General: Fix clang build

Allows building on clang to work again

.gitmodules

@@ -1,6 +1,6 @@
 [submodule "inih"]
     path = externals/inih/inih
-    url = https://github.com/svn2github/inih
+    url = https://github.com/benhoyt/inih.git
 [submodule "cubeb"]
     path = externals/cubeb
     url = https://github.com/kinetiknz/cubeb.git

CMakeLists.txt

@@ -161,7 +161,7 @@ macro(yuzu_find_packages)
     #    Cmake Pkg Prefix  Version     Conan Pkg
         "Boost             1.73        boost/1.73.0"
         "Catch2            2.13        catch2/2.13.0"
-        "fmt               7.0         fmt/7.0.3"
+        "fmt               7.1         fmt/7.1.0"
     # can't use until https://github.com/bincrafters/community/issues/1173
         #"libzip            1.5         libzip/1.5.2@bincrafters/stable"
         "lz4               1.8         lz4/1.9.2"
@@ -263,6 +263,7 @@ if (CONAN_REQUIRED_LIBS)
         libzip:with_openssl=False
         libzip:enable_windows_crypto=False
     )
+
     conan_check(VERSION 1.24.0 REQUIRED)
     # Add the bincrafters remote
     conan_add_remote(NAME bincrafters
@@ -354,6 +355,19 @@ if (NOT LIBUSB_FOUND)
     set(LIBUSB_LIBRARIES usb)
 endif()
 
+# Use system installed ffmpeg.
+if (NOT MSVC)
+    find_package(FFmpeg REQUIRED)
+else()
+    set(FFMPEG_EXT_NAME "ffmpeg-4.2.1")
+    set(FFMPEG_PATH "${CMAKE_BINARY_DIR}/externals/${FFMPEG_EXT_NAME}")
+    download_bundled_external("ffmpeg/" ${FFMPEG_EXT_NAME} "")
+    set(FFMPEG_FOUND YES)
+    set(FFMPEG_INCLUDE_DIR "${FFMPEG_PATH}/include" CACHE PATH "Path to FFmpeg headers" FORCE)
+    set(FFMPEG_LIBRARY_DIR "${FFMPEG_PATH}/bin" CACHE PATH "Path to FFmpeg library" FORCE)
+    set(FFMPEG_DLL_DIR "${FFMPEG_PATH}/bin" CACHE PATH "Path to FFmpeg dll's" FORCE)
+endif()
+
 # Prefer the -pthread flag on Linux.
 set(THREADS_PREFER_PTHREAD_FLAG ON)
 find_package(Threads REQUIRED)

CMakeModules/CopyYuzuFFmpegDeps.cmake

@@ -0,0 +1,10 @@
+function(copy_yuzu_FFmpeg_deps target_dir)
+    include(WindowsCopyFiles)
+    set(DLL_DEST "${CMAKE_BINARY_DIR}/bin/$<CONFIG>/")
+    windows_copy_files(${target_dir} ${FFMPEG_DLL_DIR} ${DLL_DEST}
+        avcodec-58.dll
+        avutil-56.dll
+        swresample-3.dll
+        swscale-5.dll
+    )
+endfunction(copy_yuzu_FFmpeg_deps)

externals/CMakeLists.txt

@@ -73,17 +73,20 @@ if (NOT LIBZIP_FOUND)
 endif()
 
 if (ENABLE_WEB_SERVICE)
-    # LibreSSL
-    set(LIBRESSL_SKIP_INSTALL ON CACHE BOOL "")
-    add_subdirectory(libressl EXCLUDE_FROM_ALL)
-    target_include_directories(ssl INTERFACE ./libressl/include)
-    target_compile_definitions(ssl PRIVATE -DHAVE_INET_NTOP)
-    get_directory_property(OPENSSL_LIBRARIES
-        DIRECTORY libressl
-        DEFINITION OPENSSL_LIBS)
-
-    # lurlparser
-    add_subdirectory(lurlparser EXCLUDE_FROM_ALL)
+    find_package(OpenSSL 1.1)
+    if (OPENSSL_FOUND)
+        set(OPENSSL_LIBRARIES OpenSSL::SSL OpenSSL::Crypto)
+    else()
+        # LibreSSL
+        set(LIBRESSL_SKIP_INSTALL ON CACHE BOOL "")
+        set(OPENSSLDIR "/etc/ssl/")
+        add_subdirectory(libressl EXCLUDE_FROM_ALL)
+        target_include_directories(ssl INTERFACE ./libressl/include)
+        target_compile_definitions(ssl PRIVATE -DHAVE_INET_NTOP)
+        get_directory_property(OPENSSL_LIBRARIES
+            DIRECTORY libressl
+            DEFINITION OPENSSL_LIBS)
+    endif()
 
     # httplib
     add_library(httplib INTERFACE)

externals/find-modules/FindFFmpeg.cmake

@@ -0,0 +1,100 @@
+# - Try to find ffmpeg libraries (libavcodec, libavformat and libavutil)
+# Once done this will define
+#
+# FFMPEG_FOUND - system has ffmpeg or libav
+# FFMPEG_INCLUDE_DIR - the ffmpeg include directory
+# FFMPEG_LIBRARIES - Link these to use ffmpeg
+# FFMPEG_LIBAVCODEC
+# FFMPEG_LIBAVFORMAT
+# FFMPEG_LIBAVUTIL
+#
+# Copyright (c) 2008 Andreas Schneider <mail@cynapses.org>
+# Modified for other libraries by Lasse Kärkkäinen <tronic>
+# Modified for Hedgewars by Stepik777
+# Modified for FFmpeg-example Tuukka Pasanen 2018
+# Modified for yuzu toastUnlimted 2020
+#
+# Redistribution and use is allowed according to the terms of the New
+# BSD license.
+#
+
+include(FindPackageHandleStandardArgs)
+
+find_package_handle_standard_args(FFMPEG
+  FOUND_VAR FFMPEG_FOUND
+  REQUIRED_VARS
+      FFMPEG_LIBRARY
+      FFMPEG_INCLUDE_DIR
+  VERSION_VAR FFMPEG_VERSION
+)
+
+if(FFMPEG_LIBRARIES AND FFMPEG_INCLUDE_DIR)
+  # in cache already
+  set(FFMPEG_FOUND TRUE)
+else()
+  # use pkg-config to get the directories and then use these values
+  # in the FIND_PATH() and FIND_LIBRARY() calls
+  find_package(PkgConfig)
+  if(PKG_CONFIG_FOUND)
+    pkg_check_modules(_FFMPEG_AVCODEC libavcodec)
+    pkg_check_modules(_FFMPEG_AVUTIL libavutil)
+    pkg_check_modules(_FFMPEG_SWSCALE libswscale)
+  endif()
+
+  find_path(FFMPEG_AVCODEC_INCLUDE_DIR
+    NAMES libavcodec/avcodec.h
+    PATHS ${_FFMPEG_AVCODEC_INCLUDE_DIRS}
+      /usr/include
+      /usr/local/include
+      /opt/local/include
+      /sw/include
+    PATH_SUFFIXES ffmpeg libav)
+
+  find_library(FFMPEG_LIBAVCODEC
+    NAMES avcodec
+    PATHS ${_FFMPEG_AVCODEC_LIBRARY_DIRS}
+      /usr/lib
+      /usr/local/lib
+      /opt/local/lib
+      /sw/lib)
+
+  find_library(FFMPEG_LIBAVUTIL
+    NAMES avutil
+    PATHS ${_FFMPEG_AVUTIL_LIBRARY_DIRS}
+      /usr/lib
+      /usr/local/lib
+      /opt/local/lib
+      /sw/lib)
+
+  find_library(FFMPEG_LIBSWSCALE
+    NAMES swscale
+    PATHS ${_FFMPEG_SWSCALE_LIBRARY_DIRS}
+      /usr/lib
+      /usr/local/lib
+      /opt/local/lib
+      /sw/lib)
+
+  if(FFMPEG_LIBAVCODEC AND FFMPEG_LIBAVUTIL AND FFMPEG_LIBSWSCALE)
+    set(FFMPEG_FOUND TRUE)
+  endif()
+
+  if(FFMPEG_FOUND)
+    set(FFMPEG_INCLUDE_DIR ${FFMPEG_AVCODEC_INCLUDE_DIR})
+    set(FFMPEG_LIBRARIES
+      ${FFMPEG_LIBAVCODEC}
+      ${FFMPEG_LIBAVUTIL}
+      ${FFMPEG_LIBSWSCALE})
+  endif()
+
+  if(FFMPEG_FOUND)
+    if(NOT FFMPEG_FIND_QUIETLY)
+      message(STATUS
+      "Found FFMPEG or Libav: ${FFMPEG_LIBRARIES}, ${FFMPEG_INCLUDE_DIR}")
+    endif()
+  else()
+    if(FFMPEG_FIND_REQUIRED)
+      message(FATAL_ERROR
+      "Could not find libavcodec or libavutil or libswscale")
+    endif()
+  endif()
+endif()

externals/httplib/README.md

@@ -1,4 +1,4 @@
-From https://github.com/yhirose/cpp-httplib/tree/fce8e6fefdab4ad48bc5b25c98e5ebfda4f3cf53
+From https://github.com/yhirose/cpp-httplib/tree/ff5677ad197947177c158fe857caff4f0e242045 with https://github.com/yhirose/cpp-httplib/pull/701
 
 MIT License
 

externals/lurlparser/CMakeLists.txt

@@ -1,8 +0,0 @@
-add_library(lurlparser
-        LUrlParser.cpp
-        LUrlParser.h
-)
-
-create_target_directory_groups(lurlparser)
-
-target_include_directories(lurlparser INTERFACE ${CMAKE_CURRENT_SOURCE_DIR})

externals/lurlparser/LUrlParser.cpp

@@ -1,265 +0,0 @@
-/*
- * Lightweight URL & URI parser (RFC 1738, RFC 3986)
- * https://github.com/corporateshark/LUrlParser
- *
- * The MIT License (MIT)
- *
- * Copyright (C) 2015 Sergey Kosarevsky (sk@linderdaum.com)
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-#include "LUrlParser.h"
-
-#include <algorithm>
-#include <cstring>
-#include <stdlib.h>
-
-// check if the scheme name is valid
-static bool IsSchemeValid( const std::string& SchemeName )
-{
-    for ( auto c : SchemeName  )
-    {
-        if ( !isalpha( c ) && c != '+' && c != '-' && c != '.' ) return false;
-    }
-
-    return true;
-}
-
-bool LUrlParser::clParseURL::GetPort( int* OutPort ) const
-{
-    if ( !IsValid() ) { return false; }
-
-    int Port = atoi( m_Port.c_str() );
-
-    if ( Port <= 0 || Port > 65535 ) { return false; }
-
-    if ( OutPort ) { *OutPort = Port; }
-
-    return true;
-}
-
-// based on RFC 1738 and RFC 3986
-LUrlParser::clParseURL LUrlParser::clParseURL::ParseURL( const std::string& URL )
-{
-    LUrlParser::clParseURL Result;
-
-    const char* CurrentString = URL.c_str();
-
-    /*
-     *	<scheme>:<scheme-specific-part>
-     *	<scheme> := [a-z\+\-\.]+
-     *	For resiliency, programs interpreting URLs should treat upper case letters as equivalent to lower case in scheme names
-     */
-
-    // try to read scheme
-    {
-        const char* LocalString = strchr( CurrentString, ':' );
-
-        if ( !LocalString )
-        {
-            return clParseURL( LUrlParserError_NoUrlCharacter );
-        }
-
-        // save the scheme name
-        Result.m_Scheme = std::string( CurrentString, LocalString - CurrentString );
-
-        if ( !IsSchemeValid( Result.m_Scheme ) )
-        {
-            return clParseURL( LUrlParserError_InvalidSchemeName );
-        }
-
-        // scheme should be lowercase
-        std::transform( Result.m_Scheme.begin(), Result.m_Scheme.end(), Result.m_Scheme.begin(), ::tolower );
-
-        // skip ':'
-        CurrentString = LocalString+1;
-    }
-
-    /*
-     *	//<user>:<password>@<host>:<port>/<url-path>
-     *	any ":", "@" and "/" must be normalized
-     */
-
-    // skip "//"
-    if ( *CurrentString++ != '/' ) return clParseURL( LUrlParserError_NoDoubleSlash );
-    if ( *CurrentString++ != '/' ) return clParseURL( LUrlParserError_NoDoubleSlash );
-
-    // check if the user name and password are specified
-    bool bHasUserName = false;
-
-    const char* LocalString = CurrentString;
-
-    while ( *LocalString )
-    {
-        if ( *LocalString == '@' )
-        {
-            // user name and password are specified
-            bHasUserName = true;
-            break;
-        }
-        else if ( *LocalString == '/' )
-        {
-            // end of <host>:<port> specification
-            bHasUserName = false;
-            break;
-        }
-
-        LocalString++;
-    }
-
-    // user name and password
-    LocalString = CurrentString;
-
-    if ( bHasUserName )
-    {
-        // read user name
-        while ( *LocalString && *LocalString != ':' && *LocalString != '@' ) LocalString++;
-
-        Result.m_UserName = std::string( CurrentString, LocalString - CurrentString );
-
-        // proceed with the current pointer
-        CurrentString = LocalString;
-
-        if ( *CurrentString == ':' )
-        {
-            // skip ':'
-            CurrentString++;
-
-            // read password
-            LocalString = CurrentString;
-
-            while ( *LocalString && *LocalString != '@' ) LocalString++;
-
-            Result.m_Password = std::string( CurrentString, LocalString - CurrentString );
-
-            CurrentString = LocalString;
-        }
-
-        // skip '@'
-        if ( *CurrentString != '@' )
-        {
-            return clParseURL( LUrlParserError_NoAtSign );
-        }
-
-        CurrentString++;
-    }
-
-    bool bHasBracket = ( *CurrentString == '[' );
-
-    // go ahead, read the host name
-    LocalString = CurrentString;
-
-    while ( *LocalString )
-    {
-        if ( bHasBracket && *LocalString == ']' )
-        {
-            // end of IPv6 address
-            LocalString++;
-            break;
-        }
-        else if ( !bHasBracket && ( *LocalString == ':' || *LocalString == '/' ) )
-        {
-            // port number is specified
-            break;
-        }
-
-        LocalString++;
-    }
-
-    Result.m_Host = std::string( CurrentString, LocalString - CurrentString );
-
-    CurrentString = LocalString;
-
-    // is port number specified?
-    if ( *CurrentString == ':' )
-    {
-        CurrentString++;
-
-        // read port number
-        LocalString = CurrentString;
-
-        while ( *LocalString && *LocalString != '/' ) LocalString++;
-
-        Result.m_Port = std::string( CurrentString, LocalString - CurrentString );
-
-        CurrentString = LocalString;
-    }
-
-    // end of string
-    if ( !*CurrentString )
-    {
-        Result.m_ErrorCode = LUrlParserError_Ok;
-
-        return Result;
-    }
-
-    // skip '/'
-    if ( *CurrentString != '/' )
-    {
-        return clParseURL( LUrlParserError_NoSlash );
-    }
-
-    CurrentString++;
-
-    // parse the path
-    LocalString = CurrentString;
-
-    while ( *LocalString && *LocalString != '#' && *LocalString != '?' ) LocalString++;
-
-    Result.m_Path = std::string( CurrentString, LocalString - CurrentString );
-
-    CurrentString = LocalString;
-
-    // check for query
-    if ( *CurrentString == '?' )
-    {
-        // skip '?'
-        CurrentString++;
-
-        // read query
-        LocalString = CurrentString;
-
-        while ( *LocalString && *LocalString != '#' ) LocalString++;
-
-        Result.m_Query = std::string( CurrentString, LocalString - CurrentString );
-
-        CurrentString = LocalString;
-    }
-
-    // check for fragment
-    if ( *CurrentString == '#' )
-    {
-        // skip '#'
-        CurrentString++;
-
-        // read fragment
-        LocalString = CurrentString;
-
-        while ( *LocalString ) LocalString++;
-
-        Result.m_Fragment = std::string( CurrentString, LocalString - CurrentString );
-
-        CurrentString = LocalString;
-    }
-
-    Result.m_ErrorCode = LUrlParserError_Ok;
-
-    return Result;
-}

externals/lurlparser/LUrlParser.h

@@ -1,78 +0,0 @@
-/*
- * Lightweight URL & URI parser (RFC 1738, RFC 3986)
- * https://github.com/corporateshark/LUrlParser
- *
- * The MIT License (MIT)
- *
- * Copyright (C) 2015 Sergey Kosarevsky (sk@linderdaum.com)
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-#pragma once
-
-#include <string>
-
-namespace LUrlParser
-{
-enum LUrlParserError
-{
-    LUrlParserError_Ok = 0,
-    LUrlParserError_Uninitialized = 1,
-    LUrlParserError_NoUrlCharacter = 2,
-    LUrlParserError_InvalidSchemeName = 3,
-    LUrlParserError_NoDoubleSlash = 4,
-    LUrlParserError_NoAtSign = 5,
-    LUrlParserError_UnexpectedEndOfLine = 6,
-    LUrlParserError_NoSlash = 7,
-};
-
-class clParseURL
-{
-public:
-    LUrlParserError m_ErrorCode;
-    std::string m_Scheme;
-    std::string m_Host;
-    std::string m_Port;
-    std::string m_Path;
-    std::string m_Query;
-    std::string m_Fragment;
-    std::string m_UserName;
-    std::string m_Password;
-
-    clParseURL()
-            : m_ErrorCode( LUrlParserError_Uninitialized )
-    {}
-
-    /// return 'true' if the parsing was successful
-    bool IsValid() const { return m_ErrorCode == LUrlParserError_Ok; }
-
-    /// helper to convert the port number to int, return 'true' if the port is valid (within the 0..65535 range)
-    bool GetPort( int* OutPort ) const;
-
-    /// parse the URL
-    static clParseURL ParseURL( const std::string& URL );
-
-private:
-    explicit clParseURL( LUrlParserError ErrorCode )
-            : m_ErrorCode( ErrorCode )
-    {}
-};
-
-} // namespace LUrlParser

externals/lurlparser/README.md

@@ -1,19 +0,0 @@
-From https://github.com/corporateshark/LUrlParser/commit/455d5e2d27e3946f11ad0328fee9ee2628e6a8e2
-
-MIT License
-
-===
-
-Lightweight URL & URI parser (RFC 1738, RFC 3986)
-
-(C) Sergey Kosarevsky, 2015
-
-@corporateshark sk@linderdaum.com
-
-http://www.linderdaum.com
-
-http://blog.linderdaum.com
-
-=============================
-
-A tiny and lightweight URL & URI parser (RFC 1738, RFC 3986) written in C++.

externals/microprofile/microprofile.h

@@ -902,8 +902,10 @@ inline uint16_t MicroProfileGetGroupIndex(MicroProfileToken t)
 #include <windows.h>
 #define snprintf _snprintf
 
+#ifdef _MSC_VER
 #pragma warning(push)
 #pragma warning(disable: 4244)
+#endif
 int64_t MicroProfileTicksPerSecondCpu()
 {
     static int64_t nTicksPerSecond = 0;
@@ -946,7 +948,11 @@ typedef HANDLE MicroProfileThread;
 DWORD _stdcall ThreadTrampoline(void* pFunc)
 {
     MicroProfileThreadFunc F = (MicroProfileThreadFunc)pFunc;
-    return (uint32_t)F(0);
+
+    // The return value of F will always return a void*, however, this is for
+    // compatibility with pthreads. The underlying "address" of the pointer
+    // is always a 32-bit value, so this cast is safe to perform.
+    return static_cast<DWORD>(reinterpret_cast<uint64_t>(F(0)));
 }
 
 inline void MicroProfileThreadStart(MicroProfileThread* pThread, MicroProfileThreadFunc Func)
@@ -1742,10 +1748,10 @@ void MicroProfileFlip()
                             }
                         }
                     }
-                    for(uint32_t i = 0; i < MICROPROFILE_MAX_GROUPS; ++i)
+                    for(uint32_t j = 0; j < MICROPROFILE_MAX_GROUPS; ++j)
                     {
-                        pLog->nGroupTicks[i] += nGroupTicks[i];
-                        pFrameGroup[i] += nGroupTicks[i];
+                        pLog->nGroupTicks[j] += nGroupTicks[j];
+                        pFrameGroup[j] += nGroupTicks[j];
                     }
                     pLog->nStackPos = nStackPos;
                 }
@@ -3328,7 +3334,7 @@ bool MicroProfileIsLocalThread(uint32_t nThreadId)
 #endif
 #else
 
-bool MicroProfileIsLocalThread(uint32_t nThreadId){return false;}
+bool MicroProfileIsLocalThread([[maybe_unused]] uint32_t nThreadId) { return false; }
 void MicroProfileStopContextSwitchTrace(){}
 void MicroProfileStartContextSwitchTrace(){}
 
@@ -3576,7 +3582,7 @@ int MicroProfileGetGpuTickReference(int64_t* pOutCpu, int64_t* pOutGpu)
 
 #undef S
 
-#ifdef _WIN32
+#ifdef _MSC_VER
 #pragma warning(pop)
 #endif
 

src/CMakeLists.txt

@@ -32,7 +32,6 @@ if (MSVC)
     # /Zc:inline          - Let codegen omit inline functions in object files
     # /Zc:throwingNew     - Let codegen assume `operator new` (without std::nothrow) will never return null
     add_compile_options(
-        /W3
         /MP
         /Zi
         /Zo
@@ -43,6 +42,13 @@ if (MSVC)
         /Zc:externConstexpr
         /Zc:inline
         /Zc:throwingNew
+
+        # Warnings
+        /W3
+        /we4547 # 'operator' : operator before comma has no effect; expected operator with side-effect
+        /we4549 # 'operator1': operator before comma has no effect; did you intend 'operator2'?
+        /we4555 # Expression has no effect; expected expression with side-effect
+        /we4834 # Discarding return value of function with 'nodiscard' attribute
     )
 
     # /GS- - No stack buffer overflow checks
@@ -56,6 +62,7 @@ else()
         -Werror=implicit-fallthrough
         -Werror=missing-declarations
         -Werror=reorder
+        -Werror=unused-result
         -Wextra
         -Wmissing-declarations
         -Wno-attributes

src/common/CMakeLists.txt

@@ -150,6 +150,8 @@ add_library(common STATIC
     scope_exit.h
     spin_lock.cpp
     spin_lock.h
+    stream.cpp
+    stream.h
     string_util.cpp
     string_util.h
     swap.h
@@ -188,6 +190,22 @@ if(ARCHITECTURE_x86_64)
     )
 endif()
 
+if (MSVC)
+  target_compile_definitions(common PRIVATE
+    # The standard library doesn't provide any replacement for codecvt yet
+    # so we can disable this deprecation warning for the time being.
+    _SILENCE_CXX17_CODECVT_HEADER_DEPRECATION_WARNING
+  )
+  target_compile_options(common PRIVATE
+    /W4
+    /WX
+  )
+else()
+  target_compile_options(common PRIVATE
+    -Werror
+  )
+endif()
+
 create_target_directory_groups(common)
 find_package(Boost 1.71 COMPONENTS context headers REQUIRED)
 

src/common/fiber.cpp

@@ -79,9 +79,9 @@ void Fiber::Exit() {
     released = true;
 }
 
-void Fiber::SetRewindPoint(std::function<void(void*)>&& rewind_func, void* start_parameter) {
+void Fiber::SetRewindPoint(std::function<void(void*)>&& rewind_func, void* rewind_param) {
     rewind_point = std::move(rewind_func);
-    rewind_parameter = start_parameter;
+    rewind_parameter = rewind_param;
 }
 
 void Fiber::Rewind() {
@@ -91,7 +91,7 @@ void Fiber::Rewind() {
     SwitchToFiber(impl->rewind_handle);
 }
 
-void Fiber::YieldTo(std::shared_ptr<Fiber>& from, std::shared_ptr<Fiber>& to) {
+void Fiber::YieldTo(std::shared_ptr<Fiber> from, std::shared_ptr<Fiber> to) {
     ASSERT_MSG(from != nullptr, "Yielding fiber is null!");
     ASSERT_MSG(to != nullptr, "Next fiber is null!");
     to->guard.lock();
@@ -161,9 +161,9 @@ Fiber::Fiber(std::function<void(void*)>&& entry_point_func, void* start_paramete
         boost::context::detail::make_fcontext(stack_base, impl->stack.size(), FiberStartFunc);
 }
 
-void Fiber::SetRewindPoint(std::function<void(void*)>&& rewind_func, void* start_parameter) {
+void Fiber::SetRewindPoint(std::function<void(void*)>&& rewind_func, void* rewind_param) {
     rewind_point = std::move(rewind_func);
-    rewind_parameter = start_parameter;
+    rewind_parameter = rewind_param;
 }
 
 Fiber::Fiber() : impl{std::make_unique<FiberImpl>()} {}
@@ -199,7 +199,7 @@ void Fiber::Rewind() {
     boost::context::detail::jump_fcontext(impl->rewind_context, this);
 }
 
-void Fiber::YieldTo(std::shared_ptr<Fiber>& from, std::shared_ptr<Fiber>& to) {
+void Fiber::YieldTo(std::shared_ptr<Fiber> from, std::shared_ptr<Fiber> to) {
     ASSERT_MSG(from != nullptr, "Yielding fiber is null!");
     ASSERT_MSG(to != nullptr, "Next fiber is null!");
     to->guard.lock();

src/common/fiber.h

@@ -41,15 +41,15 @@ public:
     Fiber(const Fiber&) = delete;
     Fiber& operator=(const Fiber&) = delete;
 
-    Fiber(Fiber&&) = default;
-    Fiber& operator=(Fiber&&) = default;
+    Fiber(Fiber&&) = delete;
+    Fiber& operator=(Fiber&&) = delete;
 
     /// Yields control from Fiber 'from' to Fiber 'to'
     /// Fiber 'from' must be the currently running fiber.
-    static void YieldTo(std::shared_ptr<Fiber>& from, std::shared_ptr<Fiber>& to);
+    static void YieldTo(std::shared_ptr<Fiber> from, std::shared_ptr<Fiber> to);
     [[nodiscard]] static std::shared_ptr<Fiber> ThreadToFiber();
 
-    void SetRewindPoint(std::function<void(void*)>&& rewind_func, void* start_parameter);
+    void SetRewindPoint(std::function<void(void*)>&& rewind_func, void* rewind_param);
 
     void Rewind();
 

src/common/file_util.cpp

@@ -472,13 +472,14 @@ u64 ScanDirectoryTree(const std::string& directory, FSTEntry& parent_entry,
 }
 
 bool DeleteDirRecursively(const std::string& directory, unsigned int recursion) {
-    const auto callback = [recursion](u64* num_entries_out, const std::string& directory,
-                                      const std::string& virtual_name) -> bool {
-        std::string new_path = directory + DIR_SEP_CHR + virtual_name;
+    const auto callback = [recursion](u64*, const std::string& directory,
+                                      const std::string& virtual_name) {
+        const std::string new_path = directory + DIR_SEP_CHR + virtual_name;
 
         if (IsDirectory(new_path)) {
-            if (recursion == 0)
+            if (recursion == 0) {
                 return false;
+            }
             return DeleteDirRecursively(new_path, recursion - 1);
         }
         return Delete(new_path);
@@ -492,7 +493,8 @@ bool DeleteDirRecursively(const std::string& directory, unsigned int recursion)
     return true;
 }
 
-void CopyDir(const std::string& source_path, const std::string& dest_path) {
+void CopyDir([[maybe_unused]] const std::string& source_path,
+             [[maybe_unused]] const std::string& dest_path) {
 #ifndef _WIN32
     if (source_path == dest_path) {
         return;
@@ -553,7 +555,7 @@ std::optional<std::string> GetCurrentDir() {
     std::string strDir = dir;
 #endif
     free(dir);
-    return std::move(strDir);
+    return strDir;
 }
 
 bool SetCurrentDir(const std::string& directory) {
@@ -772,21 +774,23 @@ 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) {
-    const std::string forbidden_characters = ".\"/\\[]:;=, ";
+    static constexpr std::string_view forbidden_characters = ".\"/\\[]:;=, ";
 
     // On a FAT32 partition, 8.3 names are stored as a 11 bytes array, filled with spaces.
     short_name = {{' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', '\0'}};
     extension = {{' ', ' ', ' ', '\0'}};
 
-    std::string::size_type point = filename.rfind('.');
-    if (point == filename.size() - 1)
+    auto point = filename.rfind('.');
+    if (point == filename.size() - 1) {
         point = filename.rfind('.', point);
+    }
 
     // Get short name.
     int j = 0;
     for (char letter : filename.substr(0, point)) {
-        if (forbidden_characters.find(letter, 0) != std::string::npos)
+        if (forbidden_characters.find(letter, 0) != std::string::npos) {
             continue;
+        }
         if (j == 8) {
             // TODO(Link Mauve): also do that for filenames containing a space.
             // TODO(Link Mauve): handle multiple files having the same short name.
@@ -794,14 +798,15 @@ void SplitFilename83(const std::string& filename, std::array<char, 9>& short_nam
             short_name[7] = '1';
             break;
         }
-        short_name[j++] = toupper(letter);
+        short_name[j++] = static_cast<char>(std::toupper(letter));
     }
 
     // Get extension.
     if (point != std::string::npos) {
         j = 0;
-        for (char letter : filename.substr(point + 1, 3))
-            extension[j++] = toupper(letter);
+        for (char letter : filename.substr(point + 1, 3)) {
+            extension[j++] = static_cast<char>(std::toupper(letter));
+        }
     }
 }
 

src/common/file_util.h

@@ -232,7 +232,7 @@ public:
 
     void Swap(IOFile& other) noexcept;
 
-    [[nodiscard]] bool Open(const std::string& filename, const char openmode[], int flags = 0);
+    bool Open(const std::string& filename, const char openmode[], int flags = 0);
     bool Close();
 
     template <typename T>

src/common/logging/backend.cpp

@@ -274,7 +274,6 @@ const char* GetLogClassName(Class log_class) {
     case Class::Count:
         break;
     }
-    UNREACHABLE();
     return "Invalid";
 }
 
@@ -293,7 +292,6 @@ const char* GetLevelName(Level log_level) {
         break;
     }
 #undef LVL
-    UNREACHABLE();
     return "Invalid";
 }
 

src/common/misc.cpp

@@ -16,16 +16,23 @@
 // Call directly after the command or use the error num.
 // This function might change the error code.
 std::string GetLastErrorMsg() {
-    static const std::size_t buff_size = 255;
+    static constexpr std::size_t buff_size = 255;
     char err_str[buff_size];
 
 #ifdef _WIN32
     FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM, nullptr, GetLastError(),
                    MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), err_str, buff_size, nullptr);
+    return std::string(err_str, buff_size);
+#elif defined(__GLIBC__) && (_GNU_SOURCE || (_POSIX_C_SOURCE < 200112L && _XOPEN_SOURCE < 600))
+    // Thread safe (GNU-specific)
+    const char* str = strerror_r(errno, err_str, buff_size);
+    return std::string(str);
 #else
     // Thread safe (XSI-compliant)
-    strerror_r(errno, err_str, buff_size);
+    const int success = strerror_r(errno, err_str, buff_size);
+    if (success != 0) {
+        return {};
+    }
+    return std::string(err_str);
 #endif
-
-    return std::string(err_str, buff_size);
 }

src/common/spin_lock.h

@@ -15,6 +15,14 @@ namespace Common {
  */
 class SpinLock {
 public:
+    SpinLock() = default;
+
+    SpinLock(const SpinLock&) = delete;
+    SpinLock& operator=(const SpinLock&) = delete;
+
+    SpinLock(SpinLock&&) = delete;
+    SpinLock& operator=(SpinLock&&) = delete;
+
     void lock();
     void unlock();
     [[nodiscard]] bool try_lock();

src/common/stream.cpp

@@ -0,0 +1,47 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <stdexcept>
+#include "common/common_types.h"
+#include "common/stream.h"
+
+namespace Common {
+
+Stream::Stream() = default;
+Stream::~Stream() = default;
+
+void Stream::Seek(s32 offset, SeekOrigin origin) {
+    if (origin == SeekOrigin::SetOrigin) {
+        if (offset < 0) {
+            position = 0;
+        } else if (position >= buffer.size()) {
+            position = buffer.size();
+        } else {
+            position = offset;
+        }
+    } else if (origin == SeekOrigin::FromCurrentPos) {
+        Seek(static_cast<s32>(position) + offset, SeekOrigin::SetOrigin);
+    } else if (origin == SeekOrigin::FromEnd) {
+        Seek(static_cast<s32>(buffer.size()) - offset, SeekOrigin::SetOrigin);
+    }
+}
+
+u8 Stream::ReadByte() {
+    if (position < buffer.size()) {
+        return buffer[position++];
+    } else {
+        throw std::out_of_range("Attempting to read a byte not within the buffer range");
+    }
+}
+
+void Stream::WriteByte(u8 byte) {
+    if (position == buffer.size()) {
+        buffer.push_back(byte);
+        position++;
+    } else {
+        buffer.insert(buffer.begin() + position, byte);
+    }
+}
+
+} // namespace Common

src/common/stream.h

@@ -0,0 +1,56 @@
+// Copyright 2020 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 {
+
+enum class SeekOrigin {
+    SetOrigin,
+    FromCurrentPos,
+    FromEnd,
+};
+
+class Stream {
+public:
+    /// Stream creates a bitstream and provides common functionality on the stream.
+    explicit Stream();
+    ~Stream();
+
+    Stream(const Stream&) = delete;
+    Stream& operator=(const Stream&) = delete;
+
+    Stream(Stream&&) = default;
+    Stream& operator=(Stream&&) = default;
+
+    /// Reposition bitstream "cursor" to the specified offset from origin
+    void Seek(s32 offset, SeekOrigin origin);
+
+    /// Reads next byte in the stream buffer and increments position
+    u8 ReadByte();
+
+    /// Writes byte at current position
+    void WriteByte(u8 byte);
+
+    [[nodiscard]] std::size_t GetPosition() const {
+        return position;
+    }
+
+    [[nodiscard]] std::vector<u8>& GetBuffer() {
+        return buffer;
+    }
+
+    [[nodiscard]] const std::vector<u8>& GetBuffer() const {
+        return buffer;
+    }
+
+private:
+    std::vector<u8> buffer;
+    std::size_t position{0};
+};
+
+} // namespace Common

src/common/string_util.cpp

@@ -8,6 +8,7 @@
 #include <cstdlib>
 #include <locale>
 #include <sstream>
+
 #include "common/common_paths.h"
 #include "common/logging/log.h"
 #include "common/string_util.h"
@@ -21,14 +22,14 @@ namespace Common {
 /// Make a string lowercase
 std::string ToLower(std::string str) {
     std::transform(str.begin(), str.end(), str.begin(),
-                   [](unsigned char c) { return std::tolower(c); });
+                   [](unsigned char c) { return static_cast<char>(std::tolower(c)); });
     return str;
 }
 
 /// Make a string uppercase
 std::string ToUpper(std::string str) {
     std::transform(str.begin(), str.end(), str.begin(),
-                   [](unsigned char c) { return std::toupper(c); });
+                   [](unsigned char c) { return static_cast<char>(std::toupper(c)); });
     return str;
 }
 

src/common/timer.cpp

@@ -142,20 +142,18 @@ std::string Timer::GetTimeFormatted() {
 // ----------------
 double Timer::GetDoubleTime() {
     // Get continuous timestamp
-    u64 TmpSeconds = static_cast<u64>(Common::Timer::GetTimeSinceJan1970().count());
-    double ms = static_cast<u64>(GetTimeMs().count()) % 1000;
+    auto tmp_seconds = static_cast<u64>(GetTimeSinceJan1970().count());
+    const auto ms = static_cast<double>(static_cast<u64>(GetTimeMs().count()) % 1000);
 
     // Remove a few years. We only really want enough seconds to make
     // sure that we are detecting actual actions, perhaps 60 seconds is
     // enough really, but I leave a year of seconds anyway, in case the
     // user's clock is incorrect or something like that.
-    TmpSeconds = TmpSeconds - (38 * 365 * 24 * 60 * 60);
+    tmp_seconds = tmp_seconds - (38 * 365 * 24 * 60 * 60);
 
     // Make a smaller integer that fits in the double
-    u32 Seconds = static_cast<u32>(TmpSeconds);
-    double TmpTime = Seconds + ms;
-
-    return TmpTime;
+    const auto seconds = static_cast<u32>(tmp_seconds);
+    return seconds + ms;
 }
 
 } // Namespace Common

src/common/wall_clock.cpp

@@ -53,7 +53,7 @@ public:
         return Common::Divide128On32(temporary, 1000000000).first;
     }
 
-    void Pause(bool is_paused) override {
+    void Pause([[maybe_unused]] bool is_paused) override {
         // Do nothing in this clock type.
     }
 

src/common/x64/native_clock.h

@@ -34,7 +34,7 @@ private:
     /// value used to reduce the native clocks accuracy as some apss rely on
     /// undefined behavior where the level of accuracy in the clock shouldn't
     /// be higher.
-    static constexpr u64 inaccuracy_mask = ~(0x400 - 1);
+    static constexpr u64 inaccuracy_mask = ~(UINT64_C(0x400) - 1);
 
     SpinLock rtsc_serialize{};
     u64 last_measure{};

src/core/CMakeLists.txt

@@ -439,6 +439,8 @@ add_library(core STATIC
     hle/service/nvdrv/devices/nvhost_gpu.h
     hle/service/nvdrv/devices/nvhost_nvdec.cpp
     hle/service/nvdrv/devices/nvhost_nvdec.h
+    hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
+    hle/service/nvdrv/devices/nvhost_nvdec_common.h
     hle/service/nvdrv/devices/nvhost_nvjpg.cpp
     hle/service/nvdrv/devices/nvhost_nvjpg.h
     hle/service/nvdrv/devices/nvhost_vic.cpp
@@ -452,6 +454,8 @@ add_library(core STATIC
     hle/service/nvdrv/nvdrv.h
     hle/service/nvdrv/nvmemp.cpp
     hle/service/nvdrv/nvmemp.h
+    hle/service/nvdrv/syncpoint_manager.cpp
+    hle/service/nvdrv/syncpoint_manager.h
     hle/service/nvflinger/buffer_queue.cpp
     hle/service/nvflinger/buffer_queue.h
     hle/service/nvflinger/nvflinger.cpp

src/core/arm/arm_interface.cpp

@@ -147,10 +147,18 @@ std::vector<ARM_Interface::BacktraceEntry> ARM_Interface::GetBacktraceFromContex
     auto fp = ctx.cpu_registers[29];
     auto lr = ctx.cpu_registers[30];
     while (true) {
-        out.push_back({"", 0, lr, 0});
-        if (!fp) {
+        out.push_back({
+            .module = "",
+            .address = 0,
+            .original_address = lr,
+            .offset = 0,
+            .name = {},
+        });
+
+        if (fp == 0) {
             break;
         }
+
         lr = memory.Read64(fp + 8) - 4;
         fp = memory.Read64(fp);
     }

src/core/core.cpp

@@ -179,16 +179,18 @@ struct System::Impl {
         arp_manager.ResetAll();
 
         telemetry_session = std::make_unique<Core::TelemetrySession>();
+
+        gpu_core = VideoCore::CreateGPU(emu_window, system);
+        if (!gpu_core) {
+            return ResultStatus::ErrorVideoCore;
+        }
+
         service_manager = std::make_shared<Service::SM::ServiceManager>(kernel);
 
         Service::Init(service_manager, system);
         GDBStub::DeferStart();
 
         interrupt_manager = std::make_unique<Core::Hardware::InterruptManager>(system);
-        gpu_core = VideoCore::CreateGPU(emu_window, system);
-        if (!gpu_core) {
-            return ResultStatus::ErrorVideoCore;
-        }
 
         // Initialize time manager, which must happen after kernel is created
         time_manager.Initialize();

src/core/cpu_manager.cpp

@@ -365,6 +365,8 @@ void CpuManager::RunThread(std::size_t core) {
     data.enter_barrier.reset();
     data.exit_barrier.reset();
     data.initialized = false;
+
+    MicroProfileOnThreadExit();
 }
 
 } // namespace Core

src/core/frontend/applets/controller.cpp

@@ -19,7 +19,7 @@ DefaultControllerApplet::DefaultControllerApplet(Service::SM::ServiceManager& se
 DefaultControllerApplet::~DefaultControllerApplet() = default;
 
 void DefaultControllerApplet::ReconfigureControllers(std::function<void()> callback,
-                                                     ControllerParameters parameters) const {
+                                                     const ControllerParameters& parameters) const {
     LOG_INFO(Service_HID, "called, deducing the best configuration based on the given parameters!");
 
     auto& npad =

src/core/frontend/applets/controller.h

@@ -38,7 +38,7 @@ public:
     virtual ~ControllerApplet();
 
     virtual void ReconfigureControllers(std::function<void()> callback,
-                                        ControllerParameters parameters) const = 0;
+                                        const ControllerParameters& parameters) const = 0;
 };
 
 class DefaultControllerApplet final : public ControllerApplet {
@@ -47,7 +47,7 @@ public:
     ~DefaultControllerApplet() override;
 
     void ReconfigureControllers(std::function<void()> callback,
-                                ControllerParameters parameters) const override;
+                                const ControllerParameters& parameters) const override;
 
 private:
     Service::SM::ServiceManager& service_manager;

src/core/hle/kernel/kernel.cpp

@@ -86,8 +86,6 @@ struct KernelCore::Impl {
         }
         cores.clear();
 
-        registered_core_threads.reset();
-
         process_list.clear();
         current_process = nullptr;
 
@@ -199,9 +197,7 @@ struct KernelCore::Impl {
         const auto it = std::find(register_host_thread_keys.begin(), end, this_id);
         ASSERT(core_id < Core::Hardware::NUM_CPU_CORES);
         ASSERT(it == end);
-        ASSERT(!registered_core_threads[core_id]);
         InsertHostThread(static_cast<u32>(core_id));
-        registered_core_threads.set(core_id);
     }
 
     void RegisterHostThread() {
@@ -332,7 +328,6 @@ struct KernelCore::Impl {
 
     // 0-3 IDs represent core threads, >3 represent others
     std::atomic<u32> registered_thread_ids{Core::Hardware::NUM_CPU_CORES};
-    std::bitset<Core::Hardware::NUM_CPU_CORES> registered_core_threads;
 
     // Number of host threads is a relatively high number to avoid overflowing
     static constexpr size_t NUM_REGISTRABLE_HOST_THREADS = 64;

src/core/hle/kernel/process.cpp

@@ -4,6 +4,7 @@
 
 #include <algorithm>
 #include <bitset>
+#include <ctime>
 #include <memory>
 #include <random>
 #include "common/alignment.h"
@@ -123,7 +124,7 @@ std::shared_ptr<Process> Process::Create(Core::System& system, std::string name,
                                                               : kernel.CreateNewUserProcessID();
     process->capabilities.InitializeForMetadatalessProcess();
 
-    std::mt19937 rng(Settings::values.rng_seed.GetValue().value_or(0));
+    std::mt19937 rng(Settings::values.rng_seed.GetValue().value_or(std::time(nullptr)));
     std::uniform_int_distribution<u64> distribution;
     std::generate(process->random_entropy.begin(), process->random_entropy.end(),
                   [&] { return distribution(rng); });

src/core/hle/service/am/am.cpp

@@ -1201,6 +1201,8 @@ IApplicationFunctions::IApplicationFunctions(Core::System& system_)
         {151, nullptr, "TryPopFromNotificationStorageChannel"},
         {160, nullptr, "GetHealthWarningDisappearedSystemEvent"},
         {170, nullptr, "SetHdcpAuthenticationActivated"},
+        {180, nullptr, "GetLaunchRequiredVersion"},
+        {181, nullptr, "UpgradeLaunchRequiredVersion"},
         {500, nullptr, "StartContinuousRecordingFlushForDebug"},
         {1000, nullptr, "CreateMovieMaker"},
         {1001, nullptr, "PrepareForJit"},

src/core/hle/service/hid/hid.cpp

@@ -260,7 +260,7 @@ Hid::Hid(Core::System& system) : ServiceFramework("hid"), system(system) {
         {404, nullptr, "HasLeftRightBattery"},
         {405, nullptr, "GetNpadInterfaceType"},
         {406, nullptr, "GetNpadLeftRightInterfaceType"},
-        {407, nullptr, "GetNpadOfHighestBatteryLevelForJoyLeft"},
+        {407, nullptr, "GetNpadOfHighestBatteryLevel"},
         {408, nullptr, "GetNpadOfHighestBatteryLevelForJoyRight"},
         {500, nullptr, "GetPalmaConnectionHandle"},
         {501, nullptr, "InitializePalma"},

src/core/hle/service/ldr/ldr.cpp

@@ -166,7 +166,7 @@ public:
             {0, &RelocatableObject::LoadNro, "LoadNro"},
             {1, &RelocatableObject::UnloadNro, "UnloadNro"},
             {2, &RelocatableObject::LoadNrr, "LoadNrr"},
-            {3, nullptr, "UnloadNrr"},
+            {3, &RelocatableObject::UnloadNrr, "UnloadNrr"},
             {4, &RelocatableObject::Initialize, "Initialize"},
             {10, nullptr, "LoadNrrEx"},
         };
@@ -272,6 +272,20 @@ public:
         rb.Push(RESULT_SUCCESS);
     }
 
+    void UnloadNrr(Kernel::HLERequestContext& ctx) {
+        IPC::RequestParser rp{ctx};
+        const auto pid = rp.Pop<u64>();
+        const auto nrr_address = rp.Pop<VAddr>();
+
+        LOG_DEBUG(Service_LDR, "called with pid={}, nrr_address={:016X}", pid, nrr_address);
+
+        nrr.erase(nrr_address);
+
+        IPC::ResponseBuilder rb{ctx, 2};
+
+        rb.Push(RESULT_SUCCESS);
+    }
+
     bool ValidateRegionForMap(Kernel::Memory::PageTable& page_table, VAddr start,
                               std::size_t size) const {
         constexpr std::size_t padding_size{4 * Kernel::Memory::PageSize};

src/core/hle/service/mii/mii.cpp

@@ -47,6 +47,7 @@ public:
             {23, nullptr, "Convert"},
             {24, nullptr, "ConvertCoreDataToCharInfo"},
             {25, nullptr, "ConvertCharInfoToCoreData"},
+            {26, nullptr, "Append"},
         };
         // clang-format on
 

src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp

@@ -15,8 +15,9 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_ctrl::nvhost_ctrl(Core::System& system, EventInterface& events_interface)
-    : nvdevice(system), events_interface{events_interface} {}
+nvhost_ctrl::nvhost_ctrl(Core::System& system, EventInterface& events_interface,
+                         SyncpointManager& syncpoint_manager)
+    : nvdevice(system), events_interface{events_interface}, syncpoint_manager{syncpoint_manager} {}
 nvhost_ctrl::~nvhost_ctrl() = default;
 
 u32 nvhost_ctrl::ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2,
@@ -36,8 +37,8 @@ u32 nvhost_ctrl::ioctl(Ioctl command, const std::vector<u8>& input, const std::v
         return IocCtrlEventRegister(input, output);
     case IoctlCommand::IocCtrlEventUnregisterCommand:
         return IocCtrlEventUnregister(input, output);
-    case IoctlCommand::IocCtrlEventSignalCommand:
-        return IocCtrlEventSignal(input, output);
+    case IoctlCommand::IocCtrlClearEventWaitCommand:
+        return IocCtrlClearEventWait(input, output);
     default:
         UNIMPLEMENTED_MSG("Unimplemented ioctl");
         return 0;
@@ -70,19 +71,33 @@ u32 nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>&
         return NvResult::BadParameter;
     }
 
+    if (syncpoint_manager.IsSyncpointExpired(params.syncpt_id, params.threshold)) {
+        params.value = syncpoint_manager.GetSyncpointMin(params.syncpt_id);
+        std::memcpy(output.data(), &params, sizeof(params));
+        return NvResult::Success;
+    }
+
+    if (const auto new_value = syncpoint_manager.RefreshSyncpoint(params.syncpt_id);
+        syncpoint_manager.IsSyncpointExpired(params.syncpt_id, params.threshold)) {
+        params.value = new_value;
+        std::memcpy(output.data(), &params, sizeof(params));
+        return NvResult::Success;
+    }
+
     auto event = events_interface.events[event_id];
     auto& gpu = system.GPU();
+
     // This is mostly to take into account unimplemented features. As synced
     // gpu is always synced.
     if (!gpu.IsAsync()) {
-        event.writable->Signal();
+        event.event.writable->Signal();
         return NvResult::Success;
     }
     auto lock = gpu.LockSync();
-    const u32 current_syncpoint_value = gpu.GetSyncpointValue(params.syncpt_id);
+    const u32 current_syncpoint_value = event.fence.value;
     const s32 diff = current_syncpoint_value - params.threshold;
     if (diff >= 0) {
-        event.writable->Signal();
+        event.event.writable->Signal();
         params.value = current_syncpoint_value;
         std::memcpy(output.data(), &params, sizeof(params));
         return NvResult::Success;
@@ -109,7 +124,7 @@ u32 nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>&
             params.value = ((params.syncpt_id & 0xfff) << 16) | 0x10000000;
         }
         params.value |= event_id;
-        event.writable->Clear();
+        event.event.writable->Clear();
         gpu.RegisterSyncptInterrupt(params.syncpt_id, target_value);
         if (!is_async && ctrl.fresh_call) {
             ctrl.must_delay = true;
@@ -154,24 +169,22 @@ u32 nvhost_ctrl::IocCtrlEventUnregister(const std::vector<u8>& input, std::vecto
     return NvResult::Success;
 }
 
-u32 nvhost_ctrl::IocCtrlEventSignal(const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_ctrl::IocCtrlClearEventWait(const std::vector<u8>& input, std::vector<u8>& output) {
     IocCtrlEventSignalParams params{};
     std::memcpy(&params, input.data(), sizeof(params));
-    // TODO(Blinkhawk): This is normally called when an NvEvents timeout on WaitSynchronization
-    // It is believed from RE to cancel the GPU Event. However, better research is required
-    u32 event_id = params.user_event_id & 0x00FF;
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called, user_event_id: {:X}", event_id);
+
+    u32 event_id = params.event_id & 0x00FF;
+    LOG_WARNING(Service_NVDRV, "cleared event wait on, event_id: {:X}", event_id);
+
     if (event_id >= MaxNvEvents) {
         return NvResult::BadParameter;
     }
     if (events_interface.status[event_id] == EventState::Waiting) {
-        auto& gpu = system.GPU();
-        if (gpu.CancelSyncptInterrupt(events_interface.assigned_syncpt[event_id],
-                                      events_interface.assigned_value[event_id])) {
-            events_interface.LiberateEvent(event_id);
-            events_interface.events[event_id].writable->Signal();
-        }
+        events_interface.LiberateEvent(event_id);
     }
+
+    syncpoint_manager.RefreshSyncpoint(events_interface.events[event_id].fence.id);
+
     return NvResult::Success;
 }
 

src/core/hle/service/nvdrv/devices/nvhost_ctrl.h

@@ -14,7 +14,8 @@ namespace Service::Nvidia::Devices {
 
 class nvhost_ctrl final : public nvdevice {
 public:
-    explicit nvhost_ctrl(Core::System& system, EventInterface& events_interface);
+    explicit nvhost_ctrl(Core::System& system, EventInterface& events_interface,
+                         SyncpointManager& syncpoint_manager);
     ~nvhost_ctrl() override;
 
     u32 ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2,
@@ -31,7 +32,7 @@ private:
         IocSyncptWaitexCommand = 0xC0100019,
         IocSyncptReadMaxCommand = 0xC008001A,
         IocGetConfigCommand = 0xC183001B,
-        IocCtrlEventSignalCommand = 0xC004001C,
+        IocCtrlClearEventWaitCommand = 0xC004001C,
         IocCtrlEventWaitCommand = 0xC010001D,
         IocCtrlEventWaitAsyncCommand = 0xC010001E,
         IocCtrlEventRegisterCommand = 0xC004001F,
@@ -94,7 +95,7 @@ private:
     static_assert(sizeof(IocGetConfigParams) == 387, "IocGetConfigParams is incorrect size");
 
     struct IocCtrlEventSignalParams {
-        u32_le user_event_id;
+        u32_le event_id;
     };
     static_assert(sizeof(IocCtrlEventSignalParams) == 4,
                   "IocCtrlEventSignalParams is incorrect size");
@@ -142,9 +143,10 @@ private:
 
     u32 IocCtrlEventUnregister(const std::vector<u8>& input, std::vector<u8>& output);
 
-    u32 IocCtrlEventSignal(const std::vector<u8>& input, std::vector<u8>& output);
+    u32 IocCtrlClearEventWait(const std::vector<u8>& input, std::vector<u8>& output);
 
     EventInterface& events_interface;
+    SyncpointManager& syncpoint_manager;
 };
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp

@@ -7,14 +7,20 @@
 #include "common/logging/log.h"
 #include "core/core.h"
 #include "core/hle/service/nvdrv/devices/nvhost_gpu.h"
+#include "core/hle/service/nvdrv/syncpoint_manager.h"
 #include "core/memory.h"
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
 
 namespace Service::Nvidia::Devices {
 
-nvhost_gpu::nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
-    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
+nvhost_gpu::nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
+                       SyncpointManager& syncpoint_manager)
+    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)), syncpoint_manager{syncpoint_manager} {
+    channel_fence.id = syncpoint_manager.AllocateSyncpoint();
+    channel_fence.value = system.GPU().GetSyncpointValue(channel_fence.id);
+}
+
 nvhost_gpu::~nvhost_gpu() = default;
 
 u32 nvhost_gpu::ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2,
@@ -126,10 +132,10 @@ u32 nvhost_gpu::AllocGPFIFOEx2(const std::vector<u8>& input, std::vector<u8>& ou
                 params.num_entries, params.flags, params.unk0, params.unk1, params.unk2,
                 params.unk3);
 
-    auto& gpu = system.GPU();
-    params.fence_out.id = assigned_syncpoints;
-    params.fence_out.value = gpu.GetSyncpointValue(assigned_syncpoints);
-    assigned_syncpoints++;
+    channel_fence.value = system.GPU().GetSyncpointValue(channel_fence.id);
+
+    params.fence_out = channel_fence;
+
     std::memcpy(output.data(), &params, output.size());
     return 0;
 }
@@ -145,37 +151,98 @@ u32 nvhost_gpu::AllocateObjectContext(const std::vector<u8>& input, std::vector<
     return 0;
 }
 
+static std::vector<Tegra::CommandHeader> BuildWaitCommandList(Fence fence) {
+    return {
+        Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceValue, 1,
+                                  Tegra::SubmissionMode::Increasing),
+        {fence.value},
+        Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceAction, 1,
+                                  Tegra::SubmissionMode::Increasing),
+        Tegra::GPU::FenceAction::Build(Tegra::GPU::FenceOperation::Acquire, fence.id),
+    };
+}
+
+static std::vector<Tegra::CommandHeader> BuildIncrementCommandList(Fence fence, u32 add_increment) {
+    std::vector<Tegra::CommandHeader> result{
+        Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceValue, 1,
+                                  Tegra::SubmissionMode::Increasing),
+        {}};
+
+    for (u32 count = 0; count < add_increment; ++count) {
+        result.emplace_back(Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceAction, 1,
+                                                      Tegra::SubmissionMode::Increasing));
+        result.emplace_back(
+            Tegra::GPU::FenceAction::Build(Tegra::GPU::FenceOperation::Increment, fence.id));
+    }
+
+    return result;
+}
+
+static std::vector<Tegra::CommandHeader> BuildIncrementWithWfiCommandList(Fence fence,
+                                                                          u32 add_increment) {
+    std::vector<Tegra::CommandHeader> result{
+        Tegra::BuildCommandHeader(Tegra::BufferMethods::WaitForInterrupt, 1,
+                                  Tegra::SubmissionMode::Increasing),
+        {}};
+    const std::vector<Tegra::CommandHeader> increment{
+        BuildIncrementCommandList(fence, add_increment)};
+
+    result.insert(result.end(), increment.begin(), increment.end());
+
+    return result;
+}
+
+u32 nvhost_gpu::SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, std::vector<u8>& output,
+                                 Tegra::CommandList&& entries) {
+    LOG_TRACE(Service_NVDRV, "called, gpfifo={:X}, num_entries={:X}, flags={:X}", params.address,
+              params.num_entries, params.flags.raw);
+
+    auto& gpu = system.GPU();
+
+    params.fence_out.id = channel_fence.id;
+
+    if (params.flags.add_wait.Value() &&
+        !syncpoint_manager.IsSyncpointExpired(params.fence_out.id, params.fence_out.value)) {
+        gpu.PushGPUEntries(Tegra::CommandList{BuildWaitCommandList(params.fence_out)});
+    }
+
+    if (params.flags.add_increment.Value() || params.flags.increment.Value()) {
+        const u32 increment_value = params.flags.increment.Value() ? params.fence_out.value : 0;
+        params.fence_out.value = syncpoint_manager.IncreaseSyncpoint(
+            params.fence_out.id, params.AddIncrementValue() + increment_value);
+    } else {
+        params.fence_out.value = syncpoint_manager.GetSyncpointMax(params.fence_out.id);
+    }
+
+    entries.RefreshIntegrityChecks(gpu);
+    gpu.PushGPUEntries(std::move(entries));
+
+    if (params.flags.add_increment.Value()) {
+        if (params.flags.suppress_wfi) {
+            gpu.PushGPUEntries(Tegra::CommandList{
+                BuildIncrementCommandList(params.fence_out, params.AddIncrementValue())});
+        } else {
+            gpu.PushGPUEntries(Tegra::CommandList{
+                BuildIncrementWithWfiCommandList(params.fence_out, params.AddIncrementValue())});
+        }
+    }
+
+    std::memcpy(output.data(), &params, sizeof(IoctlSubmitGpfifo));
+    return 0;
+}
+
 u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& output) {
     if (input.size() < sizeof(IoctlSubmitGpfifo)) {
         UNIMPLEMENTED();
     }
     IoctlSubmitGpfifo params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSubmitGpfifo));
-    LOG_TRACE(Service_NVDRV, "called, gpfifo={:X}, num_entries={:X}, flags={:X}", params.address,
-              params.num_entries, params.flags.raw);
-
-    ASSERT_MSG(input.size() == sizeof(IoctlSubmitGpfifo) +
-                                   params.num_entries * sizeof(Tegra::CommandListHeader),
-               "Incorrect input size");
 
     Tegra::CommandList entries(params.num_entries);
-    std::memcpy(entries.data(), &input[sizeof(IoctlSubmitGpfifo)],
+    std::memcpy(entries.command_lists.data(), &input[sizeof(IoctlSubmitGpfifo)],
                 params.num_entries * sizeof(Tegra::CommandListHeader));
 
-    UNIMPLEMENTED_IF(params.flags.add_wait.Value() != 0);
-    UNIMPLEMENTED_IF(params.flags.add_increment.Value() != 0);
-
-    auto& gpu = system.GPU();
-    u32 current_syncpoint_value = gpu.GetSyncpointValue(params.fence_out.id);
-    if (params.flags.increment.Value()) {
-        params.fence_out.value += current_syncpoint_value;
-    } else {
-        params.fence_out.value = current_syncpoint_value;
-    }
-    gpu.PushGPUEntries(std::move(entries));
-
-    std::memcpy(output.data(), &params, sizeof(IoctlSubmitGpfifo));
-    return 0;
+    return SubmitGPFIFOImpl(params, output, std::move(entries));
 }
 
 u32 nvhost_gpu::KickoffPB(const std::vector<u8>& input, std::vector<u8>& output,
@@ -185,31 +252,17 @@ u32 nvhost_gpu::KickoffPB(const std::vector<u8>& input, std::vector<u8>& output,
     }
     IoctlSubmitGpfifo params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSubmitGpfifo));
-    LOG_TRACE(Service_NVDRV, "called, gpfifo={:X}, num_entries={:X}, flags={:X}", params.address,
-              params.num_entries, params.flags.raw);
 
     Tegra::CommandList entries(params.num_entries);
     if (version == IoctlVersion::Version2) {
-        std::memcpy(entries.data(), input2.data(),
+        std::memcpy(entries.command_lists.data(), input2.data(),
                     params.num_entries * sizeof(Tegra::CommandListHeader));
     } else {
-        system.Memory().ReadBlock(params.address, entries.data(),
+        system.Memory().ReadBlock(params.address, entries.command_lists.data(),
                                   params.num_entries * sizeof(Tegra::CommandListHeader));
     }
-    UNIMPLEMENTED_IF(params.flags.add_wait.Value() != 0);
-    UNIMPLEMENTED_IF(params.flags.add_increment.Value() != 0);
 
-    auto& gpu = system.GPU();
-    u32 current_syncpoint_value = gpu.GetSyncpointValue(params.fence_out.id);
-    if (params.flags.increment.Value()) {
-        params.fence_out.value += current_syncpoint_value;
-    } else {
-        params.fence_out.value = current_syncpoint_value;
-    }
-    gpu.PushGPUEntries(std::move(entries));
-
-    std::memcpy(output.data(), &params, output.size());
-    return 0;
+    return SubmitGPFIFOImpl(params, output, std::move(entries));
 }
 
 u32 nvhost_gpu::GetWaitbase(const std::vector<u8>& input, std::vector<u8>& output) {

src/core/hle/service/nvdrv/devices/nvhost_gpu.h

@@ -11,6 +11,11 @@
 #include "common/swap.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
 #include "core/hle/service/nvdrv/nvdata.h"
+#include "video_core/dma_pusher.h"
+
+namespace Service::Nvidia {
+class SyncpointManager;
+}
 
 namespace Service::Nvidia::Devices {
 
@@ -21,7 +26,8 @@ constexpr u32 NVGPU_IOCTL_CHANNEL_KICKOFF_PB(0x1b);
 
 class nvhost_gpu final : public nvdevice {
 public:
-    explicit nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
+    explicit nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
+                        SyncpointManager& syncpoint_manager);
     ~nvhost_gpu() override;
 
     u32 ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2,
@@ -162,10 +168,15 @@ private:
             u32_le raw;
             BitField<0, 1, u32_le> add_wait;      // append a wait sync_point to the list
             BitField<1, 1, u32_le> add_increment; // append an increment to the list
-            BitField<2, 1, u32_le> new_hw_format; // Mostly ignored
+            BitField<2, 1, u32_le> new_hw_format; // mostly ignored
+            BitField<4, 1, u32_le> suppress_wfi;  // suppress wait for interrupt
             BitField<8, 1, u32_le> increment;     // increment the returned fence
         } flags;
         Fence fence_out; // returned new fence object for others to wait on
+
+        u32 AddIncrementValue() const {
+            return flags.add_increment.Value() << 1;
+        }
     };
     static_assert(sizeof(IoctlSubmitGpfifo) == 16 + sizeof(Fence),
                   "IoctlSubmitGpfifo is incorrect size");
@@ -190,6 +201,8 @@ private:
     u32 SetChannelPriority(const std::vector<u8>& input, std::vector<u8>& output);
     u32 AllocGPFIFOEx2(const std::vector<u8>& input, std::vector<u8>& output);
     u32 AllocateObjectContext(const std::vector<u8>& input, std::vector<u8>& output);
+    u32 SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, std::vector<u8>& output,
+                         Tegra::CommandList&& entries);
     u32 SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& output);
     u32 KickoffPB(const std::vector<u8>& input, std::vector<u8>& output,
                   const std::vector<u8>& input2, IoctlVersion version);
@@ -198,7 +211,8 @@ private:
     u32 ChannelSetTimeslice(const std::vector<u8>& input, std::vector<u8>& output);
 
     std::shared_ptr<nvmap> nvmap_dev;
-    u32 assigned_syncpoints{};
+    SyncpointManager& syncpoint_manager;
+    Fence channel_fence;
 };
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp

@@ -2,15 +2,17 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <cstring>
-
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "core/core.h"
 #include "core/hle/service/nvdrv/devices/nvhost_nvdec.h"
+#include "video_core/memory_manager.h"
+#include "video_core/renderer_base.h"
 
 namespace Service::Nvidia::Devices {
 
-nvhost_nvdec::nvhost_nvdec(Core::System& system) : nvdevice(system) {}
+nvhost_nvdec::nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
+    : nvhost_nvdec_common(system, std::move(nvmap_dev)) {}
 nvhost_nvdec::~nvhost_nvdec() = default;
 
 u32 nvhost_nvdec::ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2,
@@ -21,7 +23,7 @@ u32 nvhost_nvdec::ioctl(Ioctl command, const std::vector<u8>& input, const std::
 
     switch (static_cast<IoctlCommand>(command.raw)) {
     case IoctlCommand::IocSetNVMAPfdCommand:
-        return SetNVMAPfd(input, output);
+        return SetNVMAPfd(input);
     case IoctlCommand::IocSubmit:
         return Submit(input, output);
     case IoctlCommand::IocGetSyncpoint:
@@ -29,79 +31,29 @@ u32 nvhost_nvdec::ioctl(Ioctl command, const std::vector<u8>& input, const std::
     case IoctlCommand::IocGetWaitbase:
         return GetWaitbase(input, output);
     case IoctlCommand::IocMapBuffer:
-        return MapBuffer(input, output);
+    case IoctlCommand::IocMapBuffer2:
+    case IoctlCommand::IocMapBuffer3:
     case IoctlCommand::IocMapBufferEx:
-        return MapBufferEx(input, output);
-    case IoctlCommand::IocUnmapBufferEx:
-        return UnmapBufferEx(input, output);
+        return MapBuffer(input, output);
+    case IoctlCommand::IocUnmapBufferEx: {
+        // This command is sent when the video stream has ended, flush all video contexts
+        // This is usually sent in the folowing order: vic, nvdec, vic.
+        // Inform the GPU to clear any remaining nvdec buffers when this is detected.
+        LOG_INFO(Service_NVDRV, "NVDEC video stream ended");
+        Tegra::ChCommandHeaderList cmdlist(1);
+        cmdlist[0] = Tegra::ChCommandHeader{0xDEADB33F};
+        system.GPU().PushCommandBuffer(cmdlist);
+        [[fallthrough]]; // fallthrough to unmap buffers
+    };
+    case IoctlCommand::IocUnmapBuffer:
+    case IoctlCommand::IocUnmapBuffer2:
+    case IoctlCommand::IocUnmapBuffer3:
+        return UnmapBuffer(input, output);
+    case IoctlCommand::IocSetSubmitTimeout:
+        return SetSubmitTimeout(input, output);
     }
 
-    UNIMPLEMENTED_MSG("Unimplemented ioctl");
-    return 0;
-}
-
-u32 nvhost_nvdec::SetNVMAPfd(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlSetNvmapFD params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlSetNvmapFD));
-    LOG_DEBUG(Service_NVDRV, "called, fd={}", params.nvmap_fd);
-
-    nvmap_fd = params.nvmap_fd;
-    return 0;
-}
-
-u32 nvhost_nvdec::Submit(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlSubmit params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlSubmit));
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called");
-    std::memcpy(output.data(), &params, sizeof(IoctlSubmit));
-    return 0;
-}
-
-u32 nvhost_nvdec::GetSyncpoint(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlGetSyncpoint params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlGetSyncpoint));
-    LOG_INFO(Service_NVDRV, "called, unknown=0x{:X}", params.unknown);
-    params.value = 0; // Seems to be hard coded at 0
-    std::memcpy(output.data(), &params, sizeof(IoctlGetSyncpoint));
-    return 0;
-}
-
-u32 nvhost_nvdec::GetWaitbase(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlGetWaitbase params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlGetWaitbase));
-    LOG_INFO(Service_NVDRV, "called, unknown=0x{:X}", params.unknown);
-    params.value = 0; // Seems to be hard coded at 0
-    std::memcpy(output.data(), &params, sizeof(IoctlGetWaitbase));
-    return 0;
-}
-
-u32 nvhost_nvdec::MapBuffer(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlMapBuffer params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlMapBuffer));
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called with address={:08X}{:08X}", params.address_2,
-                params.address_1);
-    params.address_1 = 0;
-    params.address_2 = 0;
-    std::memcpy(output.data(), &params, sizeof(IoctlMapBuffer));
-    return 0;
-}
-
-u32 nvhost_nvdec::MapBufferEx(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlMapBufferEx params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlMapBufferEx));
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called with address={:08X}{:08X}", params.address_2,
-                params.address_1);
-    params.address_1 = 0;
-    params.address_2 = 0;
-    std::memcpy(output.data(), &params, sizeof(IoctlMapBufferEx));
-    return 0;
-}
-
-u32 nvhost_nvdec::UnmapBufferEx(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlUnmapBufferEx params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlUnmapBufferEx));
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called");
-    std::memcpy(output.data(), &params, sizeof(IoctlUnmapBufferEx));
+    UNIMPLEMENTED_MSG("Unimplemented ioctl 0x{:X}", command.raw);
     return 0;
 }
 

src/core/hle/service/nvdrv/devices/nvhost_nvdec.h

@@ -4,16 +4,14 @@
 
 #pragma once
 
-#include <vector>
-#include "common/common_types.h"
-#include "common/swap.h"
-#include "core/hle/service/nvdrv/devices/nvdevice.h"
+#include <memory>
+#include "core/hle/service/nvdrv/devices/nvhost_nvdec_common.h"
 
 namespace Service::Nvidia::Devices {
 
-class nvhost_nvdec final : public nvdevice {
+class nvhost_nvdec final : public nvhost_nvdec_common {
 public:
-    explicit nvhost_nvdec(Core::System& system);
+    explicit nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvhost_nvdec() override;
 
     u32 ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2,
@@ -27,62 +25,15 @@ private:
         IocGetSyncpoint = 0xC0080002,
         IocGetWaitbase = 0xC0080003,
         IocMapBuffer = 0xC01C0009,
+        IocMapBuffer2 = 0xC16C0009,
+        IocMapBuffer3 = 0xC15C0009,
         IocMapBufferEx = 0xC0A40009,
-        IocUnmapBufferEx = 0xC0A4000A,
+        IocUnmapBuffer = 0xC0A4000A,
+        IocUnmapBuffer2 = 0xC16C000A,
+        IocUnmapBufferEx = 0xC01C000A,
+        IocUnmapBuffer3 = 0xC15C000A,
+        IocSetSubmitTimeout = 0x40040007,
     };
-
-    struct IoctlSetNvmapFD {
-        u32_le nvmap_fd;
-    };
-    static_assert(sizeof(IoctlSetNvmapFD) == 0x4, "IoctlSetNvmapFD is incorrect size");
-
-    struct IoctlSubmit {
-        INSERT_PADDING_BYTES(0x40); // TODO(DarkLordZach): RE this structure
-    };
-    static_assert(sizeof(IoctlSubmit) == 0x40, "IoctlSubmit has incorrect size");
-
-    struct IoctlGetSyncpoint {
-        u32 unknown; // seems to be ignored? Nintendo added this
-        u32 value;
-    };
-    static_assert(sizeof(IoctlGetSyncpoint) == 0x08, "IoctlGetSyncpoint has incorrect size");
-
-    struct IoctlGetWaitbase {
-        u32 unknown; // seems to be ignored? Nintendo added this
-        u32 value;
-    };
-    static_assert(sizeof(IoctlGetWaitbase) == 0x08, "IoctlGetWaitbase has incorrect size");
-
-    struct IoctlMapBuffer {
-        u32 unknown;
-        u32 address_1;
-        u32 address_2;
-        INSERT_PADDING_BYTES(0x10); // TODO(DarkLordZach): RE this structure
-    };
-    static_assert(sizeof(IoctlMapBuffer) == 0x1C, "IoctlMapBuffer is incorrect size");
-
-    struct IoctlMapBufferEx {
-        u32 unknown;
-        u32 address_1;
-        u32 address_2;
-        INSERT_PADDING_BYTES(0x98); // TODO(DarkLordZach): RE this structure
-    };
-    static_assert(sizeof(IoctlMapBufferEx) == 0xA4, "IoctlMapBufferEx has incorrect size");
-
-    struct IoctlUnmapBufferEx {
-        INSERT_PADDING_BYTES(0xA4); // TODO(DarkLordZach): RE this structure
-    };
-    static_assert(sizeof(IoctlUnmapBufferEx) == 0xA4, "IoctlUnmapBufferEx has incorrect size");
-
-    u32_le nvmap_fd{};
-
-    u32 SetNVMAPfd(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 Submit(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 GetSyncpoint(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 GetWaitbase(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 MapBuffer(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 MapBufferEx(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 UnmapBufferEx(const std::vector<u8>& input, std::vector<u8>& output);
 };
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp

@@ -0,0 +1,234 @@
+// Copyright 2020 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <cstring>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "common/logging/log.h"
+#include "core/core.h"
+#include "core/hle/service/nvdrv/devices/nvhost_nvdec_common.h"
+#include "core/hle/service/nvdrv/devices/nvmap.h"
+#include "core/memory.h"
+#include "video_core/memory_manager.h"
+#include "video_core/renderer_base.h"
+
+namespace Service::Nvidia::Devices {
+
+namespace {
+// Splice vectors will copy count amount of type T from the input vector into the dst vector.
+template <typename T>
+std::size_t SpliceVectors(const std::vector<u8>& input, std::vector<T>& dst, std::size_t count,
+                          std::size_t offset) {
+    std::memcpy(dst.data(), input.data() + offset, count * sizeof(T));
+    offset += count * sizeof(T);
+    return offset;
+}
+
+// Write vectors will write data to the output buffer
+template <typename T>
+std::size_t WriteVectors(std::vector<u8>& dst, const std::vector<T>& src, std::size_t offset) {
+    std::memcpy(dst.data() + offset, src.data(), src.size() * sizeof(T));
+    offset += src.size() * sizeof(T);
+    return offset;
+}
+} // Anonymous namespace
+
+namespace NvErrCodes {
+constexpr u32 Success{};
+[[maybe_unused]] constexpr u32 OutOfMemory{static_cast<u32>(-12)};
+constexpr u32 InvalidInput{static_cast<u32>(-22)};
+} // namespace NvErrCodes
+
+nvhost_nvdec_common::nvhost_nvdec_common(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
+    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
+nvhost_nvdec_common::~nvhost_nvdec_common() = default;
+
+u32 nvhost_nvdec_common::SetNVMAPfd(const std::vector<u8>& input) {
+    IoctlSetNvmapFD params{};
+    std::memcpy(&params, input.data(), sizeof(IoctlSetNvmapFD));
+    LOG_DEBUG(Service_NVDRV, "called, fd={}", params.nvmap_fd);
+
+    nvmap_fd = params.nvmap_fd;
+    return 0;
+}
+
+u32 nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u8>& output) {
+    IoctlSubmit params{};
+    std::memcpy(&params, input.data(), sizeof(IoctlSubmit));
+    LOG_DEBUG(Service_NVDRV, "called NVDEC Submit, cmd_buffer_count={}", params.cmd_buffer_count);
+
+    // Instantiate param buffers
+    std::size_t offset = sizeof(IoctlSubmit);
+    std::vector<CommandBuffer> command_buffers(params.cmd_buffer_count);
+    std::vector<Reloc> relocs(params.relocation_count);
+    std::vector<u32> reloc_shifts(params.relocation_count);
+    std::vector<SyncptIncr> syncpt_increments(params.syncpoint_count);
+    std::vector<SyncptIncr> wait_checks(params.syncpoint_count);
+    std::vector<Fence> fences(params.fence_count);
+
+    // Splice input into their respective buffers
+    offset = SpliceVectors(input, command_buffers, params.cmd_buffer_count, offset);
+    offset = SpliceVectors(input, relocs, params.relocation_count, offset);
+    offset = SpliceVectors(input, reloc_shifts, params.relocation_count, offset);
+    offset = SpliceVectors(input, syncpt_increments, params.syncpoint_count, offset);
+    offset = SpliceVectors(input, wait_checks, params.syncpoint_count, offset);
+    offset = SpliceVectors(input, fences, params.fence_count, offset);
+
+    // TODO(ameerj): For async gpu, utilize fences for syncpoint 'max' increment
+
+    auto& gpu = system.GPU();
+
+    for (const auto& cmd_buffer : command_buffers) {
+        auto object = nvmap_dev->GetObject(cmd_buffer.memory_id);
+        ASSERT_OR_EXECUTE(object, return NvErrCodes::InvalidInput;);
+        const auto map = FindBufferMap(object->dma_map_addr);
+        if (!map) {
+            LOG_ERROR(Service_NVDRV, "Tried to submit an invalid offset 0x{:X} dma 0x{:X}",
+                      object->addr, object->dma_map_addr);
+            return 0;
+        }
+        Tegra::ChCommandHeaderList cmdlist(cmd_buffer.word_count);
+        gpu.MemoryManager().ReadBlock(map->StartAddr() + cmd_buffer.offset, cmdlist.data(),
+                                      cmdlist.size() * sizeof(u32));
+        gpu.PushCommandBuffer(cmdlist);
+    }
+
+    std::memcpy(output.data(), &params, sizeof(IoctlSubmit));
+    // Some games expect command_buffers to be written back
+    offset = sizeof(IoctlSubmit);
+    offset = WriteVectors(output, command_buffers, offset);
+    offset = WriteVectors(output, relocs, offset);
+    offset = WriteVectors(output, reloc_shifts, offset);
+    offset = WriteVectors(output, syncpt_increments, offset);
+    offset = WriteVectors(output, wait_checks, offset);
+
+    return NvErrCodes::Success;
+}
+
+u32 nvhost_nvdec_common::GetSyncpoint(const std::vector<u8>& input, std::vector<u8>& output) {
+    IoctlGetSyncpoint params{};
+    std::memcpy(&params, input.data(), sizeof(IoctlGetSyncpoint));
+    LOG_DEBUG(Service_NVDRV, "called GetSyncpoint, id={}", params.param);
+
+    // We found that implementing this causes deadlocks with async gpu, along with degraded
+    // performance. TODO: RE the nvdec async implementation
+    params.value = 0;
+    std::memcpy(output.data(), &params, sizeof(IoctlGetSyncpoint));
+
+    return NvErrCodes::Success;
+}
+
+u32 nvhost_nvdec_common::GetWaitbase(const std::vector<u8>& input, std::vector<u8>& output) {
+    IoctlGetWaitbase params{};
+    std::memcpy(&params, input.data(), sizeof(IoctlGetWaitbase));
+    params.value = 0; // Seems to be hard coded at 0
+    std::memcpy(output.data(), &params, sizeof(IoctlGetWaitbase));
+    return 0;
+}
+
+u32 nvhost_nvdec_common::MapBuffer(const std::vector<u8>& input, std::vector<u8>& output) {
+    IoctlMapBuffer params{};
+    std::memcpy(&params, input.data(), sizeof(IoctlMapBuffer));
+    std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries);
+
+    SpliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer));
+
+    auto& gpu = system.GPU();
+
+    for (auto& cmf_buff : cmd_buffer_handles) {
+        auto object{nvmap_dev->GetObject(cmf_buff.map_handle)};
+        if (!object) {
+            LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmf_buff.map_handle);
+            std::memcpy(output.data(), &params, output.size());
+            return NvErrCodes::InvalidInput;
+        }
+        if (object->dma_map_addr == 0) {
+            // NVDEC and VIC memory is in the 32-bit address space
+            // MapAllocate32 will attempt to map a lower 32-bit value in the shared gpu memory space
+            const GPUVAddr low_addr = gpu.MemoryManager().MapAllocate32(object->addr, object->size);
+            object->dma_map_addr = static_cast<u32>(low_addr);
+            // Ensure that the dma_map_addr is indeed in the lower 32-bit address space.
+            ASSERT(object->dma_map_addr == low_addr);
+        }
+        if (!object->dma_map_addr) {
+            LOG_ERROR(Service_NVDRV, "failed to map size={}", object->size);
+        } else {
+            cmf_buff.map_address = object->dma_map_addr;
+            AddBufferMap(object->dma_map_addr, object->size, object->addr,
+                         object->status == nvmap::Object::Status::Allocated);
+        }
+    }
+    std::memcpy(output.data(), &params, sizeof(IoctlMapBuffer));
+    std::memcpy(output.data() + sizeof(IoctlMapBuffer), cmd_buffer_handles.data(),
+                cmd_buffer_handles.size() * sizeof(MapBufferEntry));
+
+    return NvErrCodes::Success;
+}
+
+u32 nvhost_nvdec_common::UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& output) {
+    IoctlMapBuffer params{};
+    std::memcpy(&params, input.data(), sizeof(IoctlMapBuffer));
+    std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries);
+    SpliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer));
+
+    auto& gpu = system.GPU();
+
+    for (auto& cmf_buff : cmd_buffer_handles) {
+        const auto object{nvmap_dev->GetObject(cmf_buff.map_handle)};
+        if (!object) {
+            LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmf_buff.map_handle);
+            std::memcpy(output.data(), &params, output.size());
+            return NvErrCodes::InvalidInput;
+        }
+        if (const auto size{RemoveBufferMap(object->dma_map_addr)}; size) {
+            gpu.MemoryManager().Unmap(object->dma_map_addr, *size);
+        } else {
+            // This occurs quite frequently, however does not seem to impact functionality
+            LOG_DEBUG(Service_NVDRV, "invalid offset=0x{:X} dma=0x{:X}", object->addr,
+                      object->dma_map_addr);
+        }
+        object->dma_map_addr = 0;
+    }
+    std::memset(output.data(), 0, output.size());
+    return NvErrCodes::Success;
+}
+
+u32 nvhost_nvdec_common::SetSubmitTimeout(const std::vector<u8>& input, std::vector<u8>& output) {
+    std::memcpy(&submit_timeout, input.data(), input.size());
+    LOG_WARNING(Service_NVDRV, "(STUBBED) called");
+    return NvErrCodes::Success;
+}
+
+std::optional<nvhost_nvdec_common::BufferMap> nvhost_nvdec_common::FindBufferMap(
+    GPUVAddr gpu_addr) const {
+    const auto it = std::find_if(
+        buffer_mappings.begin(), buffer_mappings.upper_bound(gpu_addr), [&](const auto& entry) {
+            return (gpu_addr >= entry.second.StartAddr() && gpu_addr < entry.second.EndAddr());
+        });
+
+    ASSERT(it != buffer_mappings.end());
+    return it->second;
+}
+
+void nvhost_nvdec_common::AddBufferMap(GPUVAddr gpu_addr, std::size_t size, VAddr cpu_addr,
+                                       bool is_allocated) {
+    buffer_mappings.insert_or_assign(gpu_addr, BufferMap{gpu_addr, size, cpu_addr, is_allocated});
+}
+
+std::optional<std::size_t> nvhost_nvdec_common::RemoveBufferMap(GPUVAddr gpu_addr) {
+    const auto iter{buffer_mappings.find(gpu_addr)};
+    if (iter == buffer_mappings.end()) {
+        return std::nullopt;
+    }
+    std::size_t size = 0;
+    if (iter->second.IsAllocated()) {
+        size = iter->second.Size();
+    }
+    buffer_mappings.erase(iter);
+    return size;
+}
+
+} // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h

@@ -0,0 +1,168 @@
+// Copyright 2020 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <map>
+#include <vector>
+#include "common/common_types.h"
+#include "common/swap.h"
+#include "core/hle/service/nvdrv/devices/nvdevice.h"
+
+namespace Service::Nvidia::Devices {
+class nvmap;
+
+class nvhost_nvdec_common : public nvdevice {
+public:
+    explicit nvhost_nvdec_common(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
+    ~nvhost_nvdec_common() override;
+
+    virtual u32 ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2,
+                      std::vector<u8>& output, std::vector<u8>& output2, IoctlCtrl& ctrl,
+                      IoctlVersion version) = 0;
+
+protected:
+    class BufferMap final {
+    public:
+        constexpr BufferMap() = default;
+
+        constexpr BufferMap(GPUVAddr start_addr, std::size_t size)
+            : start_addr{start_addr}, end_addr{start_addr + size} {}
+
+        constexpr BufferMap(GPUVAddr start_addr, std::size_t size, VAddr cpu_addr,
+                            bool is_allocated)
+            : start_addr{start_addr}, end_addr{start_addr + size}, cpu_addr{cpu_addr},
+              is_allocated{is_allocated} {}
+
+        constexpr VAddr StartAddr() const {
+            return start_addr;
+        }
+
+        constexpr VAddr EndAddr() const {
+            return end_addr;
+        }
+
+        constexpr std::size_t Size() const {
+            return end_addr - start_addr;
+        }
+
+        constexpr VAddr CpuAddr() const {
+            return cpu_addr;
+        }
+
+        constexpr bool IsAllocated() const {
+            return is_allocated;
+        }
+
+    private:
+        GPUVAddr start_addr{};
+        GPUVAddr end_addr{};
+        VAddr cpu_addr{};
+        bool is_allocated{};
+    };
+
+    struct IoctlSetNvmapFD {
+        u32_le nvmap_fd;
+    };
+    static_assert(sizeof(IoctlSetNvmapFD) == 4, "IoctlSetNvmapFD is incorrect size");
+
+    struct IoctlSubmitCommandBuffer {
+        u32_le id;
+        u32_le offset;
+        u32_le count;
+    };
+    static_assert(sizeof(IoctlSubmitCommandBuffer) == 0xC,
+                  "IoctlSubmitCommandBuffer is incorrect size");
+    struct IoctlSubmit {
+        u32_le cmd_buffer_count;
+        u32_le relocation_count;
+        u32_le syncpoint_count;
+        u32_le fence_count;
+    };
+    static_assert(sizeof(IoctlSubmit) == 0x10, "IoctlSubmit has incorrect size");
+
+    struct CommandBuffer {
+        s32 memory_id;
+        u32 offset;
+        s32 word_count;
+    };
+    static_assert(sizeof(CommandBuffer) == 0xC, "CommandBuffer has incorrect size");
+
+    struct Reloc {
+        s32 cmdbuffer_memory;
+        s32 cmdbuffer_offset;
+        s32 target;
+        s32 target_offset;
+    };
+    static_assert(sizeof(Reloc) == 0x10, "CommandBuffer has incorrect size");
+
+    struct SyncptIncr {
+        u32 id;
+        u32 increments;
+    };
+    static_assert(sizeof(SyncptIncr) == 0x8, "CommandBuffer has incorrect size");
+
+    struct Fence {
+        u32 id;
+        u32 value;
+    };
+    static_assert(sizeof(Fence) == 0x8, "CommandBuffer has incorrect size");
+
+    struct IoctlGetSyncpoint {
+        // Input
+        u32_le param;
+        // Output
+        u32_le value;
+    };
+    static_assert(sizeof(IoctlGetSyncpoint) == 8, "IocGetIdParams has wrong size");
+
+    struct IoctlGetWaitbase {
+        u32_le unknown; // seems to be ignored? Nintendo added this
+        u32_le value;
+    };
+    static_assert(sizeof(IoctlGetWaitbase) == 0x8, "IoctlGetWaitbase is incorrect size");
+
+    struct IoctlMapBuffer {
+        u32_le num_entries;
+        u32_le data_address; // Ignored by the driver.
+        u32_le attach_host_ch_das;
+    };
+    static_assert(sizeof(IoctlMapBuffer) == 0x0C, "IoctlMapBuffer is incorrect size");
+
+    struct IocGetIdParams {
+        // Input
+        u32_le param;
+        // Output
+        u32_le value;
+    };
+    static_assert(sizeof(IocGetIdParams) == 8, "IocGetIdParams has wrong size");
+
+    // Used for mapping and unmapping command buffers
+    struct MapBufferEntry {
+        u32_le map_handle;
+        u32_le map_address;
+    };
+    static_assert(sizeof(IoctlMapBuffer) == 0x0C, "IoctlMapBuffer is incorrect size");
+
+    /// Ioctl command implementations
+    u32 SetNVMAPfd(const std::vector<u8>& input);
+    u32 Submit(const std::vector<u8>& input, std::vector<u8>& output);
+    u32 GetSyncpoint(const std::vector<u8>& input, std::vector<u8>& output);
+    u32 GetWaitbase(const std::vector<u8>& input, std::vector<u8>& output);
+    u32 MapBuffer(const std::vector<u8>& input, std::vector<u8>& output);
+    u32 UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& output);
+    u32 SetSubmitTimeout(const std::vector<u8>& input, std::vector<u8>& output);
+
+    std::optional<BufferMap> FindBufferMap(GPUVAddr gpu_addr) const;
+    void AddBufferMap(GPUVAddr gpu_addr, std::size_t size, VAddr cpu_addr, bool is_allocated);
+    std::optional<std::size_t> RemoveBufferMap(GPUVAddr gpu_addr);
+
+    u32_le nvmap_fd{};
+    u32_le submit_timeout{};
+    std::shared_ptr<nvmap> nvmap_dev;
+
+    // This is expected to be ordered, therefore we must use a map, not unordered_map
+    std::map<GPUVAddr, BufferMap> buffer_mappings;
+};
+}; // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_vic.cpp

@@ -2,15 +2,17 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <cstring>
-
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "core/core.h"
 #include "core/hle/service/nvdrv/devices/nvhost_vic.h"
+#include "video_core/memory_manager.h"
+#include "video_core/renderer_base.h"
 
 namespace Service::Nvidia::Devices {
+nvhost_vic::nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
+    : nvhost_nvdec_common(system, std::move(nvmap_dev)) {}
 
-nvhost_vic::nvhost_vic(Core::System& system) : nvdevice(system) {}
 nvhost_vic::~nvhost_vic() = default;
 
 u32 nvhost_vic::ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2,
@@ -21,7 +23,7 @@ u32 nvhost_vic::ioctl(Ioctl command, const std::vector<u8>& input, const std::ve
 
     switch (static_cast<IoctlCommand>(command.raw)) {
     case IoctlCommand::IocSetNVMAPfdCommand:
-        return SetNVMAPfd(input, output);
+        return SetNVMAPfd(input);
     case IoctlCommand::IocSubmit:
         return Submit(input, output);
     case IoctlCommand::IocGetSyncpoint:
@@ -29,83 +31,19 @@ u32 nvhost_vic::ioctl(Ioctl command, const std::vector<u8>& input, const std::ve
     case IoctlCommand::IocGetWaitbase:
         return GetWaitbase(input, output);
     case IoctlCommand::IocMapBuffer:
-        return MapBuffer(input, output);
+    case IoctlCommand::IocMapBuffer2:
+    case IoctlCommand::IocMapBuffer3:
+    case IoctlCommand::IocMapBuffer4:
     case IoctlCommand::IocMapBufferEx:
         return MapBuffer(input, output);
+    case IoctlCommand::IocUnmapBuffer:
+    case IoctlCommand::IocUnmapBuffer2:
+    case IoctlCommand::IocUnmapBuffer3:
     case IoctlCommand::IocUnmapBufferEx:
-        return UnmapBufferEx(input, output);
+        return UnmapBuffer(input, output);
     }
 
-    UNIMPLEMENTED_MSG("Unimplemented ioctl");
-    return 0;
-}
-
-u32 nvhost_vic::SetNVMAPfd(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlSetNvmapFD params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlSetNvmapFD));
-    LOG_DEBUG(Service_NVDRV, "called, fd={}", params.nvmap_fd);
-
-    nvmap_fd = params.nvmap_fd;
-    return 0;
-}
-
-u32 nvhost_vic::Submit(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlSubmit params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlSubmit));
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called");
-
-    // Workaround for Luigi's Mansion 3, as nvhost_vic is not implemented for asynch GPU
-    params.command_buffer = {};
-
-    std::memcpy(output.data(), &params, sizeof(IoctlSubmit));
-    return 0;
-}
-
-u32 nvhost_vic::GetSyncpoint(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlGetSyncpoint params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlGetSyncpoint));
-    LOG_INFO(Service_NVDRV, "called, unknown=0x{:X}", params.unknown);
-    params.value = 0; // Seems to be hard coded at 0
-    std::memcpy(output.data(), &params, sizeof(IoctlGetSyncpoint));
-    return 0;
-}
-
-u32 nvhost_vic::GetWaitbase(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlGetWaitbase params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlGetWaitbase));
-    LOG_INFO(Service_NVDRV, "called, unknown=0x{:X}", params.unknown);
-    params.value = 0; // Seems to be hard coded at 0
-    std::memcpy(output.data(), &params, sizeof(IoctlGetWaitbase));
-    return 0;
-}
-
-u32 nvhost_vic::MapBuffer(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlMapBuffer params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlMapBuffer));
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called with address={:08X}{:08X}", params.address_2,
-                params.address_1);
-    params.address_1 = 0;
-    params.address_2 = 0;
-    std::memcpy(output.data(), &params, sizeof(IoctlMapBuffer));
-    return 0;
-}
-
-u32 nvhost_vic::MapBufferEx(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlMapBufferEx params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlMapBufferEx));
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called with address={:08X}{:08X}", params.address_2,
-                params.address_1);
-    params.address_1 = 0;
-    params.address_2 = 0;
-    std::memcpy(output.data(), &params, sizeof(IoctlMapBufferEx));
-    return 0;
-}
-
-u32 nvhost_vic::UnmapBufferEx(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlUnmapBufferEx params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlUnmapBufferEx));
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called");
-    std::memcpy(output.data(), &params, sizeof(IoctlUnmapBufferEx));
+    UNIMPLEMENTED_MSG("Unimplemented ioctl 0x{:X}", command.raw);
     return 0;
 }
 

src/core/hle/service/nvdrv/devices/nvhost_vic.h

@@ -4,19 +4,15 @@
 
 #pragma once
 
-#include <array>
-#include <vector>
-#include "common/common_types.h"
-#include "common/swap.h"
-#include "core/hle/service/nvdrv/devices/nvdevice.h"
+#include "core/hle/service/nvdrv/devices/nvhost_nvdec_common.h"
 
 namespace Service::Nvidia::Devices {
+class nvmap;
 
-class nvhost_vic final : public nvdevice {
+class nvhost_vic final : public nvhost_nvdec_common {
 public:
-    explicit nvhost_vic(Core::System& system);
-    ~nvhost_vic() override;
-
+    explicit nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
+    ~nvhost_vic();
     u32 ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2,
               std::vector<u8>& output, std::vector<u8>& output2, IoctlCtrl& ctrl,
               IoctlVersion version) override;
@@ -28,74 +24,14 @@ private:
         IocGetSyncpoint = 0xC0080002,
         IocGetWaitbase = 0xC0080003,
         IocMapBuffer = 0xC01C0009,
+        IocMapBuffer2 = 0xC0340009,
+        IocMapBuffer3 = 0xC0140009,
+        IocMapBuffer4 = 0xC00C0009,
         IocMapBufferEx = 0xC03C0009,
-        IocUnmapBufferEx = 0xC03C000A,
+        IocUnmapBuffer = 0xC03C000A,
+        IocUnmapBuffer2 = 0xC034000A,
+        IocUnmapBuffer3 = 0xC00C000A,
+        IocUnmapBufferEx = 0xC01C000A,
     };
-
-    struct IoctlSetNvmapFD {
-        u32_le nvmap_fd;
-    };
-    static_assert(sizeof(IoctlSetNvmapFD) == 4, "IoctlSetNvmapFD is incorrect size");
-
-    struct IoctlSubmitCommandBuffer {
-        u32 id;
-        u32 offset;
-        u32 count;
-    };
-    static_assert(sizeof(IoctlSubmitCommandBuffer) == 0xC,
-                  "IoctlSubmitCommandBuffer is incorrect size");
-
-    struct IoctlSubmit {
-        u32 command_buffer_count;
-        u32 relocations_count;
-        u32 syncpt_count;
-        u32 wait_count;
-        std::array<IoctlSubmitCommandBuffer, 4> command_buffer;
-    };
-    static_assert(sizeof(IoctlSubmit) == 0x40, "IoctlSubmit is incorrect size");
-
-    struct IoctlGetSyncpoint {
-        u32 unknown; // seems to be ignored? Nintendo added this
-        u32 value;
-    };
-    static_assert(sizeof(IoctlGetSyncpoint) == 0x8, "IoctlGetSyncpoint is incorrect size");
-
-    struct IoctlGetWaitbase {
-        u32 unknown; // seems to be ignored? Nintendo added this
-        u32 value;
-    };
-    static_assert(sizeof(IoctlGetWaitbase) == 0x8, "IoctlGetWaitbase is incorrect size");
-
-    struct IoctlMapBuffer {
-        u32 unknown;
-        u32 address_1;
-        u32 address_2;
-        INSERT_PADDING_BYTES(0x10); // TODO(DarkLordZach): RE this structure
-    };
-    static_assert(sizeof(IoctlMapBuffer) == 0x1C, "IoctlMapBuffer is incorrect size");
-
-    struct IoctlMapBufferEx {
-        u32 unknown;
-        u32 address_1;
-        u32 address_2;
-        INSERT_PADDING_BYTES(0x30); // TODO(DarkLordZach): RE this structure
-    };
-    static_assert(sizeof(IoctlMapBufferEx) == 0x3C, "IoctlMapBufferEx is incorrect size");
-
-    struct IoctlUnmapBufferEx {
-        INSERT_PADDING_BYTES(0x3C); // TODO(DarkLordZach): RE this structure
-    };
-    static_assert(sizeof(IoctlUnmapBufferEx) == 0x3C, "IoctlUnmapBufferEx is incorrect size");
-
-    u32_le nvmap_fd{};
-
-    u32 SetNVMAPfd(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 Submit(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 GetSyncpoint(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 GetWaitbase(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 MapBuffer(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 MapBufferEx(const std::vector<u8>& input, std::vector<u8>& output);
-    u32 UnmapBufferEx(const std::vector<u8>& input, std::vector<u8>& output);
 };
-
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvmap.h

@@ -37,6 +37,7 @@ public:
         VAddr addr;
         Status status;
         u32 refcount;
+        u32 dma_map_addr;
     };
 
     std::shared_ptr<Object> GetObject(u32 handle) const {

src/core/hle/service/nvdrv/nvdrv.cpp

@@ -21,6 +21,7 @@
 #include "core/hle/service/nvdrv/interface.h"
 #include "core/hle/service/nvdrv/nvdrv.h"
 #include "core/hle/service/nvdrv/nvmemp.h"
+#include "core/hle/service/nvdrv/syncpoint_manager.h"
 #include "core/hle/service/nvflinger/nvflinger.h"
 
 namespace Service::Nvidia {
@@ -36,24 +37,26 @@ void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger
     nvflinger.SetNVDrvInstance(module_);
 }
 
-Module::Module(Core::System& system) {
+Module::Module(Core::System& system) : syncpoint_manager{system.GPU()} {
     auto& kernel = system.Kernel();
     for (u32 i = 0; i < MaxNvEvents; i++) {
         std::string event_label = fmt::format("NVDRV::NvEvent_{}", i);
-        events_interface.events[i] = Kernel::WritableEvent::CreateEventPair(kernel, event_label);
+        events_interface.events[i] = {Kernel::WritableEvent::CreateEventPair(kernel, event_label)};
         events_interface.status[i] = EventState::Free;
         events_interface.registered[i] = false;
     }
     auto nvmap_dev = std::make_shared<Devices::nvmap>(system);
     devices["/dev/nvhost-as-gpu"] = std::make_shared<Devices::nvhost_as_gpu>(system, nvmap_dev);
-    devices["/dev/nvhost-gpu"] = std::make_shared<Devices::nvhost_gpu>(system, nvmap_dev);
+    devices["/dev/nvhost-gpu"] =
+        std::make_shared<Devices::nvhost_gpu>(system, nvmap_dev, syncpoint_manager);
     devices["/dev/nvhost-ctrl-gpu"] = std::make_shared<Devices::nvhost_ctrl_gpu>(system);
     devices["/dev/nvmap"] = nvmap_dev;
     devices["/dev/nvdisp_disp0"] = std::make_shared<Devices::nvdisp_disp0>(system, nvmap_dev);
-    devices["/dev/nvhost-ctrl"] = std::make_shared<Devices::nvhost_ctrl>(system, events_interface);
-    devices["/dev/nvhost-nvdec"] = std::make_shared<Devices::nvhost_nvdec>(system);
+    devices["/dev/nvhost-ctrl"] =
+        std::make_shared<Devices::nvhost_ctrl>(system, events_interface, syncpoint_manager);
+    devices["/dev/nvhost-nvdec"] = std::make_shared<Devices::nvhost_nvdec>(system, nvmap_dev);
     devices["/dev/nvhost-nvjpg"] = std::make_shared<Devices::nvhost_nvjpg>(system);
-    devices["/dev/nvhost-vic"] = std::make_shared<Devices::nvhost_vic>(system);
+    devices["/dev/nvhost-vic"] = std::make_shared<Devices::nvhost_vic>(system, nvmap_dev);
 }
 
 Module::~Module() = default;
@@ -95,17 +98,17 @@ void Module::SignalSyncpt(const u32 syncpoint_id, const u32 value) {
         if (events_interface.assigned_syncpt[i] == syncpoint_id &&
             events_interface.assigned_value[i] == value) {
             events_interface.LiberateEvent(i);
-            events_interface.events[i].writable->Signal();
+            events_interface.events[i].event.writable->Signal();
         }
     }
 }
 
 std::shared_ptr<Kernel::ReadableEvent> Module::GetEvent(const u32 event_id) const {
-    return events_interface.events[event_id].readable;
+    return events_interface.events[event_id].event.readable;
 }
 
 std::shared_ptr<Kernel::WritableEvent> Module::GetEventWriteable(const u32 event_id) const {
-    return events_interface.events[event_id].writable;
+    return events_interface.events[event_id].event.writable;
 }
 
 } // namespace Service::Nvidia

src/core/hle/service/nvdrv/nvdrv.h

@@ -10,6 +10,7 @@
 #include "common/common_types.h"
 #include "core/hle/kernel/writable_event.h"
 #include "core/hle/service/nvdrv/nvdata.h"
+#include "core/hle/service/nvdrv/syncpoint_manager.h"
 #include "core/hle/service/service.h"
 
 namespace Core {
@@ -22,15 +23,23 @@ class NVFlinger;
 
 namespace Service::Nvidia {
 
+class SyncpointManager;
+
 namespace Devices {
 class nvdevice;
 }
 
+/// Represents an Nvidia event
+struct NvEvent {
+    Kernel::EventPair event;
+    Fence fence{};
+};
+
 struct EventInterface {
     // Mask representing currently busy events
     u64 events_mask{};
     // Each kernel event associated to an NV event
-    std::array<Kernel::EventPair, MaxNvEvents> events;
+    std::array<NvEvent, MaxNvEvents> events;
     // The status of the current NVEvent
     std::array<EventState, MaxNvEvents> status{};
     // Tells if an NVEvent is registered or not
@@ -119,6 +128,9 @@ public:
     std::shared_ptr<Kernel::WritableEvent> GetEventWriteable(u32 event_id) const;
 
 private:
+    /// Manages syncpoints on the host
+    SyncpointManager syncpoint_manager;
+
     /// Id to use for the next open file descriptor.
     u32 next_fd = 1;
 

src/core/hle/service/nvdrv/syncpoint_manager.cpp

@@ -0,0 +1,39 @@
+// Copyright 2020 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+#include "core/hle/service/nvdrv/syncpoint_manager.h"
+#include "video_core/gpu.h"
+
+namespace Service::Nvidia {
+
+SyncpointManager::SyncpointManager(Tegra::GPU& gpu) : gpu{gpu} {}
+
+SyncpointManager::~SyncpointManager() = default;
+
+u32 SyncpointManager::RefreshSyncpoint(u32 syncpoint_id) {
+    syncpoints[syncpoint_id].min = gpu.GetSyncpointValue(syncpoint_id);
+    return GetSyncpointMin(syncpoint_id);
+}
+
+u32 SyncpointManager::AllocateSyncpoint() {
+    for (u32 syncpoint_id = 1; syncpoint_id < MaxSyncPoints; syncpoint_id++) {
+        if (!syncpoints[syncpoint_id].is_allocated) {
+            syncpoints[syncpoint_id].is_allocated = true;
+            return syncpoint_id;
+        }
+    }
+    UNREACHABLE_MSG("No more available syncpoints!");
+    return {};
+}
+
+u32 SyncpointManager::IncreaseSyncpoint(u32 syncpoint_id, u32 value) {
+    for (u32 index = 0; index < value; ++index) {
+        syncpoints[syncpoint_id].max.fetch_add(1, std::memory_order_relaxed);
+    }
+
+    return GetSyncpointMax(syncpoint_id);
+}
+
+} // namespace Service::Nvidia

src/core/hle/service/nvdrv/syncpoint_manager.h

@@ -0,0 +1,85 @@
+// Copyright 2020 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <atomic>
+
+#include "common/common_types.h"
+#include "core/hle/service/nvdrv/nvdata.h"
+
+namespace Tegra {
+class GPU;
+}
+
+namespace Service::Nvidia {
+
+class SyncpointManager final {
+public:
+    explicit SyncpointManager(Tegra::GPU& gpu);
+    ~SyncpointManager();
+
+    /**
+     * Returns true if the specified syncpoint is expired for the given value.
+     * @param syncpoint_id Syncpoint ID to check.
+     * @param value Value to check against the specified syncpoint.
+     * @returns True if the specified syncpoint is expired for the given value, otherwise False.
+     */
+    bool IsSyncpointExpired(u32 syncpoint_id, u32 value) const {
+        return (GetSyncpointMax(syncpoint_id) - value) >= (GetSyncpointMin(syncpoint_id) - value);
+    }
+
+    /**
+     * Gets the lower bound for the specified syncpoint.
+     * @param syncpoint_id Syncpoint ID to get the lower bound for.
+     * @returns The lower bound for the specified syncpoint.
+     */
+    u32 GetSyncpointMin(u32 syncpoint_id) const {
+        return syncpoints[syncpoint_id].min.load(std::memory_order_relaxed);
+    }
+
+    /**
+     * Gets the uper bound for the specified syncpoint.
+     * @param syncpoint_id Syncpoint ID to get the upper bound for.
+     * @returns The upper bound for the specified syncpoint.
+     */
+    u32 GetSyncpointMax(u32 syncpoint_id) const {
+        return syncpoints[syncpoint_id].max.load(std::memory_order_relaxed);
+    }
+
+    /**
+     * Refreshes the minimum value for the specified syncpoint.
+     * @param syncpoint_id Syncpoint ID to be refreshed.
+     * @returns The new syncpoint minimum value.
+     */
+    u32 RefreshSyncpoint(u32 syncpoint_id);
+
+    /**
+     * Allocates a new syncoint.
+     * @returns The syncpoint ID for the newly allocated syncpoint.
+     */
+    u32 AllocateSyncpoint();
+
+    /**
+     * Increases the maximum value for the specified syncpoint.
+     * @param syncpoint_id Syncpoint ID to be increased.
+     * @param value Value to increase the specified syncpoint by.
+     * @returns The new syncpoint maximum value.
+     */
+    u32 IncreaseSyncpoint(u32 syncpoint_id, u32 value);
+
+private:
+    struct Syncpoint {
+        std::atomic<u32> min;
+        std::atomic<u32> max;
+        std::atomic<bool> is_allocated;
+    };
+
+    std::array<Syncpoint, MaxSyncPoints> syncpoints{};
+
+    Tegra::GPU& gpu;
+};
+
+} // namespace Service::Nvidia

src/core/hle/service/nvflinger/buffer_queue.cpp

@@ -29,6 +29,10 @@ void BufferQueue::SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer)
         .slot = slot,
         .status = Buffer::Status::Free,
         .igbp_buffer = igbp_buffer,
+        .transform = {},
+        .crop_rect = {},
+        .swap_interval = 0,
+        .multi_fence = {},
     });
 
     buffer_wait_event.writable->Signal();

src/core/hle/service/nvflinger/nvflinger.cpp

@@ -242,6 +242,10 @@ void NVFlinger::Compose() {
 
         const auto& igbp_buffer = buffer->get().igbp_buffer;
 
+        if (!system.IsPoweredOn()) {
+            return; // We are likely shutting down
+        }
+
         auto& gpu = system.GPU();
         const auto& multi_fence = buffer->get().multi_fence;
         guard->unlock();

src/core/hle/service/set/set.cpp

@@ -202,6 +202,7 @@ SET::SET() : ServiceFramework("set") {
         {8, &SET::GetQuestFlag, "GetQuestFlag"},
         {9, &SET::GetKeyCodeMap2, "GetKeyCodeMap2"},
         {10, nullptr, "GetFirmwareVersionForDebug"},
+        {11, nullptr, "GetDeviceNickName"},
     };
     // clang-format on
 

src/core/hle/service/set/set_sys.cpp

@@ -300,6 +300,8 @@ SET_SYS::SET_SYS() : ServiceFramework("set:sys") {
         {198, nullptr, "SetButtonConfigRegisteredSettingsEmbedded"},
         {199, nullptr, "GetButtonConfigRegisteredSettings"},
         {200, nullptr, "SetButtonConfigRegisteredSettings"},
+        {201, nullptr, "GetFieldTestingFlag"},
+        {202, nullptr, "SetFieldTestingFlag"},
     };
     // clang-format on
 

src/core/settings.cpp

@@ -14,7 +14,7 @@
 namespace Settings {
 
 Values values = {};
-bool configuring_global = true;
+static bool configuring_global = true;
 
 std::string GetTimeZoneString() {
     static constexpr std::array timezones{
@@ -63,6 +63,7 @@ void LogSettings() {
     log_setting("Renderer_GPUAccuracyLevel", values.gpu_accuracy.GetValue());
     log_setting("Renderer_UseAsynchronousGpuEmulation",
                 values.use_asynchronous_gpu_emulation.GetValue());
+    log_setting("Renderer_UseNvdecEmulation", values.use_nvdec_emulation.GetValue());
     log_setting("Renderer_UseVsync", values.use_vsync.GetValue());
     log_setting("Renderer_UseAssemblyShaders", values.use_assembly_shaders.GetValue());
     log_setting("Renderer_UseAsynchronousShaders", values.use_asynchronous_shaders.GetValue());
@@ -80,11 +81,12 @@ void LogSettings() {
     log_setting("Services_BCATBoxcatLocal", values.bcat_boxcat_local);
 }
 
-float Volume() {
-    if (values.audio_muted) {
-        return 0.0f;
-    }
-    return values.volume.GetValue();
+bool IsConfiguringGlobal() {
+    return configuring_global;
+}
+
+void SetConfiguringGlobal(bool is_global) {
+    configuring_global = is_global;
 }
 
 bool IsGPULevelExtreme() {
@@ -96,6 +98,13 @@ bool IsGPULevelHigh() {
            values.gpu_accuracy.GetValue() == GPUAccuracy::High;
 }
 
+float Volume() {
+    if (values.audio_muted) {
+        return 0.0f;
+    }
+    return values.volume.GetValue();
+}
+
 void RestoreGlobalState() {
     // If a game is running, DO NOT restore the global settings state
     if (Core::System::GetInstance().IsPoweredOn()) {
@@ -119,6 +128,7 @@ void RestoreGlobalState() {
     values.use_disk_shader_cache.SetGlobal(true);
     values.gpu_accuracy.SetGlobal(true);
     values.use_asynchronous_gpu_emulation.SetGlobal(true);
+    values.use_nvdec_emulation.SetGlobal(true);
     values.use_vsync.SetGlobal(true);
     values.use_assembly_shaders.SetGlobal(true);
     values.use_asynchronous_shaders.SetGlobal(true);

src/core/settings.h

@@ -33,8 +33,6 @@ enum class CPUAccuracy {
     DebugMode = 2,
 };
 
-extern bool configuring_global;
-
 template <typename Type>
 class Setting final {
 public:
@@ -111,6 +109,7 @@ struct Values {
     Setting<bool> use_disk_shader_cache;
     Setting<GPUAccuracy> gpu_accuracy;
     Setting<bool> use_asynchronous_gpu_emulation;
+    Setting<bool> use_nvdec_emulation;
     Setting<bool> use_vsync;
     Setting<bool> use_assembly_shaders;
     Setting<bool> use_asynchronous_shaders;
@@ -197,13 +196,18 @@ struct Values {
 
     // Add-Ons
     std::map<u64, std::vector<std::string>> disabled_addons;
-} extern values;
+};
 
-float Volume();
+extern Values values;
+
+bool IsConfiguringGlobal();
+void SetConfiguringGlobal(bool is_global);
 
 bool IsGPULevelExtreme();
 bool IsGPULevelHigh();
 
+float Volume();
+
 std::string GetTimeZoneString();
 
 void Apply();

src/core/telemetry_session.cpp

@@ -206,6 +206,8 @@ void TelemetrySession::AddInitialInfo(Loader::AppLoader& app_loader) {
              TranslateGPUAccuracyLevel(Settings::values.gpu_accuracy.GetValue()));
     AddField(field_type, "Renderer_UseAsynchronousGpuEmulation",
              Settings::values.use_asynchronous_gpu_emulation.GetValue());
+    AddField(field_type, "Renderer_UseNvdecEmulation",
+             Settings::values.use_nvdec_emulation.GetValue());
     AddField(field_type, "Renderer_UseVsync", Settings::values.use_vsync.GetValue());
     AddField(field_type, "Renderer_UseAssemblyShaders",
              Settings::values.use_assembly_shaders.GetValue());

src/input_common/gcadapter/gc_adapter.cpp

@@ -21,26 +21,6 @@
 
 namespace GCAdapter {
 
-// Used to loop through and assign button in poller
-constexpr std::array<PadButton, 12> PadButtonArray{
-    PadButton::PAD_BUTTON_LEFT, PadButton::PAD_BUTTON_RIGHT, PadButton::PAD_BUTTON_DOWN,
-    PadButton::PAD_BUTTON_UP,   PadButton::PAD_TRIGGER_Z,    PadButton::PAD_TRIGGER_R,
-    PadButton::PAD_TRIGGER_L,   PadButton::PAD_BUTTON_A,     PadButton::PAD_BUTTON_B,
-    PadButton::PAD_BUTTON_X,    PadButton::PAD_BUTTON_Y,     PadButton::PAD_BUTTON_START,
-};
-
-static void PadToState(const GCPadStatus& pad, GCState& out_state) {
-    for (const auto& button : PadButtonArray) {
-        const auto button_key = static_cast<u16>(button);
-        const auto button_value = (pad.button & button_key) != 0;
-        out_state.buttons.insert_or_assign(static_cast<s32>(button_key), button_value);
-    }
-
-    for (std::size_t i = 0; i < pad.axis_values.size(); ++i) {
-        out_state.axes.insert_or_assign(static_cast<u32>(i), pad.axis_values[i]);
-    }
-}
-
 Adapter::Adapter() {
     if (usb_adapter_handle != nullptr) {
         return;
@@ -49,168 +29,263 @@ Adapter::Adapter() {
 
     const int init_res = libusb_init(&libusb_ctx);
     if (init_res == LIBUSB_SUCCESS) {
-        Setup();
+        adapter_scan_thread = std::thread(&Adapter::AdapterScanThread, this);
     } else {
         LOG_ERROR(Input, "libusb could not be initialized. failed with error = {}", init_res);
     }
 }
 
-GCPadStatus Adapter::GetPadStatus(std::size_t port, const std::array<u8, 37>& adapter_payload) {
-    GCPadStatus pad = {};
-    const std::size_t offset = 1 + (9 * port);
+Adapter::~Adapter() {
+    Reset();
+}
 
-    adapter_controllers_status[port] = static_cast<ControllerTypes>(adapter_payload[offset] >> 4);
+void Adapter::AdapterInputThread() {
+    LOG_DEBUG(Input, "GC Adapter input thread started");
+    s32 payload_size{};
+    AdapterPayload adapter_payload{};
+
+    if (adapter_scan_thread.joinable()) {
+        adapter_scan_thread.join();
+    }
+
+    while (adapter_input_thread_running) {
+        libusb_interrupt_transfer(usb_adapter_handle, input_endpoint, adapter_payload.data(),
+                                  static_cast<s32>(adapter_payload.size()), &payload_size, 16);
+        if (IsPayloadCorrect(adapter_payload, payload_size)) {
+            UpdateControllers(adapter_payload);
+            UpdateVibrations();
+        }
+        std::this_thread::yield();
+    }
+
+    if (restart_scan_thread) {
+        adapter_scan_thread = std::thread(&Adapter::AdapterScanThread, this);
+        restart_scan_thread = false;
+    }
+}
+
+bool Adapter::IsPayloadCorrect(const AdapterPayload& adapter_payload, s32 payload_size) {
+    if (payload_size != static_cast<s32>(adapter_payload.size()) ||
+        adapter_payload[0] != LIBUSB_DT_HID) {
+        LOG_DEBUG(Input, "Error reading payload (size: {}, type: {:02x})", payload_size,
+                  adapter_payload[0]);
+        if (input_error_counter++ > 20) {
+            LOG_ERROR(Input, "GC adapter timeout, Is the adapter connected?");
+            adapter_input_thread_running = false;
+            restart_scan_thread = true;
+        }
+        return false;
+    }
+
+    input_error_counter = 0;
+    return true;
+}
+
+void Adapter::UpdateControllers(const AdapterPayload& adapter_payload) {
+    for (std::size_t port = 0; port < pads.size(); ++port) {
+        const std::size_t offset = 1 + (9 * port);
+        const auto type = static_cast<ControllerTypes>(adapter_payload[offset] >> 4);
+        UpdatePadType(port, type);
+        if (DeviceConnected(port)) {
+            const u8 b1 = adapter_payload[offset + 1];
+            const u8 b2 = adapter_payload[offset + 2];
+            UpdateStateButtons(port, b1, b2);
+            UpdateStateAxes(port, adapter_payload);
+            if (configuring) {
+                UpdateYuzuSettings(port);
+            }
+        }
+    }
+}
+
+void Adapter::UpdatePadType(std::size_t port, ControllerTypes pad_type) {
+    if (pads[port].type == pad_type) {
+        return;
+    }
+    // Device changed reset device and set new type
+    ResetDevice(port);
+    pads[port].type = pad_type;
+}
+
+void Adapter::UpdateStateButtons(std::size_t port, u8 b1, u8 b2) {
+    if (port >= pads.size()) {
+        return;
+    }
 
     static constexpr std::array<PadButton, 8> b1_buttons{
-        PadButton::PAD_BUTTON_A,    PadButton::PAD_BUTTON_B,    PadButton::PAD_BUTTON_X,
-        PadButton::PAD_BUTTON_Y,    PadButton::PAD_BUTTON_LEFT, PadButton::PAD_BUTTON_RIGHT,
-        PadButton::PAD_BUTTON_DOWN, PadButton::PAD_BUTTON_UP,
+        PadButton::ButtonA,    PadButton::ButtonB,     PadButton::ButtonX,    PadButton::ButtonY,
+        PadButton::ButtonLeft, PadButton::ButtonRight, PadButton::ButtonDown, PadButton::ButtonUp,
     };
 
     static constexpr std::array<PadButton, 4> b2_buttons{
-        PadButton::PAD_BUTTON_START,
-        PadButton::PAD_TRIGGER_Z,
-        PadButton::PAD_TRIGGER_R,
-        PadButton::PAD_TRIGGER_L,
+        PadButton::ButtonStart,
+        PadButton::TriggerZ,
+        PadButton::TriggerR,
+        PadButton::TriggerL,
     };
+    pads[port].buttons = 0;
+    for (std::size_t i = 0; i < b1_buttons.size(); ++i) {
+        if ((b1 & (1U << i)) != 0) {
+            pads[port].buttons =
+                static_cast<u16>(pads[port].buttons | static_cast<u16>(b1_buttons[i]));
+            pads[port].last_button = b1_buttons[i];
+        }
+    }
 
+    for (std::size_t j = 0; j < b2_buttons.size(); ++j) {
+        if ((b2 & (1U << j)) != 0) {
+            pads[port].buttons =
+                static_cast<u16>(pads[port].buttons | static_cast<u16>(b2_buttons[j]));
+            pads[port].last_button = b2_buttons[j];
+        }
+    }
+}
+
+void Adapter::UpdateStateAxes(std::size_t port, const AdapterPayload& adapter_payload) {
+    if (port >= pads.size()) {
+        return;
+    }
+
+    const std::size_t offset = 1 + (9 * port);
     static constexpr std::array<PadAxes, 6> axes{
         PadAxes::StickX,    PadAxes::StickY,      PadAxes::SubstickX,
         PadAxes::SubstickY, PadAxes::TriggerLeft, PadAxes::TriggerRight,
     };
 
-    if (adapter_controllers_status[port] == ControllerTypes::None && !get_origin[port]) {
-        // Controller may have been disconnected, recalibrate if reconnected.
-        get_origin[port] = true;
+    for (const PadAxes axis : axes) {
+        const auto index = static_cast<std::size_t>(axis);
+        const u8 axis_value = adapter_payload[offset + 3 + index];
+        if (pads[port].axis_origin[index] == 255) {
+            pads[port].axis_origin[index] = axis_value;
+        }
+        pads[port].axis_values[index] =
+            static_cast<s16>(axis_value - pads[port].axis_origin[index]);
     }
-
-    if (adapter_controllers_status[port] != ControllerTypes::None) {
-        const u8 b1 = adapter_payload[offset + 1];
-        const u8 b2 = adapter_payload[offset + 2];
-
-        for (std::size_t i = 0; i < b1_buttons.size(); ++i) {
-            if ((b1 & (1U << i)) != 0) {
-                pad.button = static_cast<u16>(pad.button | static_cast<u16>(b1_buttons[i]));
-            }
-        }
-
-        for (std::size_t j = 0; j < b2_buttons.size(); ++j) {
-            if ((b2 & (1U << j)) != 0) {
-                pad.button = static_cast<u16>(pad.button | static_cast<u16>(b2_buttons[j]));
-            }
-        }
-        for (PadAxes axis : axes) {
-            const auto index = static_cast<std::size_t>(axis);
-            pad.axis_values[index] = adapter_payload[offset + 3 + index];
-        }
-
-        if (get_origin[port]) {
-            origin_status[port].axis_values = pad.axis_values;
-            get_origin[port] = false;
-        }
-    }
-    return pad;
 }
 
-void Adapter::Read() {
-    LOG_DEBUG(Input, "GC Adapter Read() thread started");
+void Adapter::UpdateYuzuSettings(std::size_t port) {
+    if (port >= pads.size()) {
+        return;
+    }
 
-    int payload_size;
-    std::array<u8, 37> adapter_payload;
-    std::array<GCPadStatus, 4> pads;
+    constexpr u8 axis_threshold = 50;
+    GCPadStatus pad_status = {.port = port};
 
-    while (adapter_thread_running) {
-        libusb_interrupt_transfer(usb_adapter_handle, input_endpoint, adapter_payload.data(),
-                                  sizeof(adapter_payload), &payload_size, 16);
+    if (pads[port].buttons != 0) {
+        pad_status.button = pads[port].last_button;
+        pad_queue.Push(pad_status);
+    }
 
-        if (payload_size != sizeof(adapter_payload) || adapter_payload[0] != LIBUSB_DT_HID) {
-            LOG_ERROR(Input,
-                      "Error reading payload (size: {}, type: {:02x}) Is the adapter connected?",
-                      payload_size, adapter_payload[0]);
-            adapter_thread_running = false; // error reading from adapter, stop reading.
-            break;
+    // Accounting for a threshold here to ensure an intentional press
+    for (std::size_t i = 0; i < pads[port].axis_values.size(); ++i) {
+        const s16 value = pads[port].axis_values[i];
+
+        if (value > axis_threshold || value < -axis_threshold) {
+            pad_status.axis = static_cast<PadAxes>(i);
+            pad_status.axis_value = value;
+            pad_status.axis_threshold = axis_threshold;
+            pad_queue.Push(pad_status);
         }
-        for (std::size_t port = 0; port < pads.size(); ++port) {
-            pads[port] = GetPadStatus(port, adapter_payload);
-            if (DeviceConnected(port) && configuring) {
-                if (pads[port].button != 0) {
-                    pad_queue[port].Push(pads[port]);
-                }
+    }
+}
 
-                // Accounting for a threshold here to ensure an intentional press
-                for (size_t i = 0; i < pads[port].axis_values.size(); ++i) {
-                    const u8 value = pads[port].axis_values[i];
-                    const u8 origin = origin_status[port].axis_values[i];
+void Adapter::UpdateVibrations() {
+    // Use 8 states to keep the switching between on/off fast enough for
+    // a human to not notice the difference between switching from on/off
+    // More states = more rumble strengths = slower update time
+    constexpr u8 vibration_states = 8;
 
-                    if (value > origin + pads[port].THRESHOLD ||
-                        value < origin - pads[port].THRESHOLD) {
-                        pads[port].axis = static_cast<PadAxes>(i);
-                        pads[port].axis_value = pads[port].axis_values[i];
-                        pad_queue[port].Push(pads[port]);
-                    }
-                }
-            }
-            PadToState(pads[port], state[port]);
+    vibration_counter = (vibration_counter + 1) % vibration_states;
+
+    for (GCController& pad : pads) {
+        const bool vibrate = pad.rumble_amplitude > vibration_counter;
+        vibration_changed |= vibrate != pad.enable_vibration;
+        pad.enable_vibration = vibrate;
+    }
+    SendVibrations();
+}
+
+void Adapter::SendVibrations() {
+    if (!rumble_enabled || !vibration_changed) {
+        return;
+    }
+    s32 size{};
+    constexpr u8 rumble_command = 0x11;
+    const u8 p1 = pads[0].enable_vibration;
+    const u8 p2 = pads[1].enable_vibration;
+    const u8 p3 = pads[2].enable_vibration;
+    const u8 p4 = pads[3].enable_vibration;
+    std::array<u8, 5> payload = {rumble_command, p1, p2, p3, p4};
+    const int err = libusb_interrupt_transfer(usb_adapter_handle, output_endpoint, payload.data(),
+                                              static_cast<s32>(payload.size()), &size, 16);
+    if (err) {
+        LOG_DEBUG(Input, "Adapter libusb write failed: {}", libusb_error_name(err));
+        if (output_error_counter++ > 5) {
+            LOG_ERROR(Input, "GC adapter output timeout, Rumble disabled");
+            rumble_enabled = false;
         }
-        std::this_thread::yield();
+        return;
+    }
+    output_error_counter = 0;
+    vibration_changed = false;
+}
+
+bool Adapter::RumblePlay(std::size_t port, f32 amplitude) {
+    amplitude = std::clamp(amplitude, 0.0f, 1.0f);
+    const auto raw_amp = static_cast<u8>(amplitude * 0x8);
+    pads[port].rumble_amplitude = raw_amp;
+
+    return rumble_enabled;
+}
+
+void Adapter::AdapterScanThread() {
+    adapter_scan_thread_running = true;
+    adapter_input_thread_running = false;
+    if (adapter_input_thread.joinable()) {
+        adapter_input_thread.join();
+    }
+    ClearLibusbHandle();
+    ResetDevices();
+    while (adapter_scan_thread_running && !adapter_input_thread_running) {
+        Setup();
+        std::this_thread::sleep_for(std::chrono::seconds(1));
     }
 }
 
 void Adapter::Setup() {
-    // Initialize all controllers as unplugged
-    adapter_controllers_status.fill(ControllerTypes::None);
-    // Initialize all ports to store axis origin values
-    get_origin.fill(true);
+    usb_adapter_handle = libusb_open_device_with_vid_pid(libusb_ctx, 0x057e, 0x0337);
 
-    // pointer to list of connected usb devices
-    libusb_device** devices{};
-
-    // populate the list of devices, get the count
-    const ssize_t device_count = libusb_get_device_list(libusb_ctx, &devices);
-    if (device_count < 0) {
-        LOG_ERROR(Input, "libusb_get_device_list failed with error: {}", device_count);
+    if (usb_adapter_handle == NULL) {
+        return;
+    }
+    if (!CheckDeviceAccess()) {
+        ClearLibusbHandle();
         return;
     }
 
-    if (devices != nullptr) {
-        for (std::size_t index = 0; index < static_cast<std::size_t>(device_count); ++index) {
-            if (CheckDeviceAccess(devices[index])) {
-                // GC Adapter found and accessible, registering it
-                GetGCEndpoint(devices[index]);
-                break;
-            }
-        }
-        libusb_free_device_list(devices, 1);
+    libusb_device* device = libusb_get_device(usb_adapter_handle);
+
+    LOG_INFO(Input, "GC adapter is now connected");
+    // GC Adapter found and accessible, registering it
+    if (GetGCEndpoint(device)) {
+        adapter_scan_thread_running = false;
+        adapter_input_thread_running = true;
+        rumble_enabled = true;
+        input_error_counter = 0;
+        output_error_counter = 0;
+        adapter_input_thread = std::thread(&Adapter::AdapterInputThread, this);
     }
 }
 
-bool Adapter::CheckDeviceAccess(libusb_device* device) {
-    libusb_device_descriptor desc;
-    const int get_descriptor_error = libusb_get_device_descriptor(device, &desc);
-    if (get_descriptor_error) {
-        // could not acquire the descriptor, no point in trying to use it.
-        LOG_ERROR(Input, "libusb_get_device_descriptor failed with error: {}",
-                  get_descriptor_error);
-        return false;
+bool Adapter::CheckDeviceAccess() {
+    // This fixes payload problems from offbrand GCAdapters
+    const s32 control_transfer_error =
+        libusb_control_transfer(usb_adapter_handle, 0x21, 11, 0x0001, 0, nullptr, 0, 1000);
+    if (control_transfer_error < 0) {
+        LOG_ERROR(Input, "libusb_control_transfer failed with error= {}", control_transfer_error);
     }
 
-    if (desc.idVendor != 0x057e || desc.idProduct != 0x0337) {
-        // This isn't the device we are looking for.
-        return false;
-    }
-    const int open_error = libusb_open(device, &usb_adapter_handle);
-
-    if (open_error == LIBUSB_ERROR_ACCESS) {
-        LOG_ERROR(Input, "Yuzu can not gain access to this device: ID {:04X}:{:04X}.",
-                  desc.idVendor, desc.idProduct);
-        return false;
-    }
-    if (open_error) {
-        LOG_ERROR(Input, "libusb_open failed to open device with error = {}", open_error);
-        return false;
-    }
-
-    int kernel_driver_error = libusb_kernel_driver_active(usb_adapter_handle, 0);
+    s32 kernel_driver_error = libusb_kernel_driver_active(usb_adapter_handle, 0);
     if (kernel_driver_error == 1) {
         kernel_driver_error = libusb_detach_kernel_driver(usb_adapter_handle, 0);
         if (kernel_driver_error != 0 && kernel_driver_error != LIBUSB_ERROR_NOT_SUPPORTED) {
@@ -236,13 +311,13 @@ bool Adapter::CheckDeviceAccess(libusb_device* device) {
     return true;
 }
 
-void Adapter::GetGCEndpoint(libusb_device* device) {
+bool Adapter::GetGCEndpoint(libusb_device* device) {
     libusb_config_descriptor* config = nullptr;
     const int config_descriptor_return = libusb_get_config_descriptor(device, 0, &config);
     if (config_descriptor_return != LIBUSB_SUCCESS) {
         LOG_ERROR(Input, "libusb_get_config_descriptor failed with error = {}",
                   config_descriptor_return);
-        return;
+        return false;
     }
 
     for (u8 ic = 0; ic < config->bNumInterfaces; ic++) {
@@ -264,31 +339,51 @@ void Adapter::GetGCEndpoint(libusb_device* device) {
     unsigned char clear_payload = 0x13;
     libusb_interrupt_transfer(usb_adapter_handle, output_endpoint, &clear_payload,
                               sizeof(clear_payload), nullptr, 16);
-
-    adapter_thread_running = true;
-    adapter_input_thread = std::thread(&Adapter::Read, this);
+    return true;
 }
 
-Adapter::~Adapter() {
-    Reset();
-}
+void Adapter::JoinThreads() {
+    restart_scan_thread = false;
+    adapter_input_thread_running = false;
+    adapter_scan_thread_running = false;
 
-void Adapter::Reset() {
-    if (adapter_thread_running) {
-        adapter_thread_running = false;
+    if (adapter_scan_thread.joinable()) {
+        adapter_scan_thread.join();
     }
+
     if (adapter_input_thread.joinable()) {
         adapter_input_thread.join();
     }
+}
 
-    adapter_controllers_status.fill(ControllerTypes::None);
-    get_origin.fill(true);
-
+void Adapter::ClearLibusbHandle() {
     if (usb_adapter_handle) {
         libusb_release_interface(usb_adapter_handle, 1);
         libusb_close(usb_adapter_handle);
         usb_adapter_handle = nullptr;
     }
+}
+
+void Adapter::ResetDevices() {
+    for (std::size_t i = 0; i < pads.size(); ++i) {
+        ResetDevice(i);
+    }
+}
+
+void Adapter::ResetDevice(std::size_t port) {
+    pads[port].type = ControllerTypes::None;
+    pads[port].enable_vibration = false;
+    pads[port].rumble_amplitude = 0;
+    pads[port].buttons = 0;
+    pads[port].last_button = PadButton::Undefined;
+    pads[port].axis_values.fill(0);
+    pads[port].axis_origin.fill(255);
+}
+
+void Adapter::Reset() {
+    JoinThreads();
+    ClearLibusbHandle();
+    ResetDevices();
 
     if (libusb_ctx) {
         libusb_exit(libusb_ctx);
@@ -297,11 +392,11 @@ void Adapter::Reset() {
 
 std::vector<Common::ParamPackage> Adapter::GetInputDevices() const {
     std::vector<Common::ParamPackage> devices;
-    for (std::size_t port = 0; port < state.size(); ++port) {
+    for (std::size_t port = 0; port < pads.size(); ++port) {
         if (!DeviceConnected(port)) {
             continue;
         }
-        std::string name = fmt::format("Gamecube Controller {}", port);
+        std::string name = fmt::format("Gamecube Controller {}", port + 1);
         devices.emplace_back(Common::ParamPackage{
             {"class", "gcpad"},
             {"display", std::move(name)},
@@ -318,18 +413,18 @@ InputCommon::ButtonMapping Adapter::GetButtonMappingForDevice(
     // This list also excludes any button that can't be really mapped
     static constexpr std::array<std::pair<Settings::NativeButton::Values, PadButton>, 12>
         switch_to_gcadapter_button = {
-            std::pair{Settings::NativeButton::A, PadButton::PAD_BUTTON_A},
-            {Settings::NativeButton::B, PadButton::PAD_BUTTON_B},
-            {Settings::NativeButton::X, PadButton::PAD_BUTTON_X},
-            {Settings::NativeButton::Y, PadButton::PAD_BUTTON_Y},
-            {Settings::NativeButton::Plus, PadButton::PAD_BUTTON_START},
-            {Settings::NativeButton::DLeft, PadButton::PAD_BUTTON_LEFT},
-            {Settings::NativeButton::DUp, PadButton::PAD_BUTTON_UP},
-            {Settings::NativeButton::DRight, PadButton::PAD_BUTTON_RIGHT},
-            {Settings::NativeButton::DDown, PadButton::PAD_BUTTON_DOWN},
-            {Settings::NativeButton::SL, PadButton::PAD_TRIGGER_L},
-            {Settings::NativeButton::SR, PadButton::PAD_TRIGGER_R},
-            {Settings::NativeButton::R, PadButton::PAD_TRIGGER_Z},
+            std::pair{Settings::NativeButton::A, PadButton::ButtonA},
+            {Settings::NativeButton::B, PadButton::ButtonB},
+            {Settings::NativeButton::X, PadButton::ButtonX},
+            {Settings::NativeButton::Y, PadButton::ButtonY},
+            {Settings::NativeButton::Plus, PadButton::ButtonStart},
+            {Settings::NativeButton::DLeft, PadButton::ButtonLeft},
+            {Settings::NativeButton::DUp, PadButton::ButtonUp},
+            {Settings::NativeButton::DRight, PadButton::ButtonRight},
+            {Settings::NativeButton::DDown, PadButton::ButtonDown},
+            {Settings::NativeButton::SL, PadButton::TriggerL},
+            {Settings::NativeButton::SR, PadButton::TriggerR},
+            {Settings::NativeButton::R, PadButton::TriggerZ},
         };
     if (!params.Has("port")) {
         return {};
@@ -352,8 +447,10 @@ InputCommon::ButtonMapping Adapter::GetButtonMappingForDevice(
     for (const auto& [switch_button, gcadapter_axis] : switch_to_gcadapter_axis) {
         Common::ParamPackage button_params({{"engine", "gcpad"}});
         button_params.Set("port", params.Get("port", 0));
-        button_params.Set("button", static_cast<int>(PadButton::PAD_STICK));
-        button_params.Set("axis", static_cast<int>(gcadapter_axis));
+        button_params.Set("button", static_cast<s32>(PadButton::Stick));
+        button_params.Set("axis", static_cast<s32>(gcadapter_axis));
+        button_params.Set("threshold", 0.5f);
+        button_params.Set("direction", "+");
         mapping.insert_or_assign(switch_button, std::move(button_params));
     }
     return mapping;
@@ -382,46 +479,33 @@ InputCommon::AnalogMapping Adapter::GetAnalogMappingForDevice(
 }
 
 bool Adapter::DeviceConnected(std::size_t port) const {
-    return adapter_controllers_status[port] != ControllerTypes::None;
-}
-
-void Adapter::ResetDeviceType(std::size_t port) {
-    adapter_controllers_status[port] = ControllerTypes::None;
+    return pads[port].type != ControllerTypes::None;
 }
 
 void Adapter::BeginConfiguration() {
-    get_origin.fill(true);
-    for (auto& pq : pad_queue) {
-        pq.Clear();
-    }
+    pad_queue.Clear();
     configuring = true;
 }
 
 void Adapter::EndConfiguration() {
-    for (auto& pq : pad_queue) {
-        pq.Clear();
-    }
+    pad_queue.Clear();
     configuring = false;
 }
 
-std::array<Common::SPSCQueue<GCPadStatus>, 4>& Adapter::GetPadQueue() {
+Common::SPSCQueue<GCPadStatus>& Adapter::GetPadQueue() {
     return pad_queue;
 }
 
-const std::array<Common::SPSCQueue<GCPadStatus>, 4>& Adapter::GetPadQueue() const {
+const Common::SPSCQueue<GCPadStatus>& Adapter::GetPadQueue() const {
     return pad_queue;
 }
 
-std::array<GCState, 4>& Adapter::GetPadState() {
-    return state;
+GCController& Adapter::GetPadState(std::size_t port) {
+    return pads.at(port);
 }
 
-const std::array<GCState, 4>& Adapter::GetPadState() const {
-    return state;
-}
-
-int Adapter::GetOriginValue(u32 port, u32 axis) const {
-    return origin_status[port].axis_values[axis];
+const GCController& Adapter::GetPadState(std::size_t port) const {
+    return pads.at(port);
 }
 
 } // namespace GCAdapter

src/input_common/gcadapter/gc_adapter.h

@@ -19,24 +19,23 @@ struct libusb_device_handle;
 namespace GCAdapter {
 
 enum class PadButton {
-    PAD_BUTTON_LEFT = 0x0001,
-    PAD_BUTTON_RIGHT = 0x0002,
-    PAD_BUTTON_DOWN = 0x0004,
-    PAD_BUTTON_UP = 0x0008,
-    PAD_TRIGGER_Z = 0x0010,
-    PAD_TRIGGER_R = 0x0020,
-    PAD_TRIGGER_L = 0x0040,
-    PAD_BUTTON_A = 0x0100,
-    PAD_BUTTON_B = 0x0200,
-    PAD_BUTTON_X = 0x0400,
-    PAD_BUTTON_Y = 0x0800,
-    PAD_BUTTON_START = 0x1000,
+    Undefined = 0x0000,
+    ButtonLeft = 0x0001,
+    ButtonRight = 0x0002,
+    ButtonDown = 0x0004,
+    ButtonUp = 0x0008,
+    TriggerZ = 0x0010,
+    TriggerR = 0x0020,
+    TriggerL = 0x0040,
+    ButtonA = 0x0100,
+    ButtonB = 0x0200,
+    ButtonX = 0x0400,
+    ButtonY = 0x0800,
+    ButtonStart = 0x1000,
     // Below is for compatibility with "AxisButton" type
-    PAD_STICK = 0x2000,
+    Stick = 0x2000,
 };
 
-extern const std::array<PadButton, 12> PadButtonArray;
-
 enum class PadAxes : u8 {
     StickX,
     StickY,
@@ -47,87 +46,122 @@ enum class PadAxes : u8 {
     Undefined,
 };
 
+enum class ControllerTypes {
+    None,
+    Wired,
+    Wireless,
+};
+
 struct GCPadStatus {
-    u16 button{}; // Or-ed PAD_BUTTON_* and PAD_TRIGGER_* bits
+    std::size_t port{};
 
-    std::array<u8, 6> axis_values{};    // Triggers and sticks, following indices defined in PadAxes
-    static constexpr u8 THRESHOLD = 50; // Threshold for axis press for polling
+    PadButton button{PadButton::Undefined}; // Or-ed PAD_BUTTON_* and PAD_TRIGGER_* bits
 
-    u8 port{};
     PadAxes axis{PadAxes::Undefined};
-    u8 axis_value{255};
+    s16 axis_value{};
+    u8 axis_threshold{50};
 };
 
-struct GCState {
-    std::unordered_map<int, bool> buttons;
-    std::unordered_map<u32, u16> axes;
+struct GCController {
+    ControllerTypes type{};
+    bool enable_vibration{};
+    u8 rumble_amplitude{};
+    u16 buttons{};
+    PadButton last_button{};
+    std::array<s16, 6> axis_values{};
+    std::array<u8, 6> axis_origin{};
 };
 
-enum class ControllerTypes { None, Wired, Wireless };
-
 class Adapter {
 public:
-    /// Initialize the GC Adapter capture and read sequence
     Adapter();
-
-    /// Close the adapter read thread and release the adapter
     ~Adapter();
+
+    /// Request a vibration for a controlelr
+    bool RumblePlay(std::size_t port, f32 amplitude);
+
     /// Used for polling
     void BeginConfiguration();
     void EndConfiguration();
 
-    std::vector<Common::ParamPackage> GetInputDevices() const;
-    InputCommon::ButtonMapping GetButtonMappingForDevice(const Common::ParamPackage& params) const;
-    InputCommon::AnalogMapping GetAnalogMappingForDevice(const Common::ParamPackage& params) const;
+    Common::SPSCQueue<GCPadStatus>& GetPadQueue();
+    const Common::SPSCQueue<GCPadStatus>& GetPadQueue() const;
+
+    GCController& GetPadState(std::size_t port);
+    const GCController& GetPadState(std::size_t port) const;
 
     /// Returns true if there is a device connected to port
     bool DeviceConnected(std::size_t port) const;
 
-    std::array<Common::SPSCQueue<GCPadStatus>, 4>& GetPadQueue();
-    const std::array<Common::SPSCQueue<GCPadStatus>, 4>& GetPadQueue() const;
-
-    std::array<GCState, 4>& GetPadState();
-    const std::array<GCState, 4>& GetPadState() const;
-
-    int GetOriginValue(u32 port, u32 axis) const;
+    /// Used for automapping features
+    std::vector<Common::ParamPackage> GetInputDevices() const;
+    InputCommon::ButtonMapping GetButtonMappingForDevice(const Common::ParamPackage& params) const;
+    InputCommon::AnalogMapping GetAnalogMappingForDevice(const Common::ParamPackage& params) const;
 
 private:
-    GCPadStatus GetPadStatus(std::size_t port, const std::array<u8, 37>& adapter_payload);
+    using AdapterPayload = std::array<u8, 37>;
 
-    void Read();
+    void UpdatePadType(std::size_t port, ControllerTypes pad_type);
+    void UpdateControllers(const AdapterPayload& adapter_payload);
+    void UpdateYuzuSettings(std::size_t port);
+    void UpdateStateButtons(std::size_t port, u8 b1, u8 b2);
+    void UpdateStateAxes(std::size_t port, const AdapterPayload& adapter_payload);
+    void UpdateVibrations();
 
-    /// Resets status of device connected to port
-    void ResetDeviceType(std::size_t port);
+    void AdapterInputThread();
 
-    /// Returns true if we successfully gain access to GC Adapter
-    bool CheckDeviceAccess(libusb_device* device);
+    void AdapterScanThread();
 
-    /// Captures GC Adapter endpoint address,
-    void GetGCEndpoint(libusb_device* device);
+    bool IsPayloadCorrect(const AdapterPayload& adapter_payload, s32 payload_size);
 
-    /// For shutting down, clear all data, join all threads, release usb
-    void Reset();
+    // Updates vibration state of all controllers
+    void SendVibrations();
 
     /// For use in initialization, querying devices to find the adapter
     void Setup();
 
+    /// Resets status of all GC controller devices to a disconected state
+    void ResetDevices();
+
+    /// Resets status of device connected to a disconected state
+    void ResetDevice(std::size_t port);
+
+    /// Returns true if we successfully gain access to GC Adapter
+    bool CheckDeviceAccess();
+
+    /// Captures GC Adapter endpoint address
+    /// Returns true if the endpoind was set correctly
+    bool GetGCEndpoint(libusb_device* device);
+
+    /// For shutting down, clear all data, join all threads, release usb
+    void Reset();
+
+    // Join all threads
+    void JoinThreads();
+
+    // Release usb handles
+    void ClearLibusbHandle();
+
     libusb_device_handle* usb_adapter_handle = nullptr;
+    std::array<GCController, 4> pads;
+    Common::SPSCQueue<GCPadStatus> pad_queue;
 
     std::thread adapter_input_thread;
-    bool adapter_thread_running;
+    std::thread adapter_scan_thread;
+    bool adapter_input_thread_running;
+    bool adapter_scan_thread_running;
+    bool restart_scan_thread;
 
     libusb_context* libusb_ctx;
 
-    u8 input_endpoint = 0;
-    u8 output_endpoint = 0;
+    u8 input_endpoint{0};
+    u8 output_endpoint{0};
+    u8 input_error_counter{0};
+    u8 output_error_counter{0};
+    int vibration_counter{0};
 
-    bool configuring = false;
-
-    std::array<GCState, 4> state;
-    std::array<bool, 4> get_origin;
-    std::array<GCPadStatus, 4> origin_status;
-    std::array<Common::SPSCQueue<GCPadStatus>, 4> pad_queue;
-    std::array<ControllerTypes, 4> adapter_controllers_status{};
+    bool configuring{false};
+    bool rumble_enabled{true};
+    bool vibration_changed{true};
 };
-
 } // namespace GCAdapter

src/input_common/gcadapter/gc_poller.cpp

@@ -15,22 +15,30 @@ namespace InputCommon {
 
 class GCButton final : public Input::ButtonDevice {
 public:
-    explicit GCButton(u32 port_, int button_, const GCAdapter::Adapter* adapter)
+    explicit GCButton(u32 port_, s32 button_, GCAdapter::Adapter* adapter)
         : port(port_), button(button_), gcadapter(adapter) {}
 
     ~GCButton() override;
 
     bool GetStatus() const override {
         if (gcadapter->DeviceConnected(port)) {
-            return gcadapter->GetPadState()[port].buttons.at(button);
+            return (gcadapter->GetPadState(port).buttons & button) != 0;
         }
         return false;
     }
 
+    bool SetRumblePlay(f32 amp_high, f32 amp_low, f32 freq_high, f32 freq_low) const override {
+        const float amplitude = amp_high + amp_low > 2.0f ? 1.0f : (amp_high + amp_low) * 0.5f;
+        const auto new_amp =
+            static_cast<f32>(pow(amplitude, 0.5f) * (3.0f - 2.0f * pow(amplitude, 0.15f)));
+
+        return gcadapter->RumblePlay(port, new_amp);
+    }
+
 private:
     const u32 port;
-    const int button;
-    const GCAdapter::Adapter* gcadapter;
+    const s32 button;
+    GCAdapter::Adapter* gcadapter;
 };
 
 class GCAxisButton final : public Input::ButtonDevice {
@@ -38,13 +46,12 @@ public:
     explicit GCAxisButton(u32 port_, u32 axis_, float threshold_, bool trigger_if_greater_,
                           const GCAdapter::Adapter* adapter)
         : port(port_), axis(axis_), threshold(threshold_), trigger_if_greater(trigger_if_greater_),
-          gcadapter(adapter),
-          origin_value(static_cast<float>(adapter->GetOriginValue(port_, axis_))) {}
+          gcadapter(adapter) {}
 
     bool GetStatus() const override {
         if (gcadapter->DeviceConnected(port)) {
-            const float current_axis_value = gcadapter->GetPadState()[port].axes.at(axis);
-            const float axis_value = (current_axis_value - origin_value) / 128.0f;
+            const float current_axis_value = gcadapter->GetPadState(port).axis_values.at(axis);
+            const float axis_value = current_axis_value / 128.0f;
             if (trigger_if_greater) {
                 // TODO: Might be worthwile to set a slider for the trigger threshold. It is
                 // currently always set to 0.5 in configure_input_player.cpp ZL/ZR HandleClick
@@ -61,7 +68,6 @@ private:
     float threshold;
     bool trigger_if_greater;
     const GCAdapter::Adapter* gcadapter;
-    const float origin_value;
 };
 
 GCButtonFactory::GCButtonFactory(std::shared_ptr<GCAdapter::Adapter> adapter_)
@@ -73,7 +79,7 @@ std::unique_ptr<Input::ButtonDevice> GCButtonFactory::Create(const Common::Param
     const auto button_id = params.Get("button", 0);
     const auto port = static_cast<u32>(params.Get("port", 0));
 
-    constexpr int PAD_STICK_ID = static_cast<u16>(GCAdapter::PadButton::PAD_STICK);
+    constexpr s32 PAD_STICK_ID = static_cast<s32>(GCAdapter::PadButton::Stick);
 
     // button is not an axis/stick button
     if (button_id != PAD_STICK_ID) {
@@ -106,32 +112,25 @@ Common::ParamPackage GCButtonFactory::GetNextInput() const {
     Common::ParamPackage params;
     GCAdapter::GCPadStatus pad;
     auto& queue = adapter->GetPadQueue();
-    for (std::size_t port = 0; port < queue.size(); ++port) {
-        while (queue[port].Pop(pad)) {
-            // This while loop will break on the earliest detected button
-            params.Set("engine", "gcpad");
-            params.Set("port", static_cast<int>(port));
-            for (const auto& button : GCAdapter::PadButtonArray) {
-                const u16 button_value = static_cast<u16>(button);
-                if (pad.button & button_value) {
-                    params.Set("button", button_value);
-                    break;
-                }
-            }
+    while (queue.Pop(pad)) {
+        // This while loop will break on the earliest detected button
+        params.Set("engine", "gcpad");
+        params.Set("port", static_cast<s32>(pad.port));
+        if (pad.button != GCAdapter::PadButton::Undefined) {
+            params.Set("button", static_cast<u16>(pad.button));
+        }
 
-            // For Axis button implementation
-            if (pad.axis != GCAdapter::PadAxes::Undefined) {
-                params.Set("axis", static_cast<u8>(pad.axis));
-                params.Set("button", static_cast<u16>(GCAdapter::PadButton::PAD_STICK));
-                if (pad.axis_value > 128) {
-                    params.Set("direction", "+");
-                    params.Set("threshold", "0.25");
-                } else {
-                    params.Set("direction", "-");
-                    params.Set("threshold", "-0.25");
-                }
-                break;
+        // For Axis button implementation
+        if (pad.axis != GCAdapter::PadAxes::Undefined) {
+            params.Set("axis", static_cast<u8>(pad.axis));
+            params.Set("button", static_cast<u16>(GCAdapter::PadButton::Stick));
+            params.Set("threshold", "0.25");
+            if (pad.axis_value > 0) {
+                params.Set("direction", "+");
+            } else {
+                params.Set("direction", "-");
             }
+            break;
         }
     }
     return params;
@@ -152,17 +151,14 @@ public:
     explicit GCAnalog(u32 port_, u32 axis_x_, u32 axis_y_, float deadzone_,
                       const GCAdapter::Adapter* adapter, float range_)
         : port(port_), axis_x(axis_x_), axis_y(axis_y_), deadzone(deadzone_), gcadapter(adapter),
-          origin_value_x(static_cast<float>(adapter->GetOriginValue(port_, axis_x_))),
-          origin_value_y(static_cast<float>(adapter->GetOriginValue(port_, axis_y_))),
           range(range_) {}
 
     float GetAxis(u32 axis) const {
         if (gcadapter->DeviceConnected(port)) {
             std::lock_guard lock{mutex};
-            const auto origin_value = axis % 2 == 0 ? origin_value_x : origin_value_y;
             const auto axis_value =
-                static_cast<float>(gcadapter->GetPadState()[port].axes.at(axis));
-            return (axis_value - origin_value) / (100.0f * range);
+                static_cast<float>(gcadapter->GetPadState(port).axis_values.at(axis));
+            return (axis_value) / (100.0f * range);
         }
         return 0.0f;
     }
@@ -215,8 +211,6 @@ private:
     const u32 axis_y;
     const float deadzone;
     const GCAdapter::Adapter* gcadapter;
-    const float origin_value_x;
-    const float origin_value_y;
     const float range;
     mutable std::mutex mutex;
 };
@@ -254,26 +248,44 @@ void GCAnalogFactory::EndConfiguration() {
 
 Common::ParamPackage GCAnalogFactory::GetNextInput() {
     GCAdapter::GCPadStatus pad;
+    Common::ParamPackage params;
     auto& queue = adapter->GetPadQueue();
-    for (std::size_t port = 0; port < queue.size(); ++port) {
-        while (queue[port].Pop(pad)) {
-            if (pad.axis == GCAdapter::PadAxes::Undefined ||
-                std::abs((static_cast<float>(pad.axis_value) - 128.0f) / 128.0f) < 0.1f) {
-                continue;
-            }
-            // An analog device needs two axes, so we need to store the axis for later and wait for
-            // a second input event. The axes also must be from the same joystick.
-            const u8 axis = static_cast<u8>(pad.axis);
-            if (analog_x_axis == -1) {
-                analog_x_axis = axis;
-                controller_number = static_cast<int>(port);
-            } else if (analog_y_axis == -1 && analog_x_axis != axis &&
-                       controller_number == static_cast<int>(port)) {
-                analog_y_axis = axis;
-            }
+    while (queue.Pop(pad)) {
+        if (pad.button != GCAdapter::PadButton::Undefined) {
+            params.Set("engine", "gcpad");
+            params.Set("port", static_cast<s32>(pad.port));
+            params.Set("button", static_cast<u16>(pad.button));
+            return params;
+        }
+        if (pad.axis == GCAdapter::PadAxes::Undefined ||
+            std::abs(static_cast<float>(pad.axis_value) / 128.0f) < 0.1f) {
+            continue;
+        }
+        // An analog device needs two axes, so we need to store the axis for later and wait for
+        // a second input event. The axes also must be from the same joystick.
+        const u8 axis = static_cast<u8>(pad.axis);
+        if (axis == 0 || axis == 1) {
+            analog_x_axis = 0;
+            analog_y_axis = 1;
+            controller_number = static_cast<s32>(pad.port);
+            break;
+        }
+        if (axis == 2 || axis == 3) {
+            analog_x_axis = 2;
+            analog_y_axis = 3;
+            controller_number = static_cast<s32>(pad.port);
+            break;
+        }
+
+        if (analog_x_axis == -1) {
+            analog_x_axis = axis;
+            controller_number = static_cast<s32>(pad.port);
+        } else if (analog_y_axis == -1 && analog_x_axis != axis &&
+                   controller_number == static_cast<s32>(pad.port)) {
+            analog_y_axis = axis;
+            break;
         }
     }
-    Common::ParamPackage params;
     if (analog_x_axis != -1 && analog_y_axis != -1) {
         params.Set("engine", "gcpad");
         params.Set("port", controller_number);

src/tests/common/fibers.cpp

@@ -6,18 +6,40 @@
 #include <cstdlib>
 #include <functional>
 #include <memory>
+#include <mutex>
+#include <stdexcept>
 #include <thread>
 #include <unordered_map>
 #include <vector>
 
 #include <catch2/catch.hpp>
-#include <math.h>
+
 #include "common/common_types.h"
 #include "common/fiber.h"
-#include "common/spin_lock.h"
 
 namespace Common {
 
+class ThreadIds {
+public:
+    void Register(u32 id) {
+        const auto thread_id = std::this_thread::get_id();
+        std::scoped_lock lock{mutex};
+        if (ids.contains(thread_id)) {
+            throw std::logic_error{"Registering the same thread twice"};
+        }
+        ids.emplace(thread_id, id);
+    }
+
+    [[nodiscard]] u32 Get() const {
+        std::scoped_lock lock{mutex};
+        return ids.at(std::this_thread::get_id());
+    }
+
+private:
+    mutable std::mutex mutex;
+    std::unordered_map<std::thread::id, u32> ids;
+};
+
 class TestControl1 {
 public:
     TestControl1() = default;
@@ -26,7 +48,7 @@ public:
 
     void ExecuteThread(u32 id);
 
-    std::unordered_map<std::thread::id, u32> ids;
+    ThreadIds thread_ids;
     std::vector<std::shared_ptr<Common::Fiber>> thread_fibers;
     std::vector<std::shared_ptr<Common::Fiber>> work_fibers;
     std::vector<u32> items;
@@ -39,8 +61,7 @@ static void WorkControl1(void* control) {
 }
 
 void TestControl1::DoWork() {
-    std::thread::id this_id = std::this_thread::get_id();
-    u32 id = ids[this_id];
+    const u32 id = thread_ids.Get();
     u32 value = items[id];
     for (u32 i = 0; i < id; i++) {
         value++;
@@ -50,8 +71,7 @@ void TestControl1::DoWork() {
 }
 
 void TestControl1::ExecuteThread(u32 id) {
-    std::thread::id this_id = std::this_thread::get_id();
-    ids[this_id] = id;
+    thread_ids.Register(id);
     auto thread_fiber = Fiber::ThreadToFiber();
     thread_fibers[id] = thread_fiber;
     work_fibers[id] = std::make_shared<Fiber>(std::function<void(void*)>{WorkControl1}, this);
@@ -98,8 +118,7 @@ public:
             value1 += i;
         }
         Fiber::YieldTo(fiber1, fiber3);
-        std::thread::id this_id = std::this_thread::get_id();
-        u32 id = ids[this_id];
+        const u32 id = thread_ids.Get();
         assert1 = id == 1;
         value2 += 5000;
         Fiber::YieldTo(fiber1, thread_fibers[id]);
@@ -115,8 +134,7 @@ public:
     }
 
     void DoWork3() {
-        std::thread::id this_id = std::this_thread::get_id();
-        u32 id = ids[this_id];
+        const u32 id = thread_ids.Get();
         assert2 = id == 0;
         value1 += 1000;
         Fiber::YieldTo(fiber3, thread_fibers[id]);
@@ -125,14 +143,12 @@ public:
     void ExecuteThread(u32 id);
 
     void CallFiber1() {
-        std::thread::id this_id = std::this_thread::get_id();
-        u32 id = ids[this_id];
+        const u32 id = thread_ids.Get();
         Fiber::YieldTo(thread_fibers[id], fiber1);
     }
 
     void CallFiber2() {
-        std::thread::id this_id = std::this_thread::get_id();
-        u32 id = ids[this_id];
+        const u32 id = thread_ids.Get();
         Fiber::YieldTo(thread_fibers[id], fiber2);
     }
 
@@ -145,7 +161,7 @@ public:
     u32 value2{};
     std::atomic<bool> trap{true};
     std::atomic<bool> trap2{true};
-    std::unordered_map<std::thread::id, u32> ids;
+    ThreadIds thread_ids;
     std::vector<std::shared_ptr<Common::Fiber>> thread_fibers;
     std::shared_ptr<Common::Fiber> fiber1;
     std::shared_ptr<Common::Fiber> fiber2;
@@ -168,15 +184,13 @@ static void WorkControl2_3(void* control) {
 }
 
 void TestControl2::ExecuteThread(u32 id) {
-    std::thread::id this_id = std::this_thread::get_id();
-    ids[this_id] = id;
+    thread_ids.Register(id);
     auto thread_fiber = Fiber::ThreadToFiber();
     thread_fibers[id] = thread_fiber;
 }
 
 void TestControl2::Exit() {
-    std::thread::id this_id = std::this_thread::get_id();
-    u32 id = ids[this_id];
+    const u32 id = thread_ids.Get();
     thread_fibers[id]->Exit();
 }
 
@@ -228,24 +242,21 @@ public:
     void DoWork1() {
         value1 += 1;
         Fiber::YieldTo(fiber1, fiber2);
-        std::thread::id this_id = std::this_thread::get_id();
-        u32 id = ids[this_id];
+        const u32 id = thread_ids.Get();
         value3 += 1;
         Fiber::YieldTo(fiber1, thread_fibers[id]);
     }
 
     void DoWork2() {
         value2 += 1;
-        std::thread::id this_id = std::this_thread::get_id();
-        u32 id = ids[this_id];
+        const u32 id = thread_ids.Get();
         Fiber::YieldTo(fiber2, thread_fibers[id]);
     }
 
     void ExecuteThread(u32 id);
 
     void CallFiber1() {
-        std::thread::id this_id = std::this_thread::get_id();
-        u32 id = ids[this_id];
+        const u32 id = thread_ids.Get();
         Fiber::YieldTo(thread_fibers[id], fiber1);
     }
 
@@ -254,7 +265,7 @@ public:
     u32 value1{};
     u32 value2{};
     u32 value3{};
-    std::unordered_map<std::thread::id, u32> ids;
+    ThreadIds thread_ids;
     std::vector<std::shared_ptr<Common::Fiber>> thread_fibers;
     std::shared_ptr<Common::Fiber> fiber1;
     std::shared_ptr<Common::Fiber> fiber2;
@@ -271,15 +282,13 @@ static void WorkControl3_2(void* control) {
 }
 
 void TestControl3::ExecuteThread(u32 id) {
-    std::thread::id this_id = std::this_thread::get_id();
-    ids[this_id] = id;
+    thread_ids.Register(id);
     auto thread_fiber = Fiber::ThreadToFiber();
     thread_fibers[id] = thread_fiber;
 }
 
 void TestControl3::Exit() {
-    std::thread::id this_id = std::this_thread::get_id();
-    u32 id = ids[this_id];
+    const u32 id = thread_ids.Get();
     thread_fibers[id]->Exit();
 }
 

src/video_core/CMakeLists.txt

@@ -5,6 +5,24 @@ add_library(video_core STATIC
     buffer_cache/buffer_cache.h
     buffer_cache/map_interval.cpp
     buffer_cache/map_interval.h
+    cdma_pusher.cpp
+    cdma_pusher.h
+    command_classes/codecs/codec.cpp
+    command_classes/codecs/codec.h
+    command_classes/codecs/h264.cpp
+    command_classes/codecs/h264.h
+    command_classes/codecs/vp9.cpp
+    command_classes/codecs/vp9.h
+    command_classes/codecs/vp9_types.h
+    command_classes/host1x.cpp
+    command_classes/host1x.h
+    command_classes/nvdec.cpp
+    command_classes/nvdec.h
+    command_classes/nvdec_common.h
+    command_classes/sync_manager.cpp
+    command_classes/sync_manager.h
+    command_classes/vic.cpp
+    command_classes/vic.h
     compatible_formats.cpp
     compatible_formats.h
     dirty_flags.cpp
@@ -250,6 +268,14 @@ create_target_directory_groups(video_core)
 target_link_libraries(video_core PUBLIC common core)
 target_link_libraries(video_core PRIVATE glad xbyak)
 
+if (MSVC)
+    target_include_directories(video_core PRIVATE ${FFMPEG_INCLUDE_DIR})
+    target_link_libraries(video_core PUBLIC ${FFMPEG_LIBRARY_DIR}/swscale.lib ${FFMPEG_LIBRARY_DIR}/avcodec.lib ${FFMPEG_LIBRARY_DIR}/avutil.lib)
+else()
+    target_include_directories(video_core PRIVATE ${FFMPEG_INCLUDE_DIR})
+    target_link_libraries(video_core PRIVATE ${FFMPEG_LIBRARIES})
+endif()
+
 add_dependencies(video_core host_shaders)
 target_include_directories(video_core PRIVATE ${HOST_SHADERS_INCLUDE})
 
@@ -276,7 +302,10 @@ else()
     target_compile_options(video_core PRIVATE
         -Werror=conversion
         -Wno-error=sign-conversion
+        -Werror=pessimizing-move
+        -Werror=redundant-move
         -Werror=switch
+        -Werror=type-limits
         -Werror=unused-variable
 
         $<$<CXX_COMPILER_ID:GNU>:-Werror=class-memaccess>

src/video_core/cdma_pusher.cpp

@@ -0,0 +1,171 @@
+// MIT License
+//
+// Copyright (c) Ryujinx Team and Contributors
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and
+// associated documentation files (the "Software"), to deal in the Software without restriction,
+// including without limitation the rights to use, copy, modify, merge, publish, distribute,
+// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all copies or
+// substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT
+// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
+// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+
+#include "command_classes/host1x.h"
+#include "command_classes/nvdec.h"
+#include "command_classes/vic.h"
+#include "common/bit_util.h"
+#include "video_core/cdma_pusher.h"
+#include "video_core/command_classes/nvdec_common.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
+
+namespace Tegra {
+CDmaPusher::CDmaPusher(GPU& gpu)
+    : gpu(gpu), nvdec_processor(std::make_shared<Nvdec>(gpu)),
+      vic_processor(std::make_unique<Vic>(gpu, nvdec_processor)),
+      host1x_processor(std::make_unique<Host1x>(gpu)),
+      nvdec_sync(std::make_unique<SyncptIncrManager>(gpu)),
+      vic_sync(std::make_unique<SyncptIncrManager>(gpu)) {}
+
+CDmaPusher::~CDmaPusher() = default;
+
+void CDmaPusher::Push(ChCommandHeaderList&& entries) {
+    cdma_queue.push(std::move(entries));
+}
+
+void CDmaPusher::DispatchCalls() {
+    while (!cdma_queue.empty()) {
+        Step();
+    }
+}
+
+void CDmaPusher::Step() {
+    const auto entries{cdma_queue.front()};
+    cdma_queue.pop();
+
+    std::vector<u32> values(entries.size());
+    std::memcpy(values.data(), entries.data(), entries.size() * sizeof(u32));
+
+    for (const u32 value : values) {
+        if (mask != 0) {
+            const u32 lbs = Common::CountTrailingZeroes32(mask);
+            mask &= ~(1U << lbs);
+            ExecuteCommand(static_cast<u32>(offset + lbs), value);
+            continue;
+        } else if (count != 0) {
+            --count;
+            ExecuteCommand(static_cast<u32>(offset), value);
+            if (incrementing) {
+                ++offset;
+            }
+            continue;
+        }
+        const auto mode = static_cast<ChSubmissionMode>((value >> 28) & 0xf);
+        switch (mode) {
+        case ChSubmissionMode::SetClass: {
+            mask = value & 0x3f;
+            offset = (value >> 16) & 0xfff;
+            current_class = static_cast<ChClassId>((value >> 6) & 0x3ff);
+            break;
+        }
+        case ChSubmissionMode::Incrementing:
+        case ChSubmissionMode::NonIncrementing:
+            count = value & 0xffff;
+            offset = (value >> 16) & 0xfff;
+            incrementing = mode == ChSubmissionMode::Incrementing;
+            break;
+        case ChSubmissionMode::Mask:
+            mask = value & 0xffff;
+            offset = (value >> 16) & 0xfff;
+            break;
+        case ChSubmissionMode::Immediate: {
+            const u32 data = value & 0xfff;
+            offset = (value >> 16) & 0xfff;
+            ExecuteCommand(static_cast<u32>(offset), data);
+            break;
+        }
+        default:
+            UNIMPLEMENTED_MSG("ChSubmission mode {} is not implemented!", static_cast<u32>(mode));
+            break;
+        }
+    }
+}
+
+void CDmaPusher::ExecuteCommand(u32 offset, u32 data) {
+    switch (current_class) {
+    case ChClassId::NvDec:
+        ThiStateWrite(nvdec_thi_state, offset, {data});
+        switch (static_cast<ThiMethod>(offset)) {
+        case ThiMethod::IncSyncpt: {
+            LOG_DEBUG(Service_NVDRV, "NVDEC Class IncSyncpt Method");
+            const auto syncpoint_id = static_cast<u32>(data & 0xFF);
+            const auto cond = static_cast<u32>((data >> 8) & 0xFF);
+            if (cond == 0) {
+                nvdec_sync->Increment(syncpoint_id);
+            } else {
+                nvdec_sync->IncrementWhenDone(static_cast<u32>(current_class), syncpoint_id);
+                nvdec_sync->SignalDone(syncpoint_id);
+            }
+            break;
+        }
+        case ThiMethod::SetMethod1:
+            LOG_DEBUG(Service_NVDRV, "NVDEC method 0x{:X}",
+                      static_cast<u32>(nvdec_thi_state.method_0));
+            nvdec_processor->ProcessMethod(
+                static_cast<Tegra::Nvdec::Method>(nvdec_thi_state.method_0), {data});
+            break;
+        default:
+            break;
+        }
+        break;
+    case ChClassId::GraphicsVic:
+        ThiStateWrite(vic_thi_state, static_cast<u32>(offset), {data});
+        switch (static_cast<ThiMethod>(offset)) {
+        case ThiMethod::IncSyncpt: {
+            LOG_DEBUG(Service_NVDRV, "VIC Class IncSyncpt Method");
+            const auto syncpoint_id = static_cast<u32>(data & 0xFF);
+            const auto cond = static_cast<u32>((data >> 8) & 0xFF);
+            if (cond == 0) {
+                vic_sync->Increment(syncpoint_id);
+            } else {
+                vic_sync->IncrementWhenDone(static_cast<u32>(current_class), syncpoint_id);
+                vic_sync->SignalDone(syncpoint_id);
+            }
+            break;
+        }
+        case ThiMethod::SetMethod1:
+            LOG_DEBUG(Service_NVDRV, "VIC method 0x{:X}, Args=({})",
+                      static_cast<u32>(vic_thi_state.method_0), data);
+            vic_processor->ProcessMethod(static_cast<Tegra::Vic::Method>(vic_thi_state.method_0),
+                                         {data});
+            break;
+        default:
+            break;
+        }
+        break;
+    case ChClassId::Host1x:
+        // This device is mainly for syncpoint synchronization
+        LOG_DEBUG(Service_NVDRV, "Host1X Class Method");
+        host1x_processor->ProcessMethod(static_cast<Tegra::Host1x::Method>(offset), {data});
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Current class not implemented {:X}", static_cast<u32>(current_class));
+        break;
+    }
+}
+
+void CDmaPusher::ThiStateWrite(ThiRegisters& state, u32 offset, const std::vector<u32>& arguments) {
+    u8* const state_offset = reinterpret_cast<u8*>(&state) + sizeof(u32) * offset;
+    std::memcpy(state_offset, arguments.data(), sizeof(u32) * arguments.size());
+}
+
+} // namespace Tegra

src/video_core/cdma_pusher.h

@@ -0,0 +1,138 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <unordered_map>
+#include <vector>
+#include <queue>
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "video_core/command_classes/sync_manager.h"
+
+namespace Tegra {
+
+class GPU;
+class Nvdec;
+class Vic;
+class Host1x;
+
+enum class ChSubmissionMode : u32 {
+    SetClass = 0,
+    Incrementing = 1,
+    NonIncrementing = 2,
+    Mask = 3,
+    Immediate = 4,
+    Restart = 5,
+    Gather = 6,
+};
+
+enum class ChClassId : u32 {
+    NoClass = 0x0,
+    Host1x = 0x1,
+    VideoEncodeMpeg = 0x20,
+    VideoEncodeNvEnc = 0x21,
+    VideoStreamingVi = 0x30,
+    VideoStreamingIsp = 0x32,
+    VideoStreamingIspB = 0x34,
+    VideoStreamingViI2c = 0x36,
+    GraphicsVic = 0x5d,
+    Graphics3D = 0x60,
+    GraphicsGpu = 0x61,
+    Tsec = 0xe0,
+    TsecB = 0xe1,
+    NvJpg = 0xc0,
+    NvDec = 0xf0
+};
+
+enum class ChMethod : u32 {
+    Empty = 0,
+    SetMethod = 0x10,
+    SetData = 0x11,
+};
+
+union ChCommandHeader {
+    u32 raw;
+    BitField<0, 16, u32> value;
+    BitField<16, 12, ChMethod> method_offset;
+    BitField<28, 4, ChSubmissionMode> submission_mode;
+};
+static_assert(sizeof(ChCommandHeader) == sizeof(u32), "ChCommand header is an invalid size");
+
+struct ChCommand {
+    ChClassId class_id{};
+    int method_offset{};
+    std::vector<u32> arguments;
+};
+
+using ChCommandHeaderList = std::vector<Tegra::ChCommandHeader>;
+using ChCommandList = std::vector<Tegra::ChCommand>;
+
+struct ThiRegisters {
+    u32_le increment_syncpt{};
+    INSERT_PADDING_WORDS(1);
+    u32_le increment_syncpt_error{};
+    u32_le ctx_switch_incremement_syncpt{};
+    INSERT_PADDING_WORDS(4);
+    u32_le ctx_switch{};
+    INSERT_PADDING_WORDS(1);
+    u32_le ctx_syncpt_eof{};
+    INSERT_PADDING_WORDS(5);
+    u32_le method_0{};
+    u32_le method_1{};
+    INSERT_PADDING_WORDS(12);
+    u32_le int_status{};
+    u32_le int_mask{};
+};
+
+enum class ThiMethod : u32 {
+    IncSyncpt = offsetof(ThiRegisters, increment_syncpt) / sizeof(u32),
+    SetMethod0 = offsetof(ThiRegisters, method_0) / sizeof(u32),
+    SetMethod1 = offsetof(ThiRegisters, method_1) / sizeof(u32),
+};
+
+class CDmaPusher {
+public:
+    explicit CDmaPusher(GPU& gpu);
+    ~CDmaPusher();
+
+    /// Push NVDEC command buffer entries into queue
+    void Push(ChCommandHeaderList&& entries);
+
+    /// Process queued command buffer entries
+    void DispatchCalls();
+
+    /// Process one queue element
+    void Step();
+
+    /// Invoke command class devices to execute the command based on the current state
+    void ExecuteCommand(u32 offset, u32 data);
+
+private:
+    /// Write arguments value to the ThiRegisters member at the specified offset
+    void ThiStateWrite(ThiRegisters& state, u32 offset, const std::vector<u32>& arguments);
+
+    GPU& gpu;
+
+    std::shared_ptr<Tegra::Nvdec> nvdec_processor;
+    std::unique_ptr<Tegra::Vic> vic_processor;
+    std::unique_ptr<Tegra::Host1x> host1x_processor;
+    std::unique_ptr<SyncptIncrManager> nvdec_sync;
+    std::unique_ptr<SyncptIncrManager> vic_sync;
+    ChClassId current_class{};
+    ThiRegisters vic_thi_state{};
+    ThiRegisters nvdec_thi_state{};
+
+    s32 count{};
+    s32 offset{};
+    s32 mask{};
+    bool incrementing{};
+
+    // Queue of command lists to be processed
+    std::queue<ChCommandHeaderList> cdma_queue;
+};
+
+} // namespace Tegra

src/video_core/command_classes/codecs/codec.cpp

@@ -0,0 +1,115 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cstring>
+#include <fstream>
+#include <vector>
+#include "common/assert.h"
+#include "video_core/command_classes/codecs/codec.h"
+#include "video_core/command_classes/codecs/h264.h"
+#include "video_core/command_classes/codecs/vp9.h"
+#include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
+
+extern "C" {
+#include <libavutil/opt.h>
+}
+
+namespace Tegra {
+
+Codec::Codec(GPU& gpu_)
+    : gpu(gpu_), h264_decoder(std::make_unique<Decoder::H264>(gpu)),
+      vp9_decoder(std::make_unique<Decoder::VP9>(gpu)) {}
+
+Codec::~Codec() {
+    if (!initialized) {
+        return;
+    }
+    // Free libav memory
+    avcodec_send_packet(av_codec_ctx, nullptr);
+    avcodec_receive_frame(av_codec_ctx, av_frame);
+    avcodec_flush_buffers(av_codec_ctx);
+
+    av_frame_unref(av_frame);
+    av_free(av_frame);
+    avcodec_close(av_codec_ctx);
+}
+
+void Codec::SetTargetCodec(NvdecCommon::VideoCodec codec) {
+    LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", static_cast<u32>(codec));
+    current_codec = codec;
+}
+
+void Codec::StateWrite(u32 offset, u64 arguments) {
+    u8* const state_offset = reinterpret_cast<u8*>(&state) + offset * sizeof(u64);
+    std::memcpy(state_offset, &arguments, sizeof(u64));
+}
+
+void Codec::Decode() {
+    bool is_first_frame = false;
+
+    if (!initialized) {
+        if (current_codec == NvdecCommon::VideoCodec::H264) {
+            av_codec = avcodec_find_decoder(AV_CODEC_ID_H264);
+        } else if (current_codec == NvdecCommon::VideoCodec::Vp9) {
+            av_codec = avcodec_find_decoder(AV_CODEC_ID_VP9);
+        } else {
+            LOG_ERROR(Service_NVDRV, "Unknown video codec {}", static_cast<u32>(current_codec));
+            return;
+        }
+
+        av_codec_ctx = avcodec_alloc_context3(av_codec);
+        av_frame = av_frame_alloc();
+        av_opt_set(av_codec_ctx->priv_data, "tune", "zerolatency", 0);
+
+        // TODO(ameerj): libavcodec gpu hw acceleration
+
+        const auto av_error = avcodec_open2(av_codec_ctx, av_codec, nullptr);
+        if (av_error < 0) {
+            LOG_ERROR(Service_NVDRV, "avcodec_open2() Failed.");
+            av_frame_unref(av_frame);
+            av_free(av_frame);
+            avcodec_close(av_codec_ctx);
+            return;
+        }
+        initialized = true;
+        is_first_frame = true;
+    }
+    bool vp9_hidden_frame = false;
+
+    AVPacket packet{};
+    av_init_packet(&packet);
+    std::vector<u8> frame_data;
+
+    if (current_codec == NvdecCommon::VideoCodec::H264) {
+        frame_data = h264_decoder->ComposeFrameHeader(state, is_first_frame);
+    } else if (current_codec == NvdecCommon::VideoCodec::Vp9) {
+        frame_data = vp9_decoder->ComposeFrameHeader(state);
+        vp9_hidden_frame = vp9_decoder->WasFrameHidden();
+    }
+
+    packet.data = frame_data.data();
+    packet.size = static_cast<int>(frame_data.size());
+
+    avcodec_send_packet(av_codec_ctx, &packet);
+
+    if (!vp9_hidden_frame) {
+        // Only receive/store visible frames
+        avcodec_receive_frame(av_codec_ctx, av_frame);
+    }
+}
+
+AVFrame* Codec::GetCurrentFrame() {
+    return av_frame;
+}
+
+const AVFrame* Codec::GetCurrentFrame() const {
+    return av_frame;
+}
+
+NvdecCommon::VideoCodec Codec::GetCurrentCodec() const {
+    return current_codec;
+}
+
+} // namespace Tegra

src/video_core/command_classes/codecs/codec.h

@@ -0,0 +1,66 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include "common/common_types.h"
+#include "video_core/command_classes/nvdec_common.h"
+
+extern "C" {
+#if defined(__GNUC__) || defined(__clang__)
+#pragma GCC diagnostic ignored "-Wconversion"
+#endif
+#include <libavcodec/avcodec.h>
+#if defined(__GNUC__) || defined(__clang__)
+#pragma GCC diagnostic pop
+#endif
+}
+
+namespace Tegra {
+class GPU;
+struct VicRegisters;
+
+namespace Decoder {
+class H264;
+class VP9;
+} // namespace Decoder
+
+class Codec {
+public:
+    explicit Codec(GPU& gpu);
+    ~Codec();
+
+    /// Sets NVDEC video stream codec
+    void SetTargetCodec(NvdecCommon::VideoCodec codec);
+
+    /// Populate NvdecRegisters state with argument value at the provided offset
+    void StateWrite(u32 offset, u64 arguments);
+
+    /// Call decoders to construct headers, decode AVFrame with ffmpeg
+    void Decode();
+
+    /// Returns most recently decoded frame
+    [[nodiscard]] AVFrame* GetCurrentFrame();
+    [[nodiscard]] const AVFrame* GetCurrentFrame() const;
+
+    /// Returns the value of current_codec
+    [[nodiscard]] NvdecCommon::VideoCodec GetCurrentCodec() const;
+
+private:
+    bool initialized{};
+    NvdecCommon::VideoCodec current_codec{NvdecCommon::VideoCodec::None};
+
+    AVCodec* av_codec{nullptr};
+    AVCodecContext* av_codec_ctx{nullptr};
+    AVFrame* av_frame{nullptr};
+
+    GPU& gpu;
+    std::unique_ptr<Decoder::H264> h264_decoder;
+    std::unique_ptr<Decoder::VP9> vp9_decoder;
+
+    NvdecCommon::NvdecRegisters state{};
+};
+
+} // namespace Tegra

src/video_core/command_classes/codecs/h264.cpp

@@ -0,0 +1,293 @@
+// MIT License
+//
+// Copyright (c) Ryujinx Team and Contributors
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and
+// associated documentation files (the "Software"), to deal in the Software without restriction,
+// including without limitation the rights to use, copy, modify, merge, publish, distribute,
+// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all copies or
+// substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT
+// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
+// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+
+#include <array>
+#include "common/bit_util.h"
+#include "video_core/command_classes/codecs/h264.h"
+#include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
+
+namespace Tegra::Decoder {
+namespace {
+// ZigZag LUTs from libavcodec.
+constexpr std::array<u8, 64> zig_zag_direct{
+    0,  1,  8,  16, 9,  2,  3,  10, 17, 24, 32, 25, 18, 11, 4,  5,  12, 19, 26, 33, 40, 48,
+    41, 34, 27, 20, 13, 6,  7,  14, 21, 28, 35, 42, 49, 56, 57, 50, 43, 36, 29, 22, 15, 23,
+    30, 37, 44, 51, 58, 59, 52, 45, 38, 31, 39, 46, 53, 60, 61, 54, 47, 55, 62, 63,
+};
+
+constexpr std::array<u8, 16> zig_zag_scan{
+    0 + 0 * 4, 1 + 0 * 4, 0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 2 + 0 * 4, 3 + 0 * 4, 2 + 1 * 4,
+    1 + 2 * 4, 0 + 3 * 4, 1 + 3 * 4, 2 + 2 * 4, 3 + 1 * 4, 3 + 2 * 4, 2 + 3 * 4, 3 + 3 * 4,
+};
+} // Anonymous namespace
+
+H264::H264(GPU& gpu_) : gpu(gpu_) {}
+
+H264::~H264() = default;
+
+const std::vector<u8>& H264::ComposeFrameHeader(NvdecCommon::NvdecRegisters& state,
+                                                bool is_first_frame) {
+    H264DecoderContext context{};
+    gpu.MemoryManager().ReadBlock(state.picture_info_offset, &context, sizeof(H264DecoderContext));
+
+    const s32 frame_number = static_cast<s32>((context.h264_parameter_set.flags >> 46) & 0x1ffff);
+    if (!is_first_frame && frame_number != 0) {
+        frame.resize(context.frame_data_size);
+
+        gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size());
+    } else {
+        /// Encode header
+        H264BitWriter writer{};
+        writer.WriteU(1, 24);
+        writer.WriteU(0, 1);
+        writer.WriteU(3, 2);
+        writer.WriteU(7, 5);
+        writer.WriteU(100, 8);
+        writer.WriteU(0, 8);
+        writer.WriteU(31, 8);
+        writer.WriteUe(0);
+        const auto chroma_format_idc =
+            static_cast<u32>((context.h264_parameter_set.flags >> 12) & 3);
+        writer.WriteUe(chroma_format_idc);
+        if (chroma_format_idc == 3) {
+            writer.WriteBit(false);
+        }
+
+        writer.WriteUe(0);
+        writer.WriteUe(0);
+        writer.WriteBit(false); // QpprimeYZeroTransformBypassFlag
+        writer.WriteBit(false); // Scaling matrix present flag
+
+        const auto order_cnt_type = static_cast<u32>((context.h264_parameter_set.flags >> 14) & 3);
+        writer.WriteUe(static_cast<u32>((context.h264_parameter_set.flags >> 8) & 0xf));
+        writer.WriteUe(order_cnt_type);
+        if (order_cnt_type == 0) {
+            writer.WriteUe(context.h264_parameter_set.log2_max_pic_order_cnt);
+        } else if (order_cnt_type == 1) {
+            writer.WriteBit(context.h264_parameter_set.delta_pic_order_always_zero_flag != 0);
+
+            writer.WriteSe(0);
+            writer.WriteSe(0);
+            writer.WriteUe(0);
+        }
+
+        const s32 pic_height = context.h264_parameter_set.pic_height_in_map_units /
+                               (context.h264_parameter_set.frame_mbs_only_flag ? 1 : 2);
+
+        writer.WriteUe(16);
+        writer.WriteBit(false);
+        writer.WriteUe(context.h264_parameter_set.pic_width_in_mbs - 1);
+        writer.WriteUe(pic_height - 1);
+        writer.WriteBit(context.h264_parameter_set.frame_mbs_only_flag != 0);
+
+        if (!context.h264_parameter_set.frame_mbs_only_flag) {
+            writer.WriteBit(((context.h264_parameter_set.flags >> 0) & 1) != 0);
+        }
+
+        writer.WriteBit(((context.h264_parameter_set.flags >> 1) & 1) != 0);
+        writer.WriteBit(false); // Frame cropping flag
+        writer.WriteBit(false); // VUI parameter present flag
+
+        writer.End();
+
+        // H264 PPS
+        writer.WriteU(1, 24);
+        writer.WriteU(0, 1);
+        writer.WriteU(3, 2);
+        writer.WriteU(8, 5);
+
+        writer.WriteUe(0);
+        writer.WriteUe(0);
+
+        writer.WriteBit(context.h264_parameter_set.entropy_coding_mode_flag != 0);
+        writer.WriteBit(false);
+        writer.WriteUe(0);
+        writer.WriteUe(context.h264_parameter_set.num_refidx_l0_default_active);
+        writer.WriteUe(context.h264_parameter_set.num_refidx_l1_default_active);
+        writer.WriteBit(((context.h264_parameter_set.flags >> 2) & 1) != 0);
+        writer.WriteU(static_cast<s32>((context.h264_parameter_set.flags >> 32) & 0x3), 2);
+        s32 pic_init_qp = static_cast<s32>((context.h264_parameter_set.flags >> 16) & 0x3f);
+        pic_init_qp = (pic_init_qp << 26) >> 26;
+        writer.WriteSe(pic_init_qp);
+        writer.WriteSe(0);
+        s32 chroma_qp_index_offset =
+            static_cast<s32>((context.h264_parameter_set.flags >> 22) & 0x1f);
+        chroma_qp_index_offset = (chroma_qp_index_offset << 27) >> 27;
+
+        writer.WriteSe(chroma_qp_index_offset);
+        writer.WriteBit(context.h264_parameter_set.deblocking_filter_control_flag != 0);
+        writer.WriteBit(((context.h264_parameter_set.flags >> 3) & 1) != 0);
+        writer.WriteBit(context.h264_parameter_set.redundant_pic_count_flag != 0);
+        writer.WriteBit(context.h264_parameter_set.transform_8x8_mode_flag != 0);
+
+        writer.WriteBit(true);
+
+        for (s32 index = 0; index < 6; index++) {
+            writer.WriteBit(true);
+            const auto matrix_x4 =
+                std::vector<u8>(context.scaling_matrix_4.begin(), context.scaling_matrix_4.end());
+            writer.WriteScalingList(matrix_x4, index * 16, 16);
+        }
+
+        if (context.h264_parameter_set.transform_8x8_mode_flag) {
+            for (s32 index = 0; index < 2; index++) {
+                writer.WriteBit(true);
+                const auto matrix_x8 = std::vector<u8>(context.scaling_matrix_8.begin(),
+                                                       context.scaling_matrix_8.end());
+
+                writer.WriteScalingList(matrix_x8, index * 64, 64);
+            }
+        }
+
+        s32 chroma_qp_index_offset2 =
+            static_cast<s32>((context.h264_parameter_set.flags >> 27) & 0x1f);
+        chroma_qp_index_offset2 = (chroma_qp_index_offset2 << 27) >> 27;
+
+        writer.WriteSe(chroma_qp_index_offset2);
+
+        writer.End();
+
+        const auto& encoded_header = writer.GetByteArray();
+        frame.resize(encoded_header.size() + context.frame_data_size);
+        std::memcpy(frame.data(), encoded_header.data(), encoded_header.size());
+
+        gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset,
+                                      frame.data() + encoded_header.size(),
+                                      context.frame_data_size);
+    }
+
+    return frame;
+}
+
+H264BitWriter::H264BitWriter() = default;
+
+H264BitWriter::~H264BitWriter() = default;
+
+void H264BitWriter::WriteU(s32 value, s32 value_sz) {
+    WriteBits(value, value_sz);
+}
+
+void H264BitWriter::WriteSe(s32 value) {
+    WriteExpGolombCodedInt(value);
+}
+
+void H264BitWriter::WriteUe(u32 value) {
+    WriteExpGolombCodedUInt(value);
+}
+
+void H264BitWriter::End() {
+    WriteBit(true);
+    Flush();
+}
+
+void H264BitWriter::WriteBit(bool state) {
+    WriteBits(state ? 1 : 0, 1);
+}
+
+void H264BitWriter::WriteScalingList(const std::vector<u8>& list, s32 start, s32 count) {
+    std::vector<u8> scan(count);
+    if (count == 16) {
+        std::memcpy(scan.data(), zig_zag_scan.data(), scan.size());
+    } else {
+        std::memcpy(scan.data(), zig_zag_direct.data(), scan.size());
+    }
+    u8 last_scale = 8;
+
+    for (s32 index = 0; index < count; index++) {
+        const u8 value = list[start + scan[index]];
+        const s32 delta_scale = static_cast<s32>(value - last_scale);
+
+        WriteSe(delta_scale);
+
+        last_scale = value;
+    }
+}
+
+std::vector<u8>& H264BitWriter::GetByteArray() {
+    return byte_array;
+}
+
+const std::vector<u8>& H264BitWriter::GetByteArray() const {
+    return byte_array;
+}
+
+void H264BitWriter::WriteBits(s32 value, s32 bit_count) {
+    s32 value_pos = 0;
+
+    s32 remaining = bit_count;
+
+    while (remaining > 0) {
+        s32 copy_size = remaining;
+
+        const s32 free_bits = GetFreeBufferBits();
+
+        if (copy_size > free_bits) {
+            copy_size = free_bits;
+        }
+
+        const s32 mask = (1 << copy_size) - 1;
+
+        const s32 src_shift = (bit_count - value_pos) - copy_size;
+        const s32 dst_shift = (buffer_size - buffer_pos) - copy_size;
+
+        buffer |= ((value >> src_shift) & mask) << dst_shift;
+
+        value_pos += copy_size;
+        buffer_pos += copy_size;
+        remaining -= copy_size;
+    }
+}
+
+void H264BitWriter::WriteExpGolombCodedInt(s32 value) {
+    const s32 sign = value <= 0 ? 0 : 1;
+    if (value < 0) {
+        value = -value;
+    }
+    value = (value << 1) - sign;
+    WriteExpGolombCodedUInt(value);
+}
+
+void H264BitWriter::WriteExpGolombCodedUInt(u32 value) {
+    const s32 size = 32 - Common::CountLeadingZeroes32(static_cast<s32>(value + 1));
+    WriteBits(1, size);
+
+    value -= (1U << (size - 1)) - 1;
+    WriteBits(static_cast<s32>(value), size - 1);
+}
+
+s32 H264BitWriter::GetFreeBufferBits() {
+    if (buffer_pos == buffer_size) {
+        Flush();
+    }
+
+    return buffer_size - buffer_pos;
+}
+
+void H264BitWriter::Flush() {
+    if (buffer_pos == 0) {
+        return;
+    }
+    byte_array.push_back(static_cast<u8>(buffer));
+
+    buffer = 0;
+    buffer_pos = 0;
+}
+} // namespace Tegra::Decoder

src/video_core/command_classes/codecs/h264.h

@@ -0,0 +1,118 @@
+// MIT License
+//
+// Copyright (c) Ryujinx Team and Contributors
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and
+// associated documentation files (the "Software"), to deal in the Software without restriction,
+// including without limitation the rights to use, copy, modify, merge, publish, distribute,
+// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all copies or
+// substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT
+// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
+// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+
+#pragma once
+
+#include <vector>
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+#include "video_core/command_classes/nvdec_common.h"
+
+namespace Tegra {
+class GPU;
+namespace Decoder {
+
+class H264BitWriter {
+public:
+    H264BitWriter();
+    ~H264BitWriter();
+
+    /// The following Write methods are based on clause 9.1 in the H.264 specification.
+    /// WriteSe and WriteUe write in the Exp-Golomb-coded syntax
+    void WriteU(s32 value, s32 value_sz);
+    void WriteSe(s32 value);
+    void WriteUe(u32 value);
+
+    /// Finalize the bitstream
+    void End();
+
+    /// append a bit to the stream, equivalent value to the state parameter
+    void WriteBit(bool state);
+
+    /// Based on section 7.3.2.1.1.1 and Table 7-4 in the H.264 specification
+    /// Writes the scaling matrices of the sream
+    void WriteScalingList(const std::vector<u8>& list, s32 start, s32 count);
+
+    /// Return the bitstream as a vector.
+    [[nodiscard]] std::vector<u8>& GetByteArray();
+    [[nodiscard]] const std::vector<u8>& GetByteArray() const;
+
+private:
+    void WriteBits(s32 value, s32 bit_count);
+    void WriteExpGolombCodedInt(s32 value);
+    void WriteExpGolombCodedUInt(u32 value);
+    [[nodiscard]] s32 GetFreeBufferBits();
+    void Flush();
+
+    s32 buffer_size{8};
+
+    s32 buffer{};
+    s32 buffer_pos{};
+    std::vector<u8> byte_array;
+};
+
+class H264 {
+public:
+    explicit H264(GPU& gpu);
+    ~H264();
+
+    /// Compose the H264 header of the frame for FFmpeg decoding
+    [[nodiscard]] const std::vector<u8>& ComposeFrameHeader(NvdecCommon::NvdecRegisters& state,
+                                                            bool is_first_frame = false);
+
+private:
+    struct H264ParameterSet {
+        u32 log2_max_pic_order_cnt{};
+        u32 delta_pic_order_always_zero_flag{};
+        u32 frame_mbs_only_flag{};
+        u32 pic_width_in_mbs{};
+        u32 pic_height_in_map_units{};
+        INSERT_PADDING_WORDS(1);
+        u32 entropy_coding_mode_flag{};
+        u32 bottom_field_pic_order_flag{};
+        u32 num_refidx_l0_default_active{};
+        u32 num_refidx_l1_default_active{};
+        u32 deblocking_filter_control_flag{};
+        u32 redundant_pic_count_flag{};
+        u32 transform_8x8_mode_flag{};
+        INSERT_PADDING_WORDS(9);
+        u64 flags{};
+        u32 frame_number{};
+        u32 frame_number2{};
+    };
+    static_assert(sizeof(H264ParameterSet) == 0x68, "H264ParameterSet is an invalid size");
+
+    struct H264DecoderContext {
+        INSERT_PADDING_BYTES(0x48);
+        u32 frame_data_size{};
+        INSERT_PADDING_BYTES(0xc);
+        H264ParameterSet h264_parameter_set{};
+        INSERT_PADDING_BYTES(0x100);
+        std::array<u8, 0x60> scaling_matrix_4;
+        std::array<u8, 0x80> scaling_matrix_8;
+    };
+    static_assert(sizeof(H264DecoderContext) == 0x2a0, "H264DecoderContext is an invalid size");
+
+    std::vector<u8> frame;
+    GPU& gpu;
+};
+
+} // namespace Decoder
+} // namespace Tegra

src/video_core/command_classes/codecs/vp9.h

@@ -0,0 +1,196 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <vector>
+
+#include "common/common_types.h"
+#include "common/stream.h"
+#include "video_core/command_classes/codecs/vp9_types.h"
+#include "video_core/command_classes/nvdec_common.h"
+
+namespace Tegra {
+class GPU;
+enum class FrameType { KeyFrame = 0, InterFrame = 1 };
+namespace Decoder {
+
+/// The VpxRangeEncoder, and VpxBitStreamWriter classes are used to compose the
+/// VP9 header bitstreams.
+
+class VpxRangeEncoder {
+public:
+    VpxRangeEncoder();
+    ~VpxRangeEncoder();
+
+    VpxRangeEncoder(const VpxRangeEncoder&) = delete;
+    VpxRangeEncoder& operator=(const VpxRangeEncoder&) = delete;
+
+    VpxRangeEncoder(VpxRangeEncoder&&) = default;
+    VpxRangeEncoder& operator=(VpxRangeEncoder&&) = default;
+
+    /// Writes the rightmost value_size bits from value into the stream
+    void Write(s32 value, s32 value_size);
+
+    /// Writes a single bit with half probability
+    void Write(bool bit);
+
+    /// Writes a bit to the base_stream encoded with probability
+    void Write(bool bit, s32 probability);
+
+    /// Signal the end of the bitstream
+    void End();
+
+    [[nodiscard]] std::vector<u8>& GetBuffer() {
+        return base_stream.GetBuffer();
+    }
+
+    [[nodiscard]] const std::vector<u8>& GetBuffer() const {
+        return base_stream.GetBuffer();
+    }
+
+private:
+    u8 PeekByte();
+    Common::Stream base_stream{};
+    u32 low_value{};
+    u32 range{0xff};
+    s32 count{-24};
+    s32 half_probability{128};
+};
+
+class VpxBitStreamWriter {
+public:
+    VpxBitStreamWriter();
+    ~VpxBitStreamWriter();
+
+    VpxBitStreamWriter(const VpxBitStreamWriter&) = delete;
+    VpxBitStreamWriter& operator=(const VpxBitStreamWriter&) = delete;
+
+    VpxBitStreamWriter(VpxBitStreamWriter&&) = default;
+    VpxBitStreamWriter& operator=(VpxBitStreamWriter&&) = default;
+
+    /// Write an unsigned integer value
+    void WriteU(u32 value, u32 value_size);
+
+    /// Write a signed integer value
+    void WriteS(s32 value, u32 value_size);
+
+    /// Based on 6.2.10 of VP9 Spec, writes a delta coded value
+    void WriteDeltaQ(u32 value);
+
+    /// Write a single bit.
+    void WriteBit(bool state);
+
+    /// Pushes current buffer into buffer_array, resets buffer
+    void Flush();
+
+    /// Returns byte_array
+    [[nodiscard]] std::vector<u8>& GetByteArray();
+
+    /// Returns const byte_array
+    [[nodiscard]] const std::vector<u8>& GetByteArray() const;
+
+private:
+    /// Write bit_count bits from value into buffer
+    void WriteBits(u32 value, u32 bit_count);
+
+    /// Gets next available position in buffer, invokes Flush() if buffer is full
+    s32 GetFreeBufferBits();
+
+    s32 buffer_size{8};
+
+    s32 buffer{};
+    s32 buffer_pos{};
+    std::vector<u8> byte_array;
+};
+
+class VP9 {
+public:
+    explicit VP9(GPU& gpu);
+    ~VP9();
+
+    VP9(const VP9&) = delete;
+    VP9& operator=(const VP9&) = delete;
+
+    VP9(VP9&&) = default;
+    VP9& operator=(VP9&&) = delete;
+
+    /// Composes the VP9 frame from the GPU state information. Based on the official VP9 spec
+    /// documentation
+    [[nodiscard]] const std::vector<u8>& ComposeFrameHeader(NvdecCommon::NvdecRegisters& state);
+
+    /// Returns true if the most recent frame was a hidden frame.
+    [[nodiscard]] bool WasFrameHidden() const {
+        return hidden;
+    }
+
+private:
+    /// Generates compressed header probability updates in the bitstream writer
+    template <typename T, std::size_t N>
+    void WriteProbabilityUpdate(VpxRangeEncoder& writer, const std::array<T, N>& new_prob,
+                                const std::array<T, N>& old_prob);
+
+    /// Generates compressed header probability updates in the bitstream writer
+    /// If probs are not equal, WriteProbabilityDelta is invoked
+    void WriteProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob);
+
+    /// Generates compressed header probability deltas in the bitstream writer
+    void WriteProbabilityDelta(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob);
+
+    /// Inverse of 6.3.4 Decode term subexp
+    void EncodeTermSubExp(VpxRangeEncoder& writer, s32 value);
+
+    /// Writes if the value is less than the test value
+    bool WriteLessThan(VpxRangeEncoder& writer, s32 value, s32 test);
+
+    /// Writes probability updates for the Coef probabilities
+    void WriteCoefProbabilityUpdate(VpxRangeEncoder& writer, s32 tx_mode,
+                                    const std::array<u8, 2304>& new_prob,
+                                    const std::array<u8, 2304>& old_prob);
+
+    /// Write probabilities for 4-byte aligned structures
+    template <typename T, std::size_t N>
+    void WriteProbabilityUpdateAligned4(VpxRangeEncoder& writer, const std::array<T, N>& new_prob,
+                                        const std::array<T, N>& old_prob);
+
+    /// Write motion vector probability updates. 6.3.17 in the spec
+    void WriteMvProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob);
+
+    /// Returns VP9 information from NVDEC provided offset and size
+    [[nodiscard]] Vp9PictureInfo GetVp9PictureInfo(const NvdecCommon::NvdecRegisters& state);
+
+    /// Read and convert NVDEC provided entropy probs to Vp9EntropyProbs struct
+    void InsertEntropy(u64 offset, Vp9EntropyProbs& dst);
+
+    /// Returns frame to be decoded after buffering
+    [[nodiscard]] Vp9FrameContainer GetCurrentFrame(const NvdecCommon::NvdecRegisters& state);
+
+    /// Use NVDEC providied information to compose the headers for the current frame
+    [[nodiscard]] std::vector<u8> ComposeCompressedHeader();
+    [[nodiscard]] VpxBitStreamWriter ComposeUncompressedHeader();
+
+    GPU& gpu;
+    std::vector<u8> frame;
+
+    std::array<s8, 4> loop_filter_ref_deltas{};
+    std::array<s8, 2> loop_filter_mode_deltas{};
+
+    bool hidden = false;
+    s64 current_frame_number = -2; // since we buffer 2 frames
+    s32 grace_period = 6;          // frame offsets need to stabilize
+    std::array<FrameContexts, 4> frame_ctxs{};
+    Vp9FrameContainer next_frame{};
+    Vp9FrameContainer next_next_frame{};
+    bool swap_next_golden{};
+
+    Vp9PictureInfo current_frame_info{};
+    Vp9EntropyProbs prev_frame_probs{};
+
+    s32 diff_update_probability = 252;
+    s32 frame_sync_code = 0x498342;
+};
+
+} // namespace Decoder
+} // namespace Tegra

src/video_core/command_classes/codecs/vp9_types.h

@@ -0,0 +1,366 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <cstring>
+#include <vector>
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+
+namespace Tegra {
+class GPU;
+
+namespace Decoder {
+struct Vp9FrameDimensions {
+    s16 width{};
+    s16 height{};
+    s16 luma_pitch{};
+    s16 chroma_pitch{};
+};
+static_assert(sizeof(Vp9FrameDimensions) == 0x8, "Vp9 Vp9FrameDimensions is an invalid size");
+
+enum FrameFlags : u32 {
+    IsKeyFrame = 1 << 0,
+    LastFrameIsKeyFrame = 1 << 1,
+    FrameSizeChanged = 1 << 2,
+    ErrorResilientMode = 1 << 3,
+    LastShowFrame = 1 << 4,
+    IntraOnly = 1 << 5,
+};
+
+enum class MvJointType {
+    MvJointZero = 0,   /* Zero vector */
+    MvJointHnzvz = 1,  /* Vert zero, hor nonzero */
+    MvJointHzvnz = 2,  /* Hor zero, vert nonzero */
+    MvJointHnzvnz = 3, /* Both components nonzero */
+};
+enum class MvClassType {
+    MvClass0 = 0,   /* (0, 2]     integer pel */
+    MvClass1 = 1,   /* (2, 4]     integer pel */
+    MvClass2 = 2,   /* (4, 8]     integer pel */
+    MvClass3 = 3,   /* (8, 16]    integer pel */
+    MvClass4 = 4,   /* (16, 32]   integer pel */
+    MvClass5 = 5,   /* (32, 64]   integer pel */
+    MvClass6 = 6,   /* (64, 128]  integer pel */
+    MvClass7 = 7,   /* (128, 256] integer pel */
+    MvClass8 = 8,   /* (256, 512] integer pel */
+    MvClass9 = 9,   /* (512, 1024] integer pel */
+    MvClass10 = 10, /* (1024,2048] integer pel */
+};
+
+enum class BlockSize {
+    Block4x4 = 0,
+    Block4x8 = 1,
+    Block8x4 = 2,
+    Block8x8 = 3,
+    Block8x16 = 4,
+    Block16x8 = 5,
+    Block16x16 = 6,
+    Block16x32 = 7,
+    Block32x16 = 8,
+    Block32x32 = 9,
+    Block32x64 = 10,
+    Block64x32 = 11,
+    Block64x64 = 12,
+    BlockSizes = 13,
+    BlockInvalid = BlockSizes
+};
+
+enum class PredictionMode {
+    DcPred = 0,   // Average of above and left pixels
+    VPred = 1,    // Vertical
+    HPred = 2,    // Horizontal
+    D45Pred = 3,  // Directional 45  deg = round(arctan(1 / 1) * 180 / pi)
+    D135Pred = 4, // Directional 135 deg = 180 - 45
+    D117Pred = 5, // Directional 117 deg = 180 - 63
+    D153Pred = 6, // Directional 153 deg = 180 - 27
+    D207Pred = 7, // Directional 207 deg = 180 + 27
+    D63Pred = 8,  // Directional 63  deg = round(arctan(2 / 1) * 180 / pi)
+    TmPred = 9,   // True-motion
+    NearestMv = 10,
+    NearMv = 11,
+    ZeroMv = 12,
+    NewMv = 13,
+    MbModeCount = 14
+};
+
+enum class TxSize {
+    Tx4x4 = 0,   // 4x4 transform
+    Tx8x8 = 1,   // 8x8 transform
+    Tx16x16 = 2, // 16x16 transform
+    Tx32x32 = 3, // 32x32 transform
+    TxSizes = 4
+};
+
+enum class TxMode {
+    Only4X4 = 0,      // Only 4x4 transform used
+    Allow8X8 = 1,     // Allow block transform size up to 8x8
+    Allow16X16 = 2,   // Allow block transform size up to 16x16
+    Allow32X32 = 3,   // Allow block transform size up to 32x32
+    TxModeSelect = 4, // Transform specified for each block
+    TxModes = 5
+};
+
+enum class reference_mode {
+    SingleReference = 0,
+    CompoundReference = 1,
+    ReferenceModeSelect = 2,
+    ReferenceModes = 3
+};
+
+struct Segmentation {
+    u8 enabled{};
+    u8 update_map{};
+    u8 temporal_update{};
+    u8 abs_delta{};
+    std::array<u32, 8> feature_mask{};
+    std::array<std::array<s16, 4>, 8> feature_data{};
+};
+static_assert(sizeof(Segmentation) == 0x64, "Segmentation is an invalid size");
+
+struct LoopFilter {
+    u8 mode_ref_delta_enabled{};
+    std::array<s8, 4> ref_deltas{};
+    std::array<s8, 2> mode_deltas{};
+};
+static_assert(sizeof(LoopFilter) == 0x7, "LoopFilter is an invalid size");
+
+struct Vp9EntropyProbs {
+    std::array<u8, 36> y_mode_prob{};
+    std::array<u8, 64> partition_prob{};
+    std::array<u8, 2304> coef_probs{};
+    std::array<u8, 8> switchable_interp_prob{};
+    std::array<u8, 28> inter_mode_prob{};
+    std::array<u8, 4> intra_inter_prob{};
+    std::array<u8, 5> comp_inter_prob{};
+    std::array<u8, 10> single_ref_prob{};
+    std::array<u8, 5> comp_ref_prob{};
+    std::array<u8, 6> tx_32x32_prob{};
+    std::array<u8, 4> tx_16x16_prob{};
+    std::array<u8, 2> tx_8x8_prob{};
+    std::array<u8, 3> skip_probs{};
+    std::array<u8, 3> joints{};
+    std::array<u8, 2> sign{};
+    std::array<u8, 20> classes{};
+    std::array<u8, 2> class_0{};
+    std::array<u8, 20> prob_bits{};
+    std::array<u8, 12> class_0_fr{};
+    std::array<u8, 6> fr{};
+    std::array<u8, 2> class_0_hp{};
+    std::array<u8, 2> high_precision{};
+};
+static_assert(sizeof(Vp9EntropyProbs) == 0x9F4, "Vp9EntropyProbs is an invalid size");
+
+struct Vp9PictureInfo {
+    bool is_key_frame{};
+    bool intra_only{};
+    bool last_frame_was_key{};
+    bool frame_size_changed{};
+    bool error_resilient_mode{};
+    bool last_frame_shown{};
+    bool show_frame{};
+    std::array<s8, 4> ref_frame_sign_bias{};
+    s32 base_q_index{};
+    s32 y_dc_delta_q{};
+    s32 uv_dc_delta_q{};
+    s32 uv_ac_delta_q{};
+    bool lossless{};
+    s32 transform_mode{};
+    bool allow_high_precision_mv{};
+    s32 interp_filter{};
+    s32 reference_mode{};
+    s8 comp_fixed_ref{};
+    std::array<s8, 2> comp_var_ref{};
+    s32 log2_tile_cols{};
+    s32 log2_tile_rows{};
+    bool segment_enabled{};
+    bool segment_map_update{};
+    bool segment_map_temporal_update{};
+    s32 segment_abs_delta{};
+    std::array<u32, 8> segment_feature_enable{};
+    std::array<std::array<s16, 4>, 8> segment_feature_data{};
+    bool mode_ref_delta_enabled{};
+    bool use_prev_in_find_mv_refs{};
+    std::array<s8, 4> ref_deltas{};
+    std::array<s8, 2> mode_deltas{};
+    Vp9EntropyProbs entropy{};
+    Vp9FrameDimensions frame_size{};
+    u8 first_level{};
+    u8 sharpness_level{};
+    u32 bitstream_size{};
+    std::array<u64, 4> frame_offsets{};
+    std::array<bool, 4> refresh_frame{};
+};
+
+struct Vp9FrameContainer {
+    Vp9PictureInfo info{};
+    std::vector<u8> bit_stream;
+};
+
+struct PictureInfo {
+    INSERT_PADDING_WORDS(12);
+    u32 bitstream_size{};
+    INSERT_PADDING_WORDS(5);
+    Vp9FrameDimensions last_frame_size{};
+    Vp9FrameDimensions golden_frame_size{};
+    Vp9FrameDimensions alt_frame_size{};
+    Vp9FrameDimensions current_frame_size{};
+    u32 vp9_flags{};
+    std::array<s8, 4> ref_frame_sign_bias{};
+    u8 first_level{};
+    u8 sharpness_level{};
+    u8 base_q_index{};
+    u8 y_dc_delta_q{};
+    u8 uv_ac_delta_q{};
+    u8 uv_dc_delta_q{};
+    u8 lossless{};
+    u8 tx_mode{};
+    u8 allow_high_precision_mv{};
+    u8 interp_filter{};
+    u8 reference_mode{};
+    s8 comp_fixed_ref{};
+    std::array<s8, 2> comp_var_ref{};
+    u8 log2_tile_cols{};
+    u8 log2_tile_rows{};
+    Segmentation segmentation{};
+    LoopFilter loop_filter{};
+    INSERT_PADDING_BYTES(5);
+    u32 surface_params{};
+    INSERT_PADDING_WORDS(3);
+
+    [[nodiscard]] Vp9PictureInfo Convert() const {
+        return {
+            .is_key_frame = (vp9_flags & FrameFlags::IsKeyFrame) != 0,
+            .intra_only = (vp9_flags & FrameFlags::IntraOnly) != 0,
+            .last_frame_was_key = (vp9_flags & FrameFlags::LastFrameIsKeyFrame) != 0,
+            .frame_size_changed = (vp9_flags & FrameFlags::FrameSizeChanged) != 0,
+            .error_resilient_mode = (vp9_flags & FrameFlags::ErrorResilientMode) != 0,
+            .last_frame_shown = (vp9_flags & FrameFlags::LastShowFrame) != 0,
+            .ref_frame_sign_bias = ref_frame_sign_bias,
+            .base_q_index = base_q_index,
+            .y_dc_delta_q = y_dc_delta_q,
+            .uv_dc_delta_q = uv_dc_delta_q,
+            .uv_ac_delta_q = uv_ac_delta_q,
+            .lossless = lossless != 0,
+            .transform_mode = tx_mode,
+            .allow_high_precision_mv = allow_high_precision_mv != 0,
+            .interp_filter = interp_filter,
+            .reference_mode = reference_mode,
+            .comp_fixed_ref = comp_fixed_ref,
+            .comp_var_ref = comp_var_ref,
+            .log2_tile_cols = log2_tile_cols,
+            .log2_tile_rows = log2_tile_rows,
+            .segment_enabled = segmentation.enabled != 0,
+            .segment_map_update = segmentation.update_map != 0,
+            .segment_map_temporal_update = segmentation.temporal_update != 0,
+            .segment_abs_delta = segmentation.abs_delta,
+            .segment_feature_enable = segmentation.feature_mask,
+            .segment_feature_data = segmentation.feature_data,
+            .mode_ref_delta_enabled = loop_filter.mode_ref_delta_enabled != 0,
+            .use_prev_in_find_mv_refs = !(vp9_flags == (FrameFlags::ErrorResilientMode)) &&
+                                        !(vp9_flags == (FrameFlags::FrameSizeChanged)) &&
+                                        !(vp9_flags == (FrameFlags::IntraOnly)) &&
+                                        (vp9_flags == (FrameFlags::LastShowFrame)) &&
+                                        !(vp9_flags == (FrameFlags::LastFrameIsKeyFrame)),
+            .ref_deltas = loop_filter.ref_deltas,
+            .mode_deltas = loop_filter.mode_deltas,
+            .frame_size = current_frame_size,
+            .first_level = first_level,
+            .sharpness_level = sharpness_level,
+            .bitstream_size = bitstream_size,
+        };
+    }
+};
+static_assert(sizeof(PictureInfo) == 0x100, "PictureInfo is an invalid size");
+
+struct EntropyProbs {
+    INSERT_PADDING_BYTES(1024);
+    std::array<std::array<u8, 4>, 7> inter_mode_prob{};
+    std::array<u8, 4> intra_inter_prob{};
+    INSERT_PADDING_BYTES(80);
+    std::array<std::array<u8, 1>, 2> tx_8x8_prob{};
+    std::array<std::array<u8, 2>, 2> tx_16x16_prob{};
+    std::array<std::array<u8, 3>, 2> tx_32x32_prob{};
+    std::array<u8, 4> y_mode_prob_e8{};
+    std::array<std::array<u8, 8>, 4> y_mode_prob_e0e7{};
+    INSERT_PADDING_BYTES(64);
+    std::array<std::array<u8, 4>, 16> partition_prob{};
+    INSERT_PADDING_BYTES(10);
+    std::array<std::array<u8, 2>, 4> switchable_interp_prob{};
+    std::array<u8, 5> comp_inter_prob{};
+    std::array<u8, 4> skip_probs{};
+    std::array<u8, 3> joints{};
+    std::array<u8, 2> sign{};
+    std::array<std::array<u8, 1>, 2> class_0{};
+    std::array<std::array<u8, 3>, 2> fr{};
+    std::array<u8, 2> class_0_hp{};
+    std::array<u8, 2> high_precision{};
+    std::array<std::array<u8, 10>, 2> classes{};
+    std::array<std::array<std::array<u8, 3>, 2>, 2> class_0_fr{};
+    std::array<std::array<u8, 10>, 2> pred_bits{};
+    std::array<std::array<u8, 2>, 5> single_ref_prob{};
+    std::array<u8, 5> comp_ref_prob{};
+    INSERT_PADDING_BYTES(17);
+    std::array<std::array<std::array<std::array<std::array<std::array<u8, 4>, 6>, 6>, 2>, 2>, 4>
+        coef_probs{};
+
+    void Convert(Vp9EntropyProbs& fc) {
+        std::memcpy(fc.inter_mode_prob.data(), inter_mode_prob.data(), fc.inter_mode_prob.size());
+
+        std::memcpy(fc.intra_inter_prob.data(), intra_inter_prob.data(),
+                    fc.intra_inter_prob.size());
+
+        std::memcpy(fc.tx_8x8_prob.data(), tx_8x8_prob.data(), fc.tx_8x8_prob.size());
+        std::memcpy(fc.tx_16x16_prob.data(), tx_16x16_prob.data(), fc.tx_16x16_prob.size());
+        std::memcpy(fc.tx_32x32_prob.data(), tx_32x32_prob.data(), fc.tx_32x32_prob.size());
+
+        for (s32 i = 0; i < 4; i++) {
+            for (s32 j = 0; j < 9; j++) {
+                fc.y_mode_prob[j + 9 * i] = j < 8 ? y_mode_prob_e0e7[i][j] : y_mode_prob_e8[i];
+            }
+        }
+
+        std::memcpy(fc.partition_prob.data(), partition_prob.data(), fc.partition_prob.size());
+
+        std::memcpy(fc.switchable_interp_prob.data(), switchable_interp_prob.data(),
+                    fc.switchable_interp_prob.size());
+        std::memcpy(fc.comp_inter_prob.data(), comp_inter_prob.data(), fc.comp_inter_prob.size());
+        std::memcpy(fc.skip_probs.data(), skip_probs.data(), fc.skip_probs.size());
+
+        std::memcpy(fc.joints.data(), joints.data(), fc.joints.size());
+
+        std::memcpy(fc.sign.data(), sign.data(), fc.sign.size());
+        std::memcpy(fc.class_0.data(), class_0.data(), fc.class_0.size());
+        std::memcpy(fc.fr.data(), fr.data(), fc.fr.size());
+        std::memcpy(fc.class_0_hp.data(), class_0_hp.data(), fc.class_0_hp.size());
+        std::memcpy(fc.high_precision.data(), high_precision.data(), fc.high_precision.size());
+        std::memcpy(fc.classes.data(), classes.data(), fc.classes.size());
+        std::memcpy(fc.class_0_fr.data(), class_0_fr.data(), fc.class_0_fr.size());
+        std::memcpy(fc.prob_bits.data(), pred_bits.data(), fc.prob_bits.size());
+        std::memcpy(fc.single_ref_prob.data(), single_ref_prob.data(), fc.single_ref_prob.size());
+        std::memcpy(fc.comp_ref_prob.data(), comp_ref_prob.data(), fc.comp_ref_prob.size());
+
+        std::memcpy(fc.coef_probs.data(), coef_probs.data(), fc.coef_probs.size());
+    }
+};
+static_assert(sizeof(EntropyProbs) == 0xEA0, "EntropyProbs is an invalid size");
+
+enum class Ref { Last, Golden, AltRef };
+
+struct RefPoolElement {
+    s64 frame{};
+    Ref ref{};
+    bool refresh{};
+};
+
+struct FrameContexts {
+    s64 from{};
+    bool adapted{};
+    Vp9EntropyProbs probs{};
+};
+
+}; // namespace Decoder
+}; // namespace Tegra

src/video_core/command_classes/host1x.cpp

@@ -0,0 +1,39 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+#include "video_core/command_classes/host1x.h"
+#include "video_core/gpu.h"
+
+Tegra::Host1x::Host1x(GPU& gpu_) : gpu(gpu_) {}
+
+Tegra::Host1x::~Host1x() = default;
+
+void Tegra::Host1x::StateWrite(u32 offset, u32 arguments) {
+    u8* const state_offset = reinterpret_cast<u8*>(&state) + offset * sizeof(u32);
+    std::memcpy(state_offset, &arguments, sizeof(u32));
+}
+
+void Tegra::Host1x::ProcessMethod(Method method, const std::vector<u32>& arguments) {
+    StateWrite(static_cast<u32>(method), arguments[0]);
+    switch (method) {
+    case Method::WaitSyncpt:
+        Execute(arguments[0]);
+        break;
+    case Method::LoadSyncptPayload32:
+        syncpoint_value = arguments[0];
+        break;
+    case Method::WaitSyncpt32:
+        Execute(arguments[0]);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Host1x method 0x{:X}", static_cast<u32>(method));
+        break;
+    }
+}
+
+void Tegra::Host1x::Execute(u32 data) {
+    // This method waits on a valid syncpoint.
+    // TODO: Implement when proper Async is in place
+}

src/video_core/command_classes/host1x.h

@@ -0,0 +1,78 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <vector>
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+
+namespace Tegra {
+class GPU;
+class Nvdec;
+
+class Host1x {
+public:
+    struct Host1xClassRegisters {
+        u32 incr_syncpt{};
+        u32 incr_syncpt_ctrl{};
+        u32 incr_syncpt_error{};
+        INSERT_PADDING_WORDS(5);
+        u32 wait_syncpt{};
+        u32 wait_syncpt_base{};
+        u32 wait_syncpt_incr{};
+        u32 load_syncpt_base{};
+        u32 incr_syncpt_base{};
+        u32 clear{};
+        u32 wait{};
+        u32 wait_with_interrupt{};
+        u32 delay_use{};
+        u32 tick_count_high{};
+        u32 tick_count_low{};
+        u32 tick_ctrl{};
+        INSERT_PADDING_WORDS(23);
+        u32 ind_ctrl{};
+        u32 ind_off2{};
+        u32 ind_off{};
+        std::array<u32, 31> ind_data{};
+        INSERT_PADDING_WORDS(1);
+        u32 load_syncpoint_payload32{};
+        u32 stall_ctrl{};
+        u32 wait_syncpt32{};
+        u32 wait_syncpt_base32{};
+        u32 load_syncpt_base32{};
+        u32 incr_syncpt_base32{};
+        u32 stall_count_high{};
+        u32 stall_count_low{};
+        u32 xref_ctrl{};
+        u32 channel_xref_high{};
+        u32 channel_xref_low{};
+    };
+    static_assert(sizeof(Host1xClassRegisters) == 0x164, "Host1xClassRegisters is an invalid size");
+
+    enum class Method : u32 {
+        WaitSyncpt = offsetof(Host1xClassRegisters, wait_syncpt) / 4,
+        LoadSyncptPayload32 = offsetof(Host1xClassRegisters, load_syncpoint_payload32) / 4,
+        WaitSyncpt32 = offsetof(Host1xClassRegisters, wait_syncpt32) / 4,
+    };
+
+    explicit Host1x(GPU& gpu);
+    ~Host1x();
+
+    /// Writes the method into the state, Invoke Execute() if encountered
+    void ProcessMethod(Method method, const std::vector<u32>& arguments);
+
+private:
+    /// For Host1x, execute is waiting on a syncpoint previously written into the state
+    void Execute(u32 data);
+
+    /// Write argument into the provided offset
+    void StateWrite(u32 offset, u32 arguments);
+
+    u32 syncpoint_value{};
+    Host1xClassRegisters state{};
+    GPU& gpu;
+};
+
+} // namespace Tegra

src/video_core/command_classes/nvdec.cpp

@@ -0,0 +1,52 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+#include "video_core/command_classes/nvdec.h"
+#include "video_core/gpu.h"
+
+namespace Tegra {
+
+Nvdec::Nvdec(GPU& gpu_) : gpu(gpu_), codec(std::make_unique<Codec>(gpu)) {}
+
+Nvdec::~Nvdec() = default;
+
+void Nvdec::ProcessMethod(Method method, const std::vector<u32>& arguments) {
+    if (method == Method::SetVideoCodec) {
+        codec->StateWrite(static_cast<u32>(method), arguments[0]);
+    } else {
+        codec->StateWrite(static_cast<u32>(method), static_cast<u64>(arguments[0]) << 8);
+    }
+
+    switch (method) {
+    case Method::SetVideoCodec:
+        codec->SetTargetCodec(static_cast<NvdecCommon::VideoCodec>(arguments[0]));
+        break;
+    case Method::Execute:
+        Execute();
+        break;
+    }
+}
+
+AVFrame* Nvdec::GetFrame() {
+    return codec->GetCurrentFrame();
+}
+
+const AVFrame* Nvdec::GetFrame() const {
+    return codec->GetCurrentFrame();
+}
+
+void Nvdec::Execute() {
+    switch (codec->GetCurrentCodec()) {
+    case NvdecCommon::VideoCodec::H264:
+    case NvdecCommon::VideoCodec::Vp9:
+        codec->Decode();
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unknown codec {}", static_cast<u32>(codec->GetCurrentCodec()));
+        break;
+    }
+}
+
+} // namespace Tegra

src/video_core/command_classes/nvdec.h

@@ -0,0 +1,39 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <vector>
+#include "common/common_types.h"
+#include "video_core/command_classes/codecs/codec.h"
+
+namespace Tegra {
+class GPU;
+
+class Nvdec {
+public:
+    enum class Method : u32 {
+        SetVideoCodec = 0x80,
+        Execute = 0xc0,
+    };
+
+    explicit Nvdec(GPU& gpu);
+    ~Nvdec();
+
+    /// Writes the method into the state, Invoke Execute() if encountered
+    void ProcessMethod(Method method, const std::vector<u32>& arguments);
+
+    /// Return most recently decoded frame
+    [[nodiscard]] AVFrame* GetFrame();
+    [[nodiscard]] const AVFrame* GetFrame() const;
+
+private:
+    /// Invoke codec to decode a frame
+    void Execute();
+
+    GPU& gpu;
+    std::unique_ptr<Codec> codec;
+};
+} // namespace Tegra

src/video_core/command_classes/nvdec_common.h

@@ -0,0 +1,48 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+
+namespace Tegra::NvdecCommon {
+
+struct NvdecRegisters {
+    INSERT_PADDING_WORDS(256);
+    u64 set_codec_id{};
+    INSERT_PADDING_WORDS(254);
+    u64 set_platform_id{};
+    u64 picture_info_offset{};
+    u64 frame_bitstream_offset{};
+    u64 frame_number{};
+    u64 h264_slice_data_offsets{};
+    u64 h264_mv_dump_offset{};
+    INSERT_PADDING_WORDS(6);
+    u64 frame_stats_offset{};
+    u64 h264_last_surface_luma_offset{};
+    u64 h264_last_surface_chroma_offset{};
+    std::array<u64, 17> surface_luma_offset{};
+    std::array<u64, 17> surface_chroma_offset{};
+    INSERT_PADDING_WORDS(132);
+    u64 vp9_entropy_probs_offset{};
+    u64 vp9_backward_updates_offset{};
+    u64 vp9_last_frame_segmap_offset{};
+    u64 vp9_curr_frame_segmap_offset{};
+    INSERT_PADDING_WORDS(2);
+    u64 vp9_last_frame_mvs_offset{};
+    u64 vp9_curr_frame_mvs_offset{};
+    INSERT_PADDING_WORDS(2);
+};
+static_assert(sizeof(NvdecRegisters) == (0xBC0), "NvdecRegisters is incorrect size");
+
+enum class VideoCodec : u32 {
+    None = 0x0,
+    H264 = 0x3,
+    Vp8 = 0x5,
+    H265 = 0x7,
+    Vp9 = 0x9,
+};
+
+} // namespace Tegra::NvdecCommon

src/video_core/command_classes/sync_manager.cpp

@@ -0,0 +1,60 @@
+// MIT License
+//
+// Copyright (c) Ryujinx Team and Contributors
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and
+// associated documentation files (the "Software"), to deal in the Software without restriction,
+// including without limitation the rights to use, copy, modify, merge, publish, distribute,
+// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all copies or
+// substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT
+// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
+// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+
+#include <algorithm>
+#include "sync_manager.h"
+#include "video_core/gpu.h"
+
+namespace Tegra {
+SyncptIncrManager::SyncptIncrManager(GPU& gpu_) : gpu(gpu_) {}
+SyncptIncrManager::~SyncptIncrManager() = default;
+
+void SyncptIncrManager::Increment(u32 id) {
+    increments.emplace_back(0, 0, id, true);
+    IncrementAllDone();
+}
+
+u32 SyncptIncrManager::IncrementWhenDone(u32 class_id, u32 id) {
+    const u32 handle = current_id++;
+    increments.emplace_back(handle, class_id, id);
+    return handle;
+}
+
+void SyncptIncrManager::SignalDone(u32 handle) {
+    const auto done_incr =
+        std::find_if(increments.begin(), increments.end(),
+                     [handle](const SyncptIncr& incr) { return incr.id == handle; });
+    if (done_incr != increments.cend()) {
+        done_incr->complete = true;
+    }
+    IncrementAllDone();
+}
+
+void SyncptIncrManager::IncrementAllDone() {
+    std::size_t done_count = 0;
+    for (; done_count < increments.size(); ++done_count) {
+        if (!increments[done_count].complete) {
+            break;
+        }
+        gpu.IncrementSyncPoint(increments[done_count].syncpt_id);
+    }
+    increments.erase(increments.begin(), increments.begin() + done_count);
+}
+} // namespace Tegra

src/video_core/command_classes/sync_manager.h

@@ -0,0 +1,64 @@
+// MIT License
+//
+// Copyright (c) Ryujinx Team and Contributors
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and
+// associated documentation files (the "Software"), to deal in the Software without restriction,
+// including without limitation the rights to use, copy, modify, merge, publish, distribute,
+// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all copies or
+// substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT
+// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
+// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//
+
+#pragma once
+
+#include <mutex>
+#include <vector>
+#include "common/common_types.h"
+
+namespace Tegra {
+class GPU;
+struct SyncptIncr {
+    u32 id;
+    u32 class_id;
+    u32 syncpt_id;
+    bool complete;
+
+    SyncptIncr(u32 id_, u32 class_id_, u32 syncpt_id_, bool done = false)
+        : id(id_), class_id(class_id_), syncpt_id(syncpt_id_), complete(done) {}
+};
+
+class SyncptIncrManager {
+public:
+    explicit SyncptIncrManager(GPU& gpu);
+    ~SyncptIncrManager();
+
+    /// Add syncpoint id and increment all
+    void Increment(u32 id);
+
+    /// Returns a handle to increment later
+    u32 IncrementWhenDone(u32 class_id, u32 id);
+
+    /// IncrememntAllDone, including handle
+    void SignalDone(u32 handle);
+
+    /// Increment all sequential pending increments that are already done.
+    void IncrementAllDone();
+
+private:
+    std::vector<SyncptIncr> increments;
+    std::mutex increment_lock;
+    u32 current_id{};
+
+    GPU& gpu;
+};
+
+} // namespace Tegra

src/video_core/command_classes/vic.cpp

@@ -0,0 +1,180 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <array>
+#include "common/assert.h"
+#include "video_core/command_classes/nvdec.h"
+#include "video_core/command_classes/vic.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
+#include "video_core/texture_cache/surface_params.h"
+
+extern "C" {
+#include <libswscale/swscale.h>
+}
+
+namespace Tegra {
+
+Vic::Vic(GPU& gpu_, std::shared_ptr<Nvdec> nvdec_processor_)
+    : gpu(gpu_), nvdec_processor(std::move(nvdec_processor_)) {}
+Vic::~Vic() = default;
+
+void Vic::VicStateWrite(u32 offset, u32 arguments) {
+    u8* const state_offset = reinterpret_cast<u8*>(&vic_state) + offset * sizeof(u32);
+    std::memcpy(state_offset, &arguments, sizeof(u32));
+}
+
+void Vic::ProcessMethod(Method method, const std::vector<u32>& arguments) {
+    LOG_DEBUG(HW_GPU, "Vic method 0x{:X}", static_cast<u32>(method));
+    VicStateWrite(static_cast<u32>(method), arguments[0]);
+    const u64 arg = static_cast<u64>(arguments[0]) << 8;
+    switch (method) {
+    case Method::Execute:
+        Execute();
+        break;
+    case Method::SetConfigStructOffset:
+        config_struct_address = arg;
+        break;
+    case Method::SetOutputSurfaceLumaOffset:
+        output_surface_luma_address = arg;
+        break;
+    case Method::SetOutputSurfaceChromaUOffset:
+        output_surface_chroma_u_address = arg;
+        break;
+    case Method::SetOutputSurfaceChromaVOffset:
+        output_surface_chroma_v_address = arg;
+        break;
+    default:
+        break;
+    }
+}
+
+void Vic::Execute() {
+    if (output_surface_luma_address == 0) {
+        LOG_ERROR(Service_NVDRV, "VIC Luma address not set. Recieved 0x{:X}",
+                  vic_state.output_surface.luma_offset);
+        return;
+    }
+    const VicConfig config{gpu.MemoryManager().Read<u64>(config_struct_address + 0x20)};
+    const VideoPixelFormat pixel_format =
+        static_cast<VideoPixelFormat>(config.pixel_format.Value());
+    switch (pixel_format) {
+    case VideoPixelFormat::BGRA8:
+    case VideoPixelFormat::RGBA8: {
+        LOG_TRACE(Service_NVDRV, "Writing RGB Frame");
+        const auto* frame = nvdec_processor->GetFrame();
+
+        if (!frame || frame->width == 0 || frame->height == 0) {
+            return;
+        }
+        if (scaler_ctx == nullptr || frame->width != scaler_width ||
+            frame->height != scaler_height) {
+            const AVPixelFormat target_format =
+                (pixel_format == VideoPixelFormat::RGBA8) ? AV_PIX_FMT_RGBA : AV_PIX_FMT_BGRA;
+
+            sws_freeContext(scaler_ctx);
+            scaler_ctx = nullptr;
+
+            // FFmpeg returns all frames in YUV420, convert it into expected format
+            scaler_ctx =
+                sws_getContext(frame->width, frame->height, AV_PIX_FMT_YUV420P, frame->width,
+                               frame->height, target_format, 0, nullptr, nullptr, nullptr);
+
+            scaler_width = frame->width;
+            scaler_height = frame->height;
+        }
+        // Get Converted frame
+        const std::size_t linear_size = frame->width * frame->height * 4;
+
+        using AVMallocPtr = std::unique_ptr<u8, decltype(&av_free)>;
+        AVMallocPtr converted_frame_buffer{static_cast<u8*>(av_malloc(linear_size)), av_free};
+
+        const int converted_stride{frame->width * 4};
+        u8* const converted_frame_buf_addr{converted_frame_buffer.get()};
+
+        sws_scale(scaler_ctx, frame->data, frame->linesize, 0, frame->height,
+                  &converted_frame_buf_addr, &converted_stride);
+
+        const u32 blk_kind = static_cast<u32>(config.block_linear_kind);
+        if (blk_kind != 0) {
+            // swizzle pitch linear to block linear
+            const u32 block_height = static_cast<u32>(config.block_linear_height_log2);
+            const auto size = Tegra::Texture::CalculateSize(true, 4, frame->width, frame->height, 1,
+                                                            block_height, 0);
+            std::vector<u8> swizzled_data(size);
+            Tegra::Texture::CopySwizzledData(frame->width, frame->height, 1, 4, 4,
+                                             swizzled_data.data(), converted_frame_buffer.get(),
+                                             false, block_height, 0, 1);
+
+            gpu.MemoryManager().WriteBlock(output_surface_luma_address, swizzled_data.data(), size);
+            gpu.Maxwell3D().OnMemoryWrite();
+        } else {
+            // send pitch linear frame
+            gpu.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr,
+                                           linear_size);
+            gpu.Maxwell3D().OnMemoryWrite();
+        }
+        break;
+    }
+    case VideoPixelFormat::Yuv420: {
+        LOG_TRACE(Service_NVDRV, "Writing YUV420 Frame");
+
+        const auto* frame = nvdec_processor->GetFrame();
+
+        if (!frame || frame->width == 0 || frame->height == 0) {
+            return;
+        }
+
+        const std::size_t surface_width = config.surface_width_minus1 + 1;
+        const std::size_t surface_height = config.surface_height_minus1 + 1;
+        const std::size_t half_width = surface_width / 2;
+        const std::size_t half_height = config.surface_height_minus1 / 2;
+        const std::size_t aligned_width = (surface_width + 0xff) & ~0xff;
+
+        const auto* luma_ptr = frame->data[0];
+        const auto* chroma_b_ptr = frame->data[1];
+        const auto* chroma_r_ptr = frame->data[2];
+        const auto stride = frame->linesize[0];
+        const auto half_stride = frame->linesize[1];
+
+        std::vector<u8> luma_buffer(aligned_width * surface_height);
+        std::vector<u8> chroma_buffer(aligned_width * half_height);
+
+        // Populate luma buffer
+        for (std::size_t y = 0; y < surface_height - 1; ++y) {
+            std::size_t src = y * stride;
+            std::size_t dst = y * aligned_width;
+
+            std::size_t size = surface_width;
+
+            for (std::size_t offset = 0; offset < size; ++offset) {
+                luma_buffer[dst + offset] = luma_ptr[src + offset];
+            }
+        }
+        gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(),
+                                       luma_buffer.size());
+
+        // Populate chroma buffer from both channels with interleaving.
+        for (std::size_t y = 0; y < half_height; ++y) {
+            std::size_t src = y * half_stride;
+            std::size_t dst = y * aligned_width;
+
+            for (std::size_t x = 0; x < half_width; ++x) {
+                chroma_buffer[dst + x * 2] = chroma_b_ptr[src + x];
+                chroma_buffer[dst + x * 2 + 1] = chroma_r_ptr[src + x];
+            }
+        }
+        gpu.MemoryManager().WriteBlock(output_surface_chroma_u_address, chroma_buffer.data(),
+                                       chroma_buffer.size());
+        gpu.Maxwell3D().OnMemoryWrite();
+        break;
+    }
+    default:
+        UNIMPLEMENTED_MSG("Unknown video pixel format {}", config.pixel_format.Value());
+        break;
+    }
+}
+
+} // namespace Tegra

src/video_core/command_classes/vic.h

@@ -0,0 +1,110 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <vector>
+#include "common/bit_field.h"
+#include "common/common_types.h"
+
+struct SwsContext;
+
+namespace Tegra {
+class GPU;
+class Nvdec;
+
+struct PlaneOffsets {
+    u32 luma_offset{};
+    u32 chroma_u_offset{};
+    u32 chroma_v_offset{};
+};
+
+struct VicRegisters {
+    INSERT_PADDING_WORDS(64);
+    u32 nop{};
+    INSERT_PADDING_WORDS(15);
+    u32 pm_trigger{};
+    INSERT_PADDING_WORDS(47);
+    u32 set_application_id{};
+    u32 set_watchdog_timer{};
+    INSERT_PADDING_WORDS(17);
+    u32 context_save_area{};
+    u32 context_switch{};
+    INSERT_PADDING_WORDS(43);
+    u32 execute{};
+    INSERT_PADDING_WORDS(63);
+    std::array<std::array<PlaneOffsets, 8>, 8> surfacex_slots{};
+    u32 picture_index{};
+    u32 control_params{};
+    u32 config_struct_offset{};
+    u32 filter_struct_offset{};
+    u32 palette_offset{};
+    u32 hist_offset{};
+    u32 context_id{};
+    u32 fce_ucode_size{};
+    PlaneOffsets output_surface{};
+    u32 fce_ucode_offset{};
+    INSERT_PADDING_WORDS(4);
+    std::array<u32, 8> slot_context_id{};
+    INSERT_PADDING_WORDS(16);
+};
+static_assert(sizeof(VicRegisters) == 0x7A0, "VicRegisters is an invalid size");
+
+class Vic {
+public:
+    enum class Method : u32 {
+        Execute = 0xc0,
+        SetControlParams = 0x1c1,
+        SetConfigStructOffset = 0x1c2,
+        SetOutputSurfaceLumaOffset = 0x1c8,
+        SetOutputSurfaceChromaUOffset = 0x1c9,
+        SetOutputSurfaceChromaVOffset = 0x1ca
+    };
+
+    explicit Vic(GPU& gpu, std::shared_ptr<Nvdec> nvdec_processor);
+    ~Vic();
+
+    /// Write to the device state.
+    void ProcessMethod(Method method, const std::vector<u32>& arguments);
+
+private:
+    void Execute();
+
+    void VicStateWrite(u32 offset, u32 arguments);
+    VicRegisters vic_state{};
+
+    enum class VideoPixelFormat : u64_le {
+        RGBA8 = 0x1f,
+        BGRA8 = 0x20,
+        Yuv420 = 0x44,
+    };
+
+    union VicConfig {
+        u64_le raw{};
+        BitField<0, 7, u64_le> pixel_format;
+        BitField<7, 2, u64_le> chroma_loc_horiz;
+        BitField<9, 2, u64_le> chroma_loc_vert;
+        BitField<11, 4, u64_le> block_linear_kind;
+        BitField<15, 4, u64_le> block_linear_height_log2;
+        BitField<19, 3, u64_le> reserved0;
+        BitField<22, 10, u64_le> reserved1;
+        BitField<32, 14, u64_le> surface_width_minus1;
+        BitField<46, 14, u64_le> surface_height_minus1;
+    };
+
+    GPU& gpu;
+    std::shared_ptr<Tegra::Nvdec> nvdec_processor;
+
+    GPUVAddr config_struct_address{};
+    GPUVAddr output_surface_luma_address{};
+    GPUVAddr output_surface_chroma_u_address{};
+    GPUVAddr output_surface_chroma_v_address{};
+
+    SwsContext* scaler_ctx{};
+    s32 scaler_width{};
+    s32 scaler_height{};
+};
+
+} // namespace Tegra

src/video_core/dma_pusher.cpp

@@ -2,6 +2,7 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "common/cityhash.h"
 #include "common/microprofile.h"
 #include "core/core.h"
 #include "core/memory.h"
@@ -12,6 +13,20 @@
 
 namespace Tegra {
 
+void CommandList::RefreshIntegrityChecks(GPU& gpu) {
+    command_list_hashes.resize(command_lists.size());
+
+    for (std::size_t index = 0; index < command_lists.size(); ++index) {
+        const CommandListHeader command_list_header = command_lists[index];
+        std::vector<CommandHeader> command_headers(command_list_header.size);
+        gpu.MemoryManager().ReadBlockUnsafe(command_list_header.addr, command_headers.data(),
+                                            command_list_header.size * sizeof(u32));
+        command_list_hashes[index] =
+            Common::CityHash64(reinterpret_cast<char*>(command_headers.data()),
+                               command_list_header.size * sizeof(u32));
+    }
+}
+
 DmaPusher::DmaPusher(Core::System& system, GPU& gpu) : gpu{gpu}, system{system} {}
 
 DmaPusher::~DmaPusher() = default;
@@ -45,32 +60,51 @@ bool DmaPusher::Step() {
         return false;
     }
 
-    const CommandList& command_list{dma_pushbuffer.front()};
-    ASSERT_OR_EXECUTE(!command_list.empty(), {
-        // Somehow the command_list is empty, in order to avoid a crash
-        // We ignore it and assume its size is 0.
+    CommandList& command_list{dma_pushbuffer.front()};
+
+    ASSERT_OR_EXECUTE(
+        command_list.command_lists.size() || command_list.prefetch_command_list.size(), {
+            // Somehow the command_list is empty, in order to avoid a crash
+            // We ignore it and assume its size is 0.
+            dma_pushbuffer.pop();
+            dma_pushbuffer_subindex = 0;
+            return true;
+        });
+
+    if (command_list.prefetch_command_list.size()) {
+        // Prefetched command list from nvdrv, used for things like synchronization
+        command_headers = std::move(command_list.prefetch_command_list);
         dma_pushbuffer.pop();
-        dma_pushbuffer_subindex = 0;
-        return true;
-    });
-    const CommandListHeader command_list_header{command_list[dma_pushbuffer_subindex++]};
-    const GPUVAddr dma_get = command_list_header.addr;
+    } else {
+        const CommandListHeader command_list_header{
+            command_list.command_lists[dma_pushbuffer_subindex]};
+        const u64 next_hash = command_list.command_list_hashes[dma_pushbuffer_subindex++];
+        const GPUVAddr dma_get = command_list_header.addr;
 
-    if (dma_pushbuffer_subindex >= command_list.size()) {
-        // We've gone through the current list, remove it from the queue
-        dma_pushbuffer.pop();
-        dma_pushbuffer_subindex = 0;
+        if (dma_pushbuffer_subindex >= command_list.command_lists.size()) {
+            // We've gone through the current list, remove it from the queue
+            dma_pushbuffer.pop();
+            dma_pushbuffer_subindex = 0;
+        }
+
+        if (command_list_header.size == 0) {
+            return true;
+        }
+
+        // Push buffer non-empty, read a word
+        command_headers.resize(command_list_header.size);
+        gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(),
+                                            command_list_header.size * sizeof(u32));
+
+        // Integrity check
+        const u64 new_hash = Common::CityHash64(reinterpret_cast<char*>(command_headers.data()),
+                                                command_list_header.size * sizeof(u32));
+        if (new_hash != next_hash) {
+            LOG_CRITICAL(HW_GPU, "CommandList at addr=0x{:X} is corrupt, skipping!", dma_get);
+            dma_pushbuffer.pop();
+            return true;
+        }
     }
-
-    if (command_list_header.size == 0) {
-        return true;
-    }
-
-    // Push buffer non-empty, read a word
-    command_headers.resize(command_list_header.size);
-    gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(),
-                                        command_list_header.size * sizeof(u32));
-
     for (std::size_t index = 0; index < command_headers.size();) {
         const CommandHeader& command_header = command_headers[index];
 

src/video_core/dma_pusher.h

@@ -18,6 +18,8 @@ class System;
 
 namespace Tegra {
 
+class GPU;
+
 enum class SubmissionMode : u32 {
     IncreasingOld = 0,
     Increasing = 1,
@@ -27,6 +29,31 @@ enum class SubmissionMode : u32 {
     IncreaseOnce = 5
 };
 
+// Note that, traditionally, methods are treated as 4-byte addressable locations, and hence
+// their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4.
+// So the values you see in docs might be multiplied by 4.
+enum class BufferMethods : u32 {
+    BindObject = 0x0,
+    Nop = 0x2,
+    SemaphoreAddressHigh = 0x4,
+    SemaphoreAddressLow = 0x5,
+    SemaphoreSequence = 0x6,
+    SemaphoreTrigger = 0x7,
+    NotifyIntr = 0x8,
+    WrcacheFlush = 0x9,
+    Unk28 = 0xA,
+    UnkCacheFlush = 0xB,
+    RefCnt = 0x14,
+    SemaphoreAcquire = 0x1A,
+    SemaphoreRelease = 0x1B,
+    FenceValue = 0x1C,
+    FenceAction = 0x1D,
+    WaitForInterrupt = 0x1E,
+    Unk7c = 0x1F,
+    Yield = 0x20,
+    NonPullerMethods = 0x40,
+};
+
 struct CommandListHeader {
     union {
         u64 raw;
@@ -49,9 +76,26 @@ union CommandHeader {
 static_assert(std::is_standard_layout_v<CommandHeader>, "CommandHeader is not standard layout");
 static_assert(sizeof(CommandHeader) == sizeof(u32), "CommandHeader has incorrect size!");
 
-class GPU;
+inline CommandHeader BuildCommandHeader(BufferMethods method, u32 arg_count, SubmissionMode mode) {
+    CommandHeader result{};
+    result.method.Assign(static_cast<u32>(method));
+    result.arg_count.Assign(arg_count);
+    result.mode.Assign(mode);
+    return result;
+}
 
-using CommandList = std::vector<Tegra::CommandListHeader>;
+struct CommandList final {
+    CommandList() = default;
+    explicit CommandList(std::size_t size) : command_lists(size) {}
+    explicit CommandList(std::vector<Tegra::CommandHeader>&& prefetch_command_list)
+        : prefetch_command_list{std::move(prefetch_command_list)} {}
+
+    void RefreshIntegrityChecks(GPU& gpu);
+
+    std::vector<Tegra::CommandListHeader> command_lists;
+    std::vector<u64> command_list_hashes;
+    std::vector<Tegra::CommandHeader> prefetch_command_list;
+};
 
 /**
  * The DmaPusher class implements DMA submission to FIFOs, providing an area of memory that the
@@ -60,7 +104,7 @@ using CommandList = std::vector<Tegra::CommandListHeader>;
  * See https://envytools.readthedocs.io/en/latest/hw/fifo/dma-pusher.html#fifo-dma-pusher for
  * details on this implementation.
  */
-class DmaPusher {
+class DmaPusher final {
 public:
     explicit DmaPusher(Core::System& system, GPU& gpu);
     ~DmaPusher();

src/video_core/engines/shader_bytecode.h

@@ -1893,6 +1893,7 @@ public:
         ICMP_IMM,
         FCMP_RR,
         FCMP_RC,
+        FCMP_IMMR,
         MUFU,  // Multi-Function Operator
         RRO_C, // Range Reduction Operator
         RRO_R,
@@ -2205,6 +2206,7 @@ private:
             INST("0111110-0-------", Id::HSET2_IMM, Type::HalfSet, "HSET2_IMM"),
             INST("010110111010----", Id::FCMP_RR, Type::Arithmetic, "FCMP_RR"),
             INST("010010111010----", Id::FCMP_RC, Type::Arithmetic, "FCMP_RC"),
+            INST("0011011-1010----", Id::FCMP_IMMR, Type::Arithmetic, "FCMP_IMMR"),
             INST("0101000010000---", Id::MUFU, Type::Arithmetic, "MUFU"),
             INST("0100110010010---", Id::RRO_C, Type::Arithmetic, "RRO_C"),
             INST("0101110010010---", Id::RRO_R, Type::Arithmetic, "RRO_R"),

src/video_core/gpu.cpp

@@ -27,9 +27,10 @@ namespace Tegra {
 
 MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
 
-GPU::GPU(Core::System& system_, bool is_async_)
+GPU::GPU(Core::System& system_, bool is_async_, bool use_nvdec_)
     : system{system_}, memory_manager{std::make_unique<Tegra::MemoryManager>(system)},
       dma_pusher{std::make_unique<Tegra::DmaPusher>(system, *this)},
+      cdma_pusher{std::make_unique<Tegra::CDmaPusher>(*this)}, use_nvdec{use_nvdec_},
       maxwell_3d{std::make_unique<Engines::Maxwell3D>(system, *memory_manager)},
       fermi_2d{std::make_unique<Engines::Fermi2D>()},
       kepler_compute{std::make_unique<Engines::KeplerCompute>(system, *memory_manager)},
@@ -77,10 +78,18 @@ DmaPusher& GPU::DmaPusher() {
     return *dma_pusher;
 }
 
+Tegra::CDmaPusher& GPU::CDmaPusher() {
+    return *cdma_pusher;
+}
+
 const DmaPusher& GPU::DmaPusher() const {
     return *dma_pusher;
 }
 
+const Tegra::CDmaPusher& GPU::CDmaPusher() const {
+    return *cdma_pusher;
+}
+
 void GPU::WaitFence(u32 syncpoint_id, u32 value) {
     // Synced GPU, is always in sync
     if (!is_async) {
@@ -185,30 +194,6 @@ void GPU::SyncGuestHost() {
 void GPU::OnCommandListEnd() {
     renderer->Rasterizer().ReleaseFences();
 }
-// Note that, traditionally, methods are treated as 4-byte addressable locations, and hence
-// their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4.
-// So the values you see in docs might be multiplied by 4.
-enum class BufferMethods {
-    BindObject = 0x0,
-    Nop = 0x2,
-    SemaphoreAddressHigh = 0x4,
-    SemaphoreAddressLow = 0x5,
-    SemaphoreSequence = 0x6,
-    SemaphoreTrigger = 0x7,
-    NotifyIntr = 0x8,
-    WrcacheFlush = 0x9,
-    Unk28 = 0xA,
-    UnkCacheFlush = 0xB,
-    RefCnt = 0x14,
-    SemaphoreAcquire = 0x1A,
-    SemaphoreRelease = 0x1B,
-    FenceValue = 0x1C,
-    FenceAction = 0x1D,
-    Unk78 = 0x1E,
-    Unk7c = 0x1F,
-    Yield = 0x20,
-    NonPullerMethods = 0x40,
-};
 
 enum class GpuSemaphoreOperation {
     AcquireEqual = 0x1,
@@ -268,7 +253,12 @@ void GPU::CallPullerMethod(const MethodCall& method_call) {
     case BufferMethods::UnkCacheFlush:
     case BufferMethods::WrcacheFlush:
     case BufferMethods::FenceValue:
+        break;
     case BufferMethods::FenceAction:
+        ProcessFenceActionMethod();
+        break;
+    case BufferMethods::WaitForInterrupt:
+        ProcessWaitForInterruptMethod();
         break;
     case BufferMethods::SemaphoreTrigger: {
         ProcessSemaphoreTriggerMethod();
@@ -382,6 +372,25 @@ void GPU::ProcessBindMethod(const MethodCall& method_call) {
     }
 }
 
+void GPU::ProcessFenceActionMethod() {
+    switch (regs.fence_action.op) {
+    case FenceOperation::Acquire:
+        WaitFence(regs.fence_action.syncpoint_id, regs.fence_value);
+        break;
+    case FenceOperation::Increment:
+        IncrementSyncPoint(regs.fence_action.syncpoint_id);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented operation {}",
+                          static_cast<u32>(regs.fence_action.op.Value()));
+    }
+}
+
+void GPU::ProcessWaitForInterruptMethod() {
+    // TODO(bunnei) ImplementMe
+    LOG_WARNING(HW_GPU, "(STUBBED) called");
+}
+
 void GPU::ProcessSemaphoreTriggerMethod() {
     const auto semaphoreOperationMask = 0xF;
     const auto op =

src/video_core/gpu.h

@@ -13,6 +13,7 @@
 #include "common/common_types.h"
 #include "core/hle/service/nvdrv/nvdata.h"
 #include "core/hle/service/nvflinger/buffer_queue.h"
+#include "video_core/cdma_pusher.h"
 #include "video_core/dma_pusher.h"
 
 using CacheAddr = std::uintptr_t;
@@ -157,7 +158,7 @@ public:
               method_count(method_count) {}
     };
 
-    explicit GPU(Core::System& system, bool is_async);
+    explicit GPU(Core::System& system, bool is_async, bool use_nvdec);
     virtual ~GPU();
 
     /// Binds a renderer to the GPU.
@@ -209,6 +210,15 @@ public:
     /// Returns a reference to the GPU DMA pusher.
     Tegra::DmaPusher& DmaPusher();
 
+    /// Returns a const reference to the GPU DMA pusher.
+    const Tegra::DmaPusher& DmaPusher() const;
+
+    /// Returns a reference to the GPU CDMA pusher.
+    Tegra::CDmaPusher& CDmaPusher();
+
+    /// Returns a const reference to the GPU CDMA pusher.
+    const Tegra::CDmaPusher& CDmaPusher() const;
+
     VideoCore::RendererBase& Renderer() {
         return *renderer;
     }
@@ -249,8 +259,27 @@ public:
         return is_async;
     }
 
-    /// Returns a const reference to the GPU DMA pusher.
-    const Tegra::DmaPusher& DmaPusher() const;
+    bool UseNvdec() const {
+        return use_nvdec;
+    }
+
+    enum class FenceOperation : u32 {
+        Acquire = 0,
+        Increment = 1,
+    };
+
+    union FenceAction {
+        u32 raw;
+        BitField<0, 1, FenceOperation> op;
+        BitField<8, 24, u32> syncpoint_id;
+
+        static CommandHeader Build(FenceOperation op, u32 syncpoint_id) {
+            FenceAction result{};
+            result.op.Assign(op);
+            result.syncpoint_id.Assign(syncpoint_id);
+            return {result.raw};
+        }
+    };
 
     struct Regs {
         static constexpr size_t NUM_REGS = 0x40;
@@ -280,10 +309,7 @@ public:
                 u32 semaphore_acquire;
                 u32 semaphore_release;
                 u32 fence_value;
-                union {
-                    BitField<4, 4, u32> operation;
-                    BitField<8, 8, u32> id;
-                } fence_action;
+                FenceAction fence_action;
                 INSERT_UNION_PADDING_WORDS(0xE2);
 
                 // Puller state
@@ -311,6 +337,9 @@ public:
     /// Push GPU command entries to be processed
     virtual void PushGPUEntries(Tegra::CommandList&& entries) = 0;
 
+    /// Push GPU command buffer entries to be processed
+    virtual void PushCommandBuffer(Tegra::ChCommandHeaderList& entries) = 0;
+
     /// Swap buffers (render frame)
     virtual void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) = 0;
 
@@ -328,6 +357,8 @@ protected:
 
 private:
     void ProcessBindMethod(const MethodCall& method_call);
+    void ProcessFenceActionMethod();
+    void ProcessWaitForInterruptMethod();
     void ProcessSemaphoreTriggerMethod();
     void ProcessSemaphoreRelease();
     void ProcessSemaphoreAcquire();
@@ -349,7 +380,9 @@ protected:
     Core::System& system;
     std::unique_ptr<Tegra::MemoryManager> memory_manager;
     std::unique_ptr<Tegra::DmaPusher> dma_pusher;
+    std::unique_ptr<Tegra::CDmaPusher> cdma_pusher;
     std::unique_ptr<VideoCore::RendererBase> renderer;
+    const bool use_nvdec;
 
 private:
     /// Mapping of command subchannels to their bound engine ids
@@ -372,6 +405,7 @@ private:
     std::array<std::list<u32>, Service::Nvidia::MaxSyncPoints> syncpt_interrupts;
 
     std::mutex sync_mutex;
+    std::mutex device_mutex;
 
     std::condition_variable sync_cv;
 

src/video_core/gpu_asynch.cpp

@@ -10,12 +10,13 @@
 
 namespace VideoCommon {
 
-GPUAsynch::GPUAsynch(Core::System& system) : GPU{system, true}, gpu_thread{system} {}
+GPUAsynch::GPUAsynch(Core::System& system, bool use_nvdec)
+    : GPU{system, true, use_nvdec}, gpu_thread{system} {}
 
 GPUAsynch::~GPUAsynch() = default;
 
 void GPUAsynch::Start() {
-    gpu_thread.StartThread(*renderer, renderer->Context(), *dma_pusher);
+    gpu_thread.StartThread(*renderer, renderer->Context(), *dma_pusher, *cdma_pusher);
     cpu_context = renderer->GetRenderWindow().CreateSharedContext();
     cpu_context->MakeCurrent();
 }
@@ -32,6 +33,27 @@ void GPUAsynch::PushGPUEntries(Tegra::CommandList&& entries) {
     gpu_thread.SubmitList(std::move(entries));
 }
 
+void GPUAsynch::PushCommandBuffer(Tegra::ChCommandHeaderList& entries) {
+    if (!use_nvdec) {
+        return;
+    }
+    // This condition fires when a video stream ends, clear all intermediary data
+    if (entries[0].raw == 0xDEADB33F) {
+        cdma_pusher.reset();
+        return;
+    }
+    if (!cdma_pusher) {
+        cdma_pusher = std::make_unique<Tegra::CDmaPusher>(*this);
+    }
+
+    // SubmitCommandBuffer would make the nvdec operations async, this is not currently working
+    // TODO(ameerj): RE proper async nvdec operation
+    // gpu_thread.SubmitCommandBuffer(std::move(entries));
+
+    cdma_pusher->Push(std::move(entries));
+    cdma_pusher->DispatchCalls();
+}
+
 void GPUAsynch::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
     gpu_thread.SwapBuffers(framebuffer);
 }

src/video_core/gpu_asynch.h

@@ -20,13 +20,14 @@ namespace VideoCommon {
 /// Implementation of GPU interface that runs the GPU asynchronously
 class GPUAsynch final : public Tegra::GPU {
 public:
-    explicit GPUAsynch(Core::System& system);
+    explicit GPUAsynch(Core::System& system, bool use_nvdec);
     ~GPUAsynch() override;
 
     void Start() override;
     void ObtainContext() override;
     void ReleaseContext() override;
     void PushGPUEntries(Tegra::CommandList&& entries) override;
+    void PushCommandBuffer(Tegra::ChCommandHeaderList& entries) override;
     void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
     void FlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;

src/video_core/gpu_synch.cpp

@@ -7,7 +7,7 @@
 
 namespace VideoCommon {
 
-GPUSynch::GPUSynch(Core::System& system) : GPU{system, false} {}
+GPUSynch::GPUSynch(Core::System& system, bool use_nvdec) : GPU{system, false, use_nvdec} {}
 
 GPUSynch::~GPUSynch() = default;
 
@@ -26,6 +26,22 @@ void GPUSynch::PushGPUEntries(Tegra::CommandList&& entries) {
     dma_pusher->DispatchCalls();
 }
 
+void GPUSynch::PushCommandBuffer(Tegra::ChCommandHeaderList& entries) {
+    if (!use_nvdec) {
+        return;
+    }
+    // This condition fires when a video stream ends, clears all intermediary data
+    if (entries[0].raw == 0xDEADB33F) {
+        cdma_pusher.reset();
+        return;
+    }
+    if (!cdma_pusher) {
+        cdma_pusher = std::make_unique<Tegra::CDmaPusher>(*this);
+    }
+    cdma_pusher->Push(std::move(entries));
+    cdma_pusher->DispatchCalls();
+}
+
 void GPUSynch::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
     renderer->SwapBuffers(framebuffer);
 }

src/video_core/gpu_synch.h

@@ -19,13 +19,14 @@ namespace VideoCommon {
 /// Implementation of GPU interface that runs the GPU synchronously
 class GPUSynch final : public Tegra::GPU {
 public:
-    explicit GPUSynch(Core::System& system);
+    explicit GPUSynch(Core::System& system, bool use_nvdec);
     ~GPUSynch() override;
 
     void Start() override;
     void ObtainContext() override;
     void ReleaseContext() override;
     void PushGPUEntries(Tegra::CommandList&& entries) override;
+    void PushCommandBuffer(Tegra::ChCommandHeaderList& entries) override;
     void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
     void FlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;

src/video_core/gpu_thread.cpp

@@ -18,7 +18,7 @@ namespace VideoCommon::GPUThread {
 /// Runs the GPU thread
 static void RunThread(Core::System& system, VideoCore::RendererBase& renderer,
                       Core::Frontend::GraphicsContext& context, Tegra::DmaPusher& dma_pusher,
-                      SynchState& state) {
+                      SynchState& state, Tegra::CDmaPusher& cdma_pusher) {
     std::string name = "yuzu:GPU";
     MicroProfileOnThreadCreate(name.c_str());
     Common::SetCurrentThreadName(name.c_str());
@@ -42,6 +42,10 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer,
         if (const auto submit_list = std::get_if<SubmitListCommand>(&next.data)) {
             dma_pusher.Push(std::move(submit_list->entries));
             dma_pusher.DispatchCalls();
+        } else if (const auto command_list = std::get_if<SubmitChCommandEntries>(&next.data)) {
+            // NVDEC
+            cdma_pusher.Push(std::move(command_list->entries));
+            cdma_pusher.DispatchCalls();
         } else if (const auto data = std::get_if<SwapBuffersCommand>(&next.data)) {
             renderer.SwapBuffers(data->framebuffer ? &*data->framebuffer : nullptr);
         } else if (std::holds_alternative<OnCommandListEndCommand>(next.data)) {
@@ -75,15 +79,19 @@ ThreadManager::~ThreadManager() {
 
 void ThreadManager::StartThread(VideoCore::RendererBase& renderer,
                                 Core::Frontend::GraphicsContext& context,
-                                Tegra::DmaPusher& dma_pusher) {
-    thread = std::thread{RunThread,         std::ref(system),     std::ref(renderer),
-                         std::ref(context), std::ref(dma_pusher), std::ref(state)};
+                                Tegra::DmaPusher& dma_pusher, Tegra::CDmaPusher& cdma_pusher) {
+    thread = std::thread(RunThread, std::ref(system), std::ref(renderer), std::ref(context),
+                         std::ref(dma_pusher), std::ref(state), std::ref(cdma_pusher));
 }
 
 void ThreadManager::SubmitList(Tegra::CommandList&& entries) {
     PushCommand(SubmitListCommand(std::move(entries)));
 }
 
+void ThreadManager::SubmitCommandBuffer(Tegra::ChCommandHeaderList&& entries) {
+    PushCommand(SubmitChCommandEntries(std::move(entries)));
+}
+
 void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
     PushCommand(SwapBuffersCommand(framebuffer ? std::make_optional(*framebuffer) : std::nullopt));
 }

src/video_core/gpu_thread.h

@@ -37,6 +37,14 @@ struct SubmitListCommand final {
     Tegra::CommandList entries;
 };
 
+/// Command to signal to the GPU thread that a cdma command list is ready for processing
+struct SubmitChCommandEntries final {
+    explicit SubmitChCommandEntries(Tegra::ChCommandHeaderList&& entries)
+        : entries{std::move(entries)} {}
+
+    Tegra::ChCommandHeaderList entries;
+};
+
 /// Command to signal to the GPU thread that a swap buffers is pending
 struct SwapBuffersCommand final {
     explicit SwapBuffersCommand(std::optional<const Tegra::FramebufferConfig> framebuffer)
@@ -77,9 +85,9 @@ struct OnCommandListEndCommand final {};
 struct GPUTickCommand final {};
 
 using CommandData =
-    std::variant<EndProcessingCommand, SubmitListCommand, SwapBuffersCommand, FlushRegionCommand,
-                 InvalidateRegionCommand, FlushAndInvalidateRegionCommand, OnCommandListEndCommand,
-                 GPUTickCommand>;
+    std::variant<EndProcessingCommand, SubmitListCommand, SubmitChCommandEntries,
+                 SwapBuffersCommand, FlushRegionCommand, InvalidateRegionCommand,
+                 FlushAndInvalidateRegionCommand, OnCommandListEndCommand, GPUTickCommand>;
 
 struct CommandDataContainer {
     CommandDataContainer() = default;
@@ -109,11 +117,14 @@ public:
 
     /// Creates and starts the GPU thread.
     void StartThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context,
-                     Tegra::DmaPusher& dma_pusher);
+                     Tegra::DmaPusher& dma_pusher, Tegra::CDmaPusher& cdma_pusher);
 
     /// Push GPU command entries to be processed
     void SubmitList(Tegra::CommandList&& entries);
 
+    /// Push GPU CDMA command buffer entries to be processed
+    void SubmitCommandBuffer(Tegra::ChCommandHeaderList&& entries);
+
     /// Swap buffers (render frame)
     void SwapBuffers(const Tegra::FramebufferConfig* framebuffer);
 

src/video_core/memory_manager.cpp

@@ -11,6 +11,7 @@
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_base.h"
 
 namespace Tegra {
 
@@ -44,6 +45,12 @@ GPUVAddr MemoryManager::MapAllocate(VAddr cpu_addr, std::size_t size, std::size_
     return Map(cpu_addr, *FindFreeRange(size, align), size);
 }
 
+GPUVAddr MemoryManager::MapAllocate32(VAddr cpu_addr, std::size_t size) {
+    const std::optional<GPUVAddr> gpu_addr = FindFreeRange(size, 1, true);
+    ASSERT(gpu_addr);
+    return Map(cpu_addr, *gpu_addr, size);
+}
+
 void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
     if (!size) {
         return;
@@ -108,7 +115,8 @@ void MemoryManager::SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::s
     page_table[PageEntryIndex(gpu_addr)] = page_entry;
 }
 
-std::optional<GPUVAddr> MemoryManager::FindFreeRange(std::size_t size, std::size_t align) const {
+std::optional<GPUVAddr> MemoryManager::FindFreeRange(std::size_t size, std::size_t align,
+                                                     bool start_32bit_address) const {
     if (!align) {
         align = page_size;
     } else {
@@ -116,7 +124,7 @@ std::optional<GPUVAddr> MemoryManager::FindFreeRange(std::size_t size, std::size
     }
 
     u64 available_size{};
-    GPUVAddr gpu_addr{address_space_start};
+    GPUVAddr gpu_addr{start_32bit_address ? address_space_start_low : address_space_start};
     while (gpu_addr + available_size < address_space_size) {
         if (GetPageEntry(gpu_addr + available_size).IsUnmapped()) {
             available_size += page_size;

src/video_core/memory_manager.h

@@ -116,6 +116,7 @@ public:
 
     [[nodiscard]] GPUVAddr Map(VAddr cpu_addr, GPUVAddr gpu_addr, std::size_t size);
     [[nodiscard]] GPUVAddr MapAllocate(VAddr cpu_addr, std::size_t size, std::size_t align);
+    [[nodiscard]] GPUVAddr MapAllocate32(VAddr cpu_addr, std::size_t size);
     [[nodiscard]] std::optional<GPUVAddr> AllocateFixed(GPUVAddr gpu_addr, std::size_t size);
     [[nodiscard]] GPUVAddr Allocate(std::size_t size, std::size_t align);
     void Unmap(GPUVAddr gpu_addr, std::size_t size);
@@ -124,7 +125,8 @@ private:
     [[nodiscard]] PageEntry GetPageEntry(GPUVAddr gpu_addr) const;
     void SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size = page_size);
     GPUVAddr UpdateRange(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size);
-    [[nodiscard]] std::optional<GPUVAddr> FindFreeRange(std::size_t size, std::size_t align) const;
+    [[nodiscard]] std::optional<GPUVAddr> FindFreeRange(std::size_t size, std::size_t align,
+                                                        bool start_32bit_address = false) const;
 
     void TryLockPage(PageEntry page_entry, std::size_t size);
     void TryUnlockPage(PageEntry page_entry, std::size_t size);
@@ -135,6 +137,7 @@ private:
 
     static constexpr u64 address_space_size = 1ULL << 40;
     static constexpr u64 address_space_start = 1ULL << 32;
+    static constexpr u64 address_space_start_low = 1ULL << 16;
     static constexpr u64 page_bits{16};
     static constexpr u64 page_size{1 << page_bits};
     static constexpr u64 page_mask{page_size - 1};

src/video_core/renderer_opengl/gl_shader_disk_cache.cpp

@@ -317,8 +317,7 @@ std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::Lo
             return std::nullopt;
         }
     }
-
-    return std::move(entries);
+    return entries;
 }
 
 void ShaderDiskCacheOpenGL::InvalidateTransferable() {