadCallback& callback);
bool UseProgrammableVertexShader(const Pica::Regs& config, Pica::Shader::ShaderSetup& setup);
void UseTrivialVertexShader();
void UseFixedGeometryShader(const PicaFixedGSConfig& config);
void UseFixedGeometryShader(const Pica::Regs& regs);
void UseTrivialGeometryShader();
void UseFragmentShader(const PicaFSConfig& config);
void UseFragmentShader(const Pica::Regs& config);
void ApplyTo(OpenGLState& state);

10
src/video_core/renderer_opengl/renderer_opengl.cpp
View file

@ -1039,9 +1039,9 @@ static void APIENTRY DebugHandler(GLenum source, GLenum type, GLuint id, GLenum
}
/// Initialize the renderer
Core::System::ResultStatus RendererOpenGL::Init() {
VideoCore::ResultStatus RendererOpenGL::Init() {
if (!gladLoadGL()) {
return Core::System::ResultStatus::ErrorVideoCore_ErrorBelowGL33;
return VideoCore::ResultStatus::ErrorBelowGL33;
}
if (GLAD_GL_KHR_debug) {
@ -1063,18 +1063,18 @@ Core::System::ResultStatus RendererOpenGL::Init() {
telemetry_session.AddField(Telemetry::FieldType::UserSystem, "GPU_OpenGL_Version", gl_version);
if (!strcmp(gpu_vendor, "GDI Generic")) {
return Core::System::ResultStatus::ErrorVideoCore_ErrorGenericDrivers;
return VideoCore::ResultStatus::ErrorGenericDrivers;
}
if (!(GLAD_GL_VERSION_3_3 || GLAD_GL_ES_VERSION_3_1)) {
return Core::System::ResultStatus::ErrorVideoCore_ErrorBelowGL33;
return VideoCore::ResultStatus::ErrorBelowGL33;
}
InitOpenGLObjects();
RefreshRasterizerSetting();
return Core::System::ResultStatus::Success;
return VideoCore::ResultStatus::Success;
}
/// Shutdown the renderer

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

@ -48,7 +48,7 @@ public:
~RendererOpenGL() override;
/// Initialize the renderer
Core::System::ResultStatus Init() override;
VideoCore::ResultStatus Init() override;
/// Shutdown the renderer
void ShutDown() override;

6
src/video_core/shader/shader.h
View file

@ -26,6 +26,8 @@ namespace Pica::Shader {
constexpr unsigned MAX_PROGRAM_CODE_LENGTH = 4096;
constexpr unsigned MAX_SWIZZLE_DATA_LENGTH = 4096;
using ProgramCode = std::array<u32, MAX_PROGRAM_CODE_LENGTH>;
using SwizzleData = std::array<u32, MAX_SWIZZLE_DATA_LENGTH>;
struct AttributeBuffer {
alignas(16) Common::Vec4<float24> attr[16];
@ -196,8 +198,8 @@ struct Uniforms {
struct ShaderSetup {
Uniforms uniforms;
std::array<u32, MAX_PROGRAM_CODE_LENGTH> program_code;
std::array<u32, MAX_SWIZZLE_DATA_LENGTH> swizzle_data;
ProgramCode program_code;
SwizzleData swizzle_data;
/// Data private to ShaderEngines
struct EngineData {

7
src/video_core/video_core.cpp
View file

@ -22,6 +22,7 @@ std::atomic<bool> g_hw_renderer_enabled;
std::atomic<bool> g_shader_jit_enabled;
std::atomic<bool> g_hw_shader_enabled;
std::atomic<bool> g_hw_shader_accurate_mul;
std::atomic<bool> g_use_disk_shader_cache;
std::atomic<bool> g_renderer_bg_color_update_requested;
std::atomic<bool> g_renderer_sampler_update_requested;
std::atomic<bool> g_renderer_shader_update_requested;
@ -34,16 +35,16 @@ Layout::FramebufferLayout g_screenshot_framebuffer_layout;
Memory::MemorySystem* g_memory;
/// Initialize the video core
Core::System::ResultStatus Init(Frontend::EmuWindow& emu_window, Memory::MemorySystem& memory) {
ResultStatus Init(Frontend::EmuWindow& emu_window, Memory::MemorySystem& memory) {
g_memory = &memory;
Pica::Init();
OpenGL::GLES = Settings::values.use_gles;
g_renderer = std::make_unique<OpenGL::RendererOpenGL>(emu_window);
Core::System::ResultStatus result = g_renderer->Init();
ResultStatus result = g_renderer->Init();
if (result != Core::System::ResultStatus::Success) {
if (result != ResultStatus::Success) {
LOG_ERROR(Render, "initialization failed !");
} else {
LOG_DEBUG(Render, "initialized OK");

10
src/video_core/video_core.h
View file

@ -6,7 +6,6 @@
#include <atomic>
#include <memory>
#include "core/core.h"
#include "core/frontend/emu_window.h"
namespace Frontend {
@ -32,6 +31,7 @@ extern std::atomic<bool> g_hw_renderer_enabled;
extern std::atomic<bool> g_shader_jit_enabled;
extern std::atomic<bool> g_hw_shader_enabled;
extern std::atomic<bool> g_hw_shader_accurate_mul;
extern std::atomic<bool> g_use_disk_shader_cache;
extern std::atomic<bool> g_renderer_bg_color_update_requested;
extern std::atomic<bool> g_renderer_sampler_update_requested;
extern std::atomic<bool> g_renderer_shader_update_requested;
@ -43,8 +43,14 @@ extern Layout::FramebufferLayout g_screenshot_framebuffer_layout;
extern Memory::MemorySystem* g_memory;
enum class ResultStatus {
Success,
ErrorGenericDrivers,
ErrorBelowGL33,
};
/// Initialize the video core
Core::System::ResultStatus Init(Frontend::EmuWindow& emu_window, Memory::MemorySystem& memory);
ResultStatus Init(Frontend::EmuWindow& emu_window, Memory::MemorySystem& memory);
/// Shutdown the video core
void Shutdown();