"Merge Tagged PR 4836"

common_funcs: Move R_ macros to result.h

.ci/scripts/windows/docker.sh

@@ -47,3 +47,6 @@ python3 .ci/scripts/windows/scan_dll.py package/imageformats/*.dll "package/"
 EXTERNALS_PATH="$(pwd)/build/externals"
 FFMPEG_DLL_PATH="$(find ${EXTERNALS_PATH} -maxdepth 1 -type d | grep ffmpeg)/bin"
 find ${FFMPEG_DLL_PATH} -type f -regex ".*\.dll" -exec cp -v {} package/ ';'
+
+# copy libraries from yuzu.exe path
+find "$(pwd)/build/bin/" -type f -regex ".*\.dll" -exec cp -v {} package/ ';'

CMakeLists.txt

@@ -426,7 +426,7 @@ if(NOT APPLE)
 endif()
 if (NOT LIBUSB_FOUND)
     add_subdirectory(externals/libusb)
-    set(LIBUSB_INCLUDE_DIR "")
+    set(LIBUSB_INCLUDE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/externals/libusb/libusb/libusb")
     set(LIBUSB_LIBRARIES usb)
 endif()
 

externals/CMakeLists.txt

@@ -47,19 +47,21 @@ target_include_directories(unicorn-headers INTERFACE ./unicorn/include)
 
 # SDL2
 if (NOT SDL2_FOUND AND ENABLE_SDL2)
-    # Yuzu itself needs: Events Joystick Haptic Sensor Timers
-    # Yuzu-cmd also needs: Video (depends on Loadso/Dlopen)
-    set(SDL_UNUSED_SUBSYSTEMS
-        Atomic Audio Render Power Threads
-        File CPUinfo Filesystem Locale)
-    foreach(_SUB ${SDL_UNUSED_SUBSYSTEMS})
-      string(TOUPPER ${_SUB} _OPT)
-      option(SDL_${_OPT} "" OFF)
-    endforeach()
+    if (NOT WIN32)
+        # Yuzu itself needs: Events Joystick Haptic Sensor Timers
+        # Yuzu-cmd also needs: Video (depends on Loadso/Dlopen)
+        set(SDL_UNUSED_SUBSYSTEMS
+            Atomic Audio Render Power Threads
+            File CPUinfo Filesystem Locale)
+        foreach(_SUB ${SDL_UNUSED_SUBSYSTEMS})
+          string(TOUPPER ${_SUB} _OPT)
+          option(SDL_${_OPT} "" OFF)
+        endforeach()
 
+        option(HIDAPI "" ON)
+    endif()
     set(SDL_STATIC ON)
     set(SDL_SHARED OFF)
-    option(HIDAPI "" ON)
 
     add_subdirectory(SDL EXCLUDE_FROM_ALL)
     add_library(SDL2 ALIAS SDL2-static)

externals/libusb/CMakeLists.txt

@@ -1,155 +1,239 @@
-# Ensure libusb compiles with UTF-8 encoding on MSVC
-if(MSVC)
-    add_compile_options(/utf-8)
-endif()
+if (MINGW)
+    # The MinGW toolchain for some reason doesn't work with this CMakeLists file after updating to
+    # 1.0.24, so we do it the old-fashioned way for now. We may want to move native Linux toolchains
+    # to here, too (TODO lat9nq?).
 
-add_library(usb STATIC EXCLUDE_FROM_ALL
-    libusb/libusb/core.c
-    libusb/libusb/core.c
-    libusb/libusb/descriptor.c
-    libusb/libusb/hotplug.c
-    libusb/libusb/io.c
-    libusb/libusb/strerror.c
-    libusb/libusb/sync.c
-)
-set_target_properties(usb PROPERTIES VERSION 1.0.23)
-if(WIN32)
-    target_include_directories(usb
-        BEFORE
-        PUBLIC
-          libusb/libusb
+    set(LIBUSB_PREFIX "${CMAKE_CURRENT_BINARY_DIR}/libusb")
+    set(LIBUSB_SRC_DIR "${CMAKE_CURRENT_SOURCE_DIR}/libusb")
+    # Workarounds for MSYS/MinGW
+    if (MSYS)
+        # CMake on Windows passes `C:/`, but we need `/C/` or `/c/` to use `configure`
+        string(REPLACE ":/" "/" LIBUSB_SRC_DIR "${LIBUSB_SRC_DIR}")
+        set(LIBUSB_SRC_DIR "/${LIBUSB_SRC_DIR}")
 
-        PRIVATE
-          "${CMAKE_CURRENT_BINARY_DIR}"
-    )
-
-    if (NOT MINGW)
-        target_include_directories(usb BEFORE PRIVATE libusb/msvc)
+        # And now that we are using /C/ for srcdir but everything else is using C:/, we need to
+        # compile everything in the source directory, else `configure` won't think the build
+        # environment is sane.
+        set(LIBUSB_PREFIX "${LIBUSB_SRC_DIR}")
     endif()
 
-    # Works around other libraries providing their own definition of USB GUIDs (e.g. SDL2)
-    target_compile_definitions(usb PRIVATE "-DGUID_DEVINTERFACE_USB_DEVICE=(GUID){ 0xA5DCBF10, 0x6530, 0x11D2, {0x90, 0x1F, 0x00, 0xC0, 0x4F, 0xB9, 0x51, 0xED}}")
-else()
-target_include_directories(usb
-    # turns out other projects also have "config.h", so make sure the
-    # LibUSB one comes first
-    BEFORE
+    set(LIBUSB_CONFIGURE "${LIBUSB_SRC_DIR}/configure")
+    set(LIBUSB_MAKEFILE "${LIBUSB_PREFIX}/Makefile")
+    set(LIBUSB_LIBRARY "${LIBUSB_PREFIX}/libusb/.libs/libusb-1.0.dll.a")
+    set(LIBUSB_SHARED_LIBRARY "${LIBUSB_PREFIX}/libusb/.libs/libusb-1.0.dll")
+    set(LIBUSB_SHARED_LIBRARY_DEST "${CMAKE_BINARY_DIR}/bin/libusb-1.0.dll")
 
-    PUBLIC
-      libusb/libusb
+    # Causes "externals/libusb/libusb/libusb/os/windows_winusb.c:1427:2: error: conversion to non-scalar type requested", so cannot statically link it for now.
+    # set(LIBUSB_CFLAGS "-DGUID_DEVINTERFACE_USB_DEVICE=\\(GUID\\){0xA5DCBF10,0x6530,0x11D2,{0x90,0x1F,0x00,0xC0,0x4F,0xB9,0x51,0xED}}")
 
-    PRIVATE
-      "${CMAKE_CURRENT_BINARY_DIR}"
-)
-endif()
+    make_directory("${LIBUSB_PREFIX}")
 
-if(WIN32 OR CYGWIN)
-    target_sources(usb PRIVATE
-      libusb/libusb/os/threads_windows.c
-      libusb/libusb/os/windows_winusb.c
-      libusb/libusb/os/windows_usbdk.c
-      libusb/libusb/os/windows_common.c
+    add_custom_command(
+        OUTPUT
+            "${LIBUSB_LIBRARY}"
+        COMMAND
+            make
+        WORKING_DIRECTORY
+            "${LIBUSB_PREFIX}"
     )
-    set(OS_WINDOWS TRUE)
-elseif(APPLE)
-    target_sources(usb PRIVATE
-        libusb/libusb/os/darwin_usb.c
+
+    # We may use this path for other GNU toolchains, so put all of the MinGW-specific stuff here
+    if (MINGW)
+        set(LIBUSB_CONFIGURE_ARGS --host=x86_64-w64-mingw32 --build=x86_64-windows)
+    endif()
+
+    add_custom_command(
+        OUTPUT
+            "${LIBUSB_MAKEFILE}"
+        COMMAND
+            # /bin/env
+            #     CFLAGS="${LIBUSB_CFLAGS}"
+            /bin/sh "${LIBUSB_CONFIGURE}"
+                ${LIBUSB_CONFIGURE_ARGS}
+                --srcdir="${LIBUSB_SRC_DIR}"
+        WORKING_DIRECTORY
+            "${LIBUSB_PREFIX}"
     )
-    find_library(COREFOUNDATION_LIBRARY CoreFoundation)
-    find_library(IOKIT_LIBRARY IOKit)
-    find_library(OBJC_LIBRARY objc)
-    target_link_libraries(usb PRIVATE
-        ${COREFOUNDATION_LIBRARY}
-        ${IOKIT_LIBRARY}
-        ${OBJC_LIBRARY}
+
+    add_custom_command(
+        OUTPUT
+            "${LIBUSB_CONFIGURE}"
+        COMMAND
+            /bin/sh "${LIBUSB_SRC_DIR}/bootstrap.sh"
+        WORKING_DIRECTORY
+            "${LIBUSB_SRC_DIR}"
     )
-    set(OS_DARWIN TRUE)
-elseif(ANDROID)
-    target_sources(usb PRIVATE
-        libusb/libusb/os/linux_usbfs.c
-        libusb/libusb/os/linux_netlink.c
+
+    add_custom_command(
+        OUTPUT
+            "${LIBUSB_SHARED_LIBRARY_DEST}"
+        COMMAND
+            /bin/cp "${LIBUSB_SHARED_LIBRARY}" "${LIBUSB_SHARED_LIBRARY_DEST}"
     )
-    find_library(LOG_LIBRARY log)
-    target_link_libraries(usb PRIVATE ${LOG_LIBRARY})
-    set(OS_LINUX TRUE)
-elseif(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
-    target_sources(usb PRIVATE
-        libusb/libusb/os/linux_usbfs.c
+
+    add_custom_target(usb-bootstrap ALL DEPENDS "${LIBUSB_CONFIGURE}")
+    add_custom_target(usb-configure ALL DEPENDS "${LIBUSB_MAKEFILE}" usb-bootstrap)
+    add_custom_target(usb-build ALL DEPENDS "${LIBUSB_LIBRARY}" usb-configure)
+    # Workaround since static linking didn't work out -- We need to copy the DLL to the bin directory
+    add_custom_target(usb-copy ALL DEPENDS "${LIBUSB_SHARED_LIBRARY_DEST}" usb-build)
+
+    # Make `usb` alias to LIBUSB_LIBRARY
+    add_library(usb INTERFACE)
+    target_link_libraries(usb INTERFACE "${LIBUSB_LIBRARY}")
+else() # MINGW
+    # Ensure libusb compiles with UTF-8 encoding on MSVC
+    if(MSVC)
+        add_compile_options(/utf-8)
+    endif()
+
+    add_library(usb STATIC EXCLUDE_FROM_ALL
+        libusb/libusb/core.c
+        libusb/libusb/core.c
+        libusb/libusb/descriptor.c
+        libusb/libusb/hotplug.c
+        libusb/libusb/io.c
+        libusb/libusb/strerror.c
+        libusb/libusb/sync.c
     )
-    find_package(Libudev)
-    if(LIBUDEV_FOUND)
-        target_sources(usb PRIVATE
-            libusb/libusb/os/linux_udev.c
+    set_target_properties(usb PROPERTIES VERSION 1.0.24)
+    if(WIN32)
+        target_include_directories(usb
+            BEFORE
+            PUBLIC
+              libusb/libusb
+
+            PRIVATE
+              "${CMAKE_CURRENT_BINARY_DIR}"
         )
-        target_link_libraries(usb PRIVATE "${LIBUDEV_LIBRARIES}")
-        target_include_directories(usb PRIVATE "${LIBUDEV_INCLUDE_DIR}")
-        set(HAVE_LIBUDEV TRUE)
-        set(USE_UDEV TRUE)
+
+        if (NOT MINGW)
+            target_include_directories(usb BEFORE PRIVATE libusb/msvc)
+        endif()
+
+        # Works around other libraries providing their own definition of USB GUIDs (e.g. SDL2)
+        target_compile_definitions(usb PRIVATE "-DGUID_DEVINTERFACE_USB_DEVICE=(GUID){ 0xA5DCBF10, 0x6530, 0x11D2, {0x90, 0x1F, 0x00, 0xC0, 0x4F, 0xB9, 0x51, 0xED}}")
     else()
+        target_include_directories(usb
+            # turns out other projects also have "config.h", so make sure the
+            # LibUSB one comes first
+            BEFORE
+
+            PUBLIC
+              libusb/libusb
+
+            PRIVATE
+              "${CMAKE_CURRENT_BINARY_DIR}"
+        )
+    endif()
+
+    if(WIN32 OR CYGWIN)
         target_sources(usb PRIVATE
+          libusb/libusb/os/threads_windows.c
+          libusb/libusb/os/windows_winusb.c
+          libusb/libusb/os/windows_usbdk.c
+          libusb/libusb/os/windows_common.c
+        )
+        set(OS_WINDOWS TRUE)
+    elseif(APPLE)
+        target_sources(usb PRIVATE
+            libusb/libusb/os/darwin_usb.c
+        )
+        find_library(COREFOUNDATION_LIBRARY CoreFoundation)
+        find_library(IOKIT_LIBRARY IOKit)
+        find_library(OBJC_LIBRARY objc)
+        target_link_libraries(usb PRIVATE
+            ${COREFOUNDATION_LIBRARY}
+            ${IOKIT_LIBRARY}
+            ${OBJC_LIBRARY}
+        )
+        set(OS_DARWIN TRUE)
+    elseif(ANDROID)
+        target_sources(usb PRIVATE
+            libusb/libusb/os/linux_usbfs.c
             libusb/libusb/os/linux_netlink.c
         )
+        find_library(LOG_LIBRARY log)
+        target_link_libraries(usb PRIVATE ${LOG_LIBRARY})
+        set(OS_LINUX TRUE)
+    elseif(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
+        target_sources(usb PRIVATE
+            libusb/libusb/os/linux_usbfs.c
+        )
+        find_package(Libudev)
+        if(LIBUDEV_FOUND)
+            target_sources(usb PRIVATE
+                libusb/libusb/os/linux_udev.c
+            )
+            target_link_libraries(usb PRIVATE "${LIBUDEV_LIBRARIES}")
+            target_include_directories(usb PRIVATE "${LIBUDEV_INCLUDE_DIR}")
+            set(HAVE_LIBUDEV TRUE)
+            set(USE_UDEV TRUE)
+        else()
+            target_sources(usb PRIVATE
+                libusb/libusb/os/linux_netlink.c
+            )
+        endif()
+        set(OS_LINUX TRUE)
+    elseif(${CMAKE_SYSTEM_NAME} MATCHES "NetBSD")
+        target_sources(usb PRIVATE
+            libusb/libusb/os/netbsd_usb.c
+        )
+        set(OS_NETBSD TRUE)
+    elseif(${CMAKE_SYSTEM_NAME} MATCHES "OpenBSD")
+        target_sources(usb PRIVATE
+            libusb/libusb/os/openbsd_usb.c
+        )
+        set(OS_OPENBSD TRUE)
     endif()
-    set(OS_LINUX TRUE)
-elseif(${CMAKE_SYSTEM_NAME} MATCHES "NetBSD")
-    target_sources(usb PRIVATE
-        libusb/libusb/os/netbsd_usb.c
-    )
-    set(OS_NETBSD TRUE)
-elseif(${CMAKE_SYSTEM_NAME} MATCHES "OpenBSD")
-    target_sources(usb PRIVATE
-        libusb/libusb/os/openbsd_usb.c
-    )
-    set(OS_OPENBSD TRUE)
-endif()
 
-if(UNIX)
-    target_sources(usb PRIVATE
-        libusb/libusb/os/events_posix.c
-        libusb/libusb/os/threads_posix.c
-    )
-    find_package(Threads REQUIRED)
-    if(THREADS_HAVE_PTHREAD_ARG)
-      target_compile_options(usb PUBLIC "-pthread")
+    if(UNIX)
+        target_sources(usb PRIVATE
+            libusb/libusb/os/events_posix.c
+            libusb/libusb/os/threads_posix.c
+        )
+        find_package(Threads REQUIRED)
+        if(THREADS_HAVE_PTHREAD_ARG)
+          target_compile_options(usb PUBLIC "-pthread")
+        endif()
+        if(CMAKE_THREAD_LIBS_INIT)
+          target_link_libraries(usb PRIVATE "${CMAKE_THREAD_LIBS_INIT}")
+        endif()
+        set(THREADS_POSIX TRUE)
+    elseif(WIN32)
+        target_sources(usb PRIVATE
+            libusb/libusb/os/events_windows.c
+            libusb/libusb/os/threads_windows.c
+        )
     endif()
-    if(CMAKE_THREAD_LIBS_INIT)
-      target_link_libraries(usb PRIVATE "${CMAKE_THREAD_LIBS_INIT}")
+
+    include(CheckFunctionExists)
+    include(CheckIncludeFiles)
+    include(CheckTypeSize)
+    check_include_files(asm/types.h HAVE_ASM_TYPES_H)
+    check_function_exists(gettimeofday HAVE_GETTIMEOFDAY)
+    check_include_files(linux/filter.h HAVE_LINUX_FILTER_H)
+    check_include_files(linux/netlink.h HAVE_LINUX_NETLINK_H)
+    check_include_files(poll.h HAVE_POLL_H)
+    check_include_files(signal.h HAVE_SIGNAL_H)
+    check_include_files(strings.h HAVE_STRINGS_H)
+    check_type_size("struct timespec" STRUCT_TIMESPEC)
+    check_function_exists(syslog HAVE_SYSLOG_FUNC)
+    check_include_files(syslog.h HAVE_SYSLOG_H)
+    check_include_files(sys/socket.h HAVE_SYS_SOCKET_H)
+    check_include_files(sys/time.h HAVE_SYS_TIME_H)
+    check_include_files(sys/types.h HAVE_SYS_TYPES_H)
+
+    set(CMAKE_EXTRA_INCLUDE_FILES poll.h)
+    check_type_size("nfds_t" nfds_t)
+    unset(CMAKE_EXTRA_INCLUDE_FILES)
+    if(HAVE_NFDS_T)
+        set(POLL_NFDS_TYPE "nfds_t")
+    else()
+        set(POLL_NFDS_TYPE "unsigned int")
     endif()
-    set(THREADS_POSIX TRUE)
-elseif(WIN32)
-    target_sources(usb PRIVATE
-        libusb/libusb/os/events_windows.c
-        libusb/libusb/os/threads_windows.c
-    )
-endif()
 
-include(CheckFunctionExists)
-include(CheckIncludeFiles)
-include(CheckTypeSize)
-check_include_files(asm/types.h HAVE_ASM_TYPES_H)
-check_function_exists(gettimeofday HAVE_GETTIMEOFDAY)
-check_include_files(linux/filter.h HAVE_LINUX_FILTER_H)
-check_include_files(linux/netlink.h HAVE_LINUX_NETLINK_H)
-check_include_files(poll.h HAVE_POLL_H)
-check_include_files(signal.h HAVE_SIGNAL_H)
-check_include_files(strings.h HAVE_STRINGS_H)
-check_type_size("struct timespec" STRUCT_TIMESPEC)
-check_function_exists(syslog HAVE_SYSLOG_FUNC)
-check_include_files(syslog.h HAVE_SYSLOG_H)
-check_include_files(sys/socket.h HAVE_SYS_SOCKET_H)
-check_include_files(sys/time.h HAVE_SYS_TIME_H)
-check_include_files(sys/types.h HAVE_SYS_TYPES_H)
-
-set(CMAKE_EXTRA_INCLUDE_FILES poll.h)
-check_type_size("nfds_t" nfds_t)
-unset(CMAKE_EXTRA_INCLUDE_FILES)
-if(HAVE_NFDS_T)
-    set(POLL_NFDS_TYPE "nfds_t")
-else()
-    set(POLL_NFDS_TYPE "unsigned int")
-endif()
-
-check_include_files(sys/timerfd.h USBI_TIMERFD_AVAILABLE)
+    check_include_files(sys/timerfd.h USBI_TIMERFD_AVAILABLE)
 
 
-configure_file(config.h.in config.h)
+    configure_file(config.h.in config.h)
+endif() # MINGW

src/common/common_funcs.h

@@ -97,17 +97,6 @@ __declspec(dllimport) void __stdcall DebugBreak(void);
         return static_cast<T>(key) == 0;                                                           \
     }
 
-/// Evaluates a boolean expression, and returns a result unless that expression is true.
-#define R_UNLESS(expr, res)                                                                        \
-    {                                                                                              \
-        if (!(expr)) {                                                                             \
-            if (res.IsError()) {                                                                   \
-                LOG_ERROR(Kernel, "Failed with result: {}", res.raw);                              \
-            }                                                                                      \
-            return res;                                                                            \
-        }                                                                                          \
-    }
-
 #define YUZU_NON_COPYABLE(cls)                                                                     \
     cls(const cls&) = delete;                                                                      \
     cls& operator=(const cls&) = delete
@@ -116,20 +105,6 @@ __declspec(dllimport) void __stdcall DebugBreak(void);
     cls(cls&&) = delete;                                                                           \
     cls& operator=(cls&&) = delete
 
-#define R_SUCCEEDED(res) (res.IsSuccess())
-
-/// Evaluates an expression that returns a result, and returns the result if it would fail.
-#define R_TRY(res_expr)                                                                            \
-    {                                                                                              \
-        const auto _tmp_r_try_rc = (res_expr);                                                     \
-        if (_tmp_r_try_rc.IsError()) {                                                             \
-            return _tmp_r_try_rc;                                                                  \
-        }                                                                                          \
-    }
-
-/// Evaluates a boolean expression, and succeeds if that expression is true.
-#define R_SUCCEED_IF(expr) R_UNLESS(!(expr), RESULT_SUCCESS)
-
 namespace Common {
 
 [[nodiscard]] constexpr u32 MakeMagic(char a, char b, char c, char d) {

src/common/fs/fs_util.cpp

@@ -2,6 +2,8 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <algorithm>
+
 #include "common/fs/fs_util.h"
 
 namespace Common::FS {
@@ -10,4 +12,16 @@ std::u8string ToU8String(std::string_view utf8_string) {
     return std::u8string{utf8_string.begin(), utf8_string.end()};
 }
 
+std::u8string BufferToU8String(std::span<const u8> buffer) {
+    return std::u8string{buffer.begin(), std::ranges::find(buffer, u8{0})};
+}
+
+std::string ToUTF8String(std::u8string_view u8_string) {
+    return std::string{u8_string.begin(), u8_string.end()};
+}
+
+std::string PathToUTF8String(const std::filesystem::path& path) {
+    return ToUTF8String(path.u8string());
+}
+
 } // namespace Common::FS

src/common/fs/fs_util.h

@@ -5,9 +5,13 @@
 #pragma once
 
 #include <concepts>
+#include <filesystem>
+#include <span>
 #include <string>
 #include <string_view>
 
+#include "common/common_types.h"
+
 namespace Common::FS {
 
 template <typename T>
@@ -22,4 +26,33 @@ concept IsChar = std::same_as<T, char>;
  */
 [[nodiscard]] std::u8string ToU8String(std::string_view utf8_string);
 
+/**
+ * Converts a buffer of bytes to a UTF8-encoded std::u8string.
+ * This converts from the start of the buffer until the first encountered null-terminator.
+ * If no null-terminator is found, this converts the entire buffer instead.
+ *
+ * @param buffer Buffer of bytes
+ *
+ * @returns UTF-8 encoded std::u8string.
+ */
+[[nodiscard]] std::u8string BufferToU8String(std::span<const u8> buffer);
+
+/**
+ * Converts a std::u8string or std::u8string_view to a UTF-8 encoded std::string.
+ *
+ * @param u8_string UTF-8 encoded u8string
+ *
+ * @returns UTF-8 encoded std::string.
+ */
+[[nodiscard]] std::string ToUTF8String(std::u8string_view u8_string);
+
+/**
+ * Converts a filesystem path to a UTF-8 encoded std::string.
+ *
+ * @param path Filesystem path
+ *
+ * @returns UTF-8 encoded std::string.
+ */
+[[nodiscard]] std::string PathToUTF8String(const std::filesystem::path& path);
+
 } // namespace Common::FS

src/common/fs/path_util.cpp

@@ -129,12 +129,6 @@ private:
     std::unordered_map<YuzuPath, fs::path> yuzu_paths;
 };
 
-std::string PathToUTF8String(const fs::path& path) {
-    const auto utf8_string = path.u8string();
-
-    return std::string{utf8_string.begin(), utf8_string.end()};
-}
-
 bool ValidatePath(const fs::path& path) {
     if (path.empty()) {
         LOG_ERROR(Common_Filesystem, "Input path is empty, path={}", PathToUTF8String(path));

src/common/fs/path_util.h

@@ -25,15 +25,6 @@ enum class YuzuPath {
     ShaderDir,      // Where shaders are stored.
 };
 
-/**
- * Converts a filesystem path to a UTF-8 encoded std::string.
- *
- * @param path Filesystem path
- *
- * @returns UTF-8 encoded std::string.
- */
-[[nodiscard]] std::string PathToUTF8String(const std::filesystem::path& path);
-
 /**
  * Validates a given path.
  *

src/core/hle/result.h

@@ -358,3 +358,28 @@ ResultVal<std::remove_reference_t<Arg>> MakeResult(Arg&& arg) {
             return CONCAT2(check_result_L, __LINE__);                                              \
         }                                                                                          \
     } while (false)
+
+#define R_SUCCEEDED(res) (res.IsSuccess())
+
+/// Evaluates a boolean expression, and succeeds if that expression is true.
+#define R_SUCCEED_IF(expr) R_UNLESS(!(expr), RESULT_SUCCESS)
+
+/// Evaluates a boolean expression, and returns a result unless that expression is true.
+#define R_UNLESS(expr, res)                                                                        \
+    {                                                                                              \
+        if (!(expr)) {                                                                             \
+            if (res.IsError()) {                                                                   \
+                LOG_ERROR(Kernel, "Failed with result: {}", res.raw);                              \
+            }                                                                                      \
+            return res;                                                                            \
+        }                                                                                          \
+    }
+
+/// Evaluates an expression that returns a result, and returns the result if it would fail.
+#define R_TRY(res_expr)                                                                            \
+    {                                                                                              \
+        const auto _tmp_r_try_rc = (res_expr);                                                     \
+        if (_tmp_r_try_rc.IsError()) {                                                             \
+            return _tmp_r_try_rc;                                                                  \
+        }                                                                                          \
+    }

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

@@ -68,6 +68,7 @@ void nvhost_nvdec::OnOpen(DeviceFD fd) {}
 void nvhost_nvdec::OnClose(DeviceFD fd) {
     LOG_INFO(Service_NVDRV, "NVDEC video stream ended");
     system.GPU().ClearCdmaInstance();
+    system.GPU().MemoryManager().InvalidateQueuedCaches();
 }
 
 } // namespace Service::Nvidia::Devices

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

@@ -198,7 +198,13 @@ NvResult nvhost_nvdec_common::UnmapBuffer(const std::vector<u8>& input, std::vec
             return NvResult::InvalidState;
         }
         if (const auto size{RemoveBufferMap(object->dma_map_addr)}; size) {
-            gpu.MemoryManager().Unmap(object->dma_map_addr, *size);
+            if (vic_device) {
+                // UnmapVicFrame defers texture_cache invalidation of the frame address until
+                // the stream is over
+                gpu.MemoryManager().UnmapVicFrame(object->dma_map_addr, *size);
+            } else {
+                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,

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

@@ -160,6 +160,7 @@ protected:
 
     s32_le nvmap_fd{};
     u32_le submit_timeout{};
+    bool vic_device{};
     std::shared_ptr<nvmap> nvmap_dev;
     SyncpointManager& syncpoint_manager;
     std::array<u32, MaxSyncPoints> device_syncpoints{};

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

@@ -12,8 +12,9 @@
 namespace Service::Nvidia::Devices {
 nvhost_vic::nvhost_vic(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_,
                        SyncpointManager& syncpoint_manager_)
-    : nvhost_nvdec_common{system_, std::move(nvmap_dev_), syncpoint_manager_} {}
-
+    : nvhost_nvdec_common(system_, std::move(nvmap_dev_), syncpoint_manager_) {
+    vic_device = true;
+}
 nvhost_vic::~nvhost_vic() = default;
 
 NvResult nvhost_vic::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input,

src/input_common/sdl/sdl_impl.cpp

@@ -18,16 +18,6 @@
 #include <utility>
 #include <vector>
 
-// Ignore -Wimplicit-fallthrough due to https://github.com/libsdl-org/SDL/issues/4307
-#ifdef __clang__
-#pragma clang diagnostic push
-#pragma clang diagnostic ignored "-Wimplicit-fallthrough"
-#endif
-#include <SDL.h>
-#ifdef __clang__
-#pragma clang diagnostic pop
-#endif
-
 #include "common/logging/log.h"
 #include "common/math_util.h"
 #include "common/param_package.h"
@@ -214,6 +204,40 @@ public:
         sdl_controller.reset(controller);
     }
 
+    bool IsJoyconLeft() const {
+        return std::strstr(GetControllerName().c_str(), "Joy-Con Left") != nullptr;
+    }
+
+    bool IsJoyconRight() const {
+        return std::strstr(GetControllerName().c_str(), "Joy-Con Right") != nullptr;
+    }
+
+    std::string GetControllerName() const {
+        if (sdl_controller) {
+            switch (SDL_GameControllerGetType(sdl_controller.get())) {
+            case SDL_CONTROLLER_TYPE_XBOX360:
+                return "XBox 360 Controller";
+            case SDL_CONTROLLER_TYPE_XBOXONE:
+                return "XBox One Controller";
+            default:
+                break;
+            }
+            const auto name = SDL_GameControllerName(sdl_controller.get());
+            if (name) {
+                return name;
+            }
+        }
+
+        if (sdl_joystick) {
+            const auto name = SDL_JoystickName(sdl_joystick.get());
+            if (name) {
+                return name;
+            }
+        }
+
+        return "Unknown";
+    }
+
 private:
     struct State {
         std::unordered_map<int, bool> buttons;
@@ -858,23 +882,42 @@ SDLState::~SDLState() {
 std::vector<Common::ParamPackage> SDLState::GetInputDevices() {
     std::scoped_lock lock(joystick_map_mutex);
     std::vector<Common::ParamPackage> devices;
+    std::unordered_map<int, std::shared_ptr<SDLJoystick>> joycon_pairs;
     for (const auto& [key, value] : joystick_map) {
         for (const auto& joystick : value) {
-            if (auto* const controller = joystick->GetSDLGameController()) {
+            if (!joystick->GetSDLJoystick()) {
+                continue;
+            }
+            std::string name =
+                fmt::format("{} {}", joystick->GetControllerName(), joystick->GetPort());
+            devices.emplace_back(Common::ParamPackage{
+                {"class", "sdl"},
+                {"display", std::move(name)},
+                {"guid", joystick->GetGUID()},
+                {"port", std::to_string(joystick->GetPort())},
+            });
+            if (joystick->IsJoyconLeft()) {
+                joycon_pairs.insert_or_assign(joystick->GetPort(), joystick);
+            }
+        }
+    }
+
+    // Add dual controllers
+    for (const auto& [key, value] : joystick_map) {
+        for (const auto& joystick : value) {
+            if (joystick->IsJoyconRight()) {
+                if (!joycon_pairs.contains(joystick->GetPort())) {
+                    continue;
+                }
+                const auto joystick2 = joycon_pairs.at(joystick->GetPort());
+
                 std::string name =
-                    fmt::format("{} {}", GetControllerName(controller), joystick->GetPort());
-                devices.emplace_back(Common::ParamPackage{
-                    {"class", "sdl"},
-                    {"display", std::move(name)},
-                    {"guid", joystick->GetGUID()},
-                    {"port", std::to_string(joystick->GetPort())},
-                });
-            } else if (auto* const joy = joystick->GetSDLJoystick()) {
-                std::string name = fmt::format("{} {}", SDL_JoystickName(joy), joystick->GetPort());
+                    fmt::format("{} {}", "Nintendo Dual Joy-Con", joystick->GetPort());
                 devices.emplace_back(Common::ParamPackage{
                     {"class", "sdl"},
                     {"display", std::move(name)},
                     {"guid", joystick->GetGUID()},
+                    {"guid2", joystick2->GetGUID()},
                     {"port", std::to_string(joystick->GetPort())},
                 });
             }
@@ -883,17 +926,6 @@ std::vector<Common::ParamPackage> SDLState::GetInputDevices() {
     return devices;
 }
 
-std::string SDLState::GetControllerName(SDL_GameController* controller) const {
-    switch (SDL_GameControllerGetType(controller)) {
-    case SDL_CONTROLLER_TYPE_XBOX360:
-        return "XBox 360 Controller";
-    case SDL_CONTROLLER_TYPE_XBOXONE:
-        return "XBox One Controller";
-    default:
-        return SDL_GameControllerName(controller);
-    }
-}
-
 namespace {
 Common::ParamPackage BuildAnalogParamPackageForButton(int port, std::string guid, s32 axis,
                                                       float value = 0.1f) {
@@ -1073,24 +1105,48 @@ ButtonMapping SDLState::GetButtonMappingForDevice(const Common::ParamPackage& pa
         return {};
     }
     const auto joystick = GetSDLJoystickByGUID(params.Get("guid", ""), params.Get("port", 0));
+
     auto* controller = joystick->GetSDLGameController();
     if (controller == nullptr) {
         return {};
     }
 
-    const bool invert =
-        SDL_GameControllerGetType(controller) != SDL_CONTROLLER_TYPE_NINTENDO_SWITCH_PRO;
-
     // This list is missing ZL/ZR since those are not considered buttons in SDL GameController.
     // We will add those afterwards
     // This list also excludes Screenshot since theres not really a mapping for that
-    using ButtonBindings =
-        std::array<std::pair<Settings::NativeButton::Values, SDL_GameControllerButton>, 17>;
-    const ButtonBindings switch_to_sdl_button{{
-        {Settings::NativeButton::A, invert ? SDL_CONTROLLER_BUTTON_B : SDL_CONTROLLER_BUTTON_A},
-        {Settings::NativeButton::B, invert ? SDL_CONTROLLER_BUTTON_A : SDL_CONTROLLER_BUTTON_B},
-        {Settings::NativeButton::X, invert ? SDL_CONTROLLER_BUTTON_Y : SDL_CONTROLLER_BUTTON_X},
-        {Settings::NativeButton::Y, invert ? SDL_CONTROLLER_BUTTON_X : SDL_CONTROLLER_BUTTON_Y},
+    ButtonBindings switch_to_sdl_button;
+
+    if (SDL_GameControllerGetType(controller) == SDL_CONTROLLER_TYPE_NINTENDO_SWITCH_PRO) {
+        switch_to_sdl_button = GetNintendoButtonBinding(joystick);
+    } else {
+        switch_to_sdl_button = GetDefaultButtonBinding();
+    }
+
+    // Add the missing bindings for ZL/ZR
+    static constexpr ZButtonBindings switch_to_sdl_axis{{
+        {Settings::NativeButton::ZL, SDL_CONTROLLER_AXIS_TRIGGERLEFT},
+        {Settings::NativeButton::ZR, SDL_CONTROLLER_AXIS_TRIGGERRIGHT},
+    }};
+
+    // Parameters contain two joysticks return dual
+    if (params.Has("guid2")) {
+        const auto joystick2 = GetSDLJoystickByGUID(params.Get("guid2", ""), params.Get("port", 0));
+
+        if (joystick2->GetSDLGameController() != nullptr) {
+            return GetDualControllerMapping(joystick, joystick2, switch_to_sdl_button,
+                                            switch_to_sdl_axis);
+        }
+    }
+
+    return GetSingleControllerMapping(joystick, switch_to_sdl_button, switch_to_sdl_axis);
+}
+
+ButtonBindings SDLState::GetDefaultButtonBinding() const {
+    return {
+        std::pair{Settings::NativeButton::A, SDL_CONTROLLER_BUTTON_B},
+        {Settings::NativeButton::B, SDL_CONTROLLER_BUTTON_A},
+        {Settings::NativeButton::X, SDL_CONTROLLER_BUTTON_Y},
+        {Settings::NativeButton::Y, SDL_CONTROLLER_BUTTON_X},
         {Settings::NativeButton::LStick, SDL_CONTROLLER_BUTTON_LEFTSTICK},
         {Settings::NativeButton::RStick, SDL_CONTROLLER_BUTTON_RIGHTSTICK},
         {Settings::NativeButton::L, SDL_CONTROLLER_BUTTON_LEFTSHOULDER},
@@ -1104,18 +1160,51 @@ ButtonMapping SDLState::GetButtonMappingForDevice(const Common::ParamPackage& pa
         {Settings::NativeButton::SL, SDL_CONTROLLER_BUTTON_LEFTSHOULDER},
         {Settings::NativeButton::SR, SDL_CONTROLLER_BUTTON_RIGHTSHOULDER},
         {Settings::NativeButton::Home, SDL_CONTROLLER_BUTTON_GUIDE},
-    }};
+    };
+}
 
-    // Add the missing bindings for ZL/ZR
-    using ZBindings =
-        std::array<std::pair<Settings::NativeButton::Values, SDL_GameControllerAxis>, 2>;
-    static constexpr ZBindings switch_to_sdl_axis{{
-        {Settings::NativeButton::ZL, SDL_CONTROLLER_AXIS_TRIGGERLEFT},
-        {Settings::NativeButton::ZR, SDL_CONTROLLER_AXIS_TRIGGERRIGHT},
-    }};
+ButtonBindings SDLState::GetNintendoButtonBinding(
+    const std::shared_ptr<SDLJoystick>& joystick) const {
+    // Default SL/SR mapping for pro controllers
+    auto sl_button = SDL_CONTROLLER_BUTTON_LEFTSHOULDER;
+    auto sr_button = SDL_CONTROLLER_BUTTON_RIGHTSHOULDER;
 
+    if (joystick->IsJoyconLeft()) {
+        sl_button = SDL_CONTROLLER_BUTTON_PADDLE2;
+        sr_button = SDL_CONTROLLER_BUTTON_PADDLE4;
+    }
+    if (joystick->IsJoyconRight()) {
+        sl_button = SDL_CONTROLLER_BUTTON_PADDLE3;
+        sr_button = SDL_CONTROLLER_BUTTON_PADDLE1;
+    }
+
+    return {
+        std::pair{Settings::NativeButton::A, SDL_CONTROLLER_BUTTON_A},
+        {Settings::NativeButton::B, SDL_CONTROLLER_BUTTON_B},
+        {Settings::NativeButton::X, SDL_CONTROLLER_BUTTON_X},
+        {Settings::NativeButton::Y, SDL_CONTROLLER_BUTTON_Y},
+        {Settings::NativeButton::LStick, SDL_CONTROLLER_BUTTON_LEFTSTICK},
+        {Settings::NativeButton::RStick, SDL_CONTROLLER_BUTTON_RIGHTSTICK},
+        {Settings::NativeButton::L, SDL_CONTROLLER_BUTTON_LEFTSHOULDER},
+        {Settings::NativeButton::R, SDL_CONTROLLER_BUTTON_RIGHTSHOULDER},
+        {Settings::NativeButton::Plus, SDL_CONTROLLER_BUTTON_START},
+        {Settings::NativeButton::Minus, SDL_CONTROLLER_BUTTON_BACK},
+        {Settings::NativeButton::DLeft, SDL_CONTROLLER_BUTTON_DPAD_LEFT},
+        {Settings::NativeButton::DUp, SDL_CONTROLLER_BUTTON_DPAD_UP},
+        {Settings::NativeButton::DRight, SDL_CONTROLLER_BUTTON_DPAD_RIGHT},
+        {Settings::NativeButton::DDown, SDL_CONTROLLER_BUTTON_DPAD_DOWN},
+        {Settings::NativeButton::SL, sl_button},
+        {Settings::NativeButton::SR, sr_button},
+        {Settings::NativeButton::Home, SDL_CONTROLLER_BUTTON_GUIDE},
+    };
+}
+
+ButtonMapping SDLState::GetSingleControllerMapping(
+    const std::shared_ptr<SDLJoystick>& joystick, const ButtonBindings& switch_to_sdl_button,
+    const ZButtonBindings& switch_to_sdl_axis) const {
     ButtonMapping mapping;
     mapping.reserve(switch_to_sdl_button.size() + switch_to_sdl_axis.size());
+    auto* controller = joystick->GetSDLGameController();
 
     for (const auto& [switch_button, sdl_button] : switch_to_sdl_button) {
         const auto& binding = SDL_GameControllerGetBindForButton(controller, sdl_button);
@@ -1133,11 +1222,68 @@ ButtonMapping SDLState::GetButtonMappingForDevice(const Common::ParamPackage& pa
     return mapping;
 }
 
+ButtonMapping SDLState::GetDualControllerMapping(const std::shared_ptr<SDLJoystick>& joystick,
+                                                 const std::shared_ptr<SDLJoystick>& joystick2,
+                                                 const ButtonBindings& switch_to_sdl_button,
+                                                 const ZButtonBindings& switch_to_sdl_axis) const {
+    ButtonMapping mapping;
+    mapping.reserve(switch_to_sdl_button.size() + switch_to_sdl_axis.size());
+    auto* controller = joystick->GetSDLGameController();
+    auto* controller2 = joystick2->GetSDLGameController();
+
+    for (const auto& [switch_button, sdl_button] : switch_to_sdl_button) {
+        if (IsButtonOnLeftSide(switch_button)) {
+            const auto& binding = SDL_GameControllerGetBindForButton(controller2, sdl_button);
+            mapping.insert_or_assign(
+                switch_button,
+                BuildParamPackageForBinding(joystick2->GetPort(), joystick2->GetGUID(), binding));
+            continue;
+        }
+        const auto& binding = SDL_GameControllerGetBindForButton(controller, sdl_button);
+        mapping.insert_or_assign(
+            switch_button,
+            BuildParamPackageForBinding(joystick->GetPort(), joystick->GetGUID(), binding));
+    }
+    for (const auto& [switch_button, sdl_axis] : switch_to_sdl_axis) {
+        if (IsButtonOnLeftSide(switch_button)) {
+            const auto& binding = SDL_GameControllerGetBindForAxis(controller2, sdl_axis);
+            mapping.insert_or_assign(
+                switch_button,
+                BuildParamPackageForBinding(joystick2->GetPort(), joystick2->GetGUID(), binding));
+            continue;
+        }
+        const auto& binding = SDL_GameControllerGetBindForAxis(controller, sdl_axis);
+        mapping.insert_or_assign(
+            switch_button,
+            BuildParamPackageForBinding(joystick->GetPort(), joystick->GetGUID(), binding));
+    }
+
+    return mapping;
+}
+
+bool SDLState::IsButtonOnLeftSide(Settings::NativeButton::Values button) const {
+    switch (button) {
+    case Settings::NativeButton::DDown:
+    case Settings::NativeButton::DLeft:
+    case Settings::NativeButton::DRight:
+    case Settings::NativeButton::DUp:
+    case Settings::NativeButton::L:
+    case Settings::NativeButton::LStick:
+    case Settings::NativeButton::Minus:
+    case Settings::NativeButton::Screenshot:
+    case Settings::NativeButton::ZL:
+        return true;
+    default:
+        return false;
+    }
+}
+
 AnalogMapping SDLState::GetAnalogMappingForDevice(const Common::ParamPackage& params) {
     if (!params.Has("guid") || !params.Has("port")) {
         return {};
     }
     const auto joystick = GetSDLJoystickByGUID(params.Get("guid", ""), params.Get("port", 0));
+    const auto joystick2 = GetSDLJoystickByGUID(params.Get("guid2", ""), params.Get("port", 0));
     auto* controller = joystick->GetSDLGameController();
     if (controller == nullptr) {
         return {};
@@ -1148,10 +1294,17 @@ AnalogMapping SDLState::GetAnalogMappingForDevice(const Common::ParamPackage& pa
         SDL_GameControllerGetBindForAxis(controller, SDL_CONTROLLER_AXIS_LEFTX);
     const auto& binding_left_y =
         SDL_GameControllerGetBindForAxis(controller, SDL_CONTROLLER_AXIS_LEFTY);
-    mapping.insert_or_assign(Settings::NativeAnalog::LStick,
-                             BuildParamPackageForAnalog(joystick->GetPort(), joystick->GetGUID(),
-                                                        binding_left_x.value.axis,
-                                                        binding_left_y.value.axis));
+    if (params.Has("guid2")) {
+        mapping.insert_or_assign(
+            Settings::NativeAnalog::LStick,
+            BuildParamPackageForAnalog(joystick2->GetPort(), joystick2->GetGUID(),
+                                       binding_left_x.value.axis, binding_left_y.value.axis));
+    } else {
+        mapping.insert_or_assign(
+            Settings::NativeAnalog::LStick,
+            BuildParamPackageForAnalog(joystick->GetPort(), joystick->GetGUID(),
+                                       binding_left_x.value.axis, binding_left_y.value.axis));
+    }
     const auto& binding_right_x =
         SDL_GameControllerGetBindForAxis(controller, SDL_CONTROLLER_AXIS_RIGHTX);
     const auto& binding_right_y =
@@ -1168,20 +1321,32 @@ MotionMapping SDLState::GetMotionMappingForDevice(const Common::ParamPackage& pa
         return {};
     }
     const auto joystick = GetSDLJoystickByGUID(params.Get("guid", ""), params.Get("port", 0));
+    const auto joystick2 = GetSDLJoystickByGUID(params.Get("guid2", ""), params.Get("port", 0));
     auto* controller = joystick->GetSDLGameController();
     if (controller == nullptr) {
         return {};
     }
 
+    MotionMapping mapping = {};
     joystick->EnableMotion();
 
-    if (!joystick->HasGyro() && !joystick->HasAccel()) {
-        return {};
+    if (joystick->HasGyro() || joystick->HasAccel()) {
+        mapping.insert_or_assign(Settings::NativeMotion::MotionRight,
+                                 BuildMotionParam(joystick->GetPort(), joystick->GetGUID()));
+    }
+    if (params.Has("guid2")) {
+        joystick2->EnableMotion();
+        if (joystick2->HasGyro() || joystick2->HasAccel()) {
+            mapping.insert_or_assign(Settings::NativeMotion::MotionLeft,
+                                     BuildMotionParam(joystick2->GetPort(), joystick2->GetGUID()));
+        }
+    } else {
+        if (joystick->HasGyro() || joystick->HasAccel()) {
+            mapping.insert_or_assign(Settings::NativeMotion::MotionLeft,
+                                     BuildMotionParam(joystick->GetPort(), joystick->GetGUID()));
+        }
     }
 
-    MotionMapping mapping = {};
-    mapping.insert_or_assign(Settings::NativeMotion::MotionLeft,
-                             BuildMotionParam(joystick->GetPort(), joystick->GetGUID()));
     return mapping;
 }
 namespace Polling {

src/input_common/sdl/sdl_impl.h

@@ -9,6 +9,17 @@
 #include <mutex>
 #include <thread>
 #include <unordered_map>
+
+// Ignore -Wimplicit-fallthrough due to https://github.com/libsdl-org/SDL/issues/4307
+#ifdef __GNUC__
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wimplicit-fallthrough"
+#endif
+#include <SDL.h>
+#ifdef __GNUC__
+#pragma GCC diagnostic pop
+#endif
+
 #include "common/common_types.h"
 #include "common/threadsafe_queue.h"
 #include "input_common/sdl/sdl.h"
@@ -18,6 +29,11 @@ using SDL_GameController = struct _SDL_GameController;
 using SDL_Joystick = struct _SDL_Joystick;
 using SDL_JoystickID = s32;
 
+using ButtonBindings =
+    std::array<std::pair<Settings::NativeButton::Values, SDL_GameControllerButton>, 17>;
+using ZButtonBindings =
+    std::array<std::pair<Settings::NativeButton::Values, SDL_GameControllerAxis>, 2>;
+
 namespace InputCommon::SDL {
 
 class SDLAnalogFactory;
@@ -66,8 +82,25 @@ private:
     /// Needs to be called before SDL_QuitSubSystem.
     void CloseJoysticks();
 
-    /// Returns a custom name for specific controllers because the default name is not correct
-    std::string GetControllerName(SDL_GameController* controller) const;
+    /// Returns the default button bindings list for generic controllers
+    ButtonBindings GetDefaultButtonBinding() const;
+
+    /// Returns the default button bindings list for nintendo controllers
+    ButtonBindings GetNintendoButtonBinding(const std::shared_ptr<SDLJoystick>& joystick) const;
+
+    /// Returns the button mappings from a single controller
+    ButtonMapping GetSingleControllerMapping(const std::shared_ptr<SDLJoystick>& joystick,
+                                             const ButtonBindings& switch_to_sdl_button,
+                                             const ZButtonBindings& switch_to_sdl_axis) const;
+
+    /// Returns the button mappings from two different controllers
+    ButtonMapping GetDualControllerMapping(const std::shared_ptr<SDLJoystick>& joystick,
+                                           const std::shared_ptr<SDLJoystick>& joystick2,
+                                           const ButtonBindings& switch_to_sdl_button,
+                                           const ZButtonBindings& switch_to_sdl_axis) const;
+
+    /// Returns true if the button is on the left joycon
+    bool IsButtonOnLeftSide(Settings::NativeButton::Values button) const;
 
     // Set to true if SDL supports game controller subsystem
     bool has_gamecontroller = false;

src/video_core/engines/shader_bytecode.h

@@ -1785,6 +1785,8 @@ public:
         SSY,
         SYNC,
         BRK,
+        CAL,
+        RET,
         DEPBAR,
         VOTE,
         VOTE_VTG,
@@ -2108,6 +2110,8 @@ private:
             INST("1111000011111---", Id::SYNC, Type::Flow, "SYNC"),
             INST("111000110100----", Id::BRK, Type::Flow, "BRK"),
             INST("111000110000----", Id::EXIT, Type::Flow, "EXIT"),
+            INST("111000100110----", Id::CAL, Type::Flow, "CAL"),
+            INST("111000110010----", Id::RET, Type::Flow, "RET"),
             INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"),
             INST("0101000011011---", Id::VOTE, Type::Warp, "VOTE"),
             INST("0101000011100---", Id::VOTE_VTG, Type::Warp, "VOTE_VTG"),

src/video_core/memory_manager.cpp

@@ -114,6 +114,25 @@ void MemoryManager::TryUnlockPage(PageEntry page_entry, std::size_t size) {
                .IsSuccess());
 }
 
+void MemoryManager::UnmapVicFrame(GPUVAddr gpu_addr, std::size_t size) {
+    if (!size) {
+        return;
+    }
+
+    const std::optional<VAddr> cpu_addr = GpuToCpuAddress(gpu_addr);
+    ASSERT(cpu_addr);
+    rasterizer->InvalidateExceptTextureCache(*cpu_addr, size);
+    cache_invalidate_queue.push_back({*cpu_addr, size});
+
+    UpdateRange(gpu_addr, PageEntry::State::Unmapped, size);
+}
+
+void MemoryManager::InvalidateQueuedCaches() {
+    for (const auto& entry : cache_invalidate_queue) {
+        rasterizer->InvalidateTextureCache(entry.first, entry.second);
+    }
+    cache_invalidate_queue.clear();
+}
 PageEntry MemoryManager::GetPageEntry(GPUVAddr gpu_addr) const {
     return page_table[PageEntryIndex(gpu_addr)];
 }

src/video_core/memory_manager.h

@@ -123,6 +123,14 @@ public:
     [[nodiscard]] GPUVAddr Allocate(std::size_t size, std::size_t align);
     void Unmap(GPUVAddr gpu_addr, std::size_t size);
 
+    /**
+     * Some Decoded NVDEC frames require that texture cache does not get invalidated.
+     * UnmapVicFrame defers the texture cache invalidation until the stream ends
+     * by invoking InvalidateQueuedCaches to invalidate all frame texture caches.
+     */
+    void UnmapVicFrame(GPUVAddr gpu_addr, std::size_t size);
+    void InvalidateQueuedCaches();
+
 private:
     [[nodiscard]] PageEntry GetPageEntry(GPUVAddr gpu_addr) const;
     void SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size = page_size);

src/video_core/rasterizer_interface.h

@@ -69,6 +69,12 @@ public:
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     virtual void FlushRegion(VAddr addr, u64 size) = 0;
 
+    /// Notify rasterizer to flush the texture cache to Switch memory
+    virtual void InvalidateExceptTextureCache(VAddr addr, u64 size) = 0;
+
+    /// Notify rasterizer to invalidate the texture cache
+    virtual void InvalidateTextureCache(VAddr addr, u64 size) = 0;
+
     /// Check if the the specified memory area requires flushing to CPU Memory.
     virtual bool MustFlushRegion(VAddr addr, u64 size) = 0;
 

src/video_core/renderer_opengl/gl_arb_decompiler.cpp

@@ -491,6 +491,9 @@ private:
     const Registry& registry;
     const ShaderType stage;
 
+    std::shared_ptr<ShaderFunctionIR> context_func;
+    u32 ast_var_base{};
+
     std::size_t num_temporaries = 0;
     std::size_t max_temporaries = 0;
 
@@ -807,13 +810,33 @@ ARBDecompiler::ARBDecompiler(const Device& device_, const ShaderIR& ir_, const R
     : device{device_}, ir{ir_}, registry{registry_}, stage{stage_} {
     DefineGlobalMemory();
 
+    context_func = ir.GetMainFunction();
+    ast_var_base = 0;
+
     AddLine("TEMP RC;");
     AddLine("TEMP FSWZA[4];");
     AddLine("TEMP FSWZB[4];");
-    if (ir.IsDecompiled()) {
+    InitializeVariables();
+    AddLine("main:");
+    if (context_func->IsDecompiled()) {
         DecompileAST();
     } else {
         DecompileBranchMode();
+        AddLine("RET;");
+    }
+
+    const auto& subfunctions = ir.GetSubFunctions();
+    auto it = subfunctions.begin();
+    while (it != subfunctions.end()) {
+        context_func = *it;
+        AddLine("func_{}:", context_func->GetId());
+        if (context_func->IsDecompiled()) {
+            DecompileAST();
+        } else {
+            DecompileBranchMode();
+            AddLine("RET;");
+        }
+        it++;
     }
     AddLine("END");
 
@@ -1060,41 +1083,38 @@ void ARBDecompiler::InitializeVariables() {
 }
 
 void ARBDecompiler::DecompileAST() {
-    const u32 num_flow_variables = ir.GetASTNumVariables();
+    const u32 num_flow_variables = context_func->GetASTNumVariables();
     for (u32 i = 0; i < num_flow_variables; ++i) {
-        AddLine("TEMP F{};", i);
+        AddLine("TEMP F{};", i + ast_var_base);
     }
     for (u32 i = 0; i < num_flow_variables; ++i) {
-        AddLine("MOV.U F{}, {{0, 0, 0, 0}};", i);
+        AddLine("MOV.U F{}, {{0, 0, 0, 0}};", i + ast_var_base);
     }
 
-    InitializeVariables();
-
-    VisitAST(ir.GetASTProgram());
+    VisitAST(context_func->GetASTProgram());
+    ast_var_base += num_flow_variables;
 }
 
 void ARBDecompiler::DecompileBranchMode() {
     static constexpr u32 FLOW_STACK_SIZE = 20;
-    if (!ir.IsFlowStackDisabled()) {
+    if (!context_func->IsFlowStackDisabled()) {
         AddLine("TEMP SSY[{}];", FLOW_STACK_SIZE);
         AddLine("TEMP PBK[{}];", FLOW_STACK_SIZE);
         AddLine("TEMP SSY_TOP;");
         AddLine("TEMP PBK_TOP;");
     }
 
-    AddLine("TEMP PC;");
+    AddLine("TEMP PC{};", context_func->GetId());
 
-    if (!ir.IsFlowStackDisabled()) {
+    if (!context_func->IsFlowStackDisabled()) {
         AddLine("MOV.U SSY_TOP.x, 0;");
         AddLine("MOV.U PBK_TOP.x, 0;");
     }
 
-    InitializeVariables();
-
-    const auto basic_block_end = ir.GetBasicBlocks().end();
-    auto basic_block_it = ir.GetBasicBlocks().begin();
+    const auto basic_block_end = context_func->GetBasicBlocks().end();
+    auto basic_block_it = context_func->GetBasicBlocks().begin();
     const u32 first_address = basic_block_it->first;
-    AddLine("MOV.U PC.x, {};", first_address);
+    AddLine("MOV.U PC{}.x, {};", context_func->GetId(), first_address);
 
     AddLine("REP;");
 
@@ -1103,7 +1123,7 @@ void ARBDecompiler::DecompileBranchMode() {
         const auto& [address, bb] = *basic_block_it;
         ++num_blocks;
 
-        AddLine("SEQ.S.CC RC.x, PC.x, {};", address);
+        AddLine("SEQ.S.CC RC.x, PC{}.x, {};", context_func->GetId(), address);
         AddLine("IF NE.x;");
 
         VisitBlock(bb);
@@ -1114,7 +1134,7 @@ void ARBDecompiler::DecompileBranchMode() {
             const auto op = std::get_if<OperationNode>(&*bb[bb.size() - 1]);
             if (!op || op->GetCode() != OperationCode::Branch) {
                 const u32 next_address = basic_block_it->first;
-                AddLine("MOV.U PC.x, {};", next_address);
+                AddLine("MOV.U PC{}.x, {};", context_func->GetId(), next_address);
                 AddLine("CONT;");
             }
         }
@@ -1152,7 +1172,8 @@ void ARBDecompiler::VisitAST(const ASTNode& node) {
     } else if (const auto decoded = std::get_if<ASTBlockDecoded>(&*node->GetInnerData())) {
         VisitBlock(decoded->nodes);
     } else if (const auto var_set = std::get_if<ASTVarSet>(&*node->GetInnerData())) {
-        AddLine("MOV.U F{}, {};", var_set->index, VisitExpression(var_set->condition));
+        AddLine("MOV.U F{}, {};", var_set->index + ast_var_base,
+                VisitExpression(var_set->condition));
         ResetTemporaries();
     } else if (const auto do_while = std::get_if<ASTDoWhile>(&*node->GetInnerData())) {
         const std::string condition = VisitExpression(do_while->condition);
@@ -1172,7 +1193,11 @@ void ARBDecompiler::VisitAST(const ASTNode& node) {
             ResetTemporaries();
         }
         if (ast_return->kills) {
-            AddLine("KIL TR;");
+            if (stage == ShaderType::Fragment) {
+                AddLine("KIL TR;");
+            } else {
+                AddLine("RET;");
+            }
         } else {
             Exit();
         }
@@ -1219,7 +1244,7 @@ std::string ARBDecompiler::VisitExpression(const Expr& node) {
         return Visit(ir.GetConditionCode(expr->cc));
     }
     if (const auto expr = std::get_if<ExprVar>(&*node)) {
-        return fmt::format("F{}.x", expr->var_index);
+        return fmt::format("F{}.x", expr->var_index + ast_var_base);
     }
     if (const auto expr = std::get_if<ExprBoolean>(&*node)) {
         return expr->value ? "0xffffffff" : "0";
@@ -1406,6 +1431,11 @@ std::string ARBDecompiler::Visit(const Node& node) {
         return {};
     }
 
+    if (const auto func_call = std::get_if<FunctionCallNode>(&*node)) {
+        AddLine("CAL func_{};", func_call->GetFuncId());
+        return {};
+    }
+
     if ([[maybe_unused]] const auto cmt = std::get_if<CommentNode>(&*node)) {
         // Uncommenting this will generate invalid code. GLASM lacks comments.
         // AddLine("// {}", cmt->GetText());
@@ -1479,7 +1509,7 @@ std::string ARBDecompiler::GlobalMemoryPointer(const GmemNode& gmem) {
 }
 
 void ARBDecompiler::Exit() {
-    if (stage != ShaderType::Fragment) {
+    if (!context_func->IsMain() || stage != ShaderType::Fragment) {
         AddLine("RET;");
         return;
     }
@@ -2021,13 +2051,13 @@ std::string ARBDecompiler::ImageStore(Operation operation) {
 
 std::string ARBDecompiler::Branch(Operation operation) {
     const auto target = std::get<ImmediateNode>(*operation[0]);
-    AddLine("MOV.U PC.x, {};", target.GetValue());
+    AddLine("MOV.U PC{}.x, {};", context_func->GetId(), target.GetValue());
     AddLine("CONT;");
     return {};
 }
 
 std::string ARBDecompiler::BranchIndirect(Operation operation) {
-    AddLine("MOV.U PC.x, {};", Visit(operation[0]));
+    AddLine("MOV.U PC{}.x, {};", context_func->GetId(), Visit(operation[0]));
     AddLine("CONT;");
     return {};
 }
@@ -2045,7 +2075,7 @@ std::string ARBDecompiler::PopFlowStack(Operation operation) {
     const auto stack = std::get<MetaStackClass>(operation.GetMeta());
     const std::string_view stack_name = StackName(stack);
     AddLine("SUB.S {}_TOP.x, {}_TOP.x, 1;", stack_name, stack_name);
-    AddLine("MOV.U PC.x, {}[{}_TOP.x].x;", stack_name, stack_name);
+    AddLine("MOV.U PC{}.x, {}[{}_TOP.x].x;", context_func->GetId(), stack_name, stack_name);
     AddLine("CONT;");
     return {};
 }
@@ -2056,6 +2086,10 @@ std::string ARBDecompiler::Exit(Operation) {
 }
 
 std::string ARBDecompiler::Discard(Operation) {
+    if (stage != ShaderType::Fragment) {
+        AddLine("RET;");
+        return {};
+    }
     AddLine("KIL TR;");
     return {};
 }

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -544,6 +544,26 @@ void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
     query_cache.FlushRegion(addr, size);
 }
 
+void RasterizerOpenGL::InvalidateExceptTextureCache(VAddr addr, u64 size) {
+    if (addr == 0 || size == 0) {
+        return;
+    }
+    shader_cache.InvalidateRegion(addr, size);
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.WriteMemory(addr, size);
+    }
+    query_cache.InvalidateRegion(addr, size);
+}
+
+void RasterizerOpenGL::InvalidateTextureCache(VAddr addr, u64 size) {
+    if (addr == 0 || size == 0) {
+        return;
+    }
+    std::scoped_lock lock{texture_cache.mutex};
+    texture_cache.UnmapMemory(addr, size);
+}
+
 bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size) {
     std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
     if (!Settings::IsGPULevelHigh()) {

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -74,6 +74,8 @@ public:
     void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
     void FlushAll() override;
     void FlushRegion(VAddr addr, u64 size) override;
+    void InvalidateExceptTextureCache(VAddr addr, u64 size) override;
+    void InvalidateTextureCache(VAddr addr, u64 size) override;
     bool MustFlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;
     void OnCPUWrite(VAddr addr, u64 size) override;

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -435,6 +435,27 @@ public:
         DeclareCustomVariables();
         DeclarePhysicalAttributeReader();
 
+        const auto& subfunctions = ir.GetSubFunctions();
+        auto it = subfunctions.rbegin();
+        while (it != subfunctions.rend()) {
+            context_func = *it;
+            code.AddLine("void func_{}() {{", context_func->GetId());
+            ++code.scope;
+
+            if (context_func->IsDecompiled()) {
+                DecompileAST();
+            } else {
+                DecompileBranchMode();
+            }
+
+            --code.scope;
+            code.AddLine("}}");
+
+            it++;
+        }
+
+        context_func = ir.GetMainFunction();
+
         code.AddLine("void main() {{");
         ++code.scope;
 
@@ -442,7 +463,7 @@ public:
             code.AddLine("gl_Position = vec4(0.0f, 0.0f, 0.0f, 1.0f);");
         }
 
-        if (ir.IsDecompiled()) {
+        if (context_func->IsDecompiled()) {
             DecompileAST();
         } else {
             DecompileBranchMode();
@@ -462,13 +483,13 @@ private:
 
     void DecompileBranchMode() {
         // VM's program counter
-        const auto first_address = ir.GetBasicBlocks().begin()->first;
+        const auto first_address = context_func->GetBasicBlocks().begin()->first;
         code.AddLine("uint jmp_to = {}U;", first_address);
 
         // TODO(Subv): Figure out the actual depth of the flow stack, for now it seems
         // unlikely that shaders will use 20 nested SSYs and PBKs.
         constexpr u32 FLOW_STACK_SIZE = 20;
-        if (!ir.IsFlowStackDisabled()) {
+        if (!context_func->IsFlowStackDisabled()) {
             for (const auto stack : std::array{MetaStackClass::Ssy, MetaStackClass::Pbk}) {
                 code.AddLine("uint {}[{}];", FlowStackName(stack), FLOW_STACK_SIZE);
                 code.AddLine("uint {} = 0U;", FlowStackTopName(stack));
@@ -480,7 +501,7 @@ private:
 
         code.AddLine("switch (jmp_to) {{");
 
-        for (const auto& pair : ir.GetBasicBlocks()) {
+        for (const auto& pair : context_func->GetBasicBlocks()) {
             const auto& [address, bb] = pair;
             code.AddLine("case 0x{:X}U: {{", address);
             ++code.scope;
@@ -1131,6 +1152,11 @@ private:
             return {};
         }
 
+        if (const auto func_call = std::get_if<FunctionCallNode>(&*node)) {
+            code.AddLine("func_{}();", func_call->GetFuncId());
+            return {};
+        }
+
         if (const auto comment = std::get_if<CommentNode>(&*node)) {
             code.AddLine("// " + comment->GetText());
             return {};
@@ -2267,7 +2293,9 @@ private:
     }
 
     Expression Exit(Operation operation) {
-        PreExit();
+        if (context_func->IsMain()) {
+            PreExit();
+        }
         code.AddLine("return;");
         return {};
     }
@@ -2277,7 +2305,11 @@ private:
         // about unexecuted instructions that may follow this.
         code.AddLine("if (true) {{");
         ++code.scope;
-        code.AddLine("discard;");
+        if (stage != ShaderType::Fragment) {
+            code.AddLine("return;");
+        } else {
+            code.AddLine("discard;");
+        }
         --code.scope;
         code.AddLine("}}");
         return {};
@@ -2388,7 +2420,7 @@ private:
     }
 
     Expression Barrier(Operation) {
-        if (!ir.IsDecompiled()) {
+        if (!context_func->IsDecompiled()) {
             LOG_ERROR(Render_OpenGL, "barrier() used but shader is not decompiled");
             return {};
         }
@@ -2755,6 +2787,8 @@ private:
     const Header header;
     std::unordered_map<u8, VaryingTFB> transform_feedback;
 
+    std::shared_ptr<ShaderFunctionIR> context_func;
+
     ShaderWriter code;
 
     std::optional<u32> max_input_vertices;
@@ -2902,9 +2936,15 @@ public:
             decomp.code.scope++;
         }
         if (ast.kills) {
-            decomp.code.AddLine("discard;");
+            if (decomp.stage != ShaderType::Fragment) {
+                decomp.code.AddLine("return;");
+            } else {
+                decomp.code.AddLine("discard;");
+            }
         } else {
-            decomp.PreExit();
+            if (decomp.context_func->IsMain()) {
+                decomp.PreExit();
+            }
             decomp.code.AddLine("return;");
         }
         if (!is_true) {
@@ -2937,13 +2977,13 @@ private:
 };
 
 void GLSLDecompiler::DecompileAST() {
-    const u32 num_flow_variables = ir.GetASTNumVariables();
+    const u32 num_flow_variables = context_func->GetASTNumVariables();
     for (u32 i = 0; i < num_flow_variables; i++) {
         code.AddLine("bool {} = false;", GetFlowVariable(i));
     }
 
     ASTDecompiler decompiler{*this};
-    decompiler.Visit(ir.GetASTProgram());
+    decompiler.Visit(context_func->GetASTProgram());
 }
 
 } // Anonymous namespace

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -493,6 +493,26 @@ void RasterizerVulkan::FlushRegion(VAddr addr, u64 size) {
     query_cache.FlushRegion(addr, size);
 }
 
+void Vulkan::RasterizerVulkan::InvalidateExceptTextureCache(VAddr addr, u64 size) {
+    if (addr == 0 || size == 0) {
+        return;
+    }
+    pipeline_cache.InvalidateRegion(addr, size);
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.WriteMemory(addr, size);
+    }
+    query_cache.InvalidateRegion(addr, size);
+}
+
+void Vulkan::RasterizerVulkan::InvalidateTextureCache(VAddr addr, u64 size) {
+    if (addr == 0 || size == 0) {
+        return;
+    }
+    std::scoped_lock lock{texture_cache.mutex};
+    texture_cache.UnmapMemory(addr, size);
+}
+
 bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size) {
     std::scoped_lock lock{texture_cache.mutex, buffer_cache.mutex};
     if (!Settings::IsGPULevelHigh()) {

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -66,6 +66,8 @@ public:
     void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
     void FlushAll() override;
     void FlushRegion(VAddr addr, u64 size) override;
+    void InvalidateExceptTextureCache(VAddr addr, u64 size) override;
+    void InvalidateTextureCache(VAddr addr, u64 size) override;
     bool MustFlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;
     void OnCPUWrite(VAddr addr, u64 size) override;

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -406,10 +406,38 @@ private:
         binding = DeclareStorageTexels(binding);
         binding = DeclareImages(binding);
 
+        const auto& subfunctions = ir.GetSubFunctions();
+
+        labels.resize(subfunctions.size() + 1);
+        other_functions.resize(subfunctions.size());
+
+        auto it = subfunctions.rbegin();
+        while (it != subfunctions.rend()) {
+            context_func = *it;
+            other_functions[context_func->GetId() - 1] =
+                OpFunction(t_void, {}, TypeFunction(t_void));
+            AddLabel();
+
+            if (context_func->IsDecompiled()) {
+                DeclareFlowVariables();
+                DecompileAST();
+            } else {
+                AllocateLabels();
+                DecompileBranchMode();
+            }
+
+            OpReturn();
+            OpFunctionEnd();
+
+            it++;
+        }
+
+        context_func = ir.GetMainFunction();
+
         const Id main = OpFunction(t_void, {}, TypeFunction(t_void));
         AddLabel();
 
-        if (ir.IsDecompiled()) {
+        if (context_func->IsDecompiled()) {
             DeclareFlowVariables();
             DecompileAST();
         } else {
@@ -441,16 +469,18 @@ private:
     void DecompileAST();
 
     void DecompileBranchMode() {
-        const u32 first_address = ir.GetBasicBlocks().begin()->first;
-        const Id loop_label = OpLabel("loop");
-        const Id merge_label = OpLabel("merge");
+        const u32 first_address = context_func->GetBasicBlocks().begin()->first;
+        const u32 func_id = context_func->GetId();
+        const std::string func_id_msg = std::to_string(func_id);
+        const Id loop_label = OpLabel("loop_" + func_id_msg);
+        const Id merge_label = OpLabel("merge_" + func_id_msg);
         const Id dummy_label = OpLabel();
         const Id jump_label = OpLabel();
-        continue_label = OpLabel("continue");
+        continue_label = OpLabel("continue_" + func_id_msg);
 
         std::vector<Sirit::Literal> literals;
         std::vector<Id> branch_labels;
-        for (const auto& [literal, label] : labels) {
+        for (const auto& [literal, label] : labels[func_id]) {
             literals.push_back(literal);
             branch_labels.push_back(label);
         }
@@ -462,11 +492,11 @@ private:
         std::tie(ssy_flow_stack, ssy_flow_stack_top) = CreateFlowStack();
         std::tie(pbk_flow_stack, pbk_flow_stack_top) = CreateFlowStack();
 
-        Name(jmp_to, "jmp_to");
-        Name(ssy_flow_stack, "ssy_flow_stack");
-        Name(ssy_flow_stack_top, "ssy_flow_stack_top");
-        Name(pbk_flow_stack, "pbk_flow_stack");
-        Name(pbk_flow_stack_top, "pbk_flow_stack_top");
+        Name(jmp_to, "jmp_to_" + func_id_msg);
+        Name(ssy_flow_stack, "ssy_flow_stack_" + func_id_msg);
+        Name(ssy_flow_stack_top, "ssy_flow_stack_top_" + func_id_msg);
+        Name(pbk_flow_stack, "pbk_flow_stack_" + func_id_msg);
+        Name(pbk_flow_stack_top, "pbk_flow_stack_top_" + func_id_msg);
 
         DefinePrologue();
 
@@ -484,13 +514,14 @@ private:
         AddLabel(default_branch);
         OpReturn();
 
-        for (const auto& [address, bb] : ir.GetBasicBlocks()) {
-            AddLabel(labels.at(address));
+        for (const auto& [address, bb] : context_func->GetBasicBlocks()) {
+            AddLabel(labels[func_id].at(address));
 
             VisitBasicBlock(bb);
 
-            const auto next_it = labels.lower_bound(address + 1);
-            const Id next_label = next_it != labels.end() ? next_it->second : default_branch;
+            const auto next_it = labels[func_id].lower_bound(address + 1);
+            const Id next_label =
+                next_it != labels[func_id].end() ? next_it->second : default_branch;
             OpBranch(next_label);
         }
 
@@ -508,9 +539,10 @@ private:
     static constexpr auto INTERNAL_FLAGS_COUNT = static_cast<std::size_t>(InternalFlag::Amount);
 
     void AllocateLabels() {
-        for (const auto& pair : ir.GetBasicBlocks()) {
+        const u32 func_id = context_func->GetId();
+        for (const auto& pair : context_func->GetBasicBlocks()) {
             const u32 address = pair.first;
-            labels.emplace(address, OpLabel(fmt::format("label_0x{:x}", address)));
+            labels[func_id].emplace(address, OpLabel(fmt::format("label_0x{:x}", address)));
         }
     }
 
@@ -589,6 +621,14 @@ private:
         DeclareOutputVertex();
     }
 
+    void SafeKill() {
+        if (stage != ShaderType::Fragment) {
+            OpReturn();
+            return;
+        }
+        OpKill();
+    }
+
     void DeclareFragment() {
         if (stage != ShaderType::Fragment) {
             return;
@@ -656,7 +696,7 @@ private:
     }
 
     void DeclareFlowVariables() {
-        for (u32 i = 0; i < ir.GetASTNumVariables(); i++) {
+        for (u32 i = 0; i < context_func->GetASTNumVariables(); i++) {
             const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false);
             Name(id, fmt::format("flow_var_{}", static_cast<u32>(i)));
             flow_variables.emplace(i, AddGlobalVariable(id));
@@ -1333,6 +1373,12 @@ private:
             return {};
         }
 
+        if (const auto func_call = std::get_if<FunctionCallNode>(&*node)) {
+            const u32 func_id = func_call->GetFuncId();
+            OpFunctionCall(t_void, other_functions[func_id - 1]);
+            return {};
+        }
+
         if (const auto comment = std::get_if<CommentNode>(&*node)) {
             if (device.HasDebuggingToolAttached()) {
                 // We should insert comments with OpString instead of using named variables
@@ -2124,7 +2170,7 @@ private:
 
         OpBranchConditional(condition, true_label, discard_label);
         AddLabel(discard_label);
-        OpKill();
+        SafeKill();
         AddLabel(true_label);
     }
 
@@ -2175,7 +2221,9 @@ private:
     }
 
     Expression Exit(Operation operation) {
-        PreExit();
+        if (context_func->IsMain()) {
+            PreExit();
+        }
         inside_branch = true;
         if (conditional_branch_set) {
             OpReturn();
@@ -2192,12 +2240,12 @@ private:
     Expression Discard(Operation operation) {
         inside_branch = true;
         if (conditional_branch_set) {
-            OpKill();
+            SafeKill();
         } else {
             const Id dummy = OpLabel();
             OpBranch(dummy);
             AddLabel(dummy);
-            OpKill();
+            SafeKill();
             AddLabel();
         }
         return {};
@@ -2276,7 +2324,7 @@ private:
     }
 
     Expression Barrier(Operation) {
-        if (!ir.IsDecompiled()) {
+        if (!context_func->IsDecompiled()) {
             LOG_ERROR(Render_Vulkan, "OpBarrier used by shader is not decompiled");
             return {};
         }
@@ -2770,6 +2818,8 @@ private:
     const Specialization& specialization;
     std::unordered_map<u8, VaryingTFB> transform_feedback;
 
+    std::shared_ptr<ShaderFunctionIR> context_func;
+
     const Id t_void = Name(TypeVoid(), "void");
 
     const Id t_bool = Name(TypeBool(), "bool");
@@ -2896,7 +2946,8 @@ private:
     Id ssy_flow_stack{};
     Id pbk_flow_stack{};
     Id continue_label{};
-    std::map<u32, Id> labels;
+    std::vector<std::map<u32, Id>> labels;
+    std::vector<Id> other_functions;
 
     bool conditional_branch_set{};
     bool inside_branch{};
@@ -3047,9 +3098,11 @@ public:
             decomp.OpBranchConditional(condition, then_label, endif_label);
             decomp.AddLabel(then_label);
             if (ast.kills) {
-                decomp.OpKill();
+                decomp.SafeKill();
             } else {
-                decomp.PreExit();
+                if (decomp.context_func->IsMain()) {
+                    decomp.PreExit();
+                }
                 decomp.OpReturn();
             }
             decomp.AddLabel(endif_label);
@@ -3058,9 +3111,11 @@ public:
             decomp.OpBranch(next_block);
             decomp.AddLabel(next_block);
             if (ast.kills) {
-                decomp.OpKill();
+                decomp.SafeKill();
             } else {
-                decomp.PreExit();
+                if (decomp.context_func->IsMain()) {
+                    decomp.PreExit();
+                }
                 decomp.OpReturn();
             }
             decomp.AddLabel(decomp.OpLabel());
@@ -3097,7 +3152,7 @@ private:
 };
 
 void SPIRVDecompiler::DecompileAST() {
-    const u32 num_flow_variables = ir.GetASTNumVariables();
+    const u32 num_flow_variables = context_func->GetASTNumVariables();
     for (u32 i = 0; i < num_flow_variables; i++) {
         const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false);
         Name(id, fmt::format("flow_var_{}", i));
@@ -3106,7 +3161,7 @@ void SPIRVDecompiler::DecompileAST() {
 
     DefinePrologue();
 
-    const ASTNode program = ir.GetASTProgram();
+    const ASTNode program = context_func->GetASTProgram();
     ASTDecompiler decompiler{*this};
     decompiler.Visit(program);
 

src/video_core/shader/control_flow.cpp

@@ -7,6 +7,7 @@
 #include <set>
 #include <stack>
 #include <unordered_map>
+#include <unordered_set>
 #include <vector>
 
 #include "common/assert.h"
@@ -26,17 +27,29 @@ using Tegra::Shader::OpCode;
 
 constexpr s32 unassigned_branch = -2;
 
+enum class JumpLabel : u32 {
+    SSYClass = 0,
+    PBKClass = 1,
+};
+
+struct JumpItem {
+    JumpLabel type;
+    u32 address;
+
+    bool operator==(const JumpItem& other) const {
+        return std::tie(type, address) == std::tie(other.type, other.address);
+    }
+};
+
 struct Query {
     u32 address{};
-    std::stack<u32> ssy_stack{};
-    std::stack<u32> pbk_stack{};
+    std::stack<JumpItem> stack{};
 };
 
 struct BlockStack {
     BlockStack() = default;
-    explicit BlockStack(const Query& q) : ssy_stack{q.ssy_stack}, pbk_stack{q.pbk_stack} {}
-    std::stack<u32> ssy_stack{};
-    std::stack<u32> pbk_stack{};
+    explicit BlockStack(const Query& q) : stack{q.stack} {}
+    std::stack<JumpItem> stack{};
 };
 
 template <typename T, typename... Args>
@@ -65,20 +78,36 @@ struct BlockInfo {
     }
 };
 
-struct CFGRebuildState {
-    explicit CFGRebuildState(const ProgramCode& program_code_, u32 start_, Registry& registry_)
-        : program_code{program_code_}, registry{registry_}, start{start_} {}
+struct ProgramControl {
+    std::unordered_set<u32> found_functions{};
+    std::list<u32> pending_functions{};
 
+    void RegisterFunction(u32 address) {
+        if (found_functions.count(address) != 0) {
+            return;
+        }
+        found_functions.insert(address);
+        pending_functions.emplace_back(address);
+    }
+};
+
+struct CFGRebuildState {
+    explicit CFGRebuildState(ProgramControl& control_, const ProgramCode& program_code_, u32 start_,
+                             u32 base_start_, Registry& registry_)
+        : control{control_}, program_code{program_code_}, registry{registry_}, start{start_},
+          base_start{base_start_} {}
+
+    ProgramControl& control;
     const ProgramCode& program_code;
     Registry& registry;
     u32 start{};
+    u32 base_start{};
     std::vector<BlockInfo> block_info;
     std::list<u32> inspect_queries;
     std::list<Query> queries;
     std::unordered_map<u32, u32> registered;
     std::set<u32> labels;
-    std::map<u32, u32> ssy_labels;
-    std::map<u32, u32> pbk_labels;
+    std::map<u32, JumpItem> jump_labels;
     std::unordered_map<u32, BlockStack> stacks;
     ASTManager* manager{};
 };
@@ -153,7 +182,7 @@ template <typename Result, typename TestCallable, typename PackCallable>
 std::optional<Result> TrackInstruction(const CFGRebuildState& state, u32& pos, TestCallable test,
                                        PackCallable pack) {
     for (; pos >= state.start; --pos) {
-        if (IsSchedInstruction(pos, state.start)) {
+        if (IsSchedInstruction(pos, state.base_start)) {
             continue;
         }
         const Instruction instr = state.program_code[pos];
@@ -262,7 +291,7 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
             single_branch.ignore = true;
             break;
         }
-        if (IsSchedInstruction(offset, state.start)) {
+        if (IsSchedInstruction(offset, state.base_start)) {
             offset++;
             continue;
         }
@@ -274,6 +303,7 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
         }
 
         switch (opcode->get().GetId()) {
+        case OpCode::Id::RET:
         case OpCode::Id::EXIT: {
             const auto pred_index = static_cast<u32>(instr.pred.pred_index);
             single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
@@ -411,13 +441,20 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
         case OpCode::Id::SSY: {
             const u32 target = offset + instr.bra.GetBranchTarget();
             insert_label(state, target);
-            state.ssy_labels.emplace(offset, target);
+            JumpItem it = {JumpLabel::SSYClass, target};
+            state.jump_labels.emplace(offset, it);
             break;
         }
         case OpCode::Id::PBK: {
             const u32 target = offset + instr.bra.GetBranchTarget();
             insert_label(state, target);
-            state.pbk_labels.emplace(offset, target);
+            JumpItem it = {JumpLabel::PBKClass, target};
+            state.jump_labels.emplace(offset, it);
+            break;
+        }
+        case OpCode::Id::CAL: {
+            const u32 target = offset + instr.bra.GetBranchTarget();
+            state.control.RegisterFunction(target);
             break;
         }
         case OpCode::Id::BRX: {
@@ -513,7 +550,7 @@ bool TryInspectAddress(CFGRebuildState& state) {
 }
 
 bool TryQuery(CFGRebuildState& state) {
-    const auto gather_labels = [](std::stack<u32>& cc, std::map<u32, u32>& labels,
+    const auto gather_labels = [](std::stack<JumpItem>& cc, std::map<u32, JumpItem>& labels,
                                   BlockInfo& block) {
         auto gather_start = labels.lower_bound(block.start);
         const auto gather_end = labels.upper_bound(block.end);
@@ -522,6 +559,19 @@ bool TryQuery(CFGRebuildState& state) {
             ++gather_start;
         }
     };
+    const auto pop_labels = [](JumpLabel type, SingleBranch* branch, Query& query) -> bool {
+        while (!query.stack.empty() && query.stack.top().type != type) {
+            query.stack.pop();
+        }
+        if (query.stack.empty()) {
+            return false;
+        }
+        if (branch->address == unassigned_branch) {
+            branch->address = query.stack.top().address;
+        }
+        query.stack.pop();
+        return true;
+    };
     if (state.queries.empty()) {
         return false;
     }
@@ -534,8 +584,7 @@ bool TryQuery(CFGRebuildState& state) {
     // consumes a label. Schedule new queries accordingly
     if (block.visited) {
         BlockStack& stack = state.stacks[q.address];
-        const bool all_okay = (stack.ssy_stack.empty() || q.ssy_stack == stack.ssy_stack) &&
-                              (stack.pbk_stack.empty() || q.pbk_stack == stack.pbk_stack);
+        const bool all_okay = (stack.stack.empty() || q.stack == stack.stack);
         state.queries.pop_front();
         return all_okay;
     }
@@ -544,8 +593,7 @@ bool TryQuery(CFGRebuildState& state) {
 
     Query q2(q);
     state.queries.pop_front();
-    gather_labels(q2.ssy_stack, state.ssy_labels, block);
-    gather_labels(q2.pbk_stack, state.pbk_labels, block);
+    gather_labels(q2.stack, state.jump_labels, block);
     if (std::holds_alternative<SingleBranch>(*block.branch)) {
         auto* branch = std::get_if<SingleBranch>(block.branch.get());
         if (!branch->condition.IsUnconditional()) {
@@ -555,16 +603,10 @@ bool TryQuery(CFGRebuildState& state) {
 
         auto& conditional_query = state.queries.emplace_back(q2);
         if (branch->is_sync) {
-            if (branch->address == unassigned_branch) {
-                branch->address = conditional_query.ssy_stack.top();
-            }
-            conditional_query.ssy_stack.pop();
+            pop_labels(JumpLabel::SSYClass, branch, conditional_query);
         }
         if (branch->is_brk) {
-            if (branch->address == unassigned_branch) {
-                branch->address = conditional_query.pbk_stack.top();
-            }
-            conditional_query.pbk_stack.pop();
+            pop_labels(JumpLabel::PBKClass, branch, conditional_query);
         }
         conditional_query.address = branch->address;
         return true;
@@ -646,25 +688,23 @@ void DecompileShader(CFGRebuildState& state) {
     state.manager->Decompile();
 }
 
-} // Anonymous namespace
-
-std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
-                                                const CompilerSettings& settings,
-                                                Registry& registry) {
-    auto result_out = std::make_unique<ShaderCharacteristics>();
+ShaderFunction ScanFunction(ProgramControl& control, const ProgramCode& program_code,
+                            u32 start_address, u32 base_start, const CompilerSettings& settings,
+                            Registry& registry) {
+    ShaderFunction result_out{};
     if (settings.depth == CompileDepth::BruteForce) {
-        result_out->settings.depth = CompileDepth::BruteForce;
+        result_out.settings.depth = CompileDepth::BruteForce;
         return result_out;
     }
 
-    CFGRebuildState state{program_code, start_address, registry};
+    CFGRebuildState state{control, program_code, start_address, base_start, registry};
     // Inspect Code and generate blocks
     state.labels.clear();
     state.labels.emplace(start_address);
     state.inspect_queries.push_back(state.start);
     while (!state.inspect_queries.empty()) {
         if (!TryInspectAddress(state)) {
-            result_out->settings.depth = CompileDepth::BruteForce;
+            result_out.settings.depth = CompileDepth::BruteForce;
             return result_out;
         }
     }
@@ -675,7 +715,7 @@ std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
 
     if (settings.depth != CompileDepth::FlowStack) {
         // Decompile Stacks
-        state.queries.push_back(Query{state.start, {}, {}});
+        state.queries.push_back(Query{state.start, {}});
         decompiled = true;
         while (!state.queries.empty()) {
             if (!TryQuery(state)) {
@@ -705,19 +745,18 @@ std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
             state.manager->ShowCurrentState("Of Shader");
             state.manager->Clear();
         } else {
-            auto characteristics = std::make_unique<ShaderCharacteristics>();
-            characteristics->start = start_address;
-            characteristics->settings.depth = settings.depth;
-            characteristics->manager = std::move(manager);
-            characteristics->end = state.block_info.back().end + 1;
-            return characteristics;
+            result_out.start = start_address;
+            result_out.settings.depth = settings.depth;
+            result_out.manager = std::move(manager);
+            result_out.end = state.block_info.back().end + 1;
+            return result_out;
         }
     }
 
-    result_out->start = start_address;
-    result_out->settings.depth =
+    result_out.start = start_address;
+    result_out.settings.depth =
         use_flow_stack ? CompileDepth::FlowStack : CompileDepth::NoFlowStack;
-    result_out->blocks.clear();
+    result_out.blocks.clear();
     for (auto& block : state.block_info) {
         ShaderBlock new_block{};
         new_block.start = block.start;
@@ -726,20 +765,20 @@ std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
         if (!new_block.ignore_branch) {
             new_block.branch = block.branch;
         }
-        result_out->end = std::max(result_out->end, block.end);
-        result_out->blocks.push_back(new_block);
+        result_out.end = std::max(result_out.end, block.end);
+        result_out.blocks.push_back(new_block);
     }
     if (!use_flow_stack) {
-        result_out->labels = std::move(state.labels);
+        result_out.labels = std::move(state.labels);
         return result_out;
     }
 
-    auto back = result_out->blocks.begin();
+    auto back = result_out.blocks.begin();
     auto next = std::next(back);
-    while (next != result_out->blocks.end()) {
+    while (next != result_out.blocks.end()) {
         if (!state.labels.contains(next->start) && next->start == back->end + 1) {
             back->end = next->end;
-            next = result_out->blocks.erase(next);
+            next = result_out.blocks.erase(next);
             continue;
         }
         back = next;
@@ -748,4 +787,22 @@ std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
 
     return result_out;
 }
+
+} // Anonymous namespace
+
+std::unique_ptr<ShaderProgram> ScanFlow(const ProgramCode& program_code, u32 start_address,
+                                        const CompilerSettings& settings, Registry& registry) {
+    ProgramControl control{};
+    auto result_out = std::make_unique<ShaderProgram>();
+    result_out->main =
+        ScanFunction(control, program_code, start_address, start_address, settings, registry);
+    while (!control.pending_functions.empty()) {
+        u32 address = control.pending_functions.front();
+        auto fun = ScanFunction(control, program_code, address, start_address, settings, registry);
+        result_out->subfunctions.emplace(address, std::move(fun));
+        control.pending_functions.pop_front();
+    }
+    return result_out;
+}
+
 } // namespace VideoCommon::Shader

src/video_core/shader/control_flow.h

@@ -5,6 +5,7 @@
 #pragma once
 
 #include <list>
+#include <map>
 #include <optional>
 #include <set>
 #include <variant>
@@ -101,7 +102,7 @@ struct ShaderBlock {
     }
 };
 
-struct ShaderCharacteristics {
+struct ShaderFunction {
     std::list<ShaderBlock> blocks{};
     std::set<u32> labels{};
     u32 start{};
@@ -110,8 +111,12 @@ struct ShaderCharacteristics {
     CompilerSettings settings{};
 };
 
-std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
-                                                const CompilerSettings& settings,
-                                                Registry& registry);
+struct ShaderProgram {
+    ShaderFunction main;
+    std::map<u32, ShaderFunction> subfunctions;
+};
+
+std::unique_ptr<ShaderProgram> ScanFlow(const ProgramCode& program_code, u32 start_address,
+                                        const CompilerSettings& settings, Registry& registry);
 
 } // namespace VideoCommon::Shader

src/video_core/shader/decode.cpp

@@ -64,9 +64,52 @@ std::optional<u32> TryDeduceSamplerSize(const SamplerEntry& sampler_to_deduce,
 
 } // Anonymous namespace
 
+class ExprDecoder {
+public:
+    explicit ExprDecoder(ShaderIR& ir_) : ir(ir_) {}
+
+    void operator()(const ExprAnd& expr) {
+        Visit(expr.operand1);
+        Visit(expr.operand2);
+    }
+
+    void operator()(const ExprOr& expr) {
+        Visit(expr.operand1);
+        Visit(expr.operand2);
+    }
+
+    void operator()(const ExprNot& expr) {
+        Visit(expr.operand1);
+    }
+
+    void operator()(const ExprPredicate& expr) {
+        const auto pred = static_cast<Tegra::Shader::Pred>(expr.predicate);
+        if (pred != Pred::UnusedIndex && pred != Pred::NeverExecute) {
+            ir.used_predicates.insert(pred);
+        }
+    }
+
+    void operator()(const ExprCondCode& expr) {}
+
+    void operator()(const ExprVar& expr) {}
+
+    void operator()(const ExprBoolean& expr) {}
+
+    void operator()(const ExprGprEqual& expr) {
+        ir.used_registers.insert(expr.gpr);
+    }
+
+    void Visit(const Expr& node) {
+        return std::visit(*this, *node);
+    }
+
+private:
+    ShaderIR& ir;
+};
+
 class ASTDecoder {
 public:
-    explicit ASTDecoder(ShaderIR& ir_) : ir(ir_) {}
+    explicit ASTDecoder(ShaderIR& ir_) : ir(ir_), decoder(ir_) {}
 
     void operator()(ASTProgram& ast) {
         ASTNode current = ast.nodes.GetFirst();
@@ -77,6 +120,7 @@ public:
     }
 
     void operator()(ASTIfThen& ast) {
+        decoder.Visit(ast.condition);
         ASTNode current = ast.nodes.GetFirst();
         while (current) {
             Visit(current);
@@ -96,13 +140,18 @@ public:
 
     void operator()(ASTBlockDecoded& ast) {}
 
-    void operator()(ASTVarSet& ast) {}
+    void operator()(ASTVarSet& ast) {
+        decoder.Visit(ast.condition);
+    }
 
     void operator()(ASTLabel& ast) {}
 
-    void operator()(ASTGoto& ast) {}
+    void operator()(ASTGoto& ast) {
+        decoder.Visit(ast.condition);
+    }
 
     void operator()(ASTDoWhile& ast) {
+        decoder.Visit(ast.condition);
         ASTNode current = ast.nodes.GetFirst();
         while (current) {
             Visit(current);
@@ -110,9 +159,13 @@ public:
         }
     }
 
-    void operator()(ASTReturn& ast) {}
+    void operator()(ASTReturn& ast) {
+        decoder.Visit(ast.condition);
+    }
 
-    void operator()(ASTBreak& ast) {}
+    void operator()(ASTBreak& ast) {
+        decoder.Visit(ast.condition);
+    }
 
     void Visit(ASTNode& node) {
         std::visit(*this, *node->GetInnerData());
@@ -125,77 +178,113 @@ public:
 
 private:
     ShaderIR& ir;
+    ExprDecoder decoder;
 };
 
 void ShaderIR::Decode() {
+    const auto decode_function = ([this](ShaderFunction& shader_info) {
+        coverage_end = std::max<u32>(0, shader_info.end);
+        switch (shader_info.settings.depth) {
+        case CompileDepth::FlowStack: {
+            for (const auto& block : shader_info.blocks) {
+                basic_blocks.insert({block.start, DecodeRange(block.start, block.end + 1)});
+            }
+            break;
+        }
+        case CompileDepth::NoFlowStack: {
+            disable_flow_stack = true;
+            const auto insert_block = [this](NodeBlock& nodes, u32 label) {
+                if (label == static_cast<u32>(exit_branch)) {
+                    return;
+                }
+                basic_blocks.insert({label, nodes});
+            };
+            const auto& blocks = shader_info.blocks;
+            NodeBlock current_block;
+            u32 current_label = static_cast<u32>(exit_branch);
+            for (const auto& block : blocks) {
+                if (shader_info.labels.contains(block.start)) {
+                    insert_block(current_block, current_label);
+                    current_block.clear();
+                    current_label = block.start;
+                }
+                if (!block.ignore_branch) {
+                    DecodeRangeInner(current_block, block.start, block.end);
+                    InsertControlFlow(current_block, block);
+                } else {
+                    DecodeRangeInner(current_block, block.start, block.end + 1);
+                }
+            }
+            insert_block(current_block, current_label);
+            break;
+        }
+        case CompileDepth::DecompileBackwards:
+        case CompileDepth::FullDecompile: {
+            program_manager = std::move(shader_info.manager);
+            disable_flow_stack = true;
+            decompiled = true;
+            ASTDecoder decoder{*this};
+            ASTNode program = program_manager.GetProgram();
+            decoder.Visit(program);
+            break;
+        }
+        default:
+            LOG_CRITICAL(HW_GPU, "Unknown decompilation mode!");
+            [[fallthrough]];
+        case CompileDepth::BruteForce: {
+            const auto shader_end = static_cast<u32>(program_code.size());
+            coverage_begin = main_offset;
+            coverage_end = shader_end;
+            for (u32 label = main_offset; label < shader_end; ++label) {
+                basic_blocks.insert({label, DecodeRange(label, label + 1)});
+            }
+            break;
+        }
+        }
+        if (settings.depth != shader_info.settings.depth) {
+            LOG_WARNING(
+                HW_GPU,
+                "Decompiling to this setting \"{}\" failed, downgrading to this setting \"{}\"",
+                CompileDepthAsString(settings.depth),
+                CompileDepthAsString(shader_info.settings.depth));
+        }
+    });
+    const auto gen_function =
+        ([this](ShaderFunction& shader_info, u32 id) -> std::shared_ptr<ShaderFunctionIR> {
+            std::shared_ptr<ShaderFunctionIR> result;
+            if (decompiled) {
+                result = std::make_shared<ShaderFunctionIR>(std::move(program_manager), id,
+                                                            shader_info.start, shader_info.end);
+            } else {
+                result =
+                    std::make_shared<ShaderFunctionIR>(std::move(basic_blocks), disable_flow_stack,
+                                                       id, shader_info.start, shader_info.end);
+            }
+            decompiled = false;
+            disable_flow_stack = false;
+            basic_blocks.clear();
+            program_manager.Clear();
+            return result;
+        });
     std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header));
 
     decompiled = false;
     auto info = ScanFlow(program_code, main_offset, settings, registry);
-    auto& shader_info = *info;
-    coverage_begin = shader_info.start;
-    coverage_end = shader_info.end;
-    switch (shader_info.settings.depth) {
-    case CompileDepth::FlowStack: {
-        for (const auto& block : shader_info.blocks) {
-            basic_blocks.insert({block.start, DecodeRange(block.start, block.end + 1)});
-        }
-        break;
+    u32 id_start = 1;
+    for (auto& pair : info->subfunctions) {
+        func_map.emplace(pair.first, id_start);
+        id_start++;
     }
-    case CompileDepth::NoFlowStack: {
-        disable_flow_stack = true;
-        const auto insert_block = [this](NodeBlock& nodes, u32 label) {
-            if (label == static_cast<u32>(exit_branch)) {
-                return;
-            }
-            basic_blocks.insert({label, nodes});
-        };
-        const auto& blocks = shader_info.blocks;
-        NodeBlock current_block;
-        u32 current_label = static_cast<u32>(exit_branch);
-        for (const auto& block : blocks) {
-            if (shader_info.labels.contains(block.start)) {
-                insert_block(current_block, current_label);
-                current_block.clear();
-                current_label = block.start;
-            }
-            if (!block.ignore_branch) {
-                DecodeRangeInner(current_block, block.start, block.end);
-                InsertControlFlow(current_block, block);
-            } else {
-                DecodeRangeInner(current_block, block.start, block.end + 1);
-            }
-        }
-        insert_block(current_block, current_label);
-        break;
-    }
-    case CompileDepth::DecompileBackwards:
-    case CompileDepth::FullDecompile: {
-        program_manager = std::move(shader_info.manager);
-        disable_flow_stack = true;
-        decompiled = true;
-        ASTDecoder decoder{*this};
-        ASTNode program = GetASTProgram();
-        decoder.Visit(program);
-        break;
-    }
-    default:
-        LOG_CRITICAL(HW_GPU, "Unknown decompilation mode!");
-        [[fallthrough]];
-    case CompileDepth::BruteForce: {
-        const auto shader_end = static_cast<u32>(program_code.size());
-        coverage_begin = main_offset;
-        coverage_end = shader_end;
-        for (u32 label = main_offset; label < shader_end; ++label) {
-            basic_blocks.insert({label, DecodeRange(label, label + 1)});
-        }
-        break;
-    }
-    }
-    if (settings.depth != shader_info.settings.depth) {
-        LOG_WARNING(
-            HW_GPU, "Decompiling to this setting \"{}\" failed, downgrading to this setting \"{}\"",
-            CompileDepthAsString(settings.depth), CompileDepthAsString(shader_info.settings.depth));
+    coverage_begin = info->main.start;
+    coverage_end = 0;
+    decode_function(info->main);
+    main_function = gen_function(info->main, 0);
+    subfunctions.resize(info->subfunctions.size());
+    for (auto& pair : info->subfunctions) {
+        auto& func_info = pair.second;
+        decode_function(func_info);
+        u32 id = func_map[pair.first];
+        subfunctions[id - 1] = gen_function(func_info, id);
     }
 }
 

src/video_core/shader/decode/other.cpp

@@ -33,6 +33,7 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         // With the previous preconditions, this instruction is a no-operation.
         break;
     }
+    case OpCode::Id::RET:
     case OpCode::Id::EXIT: {
         const ConditionCode cc = instr.flow_condition_code;
         UNIMPLEMENTED_IF_MSG(cc != ConditionCode::T, "EXIT condition code used: {}", cc);
@@ -312,6 +313,16 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         LOG_DEBUG(HW_GPU, "DEPBAR instruction is stubbed");
         break;
     }
+    case OpCode::Id::CAL: {
+        const u32 target = pc + instr.bra.GetBranchTarget();
+        const auto it = func_map.find(target);
+        if (it == func_map.end()) {
+            UNREACHABLE();
+            break;
+        }
+        bb.push_back(FunctionCall(it->second));
+        break;
+    }
     default:
         UNIMPLEMENTED_MSG("Unhandled instruction: {}", opcode->get().GetName());
     }

src/video_core/shader/node.h

@@ -267,10 +267,11 @@ class PatchNode;
 class SmemNode;
 class GmemNode;
 class CommentNode;
+class FunctionCallNode;
 
 using NodeData = std::variant<OperationNode, ConditionalNode, GprNode, CustomVarNode, ImmediateNode,
                               InternalFlagNode, PredicateNode, AbufNode, PatchNode, CbufNode,
-                              LmemNode, SmemNode, GmemNode, CommentNode>;
+                              LmemNode, SmemNode, GmemNode, FunctionCallNode, CommentNode>;
 using Node = std::shared_ptr<NodeData>;
 using Node4 = std::array<Node, 4>;
 using NodeBlock = std::vector<Node>;
@@ -494,6 +495,18 @@ private:
     std::vector<Node> code; ///< Code to execute
 };
 
+class FunctionCallNode final : public AmendNode {
+public:
+    explicit FunctionCallNode(u32 func_id_) : func_id{func_id_} {}
+
+    [[nodiscard]] u32 GetFuncId() const {
+        return func_id;
+    }
+
+private:
+    u32 func_id; ///< Id of the function to call
+};
+
 /// A general purpose register
 class GprNode final {
 public:

src/video_core/shader/node_helper.cpp

@@ -19,6 +19,11 @@ Node Comment(std::string text) {
     return MakeNode<CommentNode>(std::move(text));
 }
 
+/// Creates a function call
+Node FunctionCall(u32 func_id) {
+    return MakeNode<FunctionCallNode>(func_id);
+}
+
 Node Immediate(u32 value) {
     return MakeNode<ImmediateNode>(value);
 }

src/video_core/shader/node_helper.h

@@ -27,6 +27,9 @@ Node Conditional(Node condition, std::vector<Node> code);
 /// Creates a commentary node
 Node Comment(std::string text);
 
+/// Creates a function call
+Node FunctionCall(u32 func_id);
+
 /// Creates an u32 immediate
 Node Immediate(u32 value);
 

src/video_core/shader/shader_ir.h

@@ -26,7 +26,7 @@ namespace VideoCommon::Shader {
 
 struct ShaderBlock;
 
-constexpr u32 MAX_PROGRAM_LENGTH = 0x1000;
+constexpr u32 MAX_PROGRAM_LENGTH = 0x2000;
 
 struct ConstBuffer {
     constexpr explicit ConstBuffer(u32 max_offset_, bool is_indirect_)
@@ -64,16 +64,68 @@ struct GlobalMemoryUsage {
     bool is_written{};
 };
 
+class ShaderFunctionIR final {
+public:
+    explicit ShaderFunctionIR(std::map<u32, NodeBlock>&& basic_blocks_, bool disable_flow_stack_,
+                              u32 id_, u32 coverage_begin_, u32 coverage_end_)
+        : basic_blocks{std::move(basic_blocks_)}, decompiled{false},
+          disable_flow_stack{disable_flow_stack_}, id{id_}, coverage_begin{coverage_begin_},
+          coverage_end{coverage_end_} {}
+    explicit ShaderFunctionIR(ASTManager&& program_manager_, u32 id_, u32 coverage_begin_,
+                              u32 coverage_end_)
+        : program_manager{std::move(program_manager_)}, decompiled{true}, disable_flow_stack{true},
+          id{id_}, coverage_begin{coverage_begin_}, coverage_end{coverage_end_} {}
+
+    const std::map<u32, NodeBlock>& GetBasicBlocks() const {
+        return basic_blocks;
+    }
+
+    [[nodiscard]] bool IsFlowStackDisabled() const {
+        return disable_flow_stack;
+    }
+
+    [[nodiscard]] bool IsDecompiled() const {
+        return decompiled;
+    }
+
+    const ASTManager& GetASTManager() const {
+        return program_manager;
+    }
+
+    [[nodiscard]] ASTNode GetASTProgram() const {
+        return program_manager.GetProgram();
+    }
+
+    [[nodiscard]] u32 GetASTNumVariables() const {
+        return program_manager.GetVariables();
+    }
+
+    [[nodiscard]] bool IsMain() const {
+        return id == 0;
+    }
+
+    [[nodiscard]] u32 GetId() const {
+        return id;
+    }
+
+private:
+    std::map<u32, NodeBlock> basic_blocks;
+    ASTManager program_manager{true, true};
+
+    bool decompiled{};
+    bool disable_flow_stack{};
+    u32 id{};
+
+    u32 coverage_begin{};
+    u32 coverage_end{};
+};
+
 class ShaderIR final {
 public:
     explicit ShaderIR(const ProgramCode& program_code_, u32 main_offset_,
                       CompilerSettings settings_, Registry& registry_);
     ~ShaderIR();
 
-    const std::map<u32, NodeBlock>& GetBasicBlocks() const {
-        return basic_blocks;
-    }
-
     const std::set<u32>& GetRegisters() const {
         return used_registers;
     }
@@ -155,26 +207,6 @@ public:
         return header;
     }
 
-    bool IsFlowStackDisabled() const {
-        return disable_flow_stack;
-    }
-
-    bool IsDecompiled() const {
-        return decompiled;
-    }
-
-    const ASTManager& GetASTManager() const {
-        return program_manager;
-    }
-
-    ASTNode GetASTProgram() const {
-        return program_manager.GetProgram();
-    }
-
-    u32 GetASTNumVariables() const {
-        return program_manager.GetVariables();
-    }
-
     u32 ConvertAddressToNvidiaSpace(u32 address) const {
         return (address - main_offset) * static_cast<u32>(sizeof(Tegra::Shader::Instruction));
     }
@@ -190,7 +222,16 @@ public:
         return num_custom_variables;
     }
 
+    std::shared_ptr<ShaderFunctionIR> GetMainFunction() const {
+        return main_function;
+    }
+
+    const std::vector<std::shared_ptr<ShaderFunctionIR>>& GetSubFunctions() const {
+        return subfunctions;
+    }
+
 private:
+    friend class ExprDecoder;
     friend class ASTDecoder;
 
     struct SamplerInfo {
@@ -453,6 +494,10 @@ private:
     std::vector<Node> amend_code;
     u32 num_custom_variables{};
 
+    std::shared_ptr<ShaderFunctionIR> main_function;
+    std::vector<std::shared_ptr<ShaderFunctionIR>> subfunctions;
+    std::unordered_map<u32, u32> func_map;
+
     std::set<u32> used_registers;
     std::set<Tegra::Shader::Pred> used_predicates;
     std::set<Tegra::Shader::Attribute::Index> used_input_attributes;

src/video_core/vulkan_common/vulkan_memory_allocator.cpp

@@ -53,6 +53,18 @@ struct Range {
     UNREACHABLE_MSG("Invalid memory usage={}", usage);
     return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
 }
+
+constexpr VkExportMemoryAllocateInfo EXPORT_ALLOCATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO,
+    .pNext = nullptr,
+#ifdef _WIN32
+    .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT,
+#elif __unix__
+    .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT,
+#else
+    .handleTypes = 0,
+#endif
+};
 } // Anonymous namespace
 
 class MemoryAllocation {
@@ -131,7 +143,7 @@ public:
 
     /// Returns whether this allocation is compatible with the arguments.
     [[nodiscard]] bool IsCompatible(VkMemoryPropertyFlags flags, u32 type_mask) const {
-        return (flags & property_flags) && (type_mask & shifted_memory_type) != 0;
+        return (flags & property_flags) == property_flags && (type_mask & shifted_memory_type) != 0;
     }
 
 private:
@@ -217,14 +229,18 @@ MemoryAllocator::~MemoryAllocator() = default;
 
 MemoryCommit MemoryAllocator::Commit(const VkMemoryRequirements& requirements, MemoryUsage usage) {
     // Find the fastest memory flags we can afford with the current requirements
-    const VkMemoryPropertyFlags flags = MemoryPropertyFlags(requirements.memoryTypeBits, usage);
+    const u32 type_mask = requirements.memoryTypeBits;
+    const VkMemoryPropertyFlags usage_flags = MemoryUsagePropertyFlags(usage);
+    const VkMemoryPropertyFlags flags = MemoryPropertyFlags(type_mask, usage_flags);
     if (std::optional<MemoryCommit> commit = TryCommit(requirements, flags)) {
         return std::move(*commit);
     }
     // Commit has failed, allocate more memory.
-    // TODO(Rodrigo): Handle out of memory situations in some way like flushing to guest memory.
-    AllocMemory(flags, requirements.memoryTypeBits, AllocationChunkSize(requirements.size));
-
+    const u64 chunk_size = AllocationChunkSize(requirements.size);
+    if (!TryAllocMemory(flags, type_mask, chunk_size)) {
+        // TODO(Rodrigo): Handle out of memory situations in some way like flushing to guest memory.
+        throw vk::Exception(VK_ERROR_OUT_OF_DEVICE_MEMORY);
+    }
     // Commit again, this time it won't fail since there's a fresh allocation above.
     // If it does, there's a bug.
     return TryCommit(requirements, flags).value();
@@ -242,26 +258,25 @@ MemoryCommit MemoryAllocator::Commit(const vk::Image& image, MemoryUsage usage)
     return commit;
 }
 
-void MemoryAllocator::AllocMemory(VkMemoryPropertyFlags flags, u32 type_mask, u64 size) {
+bool MemoryAllocator::TryAllocMemory(VkMemoryPropertyFlags flags, u32 type_mask, u64 size) {
     const u32 type = FindType(flags, type_mask).value();
-    const VkExportMemoryAllocateInfo export_allocate_info{
-        .sType = VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO,
-        .pNext = nullptr,
-#ifdef _WIN32
-        .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT,
-#elif __unix__
-        .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT,
-#else
-        .handleTypes = 0,
-#endif
-    };
-    vk::DeviceMemory memory = device.GetLogical().AllocateMemory({
+    vk::DeviceMemory memory = device.GetLogical().TryAllocateMemory({
         .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
-        .pNext = export_allocations ? &export_allocate_info : nullptr,
+        .pNext = export_allocations ? &EXPORT_ALLOCATE_INFO : nullptr,
         .allocationSize = size,
         .memoryTypeIndex = type,
     });
+    if (!memory) {
+        if ((flags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) != 0) {
+            // Try to allocate non device local memory
+            return TryAllocMemory(flags & ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, type_mask, size);
+        } else {
+            // RIP
+            return false;
+        }
+    }
     allocations.push_back(std::make_unique<MemoryAllocation>(std::move(memory), flags, size, type));
+    return true;
 }
 
 std::optional<MemoryCommit> MemoryAllocator::TryCommit(const VkMemoryRequirements& requirements,
@@ -274,24 +289,24 @@ std::optional<MemoryCommit> MemoryAllocator::TryCommit(const VkMemoryRequirement
             return commit;
         }
     }
+    if ((flags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) != 0) {
+        // Look for non device local commits on failure
+        return TryCommit(requirements, flags & ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
+    }
     return std::nullopt;
 }
 
-VkMemoryPropertyFlags MemoryAllocator::MemoryPropertyFlags(u32 type_mask, MemoryUsage usage) const {
-    return MemoryPropertyFlags(type_mask, MemoryUsagePropertyFlags(usage));
-}
-
 VkMemoryPropertyFlags MemoryAllocator::MemoryPropertyFlags(u32 type_mask,
                                                            VkMemoryPropertyFlags flags) const {
     if (FindType(flags, type_mask)) {
         // Found a memory type with those requirements
         return flags;
     }
-    if (flags & VK_MEMORY_PROPERTY_HOST_CACHED_BIT) {
+    if ((flags & VK_MEMORY_PROPERTY_HOST_CACHED_BIT) != 0) {
         // Remove host cached bit in case it's not supported
         return MemoryPropertyFlags(type_mask, flags & ~VK_MEMORY_PROPERTY_HOST_CACHED_BIT);
     }
-    if (flags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) {
+    if ((flags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) != 0) {
         // Remove device local, if it's not supported by the requested resource
         return MemoryPropertyFlags(type_mask, flags & ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
     }
@@ -302,7 +317,7 @@ VkMemoryPropertyFlags MemoryAllocator::MemoryPropertyFlags(u32 type_mask,
 std::optional<u32> MemoryAllocator::FindType(VkMemoryPropertyFlags flags, u32 type_mask) const {
     for (u32 type_index = 0; type_index < properties.memoryTypeCount; ++type_index) {
         const VkMemoryPropertyFlags type_flags = properties.memoryTypes[type_index].propertyFlags;
-        if ((type_mask & (1U << type_index)) && (type_flags & flags)) {
+        if ((type_mask & (1U << type_index)) != 0 && (type_flags & flags) == flags) {
             // The type matches in type and in the wanted properties.
             return type_index;
         }

src/video_core/vulkan_common/vulkan_memory_allocator.h

@@ -101,16 +101,13 @@ public:
     MemoryCommit Commit(const vk::Image& image, MemoryUsage usage);
 
 private:
-    /// Allocates a chunk of memory.
-    void AllocMemory(VkMemoryPropertyFlags flags, u32 type_mask, u64 size);
+    /// Tries to allocate a chunk of memory.
+    bool TryAllocMemory(VkMemoryPropertyFlags flags, u32 type_mask, u64 size);
 
     /// Tries to allocate a memory commit.
     std::optional<MemoryCommit> TryCommit(const VkMemoryRequirements& requirements,
                                           VkMemoryPropertyFlags flags);
 
-    /// Returns the fastest compatible memory property flags from a wanted usage.
-    VkMemoryPropertyFlags MemoryPropertyFlags(u32 type_mask, MemoryUsage usage) const;
-
     /// Returns the fastest compatible memory property flags from the wanted flags.
     VkMemoryPropertyFlags MemoryPropertyFlags(u32 type_mask, VkMemoryPropertyFlags flags) const;
 

src/yuzu/configuration/config.cpp

@@ -16,7 +16,7 @@
 
 namespace FS = Common::FS;
 
-Config::Config(const std::string& config_name, ConfigType config_type) : type(config_type) {
+Config::Config(std::string_view config_name, ConfigType config_type) : type(config_type) {
     global = config_type == ConfigType::GlobalConfig;
 
     Initialize(config_name);
@@ -242,7 +242,7 @@ const std::array<UISettings::Shortcut, 17> Config::default_hotkeys{{
 }};
 // clang-format on
 
-void Config::Initialize(const std::string& config_name) {
+void Config::Initialize(std::string_view config_name) {
     const auto fs_config_loc = FS::GetYuzuPath(FS::YuzuPath::ConfigDir);
     const auto config_file = fmt::format("{}.ini", config_name);
 

src/yuzu/configuration/config.h

@@ -22,7 +22,7 @@ public:
         InputProfile,
     };
 
-    explicit Config(const std::string& config_name = "qt-config",
+    explicit Config(std::string_view config_name = "qt-config",
                     ConfigType config_type = ConfigType::GlobalConfig);
     ~Config();
 
@@ -45,7 +45,7 @@ public:
     static const std::array<UISettings::Shortcut, 17> default_hotkeys;
 
 private:
-    void Initialize(const std::string& config_name);
+    void Initialize(std::string_view config_name);
 
     void ReadValues();
     void ReadPlayerValue(std::size_t player_index);

src/yuzu/configuration/configure_per_game.cpp

@@ -4,6 +4,7 @@
 
 #include <algorithm>
 #include <memory>
+#include <string>
 #include <utility>
 
 #include <QAbstractButton>
@@ -17,6 +18,7 @@
 #include <QTimer>
 #include <QTreeView>
 
+#include "common/fs/path_util.h"
 #include "core/core.h"
 #include "core/file_sys/control_metadata.h"
 #include "core/file_sys/patch_manager.h"
@@ -29,10 +31,11 @@
 #include "yuzu/uisettings.h"
 #include "yuzu/util/util.h"
 
-ConfigurePerGame::ConfigurePerGame(QWidget* parent, u64 title_id)
+ConfigurePerGame::ConfigurePerGame(QWidget* parent, u64 title_id, std::string_view file_name)
     : QDialog(parent), ui(std::make_unique<Ui::ConfigurePerGame>()), title_id(title_id) {
-    game_config = std::make_unique<Config>(fmt::format("{:016X}", title_id),
-                                           Config::ConfigType::PerGameConfig);
+    const auto config_file_name =
+        title_id == 0 ? Common::FS::GetFilename(file_name) : fmt::format("{:016X}", title_id);
+    game_config = std::make_unique<Config>(config_file_name, Config::ConfigType::PerGameConfig);
 
     Settings::SetConfiguringGlobal(false);
 

src/yuzu/configuration/configure_per_game.h

@@ -5,6 +5,7 @@
 #pragma once
 
 #include <memory>
+#include <string>
 #include <vector>
 
 #include <QDialog>
@@ -27,7 +28,7 @@ class ConfigurePerGame : public QDialog {
     Q_OBJECT
 
 public:
-    explicit ConfigurePerGame(QWidget* parent, u64 title_id);
+    explicit ConfigurePerGame(QWidget* parent, u64 title_id, std::string_view file_name);
     ~ConfigurePerGame() override;
 
     /// Save all button configurations to settings file

src/yuzu/game_list.cpp

@@ -561,11 +561,11 @@ void GameList::AddGamePopup(QMenu& context_menu, u64 program_id, const std::stri
     connect(remove_dlc, &QAction::triggered, [this, program_id]() {
         emit RemoveInstalledEntryRequested(program_id, InstalledEntryType::AddOnContent);
     });
-    connect(remove_shader_cache, &QAction::triggered, [this, program_id]() {
-        emit RemoveFileRequested(program_id, GameListRemoveTarget::ShaderCache);
+    connect(remove_shader_cache, &QAction::triggered, [this, program_id, path]() {
+        emit RemoveFileRequested(program_id, GameListRemoveTarget::ShaderCache, path);
     });
-    connect(remove_custom_config, &QAction::triggered, [this, program_id]() {
-        emit RemoveFileRequested(program_id, GameListRemoveTarget::CustomConfiguration);
+    connect(remove_custom_config, &QAction::triggered, [this, program_id, path]() {
+        emit RemoveFileRequested(program_id, GameListRemoveTarget::CustomConfiguration, path);
     });
     connect(dump_romfs, &QAction::triggered,
             [this, program_id, path]() { emit DumpRomFSRequested(program_id, path); });

src/yuzu/game_list.h

@@ -88,7 +88,8 @@ signals:
                              const std::string& game_path);
     void OpenTransferableShaderCacheRequested(u64 program_id);
     void RemoveInstalledEntryRequested(u64 program_id, InstalledEntryType type);
-    void RemoveFileRequested(u64 program_id, GameListRemoveTarget target);
+    void RemoveFileRequested(u64 program_id, GameListRemoveTarget target,
+                             std::string_view game_path);
     void DumpRomFSRequested(u64 program_id, const std::string& game_path);
     void CopyTIDRequested(u64 program_id);
     void NavigateToGamedbEntryRequested(u64 program_id,

src/yuzu/main.cpp

@@ -1334,7 +1334,10 @@ void GMainWindow::BootGame(const QString& filename, std::size_t program_index) {
 
     if (!(loader == nullptr || loader->ReadProgramId(title_id) != Loader::ResultStatus::Success)) {
         // Load per game settings
-        Config per_game_config(fmt::format("{:016X}", title_id), Config::ConfigType::PerGameConfig);
+        const auto config_file_name = title_id == 0
+                                          ? Common::FS::GetFilename(filename.toStdString())
+                                          : fmt::format("{:016X}", title_id);
+        Config per_game_config(config_file_name, Config::ConfigType::PerGameConfig);
     }
 
     ConfigureVibration::SetAllVibrationDevices();
@@ -1795,7 +1798,8 @@ void GMainWindow::RemoveAddOnContent(u64 program_id, const QString& entry_type)
                              tr("Successfully removed %1 installed DLC.").arg(count));
 }
 
-void GMainWindow::OnGameListRemoveFile(u64 program_id, GameListRemoveTarget target) {
+void GMainWindow::OnGameListRemoveFile(u64 program_id, GameListRemoveTarget target,
+                                       std::string_view game_path) {
     const QString question = [this, target] {
         switch (target) {
         case GameListRemoveTarget::ShaderCache:
@@ -1817,7 +1821,7 @@ void GMainWindow::OnGameListRemoveFile(u64 program_id, GameListRemoveTarget targ
         RemoveTransferableShaderCache(program_id);
         break;
     case GameListRemoveTarget::CustomConfiguration:
-        RemoveCustomConfiguration(program_id);
+        RemoveCustomConfiguration(program_id, game_path);
         break;
     }
 }
@@ -1842,9 +1846,12 @@ void GMainWindow::RemoveTransferableShaderCache(u64 program_id) {
     }
 }
 
-void GMainWindow::RemoveCustomConfiguration(u64 program_id) {
-    const auto custom_config_file_path = Common::FS::GetYuzuPath(Common::FS::YuzuPath::ConfigDir) /
-                                         "custom" / fmt::format("{:016X}.ini", program_id);
+void GMainWindow::RemoveCustomConfiguration(u64 program_id, std::string_view game_path) {
+    const auto config_file_name = program_id == 0
+                                      ? fmt::format("{:s}.ini", Common::FS::GetFilename(game_path))
+                                      : fmt::format("{:016X}.ini", program_id);
+    const auto custom_config_file_path =
+        Common::FS::GetYuzuPath(Common::FS::YuzuPath::ConfigDir) / "custom" / config_file_name;
 
     if (!Common::FS::Exists(custom_config_file_path)) {
         QMessageBox::warning(this, tr("Error Removing Custom Configuration"),
@@ -2635,7 +2642,7 @@ void GMainWindow::OpenPerGameConfiguration(u64 title_id, const std::string& file
     const auto v_file = Core::GetGameFileFromPath(vfs, file_name);
     const auto& system = Core::System::GetInstance();
 
-    ConfigurePerGame dialog(this, title_id);
+    ConfigurePerGame dialog(this, title_id, file_name);
     dialog.LoadFromFile(v_file);
     const auto result = dialog.exec();
 

src/yuzu/main.h

@@ -236,7 +236,8 @@ private slots:
                               const std::string& game_path);
     void OnTransferableShaderCacheOpenFile(u64 program_id);
     void OnGameListRemoveInstalledEntry(u64 program_id, InstalledEntryType type);
-    void OnGameListRemoveFile(u64 program_id, GameListRemoveTarget target);
+    void OnGameListRemoveFile(u64 program_id, GameListRemoveTarget target,
+                              std::string_view game_path);
     void OnGameListDumpRomFS(u64 program_id, const std::string& game_path);
     void OnGameListCopyTID(u64 program_id);
     void OnGameListNavigateToGamedbEntry(u64 program_id,
@@ -275,7 +276,7 @@ private:
     void RemoveUpdateContent(u64 program_id, const QString& entry_type);
     void RemoveAddOnContent(u64 program_id, const QString& entry_type);
     void RemoveTransferableShaderCache(u64 program_id);
-    void RemoveCustomConfiguration(u64 program_id);
+    void RemoveCustomConfiguration(u64 program_id, std::string_view game_path);
     std::optional<u64> SelectRomFSDumpTarget(const FileSys::ContentProvider&, u64 program_id);
     InstallResult InstallNSPXCI(const QString& filename);
     InstallResult InstallNCA(const QString& filename);