undo revert of style improvement

This reverts commit 372c314f06.
It broke formatting in an attempt to fix GCC builds.
Some other change (perhaps dependency updates) has resolved the issue.

.github/workflows/build.yml

@@ -39,7 +39,7 @@ jobs:
     - name: "Install system dependencies"
       run: |
         sudo apt update -qq
-        sudo apt install -y clang-15 cmake freeglut3-dev libgcrypt20-dev libglm-dev libgtk-3-dev libpulse-dev libsecret-1-dev libsystemd-dev libudev-dev nasm ninja-build
+        sudo apt install -y clang-15 cmake freeglut3-dev libgcrypt20-dev libglm-dev libgtk-3-dev libpulse-dev libsecret-1-dev libsystemd-dev libudev-dev nasm ninja-build libbluetooth-dev
 
     - name: "Setup cmake"
       uses: jwlawson/actions-setup-cmake@v2
@@ -96,7 +96,7 @@ jobs:
     - name: "Install system dependencies"
       run: |
         sudo apt update -qq
-        sudo apt install -y clang-15 cmake freeglut3-dev libgcrypt20-dev libglm-dev libgtk-3-dev libpulse-dev libsecret-1-dev libsystemd-dev nasm ninja-build appstream
+        sudo apt install -y clang-15 cmake freeglut3-dev libgcrypt20-dev libglm-dev libgtk-3-dev libpulse-dev libsecret-1-dev libsystemd-dev nasm ninja-build appstream libbluetooth-dev
 
     - name: "Build AppImage"
       run: |

.github/workflows/deploy_release.yml

@@ -1,4 +1,4 @@
-name: Deploy experimental release
+name: Deploy release
 on: 
   workflow_dispatch:
     inputs:
@@ -54,7 +54,7 @@ jobs:
       next_version_major: ${{ needs.calculate-version.outputs.next_version_major }}
       next_version_minor: ${{ needs.calculate-version.outputs.next_version_minor }}
   deploy:
-    name: Deploy experimental release
+    name: Deploy release
     runs-on: ubuntu-22.04
     needs: [call-release-build, calculate-version]
     steps:

.github/workflows/deploy_stable_release.yml

@@ -1,85 +0,0 @@
-name: Create new release
-on: 
-  workflow_dispatch:
-    inputs:
-      PlaceholderInput:
-        description: PlaceholderInput
-        required: false
-jobs:
-  call-release-build:
-    uses: ./.github/workflows/build.yml
-    with:
-      deploymode: release
-  deploy:
-    name: Deploy release
-    runs-on: ubuntu-20.04
-    needs: call-release-build
-    steps:
-      - uses: actions/checkout@v3
- 
-      - uses: actions/download-artifact@v4
-        with:
-          name: cemu-bin-linux-x64
-          path: cemu-bin-linux-x64
-
-      - uses: actions/download-artifact@v4
-        with:
-          name: cemu-appimage-x64
-          path: cemu-appimage-x64
-          
-      - uses: actions/download-artifact@v4
-        with:
-          name: cemu-bin-windows-x64
-          path: cemu-bin-windows-x64
-
-      - uses: actions/download-artifact@v4
-        with:
-          name: cemu-bin-macos-x64
-          path: cemu-bin-macos-x64
-
-      - name: Initialize
-        run: |
-          mkdir upload
-          sudo apt update -qq
-          sudo apt install -y zip
-          
-      - name: Get Cemu release version
-        run: |
-          gcc -o getversion .github/getversion.cpp
-          echo "Cemu CI version: $(./getversion)"
-          echo "CEMU_FOLDER_NAME=Cemu_$(./getversion)" >> $GITHUB_ENV
-          echo "CEMU_VERSION=$(./getversion)" >> $GITHUB_ENV
-          
-      - name: Create release from windows-bin
-        run: |
-          ls ./
-          ls ./bin/
-          cp -R ./bin ./${{ env.CEMU_FOLDER_NAME }}
-          mv cemu-bin-windows-x64/Cemu.exe ./${{ env.CEMU_FOLDER_NAME }}/Cemu.exe
-          zip -9 -r upload/cemu-${{ env.CEMU_VERSION }}-windows-x64.zip ${{ env.CEMU_FOLDER_NAME }}
-          rm -r ./${{ env.CEMU_FOLDER_NAME }}
-
-      - name: Create appimage
-        run: |
-          VERSION=${{ env.CEMU_VERSION }}
-          echo "Cemu Version is $VERSION"
-          ls cemu-appimage-x64
-          mv cemu-appimage-x64/Cemu-*-x86_64.AppImage upload/Cemu-$VERSION-x86_64.AppImage
-          
-      - name: Create release from ubuntu-bin
-        run: |
-          ls ./
-          ls ./bin/
-          cp -R ./bin ./${{ env.CEMU_FOLDER_NAME }}
-          mv cemu-bin-linux-x64/Cemu ./${{ env.CEMU_FOLDER_NAME }}/Cemu
-          zip -9 -r upload/cemu-${{ env.CEMU_VERSION }}-ubuntu-20.04-x64.zip ${{ env.CEMU_FOLDER_NAME }}
-          rm -r ./${{ env.CEMU_FOLDER_NAME }}
-          
-      - name: Create release from macos-bin
-        run: cp cemu-bin-macos-x64/Cemu.dmg upload/cemu-${{ env.CEMU_VERSION }}-macos-12-x64.dmg
-          
-      - name: Create release
-        run: |
-          wget -O ghr.tar.gz https://github.com/tcnksm/ghr/releases/download/v0.15.0/ghr_v0.15.0_linux_amd64.tar.gz
-          tar xvzf ghr.tar.gz; rm ghr.tar.gz
-          ghr_v0.15.0_linux_amd64/ghr -t ${{ secrets.GITHUB_TOKEN }} -n "Cemu ${{ env.CEMU_VERSION }}" -b "Changelog:" v${{ env.CEMU_VERSION }} ./upload

.github/workflows/generate_pot.yml

@@ -35,7 +35,7 @@ jobs:
           -o cemu.pot
 
       - name: Upload artifact
-        uses: actions/upload-artifact@v3
+        uses: actions/upload-artifact@v4
         with:
           name: POT file
           path: ./cemu.pot

BUILD.md

@@ -46,10 +46,10 @@ To compile Cemu, a recent enough compiler and STL with C++20 support is required
 ### Dependencies
 
 #### For Arch and derivatives:
-`sudo pacman -S --needed base-devel clang cmake freeglut git glm gtk3 libgcrypt libpulse libsecret linux-headers llvm nasm ninja systemd unzip zip`
+`sudo pacman -S --needed base-devel bluez-libs clang cmake freeglut git glm gtk3 libgcrypt libpulse libsecret linux-headers llvm nasm ninja systemd unzip zip`
 
 #### For Debian, Ubuntu and derivatives:
-`sudo apt install -y cmake curl clang-15 freeglut3-dev git libgcrypt20-dev libglm-dev libgtk-3-dev libpulse-dev libsecret-1-dev libsystemd-dev libtool nasm ninja-build`
+`sudo apt install -y cmake curl clang-15 freeglut3-dev git libbluetooth-dev libgcrypt20-dev libglm-dev libgtk-3-dev libpulse-dev libsecret-1-dev libsystemd-dev libtool nasm ninja-build`
 
 You may also need to install `libusb-1.0-0-dev` as a workaround for an issue with the vcpkg hidapi package.
 
@@ -57,7 +57,7 @@ At Step 3 in [Build Cemu using cmake and clang](#build-cemu-using-cmake-and-clan
    `cmake -S . -B build -DCMAKE_BUILD_TYPE=release -DCMAKE_C_COMPILER=/usr/bin/clang-15 -DCMAKE_CXX_COMPILER=/usr/bin/clang++-15 -G Ninja -DCMAKE_MAKE_PROGRAM=/usr/bin/ninja`
 
 #### For Fedora and derivatives:
-`sudo dnf install clang cmake cubeb-devel freeglut-devel git glm-devel gtk3-devel kernel-headers libgcrypt-devel libsecret-devel libtool libusb1-devel llvm nasm ninja-build perl-core systemd-devel zlib-devel zlib-static`
+`sudo dnf install bluez-libs-devel clang cmake cubeb-devel freeglut-devel git glm-devel gtk3-devel kernel-headers libgcrypt-devel libsecret-devel libtool libusb1-devel llvm nasm ninja-build perl-core systemd-devel wayland-protocols-devel zlib-devel zlib-static`
 
 ### Build Cemu
 
@@ -120,6 +120,9 @@ This section refers to running `cmake -S...` (truncated).
 * Compiling failed during rebuild after `git pull` with an error that mentions RPATH
    * Add the following and try running the command again:
       * `-DCMAKE_BUILD_WITH_INSTALL_RPATH=ON`
+* Environment variable `VCPKG_FORCE_SYSTEM_BINARIES` must be set.
+   * Execute the folowing and then try running the command again:
+      * `export VCPKG_FORCE_SYSTEM_BINARIES=1`
 * If you are getting a random error, read the [package-name-and-platform]-out.log and [package-name-and-platform]-err.log for the actual reason to see if you might be lacking the headers from a dependency.
 
 
@@ -189,3 +192,41 @@ Then install the dependencies:
 
 If CMake complains about Cemu already being compiled or another similar error, try deleting the `CMakeCache.txt` file inside the `build` folder and retry building.
 
+## CMake configure flags
+Some flags can be passed during CMake configure to customise which features are enabled on build.
+
+Example usage: `cmake -S . -B build -DCMAKE_BUILD_TYPE=release -DENABLE_SDL=ON -DENABLE_VULKAN=OFF`
+
+### All platforms
+| Flag               |   | Description                                                                 | Default | Note               |
+|--------------------|:--|-----------------------------------------------------------------------------|---------|--------------------|
+| ALLOW_PORTABLE     |   | Allow Cemu to use the `portable` directory to store configs and data        | ON      |                    |
+| CEMU_CXX_FLAGS     |   | Flags passed straight to the compiler, e.g. `-march=native`, `-Wall`, `/W3` | ""      |                    |
+| ENABLE_CUBEB       |   | Enable cubeb audio backend                                                  | ON      |                    |
+| ENABLE_DISCORD_RPC |   | Enable Discord Rich presence support                                        | ON      |                    |
+| ENABLE_OPENGL      |   | Enable OpenGL graphics backend                                              | ON      | Currently required |
+| ENABLE_HIDAPI      |   | Enable HIDAPI (used for Wiimote controller API)                             | ON      |                    |
+| ENABLE_SDL         |   | Enable SDLController controller API                                         | ON      | Currently required |
+| ENABLE_VCPKG       |   | Use VCPKG package manager to obtain dependencies                            | ON      |                    |
+| ENABLE_VULKAN      |   | Enable the Vulkan graphics backend                                          | ON      |                    |
+| ENABLE_WXWIDGETS   |   | Enable wxWidgets UI                                                         | ON      | Currently required |
+
+### Windows
+| Flag               | Description                       | Default | Note               |
+|--------------------|-----------------------------------|---------|--------------------|
+| ENABLE_DIRECTAUDIO | Enable DirectAudio audio backend  | ON      | Currently required |
+| ENABLE_DIRECTINPUT | Enable DirectInput controller API | ON      | Currently required |
+| ENABLE_XAUDIO      | Enable XAudio audio backend       | ON      |                    |
+| ENABLE_XINPUT      | Enable XInput controller API      | ON      |                    |
+
+### Linux
+| Flag                  | Description                                        | Default |
+|-----------------------|----------------------------------------------------|---------|
+| ENABLE_BLUEZ          | Build with Bluez (used for Wiimote controller API) | ON      |
+| ENABLE_FERAL_GAMEMODE | Enable Feral Interactive GameMode support          | ON      |
+| ENABLE_WAYLAND        | Enable Wayland support                             | ON      |
+
+### macOS
+| Flag         | Description                                    | Default |
+|--------------|------------------------------------------------|---------|
+| MACOS_BUNDLE | MacOS executable will be an application bundle | OFF     |

CMakeLists.txt

@@ -2,6 +2,7 @@ cmake_minimum_required(VERSION 3.21.1)
 
 option(ENABLE_VCPKG "Enable the vcpkg package manager" ON)
 option(MACOS_BUNDLE "The executable when built on macOS will be created as an application bundle" OFF)
+option(ALLOW_PORTABLE "Allow Cemu to be run in portable mode" ON)
 
 # used by CI script to set version:
 set(EMULATOR_VERSION_MAJOR "0" CACHE STRING "")
@@ -98,6 +99,7 @@ endif()
 if (UNIX AND NOT APPLE)
 	option(ENABLE_WAYLAND "Build with Wayland support" ON)
 	option(ENABLE_FERAL_GAMEMODE "Enables Feral Interactive GameMode Support" ON)
+	option(ENABLE_BLUEZ "Build with Bluez support" ON)
 endif()
 
 option(ENABLE_OPENGL "Enables the OpenGL backend" ON)
@@ -122,23 +124,6 @@ if (WIN32)
 endif()
 option(ENABLE_CUBEB "Enabled cubeb backend" ON)
 
-# usb hid backends
-if (WIN32)
-	option(ENABLE_NSYSHID_WINDOWS_HID "Enables the native Windows HID backend for nsyshid" ON)
-endif ()
-# libusb and windows hid backends shouldn't be active at the same time; otherwise we'd see all devices twice!
-if (NOT ENABLE_NSYSHID_WINDOWS_HID)
-	option(ENABLE_NSYSHID_LIBUSB "Enables the libusb backend for nsyshid" ON)
-else ()
-	set(ENABLE_NSYSHID_LIBUSB OFF CACHE BOOL "" FORCE)
-endif ()
-if (ENABLE_NSYSHID_WINDOWS_HID)
-	add_compile_definitions(NSYSHID_ENABLE_BACKEND_WINDOWS_HID)
-endif ()
-if (ENABLE_NSYSHID_LIBUSB)
-	add_compile_definitions(NSYSHID_ENABLE_BACKEND_LIBUSB)
-endif ()
-
 option(ENABLE_WXWIDGETS "Build with wxWidgets UI (Currently required)" ON)
 
 set(THREADS_PREFER_PTHREAD_FLAG true)
@@ -179,6 +164,12 @@ if (UNIX AND NOT APPLE)
 	endif()
 	find_package(GTK3 REQUIRED)
 
+	if(ENABLE_BLUEZ)
+		find_package(bluez REQUIRED)
+		set(ENABLE_WIIMOTE ON)
+		add_compile_definitions(HAS_BLUEZ)
+	endif()
+
 endif()
 
 if (ENABLE_VULKAN)
@@ -222,7 +213,7 @@ if (ENABLE_CUBEB)
 		option(BUILD_TOOLS "" OFF)
 		option(BUNDLE_SPEEX "" OFF)
 		set(USE_WINMM OFF CACHE BOOL "")
-		add_subdirectory("dependencies/cubeb" EXCLUDE_FROM_ALL)
+		add_subdirectory("dependencies/cubeb" EXCLUDE_FROM_ALL SYSTEM)
 		set_property(TARGET cubeb PROPERTY MSVC_RUNTIME_LIBRARY "MultiThreaded$<$<CONFIG:Debug>:Debug>")
 		add_library(cubeb::cubeb ALIAS cubeb)
 	endif()

boost.natvis

@@ -0,0 +1,26 @@
+<?xml version='1.0' encoding='utf-8'?>
+<AutoVisualizer xmlns="http://schemas.microsoft.com/vstudio/debugger/natvis/2010">
+
+    <Type Name="boost::container::small_vector&lt;*&gt;">
+        <Expand>
+            <Item Name="[size]">m_holder.m_size</Item>
+            <ArrayItems>
+                <Size>m_holder.m_size</Size>
+                <ValuePointer>m_holder.m_start</ValuePointer>
+            </ArrayItems>
+        </Expand>
+    </Type>
+
+	<Type Name="boost::container::static_vector&lt;*&gt;">
+	<DisplayString>{{ size={m_holder.m_size} }}</DisplayString>
+	<Expand>
+	  <Item Name="[size]" ExcludeView="simple">m_holder.m_size</Item>
+	  <Item Name="[capacity]" ExcludeView="simple">static_capacity</Item>
+	  <ArrayItems>
+		<Size>m_holder.m_size</Size>
+		<ValuePointer>($T1*)m_holder.storage.data</ValuePointer>
+	  </ArrayItems>
+	</Expand>
+	</Type>
+
+</AutoVisualizer>

cmake/Findbluez.cmake

@@ -0,0 +1,20 @@
+# SPDX-FileCopyrightText: 2022 Andrea Pappacoda <andrea@pappacoda.it>
+# SPDX-License-Identifier: ISC
+
+find_package(bluez CONFIG)
+if (NOT bluez_FOUND)
+  find_package(PkgConfig)
+  if (PKG_CONFIG_FOUND)
+    pkg_search_module(bluez IMPORTED_TARGET GLOBAL bluez-1.0 bluez)
+    if (bluez_FOUND)
+      add_library(bluez::bluez ALIAS PkgConfig::bluez)
+    endif ()
+  endif ()
+endif ()
+
+find_package_handle_standard_args(bluez
+    REQUIRED_VARS
+    bluez_LINK_LIBRARIES
+    bluez_FOUND
+    VERSION_VAR bluez_VERSION
+)

dependencies/vcpkg

@@ -1 +1 @@
-Subproject commit a4275b7eee79fb24ec2e135481ef5fce8b41c339
+Subproject commit 533a5fda5c0646d1771345fb572e759283444d5f

src/CMakeLists.txt

@@ -82,8 +82,8 @@ if (MACOS_BUNDLE)
 	set(MACOSX_BUNDLE_ICON_FILE "cemu.icns")
 	set(MACOSX_BUNDLE_GUI_IDENTIFIER "info.cemu.Cemu")
 	set(MACOSX_BUNDLE_BUNDLE_NAME "Cemu")
-	set(MACOSX_BUNDLE_SHORT_VERSION_STRING ${CMAKE_PROJECT_VERSION})
-	set(MACOSX_BUNDLE_BUNDLE_VERSION ${CMAKE_PROJECT_VERSION})
+	set(MACOSX_BUNDLE_SHORT_VERSION_STRING "${EMULATOR_VERSION_MAJOR}.${EMULATOR_VERSION_MINOR}.${EMULATOR_VERSION_PATCH}")
+	set(MACOSX_BUNDLE_BUNDLE_VERSION "${EMULATOR_VERSION_MAJOR}.${EMULATOR_VERSION_MINOR}.${EMULATOR_VERSION_PATCH}")
 	set(MACOSX_BUNDLE_COPYRIGHT "Copyright © 2024 Cemu Project")
 
 	set(MACOSX_BUNDLE_CATEGORY "public.app-category.games")
@@ -101,12 +101,18 @@ if (MACOS_BUNDLE)
 			COMMAND ${CMAKE_COMMAND} ARGS -E copy_directory "${CMAKE_SOURCE_DIR}/bin/${folder}" "${CMAKE_SOURCE_DIR}/bin/${OUTPUT_NAME}.app/Contents/SharedSupport/${folder}")
 	endforeach(folder)
 
+	if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+	    set(LIBUSB_PATH "${CMAKE_BINARY_DIR}/vcpkg_installed/x64-osx/debug/lib/libusb-1.0.0.dylib")
+	else()
+	    set(LIBUSB_PATH "${CMAKE_BINARY_DIR}/vcpkg_installed/x64-osx/lib/libusb-1.0.0.dylib")
+	endif()
+
 	add_custom_command (TARGET CemuBin POST_BUILD
 		COMMAND ${CMAKE_COMMAND} ARGS -E copy "/usr/local/lib/libMoltenVK.dylib" "${CMAKE_SOURCE_DIR}/bin/${OUTPUT_NAME}.app/Contents/Frameworks/libMoltenVK.dylib"
-		COMMAND ${CMAKE_COMMAND} ARGS -E copy "${CMAKE_BINARY_DIR}/vcpkg_installed/x64-osx/lib/libusb-1.0.0.dylib" "${CMAKE_SOURCE_DIR}/bin/${OUTPUT_NAME}.app/Contents/Frameworks/libusb-1.0.0.dylib"
+		COMMAND ${CMAKE_COMMAND} ARGS -E copy "${LIBUSB_PATH}" "${CMAKE_SOURCE_DIR}/bin/${OUTPUT_NAME}.app/Contents/Frameworks/libusb-1.0.0.dylib"
 		COMMAND ${CMAKE_COMMAND} ARGS -E copy "${CMAKE_SOURCE_DIR}/src/resource/update.sh" "${CMAKE_SOURCE_DIR}/bin/${OUTPUT_NAME}.app/Contents/MacOS/update.sh"
 		COMMAND bash -c "install_name_tool -add_rpath @executable_path/../Frameworks ${CMAKE_SOURCE_DIR}/bin/${OUTPUT_NAME}.app/Contents/MacOS/${OUTPUT_NAME}"
-		COMMAND bash -c "install_name_tool -change /Users/runner/work/Cemu/Cemu/build/vcpkg_installed/x64-osx/lib/libusb-1.0.0.dylib @executable_path/../Frameworks/libusb-1.0.0.dylib ${CMAKE_SOURCE_DIR}/bin/${OUTPUT_NAME}.app/Contents/MacOS/${OUTPUT_NAME}")
+		COMMAND bash -c "install_name_tool -change ${LIBUSB_PATH} @executable_path/../Frameworks/libusb-1.0.0.dylib ${CMAKE_SOURCE_DIR}/bin/${OUTPUT_NAME}.app/Contents/MacOS/${OUTPUT_NAME}")
 endif()
 
 set_target_properties(CemuBin PROPERTIES

src/Cafe/CMakeLists.txt

@@ -67,24 +67,31 @@ add_library(CemuCafe
   HW/Espresso/Recompiler/PPCFunctionBoundaryTracker.h
   HW/Espresso/Recompiler/PPCRecompiler.cpp
   HW/Espresso/Recompiler/PPCRecompiler.h
-  HW/Espresso/Recompiler/PPCRecompilerImlAnalyzer.cpp
+  HW/Espresso/Recompiler/IML/IML.h
+  HW/Espresso/Recompiler/IML/IMLSegment.cpp
+  HW/Espresso/Recompiler/IML/IMLSegment.h
+  HW/Espresso/Recompiler/IML/IMLInstruction.cpp
+  HW/Espresso/Recompiler/IML/IMLInstruction.h
+  HW/Espresso/Recompiler/IML/IMLDebug.cpp
+  HW/Espresso/Recompiler/IML/IMLAnalyzer.cpp
+  HW/Espresso/Recompiler/IML/IMLOptimizer.cpp
+  HW/Espresso/Recompiler/IML/IMLRegisterAllocator.cpp
+  HW/Espresso/Recompiler/IML/IMLRegisterAllocator.h
+  HW/Espresso/Recompiler/IML/IMLRegisterAllocatorRanges.cpp
+  HW/Espresso/Recompiler/IML/IMLRegisterAllocatorRanges.h
   HW/Espresso/Recompiler/PPCRecompilerImlGen.cpp
   HW/Espresso/Recompiler/PPCRecompilerImlGenFPU.cpp
   HW/Espresso/Recompiler/PPCRecompilerIml.h
-  HW/Espresso/Recompiler/PPCRecompilerImlOptimizer.cpp
-  HW/Espresso/Recompiler/PPCRecompilerImlRanges.cpp
-  HW/Espresso/Recompiler/PPCRecompilerImlRanges.h
-  HW/Espresso/Recompiler/PPCRecompilerImlRegisterAllocator2.cpp
-  HW/Espresso/Recompiler/PPCRecompilerImlRegisterAllocator.cpp
   HW/Espresso/Recompiler/PPCRecompilerIntermediate.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64AVX.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64BMI.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64FPU.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64Gen.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64GenFPU.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64.h
-  HW/Espresso/Recompiler/x64Emit.hpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64AVX.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64BMI.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64FPU.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64Gen.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64GenFPU.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64.h
+  HW/Espresso/Recompiler/BackendX64/X64Emit.hpp
+  HW/Espresso/Recompiler/BackendX64/x86Emitter.h
   HW/Latte/Common/RegisterSerializer.cpp
   HW/Latte/Common/RegisterSerializer.h
   HW/Latte/Common/ShaderSerializer.cpp
@@ -463,8 +470,8 @@ add_library(CemuCafe
   OS/libs/nsyshid/BackendEmulated.h
   OS/libs/nsyshid/BackendLibusb.cpp
   OS/libs/nsyshid/BackendLibusb.h
-  OS/libs/nsyshid/BackendWindowsHID.cpp
-  OS/libs/nsyshid/BackendWindowsHID.h
+  OS/libs/nsyshid/Dimensions.cpp
+  OS/libs/nsyshid/Dimensions.h
   OS/libs/nsyshid/Infinity.cpp
   OS/libs/nsyshid/Infinity.h
   OS/libs/nsyshid/Skylander.cpp
@@ -530,6 +537,12 @@ set_property(TARGET CemuCafe PROPERTY MSVC_RUNTIME_LIBRARY "MultiThreaded$<$<CON
 
 target_include_directories(CemuCafe PUBLIC "../")
 
+if (glslang_VERSION VERSION_LESS "15.0.0")
+	set(glslang_target "glslang::SPIRV")
+else()
+	set(glslang_target "glslang")
+endif()
+
 target_link_libraries(CemuCafe PRIVATE
 	CemuAsm
 	CemuAudio
@@ -545,7 +558,7 @@ target_link_libraries(CemuCafe PRIVATE
 	Boost::nowide
 	CURL::libcurl
 	fmt::fmt
-	glslang::SPIRV
+	${glslang_target}
 	ih264d
 	OpenSSL::Crypto
 	OpenSSL::SSL
@@ -561,15 +574,16 @@ if (ENABLE_WAYLAND)
   target_link_libraries(CemuCafe PUBLIC Wayland::Client)
 endif()
 
-if (ENABLE_NSYSHID_LIBUSB)
-	if (ENABLE_VCPKG)
-		find_package(PkgConfig REQUIRED)
-		pkg_check_modules(libusb REQUIRED IMPORTED_TARGET libusb-1.0)
-		target_link_libraries(CemuCafe PRIVATE PkgConfig::libusb)
-	else ()
-		find_package(libusb MODULE REQUIRED)
-		target_link_libraries(CemuCafe PRIVATE libusb::libusb)
-	endif ()
+if (ENABLE_VCPKG)
+  if(WIN32)
+    set(PKG_CONFIG_EXECUTABLE "${VCPKG_INSTALLED_DIR}/x64-windows/tools/pkgconf/pkgconf.exe")
+  endif()
+  find_package(PkgConfig REQUIRED)
+  pkg_check_modules(libusb REQUIRED IMPORTED_TARGET libusb-1.0)
+  target_link_libraries(CemuCafe PRIVATE PkgConfig::libusb)
+else ()
+  find_package(libusb MODULE REQUIRED)
+  target_link_libraries(CemuCafe PRIVATE libusb::libusb)
 endif ()
 
 if (ENABLE_WXWIDGETS)

src/Cafe/CafeSystem.cpp

@@ -9,6 +9,7 @@
 #include "audio/IAudioAPI.h"
 #include "audio/IAudioInputAPI.h"
 #include "config/ActiveSettings.h"
+#include "config/LaunchSettings.h"
 #include "Cafe/TitleList/GameInfo.h"
 #include "Cafe/GraphicPack/GraphicPack2.h"
 #include "util/helpers/SystemException.h"
@@ -396,7 +397,7 @@ void cemu_initForGame()
 		// replace any known function signatures with our HLE implementations and patch bugs in the games
 		GamePatch_scan();
 	}
-	LatteGPUState.alwaysDisplayDRC = ActiveSettings::DisplayDRCEnabled();
+	LatteGPUState.isDRCPrimary = ActiveSettings::DisplayDRCEnabled();
 	InfoLog_PrintActiveSettings();
 	Latte_Start();
 	// check for debugger entrypoint bp
@@ -637,40 +638,40 @@ namespace CafeSystem
         fsc_unmount("/cemuBossStorage/", FSC_PRIORITY_BASE);
     }
 
-	STATUS_CODE LoadAndMountForegroundTitle(TitleId titleId)
+	PREPARE_STATUS_CODE LoadAndMountForegroundTitle(TitleId titleId)
 	{
         cemuLog_log(LogType::Force, "Mounting title {:016x}", (uint64)titleId);
 		sGameInfo_ForegroundTitle = CafeTitleList::GetGameInfo(titleId);
 		if (!sGameInfo_ForegroundTitle.IsValid())
 		{
 			cemuLog_log(LogType::Force, "Mounting failed: Game meta information is either missing, inaccessible or not valid (missing or invalid .xml files in code and meta folder)");
-			return STATUS_CODE::UNABLE_TO_MOUNT;
+			return PREPARE_STATUS_CODE::UNABLE_TO_MOUNT;
 		}
 		// check base
 		TitleInfo& titleBase = sGameInfo_ForegroundTitle.GetBase();
 		if (!titleBase.IsValid())
-			return STATUS_CODE::UNABLE_TO_MOUNT;
+			return PREPARE_STATUS_CODE::UNABLE_TO_MOUNT;
 		if(!titleBase.ParseXmlInfo())
-			return STATUS_CODE::UNABLE_TO_MOUNT;
+			return PREPARE_STATUS_CODE::UNABLE_TO_MOUNT;
 		cemuLog_log(LogType::Force, "Base: {}", titleBase.GetPrintPath());
 		// mount base
 		if (!titleBase.Mount("/vol/content", "content", FSC_PRIORITY_BASE) || !titleBase.Mount(GetInternalVirtualCodeFolder(), "code", FSC_PRIORITY_BASE))
 		{
 			cemuLog_log(LogType::Force, "Mounting failed");
-			return STATUS_CODE::UNABLE_TO_MOUNT;
+			return PREPARE_STATUS_CODE::UNABLE_TO_MOUNT;
 		}
 		// check update
 		TitleInfo& titleUpdate = sGameInfo_ForegroundTitle.GetUpdate();
 		if (titleUpdate.IsValid())
 		{
 			if (!titleUpdate.ParseXmlInfo())
-				return STATUS_CODE::UNABLE_TO_MOUNT;
+				return PREPARE_STATUS_CODE::UNABLE_TO_MOUNT;
 			cemuLog_log(LogType::Force, "Update: {}", titleUpdate.GetPrintPath());
 			// mount update
 			if (!titleUpdate.Mount("/vol/content", "content", FSC_PRIORITY_PATCH) || !titleUpdate.Mount(GetInternalVirtualCodeFolder(), "code", FSC_PRIORITY_PATCH))
 			{
 				cemuLog_log(LogType::Force, "Mounting failed");
-				return STATUS_CODE::UNABLE_TO_MOUNT;
+				return PREPARE_STATUS_CODE::UNABLE_TO_MOUNT;
 			}
 		}
 		else
@@ -682,20 +683,20 @@ namespace CafeSystem
 			// todo - support for multi-title AOC
 			TitleInfo& titleAOC = aocList[0];
 			if (!titleAOC.ParseXmlInfo())
-				return STATUS_CODE::UNABLE_TO_MOUNT;
+				return PREPARE_STATUS_CODE::UNABLE_TO_MOUNT;
 			cemu_assert_debug(titleAOC.IsValid());
 			cemuLog_log(LogType::Force, "DLC: {}", titleAOC.GetPrintPath());
 			// mount AOC
 			if (!titleAOC.Mount(fmt::format("/vol/aoc{:016x}", titleAOC.GetAppTitleId()), "content", FSC_PRIORITY_PATCH))
 			{
 				cemuLog_log(LogType::Force, "Mounting failed");
-				return STATUS_CODE::UNABLE_TO_MOUNT;
+				return PREPARE_STATUS_CODE::UNABLE_TO_MOUNT;
 			}
 		}
 		else
 			cemuLog_log(LogType::Force, "DLC: Not present");
 		sForegroundTitleId = titleId;
-		return STATUS_CODE::SUCCESS;
+		return PREPARE_STATUS_CODE::SUCCESS;
 	}
 
     void UnmountForegroundTitle()
@@ -723,7 +724,7 @@ namespace CafeSystem
         }
     }
 
-	STATUS_CODE SetupExecutable()
+	PREPARE_STATUS_CODE SetupExecutable()
 	{
 		// set rpx path from cos.xml if available
 		_pathToBaseExecutable = _pathToExecutable;
@@ -755,7 +756,7 @@ namespace CafeSystem
 			}
 		}
 		LoadMainExecutable();
-		return STATUS_CODE::SUCCESS;
+		return PREPARE_STATUS_CODE::SUCCESS;
 	}
 
     void SetupMemorySpace()
@@ -769,7 +770,7 @@ namespace CafeSystem
         memory_unmapForCurrentTitle();
     }
 
-	STATUS_CODE PrepareForegroundTitle(TitleId titleId)
+	PREPARE_STATUS_CODE PrepareForegroundTitle(TitleId titleId)
 	{
 		CafeTitleList::WaitForMandatoryScan();
 		sLaunchModeIsStandalone = false;
@@ -780,21 +781,21 @@ namespace CafeSystem
         // mount mlc storage
         MountBaseDirectories();
         // mount title folders
-		STATUS_CODE r = LoadAndMountForegroundTitle(titleId);
-		if (r != STATUS_CODE::SUCCESS)
+		PREPARE_STATUS_CODE r = LoadAndMountForegroundTitle(titleId);
+		if (r != PREPARE_STATUS_CODE::SUCCESS)
 			return r;
 		gameProfile_load();
 		// setup memory space and PPC recompiler
         SetupMemorySpace();
         PPCRecompiler_init();
 		r = SetupExecutable(); // load RPX
-		if (r != STATUS_CODE::SUCCESS)
+		if (r != PREPARE_STATUS_CODE::SUCCESS)
 			return r;
 		InitVirtualMlcStorage();
-		return STATUS_CODE::SUCCESS;
+		return PREPARE_STATUS_CODE::SUCCESS;
 	}
 
-	STATUS_CODE PrepareForegroundTitleFromStandaloneRPX(const fs::path& path)
+	PREPARE_STATUS_CODE PrepareForegroundTitleFromStandaloneRPX(const fs::path& path)
 	{
 		sLaunchModeIsStandalone = true;
 		cemuLog_log(LogType::Force, "Launching executable in standalone mode due to incorrect layout or missing meta files");
@@ -812,7 +813,7 @@ namespace CafeSystem
 				if (!r)
 				{
 					cemuLog_log(LogType::Force, "Failed to mount {}", _pathToUtf8(contentPath));
-					return STATUS_CODE::UNABLE_TO_MOUNT;
+					return PREPARE_STATUS_CODE::UNABLE_TO_MOUNT;
 				}
 			}
 		}
@@ -824,7 +825,7 @@ namespace CafeSystem
 		// since a lot of systems (including save folder location) rely on a TitleId, we derive a placeholder id from the executable hash
 		auto execData = fsc_extractFile(_pathToExecutable.c_str());
 		if (!execData)
-			return STATUS_CODE::INVALID_RPX;
+			return PREPARE_STATUS_CODE::INVALID_RPX;
 		uint32 h = generateHashFromRawRPXData(execData->data(), execData->size());
 		sForegroundTitleId = 0xFFFFFFFF00000000ULL | (uint64)h;
 		cemuLog_log(LogType::Force, "Generated placeholder TitleId: {:016x}", sForegroundTitleId);
@@ -834,7 +835,7 @@ namespace CafeSystem
         // load executable
         SetupExecutable();
 		InitVirtualMlcStorage();
-		return STATUS_CODE::SUCCESS;
+		return PREPARE_STATUS_CODE::SUCCESS;
 	}
 
 	void _LaunchTitleThread()
@@ -843,7 +844,7 @@ namespace CafeSystem
 			module->TitleStart();
 		cemu_initForGame();
 		// enter scheduler
-		if (ActiveSettings::GetCPUMode() == CPUMode::MulticoreRecompiler)
+		if (ActiveSettings::GetCPUMode() == CPUMode::MulticoreRecompiler && !LaunchSettings::ForceInterpreter())
 			coreinit::OSSchedulerBegin(3);
 		else
 			coreinit::OSSchedulerBegin(1);

src/Cafe/CafeSystem.h

@@ -15,20 +15,19 @@ namespace CafeSystem
 		virtual void CafeRecreateCanvas() = 0;
 	};
 
-	enum class STATUS_CODE
+	enum class PREPARE_STATUS_CODE
 	{
 		SUCCESS,
 		INVALID_RPX,
 		UNABLE_TO_MOUNT, // failed to mount through TitleInfo (most likely caused by an invalid or outdated path)
-		//BAD_META_DATA, - the title list only stores titles with valid meta, so this error code is impossible
 	};
 
 	void Initialize();
 	void SetImplementation(SystemImplementation* impl);
     void Shutdown();
 
-	STATUS_CODE PrepareForegroundTitle(TitleId titleId);
-	STATUS_CODE PrepareForegroundTitleFromStandaloneRPX(const fs::path& path);
+	PREPARE_STATUS_CODE PrepareForegroundTitle(TitleId titleId);
+	PREPARE_STATUS_CODE PrepareForegroundTitleFromStandaloneRPX(const fs::path& path);
 	void LaunchForegroundTitle();
 	bool IsTitleRunning();
 

src/Cafe/Filesystem/FST/FST.cpp

@@ -3,8 +3,7 @@
 #include "Cemu/ncrypto/ncrypto.h"
 #include "Cafe/Filesystem/WUD/wud.h"
 #include "util/crypto/aes128.h"
-#include "openssl/evp.h" /* EVP_Digest */
-#include "openssl/sha.h" /* SHA1 / SHA256_DIGEST_LENGTH */
+#include "openssl/sha.h" /* SHA1 / SHA256 */
 #include "fstUtil.h"
 
 #include "FST.h"
@@ -141,7 +140,7 @@ struct DiscPartitionTableHeader
 	static constexpr uint32 MAGIC_VALUE = 0xCCA6E67B;
 
 	/* +0x00 */ uint32be magic;
-	/* +0x04 */ uint32be sectorSize; // must be 0x8000?
+	/* +0x04 */ uint32be blockSize; // must be 0x8000?
 	/* +0x08 */ uint8 partitionTableHash[20]; // hash of the data range at +0x800 to end of sector (0x8000)
 	/* +0x1C */ uint32be numPartitions;
 };
@@ -164,10 +163,10 @@ struct DiscPartitionHeader
 	static constexpr uint32 MAGIC_VALUE = 0xCC93A4F5;
 
 	/* +0x00 */ uint32be magic;
-	/* +0x04 */ uint32be sectorSize; // must match DISC_SECTOR_SIZE
+	/* +0x04 */ uint32be sectorSize; // must match DISC_SECTOR_SIZE for hashed blocks
 
 	/* +0x08 */ uint32be ukn008;
-	/* +0x0C */ uint32be ukn00C;
+	/* +0x0C */ uint32be ukn00C; // h3 array size?
 	/* +0x10 */ uint32be h3HashNum;
 	/* +0x14 */ uint32be fstSize; // in bytes
 	/* +0x18 */ uint32be fstSector; // relative to partition start
@@ -178,13 +177,15 @@ struct DiscPartitionHeader
 	/* +0x24 */ uint8 fstHashType;
 	/* +0x25 */ uint8 fstEncryptionType; // purpose of this isn't really understood. Maybe it controls which key is being used? (1 -> disc key, 2 -> partition key)
 
-	/* +0x26 */ uint8 versionA;
-	/* +0x27 */ uint8 ukn027; // also a version field?
+	/* +0x26 */ uint8be versionA;
+	/* +0x27 */ uint8be ukn027; // also a version field?
 
 	// there is an array at +0x40 ? Related to H3 list. Also related to value at +0x0C and h3HashNum
+	/* +0x28 */ uint8be _uknOrPadding028[0x18];
+	/* +0x40 */ uint8be h3HashArray[32]; // dynamic size. Only present if fstHashType != 0
 };
 
-static_assert(sizeof(DiscPartitionHeader) == 0x28);
+static_assert(sizeof(DiscPartitionHeader) == 0x40+0x20);
 
 bool FSTVolume::FindDiscKey(const fs::path& path, NCrypto::AesKey& discTitleKey)
 {
@@ -269,7 +270,7 @@ FSTVolume* FSTVolume::OpenFromDiscImage(const fs::path& path, NCrypto::AesKey& d
 		cemuLog_log(LogType::Force, "Disc image rejected because decryption failed");
 		return nullptr;
 	}
-	if (partitionHeader->sectorSize != DISC_SECTOR_SIZE)
+	if (partitionHeader->blockSize != DISC_SECTOR_SIZE)
 	{
 		cemuLog_log(LogType::Force, "Disc image rejected because partition sector size is invalid");
 		return nullptr;
@@ -336,6 +337,9 @@ FSTVolume* FSTVolume::OpenFromDiscImage(const fs::path& path, NCrypto::AesKey& d
 	cemu_assert_debug(partitionHeaderSI.fstEncryptionType == 1);
 	// todo - check other fields?
 
+	if(partitionHeaderSI.fstHashType == 0 && partitionHeaderSI.h3HashNum != 0)
+		cemuLog_log(LogType::Force, "FST: Partition uses unhashed blocks but stores a non-zero amount of H3 hashes");
+
 	// GM partition
 	DiscPartitionHeader partitionHeaderGM{};
 	if (!readPartitionHeader(partitionHeaderGM, gmPartitionIndex))
@@ -349,9 +353,10 @@ FSTVolume* FSTVolume::OpenFromDiscImage(const fs::path& path, NCrypto::AesKey& d
 	// if decryption is necessary
 	// load SI FST
 	dataSource->SetBaseOffset((uint64)partitionArray[siPartitionIndex].partitionAddress * DISC_SECTOR_SIZE);
-	auto siFST = OpenFST(dataSource.get(), (uint64)partitionHeaderSI.fstSector * DISC_SECTOR_SIZE, partitionHeaderSI.fstSize, &discTitleKey, static_cast<FSTVolume::ClusterHashMode>(partitionHeaderSI.fstHashType));
+	auto siFST = OpenFST(dataSource.get(), (uint64)partitionHeaderSI.fstSector * DISC_SECTOR_SIZE, partitionHeaderSI.fstSize, &discTitleKey, static_cast<FSTVolume::ClusterHashMode>(partitionHeaderSI.fstHashType), nullptr);
 	if (!siFST)
 		return nullptr;
+	cemu_assert_debug(!(siFST->HashIsDisabled() && partitionHeaderSI.h3HashNum != 0)); // if hash is disabled, no H3 data may be present
 	// load ticket file for partition that we want to decrypt
 	NCrypto::ETicketParser ticketParser;
 	std::vector<uint8> ticketData = siFST->ExtractFile(fmt::format("{:02x}/title.tik", gmPartitionIndex));
@@ -360,16 +365,32 @@ FSTVolume* FSTVolume::OpenFromDiscImage(const fs::path& path, NCrypto::AesKey& d
 		cemuLog_log(LogType::Force, "Disc image ticket file is invalid");
 		return nullptr;
 	}
+#if 0
+	// each SI partition seems to contain a title.tmd that we could parse and which should have information about the associated GM partition
+	// but the console seems to ignore this file for disc images, at least when mounting, so we shouldn't rely on it either
+	std::vector<uint8> tmdData = siFST->ExtractFile(fmt::format("{:02x}/title.tmd", gmPartitionIndex));
+	if (tmdData.empty())
+	{
+		cemuLog_log(LogType::Force, "Disc image TMD file is missing");
+		return nullptr;
+	}
+	// parse TMD
+	NCrypto::TMDParser tmdParser;
+	if (!tmdParser.parse(tmdData.data(), tmdData.size()))
+	{
+		cemuLog_log(LogType::Force, "Disc image TMD file is invalid");
+		return nullptr;
+	}
+#endif
 	delete siFST;
-
 	NCrypto::AesKey gmTitleKey;
 	ticketParser.GetTitleKey(gmTitleKey);
-
 	// load GM partition
 	dataSource->SetBaseOffset((uint64)partitionArray[gmPartitionIndex].partitionAddress * DISC_SECTOR_SIZE);
-	FSTVolume* r = OpenFST(std::move(dataSource), (uint64)partitionHeaderGM.fstSector * DISC_SECTOR_SIZE, partitionHeaderGM.fstSize, &gmTitleKey, static_cast<FSTVolume::ClusterHashMode>(partitionHeaderGM.fstHashType));
+	FSTVolume* r = OpenFST(std::move(dataSource), (uint64)partitionHeaderGM.fstSector * DISC_SECTOR_SIZE, partitionHeaderGM.fstSize, &gmTitleKey, static_cast<FSTVolume::ClusterHashMode>(partitionHeaderGM.fstHashType), nullptr);
 	if (r)
 		SET_FST_ERROR(OK);
+	cemu_assert_debug(!(r->HashIsDisabled() && partitionHeaderGM.h3HashNum != 0)); // if hash is disabled, no H3 data may be present
 	return r;
 }
 
@@ -426,15 +447,15 @@ FSTVolume* FSTVolume::OpenFromContentFolder(fs::path folderPath, ErrorCode* erro
 	}
 	// load FST
 	// fstSize = size of first cluster?
-	FSTVolume* fstVolume = FSTVolume::OpenFST(std::move(dataSource), 0, fstSize, &titleKey, fstHashMode);
+	FSTVolume* fstVolume = FSTVolume::OpenFST(std::move(dataSource), 0, fstSize, &titleKey, fstHashMode, &tmdParser);
 	if (fstVolume)
 		SET_FST_ERROR(OK);
 	return fstVolume;
 }
 
-FSTVolume* FSTVolume::OpenFST(FSTDataSource* dataSource, uint64 fstOffset, uint32 fstSize, NCrypto::AesKey* partitionTitleKey, ClusterHashMode fstHashMode)
+FSTVolume* FSTVolume::OpenFST(FSTDataSource* dataSource, uint64 fstOffset, uint32 fstSize, NCrypto::AesKey* partitionTitleKey, ClusterHashMode fstHashMode, NCrypto::TMDParser* optionalTMD)
 {
-	cemu_assert_debug(fstHashMode != ClusterHashMode::RAW || fstHashMode != ClusterHashMode::RAW2);
+	cemu_assert_debug(fstHashMode != ClusterHashMode::RAW || fstHashMode != ClusterHashMode::RAW_STREAM);
 	if (fstSize < sizeof(FSTHeader))
 		return nullptr;
 	constexpr uint64 FST_CLUSTER_OFFSET = 0;
@@ -465,6 +486,34 @@ FSTVolume* FSTVolume::OpenFST(FSTDataSource* dataSource, uint64 fstOffset, uint3
 		clusterTable[i].offset = clusterDataTable[i].offset;
 		clusterTable[i].size = clusterDataTable[i].size;
 		clusterTable[i].hashMode = static_cast<FSTVolume::ClusterHashMode>((uint8)clusterDataTable[i].hashMode);
+		clusterTable[i].hasContentHash = false; // from the TMD file (H4?)
+	}
+	// if the TMD is available (when opening .app files) we can use the extra info from it to validate unhashed clusters
+	// each content entry in the TMD corresponds to one cluster used by the FST
+	if(optionalTMD)
+	{
+		if(numCluster != optionalTMD->GetContentList().size())
+		{
+			cemuLog_log(LogType::Force, "FST: Number of clusters does not match TMD content list");
+			return nullptr;
+		}
+		auto& contentList = optionalTMD->GetContentList();
+		for(size_t i=0; i<contentList.size(); i++)
+		{
+			auto& cluster = clusterTable[i];
+			auto& content = contentList[i];
+			cluster.hasContentHash = true;
+			cluster.contentHashIsSHA1 = HAS_FLAG(contentList[i].contentFlags, NCrypto::TMDParser::TMDContentFlags::FLAG_SHA1);
+			cluster.contentSize = content.size;
+			static_assert(sizeof(content.hash32) == sizeof(cluster.contentHash32));
+			memcpy(cluster.contentHash32, content.hash32, sizeof(cluster.contentHash32));
+			// if unhashed mode, then initialize the hash context
+			if(cluster.hashMode == ClusterHashMode::RAW || cluster.hashMode == ClusterHashMode::RAW_STREAM)
+			{
+				cluster.singleHashCtx.reset(EVP_MD_CTX_new());
+				EVP_DigestInit_ex(cluster.singleHashCtx.get(), cluster.contentHashIsSHA1 ? EVP_sha1() : EVP_sha256(), nullptr);
+			}
+		}
 	}
 	// preprocess FST table
 	FSTHeader_FileEntry* fileTable = (FSTHeader_FileEntry*)(clusterDataTable + numCluster);
@@ -491,16 +540,17 @@ FSTVolume* FSTVolume::OpenFST(FSTDataSource* dataSource, uint64 fstOffset, uint3
 	fstVolume->m_offsetFactor = fstHeader->offsetFactor;
 	fstVolume->m_sectorSize = DISC_SECTOR_SIZE;
 	fstVolume->m_partitionTitlekey = *partitionTitleKey;
-	std::swap(fstVolume->m_cluster, clusterTable);
-	std::swap(fstVolume->m_entries, fstEntries);
-	std::swap(fstVolume->m_nameStringTable, nameStringTable);
+	fstVolume->m_hashIsDisabled = fstHeader->hashIsDisabled != 0;
+	fstVolume->m_cluster = std::move(clusterTable);
+	fstVolume->m_entries = std::move(fstEntries);
+	fstVolume->m_nameStringTable = std::move(nameStringTable);
 	return fstVolume;
 }
 
-FSTVolume* FSTVolume::OpenFST(std::unique_ptr<FSTDataSource> dataSource, uint64 fstOffset, uint32 fstSize, NCrypto::AesKey* partitionTitleKey, ClusterHashMode fstHashMode)
+FSTVolume* FSTVolume::OpenFST(std::unique_ptr<FSTDataSource> dataSource, uint64 fstOffset, uint32 fstSize, NCrypto::AesKey* partitionTitleKey, ClusterHashMode fstHashMode, NCrypto::TMDParser* optionalTMD)
 {
 	FSTDataSource* ds = dataSource.release();
-	FSTVolume* fstVolume = OpenFST(ds, fstOffset, fstSize, partitionTitleKey, fstHashMode);
+	FSTVolume* fstVolume = OpenFST(ds, fstOffset, fstSize, partitionTitleKey, fstHashMode, optionalTMD);
 	if (!fstVolume)
 	{
 		delete ds;
@@ -757,7 +807,7 @@ uint32 FSTVolume::ReadFile(FSTFileHandle& fileHandle, uint32 offset, uint32 size
 		return 0;
 	cemu_assert_debug(!HAS_FLAG(entry.GetFlags(), FSTEntry::FLAGS::FLAG_LINK));
 	FSTCluster& cluster = m_cluster[entry.fileInfo.clusterIndex];
-	if (cluster.hashMode == ClusterHashMode::RAW || cluster.hashMode == ClusterHashMode::RAW2)
+	if (cluster.hashMode == ClusterHashMode::RAW || cluster.hashMode == ClusterHashMode::RAW_STREAM)
 		return ReadFile_HashModeRaw(entry.fileInfo.clusterIndex, entry, offset, size, dataOut);
 	else if (cluster.hashMode == ClusterHashMode::HASH_INTERLEAVED)
 		return ReadFile_HashModeHashed(entry.fileInfo.clusterIndex, entry, offset, size, dataOut);
@@ -765,87 +815,15 @@ uint32 FSTVolume::ReadFile(FSTFileHandle& fileHandle, uint32 offset, uint32 size
 	return 0;
 }
 
-uint32 FSTVolume::ReadFile_HashModeRaw(uint32 clusterIndex, FSTEntry& entry, uint32 readOffset, uint32 readSize, void* dataOut)
-{
-	const uint32 readSizeInput = readSize;
-	uint8* dataOutU8 = (uint8*)dataOut;
-	if (readOffset >= entry.fileInfo.fileSize)
-		return 0;
-	else if ((readOffset + readSize) >= entry.fileInfo.fileSize)
-		readSize = (entry.fileInfo.fileSize - readOffset);
-
-	const FSTCluster& cluster = m_cluster[clusterIndex];
-	uint64 clusterOffset = (uint64)cluster.offset * m_sectorSize;
-	uint64 absFileOffset = entry.fileInfo.fileOffset * m_offsetFactor + readOffset;
-
-	// make sure the raw range we read is aligned to AES block size (16)
-	uint64 readAddrStart = absFileOffset & ~0xF;
-	uint64 readAddrEnd = (absFileOffset + readSize + 0xF) & ~0xF;
-
-	bool usesInitialIV = readOffset < 16;
-	if (!usesInitialIV)
-		readAddrStart -= 16; // read previous AES block since we require it for the IV
-	uint32 prePadding = (uint32)(absFileOffset - readAddrStart); // number of extra bytes we read before readOffset (for AES alignment and IV calculation)
-	uint32 postPadding = (uint32)(readAddrEnd - (absFileOffset + readSize));
-
-	uint8 readBuffer[64 * 1024];
-	// read first chunk
-	// if file read offset (readOffset) is within the first AES-block then use initial IV calculated from cluster index
-	// otherwise read previous AES-block is the IV (AES-CBC)
-	uint64 readAddrCurrent = readAddrStart;
-	uint32 rawBytesToRead = (uint32)std::min((readAddrEnd - readAddrStart), (uint64)sizeof(readBuffer));
-	if (m_dataSource->readData(clusterIndex, clusterOffset, readAddrCurrent, readBuffer, rawBytesToRead) != rawBytesToRead)
-	{
-		cemuLog_log(LogType::Force, "FST read error in raw content");
-		return 0;
-	}
-	readAddrCurrent += rawBytesToRead;
-
-	uint8 iv[16]{};
-	if (usesInitialIV)
-	{
-		// for the first AES block, the IV is initialized from cluster index
-		iv[0] = (uint8)(clusterIndex >> 8);
-		iv[1] = (uint8)(clusterIndex >> 0);
-		AES128_CBC_decrypt_updateIV(readBuffer, readBuffer, rawBytesToRead, m_partitionTitlekey.b, iv);
-		std::memcpy(dataOutU8, readBuffer + prePadding, rawBytesToRead - prePadding - postPadding);
-		dataOutU8 += (rawBytesToRead - prePadding - postPadding);
-		readSize -= (rawBytesToRead - prePadding - postPadding);
-	}
-	else
-	{
-		// IV is initialized from previous AES block (AES-CBC)
-		std::memcpy(iv, readBuffer, 16);
-		AES128_CBC_decrypt_updateIV(readBuffer + 16, readBuffer + 16, rawBytesToRead - 16, m_partitionTitlekey.b, iv);
-		std::memcpy(dataOutU8, readBuffer + prePadding, rawBytesToRead - prePadding - postPadding);
-		dataOutU8 += (rawBytesToRead - prePadding - postPadding);
-		readSize -= (rawBytesToRead - prePadding - postPadding);
-	}
-
-	// read remaining chunks
-	while (readSize > 0)
-	{
-		uint32 bytesToRead = (uint32)std::min((uint32)sizeof(readBuffer), readSize);
-		uint32 alignedBytesToRead = (bytesToRead + 15) & ~0xF;
-		if (m_dataSource->readData(clusterIndex, clusterOffset, readAddrCurrent, readBuffer, alignedBytesToRead) != alignedBytesToRead)
-		{
-			cemuLog_log(LogType::Force, "FST read error in raw content");
-			return 0;
-		}
-		AES128_CBC_decrypt_updateIV(readBuffer, readBuffer, alignedBytesToRead, m_partitionTitlekey.b, iv);
-		std::memcpy(dataOutU8, readBuffer, bytesToRead);
-		dataOutU8 += bytesToRead;
-		readSize -= bytesToRead;
-		readAddrCurrent += alignedBytesToRead;
-	}
-
-	return readSizeInput - readSize;
-}
-
 constexpr size_t BLOCK_SIZE = 0x10000;
 constexpr size_t BLOCK_HASH_SIZE = 0x0400;
 constexpr size_t BLOCK_FILE_SIZE = 0xFC00;
 
+struct FSTRawBlock
+{
+	std::vector<uint8> rawData; // unhashed block size depends on sector size field in partition header
+};
+
 struct FSTHashedBlock
 {
 	uint8 rawData[BLOCK_SIZE];
@@ -887,12 +865,160 @@ struct FSTHashedBlock
 
 static_assert(sizeof(FSTHashedBlock) == BLOCK_SIZE);
 
+struct FSTCachedRawBlock
+{
+	FSTRawBlock blockData;
+	uint8 ivForNextBlock[16];
+	uint64 lastAccess;
+};
+
 struct FSTCachedHashedBlock 
 {
 	FSTHashedBlock blockData;
 	uint64 lastAccess;
 };
 
+// Checks cache fill state and if necessary drops least recently accessed block from the cache. Optionally allows to recycle the released cache entry to cut down cost of memory allocation and clearing
+void FSTVolume::TrimCacheIfRequired(FSTCachedRawBlock** droppedRawBlock, FSTCachedHashedBlock** droppedHashedBlock)
+{
+	// calculate size used by cache
+	size_t cacheSize = 0;
+	for (auto& itr : m_cacheDecryptedRawBlocks)
+		cacheSize += itr.second->blockData.rawData.size();
+	for (auto& itr : m_cacheDecryptedHashedBlocks)
+		cacheSize += sizeof(FSTCachedHashedBlock) + sizeof(FSTHashedBlock);
+	// only trim if cache is full (larger than 2MB)
+	if (cacheSize < 2*1024*1024) // 2MB
+		return;
+	// scan both cache lists to find least recently accessed block to drop
+	auto dropRawItr = std::min_element(m_cacheDecryptedRawBlocks.begin(), m_cacheDecryptedRawBlocks.end(), [](const auto& a, const auto& b) -> bool
+										{ return a.second->lastAccess < b.second->lastAccess; });
+	auto dropHashedItr = std::min_element(m_cacheDecryptedHashedBlocks.begin(), m_cacheDecryptedHashedBlocks.end(), [](const auto& a, const auto& b) -> bool
+										{ return a.second->lastAccess < b.second->lastAccess; });
+	uint64 lastAccess = std::numeric_limits<uint64>::max();
+	if(dropRawItr != m_cacheDecryptedRawBlocks.end())
+		lastAccess = dropRawItr->second->lastAccess;
+	if(dropHashedItr != m_cacheDecryptedHashedBlocks.end())
+		lastAccess = std::min<uint64>(lastAccess, dropHashedItr->second->lastAccess);
+	if(dropRawItr != m_cacheDecryptedRawBlocks.end() && dropRawItr->second->lastAccess == lastAccess)
+	{
+		if (droppedRawBlock)
+			*droppedRawBlock = dropRawItr->second;
+		else
+			delete dropRawItr->second;
+		m_cacheDecryptedRawBlocks.erase(dropRawItr);
+		return;
+	}
+	else if(dropHashedItr != m_cacheDecryptedHashedBlocks.end() && dropHashedItr->second->lastAccess == lastAccess)
+	{
+		if (droppedHashedBlock)
+			*droppedHashedBlock = dropHashedItr->second;
+		else
+			delete dropHashedItr->second;
+		m_cacheDecryptedHashedBlocks.erase(dropHashedItr);
+	}
+}
+
+void FSTVolume::DetermineUnhashedBlockIV(uint32 clusterIndex, uint32 blockIndex, uint8 ivOut[16])
+{
+	memset(ivOut, 0, sizeof(ivOut));
+	if(blockIndex == 0)
+	{
+		ivOut[0] = (uint8)(clusterIndex >> 8);
+		ivOut[1] = (uint8)(clusterIndex >> 0);
+	}
+	else
+	{
+		// the last 16 encrypted bytes of the previous block are the IV (AES CBC)
+		// if the previous block is cached we can grab the IV from there. Otherwise we have to read the 16 bytes from the data source
+		uint32 prevBlockIndex = blockIndex - 1;
+		uint64 cacheBlockId = ((uint64)clusterIndex << (64 - 16)) | (uint64)prevBlockIndex;
+		auto itr = m_cacheDecryptedRawBlocks.find(cacheBlockId);
+		if (itr != m_cacheDecryptedRawBlocks.end())
+		{
+			memcpy(ivOut, itr->second->ivForNextBlock, 16);
+		}
+		else
+		{
+			cemu_assert(m_sectorSize >= 16);
+			uint64 clusterOffset = (uint64)m_cluster[clusterIndex].offset * m_sectorSize;
+			uint8 prevIV[16];
+			if (m_dataSource->readData(clusterIndex, clusterOffset, blockIndex * m_sectorSize - 16, prevIV, 16) != 16)
+			{
+				cemuLog_log(LogType::Force, "Failed to read IV for raw FST block");
+				m_detectedCorruption = true;
+				return;
+			}
+			memcpy(ivOut, prevIV, 16);
+		}
+	}
+}
+
+FSTCachedRawBlock* FSTVolume::GetDecryptedRawBlock(uint32 clusterIndex, uint32 blockIndex)
+{
+	FSTCluster& cluster = m_cluster[clusterIndex];
+	uint64 clusterOffset = (uint64)cluster.offset * m_sectorSize;
+	// generate id for cache
+	uint64 cacheBlockId = ((uint64)clusterIndex << (64 - 16)) | (uint64)blockIndex;
+	// lookup block in cache
+	FSTCachedRawBlock* block = nullptr;
+	auto itr = m_cacheDecryptedRawBlocks.find(cacheBlockId);
+	if (itr != m_cacheDecryptedRawBlocks.end())
+	{
+		block = itr->second;
+		block->lastAccess = ++m_cacheAccessCounter;
+		return block;
+	}
+	// if cache already full, drop least recently accessed block and recycle FSTCachedRawBlock object if possible
+	TrimCacheIfRequired(&block, nullptr);
+	if (!block)
+		block = new FSTCachedRawBlock();
+	block->blockData.rawData.resize(m_sectorSize);
+	// block not cached, read new
+	block->lastAccess = ++m_cacheAccessCounter;
+	if (m_dataSource->readData(clusterIndex, clusterOffset, blockIndex * m_sectorSize, block->blockData.rawData.data(), m_sectorSize) != m_sectorSize)
+	{
+		cemuLog_log(LogType::Force, "Failed to read raw FST block");
+		delete block;
+		m_detectedCorruption = true;
+		return nullptr;
+	}
+	// decrypt hash data
+	uint8 iv[16]{};
+	DetermineUnhashedBlockIV(clusterIndex, blockIndex, iv);
+	memcpy(block->ivForNextBlock, block->blockData.rawData.data() + m_sectorSize - 16, 16);
+	AES128_CBC_decrypt(block->blockData.rawData.data(), block->blockData.rawData.data(), m_sectorSize, m_partitionTitlekey.b, iv);
+	// if this is the next block, then hash it
+	if(cluster.hasContentHash)
+	{
+		if(cluster.singleHashNumBlocksHashed == blockIndex)
+		{
+			cemu_assert_debug(!(cluster.contentSize % m_sectorSize)); // size should be multiple of sector size? Regardless, the hashing code below can handle non-aligned sizes
+			bool isLastBlock = blockIndex == (std::max<uint32>(cluster.contentSize / m_sectorSize, 1) - 1);
+			uint32 hashSize = m_sectorSize;
+			if(isLastBlock)
+				hashSize = cluster.contentSize - (uint64)blockIndex*m_sectorSize;
+			EVP_DigestUpdate(cluster.singleHashCtx.get(), block->blockData.rawData.data(), hashSize);
+			cluster.singleHashNumBlocksHashed++;
+			if(isLastBlock)
+			{
+				uint8 hash[32];
+				EVP_DigestFinal_ex(cluster.singleHashCtx.get(), hash, nullptr);
+				if(memcmp(hash, cluster.contentHash32, cluster.contentHashIsSHA1 ? 20 : 32) != 0)
+				{
+					cemuLog_log(LogType::Force, "FST: Raw section hash mismatch");
+					delete block;
+					m_detectedCorruption = true;
+					return nullptr;
+				}
+			}
+		}
+	}
+	// register in cache
+	m_cacheDecryptedRawBlocks.emplace(cacheBlockId, block);
+	return block;
+}
+
 FSTCachedHashedBlock* FSTVolume::GetDecryptedHashedBlock(uint32 clusterIndex, uint32 blockIndex)
 {
 	const FSTCluster& cluster = m_cluster[clusterIndex];
@@ -908,22 +1034,17 @@ FSTCachedHashedBlock* FSTVolume::GetDecryptedHashedBlock(uint32 clusterIndex, ui
 		block->lastAccess = ++m_cacheAccessCounter;
 		return block;
 	}
-	// if cache already full, drop least recently accessed block (but recycle the FSTHashedBlock* object)
-	if (m_cacheDecryptedHashedBlocks.size() >= 16)
-	{
-		auto dropItr = std::min_element(m_cacheDecryptedHashedBlocks.begin(), m_cacheDecryptedHashedBlocks.end(), [](const auto& a, const auto& b) -> bool
-		{ return a.second->lastAccess < b.second->lastAccess; });
-		block = dropItr->second;
-		m_cacheDecryptedHashedBlocks.erase(dropItr);
-	}
-	else
+	// if cache already full, drop least recently accessed block and recycle FSTCachedHashedBlock object if possible
+	TrimCacheIfRequired(nullptr, &block);
+	if (!block)
 		block = new FSTCachedHashedBlock();
 	// block not cached, read new
 	block->lastAccess = ++m_cacheAccessCounter;
 	if (m_dataSource->readData(clusterIndex, clusterOffset, blockIndex * BLOCK_SIZE, block->blockData.rawData, BLOCK_SIZE) != BLOCK_SIZE)
 	{
-		cemuLog_log(LogType::Force, "Failed to read FST block");
+		cemuLog_log(LogType::Force, "Failed to read hashed FST block");
 		delete block;
+		m_detectedCorruption = true;
 		return nullptr;
 	}
 	// decrypt hash data
@@ -931,11 +1052,46 @@ FSTCachedHashedBlock* FSTVolume::GetDecryptedHashedBlock(uint32 clusterIndex, ui
 	AES128_CBC_decrypt(block->blockData.getHashData(), block->blockData.getHashData(), BLOCK_HASH_SIZE, m_partitionTitlekey.b, iv);
 	// decrypt file data
 	AES128_CBC_decrypt(block->blockData.getFileData(), block->blockData.getFileData(), BLOCK_FILE_SIZE, m_partitionTitlekey.b, block->blockData.getH0Hash(blockIndex%16));
+	// compare with H0 to verify data integrity
+	NCrypto::CHash160 h0;
+	SHA1(block->blockData.getFileData(), BLOCK_FILE_SIZE, h0.b);
+	uint32 h0Index = (blockIndex % 4096);
+	if (memcmp(h0.b, block->blockData.getH0Hash(h0Index & 0xF), sizeof(h0.b)) != 0)
+	{
+		cemuLog_log(LogType::Force, "FST: Hash H0 mismatch in hashed block (section {} index {})", clusterIndex, blockIndex);
+		delete block;
+		m_detectedCorruption = true;
+		return nullptr;
+	}
 	// register in cache
 	m_cacheDecryptedHashedBlocks.emplace(cacheBlockId, block);
 	return block;
 }
 
+uint32 FSTVolume::ReadFile_HashModeRaw(uint32 clusterIndex, FSTEntry& entry, uint32 readOffset, uint32 readSize, void* dataOut)
+{
+	uint8* dataOutU8 = (uint8*)dataOut;
+	if (readOffset >= entry.fileInfo.fileSize)
+		return 0;
+	else if ((readOffset + readSize) >= entry.fileInfo.fileSize)
+		readSize = (entry.fileInfo.fileSize - readOffset);
+	uint64 absFileOffset = entry.fileInfo.fileOffset * m_offsetFactor + readOffset;
+	uint32 remainingReadSize = readSize;
+	while (remainingReadSize > 0)
+	{
+		const FSTCachedRawBlock* rawBlock = this->GetDecryptedRawBlock(clusterIndex, absFileOffset/m_sectorSize);
+		if (!rawBlock)
+			break;
+		uint32 blockOffset = (uint32)(absFileOffset % m_sectorSize);
+		uint32 bytesToRead = std::min<uint32>(remainingReadSize, m_sectorSize - blockOffset);
+		std::memcpy(dataOutU8, rawBlock->blockData.rawData.data() + blockOffset, bytesToRead);
+		dataOutU8 += bytesToRead;
+		remainingReadSize -= bytesToRead;
+		absFileOffset += bytesToRead;
+	}
+	return readSize - remainingReadSize;
+}
+
 uint32 FSTVolume::ReadFile_HashModeHashed(uint32 clusterIndex, FSTEntry& entry, uint32 readOffset, uint32 readSize, void* dataOut)
 {
 	/*
@@ -966,7 +1122,6 @@ uint32 FSTVolume::ReadFile_HashModeHashed(uint32 clusterIndex, FSTEntry& entry,
 	*/
 
 	const FSTCluster& cluster = m_cluster[clusterIndex];
-	uint64 clusterBaseOffset = (uint64)cluster.offset * m_sectorSize;
 	uint64 fileReadOffset = entry.fileInfo.fileOffset * m_offsetFactor + readOffset;
 	uint32 blockIndex = (uint32)(fileReadOffset / BLOCK_FILE_SIZE);
 	uint32 bytesRemaining = readSize;
@@ -1019,6 +1174,8 @@ bool FSTVolume::Next(FSTDirectoryIterator& directoryIterator, FSTFileHandle& fil
 
 FSTVolume::~FSTVolume()
 {
+	for (auto& itr : m_cacheDecryptedRawBlocks)
+		delete itr.second;
 	for (auto& itr : m_cacheDecryptedHashedBlocks)
 		delete itr.second;
 	if (m_sourceIsOwned)
@@ -1115,4 +1272,4 @@ bool FSTVerifier::VerifyHashedContentFile(FileStream* fileContent, const NCrypto
 void FSTVolumeTest()
 {
 	FSTPathUnitTest();
-}
+}

src/Cafe/Filesystem/FST/FST.h

@@ -1,5 +1,6 @@
 #pragma once
 #include "Cemu/ncrypto/ncrypto.h"
+#include "openssl/evp.h"
 
 struct FSTFileHandle
 {
@@ -45,6 +46,7 @@ public:
 	~FSTVolume();
 
 	uint32 GetFileCount() const;
+	bool HasCorruption() const { return m_detectedCorruption; }
 
 	bool OpenFile(std::string_view path, FSTFileHandle& fileHandleOut, bool openOnlyFiles = false);
 
@@ -86,15 +88,25 @@ private:
 	enum class ClusterHashMode : uint8
 	{
 		RAW = 0, // raw data + encryption, no hashing?
-		RAW2 = 1, // raw data + encryption, with hash stored in tmd?
+	  	RAW_STREAM = 1, // raw data + encryption, with hash stored in tmd?
 		HASH_INTERLEAVED = 2, // hashes + raw interleaved in 0x10000 blocks (0x400 bytes of hashes at the beginning, followed by 0xFC00 bytes of data)
 	};
 
 	struct FSTCluster
 	{
+		FSTCluster() : singleHashCtx(nullptr, &EVP_MD_CTX_free) {}
+
 		uint32 offset;
 		uint32 size;
 		ClusterHashMode hashMode;
+		// extra data if TMD is available
+		bool hasContentHash;
+		uint8 contentHash32[32];
+		bool contentHashIsSHA1; // if true then it's SHA1 (with extra bytes zeroed out), otherwise it's SHA256
+		uint64 contentSize; // size of the content (in blocks)
+		// hash context for single hash mode (content hash must be available)
+		std::unique_ptr<EVP_MD_CTX, decltype(&EVP_MD_CTX_free)> singleHashCtx; // unique_ptr to make this move-only
+		uint32 singleHashNumBlocksHashed{0};
 	};
 
 	struct FSTEntry
@@ -164,17 +176,30 @@ private:
 	bool m_sourceIsOwned{};
 	uint32 m_sectorSize{}; // for cluster offsets
 	uint32 m_offsetFactor{}; // for file offsets
+	bool m_hashIsDisabled{}; // disables hash verification (for all clusters of this volume?)
 	std::vector<FSTCluster> m_cluster;
 	std::vector<FSTEntry> m_entries;
 	std::vector<char> m_nameStringTable;
 	NCrypto::AesKey m_partitionTitlekey;
+	bool m_detectedCorruption{false};
 
-	/* Cache for decrypted hashed blocks */
+	bool HashIsDisabled() const
+	{
+		return m_hashIsDisabled;
+	}
+
+	/* Cache for decrypted raw and hashed blocks */
+	std::unordered_map<uint64, struct FSTCachedRawBlock*> m_cacheDecryptedRawBlocks;
 	std::unordered_map<uint64, struct FSTCachedHashedBlock*> m_cacheDecryptedHashedBlocks;
 	uint64 m_cacheAccessCounter{};
 
+	void DetermineUnhashedBlockIV(uint32 clusterIndex, uint32 blockIndex, uint8 ivOut[16]);
+
+	struct FSTCachedRawBlock* GetDecryptedRawBlock(uint32 clusterIndex, uint32 blockIndex);
 	struct FSTCachedHashedBlock* GetDecryptedHashedBlock(uint32 clusterIndex, uint32 blockIndex);
 
+	void TrimCacheIfRequired(struct FSTCachedRawBlock** droppedRawBlock, struct FSTCachedHashedBlock** droppedHashedBlock);
+
 	/* File reading */
 	uint32 ReadFile_HashModeRaw(uint32 clusterIndex, FSTEntry& entry, uint32 readOffset, uint32 readSize, void* dataOut);
 	uint32 ReadFile_HashModeHashed(uint32 clusterIndex, FSTEntry& entry, uint32 readOffset, uint32 readSize, void* dataOut);
@@ -185,7 +210,10 @@ private:
 		/* +0x00 */ uint32be magic;
 		/* +0x04 */ uint32be offsetFactor;
 		/* +0x08 */ uint32be numCluster;
-		/* +0x0C */ uint32be ukn0C;
+		/* +0x0C */ uint8be hashIsDisabled;
+		/* +0x0D */ uint8be ukn0D;
+		/* +0x0E */ uint8be ukn0E;
+		/* +0x0F */ uint8be ukn0F;
 		/* +0x10 */ uint32be ukn10;
 		/* +0x14 */ uint32be ukn14;
 		/* +0x18 */ uint32be ukn18;
@@ -262,8 +290,8 @@ private:
 
 	static_assert(sizeof(FSTHeader_FileEntry) == 0x10);
 
-	static FSTVolume* OpenFST(FSTDataSource* dataSource, uint64 fstOffset, uint32 fstSize, NCrypto::AesKey* partitionTitleKey, ClusterHashMode fstHashMode);
-	static FSTVolume* OpenFST(std::unique_ptr<FSTDataSource> dataSource, uint64 fstOffset, uint32 fstSize, NCrypto::AesKey* partitionTitleKey, ClusterHashMode fstHashMode);
+	static FSTVolume* OpenFST(FSTDataSource* dataSource, uint64 fstOffset, uint32 fstSize, NCrypto::AesKey* partitionTitleKey, ClusterHashMode fstHashMode, NCrypto::TMDParser* optionalTMD);
+	static FSTVolume* OpenFST(std::unique_ptr<FSTDataSource> dataSource, uint64 fstOffset, uint32 fstSize, NCrypto::AesKey* partitionTitleKey, ClusterHashMode fstHashMode, NCrypto::TMDParser* optionalTMD);
 	static bool ProcessFST(FSTHeader_FileEntry* fileTable, uint32 numFileEntries, uint32 numCluster, std::vector<char>& nameStringTable, std::vector<FSTEntry>& fstEntries);
 
 	bool MatchFSTEntryName(FSTEntry& entry, std::string_view comparedName)

src/Cafe/GameProfile/GameProfile.cpp

@@ -140,7 +140,7 @@ bool gameProfile_loadEnumOption(IniParser& iniParser, const char* optionName, T&
 	for(const T& v : T())
 	{
 		// test integer option
-		if (boost::iequals(fmt::format("{}", static_cast<typename std::underlying_type<T>::type>(v)), *option_value))
+		if (boost::iequals(fmt::format("{}", fmt::underlying(v)), *option_value))
 		{
 			option = v;
 			return true;

src/Cafe/GraphicPack/GraphicPack2.cpp

@@ -345,7 +345,7 @@ GraphicPack2::GraphicPack2(fs::path rulesPath, IniParser& rules)
 			const auto preset_name = rules.FindOption("name");
 			if (!preset_name)
 			{
-				cemuLog_log(LogType::Force, "Graphic pack \"{}\": Preset in line {} skipped because it has no name option defined", m_name, rules.GetCurrentSectionLineNumber());
+				cemuLog_log(LogType::Force, "Graphic pack \"{}\": Preset in line {} skipped because it has no name option defined", GetNormalizedPathString(), rules.GetCurrentSectionLineNumber());
 				continue;
 			}
 			
@@ -369,7 +369,7 @@ GraphicPack2::GraphicPack2(fs::path rulesPath, IniParser& rules)
 			}
 			catch (const std::exception & ex)
 			{
-				cemuLog_log(LogType::Force, "Graphic pack \"{}\": Can't parse preset \"{}\": {}", m_name, *preset_name, ex.what());
+				cemuLog_log(LogType::Force, "Graphic pack \"{}\": Can't parse preset \"{}\": {}", GetNormalizedPathString(), *preset_name, ex.what());
 			}
 		}
 		else if (boost::iequals(currentSectionName, "RAM"))
@@ -383,7 +383,7 @@ GraphicPack2::GraphicPack2(fs::path rulesPath, IniParser& rules)
 				{
 					if (m_version <= 5)
 					{
-						cemuLog_log(LogType::Force, "Graphic pack \"{}\": [RAM] options are only available for graphic pack version 6 or higher", m_name, optionNameBuf);
+						cemuLog_log(LogType::Force, "Graphic pack \"{}\": [RAM] options are only available for graphic pack version 6 or higher", GetNormalizedPathString(), optionNameBuf);
 						throw std::exception();
 					}
 
@@ -393,12 +393,12 @@ GraphicPack2::GraphicPack2(fs::path rulesPath, IniParser& rules)
 					{
 						if (addrEnd <= addrStart)
 						{
-							cemuLog_log(LogType::Force, "Graphic pack \"{}\": start address (0x{:08x}) must be greater than end address (0x{:08x}) for {}", m_name, addrStart, addrEnd, optionNameBuf);
+							cemuLog_log(LogType::Force, "Graphic pack \"{}\": start address (0x{:08x}) must be greater than end address (0x{:08x}) for {}", GetNormalizedPathString(), addrStart, addrEnd, optionNameBuf);
 							throw std::exception();
 						}
 						else if ((addrStart & 0xFFF) != 0 || (addrEnd & 0xFFF) != 0)
 						{
-							cemuLog_log(LogType::Force, "Graphic pack \"{}\": addresses for %s are not aligned to 0x1000", m_name, optionNameBuf);
+							cemuLog_log(LogType::Force, "Graphic pack \"{}\": addresses for %s are not aligned to 0x1000", GetNormalizedPathString(), optionNameBuf);
 							throw std::exception();
 						}
 						else
@@ -408,7 +408,7 @@ GraphicPack2::GraphicPack2(fs::path rulesPath, IniParser& rules)
 					}
 					else
 					{
-						cemuLog_log(LogType::Force, "Graphic pack \"{}\": has invalid syntax for option {}", m_name, optionNameBuf);
+						cemuLog_log(LogType::Force, "Graphic pack \"{}\": has invalid syntax for option {}", GetNormalizedPathString(), optionNameBuf);
 						throw std::exception();
 					}
 				}
@@ -422,24 +422,32 @@ GraphicPack2::GraphicPack2(fs::path rulesPath, IniParser& rules)
 		std::unordered_map<std::string, std::vector<PresetPtr>> tmp_map;
 		
 		// all vars must be defined in the default preset vars before
-		for (const auto& entry : m_presets)
+		std::vector<std::pair<std::string, std::string>> mismatchingPresetVars;
+		for (const auto& presetEntry : m_presets)
 		{
-			tmp_map[entry->category].emplace_back(entry);
+			tmp_map[presetEntry->category].emplace_back(presetEntry);
 			
-			for (auto& kv : entry->variables)
+			for (auto& presetVar : presetEntry->variables)
 			{
-				const auto it = m_preset_vars.find(kv.first);
+				const auto it = m_preset_vars.find(presetVar.first);
 				if (it == m_preset_vars.cend())
 				{
-					cemuLog_log(LogType::Force, "Graphic pack: \"{}\" contains preset variables which are not defined in the default section", m_name);
-					throw std::exception();
+					mismatchingPresetVars.emplace_back(presetEntry->name, presetVar.first);
+					continue;
 				}
-
 				// overwrite var type with default var type
-				kv.second.first = it->second.first;
+				presetVar.second.first = it->second.first;
 			}
 		}
 
+		if(!mismatchingPresetVars.empty())
+		{
+			cemuLog_log(LogType::Force, "Graphic pack \"{}\" contains preset variables which are not defined in the [Default] section:", GetNormalizedPathString());
+			for (const auto& [presetName, varName] : mismatchingPresetVars)
+				cemuLog_log(LogType::Force, "Preset: {} Variable: {}", presetName, varName);
+			throw std::exception();
+		}
+
 		// have first entry be default active for every category if no default= is set
 		for(auto entry : get_values(tmp_map))
 		{
@@ -469,7 +477,7 @@ GraphicPack2::GraphicPack2(fs::path rulesPath, IniParser& rules)
 				auto& p2 = kv.second[i + 1];
 				if (p1->variables.size() != p2->variables.size())
 				{
-					cemuLog_log(LogType::Force, "Graphic pack: \"{}\" contains inconsistent preset variables", m_name);
+					cemuLog_log(LogType::Force, "Graphic pack: \"{}\" contains inconsistent preset variables", GetNormalizedPathString());
 					throw std::exception();
 				}
 
@@ -477,14 +485,14 @@ GraphicPack2::GraphicPack2(fs::path rulesPath, IniParser& rules)
 				std::set<std::string> keys2(get_keys(p2->variables).begin(), get_keys(p2->variables).end());
 				if (keys1 != keys2)
 				{
-					cemuLog_log(LogType::Force, "Graphic pack: \"{}\" contains inconsistent preset variables", m_name);
+					cemuLog_log(LogType::Force, "Graphic pack: \"{}\" contains inconsistent preset variables", GetNormalizedPathString());
 					throw std::exception();
 				}
 
 				if(p1->is_default)
 				{
 					if(has_default)
-						cemuLog_log(LogType::Force, "Graphic pack: \"{}\" has more than one preset with the default key set for the same category \"{}\"", m_name, p1->name);
+						cemuLog_log(LogType::Force, "Graphic pack: \"{}\" has more than one preset with the default key set for the same category \"{}\"", GetNormalizedPathString(), p1->name);
 					p1->active = true;
 					has_default = true;
 				}
@@ -960,7 +968,7 @@ bool GraphicPack2::Activate()
 				auto option_upscale = rules.FindOption("upscaleMagFilter");
 				if(option_upscale && boost::iequals(*option_upscale, "NearestNeighbor"))
 					m_output_settings.upscale_filter = LatteTextureView::MagFilter::kNearestNeighbor;
-				auto option_downscale = rules.FindOption("NearestNeighbor");
+				auto option_downscale = rules.FindOption("downscaleMinFilter");
 				if (option_downscale && boost::iequals(*option_downscale, "NearestNeighbor"))
 					m_output_settings.downscale_filter = LatteTextureView::MagFilter::kNearestNeighbor;
 			}

src/Cafe/HW/Espresso/Debugger/Debugger.cpp

@@ -8,6 +8,7 @@
 #include "gui/debugger/DebuggerWindow2.h"
 
 #include "Cafe/OS/libs/coreinit/coreinit.h"
+#include "util/helpers/helpers.h"
 
 #if BOOST_OS_WINDOWS
 #include <Windows.h>
@@ -136,11 +137,6 @@ void debugger_createCodeBreakpoint(uint32 address, uint8 bpType)
 	debugger_updateExecutionBreakpoint(address);
 }
 
-void debugger_createExecuteBreakpoint(uint32 address)
-{
-	debugger_createCodeBreakpoint(address, DEBUGGER_BP_T_NORMAL);
-}
-
 namespace coreinit
 {
 	std::vector<std::thread::native_handle_type>& OSGetSchedulerThreads();
@@ -294,8 +290,23 @@ void debugger_toggleExecuteBreakpoint(uint32 address)
 	}
 	else
 	{
-		// create new breakpoint
-		debugger_createExecuteBreakpoint(address);
+		// create new execution breakpoint
+		debugger_createCodeBreakpoint(address, DEBUGGER_BP_T_NORMAL);
+	}
+}
+
+void debugger_toggleLoggingBreakpoint(uint32 address)
+{
+	auto existingBP = debugger_getFirstBP(address, DEBUGGER_BP_T_LOGGING);
+	if (existingBP)
+	{
+		// delete existing breakpoint
+		debugger_deleteBreakpoint(existingBP);
+	}
+	else
+	{
+		// create new logging breakpoint
+		debugger_createCodeBreakpoint(address, DEBUGGER_BP_T_LOGGING);
 	}
 }
 
@@ -447,6 +458,34 @@ bool debugger_hasPatch(uint32 address)
 	return false;
 }
 
+void debugger_removePatch(uint32 address)
+{
+	for (sint32 i = 0; i < debuggerState.patches.size(); i++)
+	{
+		auto& patch = debuggerState.patches[i];
+		if (address < patch->address || address >= (patch->address + patch->length))
+			continue;
+		MPTR startAddress = patch->address;
+		MPTR endAddress = patch->address + patch->length;
+		// remove any breakpoints overlapping with the patch
+		for (auto& bp : debuggerState.breakpoints)
+		{
+			if (bp->address + 4 > startAddress && bp->address < endAddress)
+			{
+				bp->enabled = false;
+				debugger_updateExecutionBreakpoint(bp->address);
+			}
+		}
+		// restore original data
+		memcpy(MEMPTR<void>(startAddress).GetPtr(), patch->origData.data(), patch->length);
+		PPCRecompiler_invalidateRange(startAddress, endAddress);
+		// remove patch
+		delete patch;
+		debuggerState.patches.erase(debuggerState.patches.begin() + i);
+		return;
+	}
+}
+
 void debugger_stepInto(PPCInterpreter_t* hCPU, bool updateDebuggerWindow = true)
 {
 	bool isRecEnabled = ppcRecompilerEnabled;
@@ -510,7 +549,48 @@ void debugger_enterTW(PPCInterpreter_t* hCPU)
 	{
 		if (bp->bpType == DEBUGGER_BP_T_LOGGING && bp->enabled)
 		{
-			std::string logName = !bp->comment.empty() ? "Breakpoint '"+boost::nowide::narrow(bp->comment)+"'" : fmt::format("Breakpoint at 0x{:08X} (no comment)", bp->address);
+			std::string comment = !bp->comment.empty() ? boost::nowide::narrow(bp->comment) : fmt::format("Breakpoint at 0x{:08X} (no comment)", bp->address);
+
+			auto replacePlaceholders = [&](const std::string& prefix, const auto& formatFunc)
+			{
+				size_t pos = 0;
+				while ((pos = comment.find(prefix, pos)) != std::string::npos)
+				{
+					size_t endPos = comment.find('}', pos);
+					if (endPos == std::string::npos)
+						break;
+
+					try
+					{
+						if (int regNum = ConvertString<int>(comment.substr(pos + prefix.length(), endPos - pos - prefix.length())); regNum >= 0 && regNum < 32)
+						{
+							std::string replacement = formatFunc(regNum);
+							comment.replace(pos, endPos - pos + 1, replacement);
+							pos += replacement.length();
+						}
+						else
+						{
+							pos = endPos + 1;
+						}
+					}
+					catch (...)
+					{
+						pos = endPos + 1;
+					}
+				}
+			};
+
+			// Replace integer register placeholders {rX}
+			replacePlaceholders("{r", [&](int regNum) {
+				return fmt::format("0x{:08X}", hCPU->gpr[regNum]);
+			});
+
+			// Replace floating point register placeholders {fX}
+			replacePlaceholders("{f", [&](int regNum) {
+				return fmt::format("{}", hCPU->fpr[regNum].fpr);
+			});
+
+			std::string logName = "Breakpoint '" + comment + "'";
 			std::string logContext = fmt::format("Thread: {:08x} LR: 0x{:08x}", MEMPTR<OSThread_t>(coreinit::OSGetCurrentThread()).GetMPTR(), hCPU->spr.LR, cemuLog_advancedPPCLoggingEnabled() ? " Stack Trace:" : "");
 			cemuLog_log(LogType::Force, "[Debugger] {} was executed! {}", logName, logContext);
 			if (cemuLog_advancedPPCLoggingEnabled())
@@ -547,7 +627,7 @@ void debugger_enterTW(PPCInterpreter_t* hCPU)
 	debuggerState.debugSession.stepInto = false;
 	debuggerState.debugSession.stepOver = false;
 	debuggerState.debugSession.run = false;
-	while (true)
+	while (debuggerState.debugSession.isTrapped)
 	{
 		std::this_thread::sleep_for(std::chrono::milliseconds(1));
 		// check for step commands

src/Cafe/HW/Espresso/Debugger/Debugger.h

@@ -100,8 +100,8 @@ extern debuggerState_t debuggerState;
 // new API
 DebuggerBreakpoint* debugger_getFirstBP(uint32 address);
 void debugger_createCodeBreakpoint(uint32 address, uint8 bpType);
-void debugger_createExecuteBreakpoint(uint32 address);
 void debugger_toggleExecuteBreakpoint(uint32 address); // create/remove execute breakpoint
+void debugger_toggleLoggingBreakpoint(uint32 address); // create/remove logging breakpoint
 void debugger_toggleBreakpoint(uint32 address, bool state, DebuggerBreakpoint* bp);
 
 void debugger_createMemoryBreakpoint(uint32 address, bool onRead, bool onWrite);
@@ -114,6 +114,7 @@ void debugger_updateExecutionBreakpoint(uint32 address, bool forceRestore = fals
 
 void debugger_createPatch(uint32 address, std::span<uint8> patchData);
 bool debugger_hasPatch(uint32 address);
+void debugger_removePatch(uint32 address);
 
 void debugger_forceBreak(); // force breakpoint at the next possible instruction
 bool debugger_isTrapped();

src/Cafe/HW/Espresso/EspressoISA.h

@@ -91,13 +91,15 @@ namespace Espresso
 		BCCTR = 528
 	};
 
-	enum class OPCODE_31
+	enum class Opcode31
 	{
-
+		TW = 4,
+		MFTB = 371,
 	};
 
 	inline PrimaryOpcode GetPrimaryOpcode(uint32 opcode) { return (PrimaryOpcode)(opcode >> 26); };
 	inline Opcode19 GetGroup19Opcode(uint32 opcode) { return (Opcode19)((opcode >> 1) & 0x3FF); };
+	inline Opcode31 GetGroup31Opcode(uint32 opcode) { return (Opcode31)((opcode >> 1) & 0x3FF); };
 
 	struct BOField 
 	{
@@ -132,6 +134,12 @@ namespace Espresso
 		uint8 bo;
 	};
 
+	// returns true if LK bit is set, only valid for branch instructions
+	inline bool DecodeLK(uint32 opcode)
+	{
+		return (opcode & 1) != 0;
+	}
+
 	inline void _decodeForm_I(uint32 opcode, uint32& LI, bool& AA, bool& LK)
 	{
 		LI = opcode & 0x3fffffc;
@@ -183,13 +191,7 @@ namespace Espresso
 		_decodeForm_D_branch(opcode, BD, BO, BI, AA, LK);
 	}
 
-	inline void decodeOp_BCLR(uint32 opcode, BOField& BO, uint32& BI, bool& LK)
-	{
-		// form XL (with BD field expected to be zero)
-		_decodeForm_XL(opcode, BO, BI, LK);
-	}
-
-	inline void decodeOp_BCCTR(uint32 opcode, BOField& BO, uint32& BI, bool& LK)
+	inline void decodeOp_BCSPR(uint32 opcode, BOField& BO, uint32& BI, bool& LK) // BCLR and BCSPR
 	{
 		// form XL (with BD field expected to be zero)
 		_decodeForm_XL(opcode, BO, BI, LK);

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterALU.hpp

@@ -3,12 +3,12 @@ static void PPCInterpreter_setXerOV(PPCInterpreter_t* hCPU, bool hasOverflow)
 {
 	if (hasOverflow)
 	{
-		hCPU->spr.XER |= XER_SO;
-		hCPU->spr.XER |= XER_OV;
+		hCPU->xer_so = 1;
+		hCPU->xer_ov = 1;
 	}
 	else
 	{
-		hCPU->spr.XER &= ~XER_OV;
+		hCPU->xer_ov = 0;
 	}
 }
 
@@ -246,7 +246,7 @@ static void PPCInterpreter_SUBFCO(PPCInterpreter_t* hCPU, uint32 opcode)
 	uint32 a = hCPU->gpr[rA];
 	uint32 b = hCPU->gpr[rB];
 	hCPU->gpr[rD] = ~a + b + 1;
-	// update xer
+	// update carry
 	if (ppc_carry_3(~a, b, 1))
 		hCPU->xer_ca = 1;
 	else
@@ -848,8 +848,7 @@ static void PPCInterpreter_CMP(PPCInterpreter_t* hCPU, uint32 opcode)
 		hCPU->cr[cr * 4 + CR_BIT_GT] = 1;
 	else 
 		hCPU->cr[cr * 4 + CR_BIT_EQ] = 1;
-	if ((hCPU->spr.XER & XER_SO) != 0)
-		hCPU->cr[cr * 4 + CR_BIT_SO] = 1;
+	hCPU->cr[cr * 4 + CR_BIT_SO] = hCPU->xer_so;
 	PPCInterpreter_nextInstruction(hCPU);
 }
 
@@ -871,8 +870,7 @@ static void PPCInterpreter_CMPL(PPCInterpreter_t* hCPU, uint32 opcode)
 		hCPU->cr[cr * 4 + CR_BIT_GT] = 1;
 	else
 		hCPU->cr[cr * 4 + CR_BIT_EQ] = 1;
-	if ((hCPU->spr.XER & XER_SO) != 0)
-		hCPU->cr[cr * 4 + CR_BIT_SO] = 1;
+	hCPU->cr[cr * 4 + CR_BIT_SO] = hCPU->xer_so;
 	PPCInterpreter_nextInstruction(hCPU);
 }
 
@@ -895,8 +893,7 @@ static void PPCInterpreter_CMPI(PPCInterpreter_t* hCPU, uint32 opcode)
 		hCPU->cr[cr * 4 + CR_BIT_GT] = 1;
 	else 
 		hCPU->cr[cr * 4 + CR_BIT_EQ] = 1;
-	if (hCPU->spr.XER & XER_SO)
-		hCPU->cr[cr * 4 + CR_BIT_SO] = 1;
+	hCPU->cr[cr * 4 + CR_BIT_SO] = hCPU->xer_so;
 	PPCInterpreter_nextInstruction(hCPU);
 }
 
@@ -919,8 +916,7 @@ static void PPCInterpreter_CMPLI(PPCInterpreter_t* hCPU, uint32 opcode)
 		hCPU->cr[cr * 4 + CR_BIT_GT] = 1;
 	else
 		hCPU->cr[cr * 4 + CR_BIT_EQ] = 1;
-	if (hCPU->spr.XER & XER_SO)
-		hCPU->cr[cr * 4 + CR_BIT_SO] = 1;
+	hCPU->cr[cr * 4 + CR_BIT_SO] = hCPU->xer_so;
 	PPCInterpreter_nextInstruction(hCPU);
 }
 

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterFPU.cpp

@@ -32,7 +32,7 @@ espresso_frsqrte_entry_t frsqrteLookupTable[32] =
 	{0x20c1000, 0x35e},{0x1f12000, 0x332},{0x1d79000, 0x30a},{0x1bf4000, 0x2e6},
 };
 
-double frsqrte_espresso(double input)
+ATTR_MS_ABI double frsqrte_espresso(double input)
 {
 	unsigned long long x = *(unsigned long long*)&input;
 
@@ -111,7 +111,7 @@ espresso_fres_entry_t fresLookupTable[32] =
 	{0x88400, 0x11a},	{0x65000, 0x11a},	{0x41c00, 0x108},	{0x20c00, 0x106}
 };
 
-double fres_espresso(double input)
+ATTR_MS_ABI double fres_espresso(double input)
 {
 	// based on testing we know that fres uses only the first 15 bits of the mantissa
 	// seee eeee eeee mmmm mmmm mmmm mmmx xxxx ....		(s = sign, e = exponent, m = mantissa, x = not used)

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterInternal.h

@@ -50,9 +50,9 @@
 #define CR_BIT_EQ	2
 #define CR_BIT_SO	3
 
-#define XER_SO			(1<<31)	// summary overflow bit
-#define XER_OV			(1<<30)	// overflow bit
 #define XER_BIT_CA		(29)	// carry bit index. To accelerate frequent access, this bit is stored as a separate uint8
+#define XER_BIT_SO		(31)	// summary overflow, counterpart to CR SO
+#define XER_BIT_OV		(30)
 
 // FPSCR
 #define FPSCR_VXSNAN	(1<<24)
@@ -118,7 +118,8 @@
 
 static inline void ppc_update_cr0(PPCInterpreter_t* hCPU, uint32 r)
 {
-	hCPU->cr[CR_BIT_SO] = (hCPU->spr.XER&XER_SO) ? 1 : 0;
+	cemu_assert_debug(hCPU->xer_so <= 1);
+	hCPU->cr[CR_BIT_SO] = hCPU->xer_so;
 	hCPU->cr[CR_BIT_LT] = ((r != 0) ? 1 : 0) & ((r & 0x80000000) ? 1 : 0);
 	hCPU->cr[CR_BIT_EQ] = (r == 0);
 	hCPU->cr[CR_BIT_GT] = hCPU->cr[CR_BIT_EQ] ^ hCPU->cr[CR_BIT_LT] ^ 1;  // this works because EQ and LT can never be set at the same time. So the only case where GT becomes 1 is when LT=0 and EQ=0
@@ -190,8 +191,8 @@ inline double roundTo25BitAccuracy(double d)
 	return *(double*)&v;
 }
 
-double fres_espresso(double input);
-double frsqrte_espresso(double input);
+ATTR_MS_ABI double fres_espresso(double input);
+ATTR_MS_ABI double frsqrte_espresso(double input);
 
 void fcmpu_espresso(PPCInterpreter_t* hCPU, int crfD, double a, double b);
 

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterLoadStore.hpp

@@ -85,7 +85,8 @@ static void PPCInterpreter_STWCX(PPCInterpreter_t* hCPU, uint32 Opcode)
 			ppc_setCRBit(hCPU, CR_BIT_GT, 0);
 			ppc_setCRBit(hCPU, CR_BIT_EQ, 1);
 		}
-		ppc_setCRBit(hCPU, CR_BIT_SO, (hCPU->spr.XER&XER_SO) != 0 ? 1 : 0);
+		cemu_assert_debug(hCPU->xer_so <= 1);
+		ppc_setCRBit(hCPU, CR_BIT_SO, hCPU->xer_so);
 		// remove reservation
 		hCPU->reservedMemAddr = 0;
 		hCPU->reservedMemValue = 0;

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterMain.cpp

@@ -63,16 +63,24 @@ void PPCInterpreter_setDEC(PPCInterpreter_t* hCPU, uint32 newValue)
 uint32 PPCInterpreter_getXER(PPCInterpreter_t* hCPU)
 {
 	uint32 xerValue = hCPU->spr.XER;
-	xerValue &= ~(1<<XER_BIT_CA);
-	if( hCPU->xer_ca )
-		xerValue |= (1<<XER_BIT_CA);
+	xerValue &= ~(1 << XER_BIT_CA);
+	xerValue &= ~(1 << XER_BIT_SO);
+	xerValue &= ~(1 << XER_BIT_OV);
+	if (hCPU->xer_ca)
+		xerValue |= (1 << XER_BIT_CA);
+	if (hCPU->xer_so)
+		xerValue |= (1 << XER_BIT_SO);
+	if (hCPU->xer_ov)
+		xerValue |= (1 << XER_BIT_OV);
 	return xerValue;
 }
 
 void PPCInterpreter_setXER(PPCInterpreter_t* hCPU, uint32 v)
 {
 	hCPU->spr.XER = v;
-	hCPU->xer_ca = (v>>XER_BIT_CA)&1;
+	hCPU->xer_ca = (v >> XER_BIT_CA) & 1;
+	hCPU->xer_so = (v >> XER_BIT_SO) & 1;
+	hCPU->xer_ov = (v >> XER_BIT_OV) & 1;
 }
 
 uint32 PPCInterpreter_getCoreIndex(PPCInterpreter_t* hCPU)

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterOPC.cpp

@@ -5,7 +5,6 @@
 #include "Cafe/OS/libs/coreinit/coreinit_CodeGen.h"
 
 #include "../Recompiler/PPCRecompiler.h"
-#include "../Recompiler/PPCRecompilerX64.h"
 
 #include <float.h>
 #include "Cafe/HW/Latte/Core/LatteBufferCache.h"

src/Cafe/HW/Espresso/PPCState.h

@@ -49,6 +49,8 @@ struct PPCInterpreter_t
 	uint32 fpscr;
 	uint8 cr[32]; // 0 -> bit not set, 1 -> bit set (upper 7 bits of each byte must always be zero) (cr0 starts at index 0, cr1 at index 4 ..)
 	uint8 xer_ca;  // carry from xer
+	uint8 xer_so;
+	uint8 xer_ov;
 	uint8 LSQE;
 	uint8 PSE;
 	// thread remaining cycles
@@ -67,7 +69,8 @@ struct PPCInterpreter_t
 	uint32 reservedMemValue;
 	// temporary storage for recompiler
 	FPR_t temporaryFPR[8];
-	uint32 temporaryGPR[4];
+	uint32 temporaryGPR[4]; // deprecated, refactor backend dependency on this away
+	uint32 temporaryGPR_reg[4];
 	// values below this are not used by Cafe OS usermode
 	struct  
 	{

src/Cafe/HW/Espresso/Recompiler/BackendX64/BackendX64.h

@@ -1,104 +1,56 @@
 
-typedef struct  
+#include "../PPCRecompiler.h" // todo - get rid of dependency
+
+#include "x86Emitter.h"
+
+struct x64RelocEntry_t
 {
+	x64RelocEntry_t(uint32 offset, void* extraInfo) : offset(offset), extraInfo(extraInfo) {};
+
 	uint32 offset;
-	uint8  type;
 	void*  extraInfo;
-}x64RelocEntry_t;
+};
 
-typedef struct  
+struct x64GenContext_t
 {
-	uint8* codeBuffer;
-	sint32 codeBufferIndex;
-	sint32 codeBufferSize;
-	// cr state
-	sint32 activeCRRegister; // current x86 condition flags reflect this cr* register
-	sint32 activeCRState; // describes the way in which x86 flags map to the cr register (signed / unsigned)
+	IMLSegment* currentSegment{};
+	x86Assembler64* emitter;
+	sint32 m_currentInstructionEmitIndex;
+
+	x64GenContext_t()
+	{
+		emitter = new x86Assembler64();
+	}
+
+	~x64GenContext_t()
+	{
+		delete emitter;
+	}
+
+	IMLInstruction* GetNextInstruction(sint32 relativeIndex = 1)
+	{
+		sint32 index = m_currentInstructionEmitIndex + relativeIndex;
+		if(index < 0 || index >= (sint32)currentSegment->imlList.size())
+			return nullptr;
+		return currentSegment->imlList.data() + index;
+	}
+
 	// relocate offsets
-	x64RelocEntry_t* relocateOffsetTable;
-	sint32 relocateOffsetTableSize;
-	sint32 relocateOffsetTableCount;
-}x64GenContext_t;
-
-// Some of these are defined by winnt.h and gnu headers
-#undef REG_EAX
-#undef REG_ECX
-#undef REG_EDX
-#undef REG_EBX
-#undef REG_ESP
-#undef REG_EBP
-#undef REG_ESI
-#undef REG_EDI
-#undef REG_NONE
-#undef REG_RAX
-#undef REG_RCX
-#undef REG_RDX
-#undef REG_RBX
-#undef REG_RSP
-#undef REG_RBP
-#undef REG_RSI
-#undef REG_RDI
-#undef REG_R8
-#undef REG_R9
-#undef REG_R10
-#undef REG_R11
-#undef REG_R12
-#undef REG_R13
-#undef REG_R14
-#undef REG_R15
-
-#define REG_EAX		0
-#define REG_ECX		1
-#define REG_EDX		2
-#define REG_EBX		3
-#define REG_ESP		4	// reserved for low half of hCPU pointer
-#define REG_EBP		5
-#define REG_ESI		6
-#define REG_EDI		7
-#define REG_NONE	-1
-
-#define REG_RAX		0
-#define REG_RCX		1
-#define REG_RDX		2
-#define REG_RBX		3
-#define REG_RSP		4	// reserved for hCPU pointer
-#define REG_RBP		5
-#define REG_RSI		6
-#define REG_RDI		7
-#define REG_R8		8
-#define REG_R9		9
-#define REG_R10		10
-#define REG_R11		11
-#define REG_R12		12
-#define REG_R13		13 // reserved to hold pointer to memory base? (Not decided yet)
-#define REG_R14		14 // reserved as temporary register
-#define REG_R15		15 // reserved for pointer to ppcRecompilerInstanceData
-
-#define REG_AL		0
-#define REG_CL		1
-#define REG_DL		2
-#define REG_BL		3
-#define REG_AH		4
-#define REG_CH		5
-#define REG_DH		6
-#define REG_BH		7
+	std::vector<x64RelocEntry_t> relocateOffsetTable2;
+};
 
 // reserved registers
-#define REG_RESV_TEMP		(REG_R14)
-#define REG_RESV_HCPU		(REG_RSP)
-#define REG_RESV_MEMBASE	(REG_R13)
-#define REG_RESV_RECDATA	(REG_R15)
+#define REG_RESV_TEMP		(X86_REG_R14)
+#define REG_RESV_HCPU		(X86_REG_RSP)
+#define REG_RESV_MEMBASE	(X86_REG_R13)
+#define REG_RESV_RECDATA	(X86_REG_R15)
 
 // reserved floating-point registers
 #define REG_RESV_FPR_TEMP	(15)
 
+#define reg32ToReg16(__x)	(__x) // deprecated
 
-extern sint32 x64Gen_registerMap[12];
-
-#define tempToRealRegister(__x) (x64Gen_registerMap[__x])
-#define tempToRealFPRRegister(__x) (__x)
-#define reg32ToReg16(__x)	(__x)
-
+// deprecated condition flags
 enum
 {
 	X86_CONDITION_EQUAL, // or zero
@@ -119,36 +71,23 @@ enum
 	X86_CONDITION_NONE, // no condition, jump always
 };
 
-#define PPCREC_CR_TEMPORARY							(8)		// never stored
-#define PPCREC_CR_STATE_TYPE_UNSIGNED_ARITHMETIC	(0)		// for signed arithmetic operations (ADD, CMPI)
-#define PPCREC_CR_STATE_TYPE_SIGNED_ARITHMETIC		(1)		// for unsigned arithmetic operations (ADD, CMPI)
-#define PPCREC_CR_STATE_TYPE_LOGICAL				(2)		// for unsigned operations (CMPLI)
-
-#define X86_RELOC_MAKE_RELATIVE				(0)		// make code imm relative to instruction
-#define X64_RELOC_LINK_TO_PPC				(1)		// translate from ppc address to x86 offset 
-#define X64_RELOC_LINK_TO_SEGMENT			(2)		// link to beginning of segment
-
-#define PPC_X64_GPR_USABLE_REGISTERS		(16-4)
-#define PPC_X64_FPR_USABLE_REGISTERS		(16-1) // Use XMM0 - XMM14, XMM15 is the temp register
-
-
-bool PPCRecompiler_generateX64Code(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext);
-
-void PPCRecompilerX64Gen_crConditionFlags_forget(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext);
+bool PPCRecompiler_generateX64Code(struct PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext);
 
 void PPCRecompilerX64Gen_redirectRelativeJump(x64GenContext_t* x64GenContext, sint32 jumpInstructionOffset, sint32 destinationOffset);
 
 void PPCRecompilerX64Gen_generateRecompilerInterfaceFunctions();
 
-void PPCRecompilerX64Gen_imlInstruction_fpr_r_name(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, PPCRecImlInstruction_t* imlInstruction);
-void PPCRecompilerX64Gen_imlInstruction_fpr_name_r(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, PPCRecImlInstruction_t* imlInstruction);
-bool PPCRecompilerX64Gen_imlInstruction_fpr_load(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, PPCRecImlInstruction_t* imlInstruction, bool indexed);
-bool PPCRecompilerX64Gen_imlInstruction_fpr_store(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, PPCRecImlInstruction_t* imlInstruction, bool indexed);
+void PPCRecompilerX64Gen_imlInstruction_fpr_r_name(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, IMLInstruction* imlInstruction);
+void PPCRecompilerX64Gen_imlInstruction_fpr_name_r(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, IMLInstruction* imlInstruction);
+bool PPCRecompilerX64Gen_imlInstruction_fpr_load(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, IMLInstruction* imlInstruction, bool indexed);
+bool PPCRecompilerX64Gen_imlInstruction_fpr_store(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, IMLInstruction* imlInstruction, bool indexed);
 
-void PPCRecompilerX64Gen_imlInstruction_fpr_r_r(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, PPCRecImlInstruction_t* imlInstruction);
-void PPCRecompilerX64Gen_imlInstruction_fpr_r_r_r(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, PPCRecImlInstruction_t* imlInstruction);
-void PPCRecompilerX64Gen_imlInstruction_fpr_r_r_r_r(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, PPCRecImlInstruction_t* imlInstruction);
-void PPCRecompilerX64Gen_imlInstruction_fpr_r(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, PPCRecImlInstruction_t* imlInstruction);
+void PPCRecompilerX64Gen_imlInstruction_fpr_r_r(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, IMLInstruction* imlInstruction);
+void PPCRecompilerX64Gen_imlInstruction_fpr_r_r_r(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, IMLInstruction* imlInstruction);
+void PPCRecompilerX64Gen_imlInstruction_fpr_r_r_r_r(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, IMLInstruction* imlInstruction);
+void PPCRecompilerX64Gen_imlInstruction_fpr_r(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, IMLInstruction* imlInstruction);
+
+void PPCRecompilerX64Gen_imlInstruction_fpr_compare(PPCRecFunction_t* PPCRecFunction, ppcImlGenContext_t* ppcImlGenContext, x64GenContext_t* x64GenContext, IMLInstruction* imlInstruction);
 
 // ASM gen
 void x64Gen_writeU8(x64GenContext_t* x64GenContext, uint8 v);
@@ -196,9 +135,6 @@ void x64Gen_or_reg64Low8_mem8Reg64(x64GenContext_t* x64GenContext, sint32 dstReg
 void x64Gen_and_reg64Low8_mem8Reg64(x64GenContext_t* x64GenContext, sint32 dstRegister, sint32 memRegister64, sint32 memImmS32);
 void x64Gen_mov_mem8Reg64_reg64Low8(x64GenContext_t* x64GenContext, sint32 dstRegister, sint32 memRegister64, sint32 memImmS32);
 
-void x64Gen_lock_cmpxchg_mem32Reg64PlusReg64_reg64(x64GenContext_t* x64GenContext, sint32 memRegisterA64, sint32 memRegisterB64, sint32 memImmS32, sint32 srcRegister);
-void x64Gen_lock_cmpxchg_mem32Reg64_reg64(x64GenContext_t* x64GenContext, sint32 memRegister64, sint32 memImmS32, sint32 srcRegister);
-
 void x64Gen_add_reg64_reg64(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 srcRegister);
 void x64Gen_add_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 srcRegister);
 void x64Gen_add_reg64_imm32(x64GenContext_t* x64GenContext, sint32 srcRegister, uint32 immU32);
@@ -207,9 +143,6 @@ void x64Gen_sub_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 des
 void x64Gen_sub_reg64Low32_imm32(x64GenContext_t* x64GenContext, sint32 srcRegister, uint32 immU32);
 void x64Gen_sub_reg64_imm32(x64GenContext_t* x64GenContext, sint32 srcRegister, uint32 immU32);
 void x64Gen_sub_mem32reg64_imm32(x64GenContext_t* x64GenContext, sint32 memRegister, sint32 memImmS32, uint64 immU32);
-void x64Gen_sbb_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 srcRegister);
-void x64Gen_adc_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 srcRegister);
-void x64Gen_adc_reg64Low32_imm32(x64GenContext_t* x64GenContext, sint32 srcRegister, uint32 immU32);
 void x64Gen_dec_mem32(x64GenContext_t* x64GenContext, sint32 memoryRegister, uint32 memoryImmU32);
 void x64Gen_imul_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 operandRegister);
 void x64Gen_idiv_reg64Low32(x64GenContext_t* x64GenContext, sint32 operandRegister);
@@ -241,9 +174,7 @@ void x64Gen_not_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister);
 void x64Gen_neg_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister);
 void x64Gen_cdq(x64GenContext_t* x64GenContext);
 
-void x64Gen_bswap_reg64(x64GenContext_t* x64GenContext, sint32 destRegister);
 void x64Gen_bswap_reg64Lower32bit(x64GenContext_t* x64GenContext, sint32 destRegister);
-void x64Gen_bswap_reg64Lower16bit(x64GenContext_t* x64GenContext, sint32 destRegister);
 
 void x64Gen_lzcnt_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 srcRegister);
 void x64Gen_bsr_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 srcRegister);
@@ -329,4 +260,8 @@ void x64Gen_movBEZeroExtend_reg64Low16_mem16Reg64PlusReg64(x64GenContext_t* x64G
 void x64Gen_movBETruncate_mem32Reg64PlusReg64_reg64(x64GenContext_t* x64GenContext, sint32 memRegisterA64, sint32 memRegisterB64, sint32 memImmS32, sint32 srcRegister);
 
 void x64Gen_shrx_reg64_reg64_reg64(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB);
-void x64Gen_shlx_reg64_reg64_reg64(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB);
+void x64Gen_shrx_reg32_reg32_reg32(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB);
+void x64Gen_sarx_reg64_reg64_reg64(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB);
+void x64Gen_sarx_reg32_reg32_reg32(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB);
+void x64Gen_shlx_reg64_reg64_reg64(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB);
+void x64Gen_shlx_reg32_reg32_reg32(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB);

src/Cafe/HW/Espresso/Recompiler/BackendX64/BackendX64AVX.cpp

@@ -1,5 +1,4 @@
-#include "PPCRecompiler.h"
-#include "PPCRecompilerX64.h"
+#include "BackendX64.h"
 
 void _x64Gen_writeMODRMDeprecated(x64GenContext_t* x64GenContext, sint32 dataRegister, sint32 memRegisterA64, sint32 memRegisterB64, sint32 memImmS32);
 
@@ -21,11 +20,10 @@ void _x64Gen_vex128_nds(x64GenContext_t* x64GenContext, uint8 opcodeMap, uint8 a
 	x64Gen_writeU8(x64GenContext, opcode);
 }
 
-#define VEX_PP_0F		0 // guessed
+#define VEX_PP_0F		0
 #define VEX_PP_66_0F	1
-#define VEX_PP_F3_0F	2 // guessed
-#define VEX_PP_F2_0F	3 // guessed
-
+#define VEX_PP_F3_0F	2
+#define VEX_PP_F2_0F	3
 
 void x64Gen_avx_VPUNPCKHQDQ_xmm_xmm_xmm(x64GenContext_t* x64GenContext, sint32 dstRegister, sint32 srcRegisterA, sint32 srcRegisterB)
 {

src/Cafe/HW/Espresso/Recompiler/BackendX64/BackendX64BMI.cpp

@@ -1,5 +1,4 @@
-#include "PPCRecompiler.h"
-#include "PPCRecompilerX64.h"
+#include "BackendX64.h"
 
 void _x64Gen_writeMODRMDeprecated(x64GenContext_t* x64GenContext, sint32 dataRegister, sint32 memRegisterA64, sint32 memRegisterB64, sint32 memImmS32);
 
@@ -69,6 +68,34 @@ void x64Gen_shrx_reg64_reg64_reg64(x64GenContext_t* x64GenContext, sint32 regist
 	x64Gen_writeU8(x64GenContext, 0xC0 + (registerDst & 7) * 8 + (registerA & 7));
 }
 
+void x64Gen_shrx_reg32_reg32_reg32(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB)
+{
+	x64Gen_writeU8(x64GenContext, 0xC4);
+	x64Gen_writeU8(x64GenContext, 0xE2 - ((registerDst >= 8) ? 0x80 : 0) - ((registerA >= 8) ? 0x20 : 0));
+	x64Gen_writeU8(x64GenContext, 0x7B - registerB * 8);
+	x64Gen_writeU8(x64GenContext, 0xF7);
+	x64Gen_writeU8(x64GenContext, 0xC0 + (registerDst & 7) * 8 + (registerA & 7));
+}
+
+void x64Gen_sarx_reg64_reg64_reg64(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB)
+{
+	// SARX reg64, reg64, reg64
+	x64Gen_writeU8(x64GenContext, 0xC4);
+	x64Gen_writeU8(x64GenContext, 0xE2 - ((registerDst >= 8) ? 0x80 : 0) - ((registerA >= 8) ? 0x20 : 0));
+	x64Gen_writeU8(x64GenContext, 0xFA - registerB * 8);
+	x64Gen_writeU8(x64GenContext, 0xF7);
+	x64Gen_writeU8(x64GenContext, 0xC0 + (registerDst & 7) * 8 + (registerA & 7));
+}
+
+void x64Gen_sarx_reg32_reg32_reg32(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB)
+{
+	x64Gen_writeU8(x64GenContext, 0xC4);
+	x64Gen_writeU8(x64GenContext, 0xE2 - ((registerDst >= 8) ? 0x80 : 0) - ((registerA >= 8) ? 0x20 : 0));
+	x64Gen_writeU8(x64GenContext, 0x7A - registerB * 8);
+	x64Gen_writeU8(x64GenContext, 0xF7);
+	x64Gen_writeU8(x64GenContext, 0xC0 + (registerDst & 7) * 8 + (registerA & 7));
+}
+
 void x64Gen_shlx_reg64_reg64_reg64(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB)
 {
 	// SHLX reg64, reg64, reg64
@@ -77,4 +104,13 @@ void x64Gen_shlx_reg64_reg64_reg64(x64GenContext_t* x64GenContext, sint32 regist
 	x64Gen_writeU8(x64GenContext, 0xF9 - registerB * 8);
 	x64Gen_writeU8(x64GenContext, 0xF7);
 	x64Gen_writeU8(x64GenContext, 0xC0 + (registerDst & 7) * 8 + (registerA & 7));
+}
+
+void x64Gen_shlx_reg32_reg32_reg32(x64GenContext_t* x64GenContext, sint32 registerDst, sint32 registerA, sint32 registerB)
+{
+	x64Gen_writeU8(x64GenContext, 0xC4);
+	x64Gen_writeU8(x64GenContext, 0xE2 - ((registerDst >= 8) ? 0x80 : 0) - ((registerA >= 8) ? 0x20 : 0));
+	x64Gen_writeU8(x64GenContext, 0x79 - registerB * 8);
+	x64Gen_writeU8(x64GenContext, 0xF7);
+	x64Gen_writeU8(x64GenContext, 0xC0 + (registerDst & 7) * 8 + (registerA & 7));
 }

src/Cafe/HW/Espresso/Recompiler/BackendX64/BackendX64Gen.cpp

@@ -1,62 +1,31 @@
-#include "PPCRecompiler.h"
-#include "PPCRecompilerIml.h"
-#include "PPCRecompilerX64.h"
+#include "BackendX64.h"
 
 // x86/x64 extension opcodes that could be useful:
 // ANDN
 // mulx, rorx, sarx, shlx, shrx
 // PDEP, PEXT
 
-void x64Gen_checkBuffer(x64GenContext_t* x64GenContext)
-{
-	// todo
-}
-
 void x64Gen_writeU8(x64GenContext_t* x64GenContext, uint8 v)
 {
-	if( x64GenContext->codeBufferIndex+1 > x64GenContext->codeBufferSize )
-	{
-		x64GenContext->codeBufferSize *= 2;
-		x64GenContext->codeBuffer = (uint8*)realloc(x64GenContext->codeBuffer, x64GenContext->codeBufferSize);
-	}
-	*(uint8*)(x64GenContext->codeBuffer+x64GenContext->codeBufferIndex) = v;
-	x64GenContext->codeBufferIndex++;
+	x64GenContext->emitter->_emitU8(v);
 }
 
 void x64Gen_writeU16(x64GenContext_t* x64GenContext, uint32 v)
 {
-	if( x64GenContext->codeBufferIndex+2 > x64GenContext->codeBufferSize )
-	{
-		x64GenContext->codeBufferSize *= 2;
-		x64GenContext->codeBuffer = (uint8*)realloc(x64GenContext->codeBuffer, x64GenContext->codeBufferSize);
-	}
-	*(uint16*)(x64GenContext->codeBuffer+x64GenContext->codeBufferIndex) = v;
-	x64GenContext->codeBufferIndex += 2;
+	x64GenContext->emitter->_emitU16(v);
 }
 
 void x64Gen_writeU32(x64GenContext_t* x64GenContext, uint32 v)
 {
-	if( x64GenContext->codeBufferIndex+4 > x64GenContext->codeBufferSize )
-	{
-		x64GenContext->codeBufferSize *= 2;
-		x64GenContext->codeBuffer = (uint8*)realloc(x64GenContext->codeBuffer, x64GenContext->codeBufferSize);
-	}
-	*(uint32*)(x64GenContext->codeBuffer+x64GenContext->codeBufferIndex) = v;
-	x64GenContext->codeBufferIndex += 4;
+	x64GenContext->emitter->_emitU32(v);
 }
 
 void x64Gen_writeU64(x64GenContext_t* x64GenContext, uint64 v)
 {
-	if( x64GenContext->codeBufferIndex+8 > x64GenContext->codeBufferSize )
-	{
-		x64GenContext->codeBufferSize *= 2;
-		x64GenContext->codeBuffer = (uint8*)realloc(x64GenContext->codeBuffer, x64GenContext->codeBufferSize);
-	}
-	*(uint64*)(x64GenContext->codeBuffer+x64GenContext->codeBufferIndex) = v;
-	x64GenContext->codeBufferIndex += 8;
+	x64GenContext->emitter->_emitU64(v);
 }
 
-#include "x64Emit.hpp"
+#include "X64Emit.hpp"
 
 void _x64Gen_writeMODRMDeprecated(x64GenContext_t* x64GenContext, sint32 dataRegister, sint32 memRegisterA64, sint32 memRegisterB64, sint32 memImmS32)
 {
@@ -67,7 +36,7 @@ void _x64Gen_writeMODRMDeprecated(x64GenContext_t* x64GenContext, sint32 dataReg
 		forceUseOffset = true;
 	}
 
-	if (memRegisterB64 == REG_NONE)
+	if (memRegisterB64 == X86_REG_NONE)
 	{
 		// memRegisterA64 + memImmS32
 		uint8 modRM = (dataRegister & 7) * 8 + (memRegisterA64 & 7);
@@ -352,7 +321,7 @@ void x64Gen_mov_mem32Reg64_imm32(x64GenContext_t* x64GenContext, sint32 memRegis
 void x64Gen_mov_mem64Reg64_imm32(x64GenContext_t* x64GenContext, sint32 memRegister, uint32 memImmU32, uint32 dataImmU32)
 {
 	// MOV QWORD [<memReg>+<memImmU32>], dataImmU32
-	if( memRegister == REG_R14 )
+	if( memRegister == X86_REG_R14 )
 	{
 		sint32 memImmS32 = (sint32)memImmU32;
 		if( memImmS32 == 0 )
@@ -384,7 +353,7 @@ void x64Gen_mov_mem64Reg64_imm32(x64GenContext_t* x64GenContext, sint32 memRegis
 void x64Gen_mov_mem8Reg64_imm8(x64GenContext_t* x64GenContext, sint32 memRegister, uint32 memImmU32, uint8 dataImmU8)
 {
 	// MOV BYTE [<memReg64>+<memImmU32>], dataImmU8
-	if( memRegister == REG_RSP )
+	if( memRegister == X86_REG_RSP )
 	{
 		sint32 memImmS32 = (sint32)memImmU32;
 		if( memImmS32 >= -128 && memImmS32 <= 127 )
@@ -625,7 +594,7 @@ void _x64_op_reg64Low_mem8Reg64(x64GenContext_t* x64GenContext, sint32 dstRegist
 	if (memRegister64 >= 8)
 		x64Gen_writeU8(x64GenContext, 0x41);
 	x64Gen_writeU8(x64GenContext, opByte);
-	_x64Gen_writeMODRMDeprecated(x64GenContext, dstRegister, memRegister64, REG_NONE, memImmS32);
+	_x64Gen_writeMODRMDeprecated(x64GenContext, dstRegister, memRegister64, X86_REG_NONE, memImmS32);
 }
 
 void x64Gen_or_reg64Low8_mem8Reg64(x64GenContext_t* x64GenContext, sint32 dstRegister, sint32 memRegister64, sint32 memImmS32)
@@ -643,40 +612,6 @@ void x64Gen_mov_mem8Reg64_reg64Low8(x64GenContext_t* x64GenContext, sint32 dstRe
 	_x64_op_reg64Low_mem8Reg64(x64GenContext, dstRegister, memRegister64, memImmS32, 0x88);
 }
 
-void x64Gen_lock_cmpxchg_mem32Reg64PlusReg64_reg64(x64GenContext_t* x64GenContext, sint32 memRegisterA64, sint32 memRegisterB64, sint32 memImmS32, sint32 srcRegister)
-{
-	// LOCK CMPXCHG DWORD [<reg64> + <reg64> + <imm64>], <srcReg64> (low dword)
-	x64Gen_writeU8(x64GenContext, 0xF0); // LOCK prefix
-
-	if( srcRegister >= 8 || memRegisterA64 >= 8|| memRegisterB64 >= 8 )
-		x64Gen_writeU8(x64GenContext, 0x40+((srcRegister>=8)?4:0)+((memRegisterA64>=8)?1:0)+((memRegisterB64>=8)?2:0));
-
-	x64Gen_writeU8(x64GenContext, 0x0F);
-	x64Gen_writeU8(x64GenContext, 0xB1);
-
-	_x64Gen_writeMODRMDeprecated(x64GenContext, srcRegister, memRegisterA64, memRegisterB64, memImmS32);
-}
-
-void x64Gen_lock_cmpxchg_mem32Reg64_reg64(x64GenContext_t* x64GenContext, sint32 memRegister64, sint32 memImmS32, sint32 srcRegister)
-{
-	// LOCK CMPXCHG DWORD [<reg64> + <imm64>], <srcReg64> (low dword)
-	x64Gen_writeU8(x64GenContext, 0xF0); // LOCK prefix
-
-	if( srcRegister >= 8 || memRegister64 >= 8 )
-		x64Gen_writeU8(x64GenContext, 0x40+((srcRegister>=8)?4:0)+((memRegister64>=8)?1:0));
-
-	x64Gen_writeU8(x64GenContext, 0x0F);
-	x64Gen_writeU8(x64GenContext, 0xB1);
-
-	if( memImmS32 == 0 )
-	{
-		x64Gen_writeU8(x64GenContext, 0x45+(srcRegister&7)*8);
-		x64Gen_writeU8(x64GenContext, 0x00);
-	}
-	else
-		assert_dbg();
-}
-
 void x64Gen_add_reg64_reg64(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 srcRegister)
 {
 	// ADD <destReg>, <srcReg>
@@ -732,7 +667,7 @@ void x64Gen_add_reg64Low32_imm32(x64GenContext_t* x64GenContext, sint32 srcRegis
 	}
 	else
 	{
-		if( srcRegister == REG_RAX )
+		if( srcRegister == X86_REG_RAX )
 		{
 			// special EAX short form
 			x64Gen_writeU8(x64GenContext, 0x05);
@@ -772,7 +707,7 @@ void x64Gen_sub_reg64Low32_imm32(x64GenContext_t* x64GenContext, sint32 srcRegis
 	}
 	else
 	{
-		if( srcRegister == REG_RAX )
+		if( srcRegister == X86_REG_RAX )
 		{
 			// special EAX short form
 			x64Gen_writeU8(x64GenContext, 0x2D);
@@ -811,7 +746,7 @@ void x64Gen_sub_mem32reg64_imm32(x64GenContext_t* x64GenContext, sint32 memRegis
 {
 	// SUB <mem32_memReg64>, <imm32>
 	sint32 immS32 = (sint32)immU32;
-	if( memRegister == REG_RSP )
+	if( memRegister == X86_REG_RSP )
 	{
 		if( memImmS32 >= 128 )
 		{
@@ -843,64 +778,11 @@ void x64Gen_sub_mem32reg64_imm32(x64GenContext_t* x64GenContext, sint32 memRegis
 	}
 }
 
-void x64Gen_sbb_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 srcRegister)
-{
-	// SBB <destReg64_low32>, <srcReg64_low32>
-	if( destRegister >= 8 && srcRegister >= 8 )
-		x64Gen_writeU8(x64GenContext, 0x45);
-	else if( srcRegister >= 8 )
-		x64Gen_writeU8(x64GenContext, 0x44);
-	else if( destRegister >= 8 )
-		x64Gen_writeU8(x64GenContext, 0x41);
-	x64Gen_writeU8(x64GenContext, 0x19);
-	x64Gen_writeU8(x64GenContext, 0xC0+(srcRegister&7)*8+(destRegister&7));
-}
-
-void x64Gen_adc_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 srcRegister)
-{
-	// ADC <destReg64_low32>, <srcReg64_low32>
-	if( destRegister >= 8 && srcRegister >= 8 )
-		x64Gen_writeU8(x64GenContext, 0x45);
-	else if( srcRegister >= 8 )
-		x64Gen_writeU8(x64GenContext, 0x44);
-	else if( destRegister >= 8 )
-		x64Gen_writeU8(x64GenContext, 0x41);
-	x64Gen_writeU8(x64GenContext, 0x11);
-	x64Gen_writeU8(x64GenContext, 0xC0+(srcRegister&7)*8+(destRegister&7));
-}
-
-void x64Gen_adc_reg64Low32_imm32(x64GenContext_t* x64GenContext, sint32 srcRegister, uint32 immU32)
-{
-	sint32 immS32 = (sint32)immU32;
-	if( srcRegister >= 8 )
-		x64Gen_writeU8(x64GenContext, 0x41);
-	if( immS32 >= -128 && immS32 <= 127 )
-	{
-		x64Gen_writeU8(x64GenContext, 0x83);
-		x64Gen_writeU8(x64GenContext, 0xD0+(srcRegister&7));
-		x64Gen_writeU8(x64GenContext, (uint8)immS32);
-	}
-	else
-	{
-		if( srcRegister == REG_RAX )
-		{
-			// special EAX short form
-			x64Gen_writeU8(x64GenContext, 0x15);
-		}
-		else
-		{
-			x64Gen_writeU8(x64GenContext, 0x81);
-			x64Gen_writeU8(x64GenContext, 0xD0+(srcRegister&7));
-		}
-		x64Gen_writeU32(x64GenContext, immU32);
-	}
-}
-
 void x64Gen_dec_mem32(x64GenContext_t* x64GenContext, sint32 memoryRegister, uint32 memoryImmU32)
 {
 	// DEC dword [<reg64>+imm]
 	sint32 memoryImmS32 = (sint32)memoryImmU32;
-	if (memoryRegister != REG_RSP)
+	if (memoryRegister != X86_REG_RSP)
 		assert_dbg(); // not supported yet
 	if (memoryImmS32 >= -128 && memoryImmS32 <= 127)
 	{
@@ -981,7 +863,7 @@ void x64Gen_and_reg64Low32_imm32(x64GenContext_t* x64GenContext, sint32 srcRegis
 	}
 	else
 	{
-		if( srcRegister == REG_RAX )
+		if( srcRegister == X86_REG_RAX )
 		{
 			// special EAX short form
 			x64Gen_writeU8(x64GenContext, 0x25);
@@ -1026,7 +908,7 @@ void x64Gen_test_reg64Low32_imm32(x64GenContext_t* x64GenContext, sint32 srcRegi
 	sint32 immS32 = (sint32)immU32;
 	if( srcRegister >= 8 )
 		x64Gen_writeU8(x64GenContext, 0x41);
-	if( srcRegister == REG_RAX )
+	if( srcRegister == X86_REG_RAX )
 	{
 		// special EAX short form
 		x64Gen_writeU8(x64GenContext, 0xA9);
@@ -1052,7 +934,7 @@ void x64Gen_cmp_reg64Low32_imm32(x64GenContext_t* x64GenContext, sint32 srcRegis
 	}
 	else
 	{
-		if( srcRegister == REG_RAX )
+		if( srcRegister == X86_REG_RAX )
 		{
 			// special RAX short form
 			x64Gen_writeU8(x64GenContext, 0x3D);
@@ -1082,7 +964,7 @@ void x64Gen_cmp_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 des
 void x64Gen_cmp_reg64Low32_mem32reg64(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 memRegister, sint32 memImmS32)
 {
 	// CMP <destReg64_lowDWORD>, DWORD [<memRegister>+<immS32>]
-	if( memRegister == REG_RSP )
+	if( memRegister == X86_REG_RSP )
 	{
 		if( memImmS32 >= -128 && memImmS32 <= 127 )
 			assert_dbg(); // todo -> Shorter instruction form
@@ -1112,7 +994,7 @@ void x64Gen_or_reg64Low32_imm32(x64GenContext_t* x64GenContext, sint32 srcRegist
 	}
 	else
 	{
-		if( srcRegister == REG_RAX )
+		if( srcRegister == X86_REG_RAX )
 		{
 			// special EAX short form
 			x64Gen_writeU8(x64GenContext, 0x0D);
@@ -1172,7 +1054,7 @@ void x64Gen_xor_reg64Low32_imm32(x64GenContext_t* x64GenContext, sint32 srcRegis
 	}
 	else
 	{
-		if( srcRegister == REG_RAX )
+		if( srcRegister == X86_REG_RAX )
 		{
 			// special EAX short form
 			x64Gen_writeU8(x64GenContext, 0x35);
@@ -1326,16 +1208,6 @@ void x64Gen_cdq(x64GenContext_t* x64GenContext)
 	x64Gen_writeU8(x64GenContext, 0x99);
 }
 
-void x64Gen_bswap_reg64(x64GenContext_t* x64GenContext, sint32 destRegister)
-{
-	if( destRegister >= 8 )
-		x64Gen_writeU8(x64GenContext, 0x41|8);
-	else
-		x64Gen_writeU8(x64GenContext, 0x40|8);
-	x64Gen_writeU8(x64GenContext, 0x0F);
-	x64Gen_writeU8(x64GenContext, 0xC8+(destRegister&7));
-}
-
 void x64Gen_bswap_reg64Lower32bit(x64GenContext_t* x64GenContext, sint32 destRegister)
 {
 	if( destRegister >= 8 )
@@ -1344,16 +1216,6 @@ void x64Gen_bswap_reg64Lower32bit(x64GenContext_t* x64GenContext, sint32 destReg
 	x64Gen_writeU8(x64GenContext, 0xC8+(destRegister&7));
 }
 
-void x64Gen_bswap_reg64Lower16bit(x64GenContext_t* x64GenContext, sint32 destRegister)
-{
-	assert_dbg(); // do not use this instruction, it's result is always undefined. Instead use ROL <reg16>, 8
-	//x64Gen_writeU8(x64GenContext, 0x66);
-	//if( destRegister >= 8 )
-	//	x64Gen_writeU8(x64GenContext, 0x41);
-	//x64Gen_writeU8(x64GenContext, 0x0F);
-	//x64Gen_writeU8(x64GenContext, 0xC8+(destRegister&7));
-}
-
 void x64Gen_lzcnt_reg64Low32_reg64Low32(x64GenContext_t* x64GenContext, sint32 destRegister, sint32 srcRegister)
 {
 	// SSE4
@@ -1388,7 +1250,7 @@ void x64Gen_setcc_mem8(x64GenContext_t* x64GenContext, sint32 conditionType, sin
 {
 	// SETcc [<reg64>+imm]
 	sint32 memoryImmS32 = (sint32)memoryImmU32;
-	if( memoryRegister != REG_RSP )
+	if( memoryRegister != X86_REG_RSP )
 		assert_dbg(); // not supported
 	if( memoryRegister >= 8 )
 		assert_dbg(); // not supported
@@ -1627,7 +1489,7 @@ void x64Gen_bt_mem8(x64GenContext_t* x64GenContext, sint32 memoryRegister, uint3
 {
 	// BT [<reg64>+imm], bitIndex	(bit test)
 	sint32 memoryImmS32 = (sint32)memoryImmU32;
-	if( memoryRegister != REG_RSP )
+	if( memoryRegister != X86_REG_RSP )
 		assert_dbg(); // not supported yet
 	if( memoryImmS32 >= -128 && memoryImmS32 <= 127 )
 	{
@@ -1662,7 +1524,7 @@ void x64Gen_jmp_imm32(x64GenContext_t* x64GenContext, uint32 destImm32)
 
 void x64Gen_jmp_memReg64(x64GenContext_t* x64GenContext, sint32 memRegister, uint32 immU32)
 {
-	if( memRegister == REG_NONE )
+	if( memRegister == X86_REG_NONE )
 	{
 		assert_dbg();
 	}

src/Cafe/HW/Espresso/Recompiler/BackendX64/BackendX64GenFPU.cpp

@@ -1,6 +1,4 @@
-#include "PPCRecompiler.h"
-#include "PPCRecompilerIml.h"
-#include "PPCRecompilerX64.h"
+#include "BackendX64.h"
 
 void x64Gen_genSSEVEXPrefix2(x64GenContext_t* x64GenContext, sint32 xmmRegister1, sint32 xmmRegister2, bool use64BitMode)
 {
@@ -44,7 +42,7 @@ void x64Gen_movupd_xmmReg_memReg128(x64GenContext_t* x64GenContext, sint32 xmmRe
 	// SSE2
 	// move two doubles from memory into xmm register
 	// MOVUPD <xmm>, [<reg>+<imm>]
-	if( memRegister == REG_ESP )
+	if( memRegister == X86_REG_ESP )
 	{
 		// todo: Short form of instruction if memImmU32 is 0 or in -128 to 127 range
 		// 66 0F 10 84 E4 23 01 00 00
@@ -56,7 +54,7 @@ void x64Gen_movupd_xmmReg_memReg128(x64GenContext_t* x64GenContext, sint32 xmmRe
 		x64Gen_writeU8(x64GenContext, 0xE4);
 		x64Gen_writeU32(x64GenContext, memImmU32);
 	}
-	else if( memRegister == REG_NONE )
+	else if( memRegister == X86_REG_NONE )
 	{
 		assert_dbg();
 		//x64Gen_writeU8(x64GenContext, 0x66);
@@ -76,7 +74,7 @@ void x64Gen_movupd_memReg128_xmmReg(x64GenContext_t* x64GenContext, sint32 xmmRe
 	// SSE2
 	// move two doubles from memory into xmm register
 	// MOVUPD [<reg>+<imm>], <xmm>
-	if( memRegister == REG_ESP )
+	if( memRegister == X86_REG_ESP )
 	{
 		// todo: Short form of instruction if memImmU32 is 0 or in -128 to 127 range
 		x64Gen_writeU8(x64GenContext, 0x66);
@@ -87,7 +85,7 @@ void x64Gen_movupd_memReg128_xmmReg(x64GenContext_t* x64GenContext, sint32 xmmRe
 		x64Gen_writeU8(x64GenContext, 0xE4);
 		x64Gen_writeU32(x64GenContext, memImmU32);
 	}
-	else if( memRegister == REG_NONE )
+	else if( memRegister == X86_REG_NONE )
 	{
 		assert_dbg();
 		//x64Gen_writeU8(x64GenContext, 0x66);
@@ -106,7 +104,7 @@ void x64Gen_movddup_xmmReg_memReg64(x64GenContext_t* x64GenContext, sint32 xmmRe
 {
 	// SSE3
 	// move one double from memory into lower and upper half of a xmm register
-	if( memRegister == REG_RSP )
+	if( memRegister == X86_REG_RSP )
 	{
 		// MOVDDUP <xmm>, [<reg>+<imm>]
 		// todo: Short form of instruction if memImmU32 is 0 or in -128 to 127 range
@@ -119,7 +117,7 @@ void x64Gen_movddup_xmmReg_memReg64(x64GenContext_t* x64GenContext, sint32 xmmRe
 		x64Gen_writeU8(x64GenContext, 0xE4);
 		x64Gen_writeU32(x64GenContext, memImmU32);
 	}
-	else if( memRegister == REG_R15 )
+	else if( memRegister == X86_REG_R15 )
 	{
 		// MOVDDUP <xmm>, [<reg>+<imm>]
 		// todo: Short form of instruction if memImmU32 is 0 or in -128 to 127 range
@@ -131,7 +129,7 @@ void x64Gen_movddup_xmmReg_memReg64(x64GenContext_t* x64GenContext, sint32 xmmRe
 		x64Gen_writeU8(x64GenContext, 0x87+(xmmRegister&7)*8);
 		x64Gen_writeU32(x64GenContext, memImmU32);
 	}
-	else if( memRegister == REG_NONE )
+	else if( memRegister == X86_REG_NONE )
 	{
 		// MOVDDUP <xmm>, [<imm>]
 		// 36 F2 0F 12 05 - 00 00 00 00
@@ -185,7 +183,7 @@ void x64Gen_movsd_memReg64_xmmReg(x64GenContext_t* x64GenContext, sint32 xmmRegi
 {
 	// SSE2
 	// move lower 64bits (double) of xmm register to memory location
-	if( memRegister == REG_NONE )
+	if( memRegister == X86_REG_NONE )
 	{
 		// MOVSD [<imm>], <xmm>
 		// F2 0F 11 05 - 45 23 01 00
@@ -197,7 +195,7 @@ void x64Gen_movsd_memReg64_xmmReg(x64GenContext_t* x64GenContext, sint32 xmmRegi
 		//x64Gen_writeU8(x64GenContext, 0x05+xmmRegister*8);
 		//x64Gen_writeU32(x64GenContext, memImmU32);
 	}
-	else if( memRegister == REG_RSP )
+	else if( memRegister == X86_REG_RSP )
 	{
 		// MOVSD [RSP+<imm>], <xmm>
 		// F2 0F 11 84 24 - 33 22 11 00
@@ -219,7 +217,7 @@ void x64Gen_movlpd_xmmReg_memReg64(x64GenContext_t* x64GenContext, sint32 xmmReg
 {
 	// SSE3
 	// move one double from memory into lower half of a xmm register, leave upper half unchanged(?)
-	if( memRegister == REG_NONE )
+	if( memRegister == X86_REG_NONE )
 	{
 		// MOVLPD <xmm>, [<imm>]
 		//x64Gen_writeU8(x64GenContext, 0x66);
@@ -229,7 +227,7 @@ void x64Gen_movlpd_xmmReg_memReg64(x64GenContext_t* x64GenContext, sint32 xmmReg
 		//x64Gen_writeU32(x64GenContext, memImmU32);
 		assert_dbg();
 	}
-	else if( memRegister == REG_RSP )
+	else if( memRegister == X86_REG_RSP )
 	{
 		// MOVLPD <xmm>, [<reg64>+<imm>]
 		// 66 0F 12 84 24 - 33 22 11 00
@@ -348,11 +346,11 @@ void x64Gen_mulpd_xmmReg_xmmReg(x64GenContext_t* x64GenContext, sint32 xmmRegist
 void x64Gen_mulpd_xmmReg_memReg128(x64GenContext_t* x64GenContext, sint32 xmmRegister, sint32 memRegister, uint32 memImmU32)
 {
 	// SSE2
-	if (memRegister == REG_NONE)
+	if (memRegister == X86_REG_NONE)
 	{
 		assert_dbg();
 	}
-	else if (memRegister == REG_R14)
+	else if (memRegister == X86_REG_R14)
 	{
 		x64Gen_writeU8(x64GenContext, 0x66);
 		x64Gen_writeU8(x64GenContext, (xmmRegister < 8) ? 0x41 : 0x45);
@@ -404,7 +402,7 @@ void x64Gen_comisd_xmmReg_mem64Reg64(x64GenContext_t* x64GenContext, sint32 xmmR
 {
 	// SSE2
 	// compare bottom double with double from memory location
-	if( memoryReg == REG_R15 )
+	if( memoryReg == X86_REG_R15 )
 	{
 		x64Gen_writeU8(x64GenContext, 0x66);
 		x64Gen_genSSEVEXPrefix1(x64GenContext, xmmRegisterDest, true);
@@ -432,7 +430,7 @@ void x64Gen_comiss_xmmReg_mem64Reg64(x64GenContext_t* x64GenContext, sint32 xmmR
 {
 	// SSE2
 	// compare bottom float with float from memory location
-	if (memoryReg == REG_R15)
+	if (memoryReg == X86_REG_R15)
 	{
 		x64Gen_genSSEVEXPrefix1(x64GenContext, xmmRegisterDest, true);
 		x64Gen_writeU8(x64GenContext, 0x0F);
@@ -448,7 +446,7 @@ void x64Gen_orps_xmmReg_mem128Reg64(x64GenContext_t* x64GenContext, sint32 xmmRe
 {
 	// SSE2
 	// and xmm register with 128 bit value from memory
-	if( memReg == REG_R15 )
+	if( memReg == X86_REG_R15 )
 	{
 		x64Gen_genSSEVEXPrefix2(x64GenContext, memReg, xmmRegisterDest, false);
 		x64Gen_writeU8(x64GenContext, 0x0F);
@@ -464,7 +462,7 @@ void x64Gen_xorps_xmmReg_mem128Reg64(x64GenContext_t* x64GenContext, sint32 xmmR
 {
 	// SSE2
 	// xor xmm register with 128 bit value from memory
-	if( memReg == REG_R15 )
+	if( memReg == X86_REG_R15 )
 	{
 		x64Gen_genSSEVEXPrefix1(x64GenContext, xmmRegisterDest, true); // todo: should be x64Gen_genSSEVEXPrefix2() with memReg?
 		x64Gen_writeU8(x64GenContext, 0x0F);
@@ -479,11 +477,11 @@ void x64Gen_xorps_xmmReg_mem128Reg64(x64GenContext_t* x64GenContext, sint32 xmmR
 void x64Gen_andpd_xmmReg_memReg128(x64GenContext_t* x64GenContext, sint32 xmmRegister, sint32 memRegister, uint32 memImmU32)
 {
 	// SSE2
-	if (memRegister == REG_NONE)
+	if (memRegister == X86_REG_NONE)
 	{
 		assert_dbg();
 	}
-	else if (memRegister == REG_R14)
+	else if (memRegister == X86_REG_R14)
 	{
 		x64Gen_writeU8(x64GenContext, 0x66);
 		x64Gen_writeU8(x64GenContext, (xmmRegister < 8) ? 0x41 : 0x45);
@@ -502,7 +500,7 @@ void x64Gen_andps_xmmReg_mem128Reg64(x64GenContext_t* x64GenContext, sint32 xmmR
 {
 	// SSE2
 	// and xmm register with 128 bit value from memory
-	if( memReg == REG_R15 )
+	if( memReg == X86_REG_R15 )
 	{
 		x64Gen_genSSEVEXPrefix1(x64GenContext, xmmRegisterDest, true); // todo: should be x64Gen_genSSEVEXPrefix2() with memReg?
 		x64Gen_writeU8(x64GenContext, 0x0F);
@@ -528,7 +526,7 @@ void x64Gen_pcmpeqd_xmmReg_mem128Reg64(x64GenContext_t* x64GenContext, sint32 xm
 {
 	// SSE2
 	// doubleword integer compare
-	if( memReg == REG_R15 )
+	if( memReg == X86_REG_R15 )
 	{
 		x64Gen_writeU8(x64GenContext, 0x66);
 		x64Gen_genSSEVEXPrefix1(x64GenContext, xmmRegisterDest, true);
@@ -610,7 +608,7 @@ void x64Gen_cvtpi2pd_xmmReg_mem64Reg64(x64GenContext_t* x64GenContext, sint32 xm
 {
 	// SSE2
 	// converts two signed 32bit integers to two doubles
-	if( memReg == REG_RSP )
+	if( memReg == X86_REG_RSP )
 	{
 		x64Gen_writeU8(x64GenContext, 0x66);
 		x64Gen_genSSEVEXPrefix1(x64GenContext, xmmRegisterDest, false);
@@ -684,7 +682,7 @@ void x64Gen_rcpss_xmmReg_xmmReg(x64GenContext_t* x64GenContext, sint32 xmmRegist
 void x64Gen_mulss_xmmReg_memReg64(x64GenContext_t* x64GenContext, sint32 xmmRegister, sint32 memRegister, uint32 memImmU32)
 {
 	// SSE2
-	if( memRegister == REG_NONE )
+	if( memRegister == X86_REG_NONE )
 	{
 		assert_dbg();
 	}

src/Cafe/HW/Espresso/Recompiler/BackendX64/X64Emit.hpp

@@ -203,7 +203,6 @@ template<class opcodeBytes, typename TA, typename TB>
 void _x64Gen_writeMODRM_internal(x64GenContext_t* x64GenContext, TA opA, TB opB)
 {
 	static_assert(TA::getType() == MODRM_OPR_TYPE::REG);
-	x64Gen_checkBuffer(x64GenContext);
 	// REX prefix
 	// 0100 WRXB
 	if constexpr (TA::getType() == MODRM_OPR_TYPE::REG && TB::getType() == MODRM_OPR_TYPE::REG)

src/Cafe/HW/Espresso/Recompiler/IML/IML.h

@@ -0,0 +1,16 @@
+#pragma once 
+
+#include "IMLInstruction.h"
+#include "IMLSegment.h"
+
+// optimizer passes
+void IMLOptimizer_OptimizeDirectFloatCopies(struct ppcImlGenContext_t* ppcImlGenContext);
+void IMLOptimizer_OptimizeDirectIntegerCopies(struct ppcImlGenContext_t* ppcImlGenContext);
+void PPCRecompiler_optimizePSQLoadAndStore(struct ppcImlGenContext_t* ppcImlGenContext);
+
+void IMLOptimizer_StandardOptimizationPass(ppcImlGenContext_t& ppcImlGenContext);
+
+// debug
+void IMLDebug_DisassembleInstruction(const IMLInstruction& inst, std::string& disassemblyLineOut);
+void IMLDebug_DumpSegment(struct ppcImlGenContext_t* ctx, IMLSegment* imlSegment, bool printLivenessRangeInfo = false);
+void IMLDebug_Dump(struct ppcImlGenContext_t* ppcImlGenContext, bool printLivenessRangeInfo = false);

src/Cafe/HW/Espresso/Recompiler/IML/IMLAnalyzer.cpp

@@ -0,0 +1,5 @@
+#include "IML.h"
+//#include "PPCRecompilerIml.h"
+#include "util/helpers/fixedSizeList.h"
+
+#include "Cafe/HW/Espresso/Interpreter/PPCInterpreterInternal.h"

src/Cafe/HW/Espresso/Recompiler/IML/IMLDebug.cpp

@@ -0,0 +1,528 @@
+#include "IML.h"
+#include "IMLInstruction.h"
+#include "IMLSegment.h"
+#include "IMLRegisterAllocatorRanges.h"
+#include "util/helpers/StringBuf.h"
+
+#include "../PPCRecompiler.h"
+
+const char* IMLDebug_GetOpcodeName(const IMLInstruction* iml)
+{
+	static char _tempOpcodename[32];
+	uint32 op = iml->operation;
+	if (op == PPCREC_IML_OP_ASSIGN)
+		return "MOV";
+	else if (op == PPCREC_IML_OP_ADD)
+		return "ADD";
+	else if (op == PPCREC_IML_OP_ADD_WITH_CARRY)
+		return "ADC";
+	else if (op == PPCREC_IML_OP_SUB)
+		return "SUB";
+	else if (op == PPCREC_IML_OP_OR)
+		return "OR";
+	else if (op == PPCREC_IML_OP_AND)
+		return "AND";
+	else if (op == PPCREC_IML_OP_XOR)
+		return "XOR";
+	else if (op == PPCREC_IML_OP_LEFT_SHIFT)
+		return "LSH";
+	else if (op == PPCREC_IML_OP_RIGHT_SHIFT_U)
+		return "RSH";
+	else if (op == PPCREC_IML_OP_RIGHT_SHIFT_S)
+		return "ARSH";
+	else if (op == PPCREC_IML_OP_LEFT_ROTATE)
+		return "LROT";
+	else if (op == PPCREC_IML_OP_MULTIPLY_SIGNED)
+		return "MULS";
+	else if (op == PPCREC_IML_OP_DIVIDE_SIGNED)
+		return "DIVS";
+
+	sprintf(_tempOpcodename, "OP0%02x_T%d", iml->operation, iml->type);
+	return _tempOpcodename;
+}
+
+std::string IMLDebug_GetRegName(IMLReg r)
+{
+	std::string regName;
+	uint32 regId = r.GetRegID();
+	switch (r.GetRegFormat())
+	{
+	case IMLRegFormat::F32:
+		regName.append("f");
+		break;
+	case IMLRegFormat::F64:
+		regName.append("fd");
+		break;
+	case IMLRegFormat::I32:
+		regName.append("i");
+		break;
+	case IMLRegFormat::I64:
+		regName.append("r");
+		break;
+	default:
+		DEBUG_BREAK;
+	}
+	regName.append(fmt::format("{}", regId));
+	return regName;
+}
+
+void IMLDebug_AppendRegisterParam(StringBuf& strOutput, IMLReg virtualRegister, bool isLast = false)
+{
+	strOutput.add(IMLDebug_GetRegName(virtualRegister));
+	if (!isLast)
+		strOutput.add(", ");
+}
+
+void IMLDebug_AppendS32Param(StringBuf& strOutput, sint32 val, bool isLast = false)
+{
+	if (val < 0)
+	{
+		strOutput.add("-");
+		val = -val;
+	}
+	strOutput.addFmt("0x{:08x}", val);
+	if (!isLast)
+		strOutput.add(", ");
+}
+
+void IMLDebug_PrintLivenessRangeInfo(StringBuf& currentLineText, IMLSegment* imlSegment, sint32 offset)
+{
+	// pad to 70 characters
+	sint32 index = currentLineText.getLen();
+	while (index < 70)
+	{
+		currentLineText.add(" ");
+		index++;
+	}
+	raLivenessRange* subrangeItr = imlSegment->raInfo.linkedList_allSubranges;
+	while (subrangeItr)
+	{
+		if (subrangeItr->interval.start.GetInstructionIndexEx() == offset)
+		{
+			if(subrangeItr->interval.start.IsInstructionIndex() && !subrangeItr->interval.start.IsOnInputEdge())
+				currentLineText.add(".");
+			else
+				currentLineText.add("|");
+
+			currentLineText.addFmt("{:<4}", subrangeItr->GetVirtualRegister());
+		}
+		else if (subrangeItr->interval.end.GetInstructionIndexEx() == offset)
+		{
+			if(subrangeItr->interval.end.IsInstructionIndex() && !subrangeItr->interval.end.IsOnOutputEdge())
+				currentLineText.add("*    ");
+			else
+				currentLineText.add("|    ");
+		}
+		else if (subrangeItr->interval.ContainsInstructionIndexEx(offset))
+		{
+			currentLineText.add("|    ");
+		}
+		else
+		{
+			currentLineText.add("     ");
+		}
+		index += 5;
+		// next
+		subrangeItr = subrangeItr->link_allSegmentRanges.next;
+	}
+}
+
+std::string IMLDebug_GetSegmentName(ppcImlGenContext_t* ctx, IMLSegment* seg)
+{
+	if (!ctx)
+	{
+		return "<NoNameWithoutCtx>";
+	}
+	// find segment index
+	for (size_t i = 0; i < ctx->segmentList2.size(); i++)
+	{
+		if (ctx->segmentList2[i] == seg)
+		{
+			return fmt::format("Seg{:04x}", i);
+		}
+	}
+	return "<SegmentNotInCtx>";
+}
+
+std::string IMLDebug_GetConditionName(IMLCondition cond)
+{
+	switch (cond)
+	{
+	case IMLCondition::EQ:
+		return "EQ";
+	case IMLCondition::NEQ:
+		return "NEQ";
+	case IMLCondition::UNSIGNED_GT:
+		return "UGT";
+	case IMLCondition::UNSIGNED_LT:
+		return "ULT";
+	case IMLCondition::SIGNED_GT:
+		return "SGT";
+	case IMLCondition::SIGNED_LT:
+		return "SLT";
+	default:
+		cemu_assert_unimplemented();
+	}
+	return "ukn";
+}
+
+void IMLDebug_DisassembleInstruction(const IMLInstruction& inst, std::string& disassemblyLineOut)
+{
+	const sint32 lineOffsetParameters = 10;//18;
+
+	StringBuf strOutput(1024);
+	strOutput.reset();
+	if (inst.type == PPCREC_IML_TYPE_R_NAME || inst.type == PPCREC_IML_TYPE_NAME_R)
+	{
+		if (inst.type == PPCREC_IML_TYPE_R_NAME)
+			strOutput.add("R_NAME");
+		else
+			strOutput.add("NAME_R");
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+
+		if(inst.type == PPCREC_IML_TYPE_R_NAME)
+			IMLDebug_AppendRegisterParam(strOutput, inst.op_r_name.regR);
+
+		strOutput.add("name_");
+		if (inst.op_r_name.name >= PPCREC_NAME_R0 && inst.op_r_name.name < (PPCREC_NAME_R0 + 999))
+		{
+			strOutput.addFmt("r{}", inst.op_r_name.name - PPCREC_NAME_R0);
+		}
+		else if (inst.op_r_name.name >= PPCREC_NAME_FPR0 && inst.op_r_name.name < (PPCREC_NAME_FPR0 + 999))
+		{
+			strOutput.addFmt("f{}", inst.op_r_name.name - PPCREC_NAME_FPR0);
+		}
+		else if (inst.op_r_name.name >= PPCREC_NAME_SPR0 && inst.op_r_name.name < (PPCREC_NAME_SPR0 + 999))
+		{
+			strOutput.addFmt("spr{}", inst.op_r_name.name - PPCREC_NAME_SPR0);
+		}
+		else if (inst.op_r_name.name >= PPCREC_NAME_CR && inst.op_r_name.name <= PPCREC_NAME_CR_LAST)
+			strOutput.addFmt("cr{}", inst.op_r_name.name - PPCREC_NAME_CR);
+		else if (inst.op_r_name.name == PPCREC_NAME_XER_CA)
+			strOutput.add("xer.ca");
+		else if (inst.op_r_name.name == PPCREC_NAME_XER_SO)
+			strOutput.add("xer.so");
+		else if (inst.op_r_name.name == PPCREC_NAME_XER_OV)
+			strOutput.add("xer.ov");
+		else if (inst.op_r_name.name == PPCREC_NAME_CPU_MEMRES_EA)
+			strOutput.add("cpuReservation.ea");
+		else if (inst.op_r_name.name == PPCREC_NAME_CPU_MEMRES_VAL)
+			strOutput.add("cpuReservation.value");
+		else
+		{
+			strOutput.addFmt("name_ukn{}", inst.op_r_name.name);
+		}
+		if (inst.type != PPCREC_IML_TYPE_R_NAME)
+		{
+			strOutput.add(", ");
+			IMLDebug_AppendRegisterParam(strOutput, inst.op_r_name.regR, true);
+		}
+
+	}
+	else if (inst.type == PPCREC_IML_TYPE_R_R)
+	{
+		strOutput.addFmt("{}", IMLDebug_GetOpcodeName(&inst));
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r.regR);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r.regA, true);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_R_R_R)
+	{
+		strOutput.addFmt("{}", IMLDebug_GetOpcodeName(&inst));
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_r.regR);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_r.regA);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_r.regB, true);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_R_R_R_CARRY)
+	{
+		strOutput.addFmt("{}", IMLDebug_GetOpcodeName(&inst));
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_r_carry.regR);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_r_carry.regA);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_r_carry.regB);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_r_carry.regCarry, true);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_COMPARE)
+	{
+		strOutput.add("CMP ");
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_compare.regA);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_compare.regB);
+		strOutput.addFmt("{}", IMLDebug_GetConditionName(inst.op_compare.cond));
+		strOutput.add(" -> ");
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_compare.regR, true);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_COMPARE_S32)
+	{
+		strOutput.add("CMP ");
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_compare_s32.regA);
+		strOutput.addFmt("{}", inst.op_compare_s32.immS32);
+		strOutput.addFmt(", {}", IMLDebug_GetConditionName(inst.op_compare_s32.cond));
+		strOutput.add(" -> ");
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_compare_s32.regR, true);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_CONDITIONAL_JUMP)
+	{
+		strOutput.add("CJUMP ");
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_conditional_jump.registerBool, true);
+		if (!inst.op_conditional_jump.mustBeTrue)
+			strOutput.add("(inverted)");
+	}
+	else if (inst.type == PPCREC_IML_TYPE_JUMP)
+	{
+		strOutput.add("JUMP");
+	}
+	else if (inst.type == PPCREC_IML_TYPE_R_R_S32)
+	{
+		strOutput.addFmt("{}", IMLDebug_GetOpcodeName(&inst));
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_s32.regR);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_s32.regA);
+		IMLDebug_AppendS32Param(strOutput, inst.op_r_r_s32.immS32, true);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_R_R_S32_CARRY)
+	{
+		strOutput.addFmt("{}", IMLDebug_GetOpcodeName(&inst));
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_s32_carry.regR);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_s32_carry.regA);
+		IMLDebug_AppendS32Param(strOutput, inst.op_r_r_s32_carry.immS32);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_r_s32_carry.regCarry, true);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_R_S32)
+	{
+		strOutput.addFmt("{}", IMLDebug_GetOpcodeName(&inst));
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_r_immS32.regR);
+		IMLDebug_AppendS32Param(strOutput, inst.op_r_immS32.immS32, true);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_LOAD || inst.type == PPCREC_IML_TYPE_STORE ||
+			 inst.type == PPCREC_IML_TYPE_LOAD_INDEXED || inst.type == PPCREC_IML_TYPE_STORE_INDEXED)
+	{
+		if (inst.type == PPCREC_IML_TYPE_LOAD || inst.type == PPCREC_IML_TYPE_LOAD_INDEXED)
+			strOutput.add("LD_");
+		else
+			strOutput.add("ST_");
+
+		if (inst.op_storeLoad.flags2.signExtend)
+			strOutput.add("S");
+		else
+			strOutput.add("U");
+		strOutput.addFmt("{}", inst.op_storeLoad.copyWidth);
+
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_storeLoad.registerData);
+
+		if (inst.type == PPCREC_IML_TYPE_LOAD_INDEXED || inst.type == PPCREC_IML_TYPE_STORE_INDEXED)
+			strOutput.addFmt("[{}+{}]", IMLDebug_GetRegName(inst.op_storeLoad.registerMem), IMLDebug_GetRegName(inst.op_storeLoad.registerMem2));
+		else
+			strOutput.addFmt("[{}+{}]", IMLDebug_GetRegName(inst.op_storeLoad.registerMem), inst.op_storeLoad.immS32);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_ATOMIC_CMP_STORE)
+	{
+		strOutput.add("ATOMIC_ST_U32");
+
+		while ((sint32)strOutput.getLen() < lineOffsetParameters)
+			strOutput.add(" ");
+
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_atomic_compare_store.regEA);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_atomic_compare_store.regCompareValue);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_atomic_compare_store.regWriteValue);
+		IMLDebug_AppendRegisterParam(strOutput, inst.op_atomic_compare_store.regBoolOut, true);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_NO_OP)
+	{
+		strOutput.add("NOP");
+	}
+	else if (inst.type == PPCREC_IML_TYPE_MACRO)
+	{
+		if (inst.operation == PPCREC_IML_MACRO_B_TO_REG)
+		{
+			strOutput.addFmt("MACRO B_TO_REG {}", IMLDebug_GetRegName(inst.op_macro.paramReg));
+		}
+		else if (inst.operation == PPCREC_IML_MACRO_BL)
+		{
+			strOutput.addFmt("MACRO BL 0x{:08x} -> 0x{:08x} cycles (depr): {}", inst.op_macro.param, inst.op_macro.param2, (sint32)inst.op_macro.paramU16);
+		}
+		else if (inst.operation == PPCREC_IML_MACRO_B_FAR)
+		{
+			strOutput.addFmt("MACRO B_FAR 0x{:08x} -> 0x{:08x} cycles (depr): {}", inst.op_macro.param, inst.op_macro.param2, (sint32)inst.op_macro.paramU16);
+		}
+		else if (inst.operation == PPCREC_IML_MACRO_LEAVE)
+		{
+			strOutput.addFmt("MACRO LEAVE ppc: 0x{:08x}", inst.op_macro.param);
+		}
+		else if (inst.operation == PPCREC_IML_MACRO_HLE)
+		{
+			strOutput.addFmt("MACRO HLE ppcAddr: 0x{:08x} funcId: 0x{:08x}", inst.op_macro.param, inst.op_macro.param2);
+		}
+		else if (inst.operation == PPCREC_IML_MACRO_COUNT_CYCLES)
+		{
+			strOutput.addFmt("MACRO COUNT_CYCLES cycles: {}", inst.op_macro.param);
+		}
+		else
+		{
+			strOutput.addFmt("MACRO ukn operation {}", inst.operation);
+		}
+	}
+	else if (inst.type == PPCREC_IML_TYPE_FPR_LOAD)
+	{
+		strOutput.addFmt("{} = ", IMLDebug_GetRegName(inst.op_storeLoad.registerData));
+		if (inst.op_storeLoad.flags2.signExtend)
+			strOutput.add("S");
+		else
+			strOutput.add("U");
+		strOutput.addFmt("{} [{}+{}] mode {}", inst.op_storeLoad.copyWidth / 8, IMLDebug_GetRegName(inst.op_storeLoad.registerMem), inst.op_storeLoad.immS32, inst.op_storeLoad.mode);
+		if (inst.op_storeLoad.flags2.notExpanded)
+		{
+			strOutput.addFmt(" <No expand>");
+		}
+	}
+	else if (inst.type == PPCREC_IML_TYPE_FPR_STORE)
+	{
+		if (inst.op_storeLoad.flags2.signExtend)
+			strOutput.add("S");
+		else
+			strOutput.add("U");
+		strOutput.addFmt("{} [t{}+{}]", inst.op_storeLoad.copyWidth / 8, inst.op_storeLoad.registerMem.GetRegID(), inst.op_storeLoad.immS32);
+		strOutput.addFmt(" = {} mode {}", IMLDebug_GetRegName(inst.op_storeLoad.registerData), inst.op_storeLoad.mode);
+	}
+	else if (inst.type == PPCREC_IML_TYPE_FPR_R_R)
+	{
+		strOutput.addFmt("{:>6} ", IMLDebug_GetOpcodeName(&inst));
+		strOutput.addFmt("{}, {}", IMLDebug_GetRegName(inst.op_fpr_r_r.regR), IMLDebug_GetRegName(inst.op_fpr_r_r.regA));
+	}
+	else if (inst.type == PPCREC_IML_TYPE_FPR_R_R_R_R)
+	{
+		strOutput.addFmt("{:>6} ", IMLDebug_GetOpcodeName(&inst));
+		strOutput.addFmt("{}, {}, {}, {}", IMLDebug_GetRegName(inst.op_fpr_r_r_r_r.regR), IMLDebug_GetRegName(inst.op_fpr_r_r_r_r.regA), IMLDebug_GetRegName(inst.op_fpr_r_r_r_r.regB), IMLDebug_GetRegName(inst.op_fpr_r_r_r_r.regC));
+	}
+	else if (inst.type == PPCREC_IML_TYPE_FPR_R_R_R)
+	{
+		strOutput.addFmt("{:>6} ", IMLDebug_GetOpcodeName(&inst));
+		strOutput.addFmt("{}, {}, {}", IMLDebug_GetRegName(inst.op_fpr_r_r_r.regR), IMLDebug_GetRegName(inst.op_fpr_r_r_r.regA), IMLDebug_GetRegName(inst.op_fpr_r_r_r.regB));
+	}
+	else if (inst.type == PPCREC_IML_TYPE_CJUMP_CYCLE_CHECK)
+	{
+		strOutput.addFmt("CYCLE_CHECK");
+	}
+	else if (inst.type == PPCREC_IML_TYPE_X86_EFLAGS_JCC)
+	{
+		strOutput.addFmt("X86_JCC {}", IMLDebug_GetConditionName(inst.op_x86_eflags_jcc.cond));
+	}
+	else
+	{
+		strOutput.addFmt("Unknown iml type {}", inst.type);
+	}
+	disassemblyLineOut.assign(strOutput.c_str());
+}
+
+void IMLDebug_DumpSegment(ppcImlGenContext_t* ctx, IMLSegment* imlSegment, bool printLivenessRangeInfo)
+{
+	StringBuf strOutput(4096);
+
+	strOutput.addFmt("SEGMENT {} | PPC=0x{:08x} Loop-depth {}", IMLDebug_GetSegmentName(ctx, imlSegment), imlSegment->ppcAddress, imlSegment->loopDepth);
+	if (imlSegment->isEnterable)
+	{
+		strOutput.addFmt(" ENTERABLE (0x{:08x})", imlSegment->enterPPCAddress);
+	}
+	if (imlSegment->deadCodeEliminationHintSeg)
+	{
+		strOutput.addFmt(" InheritOverwrite: {}", IMLDebug_GetSegmentName(ctx, imlSegment->deadCodeEliminationHintSeg));
+	}
+	cemuLog_log(LogType::Force, "{}", strOutput.c_str());
+
+	if (printLivenessRangeInfo)
+	{
+		strOutput.reset();
+		IMLDebug_PrintLivenessRangeInfo(strOutput, imlSegment, RA_INTER_RANGE_START);
+		cemuLog_log(LogType::Force, "{}", strOutput.c_str());
+	}
+	//debug_printf("\n");
+	strOutput.reset();
+
+	std::string disassemblyLine;
+	for (sint32 i = 0; i < imlSegment->imlList.size(); i++)
+	{
+		const IMLInstruction& inst = imlSegment->imlList[i];
+		// don't log NOP instructions
+		if (inst.type == PPCREC_IML_TYPE_NO_OP)
+			continue;
+		strOutput.reset();
+		strOutput.addFmt("{:02x} ", i);
+		//cemuLog_log(LogType::Force, "{:02x} ", i);
+		disassemblyLine.clear();
+		IMLDebug_DisassembleInstruction(inst, disassemblyLine);
+		strOutput.add(disassemblyLine);
+		if (printLivenessRangeInfo)
+		{
+			IMLDebug_PrintLivenessRangeInfo(strOutput, imlSegment, i);
+		}
+		cemuLog_log(LogType::Force, "{}", strOutput.c_str());
+	}
+	// all ranges
+	if (printLivenessRangeInfo)
+	{
+		strOutput.reset();
+		strOutput.add("Ranges-VirtReg                                                        ");
+		raLivenessRange* subrangeItr = imlSegment->raInfo.linkedList_allSubranges;
+		while (subrangeItr)
+		{
+			strOutput.addFmt("v{:<4}", (uint32)subrangeItr->GetVirtualRegister());
+			subrangeItr = subrangeItr->link_allSegmentRanges.next;
+		}
+		cemuLog_log(LogType::Force, "{}", strOutput.c_str());
+		strOutput.reset();
+		strOutput.add("Ranges-PhysReg                                                        ");
+		subrangeItr = imlSegment->raInfo.linkedList_allSubranges;
+		while (subrangeItr)
+		{
+			strOutput.addFmt("p{:<4}", subrangeItr->GetPhysicalRegister());
+			subrangeItr = subrangeItr->link_allSegmentRanges.next;
+		}
+		cemuLog_log(LogType::Force, "{}", strOutput.c_str());
+	}
+	// branch info
+	strOutput.reset();
+	strOutput.add("Links from: ");
+	for (sint32 i = 0; i < imlSegment->list_prevSegments.size(); i++)
+	{
+		if (i)
+			strOutput.add(", ");
+		strOutput.addFmt("{}", IMLDebug_GetSegmentName(ctx, imlSegment->list_prevSegments[i]).c_str());
+	}
+	cemuLog_log(LogType::Force, "{}", strOutput.c_str());
+	if (imlSegment->nextSegmentBranchNotTaken)
+		cemuLog_log(LogType::Force, "BranchNotTaken: {}", IMLDebug_GetSegmentName(ctx, imlSegment->nextSegmentBranchNotTaken).c_str());
+	if (imlSegment->nextSegmentBranchTaken)
+		cemuLog_log(LogType::Force, "BranchTaken: {}", IMLDebug_GetSegmentName(ctx, imlSegment->nextSegmentBranchTaken).c_str());
+	if (imlSegment->nextSegmentIsUncertain)
+		cemuLog_log(LogType::Force, "Dynamic target");
+}
+
+void IMLDebug_Dump(ppcImlGenContext_t* ppcImlGenContext, bool printLivenessRangeInfo)
+{
+	for (size_t i = 0; i < ppcImlGenContext->segmentList2.size(); i++)
+	{
+		IMLDebug_DumpSegment(ppcImlGenContext, ppcImlGenContext->segmentList2[i], printLivenessRangeInfo);
+		cemuLog_log(LogType::Force, "");
+	}
+}

src/Cafe/HW/Espresso/Recompiler/IML/IMLInstruction.cpp

@@ -0,0 +1,669 @@
+#include "IMLInstruction.h"
+#include "IML.h"
+
+#include "../PPCRecompiler.h"
+#include "../PPCRecompilerIml.h"
+
+// return true if an instruction has side effects on top of just reading and writing registers
+bool IMLInstruction::HasSideEffects() const
+{
+	bool hasSideEffects = true;
+	if(type == PPCREC_IML_TYPE_R_R || type == PPCREC_IML_TYPE_R_R_S32 || type == PPCREC_IML_TYPE_COMPARE || type == PPCREC_IML_TYPE_COMPARE_S32)
+		hasSideEffects = false;
+	// todo - add more cases
+	return hasSideEffects;
+}
+
+void IMLInstruction::CheckRegisterUsage(IMLUsedRegisters* registersUsed) const
+{
+	registersUsed->readGPR1 = IMLREG_INVALID;
+	registersUsed->readGPR2 = IMLREG_INVALID;
+	registersUsed->readGPR3 = IMLREG_INVALID;
+	registersUsed->readGPR4 = IMLREG_INVALID;
+	registersUsed->writtenGPR1 = IMLREG_INVALID;
+	registersUsed->writtenGPR2 = IMLREG_INVALID;
+	if (type == PPCREC_IML_TYPE_R_NAME)
+	{
+		registersUsed->writtenGPR1 = op_r_name.regR;
+	}
+	else if (type == PPCREC_IML_TYPE_NAME_R)
+	{
+		registersUsed->readGPR1 = op_r_name.regR;
+	}
+	else if (type == PPCREC_IML_TYPE_R_R)
+	{
+		if (operation == PPCREC_IML_OP_X86_CMP)
+		{
+			// both operands are read only
+			registersUsed->readGPR1 = op_r_r.regR;
+			registersUsed->readGPR2 = op_r_r.regA;
+		}
+		else if (
+			operation == PPCREC_IML_OP_ASSIGN ||
+			operation == PPCREC_IML_OP_ENDIAN_SWAP ||
+			operation == PPCREC_IML_OP_CNTLZW ||
+			operation == PPCREC_IML_OP_NOT ||
+			operation == PPCREC_IML_OP_NEG ||
+			operation == PPCREC_IML_OP_ASSIGN_S16_TO_S32 ||
+			operation == PPCREC_IML_OP_ASSIGN_S8_TO_S32)
+		{
+			// result is written, operand is read
+			registersUsed->writtenGPR1 = op_r_r.regR;
+			registersUsed->readGPR1 = op_r_r.regA;
+		}
+		else
+			cemu_assert_unimplemented();
+	}
+	else if (type == PPCREC_IML_TYPE_R_S32)
+	{
+		cemu_assert_debug(operation != PPCREC_IML_OP_ADD &&
+			operation != PPCREC_IML_OP_SUB &&
+			operation != PPCREC_IML_OP_AND &&
+			operation != PPCREC_IML_OP_OR &&
+			operation != PPCREC_IML_OP_XOR); // deprecated, use r_r_s32 for these
+
+		if (operation == PPCREC_IML_OP_LEFT_ROTATE)
+		{
+			// register operand is read and write
+			registersUsed->readGPR1 = op_r_immS32.regR;
+			registersUsed->writtenGPR1 = op_r_immS32.regR;
+		}
+		else if (operation == PPCREC_IML_OP_X86_CMP)
+		{
+			// register operand is read only
+			registersUsed->readGPR1 = op_r_immS32.regR;
+		}
+		else
+		{
+			// register operand is write only
+			// todo - use explicit lists, avoid default cases
+			registersUsed->writtenGPR1 = op_r_immS32.regR;
+		}
+	}
+	else if (type == PPCREC_IML_TYPE_R_R_S32)
+	{
+		registersUsed->writtenGPR1 = op_r_r_s32.regR;
+		registersUsed->readGPR1 = op_r_r_s32.regA;
+	}
+	else if (type == PPCREC_IML_TYPE_R_R_S32_CARRY)
+	{
+		registersUsed->writtenGPR1 = op_r_r_s32_carry.regR;
+		registersUsed->readGPR1 = op_r_r_s32_carry.regA;
+		// some operations read carry
+		switch (operation)
+		{
+		case PPCREC_IML_OP_ADD_WITH_CARRY:
+			registersUsed->readGPR2 = op_r_r_s32_carry.regCarry;
+			break;
+		case PPCREC_IML_OP_ADD:
+			break;
+		default:
+			cemu_assert_unimplemented();
+		}
+		// carry is always written
+		registersUsed->writtenGPR2 = op_r_r_s32_carry.regCarry;
+	}
+	else if (type == PPCREC_IML_TYPE_R_R_R)
+	{
+		// in all cases result is written and other operands are read only
+		// with the exception of XOR, where if regA == regB then all bits are zeroed out. So we don't consider it a read
+		registersUsed->writtenGPR1 = op_r_r_r.regR;
+		if(!(operation == PPCREC_IML_OP_XOR && op_r_r_r.regA == op_r_r_r.regB))
+		{
+			registersUsed->readGPR1 = op_r_r_r.regA;
+			registersUsed->readGPR2 = op_r_r_r.regB;
+		}
+	}
+	else if (type == PPCREC_IML_TYPE_R_R_R_CARRY)
+	{
+		registersUsed->writtenGPR1 = op_r_r_r_carry.regR;
+		registersUsed->readGPR1 = op_r_r_r_carry.regA;
+		registersUsed->readGPR2 = op_r_r_r_carry.regB;
+		// some operations read carry
+		switch (operation)
+		{
+		case PPCREC_IML_OP_ADD_WITH_CARRY:
+			registersUsed->readGPR3 = op_r_r_r_carry.regCarry;
+			break;
+		case PPCREC_IML_OP_ADD:
+			break;
+		default:
+			cemu_assert_unimplemented();
+		}
+		// carry is always written
+		registersUsed->writtenGPR2 = op_r_r_r_carry.regCarry;
+	}
+	else if (type == PPCREC_IML_TYPE_CJUMP_CYCLE_CHECK)
+	{
+		// no effect on registers
+	}
+	else if (type == PPCREC_IML_TYPE_NO_OP)
+	{
+		// no effect on registers
+	}
+	else if (type == PPCREC_IML_TYPE_MACRO)
+	{
+		if (operation == PPCREC_IML_MACRO_BL || operation == PPCREC_IML_MACRO_B_FAR || operation == PPCREC_IML_MACRO_LEAVE || operation == PPCREC_IML_MACRO_DEBUGBREAK || operation == PPCREC_IML_MACRO_COUNT_CYCLES || operation == PPCREC_IML_MACRO_HLE)
+		{
+			// no effect on registers
+		}
+		else if (operation == PPCREC_IML_MACRO_B_TO_REG)
+		{
+			cemu_assert_debug(op_macro.paramReg.IsValid());
+			registersUsed->readGPR1 = op_macro.paramReg;
+		}
+		else
+			cemu_assert_unimplemented();
+	}
+	else if (type == PPCREC_IML_TYPE_COMPARE)
+	{
+		registersUsed->readGPR1 = op_compare.regA;
+		registersUsed->readGPR2 = op_compare.regB;
+		registersUsed->writtenGPR1 = op_compare.regR;
+	}
+	else if (type == PPCREC_IML_TYPE_COMPARE_S32)
+	{
+		registersUsed->readGPR1 = op_compare_s32.regA;
+		registersUsed->writtenGPR1 = op_compare_s32.regR;
+	}
+	else if (type == PPCREC_IML_TYPE_CONDITIONAL_JUMP)
+	{
+		registersUsed->readGPR1 = op_conditional_jump.registerBool;
+	}
+	else if (type == PPCREC_IML_TYPE_JUMP)
+	{
+		// no registers affected
+	}
+	else if (type == PPCREC_IML_TYPE_LOAD)
+	{
+		registersUsed->writtenGPR1 = op_storeLoad.registerData;
+		if (op_storeLoad.registerMem.IsValid())
+			registersUsed->readGPR1 = op_storeLoad.registerMem;
+	}
+	else if (type == PPCREC_IML_TYPE_LOAD_INDEXED)
+	{
+		registersUsed->writtenGPR1 = op_storeLoad.registerData;
+		if (op_storeLoad.registerMem.IsValid())
+			registersUsed->readGPR1 = op_storeLoad.registerMem;
+		if (op_storeLoad.registerMem2.IsValid())
+			registersUsed->readGPR2 = op_storeLoad.registerMem2;
+	}
+	else if (type == PPCREC_IML_TYPE_STORE)
+	{
+		registersUsed->readGPR1 = op_storeLoad.registerData;
+		if (op_storeLoad.registerMem.IsValid())
+			registersUsed->readGPR2 = op_storeLoad.registerMem;
+	}
+	else if (type == PPCREC_IML_TYPE_STORE_INDEXED)
+	{
+		registersUsed->readGPR1 = op_storeLoad.registerData;
+		if (op_storeLoad.registerMem.IsValid())
+			registersUsed->readGPR2 = op_storeLoad.registerMem;
+		if (op_storeLoad.registerMem2.IsValid())
+			registersUsed->readGPR3 = op_storeLoad.registerMem2;
+	}
+	else if (type == PPCREC_IML_TYPE_ATOMIC_CMP_STORE)
+	{
+		registersUsed->readGPR1 = op_atomic_compare_store.regEA;
+		registersUsed->readGPR2 = op_atomic_compare_store.regCompareValue;
+		registersUsed->readGPR3 = op_atomic_compare_store.regWriteValue;
+		registersUsed->writtenGPR1 = op_atomic_compare_store.regBoolOut;
+	}
+	else if (type == PPCREC_IML_TYPE_CALL_IMM)
+	{
+		if (op_call_imm.regParam0.IsValid())
+			registersUsed->readGPR1 = op_call_imm.regParam0;
+		if (op_call_imm.regParam1.IsValid())
+			registersUsed->readGPR2 = op_call_imm.regParam1;
+		if (op_call_imm.regParam2.IsValid())
+			registersUsed->readGPR3 = op_call_imm.regParam2;
+		registersUsed->writtenGPR1 = op_call_imm.regReturn;
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_LOAD)
+	{
+		// fpr load operation
+		registersUsed->writtenGPR1 = op_storeLoad.registerData;
+		// address is in gpr register
+		if (op_storeLoad.registerMem.IsValid())
+			registersUsed->readGPR1 = op_storeLoad.registerMem;
+		// determine partially written result
+		switch (op_storeLoad.mode)
+		{
+		case PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0:
+		case PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0_PS1:
+			cemu_assert_debug(op_storeLoad.registerGQR.IsValid());
+			registersUsed->readGPR2 = op_storeLoad.registerGQR;
+			break;
+		case PPCREC_FPR_LD_MODE_DOUBLE_INTO_PS0:
+			// PS1 remains the same
+			cemu_assert_debug(op_storeLoad.registerGQR.IsInvalid());
+			registersUsed->readGPR2 = op_storeLoad.registerData;
+			break;
+		case PPCREC_FPR_LD_MODE_SINGLE_INTO_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_FLOAT_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_FLOAT_PS0:
+		case PPCREC_FPR_LD_MODE_PSQ_S16_PS0:
+		case PPCREC_FPR_LD_MODE_PSQ_S16_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_U16_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_U16_PS0:
+		case PPCREC_FPR_LD_MODE_PSQ_S8_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_U8_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_U8_PS0:
+		case PPCREC_FPR_LD_MODE_PSQ_S8_PS0:
+			cemu_assert_debug(op_storeLoad.registerGQR.IsInvalid());
+			break;
+		default:
+			cemu_assert_unimplemented();
+		}
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_LOAD_INDEXED)
+	{
+		// fpr load operation
+		registersUsed->writtenGPR1 = op_storeLoad.registerData;
+		// address is in gpr registers
+		if (op_storeLoad.registerMem.IsValid())
+			registersUsed->readGPR1 = op_storeLoad.registerMem;
+		if (op_storeLoad.registerMem2.IsValid())
+			registersUsed->readGPR2 = op_storeLoad.registerMem2;
+		// determine partially written result
+		switch (op_storeLoad.mode)
+		{
+		case PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0:
+		case PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0_PS1:
+			cemu_assert_debug(op_storeLoad.registerGQR.IsValid());
+			registersUsed->readGPR3 = op_storeLoad.registerGQR;
+			break;
+		case PPCREC_FPR_LD_MODE_DOUBLE_INTO_PS0:
+			// PS1 remains the same
+			cemu_assert_debug(op_storeLoad.registerGQR.IsInvalid());
+			registersUsed->readGPR3 = op_storeLoad.registerData;
+			break;
+		case PPCREC_FPR_LD_MODE_SINGLE_INTO_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_FLOAT_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_FLOAT_PS0:
+		case PPCREC_FPR_LD_MODE_PSQ_S16_PS0:
+		case PPCREC_FPR_LD_MODE_PSQ_S16_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_U16_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_U16_PS0:
+		case PPCREC_FPR_LD_MODE_PSQ_S8_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_U8_PS0_PS1:
+		case PPCREC_FPR_LD_MODE_PSQ_U8_PS0:
+			cemu_assert_debug(op_storeLoad.registerGQR.IsInvalid());
+			break;
+		default:
+			cemu_assert_unimplemented();
+		}
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_STORE)
+	{
+		// fpr store operation
+		registersUsed->readGPR1 = op_storeLoad.registerData;
+		if (op_storeLoad.registerMem.IsValid())
+			registersUsed->readGPR2 = op_storeLoad.registerMem;
+		// PSQ generic stores also access GQR
+		switch (op_storeLoad.mode)
+		{
+		case PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0:
+		case PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0_PS1:
+			cemu_assert_debug(op_storeLoad.registerGQR.IsValid());
+			registersUsed->readGPR3 = op_storeLoad.registerGQR;
+			break;
+		default:
+			cemu_assert_debug(op_storeLoad.registerGQR.IsInvalid());
+			break;
+		}
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_STORE_INDEXED)
+	{
+		// fpr store operation
+		registersUsed->readGPR1 = op_storeLoad.registerData;
+		// address is in gpr registers
+		if (op_storeLoad.registerMem.IsValid())
+			registersUsed->readGPR2 = op_storeLoad.registerMem;
+		if (op_storeLoad.registerMem2.IsValid())
+			registersUsed->readGPR3 = op_storeLoad.registerMem2;
+		// PSQ generic stores also access GQR
+		switch (op_storeLoad.mode)
+		{
+		case PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0:
+		case PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0_PS1:
+			cemu_assert_debug(op_storeLoad.registerGQR.IsValid());
+			registersUsed->readGPR4 = op_storeLoad.registerGQR;
+			break;
+		default:
+			cemu_assert_debug(op_storeLoad.registerGQR.IsInvalid());
+			break;
+		}
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_R_R)
+	{
+		// fpr operation
+		if (operation == PPCREC_IML_OP_FPR_COPY_BOTTOM_TO_BOTTOM_AND_TOP ||
+			operation == PPCREC_IML_OP_FPR_COPY_TOP_TO_BOTTOM_AND_TOP ||
+			operation == PPCREC_IML_OP_FPR_COPY_BOTTOM_AND_TOP_SWAPPED ||
+			operation == PPCREC_IML_OP_ASSIGN ||
+			operation == PPCREC_IML_OP_FPR_NEGATE_PAIR ||
+			operation == PPCREC_IML_OP_FPR_ABS_PAIR ||
+			operation == PPCREC_IML_OP_FPR_FRES_PAIR ||
+			operation == PPCREC_IML_OP_FPR_FRSQRTE_PAIR)
+		{
+			// operand read, result written
+			registersUsed->readGPR1 = op_fpr_r_r.regA;
+			registersUsed->writtenGPR1 = op_fpr_r_r.regR;
+		}
+		else if (
+			operation == PPCREC_IML_OP_FPR_COPY_BOTTOM_TO_BOTTOM ||
+			operation == PPCREC_IML_OP_FPR_COPY_BOTTOM_TO_TOP ||
+			operation == PPCREC_IML_OP_FPR_COPY_TOP_TO_TOP ||
+			operation == PPCREC_IML_OP_FPR_COPY_TOP_TO_BOTTOM ||
+			operation == PPCREC_IML_OP_FPR_EXPAND_BOTTOM32_TO_BOTTOM64_AND_TOP64 ||
+			operation == PPCREC_IML_OP_FPR_BOTTOM_FCTIWZ ||
+			operation == PPCREC_IML_OP_FPR_BOTTOM_RECIPROCAL_SQRT
+			)
+		{
+			// operand read, result read and (partially) written
+			registersUsed->readGPR1 = op_fpr_r_r.regA;
+			registersUsed->readGPR2 = op_fpr_r_r.regR;
+			registersUsed->writtenGPR1 = op_fpr_r_r.regR;
+		}
+		else if (operation == PPCREC_IML_OP_FPR_MULTIPLY_BOTTOM ||
+			operation == PPCREC_IML_OP_FPR_MULTIPLY_PAIR ||
+			operation == PPCREC_IML_OP_FPR_DIVIDE_BOTTOM ||
+			operation == PPCREC_IML_OP_FPR_DIVIDE_PAIR ||
+			operation == PPCREC_IML_OP_FPR_ADD_BOTTOM ||
+			operation == PPCREC_IML_OP_FPR_ADD_PAIR ||
+			operation == PPCREC_IML_OP_FPR_SUB_PAIR ||
+			operation == PPCREC_IML_OP_FPR_SUB_BOTTOM)
+		{
+			// operand read, result read and written
+			registersUsed->readGPR1 = op_fpr_r_r.regA;
+			registersUsed->readGPR2 = op_fpr_r_r.regR;
+			registersUsed->writtenGPR1 = op_fpr_r_r.regR;
+
+		}
+		else if (operation == PPCREC_IML_OP_FPR_FCMPU_BOTTOM ||
+			operation == PPCREC_IML_OP_FPR_FCMPU_TOP ||
+			operation == PPCREC_IML_OP_FPR_FCMPO_BOTTOM)
+		{
+			// operand read, result read
+			registersUsed->readGPR1 = op_fpr_r_r.regA;
+			registersUsed->readGPR2 = op_fpr_r_r.regR;
+		}
+		else
+			cemu_assert_unimplemented();
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_R_R_R)
+	{
+		// fpr operation
+		registersUsed->readGPR1 = op_fpr_r_r_r.regA;
+		registersUsed->readGPR2 = op_fpr_r_r_r.regB;
+		registersUsed->writtenGPR1 = op_fpr_r_r_r.regR;
+		// handle partially written result
+		switch (operation)
+		{
+		case PPCREC_IML_OP_FPR_MULTIPLY_BOTTOM:
+		case PPCREC_IML_OP_FPR_ADD_BOTTOM:
+		case PPCREC_IML_OP_FPR_SUB_BOTTOM:
+			registersUsed->readGPR3 = op_fpr_r_r_r.regR;
+			break;
+		case PPCREC_IML_OP_FPR_SUB_PAIR:
+			break;
+		default:
+			cemu_assert_unimplemented();
+		}
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_R_R_R_R)
+	{
+		// fpr operation
+		registersUsed->readGPR1 = op_fpr_r_r_r_r.regA;
+		registersUsed->readGPR2 = op_fpr_r_r_r_r.regB;
+		registersUsed->readGPR3 = op_fpr_r_r_r_r.regC;
+		registersUsed->writtenGPR1 = op_fpr_r_r_r_r.regR;
+		// handle partially written result
+		switch (operation)
+		{
+		case PPCREC_IML_OP_FPR_SELECT_BOTTOM:
+			registersUsed->readGPR4 = op_fpr_r_r_r_r.regR;
+			break;
+		case PPCREC_IML_OP_FPR_SUM0:
+		case PPCREC_IML_OP_FPR_SUM1:
+		case PPCREC_IML_OP_FPR_SELECT_PAIR:
+			break;
+		default:
+			cemu_assert_unimplemented();
+		}
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_R)
+	{
+		// fpr operation
+		if (operation == PPCREC_IML_OP_FPR_NEGATE_BOTTOM ||
+			operation == PPCREC_IML_OP_FPR_ABS_BOTTOM ||
+			operation == PPCREC_IML_OP_FPR_NEGATIVE_ABS_BOTTOM ||
+			operation == PPCREC_IML_OP_FPR_EXPAND_BOTTOM32_TO_BOTTOM64_AND_TOP64 ||
+			operation == PPCREC_IML_OP_FPR_ROUND_TO_SINGLE_PRECISION_BOTTOM ||
+			operation == PPCREC_IML_OP_FPR_ROUND_TO_SINGLE_PRECISION_PAIR)
+		{
+			registersUsed->readGPR1 = op_fpr_r.regR;
+			registersUsed->writtenGPR1 = op_fpr_r.regR;
+		}
+		else
+			cemu_assert_unimplemented();
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_COMPARE)
+	{
+		registersUsed->writtenGPR1 = op_fpr_compare.regR;
+		registersUsed->readGPR1 = op_fpr_compare.regA;
+		registersUsed->readGPR2 = op_fpr_compare.regB;
+	}
+	else if (type == PPCREC_IML_TYPE_X86_EFLAGS_JCC)
+	{
+		// no registers read or written (except for the implicit eflags)
+	}
+	else
+	{
+		cemu_assert_unimplemented();
+	}
+}
+
+IMLReg replaceRegisterIdMultiple(IMLReg reg, const std::unordered_map<IMLRegID, IMLRegID>& translationTable)
+{
+	if (reg.IsInvalid())
+		return reg;
+	const auto& it = translationTable.find(reg.GetRegID());
+	cemu_assert_debug(it != translationTable.cend());
+	IMLReg alteredReg = reg;
+	alteredReg.SetRegID(it->second);
+	return alteredReg;
+}
+
+void IMLInstruction::RewriteGPR(const std::unordered_map<IMLRegID, IMLRegID>& translationTable)
+{
+	if (type == PPCREC_IML_TYPE_R_NAME)
+	{
+		op_r_name.regR = replaceRegisterIdMultiple(op_r_name.regR, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_NAME_R)
+	{
+		op_r_name.regR = replaceRegisterIdMultiple(op_r_name.regR, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_R_R)
+	{
+		op_r_r.regR = replaceRegisterIdMultiple(op_r_r.regR, translationTable);
+		op_r_r.regA = replaceRegisterIdMultiple(op_r_r.regA, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_R_S32)
+	{
+		op_r_immS32.regR = replaceRegisterIdMultiple(op_r_immS32.regR, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_R_R_S32)
+	{
+		op_r_r_s32.regR = replaceRegisterIdMultiple(op_r_r_s32.regR, translationTable);
+		op_r_r_s32.regA = replaceRegisterIdMultiple(op_r_r_s32.regA, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_R_R_S32_CARRY)
+	{
+		op_r_r_s32_carry.regR = replaceRegisterIdMultiple(op_r_r_s32_carry.regR, translationTable);
+		op_r_r_s32_carry.regA = replaceRegisterIdMultiple(op_r_r_s32_carry.regA, translationTable);
+		op_r_r_s32_carry.regCarry = replaceRegisterIdMultiple(op_r_r_s32_carry.regCarry, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_R_R_R)
+	{
+		op_r_r_r.regR = replaceRegisterIdMultiple(op_r_r_r.regR, translationTable);
+		op_r_r_r.regA = replaceRegisterIdMultiple(op_r_r_r.regA, translationTable);
+		op_r_r_r.regB = replaceRegisterIdMultiple(op_r_r_r.regB, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_R_R_R_CARRY)
+	{
+		op_r_r_r_carry.regR = replaceRegisterIdMultiple(op_r_r_r_carry.regR, translationTable);
+		op_r_r_r_carry.regA = replaceRegisterIdMultiple(op_r_r_r_carry.regA, translationTable);
+		op_r_r_r_carry.regB = replaceRegisterIdMultiple(op_r_r_r_carry.regB, translationTable);
+		op_r_r_r_carry.regCarry = replaceRegisterIdMultiple(op_r_r_r_carry.regCarry, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_COMPARE)
+	{
+		op_compare.regR = replaceRegisterIdMultiple(op_compare.regR, translationTable);
+		op_compare.regA = replaceRegisterIdMultiple(op_compare.regA, translationTable);
+		op_compare.regB = replaceRegisterIdMultiple(op_compare.regB, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_COMPARE_S32)
+	{
+		op_compare_s32.regR = replaceRegisterIdMultiple(op_compare_s32.regR, translationTable);
+		op_compare_s32.regA = replaceRegisterIdMultiple(op_compare_s32.regA, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_CONDITIONAL_JUMP)
+	{
+		op_conditional_jump.registerBool = replaceRegisterIdMultiple(op_conditional_jump.registerBool, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_CJUMP_CYCLE_CHECK || type == PPCREC_IML_TYPE_JUMP)
+	{
+		// no effect on registers
+	}
+	else if (type == PPCREC_IML_TYPE_NO_OP)
+	{
+		// no effect on registers
+	}
+	else if (type == PPCREC_IML_TYPE_MACRO)
+	{
+		if (operation == PPCREC_IML_MACRO_BL || operation == PPCREC_IML_MACRO_B_FAR || operation == PPCREC_IML_MACRO_LEAVE || operation == PPCREC_IML_MACRO_DEBUGBREAK || operation == PPCREC_IML_MACRO_HLE || operation == PPCREC_IML_MACRO_COUNT_CYCLES)
+		{
+			// no effect on registers
+		}
+		else if (operation == PPCREC_IML_MACRO_B_TO_REG)
+		{
+			op_macro.paramReg = replaceRegisterIdMultiple(op_macro.paramReg, translationTable);
+		}
+		else
+		{
+			cemu_assert_unimplemented();
+		}
+	}
+	else if (type == PPCREC_IML_TYPE_LOAD)
+	{
+		op_storeLoad.registerData = replaceRegisterIdMultiple(op_storeLoad.registerData, translationTable);
+		if (op_storeLoad.registerMem.IsValid())
+		{
+			op_storeLoad.registerMem = replaceRegisterIdMultiple(op_storeLoad.registerMem, translationTable);
+		}
+	}
+	else if (type == PPCREC_IML_TYPE_LOAD_INDEXED)
+	{
+		op_storeLoad.registerData = replaceRegisterIdMultiple(op_storeLoad.registerData, translationTable);
+		if (op_storeLoad.registerMem.IsValid())
+			op_storeLoad.registerMem = replaceRegisterIdMultiple(op_storeLoad.registerMem, translationTable);
+		if (op_storeLoad.registerMem2.IsValid())
+			op_storeLoad.registerMem2 = replaceRegisterIdMultiple(op_storeLoad.registerMem2, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_STORE)
+	{
+		op_storeLoad.registerData = replaceRegisterIdMultiple(op_storeLoad.registerData, translationTable);
+		if (op_storeLoad.registerMem.IsValid())
+			op_storeLoad.registerMem = replaceRegisterIdMultiple(op_storeLoad.registerMem, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_STORE_INDEXED)
+	{
+		op_storeLoad.registerData = replaceRegisterIdMultiple(op_storeLoad.registerData, translationTable);
+		if (op_storeLoad.registerMem.IsValid())
+			op_storeLoad.registerMem = replaceRegisterIdMultiple(op_storeLoad.registerMem, translationTable);
+		if (op_storeLoad.registerMem2.IsValid())
+			op_storeLoad.registerMem2 = replaceRegisterIdMultiple(op_storeLoad.registerMem2, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_ATOMIC_CMP_STORE)
+	{
+		op_atomic_compare_store.regEA = replaceRegisterIdMultiple(op_atomic_compare_store.regEA, translationTable);
+		op_atomic_compare_store.regCompareValue = replaceRegisterIdMultiple(op_atomic_compare_store.regCompareValue, translationTable);
+		op_atomic_compare_store.regWriteValue = replaceRegisterIdMultiple(op_atomic_compare_store.regWriteValue, translationTable);
+		op_atomic_compare_store.regBoolOut = replaceRegisterIdMultiple(op_atomic_compare_store.regBoolOut, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_CALL_IMM)
+	{
+		op_call_imm.regReturn = replaceRegisterIdMultiple(op_call_imm.regReturn, translationTable);
+		if (op_call_imm.regParam0.IsValid())
+			op_call_imm.regParam0 = replaceRegisterIdMultiple(op_call_imm.regParam0, translationTable);
+		if (op_call_imm.regParam1.IsValid())
+			op_call_imm.regParam1 = replaceRegisterIdMultiple(op_call_imm.regParam1, translationTable);
+		if (op_call_imm.regParam2.IsValid())
+			op_call_imm.regParam2 = replaceRegisterIdMultiple(op_call_imm.regParam2, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_LOAD)
+	{
+		op_storeLoad.registerData = replaceRegisterIdMultiple(op_storeLoad.registerData, translationTable);
+		op_storeLoad.registerMem = replaceRegisterIdMultiple(op_storeLoad.registerMem, translationTable);
+		op_storeLoad.registerGQR = replaceRegisterIdMultiple(op_storeLoad.registerGQR, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_LOAD_INDEXED)
+	{
+		op_storeLoad.registerData = replaceRegisterIdMultiple(op_storeLoad.registerData, translationTable);
+		op_storeLoad.registerMem = replaceRegisterIdMultiple(op_storeLoad.registerMem, translationTable);
+		op_storeLoad.registerMem2 = replaceRegisterIdMultiple(op_storeLoad.registerMem2, translationTable);
+		op_storeLoad.registerGQR = replaceRegisterIdMultiple(op_storeLoad.registerGQR, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_STORE)
+	{
+		op_storeLoad.registerData = replaceRegisterIdMultiple(op_storeLoad.registerData, translationTable);
+		op_storeLoad.registerMem = replaceRegisterIdMultiple(op_storeLoad.registerMem, translationTable);
+		op_storeLoad.registerGQR = replaceRegisterIdMultiple(op_storeLoad.registerGQR, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_STORE_INDEXED)
+	{
+		op_storeLoad.registerData = replaceRegisterIdMultiple(op_storeLoad.registerData, translationTable);
+		op_storeLoad.registerMem = replaceRegisterIdMultiple(op_storeLoad.registerMem, translationTable);
+		op_storeLoad.registerMem2 = replaceRegisterIdMultiple(op_storeLoad.registerMem2, translationTable);
+		op_storeLoad.registerGQR = replaceRegisterIdMultiple(op_storeLoad.registerGQR, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_R)
+	{
+		op_fpr_r.regR = replaceRegisterIdMultiple(op_fpr_r.regR, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_R_R)
+	{
+		op_fpr_r_r.regR = replaceRegisterIdMultiple(op_fpr_r_r.regR, translationTable);
+		op_fpr_r_r.regA = replaceRegisterIdMultiple(op_fpr_r_r.regA, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_R_R_R)
+	{
+		op_fpr_r_r_r.regR = replaceRegisterIdMultiple(op_fpr_r_r_r.regR, translationTable);
+		op_fpr_r_r_r.regA = replaceRegisterIdMultiple(op_fpr_r_r_r.regA, translationTable);
+		op_fpr_r_r_r.regB = replaceRegisterIdMultiple(op_fpr_r_r_r.regB, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_R_R_R_R)
+	{
+		op_fpr_r_r_r_r.regR = replaceRegisterIdMultiple(op_fpr_r_r_r_r.regR, translationTable);
+		op_fpr_r_r_r_r.regA = replaceRegisterIdMultiple(op_fpr_r_r_r_r.regA, translationTable);
+		op_fpr_r_r_r_r.regB = replaceRegisterIdMultiple(op_fpr_r_r_r_r.regB, translationTable);
+		op_fpr_r_r_r_r.regC = replaceRegisterIdMultiple(op_fpr_r_r_r_r.regC, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_FPR_COMPARE)
+	{
+		op_fpr_compare.regA = replaceRegisterIdMultiple(op_fpr_compare.regA, translationTable);
+		op_fpr_compare.regB = replaceRegisterIdMultiple(op_fpr_compare.regB, translationTable);
+		op_fpr_compare.regR = replaceRegisterIdMultiple(op_fpr_compare.regR, translationTable);
+	}
+	else if (type == PPCREC_IML_TYPE_X86_EFLAGS_JCC)
+	{
+		// no registers read or written (except for the implicit eflags)
+	}
+	else
+	{
+		cemu_assert_unimplemented();
+	}
+}

src/Cafe/HW/Espresso/Recompiler/IML/IMLInstruction.h

@@ -0,0 +1,785 @@
+#pragma once
+
+using IMLRegID = uint16; // 16 bit ID
+using IMLPhysReg = sint32; // arbitrary value that is up to the architecture backend, usually this will be the register index. A value of -1 is reserved and means not assigned
+
+// format of IMLReg:
+// 0-15		(16 bit)	IMLRegID
+// 19-23	(5 bit)		Offset				In elements, for SIMD registers
+// 24-27	(4 bit)		IMLRegFormat		RegFormat
+// 28-31	(4 bit)		IMLRegFormat		BaseFormat
+
+enum class IMLRegFormat : uint8
+{
+	INVALID_FORMAT,
+	I64,
+	I32,
+	I16,
+	I8,
+	// I1 ?
+	F64,
+	F32,
+	TYPE_COUNT,
+};
+
+class IMLReg
+{
+public:
+	IMLReg()
+	{
+		m_raw = 0; // 0 is invalid
+	}
+
+	IMLReg(IMLRegFormat baseRegFormat, IMLRegFormat regFormat, uint8 viewOffset, IMLRegID regId)
+	{
+		m_raw = 0;
+		m_raw |= ((uint8)baseRegFormat << 28);
+		m_raw |= ((uint8)regFormat << 24);
+		m_raw |= (uint32)regId;
+	}
+
+	IMLReg(IMLReg&& baseReg, IMLRegFormat viewFormat, uint8 viewOffset, IMLRegID regId)
+	{
+		DEBUG_BREAK;
+		//m_raw = 0;
+		//m_raw |= ((uint8)baseRegFormat << 28);
+		//m_raw |= ((uint8)viewFormat << 24);
+		//m_raw |= (uint32)regId;
+	}
+
+	IMLReg(const IMLReg& other) : m_raw(other.m_raw) {}
+
+	IMLRegFormat GetBaseFormat() const
+	{
+		return (IMLRegFormat)((m_raw >> 28) & 0xF);
+	}
+
+	IMLRegFormat GetRegFormat() const
+	{
+		return (IMLRegFormat)((m_raw >> 24) & 0xF);
+	}
+
+	IMLRegID GetRegID() const
+	{
+		cemu_assert_debug(GetBaseFormat() != IMLRegFormat::INVALID_FORMAT);
+		cemu_assert_debug(GetRegFormat() != IMLRegFormat::INVALID_FORMAT);
+		return (IMLRegID)(m_raw & 0xFFFF);
+	}
+
+	void SetRegID(IMLRegID regId)
+	{
+		cemu_assert_debug(regId <= 0xFFFF);
+		m_raw &= ~0xFFFF;
+		m_raw |= (uint32)regId;
+	}
+
+	bool IsInvalid() const
+	{
+		return GetBaseFormat() == IMLRegFormat::INVALID_FORMAT;
+	}
+
+	bool IsValid() const
+	{
+		return GetBaseFormat() != IMLRegFormat::INVALID_FORMAT;
+	}
+
+	bool IsValidAndSameRegID(IMLRegID regId) const
+	{
+		return IsValid() && GetRegID() == regId;
+	}
+
+	// compare all fields
+	bool operator==(const IMLReg& other) const
+	{
+		return m_raw == other.m_raw;
+	}
+
+private:
+	uint32 m_raw;
+};
+
+static const IMLReg IMLREG_INVALID(IMLRegFormat::INVALID_FORMAT, IMLRegFormat::INVALID_FORMAT, 0, 0);
+static const IMLRegID IMLRegID_INVALID(0xFFFF);
+
+using IMLName = uint32;
+
+enum
+{
+	PPCREC_IML_OP_ASSIGN,			// '=' operator
+	PPCREC_IML_OP_ENDIAN_SWAP,		// '=' operator with 32bit endian swap
+	PPCREC_IML_OP_MULTIPLY_SIGNED,  // '*' operator (signed multiply)
+	PPCREC_IML_OP_MULTIPLY_HIGH_UNSIGNED, // unsigned 64bit multiply, store only high 32bit-word of result
+	PPCREC_IML_OP_MULTIPLY_HIGH_SIGNED, // signed 64bit multiply, store only high 32bit-word of result
+	PPCREC_IML_OP_DIVIDE_SIGNED,	// '/' operator (signed divide)
+	PPCREC_IML_OP_DIVIDE_UNSIGNED,	// '/' operator (unsigned divide)
+
+	// binary operation
+	PPCREC_IML_OP_OR,				// '|' operator
+	PPCREC_IML_OP_AND,				// '&' operator
+	PPCREC_IML_OP_XOR,				// '^' operator
+	PPCREC_IML_OP_LEFT_ROTATE,		// left rotate operator
+	PPCREC_IML_OP_LEFT_SHIFT,		// shift left operator
+	PPCREC_IML_OP_RIGHT_SHIFT_U,	// right shift operator (unsigned)
+	PPCREC_IML_OP_RIGHT_SHIFT_S,	// right shift operator (signed)
+	// ppc
+	PPCREC_IML_OP_SLW,				// SLW (shift based on register by up to 63 bits)
+	PPCREC_IML_OP_SRW,				// SRW (shift based on register by up to 63 bits)
+	PPCREC_IML_OP_CNTLZW,
+	// FPU
+	PPCREC_IML_OP_FPR_ADD_BOTTOM,
+	PPCREC_IML_OP_FPR_ADD_PAIR,
+	PPCREC_IML_OP_FPR_SUB_PAIR,
+	PPCREC_IML_OP_FPR_SUB_BOTTOM,
+	PPCREC_IML_OP_FPR_MULTIPLY_BOTTOM,
+	PPCREC_IML_OP_FPR_MULTIPLY_PAIR,
+	PPCREC_IML_OP_FPR_DIVIDE_BOTTOM,
+	PPCREC_IML_OP_FPR_DIVIDE_PAIR,
+	PPCREC_IML_OP_FPR_COPY_BOTTOM_TO_BOTTOM_AND_TOP,
+	PPCREC_IML_OP_FPR_COPY_TOP_TO_BOTTOM_AND_TOP,
+	PPCREC_IML_OP_FPR_COPY_BOTTOM_TO_BOTTOM,
+	PPCREC_IML_OP_FPR_COPY_BOTTOM_TO_TOP, // leave bottom of destination untouched
+	PPCREC_IML_OP_FPR_COPY_TOP_TO_TOP, // leave bottom of destination untouched
+	PPCREC_IML_OP_FPR_COPY_TOP_TO_BOTTOM, // leave top of destination untouched
+	PPCREC_IML_OP_FPR_COPY_BOTTOM_AND_TOP_SWAPPED,
+	PPCREC_IML_OP_FPR_EXPAND_BOTTOM32_TO_BOTTOM64_AND_TOP64, // expand bottom f32 to f64 in bottom and top half
+	PPCREC_IML_OP_FPR_FCMPO_BOTTOM, // deprecated
+	PPCREC_IML_OP_FPR_FCMPU_BOTTOM, // deprecated
+	PPCREC_IML_OP_FPR_FCMPU_TOP, // deprecated
+	PPCREC_IML_OP_FPR_NEGATE_BOTTOM,
+	PPCREC_IML_OP_FPR_NEGATE_PAIR,
+	PPCREC_IML_OP_FPR_ABS_BOTTOM, // abs(fp0)
+	PPCREC_IML_OP_FPR_ABS_PAIR,
+	PPCREC_IML_OP_FPR_FRES_PAIR, // 1.0/fp approx (Espresso accuracy)
+	PPCREC_IML_OP_FPR_FRSQRTE_PAIR, // 1.0/sqrt(fp) approx (Espresso accuracy)
+	PPCREC_IML_OP_FPR_NEGATIVE_ABS_BOTTOM, // -abs(fp0)
+	PPCREC_IML_OP_FPR_ROUND_TO_SINGLE_PRECISION_BOTTOM, // round 64bit double to 64bit double with 32bit float precision (in bottom half of xmm register)
+	PPCREC_IML_OP_FPR_ROUND_TO_SINGLE_PRECISION_PAIR, // round two 64bit doubles to 64bit double with 32bit float precision
+	PPCREC_IML_OP_FPR_BOTTOM_RECIPROCAL_SQRT,
+	PPCREC_IML_OP_FPR_BOTTOM_FCTIWZ,
+	PPCREC_IML_OP_FPR_SELECT_BOTTOM, // selectively copy bottom value from operand B or C based on value in operand A
+	PPCREC_IML_OP_FPR_SELECT_PAIR, // selectively copy top/bottom from operand B or C based on value in top/bottom of operand A
+	// PS
+	PPCREC_IML_OP_FPR_SUM0,
+	PPCREC_IML_OP_FPR_SUM1,
+
+
+	// R_R_R only
+
+	// R_R_S32 only
+
+	// R_R_R + R_R_S32
+	PPCREC_IML_OP_ADD, // also R_R_R_CARRY
+	PPCREC_IML_OP_SUB,
+
+	// R_R only
+	PPCREC_IML_OP_NOT,
+	PPCREC_IML_OP_NEG,
+	PPCREC_IML_OP_ASSIGN_S16_TO_S32,
+	PPCREC_IML_OP_ASSIGN_S8_TO_S32,
+
+	// R_R_R_carry
+	PPCREC_IML_OP_ADD_WITH_CARRY, // similar to ADD but also adds carry bit (0 or 1)
+
+	// X86 extension
+	PPCREC_IML_OP_X86_CMP, // R_R and R_S32
+
+	PPCREC_IML_OP_INVALID
+};
+
+#define PPCREC_IML_OP_FPR_COPY_PAIR (PPCREC_IML_OP_ASSIGN)
+
+enum
+{
+	PPCREC_IML_MACRO_B_TO_REG,		// branch to PPC address in register (used for BCCTR, BCLR)
+
+	PPCREC_IML_MACRO_BL,			// call to different function (can be within same function)
+	PPCREC_IML_MACRO_B_FAR,			// branch to different function
+	PPCREC_IML_MACRO_COUNT_CYCLES,	// decrease current remaining thread cycles by a certain amount
+	PPCREC_IML_MACRO_HLE,			// HLE function call
+	PPCREC_IML_MACRO_LEAVE,			// leaves recompiler and switches to interpeter
+	// debugging
+	PPCREC_IML_MACRO_DEBUGBREAK,	// throws a debugbreak
+};
+
+enum class IMLCondition : uint8
+{
+	EQ,
+	NEQ,
+	SIGNED_GT,
+	SIGNED_LT,
+	UNSIGNED_GT,
+	UNSIGNED_LT,
+
+	// floating point conditions
+	UNORDERED_GT, // a > b, false if either is NaN
+	UNORDERED_LT, // a < b, false if either is NaN
+	UNORDERED_EQ, // a == b, false if either is NaN
+	UNORDERED_U, // unordered (true if either operand is NaN)
+
+	ORDERED_GT,
+	ORDERED_LT,
+	ORDERED_EQ,
+	ORDERED_U
+};
+
+enum
+{
+	PPCREC_IML_TYPE_NONE,
+	PPCREC_IML_TYPE_NO_OP,				// no-op instruction
+	PPCREC_IML_TYPE_R_R,				// r* = (op) *r			(can also be r* (op) *r) 
+	PPCREC_IML_TYPE_R_R_R,				// r* = r* (op) r*
+	PPCREC_IML_TYPE_R_R_R_CARRY,		// r* = r* (op) r*		(reads and/or updates carry)
+	PPCREC_IML_TYPE_R_R_S32,			// r* = r* (op) s32*
+	PPCREC_IML_TYPE_R_R_S32_CARRY,		// r* = r* (op) s32*	(reads and/or updates carry)
+	PPCREC_IML_TYPE_LOAD,				// r* = [r*+s32*]
+	PPCREC_IML_TYPE_LOAD_INDEXED,		// r* = [r*+r*]
+	PPCREC_IML_TYPE_STORE,				// [r*+s32*] = r*
+	PPCREC_IML_TYPE_STORE_INDEXED,		// [r*+r*] = r*
+	PPCREC_IML_TYPE_R_NAME,				// r* = name
+	PPCREC_IML_TYPE_NAME_R,				// name* = r*
+	PPCREC_IML_TYPE_R_S32,				// r* (op) imm
+	PPCREC_IML_TYPE_MACRO,
+	PPCREC_IML_TYPE_CJUMP_CYCLE_CHECK,	// jumps only if remaining thread cycles < 0
+
+	// conditions and branches
+	PPCREC_IML_TYPE_COMPARE,			// r* = r* CMP[cond] r*
+	PPCREC_IML_TYPE_COMPARE_S32,		// r* = r* CMP[cond] imm
+	PPCREC_IML_TYPE_JUMP,				// jump always
+	PPCREC_IML_TYPE_CONDITIONAL_JUMP,	// jump conditionally based on boolean value in register
+
+	// atomic
+	PPCREC_IML_TYPE_ATOMIC_CMP_STORE,
+
+	// function call
+	PPCREC_IML_TYPE_CALL_IMM,			// call to fixed immediate address
+
+	// FPR
+	PPCREC_IML_TYPE_FPR_LOAD,			// r* = (bitdepth) [r*+s32*] (single or paired single mode)
+	PPCREC_IML_TYPE_FPR_LOAD_INDEXED,	// r* = (bitdepth) [r*+r*] (single or paired single mode)
+	PPCREC_IML_TYPE_FPR_STORE,			// (bitdepth) [r*+s32*] = r* (single or paired single mode)
+	PPCREC_IML_TYPE_FPR_STORE_INDEXED,	// (bitdepth) [r*+r*] = r* (single or paired single mode)
+	PPCREC_IML_TYPE_FPR_R_R,
+	PPCREC_IML_TYPE_FPR_R_R_R,
+	PPCREC_IML_TYPE_FPR_R_R_R_R,
+	PPCREC_IML_TYPE_FPR_R,
+
+	PPCREC_IML_TYPE_FPR_COMPARE,		// r* = r* CMP[cond] r*
+
+	// X86 specific
+	PPCREC_IML_TYPE_X86_EFLAGS_JCC,
+};
+
+enum // IMLName
+{
+	PPCREC_NAME_NONE,
+	PPCREC_NAME_TEMPORARY = 1000,
+	PPCREC_NAME_R0 = 2000,
+	PPCREC_NAME_SPR0 = 3000,
+	PPCREC_NAME_FPR0 = 4000,
+	PPCREC_NAME_TEMPORARY_FPR0 = 5000, // 0 to 7
+	PPCREC_NAME_XER_CA = 6000, // carry bit from XER
+	PPCREC_NAME_XER_OV = 6001, // overflow bit from XER
+	PPCREC_NAME_XER_SO = 6002, // summary overflow bit from XER
+	PPCREC_NAME_CR = 7000, // CR register bits (31 to 0)
+	PPCREC_NAME_CR_LAST = PPCREC_NAME_CR+31,
+	PPCREC_NAME_CPU_MEMRES_EA = 8000,
+	PPCREC_NAME_CPU_MEMRES_VAL = 8001
+};
+
+#define PPC_REC_INVALID_REGISTER	0xFF	// deprecated. Use IMLREG_INVALID instead
+
+enum
+{
+	// fpr load
+	PPCREC_FPR_LD_MODE_SINGLE_INTO_PS0,
+	PPCREC_FPR_LD_MODE_SINGLE_INTO_PS0_PS1,
+	PPCREC_FPR_LD_MODE_DOUBLE_INTO_PS0,
+	PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0,
+	PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0_PS1,
+	PPCREC_FPR_LD_MODE_PSQ_FLOAT_PS0,
+	PPCREC_FPR_LD_MODE_PSQ_FLOAT_PS0_PS1,
+	PPCREC_FPR_LD_MODE_PSQ_S16_PS0,
+	PPCREC_FPR_LD_MODE_PSQ_S16_PS0_PS1,
+	PPCREC_FPR_LD_MODE_PSQ_U16_PS0,
+	PPCREC_FPR_LD_MODE_PSQ_U16_PS0_PS1,
+	PPCREC_FPR_LD_MODE_PSQ_S8_PS0,
+	PPCREC_FPR_LD_MODE_PSQ_S8_PS0_PS1,
+	PPCREC_FPR_LD_MODE_PSQ_U8_PS0,
+	PPCREC_FPR_LD_MODE_PSQ_U8_PS0_PS1,
+	// fpr store
+	PPCREC_FPR_ST_MODE_SINGLE_FROM_PS0, // store 1 single precision float from ps0
+	PPCREC_FPR_ST_MODE_DOUBLE_FROM_PS0, // store 1 double precision float from ps0
+
+	PPCREC_FPR_ST_MODE_UI32_FROM_PS0, // store raw low-32bit of PS0
+
+	PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0_PS1,
+	PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0,
+	PPCREC_FPR_ST_MODE_PSQ_FLOAT_PS0_PS1,
+	PPCREC_FPR_ST_MODE_PSQ_FLOAT_PS0,
+	PPCREC_FPR_ST_MODE_PSQ_S8_PS0,
+	PPCREC_FPR_ST_MODE_PSQ_S8_PS0_PS1,
+	PPCREC_FPR_ST_MODE_PSQ_U8_PS0,
+	PPCREC_FPR_ST_MODE_PSQ_U8_PS0_PS1,
+	PPCREC_FPR_ST_MODE_PSQ_U16_PS0,
+	PPCREC_FPR_ST_MODE_PSQ_U16_PS0_PS1,
+	PPCREC_FPR_ST_MODE_PSQ_S16_PS0,
+	PPCREC_FPR_ST_MODE_PSQ_S16_PS0_PS1,
+};
+
+struct IMLUsedRegisters
+{
+	IMLUsedRegisters() {};
+
+	bool IsWrittenByRegId(IMLRegID regId) const
+	{
+		if (writtenGPR1.IsValid() && writtenGPR1.GetRegID() == regId)
+			return true;
+		if (writtenGPR2.IsValid() && writtenGPR2.GetRegID() == regId)
+			return true;
+		return false;
+	}
+
+	bool IsBaseGPRWritten(IMLReg imlReg) const
+	{
+		cemu_assert_debug(imlReg.IsValid());
+		auto regId = imlReg.GetRegID();
+		return IsWrittenByRegId(regId);
+	}
+
+	template<typename Fn>
+	void ForEachWrittenGPR(Fn F) const
+	{
+		if (writtenGPR1.IsValid())
+			F(writtenGPR1);
+		if (writtenGPR2.IsValid())
+			F(writtenGPR2);
+	}
+
+	template<typename Fn>
+	void ForEachReadGPR(Fn F) const
+	{
+		if (readGPR1.IsValid())
+			F(readGPR1);
+		if (readGPR2.IsValid())
+			F(readGPR2);
+		if (readGPR3.IsValid())
+			F(readGPR3);
+		if (readGPR4.IsValid())
+			F(readGPR4);
+	}
+
+	template<typename Fn>
+	void ForEachAccessedGPR(Fn F) const
+	{
+		// GPRs
+		if (readGPR1.IsValid())
+			F(readGPR1, false);
+		if (readGPR2.IsValid())
+			F(readGPR2, false);
+		if (readGPR3.IsValid())
+			F(readGPR3, false);
+		if (readGPR4.IsValid())
+			F(readGPR4, false);
+		if (writtenGPR1.IsValid())
+			F(writtenGPR1, true);
+		if (writtenGPR2.IsValid())
+			F(writtenGPR2, true);
+	}
+
+	IMLReg readGPR1;
+	IMLReg readGPR2;
+	IMLReg readGPR3;
+	IMLReg readGPR4;
+	IMLReg writtenGPR1;
+	IMLReg writtenGPR2;
+};
+
+struct IMLInstruction
+{
+	IMLInstruction() {}
+	IMLInstruction(const IMLInstruction& other) 
+	{
+		memcpy(this, &other, sizeof(IMLInstruction));
+	}
+
+	uint8 type;
+	uint8 operation;
+	union
+	{
+		struct
+		{
+			uint8 _padding[7];
+		}padding;
+		struct
+		{
+			IMLReg regR;
+			IMLReg regA;
+		}op_r_r;
+		struct
+		{
+			IMLReg regR;
+			IMLReg regA;
+			IMLReg regB;
+		}op_r_r_r;
+		struct
+		{
+			IMLReg regR;
+			IMLReg regA;
+			IMLReg regB;
+			IMLReg regCarry;
+		}op_r_r_r_carry;
+		struct
+		{
+			IMLReg regR;
+			IMLReg regA;
+			sint32 immS32;
+		}op_r_r_s32;
+		struct
+		{
+			IMLReg regR;
+			IMLReg regA;
+			IMLReg regCarry;
+			sint32 immS32;
+		}op_r_r_s32_carry;
+		struct
+		{
+			IMLReg regR;
+			IMLName name;
+		}op_r_name; // alias op_name_r
+		struct
+		{
+			IMLReg regR;
+			sint32 immS32;
+		}op_r_immS32;
+		struct
+		{
+			uint32 param;
+			uint32 param2;
+			uint16 paramU16;
+			IMLReg paramReg;
+		}op_macro;
+		struct
+		{
+			IMLReg registerData;
+			IMLReg registerMem;
+			IMLReg registerMem2;
+			IMLReg registerGQR;
+			uint8 copyWidth;
+			struct
+			{
+				bool swapEndian : 1;
+				bool signExtend : 1;
+				bool notExpanded : 1; // for floats
+			}flags2;
+			uint8 mode; // transfer mode (copy width, ps0/ps1 behavior)
+			sint32 immS32;
+		}op_storeLoad;
+		struct
+		{
+			uintptr_t callAddress;
+			IMLReg regParam0;
+			IMLReg regParam1;
+			IMLReg regParam2;
+			IMLReg regReturn;
+		}op_call_imm;
+		struct
+		{
+			IMLReg regR;
+			IMLReg regA;
+		}op_fpr_r_r;
+		struct
+		{
+			IMLReg regR;
+			IMLReg regA;
+			IMLReg regB;
+		}op_fpr_r_r_r;
+		struct
+		{
+			IMLReg regR;
+			IMLReg regA;
+			IMLReg regB;
+			IMLReg regC;
+		}op_fpr_r_r_r_r;
+		struct
+		{
+			IMLReg regR;
+		}op_fpr_r;
+		struct
+		{
+			IMLReg regR; // stores the boolean result of the comparison
+			IMLReg regA;
+			IMLReg regB;
+			IMLCondition cond;
+		}op_fpr_compare;
+		struct
+		{
+			IMLReg regR; // stores the boolean result of the comparison
+			IMLReg regA;
+			IMLReg regB;
+			IMLCondition cond;
+		}op_compare;
+		struct
+		{
+			IMLReg regR; // stores the boolean result of the comparison
+			IMLReg regA;
+			sint32 immS32;
+			IMLCondition cond;
+		}op_compare_s32;
+		struct
+		{
+			IMLReg registerBool;
+			bool mustBeTrue;
+		}op_conditional_jump;
+		struct  
+		{
+			IMLReg regEA;
+			IMLReg regCompareValue;
+			IMLReg regWriteValue;
+			IMLReg regBoolOut;
+		}op_atomic_compare_store;
+		// conditional operations (emitted if supported by target platform)
+		struct
+		{
+			// r_s32
+			IMLReg regR;
+			sint32 immS32;
+			// condition
+			uint8 crRegisterIndex;
+			uint8 crBitIndex;
+			bool  bitMustBeSet;
+		}op_conditional_r_s32;
+		// X86 specific
+		struct
+		{
+			IMLCondition cond;
+			bool invertedCondition;
+		}op_x86_eflags_jcc;
+	};
+
+	bool IsSuffixInstruction() const
+	{
+		if (type == PPCREC_IML_TYPE_MACRO && operation == PPCREC_IML_MACRO_BL ||
+			type == PPCREC_IML_TYPE_MACRO && operation == PPCREC_IML_MACRO_B_FAR ||
+			type == PPCREC_IML_TYPE_MACRO && operation == PPCREC_IML_MACRO_B_TO_REG ||
+			type == PPCREC_IML_TYPE_MACRO && operation == PPCREC_IML_MACRO_LEAVE ||
+			type == PPCREC_IML_TYPE_MACRO && operation == PPCREC_IML_MACRO_HLE ||
+			type == PPCREC_IML_TYPE_CJUMP_CYCLE_CHECK ||
+			type == PPCREC_IML_TYPE_JUMP ||
+			type == PPCREC_IML_TYPE_CONDITIONAL_JUMP ||
+			type == PPCREC_IML_TYPE_X86_EFLAGS_JCC)
+			return true;
+		return false;
+	}
+
+	// instruction setters
+	void make_no_op()
+	{
+		type = PPCREC_IML_TYPE_NO_OP;
+		operation = 0;
+	}
+
+	void make_r_name(IMLReg regR, IMLName name)
+	{
+		cemu_assert_debug(regR.GetBaseFormat() == regR.GetRegFormat()); // for name load/store instructions the register must match the base format
+		type = PPCREC_IML_TYPE_R_NAME;
+		operation = PPCREC_IML_OP_ASSIGN;
+		op_r_name.regR = regR;
+		op_r_name.name = name;
+	}
+
+	void make_name_r(IMLName name, IMLReg regR)
+	{
+		cemu_assert_debug(regR.GetBaseFormat() == regR.GetRegFormat()); // for name load/store instructions the register must match the base format
+		type = PPCREC_IML_TYPE_NAME_R;
+		operation = PPCREC_IML_OP_ASSIGN;
+		op_r_name.regR = regR;
+		op_r_name.name = name;
+	}
+
+	void make_debugbreak(uint32 currentPPCAddress = 0)
+	{
+		make_macro(PPCREC_IML_MACRO_DEBUGBREAK, 0, currentPPCAddress, 0, IMLREG_INVALID);
+	}
+
+	void make_macro(uint32 macroId, uint32 param, uint32 param2, uint16 paramU16, IMLReg regParam)
+	{
+		this->type = PPCREC_IML_TYPE_MACRO;
+		this->operation = macroId;
+		this->op_macro.param = param;
+		this->op_macro.param2 = param2;
+		this->op_macro.paramU16 = paramU16;
+		this->op_macro.paramReg = regParam;
+	}
+
+	void make_cjump_cycle_check()
+	{
+		this->type = PPCREC_IML_TYPE_CJUMP_CYCLE_CHECK;
+		this->operation = 0;
+	}
+
+	void make_r_r(uint32 operation, IMLReg regR, IMLReg regA)
+	{
+		this->type = PPCREC_IML_TYPE_R_R;
+		this->operation = operation;
+		this->op_r_r.regR = regR;
+		this->op_r_r.regA = regA;
+	}
+
+	void make_r_s32(uint32 operation, IMLReg regR, sint32 immS32)
+	{
+		this->type = PPCREC_IML_TYPE_R_S32;
+		this->operation = operation;
+		this->op_r_immS32.regR = regR;
+		this->op_r_immS32.immS32 = immS32;
+	}
+
+	void make_r_r_r(uint32 operation, IMLReg regR, IMLReg regA, IMLReg regB)
+	{
+		this->type = PPCREC_IML_TYPE_R_R_R;
+		this->operation = operation;
+		this->op_r_r_r.regR = regR;
+		this->op_r_r_r.regA = regA;
+		this->op_r_r_r.regB = regB;
+	}
+
+	void make_r_r_r_carry(uint32 operation, IMLReg regR, IMLReg regA, IMLReg regB, IMLReg regCarry)
+	{
+		this->type = PPCREC_IML_TYPE_R_R_R_CARRY;
+		this->operation = operation;
+		this->op_r_r_r_carry.regR = regR;
+		this->op_r_r_r_carry.regA = regA;
+		this->op_r_r_r_carry.regB = regB;
+		this->op_r_r_r_carry.regCarry = regCarry;
+	}
+
+	void make_r_r_s32(uint32 operation, IMLReg regR, IMLReg regA, sint32 immS32)
+	{
+		this->type = PPCREC_IML_TYPE_R_R_S32;
+		this->operation = operation;
+		this->op_r_r_s32.regR = regR;
+		this->op_r_r_s32.regA = regA;
+		this->op_r_r_s32.immS32 = immS32;
+	}
+
+	void make_r_r_s32_carry(uint32 operation, IMLReg regR, IMLReg regA, sint32 immS32, IMLReg regCarry)
+	{
+		this->type = PPCREC_IML_TYPE_R_R_S32_CARRY;
+		this->operation = operation;
+		this->op_r_r_s32_carry.regR = regR;
+		this->op_r_r_s32_carry.regA = regA;
+		this->op_r_r_s32_carry.immS32 = immS32;
+		this->op_r_r_s32_carry.regCarry = regCarry;
+	}
+
+	void make_compare(IMLReg regA, IMLReg regB, IMLReg regR, IMLCondition cond)
+	{
+		this->type = PPCREC_IML_TYPE_COMPARE;
+		this->operation = PPCREC_IML_OP_INVALID;
+		this->op_compare.regR = regR;
+		this->op_compare.regA = regA;
+		this->op_compare.regB = regB;
+		this->op_compare.cond = cond;
+	}
+
+	void make_compare_s32(IMLReg regA, sint32 immS32, IMLReg regR, IMLCondition cond)
+	{
+		this->type = PPCREC_IML_TYPE_COMPARE_S32;
+		this->operation = PPCREC_IML_OP_INVALID;
+		this->op_compare_s32.regR = regR;
+		this->op_compare_s32.regA = regA;
+		this->op_compare_s32.immS32 = immS32;
+		this->op_compare_s32.cond = cond;
+	}
+
+	void make_conditional_jump(IMLReg regBool, bool mustBeTrue)
+	{
+		this->type = PPCREC_IML_TYPE_CONDITIONAL_JUMP;
+		this->operation = PPCREC_IML_OP_INVALID;
+		this->op_conditional_jump.registerBool = regBool;
+		this->op_conditional_jump.mustBeTrue = mustBeTrue;
+	}
+
+	void make_jump()
+	{
+		this->type = PPCREC_IML_TYPE_JUMP;
+		this->operation = PPCREC_IML_OP_INVALID;
+	}
+
+	// load from memory
+	void make_r_memory(IMLReg regD, IMLReg regMem, sint32 immS32, uint32 copyWidth, bool signExtend, bool switchEndian)
+	{
+		this->type = PPCREC_IML_TYPE_LOAD;
+		this->operation = 0;
+		this->op_storeLoad.registerData = regD;
+		this->op_storeLoad.registerMem = regMem;
+		this->op_storeLoad.immS32 = immS32;
+		this->op_storeLoad.copyWidth = copyWidth;
+		this->op_storeLoad.flags2.swapEndian = switchEndian;
+		this->op_storeLoad.flags2.signExtend = signExtend;
+	}
+
+	// store to memory
+	void make_memory_r(IMLReg regS, IMLReg regMem, sint32 immS32, uint32 copyWidth, bool switchEndian)
+	{
+		this->type = PPCREC_IML_TYPE_STORE;
+		this->operation = 0;
+		this->op_storeLoad.registerData = regS;
+		this->op_storeLoad.registerMem = regMem;
+		this->op_storeLoad.immS32 = immS32;
+		this->op_storeLoad.copyWidth = copyWidth;
+		this->op_storeLoad.flags2.swapEndian = switchEndian;
+		this->op_storeLoad.flags2.signExtend = false;
+	}
+
+	void make_atomic_cmp_store(IMLReg regEA, IMLReg regCompareValue, IMLReg regWriteValue, IMLReg regSuccessOutput)
+	{
+		this->type = PPCREC_IML_TYPE_ATOMIC_CMP_STORE;
+		this->operation = 0;
+		this->op_atomic_compare_store.regEA = regEA;
+		this->op_atomic_compare_store.regCompareValue = regCompareValue;
+		this->op_atomic_compare_store.regWriteValue = regWriteValue;
+		this->op_atomic_compare_store.regBoolOut = regSuccessOutput;
+	}
+
+	void make_call_imm(uintptr_t callAddress, IMLReg param0, IMLReg param1, IMLReg param2, IMLReg regReturn)
+	{
+		this->type = PPCREC_IML_TYPE_CALL_IMM;
+		this->operation = 0;
+		this->op_call_imm.callAddress = callAddress;
+		this->op_call_imm.regParam0 = param0;
+		this->op_call_imm.regParam1 = param1;
+		this->op_call_imm.regParam2 = param2;
+		this->op_call_imm.regReturn = regReturn;
+	}
+
+	void make_fpr_compare(IMLReg regA, IMLReg regB, IMLReg regR, IMLCondition cond)
+	{
+		this->type = PPCREC_IML_TYPE_FPR_COMPARE;
+		this->operation = -999;
+		this->op_fpr_compare.regR = regR;
+		this->op_fpr_compare.regA = regA;
+		this->op_fpr_compare.regB = regB;
+		this->op_fpr_compare.cond = cond;
+	}
+
+	/* X86 specific */
+	void make_x86_eflags_jcc(IMLCondition cond, bool invertedCondition)
+	{
+		this->type = PPCREC_IML_TYPE_X86_EFLAGS_JCC;
+		this->operation = -999;
+		this->op_x86_eflags_jcc.cond = cond;
+		this->op_x86_eflags_jcc.invertedCondition = invertedCondition;
+	}
+
+	void CheckRegisterUsage(IMLUsedRegisters* registersUsed) const;
+	bool HasSideEffects() const; // returns true if the instruction has side effects beyond just reading and writing registers. Dead code elimination uses this to know if an instruction can be dropped when the regular register outputs are not used
+
+	void RewriteGPR(const std::unordered_map<IMLRegID, IMLRegID>& translationTable);
+};
+
+// architecture specific constants
+namespace IMLArchX86
+{
+	static constexpr int PHYSREG_GPR_BASE = 0;
+	static constexpr int PHYSREG_FPR_BASE = 16;
+};

src/Cafe/HW/Espresso/Recompiler/IML/IMLOptimizer.cpp

@@ -0,0 +1,794 @@
+#include "Cafe/HW/Espresso/Interpreter/PPCInterpreterInternal.h"
+#include "Cafe/HW/Espresso/Recompiler/IML/IML.h"
+#include "Cafe/HW/Espresso/Recompiler/IML/IMLInstruction.h"
+
+#include "../PPCRecompiler.h"
+#include "../PPCRecompilerIml.h"
+#include "../BackendX64/BackendX64.h"
+
+#include "Common/FileStream.h"
+
+#include <boost/container/static_vector.hpp>
+#include <boost/container/small_vector.hpp>
+
+IMLReg _FPRRegFromID(IMLRegID regId)
+{
+	return IMLReg(IMLRegFormat::F64, IMLRegFormat::F64, 0, regId);
+}
+
+void PPCRecompiler_optimizeDirectFloatCopiesScanForward(ppcImlGenContext_t* ppcImlGenContext, IMLSegment* imlSegment, sint32 imlIndexLoad, IMLReg fprReg)
+{
+	IMLRegID fprIndex = fprReg.GetRegID();
+
+	IMLInstruction* imlInstructionLoad = imlSegment->imlList.data() + imlIndexLoad;
+	if (imlInstructionLoad->op_storeLoad.flags2.notExpanded)
+		return;
+
+	IMLUsedRegisters registersUsed;
+	sint32 scanRangeEnd = std::min<sint32>(imlIndexLoad + 25, imlSegment->imlList.size()); // don't scan too far (saves performance and also the chances we can merge the load+store become low at high distances)
+	bool foundMatch = false;
+	sint32 lastStore = -1;
+	for (sint32 i = imlIndexLoad + 1; i < scanRangeEnd; i++)
+	{
+		IMLInstruction* imlInstruction = imlSegment->imlList.data() + i;
+		if (imlInstruction->IsSuffixInstruction())
+			break;
+		// check if FPR is stored
+		if ((imlInstruction->type == PPCREC_IML_TYPE_FPR_STORE && imlInstruction->op_storeLoad.mode == PPCREC_FPR_ST_MODE_SINGLE_FROM_PS0) ||
+			(imlInstruction->type == PPCREC_IML_TYPE_FPR_STORE_INDEXED && imlInstruction->op_storeLoad.mode == PPCREC_FPR_ST_MODE_SINGLE_FROM_PS0))
+		{
+			if (imlInstruction->op_storeLoad.registerData.GetRegID() == fprIndex)
+			{
+				if (foundMatch == false)
+				{
+					// flag the load-single instruction as "don't expand" (leave single value as-is)
+					imlInstructionLoad->op_storeLoad.flags2.notExpanded = true;
+				}
+				// also set the flag for the store instruction
+				IMLInstruction* imlInstructionStore = imlInstruction;
+				imlInstructionStore->op_storeLoad.flags2.notExpanded = true;
+
+				foundMatch = true;
+				lastStore = i + 1;
+
+				continue;
+			}
+		}
+
+		// check if FPR is overwritten (we can actually ignore read operations?)
+		imlInstruction->CheckRegisterUsage(&registersUsed);
+		if (registersUsed.writtenGPR1.IsValidAndSameRegID(fprIndex) || registersUsed.writtenGPR2.IsValidAndSameRegID(fprIndex))
+			break;
+		if (registersUsed.readGPR1.IsValidAndSameRegID(fprIndex))
+			break;
+		if (registersUsed.readGPR2.IsValidAndSameRegID(fprIndex))
+			break;
+		if (registersUsed.readGPR3.IsValidAndSameRegID(fprIndex))
+			break;
+		if (registersUsed.readGPR4.IsValidAndSameRegID(fprIndex))
+			break;
+	}
+
+	if (foundMatch)
+	{
+		// insert expand instruction after store
+		IMLInstruction* newExpand = PPCRecompiler_insertInstruction(imlSegment, lastStore);
+		PPCRecompilerImlGen_generateNewInstruction_fpr_r(ppcImlGenContext, newExpand, PPCREC_IML_OP_FPR_EXPAND_BOTTOM32_TO_BOTTOM64_AND_TOP64, _FPRRegFromID(fprIndex));
+	}
+}
+
+/*
+* Scans for patterns:
+* <Load sp float into register f>
+* <Random unrelated instructions>
+* <Store sp float from register f>
+* For these patterns the store and load is modified to work with un-extended values (float remains as float, no double conversion)
+* The float->double extension is then executed later
+* Advantages:
+* Keeps denormals and other special float values intact
+* Slightly improves performance
+*/
+void IMLOptimizer_OptimizeDirectFloatCopies(ppcImlGenContext_t* ppcImlGenContext)
+{
+	for (IMLSegment* segIt : ppcImlGenContext->segmentList2)
+	{
+		for (sint32 i = 0; i < segIt->imlList.size(); i++)
+		{
+			IMLInstruction* imlInstruction = segIt->imlList.data() + i;
+			if (imlInstruction->type == PPCREC_IML_TYPE_FPR_LOAD && imlInstruction->op_storeLoad.mode == PPCREC_FPR_LD_MODE_SINGLE_INTO_PS0_PS1)
+			{
+				PPCRecompiler_optimizeDirectFloatCopiesScanForward(ppcImlGenContext, segIt, i, imlInstruction->op_storeLoad.registerData);
+			}
+			else if (imlInstruction->type == PPCREC_IML_TYPE_FPR_LOAD_INDEXED && imlInstruction->op_storeLoad.mode == PPCREC_FPR_LD_MODE_SINGLE_INTO_PS0_PS1)
+			{
+				PPCRecompiler_optimizeDirectFloatCopiesScanForward(ppcImlGenContext, segIt, i, imlInstruction->op_storeLoad.registerData);
+			}
+		}
+	}
+}
+
+void PPCRecompiler_optimizeDirectIntegerCopiesScanForward(ppcImlGenContext_t* ppcImlGenContext, IMLSegment* imlSegment, sint32 imlIndexLoad, IMLReg gprReg)
+{
+	cemu_assert_debug(gprReg.GetBaseFormat() == IMLRegFormat::I64); // todo - proper handling required for non-standard sizes
+	cemu_assert_debug(gprReg.GetRegFormat() == IMLRegFormat::I32);
+
+	IMLRegID gprIndex = gprReg.GetRegID();
+	IMLInstruction* imlInstructionLoad = imlSegment->imlList.data() + imlIndexLoad;
+	if ( imlInstructionLoad->op_storeLoad.flags2.swapEndian == false )
+		return;
+	bool foundMatch = false;
+	IMLUsedRegisters registersUsed;
+	sint32 scanRangeEnd = std::min<sint32>(imlIndexLoad + 25, imlSegment->imlList.size()); // don't scan too far (saves performance and also the chances we can merge the load+store become low at high distances)
+	sint32 i = imlIndexLoad + 1;
+	for (; i < scanRangeEnd; i++)
+	{
+		IMLInstruction* imlInstruction = imlSegment->imlList.data() + i;
+		if (imlInstruction->IsSuffixInstruction())
+			break;
+		// check if GPR is stored
+		if ((imlInstruction->type == PPCREC_IML_TYPE_STORE && imlInstruction->op_storeLoad.copyWidth == 32 ) )
+		{
+			if (imlInstruction->op_storeLoad.registerMem.GetRegID() == gprIndex)
+				break;
+			if (imlInstruction->op_storeLoad.registerData.GetRegID() == gprIndex)
+			{
+				IMLInstruction* imlInstructionStore = imlInstruction;
+				if (foundMatch == false)
+				{
+					// switch the endian swap flag for the load instruction
+					imlInstructionLoad->op_storeLoad.flags2.swapEndian = !imlInstructionLoad->op_storeLoad.flags2.swapEndian;
+					foundMatch = true;
+				}
+				// switch the endian swap flag for the store instruction
+				imlInstructionStore->op_storeLoad.flags2.swapEndian = !imlInstructionStore->op_storeLoad.flags2.swapEndian;
+				// keep scanning
+				continue;
+			}
+		}
+		// check if GPR is accessed
+		imlInstruction->CheckRegisterUsage(&registersUsed);
+		if (registersUsed.readGPR1.IsValidAndSameRegID(gprIndex) ||
+			registersUsed.readGPR2.IsValidAndSameRegID(gprIndex) ||
+			registersUsed.readGPR3.IsValidAndSameRegID(gprIndex))
+		{
+			break;
+		}
+		if (registersUsed.IsBaseGPRWritten(gprReg))
+			return; // GPR overwritten, we don't need to byte swap anymore
+	}
+	if (foundMatch)
+	{
+		PPCRecompiler_insertInstruction(imlSegment, i)->make_r_r(PPCREC_IML_OP_ENDIAN_SWAP, gprReg, gprReg);
+	}
+}
+
+/*
+* Scans for patterns:
+* <Load sp integer into register r>
+* <Random unrelated instructions>
+* <Store sp integer from register r>
+* For these patterns the store and load is modified to work with non-swapped values
+* The big_endian->little_endian conversion is then executed later
+* Advantages:
+* Slightly improves performance
+*/
+void IMLOptimizer_OptimizeDirectIntegerCopies(ppcImlGenContext_t* ppcImlGenContext)
+{
+	for (IMLSegment* segIt : ppcImlGenContext->segmentList2)
+	{
+		for (sint32 i = 0; i < segIt->imlList.size(); i++)
+		{
+			IMLInstruction* imlInstruction = segIt->imlList.data() + i;
+			if (imlInstruction->type == PPCREC_IML_TYPE_LOAD && imlInstruction->op_storeLoad.copyWidth == 32 && imlInstruction->op_storeLoad.flags2.swapEndian )
+			{
+				PPCRecompiler_optimizeDirectIntegerCopiesScanForward(ppcImlGenContext, segIt, i, imlInstruction->op_storeLoad.registerData);
+			}
+		}
+	}
+}
+
+IMLName PPCRecompilerImlGen_GetRegName(ppcImlGenContext_t* ppcImlGenContext, IMLReg reg);
+
+sint32 _getGQRIndexFromRegister(ppcImlGenContext_t* ppcImlGenContext, IMLReg gqrReg)
+{
+	if (gqrReg.IsInvalid())
+		return -1;
+	sint32 namedReg = PPCRecompilerImlGen_GetRegName(ppcImlGenContext, gqrReg);
+	if (namedReg >= (PPCREC_NAME_SPR0 + SPR_UGQR0) && namedReg <= (PPCREC_NAME_SPR0 + SPR_UGQR7))
+	{
+		return namedReg - (PPCREC_NAME_SPR0 + SPR_UGQR0);
+	}
+	else
+	{
+		cemu_assert_suspicious();
+	}
+	return -1;
+}
+
+bool PPCRecompiler_isUGQRValueKnown(ppcImlGenContext_t* ppcImlGenContext, sint32 gqrIndex, uint32& gqrValue)
+{
+	// UGQR 2 to 7 are initialized by the OS and we assume that games won't ever permanently touch those
+	// todo - hack - replace with more accurate solution
+	if (gqrIndex == 2)
+		gqrValue = 0x00040004;
+	else if (gqrIndex == 3)
+		gqrValue = 0x00050005;
+	else if (gqrIndex == 4)
+		gqrValue = 0x00060006;
+	else if (gqrIndex == 5)
+		gqrValue = 0x00070007;
+	else
+		return false;
+	return true;
+}
+
+/*
+ * If value of GQR can be predicted for a given PSQ load or store instruction then replace it with an optimized version
+ */
+void PPCRecompiler_optimizePSQLoadAndStore(ppcImlGenContext_t* ppcImlGenContext)
+{
+	for (IMLSegment* segIt : ppcImlGenContext->segmentList2) 
+	{
+		for(IMLInstruction& instIt : segIt->imlList)
+		{
+			if (instIt.type == PPCREC_IML_TYPE_FPR_LOAD || instIt.type == PPCREC_IML_TYPE_FPR_LOAD_INDEXED)
+			{
+				if(instIt.op_storeLoad.mode != PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0 &&
+					instIt.op_storeLoad.mode != PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0_PS1 )
+					continue;
+				// get GQR value
+				cemu_assert_debug(instIt.op_storeLoad.registerGQR.IsValid());
+				sint32 gqrIndex = _getGQRIndexFromRegister(ppcImlGenContext, instIt.op_storeLoad.registerGQR);
+				cemu_assert(gqrIndex >= 0);
+				if (ppcImlGenContext->tracking.modifiesGQR[gqrIndex])
+					continue;
+				uint32 gqrValue;
+				if (!PPCRecompiler_isUGQRValueKnown(ppcImlGenContext, gqrIndex, gqrValue))
+					continue;
+
+				uint32 formatType = (gqrValue >> 16) & 7;
+				uint32 scale = (gqrValue >> 24) & 0x3F;
+				if (scale != 0)
+					continue; // only generic handler supports scale
+				if (instIt.op_storeLoad.mode == PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0)
+				{
+					if (formatType == 0)
+						instIt.op_storeLoad.mode = PPCREC_FPR_LD_MODE_PSQ_FLOAT_PS0;
+					else if (formatType == 4)
+						instIt.op_storeLoad.mode = PPCREC_FPR_LD_MODE_PSQ_U8_PS0;
+					else if (formatType == 5)
+						instIt.op_storeLoad.mode = PPCREC_FPR_LD_MODE_PSQ_U16_PS0;
+					else if (formatType == 6)
+						instIt.op_storeLoad.mode = PPCREC_FPR_LD_MODE_PSQ_S8_PS0;
+					else if (formatType == 7)
+						instIt.op_storeLoad.mode = PPCREC_FPR_LD_MODE_PSQ_S16_PS0;
+					if (instIt.op_storeLoad.mode != PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0)
+						instIt.op_storeLoad.registerGQR = IMLREG_INVALID;
+				}
+				else if (instIt.op_storeLoad.mode == PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0_PS1)
+				{
+					if (formatType == 0)
+						instIt.op_storeLoad.mode = PPCREC_FPR_LD_MODE_PSQ_FLOAT_PS0_PS1;
+					else if (formatType == 4)
+						instIt.op_storeLoad.mode = PPCREC_FPR_LD_MODE_PSQ_U8_PS0_PS1;
+					else if (formatType == 5)
+						instIt.op_storeLoad.mode = PPCREC_FPR_LD_MODE_PSQ_U16_PS0_PS1;
+					else if (formatType == 6)
+						instIt.op_storeLoad.mode = PPCREC_FPR_LD_MODE_PSQ_S8_PS0_PS1;
+					else if (formatType == 7)
+						instIt.op_storeLoad.mode = PPCREC_FPR_LD_MODE_PSQ_S16_PS0_PS1;
+					if (instIt.op_storeLoad.mode != PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0_PS1)
+						instIt.op_storeLoad.registerGQR = IMLREG_INVALID;
+				}
+			}
+			else if (instIt.type == PPCREC_IML_TYPE_FPR_STORE || instIt.type == PPCREC_IML_TYPE_FPR_STORE_INDEXED)
+			{
+				if(instIt.op_storeLoad.mode != PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0 &&
+					instIt.op_storeLoad.mode != PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0_PS1)
+					continue;
+				// get GQR value
+				cemu_assert_debug(instIt.op_storeLoad.registerGQR.IsValid());
+				sint32 gqrIndex = _getGQRIndexFromRegister(ppcImlGenContext, instIt.op_storeLoad.registerGQR);
+				cemu_assert(gqrIndex >= 0 && gqrIndex < 8);
+				if (ppcImlGenContext->tracking.modifiesGQR[gqrIndex])
+					continue;
+				uint32 gqrValue;
+				if(!PPCRecompiler_isUGQRValueKnown(ppcImlGenContext, gqrIndex, gqrValue))
+					continue;
+				uint32 formatType = (gqrValue >> 16) & 7;
+				uint32 scale = (gqrValue >> 24) & 0x3F;
+				if (scale != 0)
+					continue; // only generic handler supports scale
+				if (instIt.op_storeLoad.mode == PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0)
+				{
+					if (formatType == 0)
+						instIt.op_storeLoad.mode = PPCREC_FPR_ST_MODE_PSQ_FLOAT_PS0;
+					else if (formatType == 4)
+						instIt.op_storeLoad.mode = PPCREC_FPR_ST_MODE_PSQ_U8_PS0;
+					else if (formatType == 5)
+						instIt.op_storeLoad.mode = PPCREC_FPR_ST_MODE_PSQ_U16_PS0;
+					else if (formatType == 6)
+						instIt.op_storeLoad.mode = PPCREC_FPR_ST_MODE_PSQ_S8_PS0;
+					else if (formatType == 7)
+						instIt.op_storeLoad.mode = PPCREC_FPR_ST_MODE_PSQ_S16_PS0;
+					if (instIt.op_storeLoad.mode != PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0)
+						instIt.op_storeLoad.registerGQR = IMLREG_INVALID;
+				}
+				else if (instIt.op_storeLoad.mode == PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0_PS1)
+				{
+					if (formatType == 0)
+						instIt.op_storeLoad.mode = PPCREC_FPR_ST_MODE_PSQ_FLOAT_PS0_PS1;
+					else if (formatType == 4)
+						instIt.op_storeLoad.mode = PPCREC_FPR_ST_MODE_PSQ_U8_PS0_PS1;
+					else if (formatType == 5)
+						instIt.op_storeLoad.mode = PPCREC_FPR_ST_MODE_PSQ_U16_PS0_PS1;
+					else if (formatType == 6)
+						instIt.op_storeLoad.mode = PPCREC_FPR_ST_MODE_PSQ_S8_PS0_PS1;
+					else if (formatType == 7)
+						instIt.op_storeLoad.mode = PPCREC_FPR_ST_MODE_PSQ_S16_PS0_PS1;
+					if (instIt.op_storeLoad.mode != PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0_PS1)
+						instIt.op_storeLoad.registerGQR = IMLREG_INVALID;
+				}
+			}
+		}
+	}
+}
+
+// analyses register dependencies across the entire function
+// per segment this will generate information about which registers need to be preserved and which ones don't (e.g. are overwritten)
+class IMLOptimizerRegIOAnalysis
+{
+  public:
+	// constructor with segment pointer list as span
+	IMLOptimizerRegIOAnalysis(std::span<IMLSegment*> segmentList, uint32 maxRegId) : m_segmentList(segmentList), m_maxRegId(maxRegId)
+	{
+		m_segRegisterInOutList.resize(segmentList.size());
+	}
+
+	struct IMLSegmentRegisterInOut
+	{
+		// todo - since our register ID range is usually pretty small (<64) we could use integer bitmasks to accelerate this? There is a helper class used in RA code already
+		std::unordered_set<IMLRegID> regWritten; // registers which are modified in this segment
+		std::unordered_set<IMLRegID> regImported; // registers which are read in this segment before they are written (importing value from previous segments)
+		std::unordered_set<IMLRegID> regForward; // registers which are not read or written in this segment, but are imported into a later segment (propagated info)
+	};
+
+	// calculate which registers are imported (read-before-written) and forwarded (read-before-written by a later segment) per segment
+	// then in a second step propagate the dependencies across linked segments
+	void ComputeDepedencies()
+	{
+		std::vector<IMLSegmentRegisterInOut>& segRegisterInOutList = m_segRegisterInOutList;
+		IMLSegmentRegisterInOut* segIO = segRegisterInOutList.data();
+		uint32 index = 0;
+		for(auto& seg : m_segmentList)
+		{
+			seg->momentaryIndex = index;
+			index++;
+			for(auto& instr : seg->imlList)
+			{
+				IMLUsedRegisters registerUsage;
+				instr.CheckRegisterUsage(&registerUsage);
+				// registers are considered imported if they are read before being written in this seg
+				registerUsage.ForEachReadGPR([&](IMLReg gprReg) {
+					IMLRegID gprId = gprReg.GetRegID();
+					if (!segIO->regWritten.contains(gprId))
+					{
+						segIO->regImported.insert(gprId);
+					}
+				});
+				registerUsage.ForEachWrittenGPR([&](IMLReg gprReg) {
+					IMLRegID gprId = gprReg.GetRegID();
+					segIO->regWritten.insert(gprId);
+				});
+			}
+			segIO++;
+		}
+		// for every exit segment, import all registers
+		for(auto& seg : m_segmentList)
+		{
+			if (!seg->nextSegmentIsUncertain)
+				continue;
+			if(seg->deadCodeEliminationHintSeg)
+				continue;
+			IMLSegmentRegisterInOut& segIO = segRegisterInOutList[seg->momentaryIndex];
+			for(uint32 i=0; i<=m_maxRegId; i++)
+			{
+				segIO.regImported.insert((IMLRegID)i);
+			}
+		}
+		// broadcast dependencies across segment chains
+		std::unordered_set<uint32> segIdsWhichNeedUpdate;
+		for (uint32 i = 0; i < m_segmentList.size(); i++)
+		{
+			segIdsWhichNeedUpdate.insert(i);
+		}
+		while(!segIdsWhichNeedUpdate.empty())
+		{
+			auto firstIt = segIdsWhichNeedUpdate.begin();
+			uint32 segId = *firstIt;
+			segIdsWhichNeedUpdate.erase(firstIt);
+			// forward regImported and regForward to earlier segments into their regForward, unless the register is written
+			auto& curSeg = m_segmentList[segId];
+			IMLSegmentRegisterInOut& curSegIO = segRegisterInOutList[segId];
+			for(auto& prevSeg : curSeg->list_prevSegments)
+			{
+				IMLSegmentRegisterInOut& prevSegIO = segRegisterInOutList[prevSeg->momentaryIndex];
+				bool prevSegChanged = false;
+				for(auto& regId : curSegIO.regImported)
+				{
+					if (!prevSegIO.regWritten.contains(regId))
+						prevSegChanged |= prevSegIO.regForward.insert(regId).second;
+				}
+				for(auto& regId : curSegIO.regForward)
+				{
+					if (!prevSegIO.regWritten.contains(regId))
+						prevSegChanged |= prevSegIO.regForward.insert(regId).second;
+				}
+				if(prevSegChanged)
+					segIdsWhichNeedUpdate.insert(prevSeg->momentaryIndex);
+			}
+			// same for hint links
+			for(auto& prevSeg : curSeg->list_deadCodeHintBy)
+			{
+				IMLSegmentRegisterInOut& prevSegIO = segRegisterInOutList[prevSeg->momentaryIndex];
+				bool prevSegChanged = false;
+				for(auto& regId : curSegIO.regImported)
+				{
+					if (!prevSegIO.regWritten.contains(regId))
+						prevSegChanged |= prevSegIO.regForward.insert(regId).second;
+				}
+				for(auto& regId : curSegIO.regForward)
+				{
+					if (!prevSegIO.regWritten.contains(regId))
+						prevSegChanged |= prevSegIO.regForward.insert(regId).second;
+				}
+				if(prevSegChanged)
+					segIdsWhichNeedUpdate.insert(prevSeg->momentaryIndex);
+			}
+		}
+	}
+
+	std::unordered_set<IMLRegID> GetRegistersNeededAtEndOfSegment(IMLSegment& seg)
+	{
+		std::unordered_set<IMLRegID> regsNeeded;
+		if(seg.nextSegmentIsUncertain)
+		{
+			if(seg.deadCodeEliminationHintSeg)
+			{
+				auto& nextSegIO = m_segRegisterInOutList[seg.deadCodeEliminationHintSeg->momentaryIndex];
+				regsNeeded.insert(nextSegIO.regImported.begin(), nextSegIO.regImported.end());
+				regsNeeded.insert(nextSegIO.regForward.begin(), nextSegIO.regForward.end());
+			}
+			else
+			{
+				// add all regs
+				for(uint32 i = 0; i <= m_maxRegId; i++)
+					regsNeeded.insert(i);
+			}
+			return regsNeeded;
+		}
+		if(seg.nextSegmentBranchTaken)
+		{
+			auto& nextSegIO = m_segRegisterInOutList[seg.nextSegmentBranchTaken->momentaryIndex];
+			regsNeeded.insert(nextSegIO.regImported.begin(), nextSegIO.regImported.end());
+			regsNeeded.insert(nextSegIO.regForward.begin(), nextSegIO.regForward.end());
+		}
+		if(seg.nextSegmentBranchNotTaken)
+		{
+			auto& nextSegIO = m_segRegisterInOutList[seg.nextSegmentBranchNotTaken->momentaryIndex];
+			regsNeeded.insert(nextSegIO.regImported.begin(), nextSegIO.regImported.end());
+			regsNeeded.insert(nextSegIO.regForward.begin(), nextSegIO.regForward.end());
+		}
+		return regsNeeded;
+	}
+
+	bool IsRegisterNeededAtEndOfSegment(IMLSegment& seg, IMLRegID regId)
+	{
+		if(seg.nextSegmentIsUncertain)
+		{
+			if(!seg.deadCodeEliminationHintSeg)
+				return true;
+			auto& nextSegIO = m_segRegisterInOutList[seg.deadCodeEliminationHintSeg->momentaryIndex];
+			if(nextSegIO.regImported.contains(regId))
+				return true;
+			if(nextSegIO.regForward.contains(regId))
+				return true;
+			return false;
+		}
+		if(seg.nextSegmentBranchTaken)
+		{
+			auto& nextSegIO = m_segRegisterInOutList[seg.nextSegmentBranchTaken->momentaryIndex];
+			if(nextSegIO.regImported.contains(regId))
+				return true;
+			if(nextSegIO.regForward.contains(regId))
+				return true;
+		}
+		if(seg.nextSegmentBranchNotTaken)
+		{
+			auto& nextSegIO = m_segRegisterInOutList[seg.nextSegmentBranchNotTaken->momentaryIndex];
+			if(nextSegIO.regImported.contains(regId))
+				return true;
+			if(nextSegIO.regForward.contains(regId))
+				return true;
+		}
+		return false;
+	}
+
+  private:
+	std::span<IMLSegment*> m_segmentList;
+	uint32 m_maxRegId;
+
+	std::vector<IMLSegmentRegisterInOut> m_segRegisterInOutList;
+
+};
+
+// scan backwards starting from index and return the index of the first found instruction which writes to the given register (by id)
+sint32 IMLUtil_FindInstructionWhichWritesRegister(IMLSegment& seg, sint32 startIndex, IMLReg reg, sint32 maxScanDistance = -1)
+{
+	sint32 endIndex = std::max<sint32>(startIndex - maxScanDistance, 0);
+	for (sint32 i = startIndex; i >= endIndex; i--)
+	{
+		IMLInstruction& imlInstruction = seg.imlList[i];
+		IMLUsedRegisters registersUsed;
+		imlInstruction.CheckRegisterUsage(&registersUsed);
+		if (registersUsed.IsBaseGPRWritten(reg))
+			return i;
+	}
+	return -1;
+}
+
+// returns true if the instruction can safely be moved while keeping ordering constraints and data dependencies intact
+// initialIndex is inclusive, targetIndex is exclusive
+bool IMLUtil_CanMoveInstructionTo(IMLSegment& seg, sint32 initialIndex, sint32 targetIndex)
+{
+	boost::container::static_vector<IMLRegID, 8> regsWritten;
+	boost::container::static_vector<IMLRegID, 8> regsRead;
+	// get list of read and written registers
+	IMLUsedRegisters registersUsed;
+	seg.imlList[initialIndex].CheckRegisterUsage(&registersUsed);
+	registersUsed.ForEachAccessedGPR([&](IMLReg reg, bool isWritten) {
+		if (isWritten)
+			regsWritten.push_back(reg.GetRegID());
+		else
+			regsRead.push_back(reg.GetRegID());
+	});
+	// check all the instructions inbetween
+	if(initialIndex < targetIndex)
+	{
+		sint32 scanStartIndex = initialIndex+1; // +1 to skip the moving instruction itself
+		sint32 scanEndIndex = targetIndex;
+		for (sint32 i = scanStartIndex; i < scanEndIndex; i++)
+		{
+			IMLUsedRegisters registersUsed;
+			seg.imlList[i].CheckRegisterUsage(&registersUsed);
+			// in order to be able to move an instruction past another instruction, any of the read registers must not be modified (written)
+			// and any of it's written registers must not be read
+			bool canMove = true;
+			registersUsed.ForEachAccessedGPR([&](IMLReg reg, bool isWritten) {
+				IMLRegID regId = reg.GetRegID();
+				if (!isWritten)
+					canMove = canMove && std::find(regsWritten.begin(), regsWritten.end(), regId) == regsWritten.end();
+				else
+					canMove = canMove && std::find(regsRead.begin(), regsRead.end(), regId) == regsRead.end();
+			});
+			if(!canMove)
+				return false;
+		}
+	}
+	else
+	{
+		cemu_assert_unimplemented(); // backwards scan is todo
+		return false;
+	}
+	return true;
+}
+
+sint32 IMLUtil_CountRegisterReadsInRange(IMLSegment& seg, sint32 scanStartIndex, sint32 scanEndIndex, IMLRegID regId)
+{
+	cemu_assert_debug(scanStartIndex <= scanEndIndex);
+	cemu_assert_debug(scanEndIndex < seg.imlList.size());
+	sint32 count = 0;
+	for (sint32 i = scanStartIndex; i <= scanEndIndex; i++)
+	{
+		IMLUsedRegisters registersUsed;
+		seg.imlList[i].CheckRegisterUsage(&registersUsed);
+		registersUsed.ForEachReadGPR([&](IMLReg reg) {
+			if (reg.GetRegID() == regId)
+				count++;
+		});
+	}
+	return count;
+}
+
+// move instruction from one index to another
+// instruction will be inserted before the instruction at targetIndex
+// returns the new instruction index of the moved instruction
+sint32 IMLUtil_MoveInstructionTo(IMLSegment& seg, sint32 initialIndex, sint32 targetIndex)
+{
+	cemu_assert_debug(initialIndex != targetIndex);
+	IMLInstruction temp = seg.imlList[initialIndex];
+	if (initialIndex < targetIndex)
+	{
+		cemu_assert_debug(targetIndex > 0);
+		targetIndex--;
+		for(size_t i=initialIndex; i<targetIndex; i++)
+			seg.imlList[i] = seg.imlList[i+1];
+		seg.imlList[targetIndex] = temp;
+		return targetIndex;
+	}
+	else
+	{
+		cemu_assert_unimplemented(); // testing needed
+		std::copy(seg.imlList.begin() + targetIndex, seg.imlList.begin() + initialIndex, seg.imlList.begin() + targetIndex + 1);
+		seg.imlList[targetIndex] = temp;
+		return targetIndex;
+	}
+}
+
+// x86 specific
+bool IMLOptimizerX86_ModifiesEFlags(IMLInstruction& inst)
+{
+	// this is a very conservative implementation. There are more cases but this is good enough for now
+	if(inst.type == PPCREC_IML_TYPE_NAME_R || inst.type == PPCREC_IML_TYPE_R_NAME)
+		return false;
+	if((inst.type == PPCREC_IML_TYPE_R_R || inst.type == PPCREC_IML_TYPE_R_S32) && inst.operation == PPCREC_IML_OP_ASSIGN)
+		return false;
+	return true; // if we dont know for sure, assume it does
+}
+
+void IMLOptimizer_DebugPrintSeg(ppcImlGenContext_t& ppcImlGenContext, IMLSegment& seg)
+{
+	printf("----------------\n");
+	IMLDebug_DumpSegment(&ppcImlGenContext, &seg);
+	fflush(stdout);
+}
+
+void IMLOptimizer_RemoveDeadCodeFromSegment(IMLOptimizerRegIOAnalysis& regIoAnalysis, IMLSegment& seg)
+{
+	// algorithm works like this:
+	// Calculate which registers need to be preserved at the end of each segment
+	// Then for each segment:
+	// - Iterate instructions backwards
+	// - Maintain a list of registers which are read at a later point (initially this is the list from the first step)
+	// - If an instruction only modifies registers which are not in the read list and has no side effects, then it is dead code and can be replaced with a no-op
+
+	std::unordered_set<IMLRegID> regsNeeded = regIoAnalysis.GetRegistersNeededAtEndOfSegment(seg);
+
+	// start with suffix instruction
+	if(seg.HasSuffixInstruction())
+	{
+		IMLInstruction& imlInstruction = seg.imlList[seg.GetSuffixInstructionIndex()];
+		IMLUsedRegisters registersUsed;
+		imlInstruction.CheckRegisterUsage(&registersUsed);
+		registersUsed.ForEachWrittenGPR([&](IMLReg reg) {
+			regsNeeded.erase(reg.GetRegID());
+		});
+		registersUsed.ForEachReadGPR([&](IMLReg reg) {
+			regsNeeded.insert(reg.GetRegID());
+		});
+	}
+	// iterate instructions backwards
+	for (sint32 i = seg.imlList.size() - (seg.HasSuffixInstruction() ? 2:1); i >= 0; i--)
+	{
+		IMLInstruction& imlInstruction = seg.imlList[i];
+		IMLUsedRegisters registersUsed;
+		imlInstruction.CheckRegisterUsage(&registersUsed);
+		// register read -> remove from overwritten list
+		// register written -> add to overwritten list
+
+		// check if this instruction only writes registers which will never be read
+		bool onlyWritesRedundantRegisters = true;
+		registersUsed.ForEachWrittenGPR([&](IMLReg reg) {
+			if (regsNeeded.contains(reg.GetRegID()))
+				onlyWritesRedundantRegisters = false;
+		});
+		// check if any of the written registers are read after this point
+		registersUsed.ForEachWrittenGPR([&](IMLReg reg) {
+			regsNeeded.erase(reg.GetRegID());
+		});
+		registersUsed.ForEachReadGPR([&](IMLReg reg) {
+			regsNeeded.insert(reg.GetRegID());
+		});
+		if(!imlInstruction.HasSideEffects() && onlyWritesRedundantRegisters)
+		{
+			imlInstruction.make_no_op();
+		}
+	}
+}
+
+void IMLOptimizerX86_SubstituteCJumpForEflagsJump(IMLOptimizerRegIOAnalysis& regIoAnalysis, IMLSegment& seg)
+{
+	// convert and optimize bool condition jumps to eflags condition jumps
+	// - Moves eflag setter (e.g. cmp) closer to eflags consumer (conditional jump) if necessary. If not possible but required then exit early
+	// - Since we only rely on eflags, the boolean register can be optimized out if DCE considers it unused
+	// - Further detect and optimize patterns like DEC + CMP + JCC into fused ops (todo)
+
+	// check if this segment ends with a conditional jump
+	if(!seg.HasSuffixInstruction())
+		return;
+	sint32 cjmpInstIndex = seg.GetSuffixInstructionIndex();
+	if(cjmpInstIndex < 0)
+		return;
+	IMLInstruction& cjumpInstr = seg.imlList[cjmpInstIndex];
+	if( cjumpInstr.type != PPCREC_IML_TYPE_CONDITIONAL_JUMP )
+		return;
+	IMLReg regCondBool = cjumpInstr.op_conditional_jump.registerBool;
+	bool invertedCondition = !cjumpInstr.op_conditional_jump.mustBeTrue;
+	// find the instruction which sets the bool
+	sint32 cmpInstrIndex = IMLUtil_FindInstructionWhichWritesRegister(seg, cjmpInstIndex-1, regCondBool, 20);
+	if(cmpInstrIndex < 0)
+		return;
+	// check if its an instruction combo which can be optimized (currently only cmp + cjump) and get the condition
+	IMLInstruction& condSetterInstr = seg.imlList[cmpInstrIndex];
+	IMLCondition cond;
+	if(condSetterInstr.type == PPCREC_IML_TYPE_COMPARE)
+		cond = condSetterInstr.op_compare.cond;
+	else if(condSetterInstr.type == PPCREC_IML_TYPE_COMPARE_S32)
+		cond = condSetterInstr.op_compare_s32.cond;
+	else
+		return;
+	// check if instructions inbetween modify eflags
+	sint32 indexEflagsSafeStart = -1; // index of the first instruction which does not modify eflags up to cjump
+	for(sint32 i = cjmpInstIndex-1; i > cmpInstrIndex; i--)
+	{
+		if(IMLOptimizerX86_ModifiesEFlags(seg.imlList[i]))
+		{
+			indexEflagsSafeStart = i+1;
+			break;
+		}
+	}
+	if(indexEflagsSafeStart >= 0)
+	{
+		cemu_assert(indexEflagsSafeStart > 0);
+		// there are eflags-modifying instructions inbetween the bool setter and cjump
+		// try to move the eflags setter close enough to the cjump (to indexEflagsSafeStart)
+		bool canMove = IMLUtil_CanMoveInstructionTo(seg, cmpInstrIndex, indexEflagsSafeStart);
+		if(!canMove)
+		{
+			return;
+		}
+		else
+		{
+			cmpInstrIndex = IMLUtil_MoveInstructionTo(seg, cmpInstrIndex, indexEflagsSafeStart);
+		}
+	}
+	// we can turn the jump into an eflags jump
+	cjumpInstr.make_x86_eflags_jcc(cond, invertedCondition);
+
+	if (IMLUtil_CountRegisterReadsInRange(seg, cmpInstrIndex, cjmpInstIndex, regCondBool.GetRegID()) > 1 || regIoAnalysis.IsRegisterNeededAtEndOfSegment(seg, regCondBool.GetRegID()))
+		return; // bool register is used beyond the CMP, we can't drop it
+
+	auto& cmpInstr = seg.imlList[cmpInstrIndex];
+	cemu_assert_debug(cmpInstr.type == PPCREC_IML_TYPE_COMPARE || cmpInstr.type == PPCREC_IML_TYPE_COMPARE_S32);
+	if(cmpInstr.type == PPCREC_IML_TYPE_COMPARE)
+	{
+		IMLReg regA = cmpInstr.op_compare.regA;
+		IMLReg regB = cmpInstr.op_compare.regB;
+		seg.imlList[cmpInstrIndex].make_r_r(PPCREC_IML_OP_X86_CMP, regA, regB);
+	}
+	else
+	{
+		IMLReg regA = cmpInstr.op_compare_s32.regA;
+		sint32 val = cmpInstr.op_compare_s32.immS32;
+		seg.imlList[cmpInstrIndex].make_r_s32(PPCREC_IML_OP_X86_CMP, regA, val);
+	}
+
+}
+
+void IMLOptimizer_StandardOptimizationPassForSegment(IMLOptimizerRegIOAnalysis& regIoAnalysis, IMLSegment& seg)
+{
+	IMLOptimizer_RemoveDeadCodeFromSegment(regIoAnalysis, seg);
+
+	// x86 specific optimizations
+	IMLOptimizerX86_SubstituteCJumpForEflagsJump(regIoAnalysis, seg); // this pass should be applied late since it creates invisible eflags dependencies (which would break further register dependency analysis)
+}
+
+void IMLOptimizer_StandardOptimizationPass(ppcImlGenContext_t& ppcImlGenContext)
+{
+	IMLOptimizerRegIOAnalysis regIoAnalysis(ppcImlGenContext.segmentList2, ppcImlGenContext.GetMaxRegId());
+	regIoAnalysis.ComputeDepedencies();
+	for (IMLSegment* segIt : ppcImlGenContext.segmentList2)
+	{
+		IMLOptimizer_StandardOptimizationPassForSegment(regIoAnalysis, *segIt);
+	}
+}

src/Cafe/HW/Espresso/Recompiler/IML/IMLRegisterAllocator.h

@@ -0,0 +1,125 @@
+#pragma once
+
+// container for storing a set of register indices
+// specifically optimized towards storing typical range of physical register indices (expected to be below 64)
+class IMLPhysRegisterSet
+{
+public:
+	void SetAvailable(uint32 index)
+	{
+		cemu_assert_debug(index < 64);
+		m_regBitmask |= ((uint64)1 << index);
+	}
+
+	void SetReserved(uint32 index)
+	{
+		cemu_assert_debug(index < 64);
+		m_regBitmask &= ~((uint64)1 << index);
+	}
+
+	void SetAllAvailable()
+	{
+		m_regBitmask = ~0ull;
+	}
+
+	bool HasAllAvailable() const
+	{
+		return m_regBitmask == ~0ull;
+	}
+
+	bool IsAvailable(uint32 index) const
+	{
+		return (m_regBitmask & ((uint64)1 << index)) != 0;
+	}
+
+	IMLPhysRegisterSet& operator&=(const IMLPhysRegisterSet& other)
+	{
+		this->m_regBitmask &= other.m_regBitmask;
+		return *this;
+	}
+
+	IMLPhysRegisterSet& operator=(const IMLPhysRegisterSet& other)
+	{
+		this->m_regBitmask = other.m_regBitmask;
+		return *this;
+	}
+
+	void RemoveRegisters(const IMLPhysRegisterSet& other)
+	{
+		this->m_regBitmask &= ~other.m_regBitmask;
+	}
+
+	bool HasAnyAvailable() const
+	{
+		return m_regBitmask != 0;
+	}
+
+	bool HasExactlyOneAvailable() const
+	{
+		return m_regBitmask != 0 && (m_regBitmask & (m_regBitmask - 1)) == 0;
+	}
+
+	// returns index of first available register. Do not call when HasAnyAvailable() == false
+	IMLPhysReg GetFirstAvailableReg()
+	{
+		cemu_assert_debug(m_regBitmask != 0);
+		sint32 regIndex = 0;
+		auto tmp = m_regBitmask;
+		while ((tmp & 0xFF) == 0)
+		{
+			regIndex += 8;
+			tmp >>= 8;
+		}
+		while ((tmp & 0x1) == 0)
+		{
+			regIndex++;
+			tmp >>= 1;
+		}
+		return regIndex;
+	}
+
+	// returns index of next available register (search includes any register index >= startIndex)
+	// returns -1 if there is no more register
+	IMLPhysReg GetNextAvailableReg(sint32 startIndex) const
+	{
+		if (startIndex >= 64)
+			return -1;
+		uint32 regIndex = startIndex;
+		auto tmp = m_regBitmask;
+		tmp >>= regIndex;
+		if (!tmp)
+			return -1;
+		while ((tmp & 0xFF) == 0)
+		{
+			regIndex += 8;
+			tmp >>= 8;
+		}
+		while ((tmp & 0x1) == 0)
+		{
+			regIndex++;
+			tmp >>= 1;
+		}
+		return regIndex;
+	}
+
+	sint32 CountAvailableRegs() const
+	{
+		return std::popcount(m_regBitmask);
+	}
+
+private:
+	uint64 m_regBitmask{ 0 };
+};
+
+struct IMLRegisterAllocatorParameters
+{
+	inline IMLPhysRegisterSet& GetPhysRegPool(IMLRegFormat regFormat)
+	{
+		return perTypePhysPool[stdx::to_underlying(regFormat)];
+	}
+
+	IMLPhysRegisterSet perTypePhysPool[stdx::to_underlying(IMLRegFormat::TYPE_COUNT)];
+	std::unordered_map<IMLRegID, IMLName> regIdToName;
+};
+
+void IMLRegisterAllocator_AllocateRegisters(ppcImlGenContext_t* ppcImlGenContext, IMLRegisterAllocatorParameters& raParam);

src/Cafe/HW/Espresso/Recompiler/IML/IMLRegisterAllocatorRanges.cpp

@@ -0,0 +1,635 @@
+#include "../PPCRecompiler.h"
+#include "../PPCRecompilerIml.h"
+#include "IMLRegisterAllocatorRanges.h"
+#include "util/helpers/MemoryPool.h"
+
+uint32 IMLRA_GetNextIterationIndex();
+
+IMLRegID raLivenessRange::GetVirtualRegister() const
+{
+	return virtualRegister;
+}
+
+sint32 raLivenessRange::GetPhysicalRegister() const
+{
+	return physicalRegister;
+}
+
+IMLName raLivenessRange::GetName() const
+{
+	return name;
+}
+
+void raLivenessRange::SetPhysicalRegister(IMLPhysReg physicalRegister)
+{
+	this->physicalRegister = physicalRegister;
+}
+
+void raLivenessRange::SetPhysicalRegisterForCluster(IMLPhysReg physicalRegister)
+{
+	auto clusterRanges = GetAllSubrangesInCluster();
+	for(auto& range : clusterRanges)
+		range->physicalRegister = physicalRegister;
+}
+
+boost::container::small_vector<raLivenessRange*, 128> raLivenessRange::GetAllSubrangesInCluster()
+{
+	uint32 iterationIndex = IMLRA_GetNextIterationIndex();
+	boost::container::small_vector<raLivenessRange*, 128> subranges;
+	subranges.push_back(this);
+	this->lastIterationIndex = iterationIndex;
+	size_t i = 0;
+	while(i<subranges.size())
+	{
+		raLivenessRange* cur = subranges[i];
+		i++;
+		// check successors
+		if(cur->subrangeBranchTaken && cur->subrangeBranchTaken->lastIterationIndex != iterationIndex)
+		{
+			cur->subrangeBranchTaken->lastIterationIndex = iterationIndex;
+			subranges.push_back(cur->subrangeBranchTaken);
+		}
+		if(cur->subrangeBranchNotTaken && cur->subrangeBranchNotTaken->lastIterationIndex != iterationIndex)
+		{
+			cur->subrangeBranchNotTaken->lastIterationIndex = iterationIndex;
+			subranges.push_back(cur->subrangeBranchNotTaken);
+		}
+		// check predecessors
+		for(auto& prev : cur->previousRanges)
+		{
+			if(prev->lastIterationIndex != iterationIndex)
+			{
+				prev->lastIterationIndex = iterationIndex;
+				subranges.push_back(prev);
+			}
+		}
+	}
+	return subranges;
+}
+
+void raLivenessRange::GetAllowedRegistersExRecursive(raLivenessRange* range, uint32 iterationIndex, IMLPhysRegisterSet& allowedRegs)
+{
+	range->lastIterationIndex = iterationIndex;
+	for (auto& it : range->list_fixedRegRequirements)
+		allowedRegs &= it.allowedReg;
+	// check successors
+	if (range->subrangeBranchTaken && range->subrangeBranchTaken->lastIterationIndex != iterationIndex)
+		GetAllowedRegistersExRecursive(range->subrangeBranchTaken, iterationIndex, allowedRegs);
+	if (range->subrangeBranchNotTaken && range->subrangeBranchNotTaken->lastIterationIndex != iterationIndex)
+		GetAllowedRegistersExRecursive(range->subrangeBranchNotTaken, iterationIndex, allowedRegs);
+	// check predecessors
+	for (auto& prev : range->previousRanges)
+	{
+		if (prev->lastIterationIndex != iterationIndex)
+			GetAllowedRegistersExRecursive(prev, iterationIndex, allowedRegs);
+	}
+};
+
+bool raLivenessRange::GetAllowedRegistersEx(IMLPhysRegisterSet& allowedRegisters)
+{
+	uint32 iterationIndex = IMLRA_GetNextIterationIndex();
+	allowedRegisters.SetAllAvailable();
+	GetAllowedRegistersExRecursive(this, iterationIndex, allowedRegisters);
+	return !allowedRegisters.HasAllAvailable();
+}
+
+IMLPhysRegisterSet raLivenessRange::GetAllowedRegisters(IMLPhysRegisterSet regPool)
+{
+	IMLPhysRegisterSet fixedRegRequirements = regPool;
+	if(interval.ExtendsPreviousSegment() || interval.ExtendsIntoNextSegment())
+	{
+		auto clusterRanges = GetAllSubrangesInCluster();
+		for(auto& subrange : clusterRanges)
+		{
+			for(auto& fixedRegLoc : subrange->list_fixedRegRequirements)
+				fixedRegRequirements &= fixedRegLoc.allowedReg;
+		}
+		return fixedRegRequirements;
+	}
+	for(auto& fixedRegLoc : list_fixedRegRequirements)
+		fixedRegRequirements &= fixedRegLoc.allowedReg;
+	return fixedRegRequirements;
+}
+
+void PPCRecRARange_addLink_perVirtualGPR(std::unordered_map<IMLRegID, raLivenessRange*>& root, raLivenessRange* subrange)
+{
+	IMLRegID regId = subrange->GetVirtualRegister();
+	auto it = root.find(regId);
+	if (it == root.end())
+	{
+		// new single element
+		root.try_emplace(regId, subrange);
+		subrange->link_sameVirtualRegister.prev = nullptr;
+		subrange->link_sameVirtualRegister.next = nullptr;
+	}
+	else
+	{
+		// insert in first position
+		raLivenessRange* priorFirst = it->second;
+		subrange->link_sameVirtualRegister.next = priorFirst;
+		it->second = subrange;
+		subrange->link_sameVirtualRegister.prev = nullptr;
+		priorFirst->link_sameVirtualRegister.prev = subrange;
+	}
+}
+
+void PPCRecRARange_addLink_allSegmentRanges(raLivenessRange** root, raLivenessRange* subrange)
+{
+	subrange->link_allSegmentRanges.next = *root;
+	if (*root)
+		(*root)->link_allSegmentRanges.prev = subrange;
+	subrange->link_allSegmentRanges.prev = nullptr;
+	*root = subrange;
+}
+
+void PPCRecRARange_removeLink_perVirtualGPR(std::unordered_map<IMLRegID, raLivenessRange*>& root, raLivenessRange* subrange)
+{
+#ifdef CEMU_DEBUG_ASSERT
+	raLivenessRange* cur = root.find(subrange->GetVirtualRegister())->second;
+	bool hasRangeFound = false;
+	while(cur)
+	{
+		if(cur == subrange)
+		{
+			hasRangeFound = true;
+			break;
+		}
+		cur = cur->link_sameVirtualRegister.next;
+	}
+	cemu_assert_debug(hasRangeFound);
+#endif
+	IMLRegID regId = subrange->GetVirtualRegister();
+	raLivenessRange* nextRange = subrange->link_sameVirtualRegister.next;
+	raLivenessRange* prevRange = subrange->link_sameVirtualRegister.prev;
+	raLivenessRange* newBase = prevRange ? prevRange : nextRange;
+	if (prevRange)
+		prevRange->link_sameVirtualRegister.next = subrange->link_sameVirtualRegister.next;
+	if (nextRange)
+		nextRange->link_sameVirtualRegister.prev = subrange->link_sameVirtualRegister.prev;
+
+	if (!prevRange)
+	{
+		if (nextRange)
+		{
+			root.find(regId)->second = nextRange;
+		}
+		else
+		{
+			cemu_assert_debug(root.find(regId)->second == subrange);
+			root.erase(regId);
+		}
+	}
+#ifdef CEMU_DEBUG_ASSERT
+	subrange->link_sameVirtualRegister.prev = (raLivenessRange*)1;
+	subrange->link_sameVirtualRegister.next = (raLivenessRange*)1;
+#endif
+}
+
+void PPCRecRARange_removeLink_allSegmentRanges(raLivenessRange** root, raLivenessRange* subrange)
+{
+	raLivenessRange* tempPrev = subrange->link_allSegmentRanges.prev;
+	if (subrange->link_allSegmentRanges.prev)
+		subrange->link_allSegmentRanges.prev->link_allSegmentRanges.next = subrange->link_allSegmentRanges.next;
+	else
+		(*root) = subrange->link_allSegmentRanges.next;
+	if (subrange->link_allSegmentRanges.next)
+		subrange->link_allSegmentRanges.next->link_allSegmentRanges.prev = tempPrev;
+#ifdef CEMU_DEBUG_ASSERT
+	subrange->link_allSegmentRanges.prev = (raLivenessRange*)1;
+	subrange->link_allSegmentRanges.next = (raLivenessRange*)1;
+#endif
+}
+
+MemoryPoolPermanentObjects<raLivenessRange> memPool_livenessSubrange(4096);
+
+// startPosition and endPosition are inclusive
+raLivenessRange* IMLRA_CreateRange(ppcImlGenContext_t* ppcImlGenContext, IMLSegment* imlSegment, IMLRegID virtualRegister, IMLName name, raInstructionEdge startPosition, raInstructionEdge endPosition)
+{
+	raLivenessRange* range = memPool_livenessSubrange.acquireObj();
+	range->previousRanges.clear();
+	range->list_accessLocations.clear();
+	range->list_fixedRegRequirements.clear();
+	range->imlSegment = imlSegment;
+
+	cemu_assert_debug(startPosition <= endPosition);
+	range->interval.start = startPosition;
+	range->interval.end = endPosition;
+
+	// register mapping
+	range->virtualRegister = virtualRegister;
+	range->name = name;
+	range->physicalRegister = -1;
+	// default values
+	range->hasStore = false;
+	range->hasStoreDelayed = false;
+	range->lastIterationIndex = 0;
+	range->subrangeBranchNotTaken = nullptr;
+	range->subrangeBranchTaken = nullptr;
+	cemu_assert_debug(range->previousRanges.empty());
+	range->_noLoad = false;
+	// add to segment linked lists
+	PPCRecRARange_addLink_perVirtualGPR(imlSegment->raInfo.linkedList_perVirtualRegister, range);
+	PPCRecRARange_addLink_allSegmentRanges(&imlSegment->raInfo.linkedList_allSubranges, range);
+	return range;
+}
+
+void _unlinkSubrange(raLivenessRange* range)
+{
+	IMLSegment* imlSegment = range->imlSegment;
+	PPCRecRARange_removeLink_perVirtualGPR(imlSegment->raInfo.linkedList_perVirtualRegister, range);
+	PPCRecRARange_removeLink_allSegmentRanges(&imlSegment->raInfo.linkedList_allSubranges, range);
+	// unlink reverse references
+	if(range->subrangeBranchTaken)
+		range->subrangeBranchTaken->previousRanges.erase(std::find(range->subrangeBranchTaken->previousRanges.begin(), range->subrangeBranchTaken->previousRanges.end(), range));
+	if(range->subrangeBranchNotTaken)
+		range->subrangeBranchNotTaken->previousRanges.erase(std::find(range->subrangeBranchNotTaken->previousRanges.begin(), range->subrangeBranchNotTaken->previousRanges.end(), range));
+	range->subrangeBranchTaken = (raLivenessRange*)(uintptr_t)-1;
+	range->subrangeBranchNotTaken = (raLivenessRange*)(uintptr_t)-1;
+	// remove forward references
+	for(auto& prev : range->previousRanges)
+	{
+		if(prev->subrangeBranchTaken == range)
+			prev->subrangeBranchTaken = nullptr;
+		if(prev->subrangeBranchNotTaken == range)
+			prev->subrangeBranchNotTaken = nullptr;
+	}
+	range->previousRanges.clear();
+}
+
+void IMLRA_DeleteRange(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange* range)
+{
+	_unlinkSubrange(range);
+	range->list_accessLocations.clear();
+	range->list_fixedRegRequirements.clear();
+	memPool_livenessSubrange.releaseObj(range);
+}
+
+void IMLRA_DeleteRangeCluster(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange* range)
+{
+	auto clusterRanges = range->GetAllSubrangesInCluster();
+	for (auto& subrange : clusterRanges)
+		IMLRA_DeleteRange(ppcImlGenContext, subrange);
+}
+
+void IMLRA_DeleteAllRanges(ppcImlGenContext_t* ppcImlGenContext)
+{
+	for(auto& seg : ppcImlGenContext->segmentList2)
+	{
+		raLivenessRange* cur;
+		while(cur = seg->raInfo.linkedList_allSubranges)
+			IMLRA_DeleteRange(ppcImlGenContext, cur);
+		seg->raInfo.linkedList_allSubranges = nullptr;
+		seg->raInfo.linkedList_perVirtualRegister.clear();
+	}
+}
+
+void IMLRA_MergeSubranges(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange* subrange, raLivenessRange* absorbedSubrange)
+{
+#ifdef CEMU_DEBUG_ASSERT
+	PPCRecRA_debugValidateSubrange(subrange);
+	PPCRecRA_debugValidateSubrange(absorbedSubrange);
+	if (subrange->imlSegment != absorbedSubrange->imlSegment)
+		assert_dbg();
+	cemu_assert_debug(subrange->interval.end == absorbedSubrange->interval.start);
+
+	if (subrange->subrangeBranchTaken || subrange->subrangeBranchNotTaken)
+		assert_dbg();
+	if (subrange == absorbedSubrange)
+		assert_dbg();
+#endif
+	// update references
+	subrange->subrangeBranchTaken = absorbedSubrange->subrangeBranchTaken;
+	subrange->subrangeBranchNotTaken = absorbedSubrange->subrangeBranchNotTaken;
+	absorbedSubrange->subrangeBranchTaken = nullptr;
+	absorbedSubrange->subrangeBranchNotTaken = nullptr;
+	if(subrange->subrangeBranchTaken)
+		*std::find(subrange->subrangeBranchTaken->previousRanges.begin(), subrange->subrangeBranchTaken->previousRanges.end(), absorbedSubrange) = subrange;
+	if(subrange->subrangeBranchNotTaken)
+		*std::find(subrange->subrangeBranchNotTaken->previousRanges.begin(), subrange->subrangeBranchNotTaken->previousRanges.end(), absorbedSubrange) = subrange;
+
+	// merge usage locations
+	for (auto& accessLoc : absorbedSubrange->list_accessLocations)
+		subrange->list_accessLocations.push_back(accessLoc);
+	absorbedSubrange->list_accessLocations.clear();
+	// merge fixed reg locations
+#ifdef CEMU_DEBUG_ASSERT
+	if(!subrange->list_fixedRegRequirements.empty() && !absorbedSubrange->list_fixedRegRequirements.empty())
+	{
+		cemu_assert_debug(subrange->list_fixedRegRequirements.back().pos < absorbedSubrange->list_fixedRegRequirements.front().pos);
+	}
+#endif
+	for (auto& fixedReg : absorbedSubrange->list_fixedRegRequirements)
+		subrange->list_fixedRegRequirements.push_back(fixedReg);
+	absorbedSubrange->list_fixedRegRequirements.clear();
+
+	subrange->interval.end = absorbedSubrange->interval.end;
+
+	PPCRecRA_debugValidateSubrange(subrange);
+
+	IMLRA_DeleteRange(ppcImlGenContext, absorbedSubrange);
+}
+
+// remove all inter-segment connections from the range cluster and split it into local ranges. Ranges are trimmed and if they have no access location they will be removed
+void IMLRA_ExplodeRangeCluster(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange* originRange)
+{
+	cemu_assert_debug(originRange->interval.ExtendsPreviousSegment() || originRange->interval.ExtendsIntoNextSegment()); // only call this on ranges that span multiple segments
+	auto clusterRanges = originRange->GetAllSubrangesInCluster();
+	for (auto& subrange : clusterRanges)
+	{
+		if (subrange->list_accessLocations.empty())
+			continue;
+		raInterval interval;
+		interval.SetInterval(subrange->list_accessLocations.front().pos, subrange->list_accessLocations.back().pos);
+		raLivenessRange* newSubrange = IMLRA_CreateRange(ppcImlGenContext, subrange->imlSegment, subrange->GetVirtualRegister(), subrange->GetName(), interval.start, interval.end);
+		// copy locations and fixed reg indices
+		newSubrange->list_accessLocations = subrange->list_accessLocations;
+		newSubrange->list_fixedRegRequirements = subrange->list_fixedRegRequirements;
+		if(originRange->HasPhysicalRegister())
+		{
+			cemu_assert_debug(subrange->list_fixedRegRequirements.empty()); // avoid unassigning a register from a range with a fixed register requirement
+		}
+		// validate
+		if(!newSubrange->list_accessLocations.empty())
+		{
+			cemu_assert_debug(newSubrange->list_accessLocations.front().pos >= newSubrange->interval.start);
+			cemu_assert_debug(newSubrange->list_accessLocations.back().pos <= newSubrange->interval.end);
+		}
+		if(!newSubrange->list_fixedRegRequirements.empty())
+		{
+			cemu_assert_debug(newSubrange->list_fixedRegRequirements.front().pos >= newSubrange->interval.start); // fixed register requirements outside of the actual access range probably means there is a mistake in GetInstructionFixedRegisters()
+			cemu_assert_debug(newSubrange->list_fixedRegRequirements.back().pos <= newSubrange->interval.end);
+		}
+	}
+	// delete the original range cluster
+	IMLRA_DeleteRangeCluster(ppcImlGenContext, originRange);
+}
+
+#ifdef CEMU_DEBUG_ASSERT
+void PPCRecRA_debugValidateSubrange(raLivenessRange* range)
+{
+	// validate subrange
+	if (range->subrangeBranchTaken && range->subrangeBranchTaken->imlSegment != range->imlSegment->nextSegmentBranchTaken)
+		assert_dbg();
+	if (range->subrangeBranchNotTaken && range->subrangeBranchNotTaken->imlSegment != range->imlSegment->nextSegmentBranchNotTaken)
+		assert_dbg();
+
+	if(range->subrangeBranchTaken || range->subrangeBranchNotTaken)
+	{
+		cemu_assert_debug(range->interval.end.ConnectsToNextSegment());
+	}
+	if(!range->previousRanges.empty())
+	{
+		cemu_assert_debug(range->interval.start.ConnectsToPreviousSegment());
+	}
+	// validate locations
+	if (!range->list_accessLocations.empty())
+	{
+		cemu_assert_debug(range->list_accessLocations.front().pos >= range->interval.start);
+		cemu_assert_debug(range->list_accessLocations.back().pos <= range->interval.end);
+	}
+	// validate fixed reg requirements
+	if (!range->list_fixedRegRequirements.empty())
+	{
+		cemu_assert_debug(range->list_fixedRegRequirements.front().pos >= range->interval.start);
+		cemu_assert_debug(range->list_fixedRegRequirements.back().pos <= range->interval.end);
+		for(sint32 i = 0; i < (sint32)range->list_fixedRegRequirements.size()-1; i++)
+			cemu_assert_debug(range->list_fixedRegRequirements[i].pos < range->list_fixedRegRequirements[i+1].pos);
+	}
+
+}
+#else
+void PPCRecRA_debugValidateSubrange(raLivenessRange* range) {}
+#endif
+
+// trim start and end of range to match first and last read/write locations
+// does not trim start/endpoints which extend into the next/previous segment
+void IMLRA_TrimRangeToUse(raLivenessRange* range)
+{
+	if(range->list_accessLocations.empty())
+	{
+		// special case where we trim ranges extending from other segments to a single instruction edge
+		cemu_assert_debug(!range->interval.start.IsInstructionIndex() || !range->interval.end.IsInstructionIndex());
+		if(range->interval.start.IsInstructionIndex())
+			range->interval.start = range->interval.end;
+		if(range->interval.end.IsInstructionIndex())
+			range->interval.end = range->interval.start;
+		return;
+	}
+	// trim start and end
+	raInterval prevInterval = range->interval;
+	if(range->interval.start.IsInstructionIndex())
+		range->interval.start = range->list_accessLocations.front().pos;
+	if(range->interval.end.IsInstructionIndex())
+		range->interval.end = range->list_accessLocations.back().pos;
+	// extra checks
+#ifdef CEMU_DEBUG_ASSERT
+	cemu_assert_debug(range->interval.start <= range->interval.end);
+	for(auto& loc : range->list_accessLocations)
+	{
+		cemu_assert_debug(range->interval.ContainsEdge(loc.pos));
+	}
+	cemu_assert_debug(prevInterval.ContainsWholeInterval(range->interval));
+#endif
+}
+
+// split range at the given position
+// After the split there will be two ranges:
+// head -> subrange is shortened to end at splitIndex (exclusive)
+// tail -> a new subrange that ranges from splitIndex (inclusive) to the end of the original subrange
+// if head has a physical register assigned it will not carry over to tail
+// The return value is the tail range
+// If trimToUsage is true, the end of the head subrange and the start of the tail subrange will be shrunk to fit the read/write locations within. If there are no locations then the range will be deleted
+raLivenessRange* IMLRA_SplitRange(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange*& subrange, raInstructionEdge splitPosition, bool trimToUsage)
+{
+	cemu_assert_debug(splitPosition.IsInstructionIndex());
+	cemu_assert_debug(!subrange->interval.IsNextSegmentOnly() && !subrange->interval.IsPreviousSegmentOnly());
+	cemu_assert_debug(subrange->interval.ContainsEdge(splitPosition));
+	// determine new intervals
+	raInterval headInterval, tailInterval;
+	headInterval.SetInterval(subrange->interval.start, splitPosition-1);
+	tailInterval.SetInterval(splitPosition, subrange->interval.end);
+	cemu_assert_debug(headInterval.start <= headInterval.end);
+	cemu_assert_debug(tailInterval.start <= tailInterval.end);
+	// create tail
+	raLivenessRange* tailSubrange = IMLRA_CreateRange(ppcImlGenContext, subrange->imlSegment, subrange->GetVirtualRegister(), subrange->GetName(), tailInterval.start, tailInterval.end);
+	tailSubrange->SetPhysicalRegister(subrange->GetPhysicalRegister());
+	// carry over branch targets and update reverse references
+	tailSubrange->subrangeBranchTaken = subrange->subrangeBranchTaken;
+	tailSubrange->subrangeBranchNotTaken = subrange->subrangeBranchNotTaken;
+	subrange->subrangeBranchTaken = nullptr;
+	subrange->subrangeBranchNotTaken = nullptr;
+	if(tailSubrange->subrangeBranchTaken)
+		*std::find(tailSubrange->subrangeBranchTaken->previousRanges.begin(), tailSubrange->subrangeBranchTaken->previousRanges.end(), subrange) = tailSubrange;
+	if(tailSubrange->subrangeBranchNotTaken)
+		*std::find(tailSubrange->subrangeBranchNotTaken->previousRanges.begin(), tailSubrange->subrangeBranchNotTaken->previousRanges.end(), subrange) = tailSubrange;
+	// we assume that list_locations is ordered by instruction index and contains no duplicate indices, so lets check that here just in case
+#ifdef CEMU_DEBUG_ASSERT
+	if(subrange->list_accessLocations.size() > 1)
+	{
+		for(size_t i=0; i<subrange->list_accessLocations.size()-1; i++)
+		{
+			cemu_assert_debug(subrange->list_accessLocations[i].pos < subrange->list_accessLocations[i+1].pos);
+		}
+	}
+#endif
+	// split locations
+	auto it = std::lower_bound(
+		subrange->list_accessLocations.begin(), subrange->list_accessLocations.end(), splitPosition,
+		[](const raAccessLocation& accessLoc, raInstructionEdge value) { return accessLoc.pos < value; }
+	);
+	size_t originalCount = subrange->list_accessLocations.size();
+	tailSubrange->list_accessLocations.insert(tailSubrange->list_accessLocations.end(), it, subrange->list_accessLocations.end());
+	subrange->list_accessLocations.erase(it, subrange->list_accessLocations.end());
+	cemu_assert_debug(subrange->list_accessLocations.empty() || subrange->list_accessLocations.back().pos < splitPosition);
+	cemu_assert_debug(tailSubrange->list_accessLocations.empty() || tailSubrange->list_accessLocations.front().pos >= splitPosition);
+	cemu_assert_debug(subrange->list_accessLocations.size() + tailSubrange->list_accessLocations.size() == originalCount);
+	// split fixed reg requirements
+	for (sint32 i = 0; i < subrange->list_fixedRegRequirements.size(); i++)
+	{
+		raFixedRegRequirement* fixedReg = subrange->list_fixedRegRequirements.data() + i;
+		if (tailInterval.ContainsEdge(fixedReg->pos))
+		{
+			tailSubrange->list_fixedRegRequirements.push_back(*fixedReg);
+		}
+	}
+	// remove tail fixed reg requirements from head
+	for (sint32 i = 0; i < subrange->list_fixedRegRequirements.size(); i++)
+	{
+		raFixedRegRequirement* fixedReg = subrange->list_fixedRegRequirements.data() + i;
+		if (!headInterval.ContainsEdge(fixedReg->pos))
+		{
+			subrange->list_fixedRegRequirements.resize(i);
+			break;
+		}
+	}
+	// adjust intervals
+	subrange->interval = headInterval;
+	tailSubrange->interval = tailInterval;
+	// trim to hole
+	if(trimToUsage)
+	{
+		if(subrange->list_accessLocations.empty() && (subrange->interval.start.IsInstructionIndex() && subrange->interval.end.IsInstructionIndex()))
+		{
+			IMLRA_DeleteRange(ppcImlGenContext, subrange);
+			subrange = nullptr;
+		}
+		else
+		{
+			IMLRA_TrimRangeToUse(subrange);
+		}
+		if(tailSubrange->list_accessLocations.empty() && (tailSubrange->interval.start.IsInstructionIndex() && tailSubrange->interval.end.IsInstructionIndex()))
+		{
+			IMLRA_DeleteRange(ppcImlGenContext, tailSubrange);
+			tailSubrange = nullptr;
+		}
+		else
+		{
+			IMLRA_TrimRangeToUse(tailSubrange);
+		}
+	}
+	// validation
+	cemu_assert_debug(!subrange || subrange->interval.start <= subrange->interval.end);
+	cemu_assert_debug(!tailSubrange || tailSubrange->interval.start <= tailSubrange->interval.end);
+	cemu_assert_debug(!tailSubrange || tailSubrange->interval.start >= splitPosition);
+	if (!trimToUsage)
+		cemu_assert_debug(!tailSubrange || tailSubrange->interval.start == splitPosition);
+
+	if(subrange)
+		PPCRecRA_debugValidateSubrange(subrange);
+	if(tailSubrange)
+		PPCRecRA_debugValidateSubrange(tailSubrange);
+	return tailSubrange;
+}
+
+sint32 IMLRA_GetSegmentReadWriteCost(IMLSegment* imlSegment)
+{
+	sint32 v = imlSegment->loopDepth + 1;
+	v *= 5;
+	return v*v; // 25, 100, 225, 400
+}
+
+// calculate additional cost of range that it would have after calling _ExplodeRange() on it
+sint32 IMLRA_CalculateAdditionalCostOfRangeExplode(raLivenessRange* subrange)
+{
+	auto ranges = subrange->GetAllSubrangesInCluster();
+	sint32 cost = 0;//-PPCRecRARange_estimateTotalCost(ranges);
+	for (auto& subrange : ranges)
+	{
+		if (subrange->list_accessLocations.empty())
+			continue; // this range would be deleted and thus has no cost
+		sint32 segmentLoadStoreCost = IMLRA_GetSegmentReadWriteCost(subrange->imlSegment);
+		bool hasAdditionalLoad = subrange->interval.ExtendsPreviousSegment();
+		bool hasAdditionalStore = subrange->interval.ExtendsIntoNextSegment();
+		if(hasAdditionalLoad && subrange->list_accessLocations.front().IsWrite()) // if written before read then a load isn't necessary
+		{
+			cemu_assert_debug(!subrange->list_accessLocations.front().IsRead());
+			cost += segmentLoadStoreCost;
+		}
+		if(hasAdditionalStore)
+		{
+			bool hasWrite = std::find_if(subrange->list_accessLocations.begin(), subrange->list_accessLocations.end(), [](const raAccessLocation& loc) { return loc.IsWrite(); }) != subrange->list_accessLocations.end();
+			if(!hasWrite) // ranges which don't modify their value do not need to be stored
+				cost += segmentLoadStoreCost;
+		}
+	}
+	// todo - properly calculating all the data-flow dependency based costs is more complex so this currently is an approximation
+	return cost;
+}
+
+sint32 IMLRA_CalculateAdditionalCostAfterSplit(raLivenessRange* subrange, raInstructionEdge splitPosition)
+{
+	// validation
+#ifdef CEMU_DEBUG_ASSERT
+	if (subrange->interval.ExtendsIntoNextSegment())
+		assert_dbg();
+#endif
+	cemu_assert_debug(splitPosition.IsInstructionIndex());
+
+	sint32 cost = 0;
+	// find split position in location list
+	if (subrange->list_accessLocations.empty())
+		return 0;
+	if (splitPosition <= subrange->list_accessLocations.front().pos)
+		return 0;
+	if (splitPosition > subrange->list_accessLocations.back().pos)
+		return 0;
+
+	size_t firstTailLocationIndex = 0;
+	for (size_t i = 0; i < subrange->list_accessLocations.size(); i++)
+	{
+		if (subrange->list_accessLocations[i].pos >= splitPosition)
+		{
+			firstTailLocationIndex = i;
+			break;
+		}
+	}
+	std::span<raAccessLocation> headLocations{subrange->list_accessLocations.data(), firstTailLocationIndex};
+	std::span<raAccessLocation> tailLocations{subrange->list_accessLocations.data() + firstTailLocationIndex, subrange->list_accessLocations.size() - firstTailLocationIndex};
+	cemu_assert_debug(headLocations.empty() || headLocations.back().pos < splitPosition);
+	cemu_assert_debug(tailLocations.empty() || tailLocations.front().pos >= splitPosition);
+
+	sint32 segmentLoadStoreCost = IMLRA_GetSegmentReadWriteCost(subrange->imlSegment);
+
+	auto CalculateCostFromLocationRange = [segmentLoadStoreCost](std::span<raAccessLocation> locations, bool trackLoadCost = true, bool trackStoreCost = true) -> sint32
+	{
+		if(locations.empty())
+			return 0;
+		sint32 cost = 0;
+		if(locations.front().IsRead() && trackLoadCost)
+			cost += segmentLoadStoreCost; // not overwritten, so there is a load cost
+		bool hasWrite = std::find_if(locations.begin(), locations.end(), [](const raAccessLocation& loc) { return loc.IsWrite(); }) != locations.end();
+		if(hasWrite && trackStoreCost)
+			cost += segmentLoadStoreCost; // modified, so there is a store cost
+		return cost;
+	};
+
+	sint32 baseCost = CalculateCostFromLocationRange(subrange->list_accessLocations);
+
+	bool tailOverwritesValue = !tailLocations.empty() && !tailLocations.front().IsRead() && tailLocations.front().IsWrite();
+
+	sint32 newCost = CalculateCostFromLocationRange(headLocations) + CalculateCostFromLocationRange(tailLocations, !tailOverwritesValue, true);
+	cemu_assert_debug(newCost >= baseCost);
+	cost = newCost - baseCost;
+
+	return cost;
+}

src/Cafe/HW/Espresso/Recompiler/IML/IMLRegisterAllocatorRanges.h

@@ -0,0 +1,364 @@
+#pragma once
+#include "IMLRegisterAllocator.h"
+
+struct raLivenessSubrangeLink
+{
+	struct raLivenessRange* prev;
+	struct raLivenessRange* next;
+};
+
+struct raInstructionEdge
+{
+	friend struct raInterval;
+public:
+	raInstructionEdge()
+	{
+		index = 0;
+	}
+
+	raInstructionEdge(sint32 instructionIndex, bool isInputEdge)
+	{
+		Set(instructionIndex, isInputEdge);
+	}
+
+	void Set(sint32 instructionIndex, bool isInputEdge)
+	{
+		if(instructionIndex == RA_INTER_RANGE_START || instructionIndex == RA_INTER_RANGE_END)
+		{
+			index = instructionIndex;
+			return;
+		}
+		index = instructionIndex * 2 + (isInputEdge ? 0 : 1);
+		cemu_assert_debug(index >= 0 && index < 0x100000*2); // make sure index value is sane
+	}
+
+	void SetRaw(sint32 index)
+	{
+		this->index = index;
+		cemu_assert_debug(index == RA_INTER_RANGE_START || index == RA_INTER_RANGE_END || (index >= 0 && index < 0x100000*2)); // make sure index value is sane
+	}
+
+	// sint32 GetRaw()
+	// {
+	// 	this->index = index;
+	// }
+
+	std::string GetDebugString()
+	{
+		if(index == RA_INTER_RANGE_START)
+			return "RA_START";
+		else if(index == RA_INTER_RANGE_END)
+			return "RA_END";
+		std::string str = fmt::format("{}", GetInstructionIndex());
+		if(IsOnInputEdge())
+			str += "i";
+		else if(IsOnOutputEdge())
+			str += "o";
+		return str;
+	}
+
+	sint32 GetInstructionIndex() const
+	{
+		cemu_assert_debug(index != RA_INTER_RANGE_START && index != RA_INTER_RANGE_END);
+		return index >> 1;
+	}
+
+	// returns instruction index or RA_INTER_RANGE_START/RA_INTER_RANGE_END
+	sint32 GetInstructionIndexEx() const
+	{
+		if(index == RA_INTER_RANGE_START || index == RA_INTER_RANGE_END)
+			return index;
+		return index >> 1;
+	}
+
+	sint32 GetRaw() const
+	{
+		return index;
+	}
+
+	bool IsOnInputEdge() const
+	{
+		cemu_assert_debug(index != RA_INTER_RANGE_START && index != RA_INTER_RANGE_END);
+		return (index&1) == 0;
+	}
+
+	bool IsOnOutputEdge() const
+	{
+		cemu_assert_debug(index != RA_INTER_RANGE_START && index != RA_INTER_RANGE_END);
+		return (index&1) != 0;
+	}
+
+	bool ConnectsToPreviousSegment() const
+	{
+		return index == RA_INTER_RANGE_START;
+	}
+
+	bool ConnectsToNextSegment() const
+	{
+		return index == RA_INTER_RANGE_END;
+	}
+
+	bool IsInstructionIndex() const
+	{
+		return index != RA_INTER_RANGE_START && index != RA_INTER_RANGE_END;
+	}
+
+	// comparison operators
+	bool operator>(const raInstructionEdge& other) const
+	{
+		return index > other.index;
+	}
+	bool operator<(const raInstructionEdge& other) const
+	{
+		return index < other.index;
+	}
+	bool operator<=(const raInstructionEdge& other) const
+	{
+		return index <= other.index;
+	}
+	bool operator>=(const raInstructionEdge& other) const
+	{
+		return index >= other.index;
+	}
+	bool operator==(const raInstructionEdge& other) const
+	{
+		return index == other.index;
+	}
+
+	raInstructionEdge operator+(sint32 offset) const
+	{
+		cemu_assert_debug(IsInstructionIndex());
+		cemu_assert_debug(offset >= 0 && offset < RA_INTER_RANGE_END);
+		raInstructionEdge edge;
+		edge.index = index + offset;
+		return edge;
+	}
+
+	raInstructionEdge operator-(sint32 offset) const
+	{
+		cemu_assert_debug(IsInstructionIndex());
+		cemu_assert_debug(offset >= 0 && offset < RA_INTER_RANGE_END);
+		raInstructionEdge edge;
+		edge.index = index - offset;
+		return edge;
+	}
+
+	raInstructionEdge& operator++()
+	{
+		cemu_assert_debug(IsInstructionIndex());
+		index++;
+		return *this;
+	}
+
+private:
+	sint32 index; // can also be RA_INTER_RANGE_START or RA_INTER_RANGE_END, otherwise contains instruction index * 2
+
+};
+
+struct raAccessLocation
+{
+	raAccessLocation(raInstructionEdge pos) : pos(pos) {}
+
+	bool IsRead() const
+	{
+		return pos.IsOnInputEdge();
+	}
+
+	bool IsWrite() const
+	{
+		return pos.IsOnOutputEdge();
+	}
+
+	raInstructionEdge pos;
+};
+
+struct raInterval
+{
+	raInterval()
+	{
+
+	}
+
+	raInterval(raInstructionEdge start, raInstructionEdge end)
+	{
+		SetInterval(start, end);
+	}
+
+	// isStartOnInput = Input+Output edge on first instruction. If false then only output
+	// isEndOnOutput = Input+Output edge on last instruction. If false then only input
+	void SetInterval(sint32 start, bool isStartOnInput, sint32 end, bool isEndOnOutput)
+	{
+		this->start.Set(start, isStartOnInput);
+		this->end.Set(end, !isEndOnOutput);
+	}
+
+	void SetInterval(raInstructionEdge start, raInstructionEdge end)
+	{
+		cemu_assert_debug(start <= end);
+		this->start = start;
+		this->end = end;
+	}
+
+	void SetStart(const raInstructionEdge& edge)
+	{
+		start = edge;
+	}
+
+	void SetEnd(const raInstructionEdge& edge)
+	{
+		end = edge;
+	}
+
+	sint32 GetStartIndex() const
+	{
+		return start.GetInstructionIndex();
+	}
+
+	sint32 GetEndIndex() const
+	{
+		return end.GetInstructionIndex();
+	}
+
+	bool ExtendsPreviousSegment() const
+	{
+		return start.ConnectsToPreviousSegment();
+	}
+
+	bool ExtendsIntoNextSegment() const
+	{
+		return end.ConnectsToNextSegment();
+	}
+
+	bool IsNextSegmentOnly() const
+	{
+		return start.ConnectsToNextSegment() && end.ConnectsToNextSegment();
+	}
+
+	bool IsPreviousSegmentOnly() const
+	{
+		return start.ConnectsToPreviousSegment() && end.ConnectsToPreviousSegment();
+	}
+
+	// returns true if range is contained within a single segment
+	bool IsLocal() const
+	{
+		return start.GetRaw() > RA_INTER_RANGE_START && end.GetRaw() < RA_INTER_RANGE_END;
+	}
+
+	bool ContainsInstructionIndex(sint32 instructionIndex) const
+	{
+		cemu_assert_debug(instructionIndex != RA_INTER_RANGE_START && instructionIndex != RA_INTER_RANGE_END);
+		return instructionIndex >= start.GetInstructionIndexEx() && instructionIndex <= end.GetInstructionIndexEx();
+	}
+
+	// similar to ContainsInstructionIndex, but allows RA_INTER_RANGE_START/END as input
+	bool ContainsInstructionIndexEx(sint32 instructionIndex) const
+	{
+		if(instructionIndex == RA_INTER_RANGE_START)
+			return start.ConnectsToPreviousSegment();
+		if(instructionIndex == RA_INTER_RANGE_END)
+			return end.ConnectsToNextSegment();
+		return instructionIndex >= start.GetInstructionIndexEx() && instructionIndex <= end.GetInstructionIndexEx();
+	}
+
+	bool ContainsEdge(const raInstructionEdge& edge) const
+	{
+		return edge >= start && edge <= end;
+	}
+
+	bool ContainsWholeInterval(const raInterval& other) const
+	{
+		return other.start >= start && other.end <= end;
+	}
+
+	bool IsOverlapping(const raInterval& other) const
+	{
+		return start <= other.end && end >= other.start;
+	}
+
+	sint32 GetPreciseDistance()
+	{
+		cemu_assert_debug(!start.ConnectsToNextSegment()); // how to handle this?
+		if(start == end)
+			return 1;
+		cemu_assert_debug(!end.ConnectsToPreviousSegment() && !end.ConnectsToNextSegment());
+		if(start.ConnectsToPreviousSegment())
+			return end.GetRaw() + 1;
+
+		return end.GetRaw() - start.GetRaw() + 1; // +1 because end is inclusive
+	}
+
+//private: not making these directly accessible only forces us to create loads of verbose getters and setters
+	raInstructionEdge start;
+	raInstructionEdge end;
+};
+
+struct raFixedRegRequirement
+{
+	raInstructionEdge pos;
+	IMLPhysRegisterSet allowedReg;
+};
+
+struct raLivenessRange
+{
+	IMLSegment* imlSegment;
+	raInterval interval;
+
+	// dirty state tracking
+	bool _noLoad;
+	bool hasStore;
+	bool hasStoreDelayed;
+	// next
+	raLivenessRange* subrangeBranchTaken;
+	raLivenessRange* subrangeBranchNotTaken;
+	// reverse counterpart of BranchTaken/BranchNotTaken
+	boost::container::small_vector<raLivenessRange*, 4> previousRanges;
+	// processing
+	uint32 lastIterationIndex;
+	// instruction read/write locations
+	std::vector<raAccessLocation> list_accessLocations;
+	// ordered list of all raInstructionEdge indices which require a fixed register
+	std::vector<raFixedRegRequirement> list_fixedRegRequirements;
+	// linked list (subranges with same GPR virtual register)
+	raLivenessSubrangeLink link_sameVirtualRegister;
+	// linked list (all subranges for this segment)
+	raLivenessSubrangeLink link_allSegmentRanges;
+	// register info
+	IMLRegID virtualRegister;
+	IMLName name;
+	// register allocator result
+	IMLPhysReg physicalRegister;
+
+	boost::container::small_vector<raLivenessRange*, 128> GetAllSubrangesInCluster();
+	bool GetAllowedRegistersEx(IMLPhysRegisterSet& allowedRegisters); // if the cluster has fixed register requirements in any instruction this returns the combined register mask. Otherwise returns false in which case allowedRegisters is left undefined
+	IMLPhysRegisterSet GetAllowedRegisters(IMLPhysRegisterSet regPool); // return regPool with fixed register requirements filtered out
+
+	IMLRegID GetVirtualRegister() const;
+	sint32 GetPhysicalRegister() const;
+	bool HasPhysicalRegister() const { return physicalRegister >= 0; }
+	IMLName GetName() const;
+	void SetPhysicalRegister(IMLPhysReg physicalRegister);
+	void SetPhysicalRegisterForCluster(IMLPhysReg physicalRegister);
+	void UnsetPhysicalRegister() { physicalRegister = -1; }
+
+  private:
+	void GetAllowedRegistersExRecursive(raLivenessRange* range, uint32 iterationIndex, IMLPhysRegisterSet& allowedRegs);
+};
+
+raLivenessRange* IMLRA_CreateRange(ppcImlGenContext_t* ppcImlGenContext, IMLSegment* imlSegment, IMLRegID virtualRegister, IMLName name, raInstructionEdge startPosition, raInstructionEdge endPosition);
+void IMLRA_DeleteRange(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange* subrange);
+void IMLRA_DeleteAllRanges(ppcImlGenContext_t* ppcImlGenContext);
+
+void IMLRA_ExplodeRangeCluster(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange* originRange);
+
+void IMLRA_MergeSubranges(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange* subrange, raLivenessRange* absorbedSubrange);
+
+raLivenessRange* IMLRA_SplitRange(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange*& subrange, raInstructionEdge splitPosition, bool trimToUsage = false);
+
+void PPCRecRA_debugValidateSubrange(raLivenessRange* subrange);
+
+// cost estimation
+sint32 IMLRA_GetSegmentReadWriteCost(IMLSegment* imlSegment);
+sint32 IMLRA_CalculateAdditionalCostOfRangeExplode(raLivenessRange* subrange);
+//sint32 PPCRecRARange_estimateAdditionalCostAfterSplit(raLivenessRange* subrange, sint32 splitIndex);
+sint32 IMLRA_CalculateAdditionalCostAfterSplit(raLivenessRange* subrange, raInstructionEdge splitPosition);

src/Cafe/HW/Espresso/Recompiler/IML/IMLSegment.cpp

@@ -0,0 +1,133 @@
+#include "IMLInstruction.h"
+#include "IMLSegment.h"
+
+void IMLSegment::SetEnterable(uint32 enterAddress)
+{
+	cemu_assert_debug(!isEnterable || enterPPCAddress == enterAddress);
+	isEnterable = true;
+	enterPPCAddress = enterAddress;
+}
+
+bool IMLSegment::HasSuffixInstruction() const
+{
+	if (imlList.empty())
+		return false;
+	const IMLInstruction& imlInstruction = imlList.back();
+	return imlInstruction.IsSuffixInstruction();
+}
+
+sint32 IMLSegment::GetSuffixInstructionIndex() const
+{
+	cemu_assert_debug(HasSuffixInstruction());
+	return (sint32)(imlList.size() - 1);
+}
+
+IMLInstruction* IMLSegment::GetLastInstruction()
+{
+	if (imlList.empty())
+		return nullptr;
+	return &imlList.back();
+}
+
+void IMLSegment::SetLinkBranchNotTaken(IMLSegment* imlSegmentDst)
+{
+	if (nextSegmentBranchNotTaken)
+		nextSegmentBranchNotTaken->list_prevSegments.erase(std::find(nextSegmentBranchNotTaken->list_prevSegments.begin(), nextSegmentBranchNotTaken->list_prevSegments.end(), this));
+	nextSegmentBranchNotTaken = imlSegmentDst;
+	if(imlSegmentDst)
+		imlSegmentDst->list_prevSegments.push_back(this);
+}
+
+void IMLSegment::SetLinkBranchTaken(IMLSegment* imlSegmentDst)
+{
+	if (nextSegmentBranchTaken)
+		nextSegmentBranchTaken->list_prevSegments.erase(std::find(nextSegmentBranchTaken->list_prevSegments.begin(), nextSegmentBranchTaken->list_prevSegments.end(), this));
+	nextSegmentBranchTaken = imlSegmentDst;
+	if (imlSegmentDst)
+		imlSegmentDst->list_prevSegments.push_back(this);
+}
+
+IMLInstruction* IMLSegment::AppendInstruction()
+{
+	IMLInstruction& inst = imlList.emplace_back();
+	memset(&inst, 0, sizeof(IMLInstruction));
+	return &inst;
+}
+
+void IMLSegment_SetLinkBranchNotTaken(IMLSegment* imlSegmentSrc, IMLSegment* imlSegmentDst)
+{
+	// make sure segments aren't already linked
+	if (imlSegmentSrc->nextSegmentBranchNotTaken == imlSegmentDst)
+		return;
+	// add as next segment for source
+	if (imlSegmentSrc->nextSegmentBranchNotTaken != nullptr)
+		assert_dbg();
+	imlSegmentSrc->nextSegmentBranchNotTaken = imlSegmentDst;
+	// add as previous segment for destination
+	imlSegmentDst->list_prevSegments.push_back(imlSegmentSrc);
+}
+
+void IMLSegment_SetLinkBranchTaken(IMLSegment* imlSegmentSrc, IMLSegment* imlSegmentDst)
+{
+	// make sure segments aren't already linked
+	if (imlSegmentSrc->nextSegmentBranchTaken == imlSegmentDst)
+		return;
+	// add as next segment for source
+	if (imlSegmentSrc->nextSegmentBranchTaken != nullptr)
+		assert_dbg();
+	imlSegmentSrc->nextSegmentBranchTaken = imlSegmentDst;
+	// add as previous segment for destination
+	imlSegmentDst->list_prevSegments.push_back(imlSegmentSrc);
+}
+
+void IMLSegment_RemoveLink(IMLSegment* imlSegmentSrc, IMLSegment* imlSegmentDst)
+{
+	if (imlSegmentSrc->nextSegmentBranchNotTaken == imlSegmentDst)
+	{
+		imlSegmentSrc->nextSegmentBranchNotTaken = nullptr;
+	}
+	else if (imlSegmentSrc->nextSegmentBranchTaken == imlSegmentDst)
+	{
+		imlSegmentSrc->nextSegmentBranchTaken = nullptr;
+	}
+	else
+		assert_dbg();
+
+	bool matchFound = false;
+	for (sint32 i = 0; i < imlSegmentDst->list_prevSegments.size(); i++)
+	{
+		if (imlSegmentDst->list_prevSegments[i] == imlSegmentSrc)
+		{
+			imlSegmentDst->list_prevSegments.erase(imlSegmentDst->list_prevSegments.begin() + i);
+			matchFound = true;
+			break;
+		}
+	}
+	if (matchFound == false)
+		assert_dbg();
+}
+
+/*
+ * Replaces all links to segment orig with linkts to segment new
+ */
+void IMLSegment_RelinkInputSegment(IMLSegment* imlSegmentOrig, IMLSegment* imlSegmentNew)
+{
+	while (imlSegmentOrig->list_prevSegments.size() != 0)
+	{
+		IMLSegment* prevSegment = imlSegmentOrig->list_prevSegments[0];
+		if (prevSegment->nextSegmentBranchNotTaken == imlSegmentOrig)
+		{
+			IMLSegment_RemoveLink(prevSegment, imlSegmentOrig);
+			IMLSegment_SetLinkBranchNotTaken(prevSegment, imlSegmentNew);
+		}
+		else if (prevSegment->nextSegmentBranchTaken == imlSegmentOrig)
+		{
+			IMLSegment_RemoveLink(prevSegment, imlSegmentOrig);
+			IMLSegment_SetLinkBranchTaken(prevSegment, imlSegmentNew);
+		}
+		else
+		{
+			assert_dbg();
+		}
+	}
+}

src/Cafe/HW/Espresso/Recompiler/IML/IMLSegment.h

@@ -0,0 +1,193 @@
+#pragma once
+#include "IMLInstruction.h"
+
+#include <boost/container/small_vector.hpp>
+
+// special values to mark the index of ranges that reach across the segment border
+#define RA_INTER_RANGE_START	(-1)
+#define RA_INTER_RANGE_END		(0x70000000)
+
+struct IMLSegmentPoint
+{
+	friend struct IMLSegmentInterval;
+
+	sint32 index;
+	struct IMLSegment* imlSegment; // do we really need to track this? SegmentPoints are always accessed via the segment that they are part of
+	IMLSegmentPoint* next;
+	IMLSegmentPoint* prev;
+
+	// the index is the instruction index times two.
+	// this gives us the ability to cover half an instruction with RA ranges
+	// covering only the first half of an instruction (0-0) means that the register is read, but not preserved
+	// covering first and the second half means the register is read and preserved
+	// covering only the second half means the register is written but not read
+
+	sint32 GetInstructionIndex() const
+	{
+		return index;
+	}
+
+	void SetInstructionIndex(sint32 index)
+	{
+		this->index = index;
+	}
+
+	void ShiftIfAfter(sint32 instructionIndex, sint32 shiftCount)
+	{
+		if (!IsPreviousSegment() && !IsNextSegment())
+		{
+			if (GetInstructionIndex() >= instructionIndex)
+				index += shiftCount;
+		}
+	}
+
+	void DecrementByOneInstruction()
+	{
+		index--;
+	}
+
+	// the segment point can point beyond the first and last instruction which indicates that it is an infinite range reaching up to the previous or next segment
+	bool IsPreviousSegment() const { return index == RA_INTER_RANGE_START; }
+	bool IsNextSegment() const { return index == RA_INTER_RANGE_END; }
+
+	// overload operand > and <
+	bool operator>(const IMLSegmentPoint& other) const { return index > other.index; }
+	bool operator<(const IMLSegmentPoint& other) const { return index < other.index; }
+	bool operator==(const IMLSegmentPoint& other) const { return index == other.index; }
+	bool operator!=(const IMLSegmentPoint& other) const { return index != other.index; }
+
+	// overload comparison operands for sint32
+	bool operator>(const sint32 other) const { return index > other; }
+	bool operator<(const sint32 other) const { return index < other; }
+	bool operator<=(const sint32 other) const { return index <= other; }
+	bool operator>=(const sint32 other) const { return index >= other; }
+};
+
+struct IMLSegmentInterval
+{
+	IMLSegmentPoint start;
+	IMLSegmentPoint end;
+
+	bool ContainsInstructionIndex(sint32 offset) const { return start <= offset && end > offset; }
+
+	bool IsRangeOverlapping(const IMLSegmentInterval& other)
+	{
+		// todo - compare the raw index
+		sint32 r1start = this->start.GetInstructionIndex();
+		sint32 r1end = this->end.GetInstructionIndex();
+		sint32 r2start = other.start.GetInstructionIndex();
+		sint32 r2end = other.end.GetInstructionIndex();
+		if (r1start < r2end && r1end > r2start)
+			return true;
+		if (this->start.IsPreviousSegment() && r1start == r2start)
+			return true;
+		if (this->end.IsNextSegment() && r1end == r2end)
+			return true;
+		return false;
+	}
+
+	bool ExtendsIntoPreviousSegment() const
+	{
+		return start.IsPreviousSegment();
+	}
+
+	bool ExtendsIntoNextSegment() const
+	{
+		return end.IsNextSegment();
+	}
+
+	bool IsNextSegmentOnly() const
+	{
+		if(!start.IsNextSegment())
+			return false;
+		cemu_assert_debug(end.IsNextSegment());
+		return true;
+	}
+
+	bool IsPreviousSegmentOnly() const
+	{
+		if (!end.IsPreviousSegment())
+			return false;
+		cemu_assert_debug(start.IsPreviousSegment());
+		return true;
+	}
+
+	sint32 GetDistance() const
+	{
+		// todo - assert if either start or end is outside the segment
+		// we may also want to switch this to raw indices?
+		return end.GetInstructionIndex() - start.GetInstructionIndex();
+	}
+};
+
+struct PPCSegmentRegisterAllocatorInfo_t
+{
+	// used during loop detection
+	bool isPartOfProcessedLoop{}; 
+	sint32 lastIterationIndex{};
+	// linked lists
+	struct raLivenessRange* linkedList_allSubranges{};
+	std::unordered_map<IMLRegID, struct raLivenessRange*> linkedList_perVirtualRegister;
+};
+
+struct IMLSegment
+{
+	sint32 momentaryIndex{}; // index in segment list, generally not kept up to date except if needed (necessary for loop detection)
+	sint32 loopDepth{};
+	uint32 ppcAddress{}; // ppc address (0xFFFFFFFF if not associated with an address)
+	uint32 x64Offset{}; // x64 code offset of segment start
+	// list of intermediate instructions in this segment
+	std::vector<IMLInstruction> imlList;
+	// segment link
+	IMLSegment* nextSegmentBranchNotTaken{}; // this is also the default for segments where there is no branch
+	IMLSegment* nextSegmentBranchTaken{};
+	bool nextSegmentIsUncertain{};
+	std::vector<IMLSegment*> list_prevSegments{};
+	// source for overwrite analysis (if nextSegmentIsUncertain is true)
+	// sometimes a segment is marked as an exit point, but for the purposes of dead code elimination we know the next segment
+	IMLSegment* deadCodeEliminationHintSeg{};
+	std::vector<IMLSegment*> list_deadCodeHintBy{};
+	// enterable segments
+	bool isEnterable{}; // this segment can be entered from outside the recompiler (no preloaded registers necessary)
+	uint32 enterPPCAddress{}; // used if isEnterable is true
+	// register allocator info
+	PPCSegmentRegisterAllocatorInfo_t raInfo{};
+	// segment state API
+	void SetEnterable(uint32 enterAddress);
+	void SetLinkBranchNotTaken(IMLSegment* imlSegmentDst);
+	void SetLinkBranchTaken(IMLSegment* imlSegmentDst);
+
+	IMLSegment* GetBranchTaken()
+	{
+		return nextSegmentBranchTaken;
+	}
+
+	IMLSegment* GetBranchNotTaken()
+	{
+		return nextSegmentBranchNotTaken;
+	}
+
+	void SetNextSegmentForOverwriteHints(IMLSegment* seg)
+	{
+		cemu_assert_debug(!deadCodeEliminationHintSeg);
+		deadCodeEliminationHintSeg = seg;
+		if (seg)
+			seg->list_deadCodeHintBy.push_back(this);
+	}
+
+	// instruction API
+	IMLInstruction* AppendInstruction();
+
+	bool HasSuffixInstruction() const;
+	sint32 GetSuffixInstructionIndex() const;
+	IMLInstruction* GetLastInstruction();
+
+	// segment points
+	IMLSegmentPoint* segmentPointList{};
+};
+
+
+void IMLSegment_SetLinkBranchNotTaken(IMLSegment* imlSegmentSrc, IMLSegment* imlSegmentDst);
+void IMLSegment_SetLinkBranchTaken(IMLSegment* imlSegmentSrc, IMLSegment* imlSegmentDst);
+void IMLSegment_RelinkInputSegment(IMLSegment* imlSegmentOrig, IMLSegment* imlSegmentNew);
+void IMLSegment_RemoveLink(IMLSegment* imlSegmentSrc, IMLSegment* imlSegmentDst);

src/Cafe/HW/Espresso/Recompiler/PPCFunctionBoundaryTracker.h

@@ -21,6 +21,16 @@ public:
 	};
 
 public:
+	~PPCFunctionBoundaryTracker()
+	{
+		while (!map_ranges.empty())
+		{
+			PPCRange_t* range = *map_ranges.begin();
+			delete range;
+			map_ranges.erase(map_ranges.begin());
+		}
+	}
+
 	void trackStartPoint(MPTR startAddress)
 	{
 		processRange(startAddress, nullptr, nullptr);
@@ -40,10 +50,34 @@ public:
 		return false;
 	}
 
+	std::vector<PPCRange_t> GetRanges()
+	{
+		std::vector<PPCRange_t> r;
+		for (auto& it : map_ranges)
+			r.emplace_back(*it);
+		return r;
+	}
+
+	bool ContainsAddress(uint32 addr) const
+	{
+		for (auto& it : map_ranges)
+		{
+			if (addr >= it->startAddress && addr < it->getEndAddress())
+				return true;
+		}
+		return false;
+	}
+
+	const std::set<uint32>& GetBranchTargets() const
+	{
+		return map_branchTargetsAll;
+	}
+
 private:
 	void addBranchDestination(PPCRange_t* sourceRange, MPTR address)
 	{
-		map_branchTargets.emplace(address);
+		map_queuedBranchTargets.emplace(address);
+		map_branchTargetsAll.emplace(address);
 	}
 
 	// process flow of instruction
@@ -114,7 +148,7 @@ private:
 				Espresso::BOField BO;
 				uint32 BI;
 				bool LK;
-				Espresso::decodeOp_BCLR(opcode, BO, BI, LK);
+				Espresso::decodeOp_BCSPR(opcode, BO, BI, LK);
 				if (BO.branchAlways() && !LK)
 				{
 					// unconditional BLR
@@ -218,7 +252,7 @@ private:
 		auto rangeItr = map_ranges.begin();
 
 		PPCRange_t* previousRange = nullptr;
-		for (std::set<uint32_t>::const_iterator targetItr = map_branchTargets.begin() ; targetItr != map_branchTargets.end(); )
+		for (std::set<uint32_t>::const_iterator targetItr = map_queuedBranchTargets.begin() ; targetItr != map_queuedBranchTargets.end(); )
 		{
 			while (rangeItr != map_ranges.end() && ((*rangeItr)->startAddress + (*rangeItr)->length) <= (*targetItr))
 			{
@@ -239,7 +273,7 @@ private:
 				(*targetItr) < ((*rangeItr)->startAddress + (*rangeItr)->length))
 			{
 				// delete visited targets
-				targetItr = map_branchTargets.erase(targetItr);
+				targetItr = map_queuedBranchTargets.erase(targetItr);
 				continue;
 			}
 
@@ -289,5 +323,6 @@ private:
 	};
 
 	std::set<PPCRange_t*, RangePtrCmp> map_ranges;
-	std::set<uint32> map_branchTargets;
+	std::set<uint32> map_queuedBranchTargets;
+	std::set<uint32> map_branchTargetsAll;
 };

src/Cafe/HW/Espresso/Recompiler/PPCRecompiler.cpp

@@ -2,7 +2,6 @@
 #include "PPCFunctionBoundaryTracker.h"
 #include "PPCRecompiler.h"
 #include "PPCRecompilerIml.h"
-#include "PPCRecompilerX64.h"
 #include "Cafe/OS/RPL/rpl.h"
 #include "util/containers/RangeStore.h"
 #include "Cafe/OS/libs/coreinit/coreinit_CodeGen.h"
@@ -14,6 +13,14 @@
 #include "util/helpers/helpers.h"
 #include "util/MemMapper/MemMapper.h"
 
+#include "IML/IML.h"
+#include "IML/IMLRegisterAllocator.h"
+#include "BackendX64/BackendX64.h"
+#include "util/highresolutiontimer/HighResolutionTimer.h"
+
+#define PPCREC_FORCE_SYNCHRONOUS_COMPILATION	0 // if 1, then function recompilation will block and execute on the thread that called PPCRecompiler_visitAddressNoBlock
+#define PPCREC_LOG_RECOMPILATION_RESULTS		0
+
 struct PPCInvalidationRange
 {
 	MPTR startAddress;
@@ -37,11 +44,36 @@ void ATTR_MS_ABI (*PPCRecompiler_leaveRecompilerCode_unvisited)();
 
 PPCRecompilerInstanceData_t* ppcRecompilerInstanceData;
 
+#if PPCREC_FORCE_SYNCHRONOUS_COMPILATION
+static std::mutex s_singleRecompilationMutex;
+#endif
+
 bool ppcRecompilerEnabled = false;
 
+void PPCRecompiler_recompileAtAddress(uint32 address);
+
 // this function does never block and can fail if the recompiler lock cannot be acquired immediately
 void PPCRecompiler_visitAddressNoBlock(uint32 enterAddress)
 {
+#if PPCREC_FORCE_SYNCHRONOUS_COMPILATION
+	if (ppcRecompilerInstanceData->ppcRecompilerDirectJumpTable[enterAddress / 4] != PPCRecompiler_leaveRecompilerCode_unvisited)
+		return;
+	PPCRecompilerState.recompilerSpinlock.lock();
+	if (ppcRecompilerInstanceData->ppcRecompilerDirectJumpTable[enterAddress / 4] != PPCRecompiler_leaveRecompilerCode_unvisited)
+	{
+		PPCRecompilerState.recompilerSpinlock.unlock();
+		return;
+	}
+	ppcRecompilerInstanceData->ppcRecompilerDirectJumpTable[enterAddress / 4] = PPCRecompiler_leaveRecompilerCode_visited;
+	PPCRecompilerState.recompilerSpinlock.unlock();
+	s_singleRecompilationMutex.lock();
+	if (ppcRecompilerInstanceData->ppcRecompilerDirectJumpTable[enterAddress / 4] == PPCRecompiler_leaveRecompilerCode_visited)
+	{
+		PPCRecompiler_recompileAtAddress(enterAddress);
+	}
+	s_singleRecompilationMutex.unlock();
+	return;
+#endif
 	// quick read-only check without lock
 	if (ppcRecompilerInstanceData->ppcRecompilerDirectJumpTable[enterAddress / 4] != PPCRecompiler_leaveRecompilerCode_unvisited)
 		return;
@@ -127,15 +159,15 @@ void PPCRecompiler_attemptEnter(PPCInterpreter_t* hCPU, uint32 enterAddress)
 		PPCRecompiler_enter(hCPU, funcPtr);
 	}
 }
+bool PPCRecompiler_ApplyIMLPasses(ppcImlGenContext_t& ppcImlGenContext);
 
-PPCRecFunction_t* PPCRecompiler_recompileFunction(PPCFunctionBoundaryTracker::PPCRange_t range, std::set<uint32>& entryAddresses, std::vector<std::pair<MPTR, uint32>>& entryPointsOut)
+PPCRecFunction_t* PPCRecompiler_recompileFunction(PPCFunctionBoundaryTracker::PPCRange_t range, std::set<uint32>& entryAddresses, std::vector<std::pair<MPTR, uint32>>& entryPointsOut, PPCFunctionBoundaryTracker& boundaryTracker)
 {
 	if (range.startAddress >= PPC_REC_CODE_AREA_END)
 	{
 		cemuLog_log(LogType::Force, "Attempting to recompile function outside of allowed code area");
 		return nullptr;
 	}
-
 	uint32 codeGenRangeStart;
 	uint32 codeGenRangeSize = 0;
 	coreinit::OSGetCodegenVirtAddrRangeInternal(codeGenRangeStart, codeGenRangeSize);
@@ -153,29 +185,61 @@ PPCRecFunction_t* PPCRecompiler_recompileFunction(PPCFunctionBoundaryTracker::PP
 	PPCRecFunction_t* ppcRecFunc = new PPCRecFunction_t();
 	ppcRecFunc->ppcAddress = range.startAddress;
 	ppcRecFunc->ppcSize = range.length;
+
+#if PPCREC_LOG_RECOMPILATION_RESULTS
+	BenchmarkTimer bt;
+	bt.Start();
+#endif
+
 	// generate intermediate code
 	ppcImlGenContext_t ppcImlGenContext = { 0 };
-	bool compiledSuccessfully = PPCRecompiler_generateIntermediateCode(ppcImlGenContext, ppcRecFunc, entryAddresses);
+	ppcImlGenContext.debug_entryPPCAddress = range.startAddress;
+	bool compiledSuccessfully = PPCRecompiler_generateIntermediateCode(ppcImlGenContext, ppcRecFunc, entryAddresses, boundaryTracker);
 	if (compiledSuccessfully == false)
 	{
-		// todo: Free everything
-		PPCRecompiler_freeContext(&ppcImlGenContext);
 		delete ppcRecFunc;
-		return NULL;
+		return nullptr;
 	}
+
+	uint32 ppcRecLowerAddr = LaunchSettings::GetPPCRecLowerAddr();
+	uint32 ppcRecUpperAddr = LaunchSettings::GetPPCRecUpperAddr();
+
+	if (ppcRecLowerAddr != 0 && ppcRecUpperAddr != 0)
+	{
+		if (ppcRecFunc->ppcAddress < ppcRecLowerAddr || ppcRecFunc->ppcAddress > ppcRecUpperAddr)
+		{
+			delete ppcRecFunc;
+			return nullptr;
+		}
+	}
+
+	// apply passes
+	if (!PPCRecompiler_ApplyIMLPasses(ppcImlGenContext))
+	{
+		delete ppcRecFunc;
+		return nullptr;
+	}
+
 	// emit x64 code
 	bool x64GenerationSuccess = PPCRecompiler_generateX64Code(ppcRecFunc, &ppcImlGenContext);
 	if (x64GenerationSuccess == false)
 	{
-		PPCRecompiler_freeContext(&ppcImlGenContext);
 		return nullptr;
 	}
+	if (ActiveSettings::DumpRecompilerFunctionsEnabled())
+	{
+		FileStream* fs = FileStream::createFile2(ActiveSettings::GetUserDataPath(fmt::format("dump/recompiler/ppc_{:08x}.bin", ppcRecFunc->ppcAddress)));
+		if (fs)
+		{
+			fs->writeData(ppcRecFunc->x86Code, ppcRecFunc->x86Size);
+			delete fs;
+		}
+	}
 
 	// collect list of PPC-->x64 entry points
 	entryPointsOut.clear();
-	for (sint32 s = 0; s < ppcImlGenContext.segmentListCount; s++)
+	for(IMLSegment* imlSegment : ppcImlGenContext.segmentList2)
 	{
-		PPCRecImlSegment_t* imlSegment = ppcImlGenContext.segmentList[s];
 		if (imlSegment->isEnterable == false)
 			continue;
 
@@ -185,10 +249,83 @@ PPCRecFunction_t* PPCRecompiler_recompileFunction(PPCFunctionBoundaryTracker::PP
 		entryPointsOut.emplace_back(ppcEnterOffset, x64Offset);
 	}
 
-	PPCRecompiler_freeContext(&ppcImlGenContext);
+#if PPCREC_LOG_RECOMPILATION_RESULTS
+	bt.Stop();
+	uint32 codeHash = 0;
+	for (uint32 i = 0; i < ppcRecFunc->x86Size; i++)
+	{
+		codeHash = _rotr(codeHash, 3);
+		codeHash += ((uint8*)ppcRecFunc->x86Code)[i];
+	}
+	cemuLog_log(LogType::Force, "[Recompiler] PPC 0x{:08x} -> x64: 0x{:x} Took {:.4}ms | Size {:04x} CodeHash {:08x}", (uint32)ppcRecFunc->ppcAddress, (uint64)(uintptr_t)ppcRecFunc->x86Code, bt.GetElapsedMilliseconds(), ppcRecFunc->x86Size, codeHash);
+#endif
+
 	return ppcRecFunc;
 }
 
+void PPCRecompiler_NativeRegisterAllocatorPass(ppcImlGenContext_t& ppcImlGenContext)
+{
+	IMLRegisterAllocatorParameters raParam;
+
+	for (auto& it : ppcImlGenContext.mappedRegs)
+		raParam.regIdToName.try_emplace(it.second.GetRegID(), it.first);
+
+	auto& gprPhysPool = raParam.GetPhysRegPool(IMLRegFormat::I64);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_RAX);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_RDX);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_RBX);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_RBP);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_RSI);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_RDI);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_R8);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_R9);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_R10);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_R11);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_R12);
+	gprPhysPool.SetAvailable(IMLArchX86::PHYSREG_GPR_BASE + X86_REG_RCX);
+
+	// add XMM registers, except XMM15 which is the temporary register
+	auto& fprPhysPool = raParam.GetPhysRegPool(IMLRegFormat::F64);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 0);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 1);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 2);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 3);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 4);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 5);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 6);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 7);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 8);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 9);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 10);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 11);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 12);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 13);
+	fprPhysPool.SetAvailable(IMLArchX86::PHYSREG_FPR_BASE + 14);
+
+	IMLRegisterAllocator_AllocateRegisters(&ppcImlGenContext, raParam);
+}
+
+bool PPCRecompiler_ApplyIMLPasses(ppcImlGenContext_t& ppcImlGenContext)
+{
+	// isolate entry points from function flow (enterable segments must not be the target of any other segment)
+	// this simplifies logic during register allocation
+	PPCRecompilerIML_isolateEnterableSegments(&ppcImlGenContext);
+
+	// if GQRs can be predicted, optimize PSQ load/stores
+	PPCRecompiler_optimizePSQLoadAndStore(&ppcImlGenContext);
+
+	// merge certain float load+store patterns (must happen before FPR register remapping)
+	IMLOptimizer_OptimizeDirectFloatCopies(&ppcImlGenContext);
+	// delay byte swapping for certain load+store patterns
+	IMLOptimizer_OptimizeDirectIntegerCopies(&ppcImlGenContext);
+
+	IMLOptimizer_StandardOptimizationPass(ppcImlGenContext);
+
+	PPCRecompiler_NativeRegisterAllocatorPass(ppcImlGenContext);
+
+	return true;
+}
+
 bool PPCRecompiler_makeRecompiledFunctionActive(uint32 initialEntryPoint, PPCFunctionBoundaryTracker::PPCRange_t& range, PPCRecFunction_t* ppcRecFunc, std::vector<std::pair<MPTR, uint32>>& entryPoints)
 {
 	// update jump table
@@ -202,7 +339,7 @@ bool PPCRecompiler_makeRecompiledFunctionActive(uint32 initialEntryPoint, PPCFun
 		return false;
 	}
 
-	// check if the current range got invalidated in the time it took to recompile it
+	// check if the current range got invalidated during the time it took to recompile it
 	bool isInvalidated = false;
 	for (auto& invRange : PPCRecompilerState.invalidationRanges)
 	{
@@ -280,7 +417,7 @@ void PPCRecompiler_recompileAtAddress(uint32 address)
 	PPCRecompilerState.recompilerSpinlock.unlock();
 
 	std::vector<std::pair<MPTR, uint32>> functionEntryPoints;
-	auto func = PPCRecompiler_recompileFunction(range, entryAddresses, functionEntryPoints);
+	auto func = PPCRecompiler_recompileFunction(range, entryAddresses, functionEntryPoints, funcBoundaries);
 
 	if (!func)
 	{
@@ -295,6 +432,10 @@ std::atomic_bool s_recompilerThreadStopSignal{false};
 void PPCRecompiler_thread()
 {
 	SetThreadName("PPCRecompiler");
+#if PPCREC_FORCE_SYNCHRONOUS_COMPILATION
+	return;
+#endif
+
 	while (true)
 	{
         if(s_recompilerThreadStopSignal)
@@ -475,6 +616,41 @@ void PPCRecompiler_invalidateRange(uint32 startAddr, uint32 endAddr)
 #if defined(ARCH_X86_64)
 void PPCRecompiler_initPlatform()
 {
+	ppcRecompilerInstanceData->_x64XMM_xorNegateMaskBottom[0] = 1ULL << 63ULL;
+	ppcRecompilerInstanceData->_x64XMM_xorNegateMaskBottom[1] = 0ULL;
+	ppcRecompilerInstanceData->_x64XMM_xorNegateMaskPair[0] = 1ULL << 63ULL;
+	ppcRecompilerInstanceData->_x64XMM_xorNegateMaskPair[1] = 1ULL << 63ULL;
+	ppcRecompilerInstanceData->_x64XMM_xorNOTMask[0] = 0xFFFFFFFFFFFFFFFFULL;
+	ppcRecompilerInstanceData->_x64XMM_xorNOTMask[1] = 0xFFFFFFFFFFFFFFFFULL;
+	ppcRecompilerInstanceData->_x64XMM_andAbsMaskBottom[0] = ~(1ULL << 63ULL);
+	ppcRecompilerInstanceData->_x64XMM_andAbsMaskBottom[1] = ~0ULL;
+	ppcRecompilerInstanceData->_x64XMM_andAbsMaskPair[0] = ~(1ULL << 63ULL);
+	ppcRecompilerInstanceData->_x64XMM_andAbsMaskPair[1] = ~(1ULL << 63ULL);
+	ppcRecompilerInstanceData->_x64XMM_andFloatAbsMaskBottom[0] = ~(1 << 31);
+	ppcRecompilerInstanceData->_x64XMM_andFloatAbsMaskBottom[1] = 0xFFFFFFFF;
+	ppcRecompilerInstanceData->_x64XMM_andFloatAbsMaskBottom[2] = 0xFFFFFFFF;
+	ppcRecompilerInstanceData->_x64XMM_andFloatAbsMaskBottom[3] = 0xFFFFFFFF;
+	ppcRecompilerInstanceData->_x64XMM_singleWordMask[0] = 0xFFFFFFFFULL;
+	ppcRecompilerInstanceData->_x64XMM_singleWordMask[1] = 0ULL;
+	ppcRecompilerInstanceData->_x64XMM_constDouble1_1[0] = 1.0;
+	ppcRecompilerInstanceData->_x64XMM_constDouble1_1[1] = 1.0;
+	ppcRecompilerInstanceData->_x64XMM_constDouble0_0[0] = 0.0;
+	ppcRecompilerInstanceData->_x64XMM_constDouble0_0[1] = 0.0;
+	ppcRecompilerInstanceData->_x64XMM_constFloat0_0[0] = 0.0f;
+	ppcRecompilerInstanceData->_x64XMM_constFloat0_0[1] = 0.0f;
+	ppcRecompilerInstanceData->_x64XMM_constFloat1_1[0] = 1.0f;
+	ppcRecompilerInstanceData->_x64XMM_constFloat1_1[1] = 1.0f;
+	*(uint32*)&ppcRecompilerInstanceData->_x64XMM_constFloatMin[0] = 0x00800000;
+	*(uint32*)&ppcRecompilerInstanceData->_x64XMM_constFloatMin[1] = 0x00800000;
+	ppcRecompilerInstanceData->_x64XMM_flushDenormalMask1[0] = 0x7F800000;
+	ppcRecompilerInstanceData->_x64XMM_flushDenormalMask1[1] = 0x7F800000;
+	ppcRecompilerInstanceData->_x64XMM_flushDenormalMask1[2] = 0x7F800000;
+	ppcRecompilerInstanceData->_x64XMM_flushDenormalMask1[3] = 0x7F800000;
+	ppcRecompilerInstanceData->_x64XMM_flushDenormalMaskResetSignBits[0] = ~0x80000000;
+	ppcRecompilerInstanceData->_x64XMM_flushDenormalMaskResetSignBits[1] = ~0x80000000;
+	ppcRecompilerInstanceData->_x64XMM_flushDenormalMaskResetSignBits[2] = ~0x80000000;
+	ppcRecompilerInstanceData->_x64XMM_flushDenormalMaskResetSignBits[3] = ~0x80000000;
+
 	// mxcsr
 	ppcRecompilerInstanceData->_x64XMM_mxCsr_ftzOn = 0x1F80 | 0x8000;
 	ppcRecompilerInstanceData->_x64XMM_mxCsr_ftzOff = 0x1F80;
@@ -512,42 +688,6 @@ void PPCRecompiler_init()
     PPCRecompiler_allocateRange(mmuRange_TRAMPOLINE_AREA.getBase(), mmuRange_TRAMPOLINE_AREA.getSize());
     PPCRecompiler_allocateRange(mmuRange_CODECAVE.getBase(), mmuRange_CODECAVE.getSize());
 
-	// init x64 recompiler instance data
-	ppcRecompilerInstanceData->_x64XMM_xorNegateMaskBottom[0] = 1ULL << 63ULL;
-	ppcRecompilerInstanceData->_x64XMM_xorNegateMaskBottom[1] = 0ULL;
-	ppcRecompilerInstanceData->_x64XMM_xorNegateMaskPair[0] = 1ULL << 63ULL;
-	ppcRecompilerInstanceData->_x64XMM_xorNegateMaskPair[1] = 1ULL << 63ULL;
-	ppcRecompilerInstanceData->_x64XMM_xorNOTMask[0] = 0xFFFFFFFFFFFFFFFFULL;
-	ppcRecompilerInstanceData->_x64XMM_xorNOTMask[1] = 0xFFFFFFFFFFFFFFFFULL;
-	ppcRecompilerInstanceData->_x64XMM_andAbsMaskBottom[0] = ~(1ULL << 63ULL);
-	ppcRecompilerInstanceData->_x64XMM_andAbsMaskBottom[1] = ~0ULL;
-	ppcRecompilerInstanceData->_x64XMM_andAbsMaskPair[0] = ~(1ULL << 63ULL);
-	ppcRecompilerInstanceData->_x64XMM_andAbsMaskPair[1] = ~(1ULL << 63ULL);
-	ppcRecompilerInstanceData->_x64XMM_andFloatAbsMaskBottom[0] = ~(1 << 31);
-	ppcRecompilerInstanceData->_x64XMM_andFloatAbsMaskBottom[1] = 0xFFFFFFFF;
-	ppcRecompilerInstanceData->_x64XMM_andFloatAbsMaskBottom[2] = 0xFFFFFFFF;
-	ppcRecompilerInstanceData->_x64XMM_andFloatAbsMaskBottom[3] = 0xFFFFFFFF;
-	ppcRecompilerInstanceData->_x64XMM_singleWordMask[0] = 0xFFFFFFFFULL;
-	ppcRecompilerInstanceData->_x64XMM_singleWordMask[1] = 0ULL;
-	ppcRecompilerInstanceData->_x64XMM_constDouble1_1[0] = 1.0;
-	ppcRecompilerInstanceData->_x64XMM_constDouble1_1[1] = 1.0;
-	ppcRecompilerInstanceData->_x64XMM_constDouble0_0[0] = 0.0;
-	ppcRecompilerInstanceData->_x64XMM_constDouble0_0[1] = 0.0;
-	ppcRecompilerInstanceData->_x64XMM_constFloat0_0[0] = 0.0f;
-	ppcRecompilerInstanceData->_x64XMM_constFloat0_0[1] = 0.0f;
-	ppcRecompilerInstanceData->_x64XMM_constFloat1_1[0] = 1.0f;
-	ppcRecompilerInstanceData->_x64XMM_constFloat1_1[1] = 1.0f;
-	*(uint32*)&ppcRecompilerInstanceData->_x64XMM_constFloatMin[0] = 0x00800000;
-	*(uint32*)&ppcRecompilerInstanceData->_x64XMM_constFloatMin[1] = 0x00800000;
-	ppcRecompilerInstanceData->_x64XMM_flushDenormalMask1[0] = 0x7F800000;
-	ppcRecompilerInstanceData->_x64XMM_flushDenormalMask1[1] = 0x7F800000;
-	ppcRecompilerInstanceData->_x64XMM_flushDenormalMask1[2] = 0x7F800000;
-	ppcRecompilerInstanceData->_x64XMM_flushDenormalMask1[3] = 0x7F800000;
-	ppcRecompilerInstanceData->_x64XMM_flushDenormalMaskResetSignBits[0] = ~0x80000000;
-	ppcRecompilerInstanceData->_x64XMM_flushDenormalMaskResetSignBits[1] = ~0x80000000;
-	ppcRecompilerInstanceData->_x64XMM_flushDenormalMaskResetSignBits[2] = ~0x80000000;
-	ppcRecompilerInstanceData->_x64XMM_flushDenormalMaskResetSignBits[3] = ~0x80000000;
-
 	// setup GQR scale tables
 
 	for (uint32 i = 0; i < 32; i++)
@@ -623,4 +763,4 @@ void PPCRecompiler_Shutdown()
         // mark as unmapped
         ppcRecompiler_reservedBlockMask[i] = false;
     }
-}
+}

src/Cafe/HW/Espresso/Recompiler/PPCRecompiler.h

@@ -1,4 +1,4 @@
-#include <vector>
+#pragma once
 
 #define PPC_REC_CODE_AREA_START		(0x00000000) // lower bound of executable memory area. Recompiler expects this address to be 0
 #define PPC_REC_CODE_AREA_END		(0x10000000) // upper bound of executable memory area
@@ -6,336 +6,113 @@
 
 #define PPC_REC_ALIGN_TO_4MB(__v)	(((__v)+4*1024*1024-1)&~(4*1024*1024-1))
 
-#define PPC_REC_MAX_VIRTUAL_GPR		(40) // enough to store 32 GPRs + a few SPRs + temp registers (usually only 1-2)
+#define PPC_REC_MAX_VIRTUAL_GPR		(40 + 32) // enough to store 32 GPRs + a few SPRs + temp registers (usually only 1-2)
 
-typedef struct  
+struct ppcRecRange_t
 {
 	uint32 ppcAddress;
 	uint32 ppcSize;
-	//void* x86Start;
-	//size_t x86Size;
 	void* storedRange;
-}ppcRecRange_t;
+};
 
-typedef struct  
+struct PPCRecFunction_t
 {
 	uint32 ppcAddress;
 	uint32 ppcSize; // ppc code size of function
 	void*  x86Code; // pointer to x86 code
 	size_t x86Size;
 	std::vector<ppcRecRange_t> list_ranges;
-}PPCRecFunction_t;
-
-#define PPCREC_IML_OP_FLAG_SIGNEXTEND			(1<<0)
-#define PPCREC_IML_OP_FLAG_SWITCHENDIAN			(1<<1)
-#define PPCREC_IML_OP_FLAG_NOT_EXPANDED			(1<<2) // set single-precision load instructions to indicate that the value should not be rounded to double-precision 
-#define PPCREC_IML_OP_FLAG_UNUSED				(1<<7) // used to mark instructions that are not used
-
-typedef struct  
-{
-	uint8 type;
-	uint8 operation;
-	uint8 crRegister; // set to 0xFF if not set, not all IML instruction types support cr.
-	uint8 crMode; // only used when crRegister is valid, used to differentiate between various forms of condition flag set/clear behavior
-	uint32 crIgnoreMask; // bit set for every respective CR bit that doesn't need to be updated
-	uint32 associatedPPCAddress; // ppc address that is associated with this instruction
-	union
-	{
-		struct  
-		{
-			uint8 _padding[7];
-		}padding;
-		struct  
-		{
-			// R (op) A [update cr* in mode *]
-			uint8 registerResult;
-			uint8 registerA;
-		}op_r_r;
-		struct  
-		{
-			// R = A (op) B [update cr* in mode *]
-			uint8 registerResult;
-			uint8 registerA;
-			uint8 registerB;
-		}op_r_r_r;
-		struct  
-		{
-			// R = A (op) immS32 [update cr* in mode *]
-			uint8 registerResult;
-			uint8 registerA;
-			sint32 immS32;
-		}op_r_r_s32;
-		struct  
-		{
-			// R/F = NAME or NAME = R/F
-			uint8 registerIndex;
-			uint8 copyWidth;
-			uint32 name;
-			uint8 flags;
-		}op_r_name;
-		struct  
-		{
-			// R (op) s32 [update cr* in mode *]
-			uint8 registerIndex;
-			sint32 immS32;
-		}op_r_immS32;
-		struct  
-		{
-			uint32 address;
-			uint8 flags;
-		}op_jumpmark;
-		struct  
-		{
-			uint32 param;
-			uint32 param2;
-			uint16 paramU16;
-		}op_macro;
-		struct  
-		{
-			uint32 jumpmarkAddress;
-			bool jumpAccordingToSegment; //PPCRecImlSegment_t* destinationSegment; // if set, this replaces jumpmarkAddress
-			uint8 condition; // only used when crRegisterIndex is 8 or above (update: Apparently only used to mark jumps without a condition? -> Cleanup)
-			uint8 crRegisterIndex;
-			uint8 crBitIndex;
-			bool  bitMustBeSet;
-		}op_conditionalJump;
-		struct
-		{
-			uint8 registerData;
-			uint8 registerMem;
-			uint8 registerMem2;
-			uint8 registerGQR;
-			uint8 copyWidth;
-			//uint8 flags;
-			struct  
-			{
-				bool swapEndian : 1;
-				bool signExtend : 1;
-				bool notExpanded : 1; // for floats
-			}flags2;
-			uint8 mode; // transfer mode (copy width, ps0/ps1 behavior)
-			sint32 immS32;
-		}op_storeLoad;
-		struct
-		{
-			struct
-			{
-				uint8 registerMem;
-				sint32 immS32;
-			}src;
-			struct
-			{
-				uint8 registerMem;
-				sint32 immS32;
-			}dst;
-			uint8 copyWidth;
-		}op_mem2mem;
-		struct  
-		{
-			uint8 registerResult;
-			uint8 registerOperand;
-			uint8 flags;
-		}op_fpr_r_r;
-		struct  
-		{
-			uint8 registerResult;
-			uint8 registerOperandA;
-			uint8 registerOperandB;
-			uint8 flags;
-		}op_fpr_r_r_r;
-		struct  
-		{
-			uint8 registerResult;
-			uint8 registerOperandA;
-			uint8 registerOperandB;
-			uint8 registerOperandC;
-			uint8 flags;
-		}op_fpr_r_r_r_r;
-		struct  
-		{
-			uint8 registerResult;
-			//uint8 flags;
-		}op_fpr_r;
-		struct  
-		{
-			uint32 ppcAddress;
-			uint32 x64Offset;
-		}op_ppcEnter;
-		struct  
-		{
-			uint8 crD; // crBitIndex (result)
-			uint8 crA; // crBitIndex
-			uint8 crB; // crBitIndex
-		}op_cr;
-		// conditional operations (emitted if supported by target platform)
-		struct
-		{
-			// r_s32
-			uint8 registerIndex;
-			sint32 immS32;
-			// condition
-			uint8 crRegisterIndex;
-			uint8 crBitIndex;
-			bool  bitMustBeSet;
-		}op_conditional_r_s32;
-	};
-}PPCRecImlInstruction_t;
-
-typedef struct _PPCRecImlSegment_t PPCRecImlSegment_t;
-
-typedef struct _ppcRecompilerSegmentPoint_t
-{
-	sint32 index;
-	PPCRecImlSegment_t* imlSegment;
-	_ppcRecompilerSegmentPoint_t* next;
-	_ppcRecompilerSegmentPoint_t* prev;
-}ppcRecompilerSegmentPoint_t;
-
-struct raLivenessLocation_t
-{
-	sint32 index;
-	bool isRead;
-	bool isWrite;
-
-	raLivenessLocation_t() = default;
-
-	raLivenessLocation_t(sint32 index, bool isRead, bool isWrite)
-		: index(index), isRead(isRead), isWrite(isWrite) {};
 };
 
-struct raLivenessSubrangeLink_t
-{
-	struct raLivenessSubrange_t* prev;
-	struct raLivenessSubrange_t* next;
-};
+#include "Cafe/HW/Espresso/Recompiler/IML/IMLInstruction.h"
+#include "Cafe/HW/Espresso/Recompiler/IML/IMLSegment.h"
 
-struct raLivenessSubrange_t
-{
-	struct raLivenessRange_t* range;
-	PPCRecImlSegment_t* imlSegment;
-	ppcRecompilerSegmentPoint_t start;
-	ppcRecompilerSegmentPoint_t end;
-	// dirty state tracking
-	bool _noLoad;
-	bool hasStore;
-	bool hasStoreDelayed;
-	// next
-	raLivenessSubrange_t* subrangeBranchTaken;
-	raLivenessSubrange_t* subrangeBranchNotTaken;
-	// processing
-	uint32 lastIterationIndex;
-	// instruction locations
-	std::vector<raLivenessLocation_t> list_locations;
-	// linked list (subranges with same GPR virtual register)
-	raLivenessSubrangeLink_t link_sameVirtualRegisterGPR;
-	// linked list (all subranges for this segment)
-	raLivenessSubrangeLink_t link_segmentSubrangesGPR;
-};
-
-struct raLivenessRange_t
-{
-	sint32 virtualRegister;
-	sint32 physicalRegister;
-	sint32 name;
-	std::vector<raLivenessSubrange_t*> list_subranges;
-};
-
-struct PPCSegmentRegisterAllocatorInfo_t
-{
-	// analyzer stage
-	bool isPartOfProcessedLoop{}; // used during loop detection
-	sint32 lastIterationIndex{};
-	// linked lists
-	raLivenessSubrange_t* linkedList_allSubranges{};
-	raLivenessSubrange_t* linkedList_perVirtualGPR[PPC_REC_MAX_VIRTUAL_GPR]{};
-};
-
-struct PPCRecVGPRDistances_t
-{
-	struct _RegArrayEntry
-	{
-		sint32 usageStart{};
-		sint32 usageEnd{};
-	}reg[PPC_REC_MAX_VIRTUAL_GPR];
-	bool isProcessed[PPC_REC_MAX_VIRTUAL_GPR]{};
-};
-
-typedef struct _PPCRecImlSegment_t
-{
-	sint32 momentaryIndex{}; // index in segment list, generally not kept up to date except if needed (necessary for loop detection)
-	sint32 startOffset{}; // offset to first instruction in iml instruction list
-	sint32 count{}; // number of instructions in segment
-	uint32 ppcAddress{}; // ppc address (0xFFFFFFFF if not associated with an address)
-	uint32 x64Offset{}; // x64 code offset of segment start
-	uint32 cycleCount{}; // number of PPC cycles required to execute this segment (roughly)
-	// list of intermediate instructions in this segment
-	PPCRecImlInstruction_t* imlList{};
-	sint32 imlListSize{};
-	sint32 imlListCount{};
-	// segment link
-	_PPCRecImlSegment_t* nextSegmentBranchNotTaken{}; // this is also the default for segments where there is no branch
-	_PPCRecImlSegment_t* nextSegmentBranchTaken{};
-	bool nextSegmentIsUncertain{};
-	sint32 loopDepth{};
-	//sList_t* list_prevSegments;
-	std::vector<_PPCRecImlSegment_t*> list_prevSegments{};
-	// PPC range of segment
-	uint32 ppcAddrMin{};
-	uint32 ppcAddrMax{};
-	// enterable segments
-	bool isEnterable{}; // this segment can be entered from outside the recompiler (no preloaded registers necessary)
-	uint32 enterPPCAddress{}; // used if isEnterable is true
-	// jump destination segments
-	bool isJumpDestination{}; // segment is a destination for one or more (conditional) jumps
-	uint32 jumpDestinationPPCAddress{};
-	// PPC FPR use mask
-	bool ppcFPRUsed[32]{}; // same as ppcGPRUsed, but for FPR
-	// CR use mask
-	uint32 crBitsInput{}; // bits that are expected to be set from the previous segment (read in this segment but not overwritten)
-	uint32 crBitsRead{}; // all bits that are read in this segment
-	uint32 crBitsWritten{}; // bits that are written in this segment
-	// register allocator info
-	PPCSegmentRegisterAllocatorInfo_t raInfo{};
-	PPCRecVGPRDistances_t raDistances{};
-	bool raRangeExtendProcessed{};
-	// segment points
-	ppcRecompilerSegmentPoint_t* segmentPointList{};
-}PPCRecImlSegment_t;
+struct IMLInstruction* PPCRecompilerImlGen_generateNewEmptyInstruction(struct ppcImlGenContext_t* ppcImlGenContext);
 
 struct ppcImlGenContext_t
 {
-	PPCRecFunction_t* functionRef;
+	class PPCFunctionBoundaryTracker* boundaryTracker;
 	uint32* currentInstruction;
 	uint32  ppcAddressOfCurrentInstruction;
+	IMLSegment* currentOutputSegment;
+	struct PPCBasicBlockInfo* currentBasicBlock{};
 	// fpr mode
 	bool LSQE{ true };
 	bool PSE{ true };
 	// cycle counter
 	uint32 cyclesSinceLastBranch; // used to track ppc cycles
-	// temporary general purpose registers
-	uint32 mappedRegister[PPC_REC_MAX_VIRTUAL_GPR];
-	// temporary floating point registers (single and double precision)
-	uint32 mappedFPRRegister[256];
-	// list of intermediate instructions
-	PPCRecImlInstruction_t* imlList;
-	sint32 imlListSize;
-	sint32 imlListCount;
+	std::unordered_map<IMLName, IMLReg> mappedRegs;
+
+	uint32 GetMaxRegId() const
+	{
+		if (mappedRegs.empty())
+			return 0;
+		return mappedRegs.size()-1;
+	}
+
 	// list of segments
-	PPCRecImlSegment_t** segmentList;
-	sint32 segmentListSize;
-	sint32 segmentListCount;
+	std::vector<IMLSegment*> segmentList2;
 	// code generation control
 	bool hasFPUInstruction; // if true, PPCEnter macro will create FP_UNAVAIL checks -> Not needed in user mode
-	// register allocator info
-	struct  
-	{
-		std::vector<raLivenessRange_t*> list_ranges;
-	}raInfo;
 	// analysis info
 	struct  
 	{
 		bool modifiesGQR[8];
 	}tracking;
+	// debug helpers
+	uint32 debug_entryPPCAddress{0};
+
+	~ppcImlGenContext_t()
+	{
+		for (IMLSegment* imlSegment : segmentList2)
+			delete imlSegment;
+		segmentList2.clear();
+	}
+
+	// append raw instruction
+	IMLInstruction& emitInst()
+	{
+		return *PPCRecompilerImlGen_generateNewEmptyInstruction(this);
+	}
+
+	IMLSegment* NewSegment()
+	{
+		IMLSegment* seg = new IMLSegment();
+		segmentList2.emplace_back(seg);
+		return seg;
+	}
+
+	size_t GetSegmentIndex(IMLSegment* seg)
+	{
+		for (size_t i = 0; i < segmentList2.size(); i++)
+		{
+			if (segmentList2[i] == seg)
+				return i;
+		}
+		cemu_assert_error();
+		return 0;
+	}
+
+	IMLSegment* InsertSegment(size_t index)
+	{
+		IMLSegment* newSeg = new IMLSegment();
+		segmentList2.insert(segmentList2.begin() + index, 1, newSeg);
+		return newSeg;
+	}
+
+	std::span<IMLSegment*> InsertSegments(size_t index, size_t count)
+	{
+		segmentList2.insert(segmentList2.begin() + index, count, {});
+		for (size_t i = index; i < (index + count); i++)
+			segmentList2[i] = new IMLSegment();
+		return { segmentList2.data() + index, count};
+	}
+
+	void UpdateSegmentIndices()
+	{
+		for (size_t i = 0; i < segmentList2.size(); i++)
+			segmentList2[i]->momentaryIndex = (sint32)i;
+	}
 };
 
 typedef void ATTR_MS_ABI (*PPCREC_JUMP_ENTRY)();
@@ -385,8 +162,6 @@ extern void ATTR_MS_ABI (*PPCRecompiler_leaveRecompilerCode_unvisited)();
 
 #define PPC_REC_INVALID_FUNCTION	((PPCRecFunction_t*)-1)
 
-// todo - move some of the stuff above into PPCRecompilerInternal.h
-
 // recompiler interface
 
 void PPCRecompiler_recompileIfUnvisited(uint32 enterAddress);

src/Cafe/HW/Espresso/Recompiler/PPCRecompilerIml.h

@@ -1,275 +1,29 @@
+bool PPCRecompiler_generateIntermediateCode(ppcImlGenContext_t& ppcImlGenContext, PPCRecFunction_t* PPCRecFunction, std::set<uint32>& entryAddresses, class PPCFunctionBoundaryTracker& boundaryTracker);
 
-#define PPCREC_CR_REG_TEMP			8 // there are only 8 cr registers (0-7) we use the 8th as temporary cr register that is never stored (BDNZ instruction for example)
+IMLSegment* PPCIMLGen_CreateSplitSegmentAtEnd(ppcImlGenContext_t& ppcImlGenContext, PPCBasicBlockInfo& basicBlockInfo);
+IMLSegment* PPCIMLGen_CreateNewSegmentAsBranchTarget(ppcImlGenContext_t& ppcImlGenContext, PPCBasicBlockInfo& basicBlockInfo);
 
-enum
-{
-	PPCREC_IML_OP_ASSIGN,			// '=' operator
-	PPCREC_IML_OP_ENDIAN_SWAP,		// '=' operator with 32bit endian swap
-	PPCREC_IML_OP_ADD,				// '+' operator
-	PPCREC_IML_OP_SUB,				// '-' operator
-	PPCREC_IML_OP_SUB_CARRY_UPDATE_CARRY, // complex operation, result = operand + ~operand2 + carry bit, updates carry bit
-	PPCREC_IML_OP_COMPARE_SIGNED,	// arithmetic/signed comparison operator (updates cr)
-	PPCREC_IML_OP_COMPARE_UNSIGNED, // logical/unsigned comparison operator (updates cr)
-	PPCREC_IML_OP_MULTIPLY_SIGNED,  // '*' operator (signed multiply)
-	PPCREC_IML_OP_MULTIPLY_HIGH_UNSIGNED, // unsigned 64bit multiply, store only high 32bit-word of result
-	PPCREC_IML_OP_MULTIPLY_HIGH_SIGNED, // signed 64bit multiply, store only high 32bit-word of result
-	PPCREC_IML_OP_DIVIDE_SIGNED,	// '/' operator (signed divide)
-	PPCREC_IML_OP_DIVIDE_UNSIGNED,	// '/' operator (unsigned divide)
-	PPCREC_IML_OP_ADD_CARRY,		// complex operation, result = operand + carry bit, updates carry bit
-	PPCREC_IML_OP_ADD_CARRY_ME,		// complex operation, result = operand + carry bit + (-1), updates carry bit
-	PPCREC_IML_OP_ADD_UPDATE_CARRY,	// '+' operator but also updates carry flag
-	PPCREC_IML_OP_ADD_CARRY_UPDATE_CARRY, // '+' operator and also adds carry, updates carry flag
-	// assign operators with cast
-	PPCREC_IML_OP_ASSIGN_S16_TO_S32, // copy 16bit and sign extend
-	PPCREC_IML_OP_ASSIGN_S8_TO_S32,  // copy 8bit and sign extend
-	// binary operation
-	PPCREC_IML_OP_OR,				// '|' operator
-	PPCREC_IML_OP_ORC,				// '|' operator, second operand is complemented first
-	PPCREC_IML_OP_AND,				// '&' operator
-	PPCREC_IML_OP_XOR,				// '^' operator
-	PPCREC_IML_OP_LEFT_ROTATE,		// left rotate operator
-	PPCREC_IML_OP_LEFT_SHIFT,		// shift left operator
-	PPCREC_IML_OP_RIGHT_SHIFT,		// right shift operator (unsigned)
-	PPCREC_IML_OP_NOT,				// complement each bit
-	PPCREC_IML_OP_NEG,				// negate
-	// ppc
-	PPCREC_IML_OP_RLWIMI,			// RLWIMI instruction (rotate, merge based on mask)
-	PPCREC_IML_OP_SRAW,				// SRAWI/SRAW instruction (algebraic shift right, sets ca flag)
-	PPCREC_IML_OP_SLW,				// SLW (shift based on register by up to 63 bits)
-	PPCREC_IML_OP_SRW,				// SRW (shift based on register by up to 63 bits)
-	PPCREC_IML_OP_CNTLZW,
-	PPCREC_IML_OP_SUBFC,			// SUBFC and SUBFIC (subtract from and set carry)
-	PPCREC_IML_OP_DCBZ,				// clear 32 bytes aligned to 0x20
-	PPCREC_IML_OP_MFCR,				// copy cr to gpr
-	PPCREC_IML_OP_MTCRF,			// copy gpr to cr (with mask)
-	// condition register
-	PPCREC_IML_OP_CR_CLEAR,			// clear cr bit
-	PPCREC_IML_OP_CR_SET,			// set cr bit
-	PPCREC_IML_OP_CR_OR,			// OR cr bits
-	PPCREC_IML_OP_CR_ORC,			// OR cr bits, complement second input operand bit first
-	PPCREC_IML_OP_CR_AND,			// AND cr bits
-	PPCREC_IML_OP_CR_ANDC,			// AND cr bits, complement second input operand bit first
-	// FPU
-	PPCREC_IML_OP_FPR_ADD_BOTTOM,
-	PPCREC_IML_OP_FPR_ADD_PAIR,
-	PPCREC_IML_OP_FPR_SUB_PAIR,
-	PPCREC_IML_OP_FPR_SUB_BOTTOM,
-	PPCREC_IML_OP_FPR_MULTIPLY_BOTTOM,
-	PPCREC_IML_OP_FPR_MULTIPLY_PAIR,
-	PPCREC_IML_OP_FPR_DIVIDE_BOTTOM,
-	PPCREC_IML_OP_FPR_DIVIDE_PAIR,
-	PPCREC_IML_OP_FPR_COPY_BOTTOM_TO_BOTTOM_AND_TOP,
-	PPCREC_IML_OP_FPR_COPY_TOP_TO_BOTTOM_AND_TOP,
-	PPCREC_IML_OP_FPR_COPY_BOTTOM_TO_BOTTOM,
-	PPCREC_IML_OP_FPR_COPY_BOTTOM_TO_TOP, // leave bottom of destination untouched
-	PPCREC_IML_OP_FPR_COPY_TOP_TO_TOP, // leave bottom of destination untouched
-	PPCREC_IML_OP_FPR_COPY_TOP_TO_BOTTOM, // leave top of destination untouched
-	PPCREC_IML_OP_FPR_COPY_BOTTOM_AND_TOP_SWAPPED,
-	PPCREC_IML_OP_FPR_EXPAND_BOTTOM32_TO_BOTTOM64_AND_TOP64, // expand bottom f32 to f64 in bottom and top half
-	PPCREC_IML_OP_FPR_BOTTOM_FRES_TO_BOTTOM_AND_TOP, // calculate reciprocal with Espresso accuracy of source bottom half and write result to destination bottom and top half
-	PPCREC_IML_OP_FPR_FCMPO_BOTTOM,
-	PPCREC_IML_OP_FPR_FCMPU_BOTTOM,
-	PPCREC_IML_OP_FPR_FCMPU_TOP,
-	PPCREC_IML_OP_FPR_NEGATE_BOTTOM,
-	PPCREC_IML_OP_FPR_NEGATE_PAIR,
-	PPCREC_IML_OP_FPR_ABS_BOTTOM, // abs(fp0)
-	PPCREC_IML_OP_FPR_ABS_PAIR,
-	PPCREC_IML_OP_FPR_FRES_PAIR, // 1.0/fp approx (Espresso accuracy)
-	PPCREC_IML_OP_FPR_FRSQRTE_PAIR, // 1.0/sqrt(fp) approx (Espresso accuracy)
-	PPCREC_IML_OP_FPR_NEGATIVE_ABS_BOTTOM, // -abs(fp0)
-	PPCREC_IML_OP_FPR_ROUND_TO_SINGLE_PRECISION_BOTTOM, // round 64bit double to 64bit double with 32bit float precision (in bottom half of xmm register)
-	PPCREC_IML_OP_FPR_ROUND_TO_SINGLE_PRECISION_PAIR, // round two 64bit doubles to 64bit double with 32bit float precision
-	PPCREC_IML_OP_FPR_BOTTOM_RECIPROCAL_SQRT,
-	PPCREC_IML_OP_FPR_BOTTOM_FCTIWZ,
-	PPCREC_IML_OP_FPR_SELECT_BOTTOM, // selectively copy bottom value from operand B or C based on value in operand A
-	PPCREC_IML_OP_FPR_SELECT_PAIR, // selectively copy top/bottom from operand B or C based on value in top/bottom of operand A
-	// PS
-	PPCREC_IML_OP_FPR_SUM0,
-	PPCREC_IML_OP_FPR_SUM1,
-};
+void PPCIMLGen_AssertIfNotLastSegmentInstruction(ppcImlGenContext_t& ppcImlGenContext);
 
-#define PPCREC_IML_OP_FPR_COPY_PAIR (PPCREC_IML_OP_ASSIGN)
-
-enum
-{
-	PPCREC_IML_MACRO_BLR,			// macro for BLR instruction code
-	PPCREC_IML_MACRO_BLRL,			// macro for BLRL instruction code
-	PPCREC_IML_MACRO_BCTR,			// macro for BCTR instruction code
-	PPCREC_IML_MACRO_BCTRL,			// macro for BCTRL instruction code
-	PPCREC_IML_MACRO_BL,			// call to different function (can be within same function)
-	PPCREC_IML_MACRO_B_FAR,			// branch to different function
-	PPCREC_IML_MACRO_COUNT_CYCLES,	// decrease current remaining thread cycles by a certain amount
-	PPCREC_IML_MACRO_HLE,			// HLE function call
-	PPCREC_IML_MACRO_MFTB,			// get TB register value (low or high)
-	PPCREC_IML_MACRO_LEAVE,			// leaves recompiler and switches to interpeter
-	// debugging
-	PPCREC_IML_MACRO_DEBUGBREAK,	// throws a debugbreak
-};
-
-enum
-{
-	PPCREC_JUMP_CONDITION_NONE,
-	PPCREC_JUMP_CONDITION_E, // equal / zero
-	PPCREC_JUMP_CONDITION_NE, // not equal / not zero
-	PPCREC_JUMP_CONDITION_LE, // less or equal
-	PPCREC_JUMP_CONDITION_L, // less
-	PPCREC_JUMP_CONDITION_GE, // greater or equal
-	PPCREC_JUMP_CONDITION_G, // greater
-	// special case:
-	PPCREC_JUMP_CONDITION_SUMMARYOVERFLOW, // needs special handling
-	PPCREC_JUMP_CONDITION_NSUMMARYOVERFLOW, // not summaryoverflow
-
-};
-
-enum
-{
-	PPCREC_CR_MODE_COMPARE_SIGNED,
-	PPCREC_CR_MODE_COMPARE_UNSIGNED, // alias logic compare
-	// others: 	PPCREC_CR_MODE_ARITHMETIC,
-	PPCREC_CR_MODE_ARITHMETIC, // arithmetic use (for use with add/sub instructions without generating extra code)
-	PPCREC_CR_MODE_LOGICAL,
-};
-
-enum
-{
-	PPCREC_IML_TYPE_NONE,
-	PPCREC_IML_TYPE_NO_OP,				// no-op instruction
-	PPCREC_IML_TYPE_JUMPMARK,			// possible jump destination (generated before each ppc instruction)
-	PPCREC_IML_TYPE_R_R,				// r* (op) *r
-	PPCREC_IML_TYPE_R_R_R,				// r* = r* (op) r*
-	PPCREC_IML_TYPE_R_R_S32,			// r* = r* (op) s32*
-	PPCREC_IML_TYPE_LOAD,				// r* = [r*+s32*]
-	PPCREC_IML_TYPE_LOAD_INDEXED,		// r* = [r*+r*]
-	PPCREC_IML_TYPE_STORE,				// [r*+s32*] = r*
-	PPCREC_IML_TYPE_STORE_INDEXED,		// [r*+r*] = r*
-	PPCREC_IML_TYPE_R_NAME,				// r* = name
-	PPCREC_IML_TYPE_NAME_R,				// name* = r*
-	PPCREC_IML_TYPE_R_S32,				// r* (op) imm
-	PPCREC_IML_TYPE_MACRO,
-	PPCREC_IML_TYPE_CJUMP,				// conditional jump
-	PPCREC_IML_TYPE_CJUMP_CYCLE_CHECK,	// jumps only if remaining thread cycles >= 0
-	PPCREC_IML_TYPE_PPC_ENTER,			// used to mark locations that should be written to recompilerCallTable
-	PPCREC_IML_TYPE_CR,					// condition register specific operations (one or more operands)
-	// conditional
-	PPCREC_IML_TYPE_CONDITIONAL_R_S32,
-	// FPR
-	PPCREC_IML_TYPE_FPR_R_NAME,			// name = f*
-	PPCREC_IML_TYPE_FPR_NAME_R,			// f* = name
-	PPCREC_IML_TYPE_FPR_LOAD,			// r* = (bitdepth) [r*+s32*] (single or paired single mode)
-	PPCREC_IML_TYPE_FPR_LOAD_INDEXED,	// r* = (bitdepth) [r*+r*] (single or paired single mode)
-	PPCREC_IML_TYPE_FPR_STORE,			// (bitdepth) [r*+s32*] = r* (single or paired single mode)
-	PPCREC_IML_TYPE_FPR_STORE_INDEXED,	// (bitdepth) [r*+r*] = r* (single or paired single mode)
-	PPCREC_IML_TYPE_FPR_R_R,
-	PPCREC_IML_TYPE_FPR_R_R_R,
-	PPCREC_IML_TYPE_FPR_R_R_R_R,
-	PPCREC_IML_TYPE_FPR_R,
-	// special
-	PPCREC_IML_TYPE_MEM2MEM,			// memory to memory copy (deprecated)
-
-};
-
-enum
-{
-	PPCREC_NAME_NONE,
-	PPCREC_NAME_TEMPORARY,
-	PPCREC_NAME_R0 = 1000,
-	PPCREC_NAME_SPR0 = 2000,
-	PPCREC_NAME_FPR0 = 3000,
-	PPCREC_NAME_TEMPORARY_FPR0 = 4000, // 0 to 7
-	//PPCREC_NAME_CR0 = 3000, // value mapped condition register (usually it isn't needed and can be optimized away)
-};
-
-// special cases for LOAD/STORE
-#define PPC_REC_LOAD_LWARX_MARKER	(100)	// lwarx instruction (similar to LWZX but sets reserved address/value)
-#define PPC_REC_STORE_STWCX_MARKER	(100)	// stwcx instruction (similar to STWX but writes only if reservation from LWARX is valid)
-#define PPC_REC_STORE_STSWI_1		(200)   // stswi nb = 1
-#define PPC_REC_STORE_STSWI_2		(201)   // stswi nb = 2
-#define PPC_REC_STORE_STSWI_3		(202)   // stswi nb = 3
-#define PPC_REC_STORE_LSWI_1		(200)   // lswi nb = 1
-#define PPC_REC_STORE_LSWI_2		(201)   // lswi nb = 2
-#define PPC_REC_STORE_LSWI_3		(202)   // lswi nb = 3
-
-#define PPC_REC_INVALID_REGISTER		0xFF
-
-#define PPCREC_CR_BIT_LT	0
-#define PPCREC_CR_BIT_GT	1
-#define PPCREC_CR_BIT_EQ	2
-#define PPCREC_CR_BIT_SO	3
-
-enum
-{
-	// fpr load
-	PPCREC_FPR_LD_MODE_SINGLE_INTO_PS0,
-	PPCREC_FPR_LD_MODE_SINGLE_INTO_PS0_PS1,
-	PPCREC_FPR_LD_MODE_DOUBLE_INTO_PS0,
-	PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0,
-	PPCREC_FPR_LD_MODE_PSQ_GENERIC_PS0_PS1,
-	PPCREC_FPR_LD_MODE_PSQ_FLOAT_PS0,
-	PPCREC_FPR_LD_MODE_PSQ_FLOAT_PS0_PS1,
-	PPCREC_FPR_LD_MODE_PSQ_S16_PS0,
-	PPCREC_FPR_LD_MODE_PSQ_S16_PS0_PS1,
-	PPCREC_FPR_LD_MODE_PSQ_U16_PS0,
-	PPCREC_FPR_LD_MODE_PSQ_U16_PS0_PS1,
-	PPCREC_FPR_LD_MODE_PSQ_S8_PS0,
-	PPCREC_FPR_LD_MODE_PSQ_S8_PS0_PS1,
-	PPCREC_FPR_LD_MODE_PSQ_U8_PS0,
-	PPCREC_FPR_LD_MODE_PSQ_U8_PS0_PS1,
-	// fpr store
-	PPCREC_FPR_ST_MODE_SINGLE_FROM_PS0, // store 1 single precision float from ps0
-	PPCREC_FPR_ST_MODE_DOUBLE_FROM_PS0, // store 1 double precision float from ps0
-
-	PPCREC_FPR_ST_MODE_UI32_FROM_PS0, // store raw low-32bit of PS0
-
-	PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0_PS1,
-	PPCREC_FPR_ST_MODE_PSQ_GENERIC_PS0,
-	PPCREC_FPR_ST_MODE_PSQ_FLOAT_PS0_PS1,
-	PPCREC_FPR_ST_MODE_PSQ_FLOAT_PS0,
-	PPCREC_FPR_ST_MODE_PSQ_S8_PS0,
-	PPCREC_FPR_ST_MODE_PSQ_S8_PS0_PS1,
-	PPCREC_FPR_ST_MODE_PSQ_U8_PS0, 
-	PPCREC_FPR_ST_MODE_PSQ_U8_PS0_PS1,
-	PPCREC_FPR_ST_MODE_PSQ_U16_PS0,
-	PPCREC_FPR_ST_MODE_PSQ_U16_PS0_PS1,
-	PPCREC_FPR_ST_MODE_PSQ_S16_PS0,
-	PPCREC_FPR_ST_MODE_PSQ_S16_PS0_PS1,
-};
-
-bool PPCRecompiler_generateIntermediateCode(ppcImlGenContext_t& ppcImlGenContext, PPCRecFunction_t* PPCRecFunction, std::set<uint32>& entryAddresses);
-void PPCRecompiler_freeContext(ppcImlGenContext_t* ppcImlGenContext); // todo - move to destructor
-
-PPCRecImlInstruction_t* PPCRecompilerImlGen_generateNewEmptyInstruction(ppcImlGenContext_t* ppcImlGenContext);
-void PPCRecompiler_pushBackIMLInstructions(PPCRecImlSegment_t* imlSegment, sint32 index, sint32 shiftBackCount);
-PPCRecImlInstruction_t* PPCRecompiler_insertInstruction(PPCRecImlSegment_t* imlSegment, sint32 index);
+IMLInstruction* PPCRecompilerImlGen_generateNewEmptyInstruction(ppcImlGenContext_t* ppcImlGenContext);
+void PPCRecompiler_pushBackIMLInstructions(IMLSegment* imlSegment, sint32 index, sint32 shiftBackCount);
+IMLInstruction* PPCRecompiler_insertInstruction(IMLSegment* imlSegment, sint32 index);
 
 void PPCRecompilerIml_insertSegments(ppcImlGenContext_t* ppcImlGenContext, sint32 index, sint32 count);
 
-void PPCRecompilerIml_setSegmentPoint(ppcRecompilerSegmentPoint_t* segmentPoint, PPCRecImlSegment_t* imlSegment, sint32 index);
-void PPCRecompilerIml_removeSegmentPoint(ppcRecompilerSegmentPoint_t* segmentPoint);
+void PPCRecompilerIml_setSegmentPoint(IMLSegmentPoint* segmentPoint, IMLSegment* imlSegment, sint32 index);
+void PPCRecompilerIml_removeSegmentPoint(IMLSegmentPoint* segmentPoint);
 
 // GPR register management
-uint32 PPCRecompilerImlGen_loadRegister(ppcImlGenContext_t* ppcImlGenContext, uint32 mappedName, bool loadNew = false);
-uint32 PPCRecompilerImlGen_loadOverwriteRegister(ppcImlGenContext_t* ppcImlGenContext, uint32 mappedName);
+IMLReg PPCRecompilerImlGen_loadRegister(ppcImlGenContext_t* ppcImlGenContext, uint32 mappedName);
 
 // FPR register management
-uint32 PPCRecompilerImlGen_loadFPRRegister(ppcImlGenContext_t* ppcImlGenContext, uint32 mappedName, bool loadNew = false);
-uint32 PPCRecompilerImlGen_loadOverwriteFPRRegister(ppcImlGenContext_t* ppcImlGenContext, uint32 mappedName);
+IMLReg PPCRecompilerImlGen_loadFPRRegister(ppcImlGenContext_t* ppcImlGenContext, uint32 mappedName, bool loadNew = false);
+IMLReg PPCRecompilerImlGen_loadOverwriteFPRRegister(ppcImlGenContext_t* ppcImlGenContext, uint32 mappedName);
 
 // IML instruction generation
-void PPCRecompilerImlGen_generateNewInstruction_jump(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlInstruction_t* imlInstruction, uint32 jumpmarkAddress);
-void PPCRecompilerImlGen_generateNewInstruction_jumpSegment(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlInstruction_t* imlInstruction);
-
-void PPCRecompilerImlGen_generateNewInstruction_r_s32(ppcImlGenContext_t* ppcImlGenContext, uint32 operation, uint8 registerIndex, sint32 immS32, uint32 copyWidth, bool signExtend, bool bigEndian, uint8 crRegister, uint32 crMode);
-void PPCRecompilerImlGen_generateNewInstruction_conditional_r_s32(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlInstruction_t* imlInstruction, uint32 operation, uint8 registerIndex, sint32 immS32, uint32 crRegisterIndex, uint32 crBitIndex, bool bitMustBeSet);
-void PPCRecompilerImlGen_generateNewInstruction_r_r(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlInstruction_t* imlInstruction, uint32 operation, uint8 registerResult, uint8 registerA, uint8 crRegister = PPC_REC_INVALID_REGISTER, uint8 crMode = 0);
-
-
-
-// IML instruction generation (new style, can generate new instructions but also overwrite existing ones)
-
-void PPCRecompilerImlGen_generateNewInstruction_noOp(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlInstruction_t* imlInstruction);
-void PPCRecompilerImlGen_generateNewInstruction_memory_memory(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlInstruction_t* imlInstruction, uint8 srcMemReg, sint32 srcImmS32, uint8 dstMemReg, sint32 dstImmS32, uint8 copyWidth);
-
-void PPCRecompilerImlGen_generateNewInstruction_fpr_r(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlInstruction_t* imlInstruction, sint32 operation, uint8 registerResult, sint32 crRegister = PPC_REC_INVALID_REGISTER);
+void PPCRecompilerImlGen_generateNewInstruction_conditional_r_s32(ppcImlGenContext_t* ppcImlGenContext, IMLInstruction* imlInstruction, uint32 operation, IMLReg registerIndex, sint32 immS32, uint32 crRegisterIndex, uint32 crBitIndex, bool bitMustBeSet);
+void PPCRecompilerImlGen_generateNewInstruction_fpr_r(ppcImlGenContext_t* ppcImlGenContext, IMLInstruction* imlInstruction, sint32 operation, IMLReg registerResult);
 
 // IML generation - FPU
 bool PPCRecompilerImlGen_LFS(ppcImlGenContext_t* ppcImlGenContext, uint32 opcode);
@@ -347,76 +101,4 @@ bool PPCRecompilerImlGen_PS_CMPU1(ppcImlGenContext_t* ppcImlGenContext, uint32 o
 
 // IML general
 
-bool PPCRecompiler_isSuffixInstruction(PPCRecImlInstruction_t* iml);
-void PPCRecompilerIML_linkSegments(ppcImlGenContext_t* ppcImlGenContext);
-void PPCRecompilerIml_setLinkBranchNotTaken(PPCRecImlSegment_t* imlSegmentSrc, PPCRecImlSegment_t* imlSegmentDst);
-void PPCRecompilerIml_setLinkBranchTaken(PPCRecImlSegment_t* imlSegmentSrc, PPCRecImlSegment_t* imlSegmentDst);
-void PPCRecompilerIML_relinkInputSegment(PPCRecImlSegment_t* imlSegmentOrig, PPCRecImlSegment_t* imlSegmentNew);
-void PPCRecompilerIML_removeLink(PPCRecImlSegment_t* imlSegmentSrc, PPCRecImlSegment_t* imlSegmentDst);
 void PPCRecompilerIML_isolateEnterableSegments(ppcImlGenContext_t* ppcImlGenContext);
-
-PPCRecImlInstruction_t* PPCRecompilerIML_getLastInstruction(PPCRecImlSegment_t* imlSegment);
-
-// IML analyzer
-typedef struct
-{
-	uint32 readCRBits;
-	uint32 writtenCRBits;
-}PPCRecCRTracking_t;
-
-bool PPCRecompilerImlAnalyzer_isTightFiniteLoop(PPCRecImlSegment_t* imlSegment);
-bool PPCRecompilerImlAnalyzer_canTypeWriteCR(PPCRecImlInstruction_t* imlInstruction);
-void PPCRecompilerImlAnalyzer_getCRTracking(PPCRecImlInstruction_t* imlInstruction, PPCRecCRTracking_t* crTracking);
-
-// IML optimizer
-bool PPCRecompiler_reduceNumberOfFPRRegisters(ppcImlGenContext_t* ppcImlGenContext);
-
-bool PPCRecompiler_manageFPRRegisters(ppcImlGenContext_t* ppcImlGenContext);
-
-void PPCRecompiler_removeRedundantCRUpdates(ppcImlGenContext_t* ppcImlGenContext);
-void PPCRecompiler_optimizeDirectFloatCopies(ppcImlGenContext_t* ppcImlGenContext);
-void PPCRecompiler_optimizeDirectIntegerCopies(ppcImlGenContext_t* ppcImlGenContext);
-
-void PPCRecompiler_optimizePSQLoadAndStore(ppcImlGenContext_t* ppcImlGenContext);
-
-// IML register allocator
-void PPCRecompilerImm_allocateRegisters(ppcImlGenContext_t* ppcImlGenContext);
-
-// late optimizations
-void PPCRecompiler_reorderConditionModifyInstructions(ppcImlGenContext_t* ppcImlGenContext);
-
-// debug
-
-void PPCRecompiler_dumpIMLSegment(PPCRecImlSegment_t* imlSegment, sint32 segmentIndex, bool printLivenessRangeInfo = false);
-
-
-typedef struct
-{
-	union
-	{
-		struct
-		{
-			sint16 readNamedReg1;
-			sint16 readNamedReg2;
-			sint16 readNamedReg3;
-			sint16 writtenNamedReg1;
-		};
-		sint16 gpr[4]; // 3 read + 1 write
-	};
-	// FPR
-	union
-	{
-		struct
-		{
-			// note: If destination operand is not fully written, it will be added as a read FPR as well
-			sint16 readFPR1;
-			sint16 readFPR2;
-			sint16 readFPR3;
-			sint16 readFPR4; // usually this is set to the result FPR if only partially overwritten
-			sint16 writtenFPR1;
-		};
-		sint16 fpr[4];
-	};
-}PPCImlOptimizerUsedRegisters_t;
-
-void PPCRecompiler_checkRegisterUsage(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlInstruction_t* imlInstruction, PPCImlOptimizerUsedRegisters_t* registersUsed);

src/Cafe/HW/Espresso/Recompiler/PPCRecompilerImlAnalyzer.cpp

@@ -1,137 +0,0 @@
-#include "PPCRecompiler.h"
-#include "PPCRecompilerIml.h"
-#include "util/helpers/fixedSizeList.h"
-#include "Cafe/HW/Espresso/Interpreter/PPCInterpreterInternal.h"
-
-/*
- * Initializes a single segment and returns true if it is a finite loop
- */
-bool PPCRecompilerImlAnalyzer_isTightFiniteLoop(PPCRecImlSegment_t* imlSegment)
-{
-	bool isTightFiniteLoop = false;
-	// base criteria, must jump to beginning of same segment
-	if (imlSegment->nextSegmentBranchTaken != imlSegment)
-		return false;
-	// loops using BDNZ are assumed to always be finite
-	for (sint32 t = 0; t < imlSegment->imlListCount; t++)
-	{
-		if (imlSegment->imlList[t].type == PPCREC_IML_TYPE_R_S32 && imlSegment->imlList[t].operation == PPCREC_IML_OP_SUB && imlSegment->imlList[t].crRegister == 8)
-		{
-			return true;
-		}
-	}
-	// for non-BDNZ loops, check for common patterns
-	// risky approach, look for ADD/SUB operations and assume that potential overflow means finite (does not include r_r_s32 ADD/SUB)
-	// this catches most loops with load-update and store-update instructions, but also those with decrementing counters
-	FixedSizeList<sint32, 64, true> list_modifiedRegisters;
-	for (sint32 t = 0; t < imlSegment->imlListCount; t++)
-	{
-		if (imlSegment->imlList[t].type == PPCREC_IML_TYPE_R_S32 && (imlSegment->imlList[t].operation == PPCREC_IML_OP_ADD || imlSegment->imlList[t].operation == PPCREC_IML_OP_SUB) )
-		{
-			list_modifiedRegisters.addUnique(imlSegment->imlList[t].op_r_immS32.registerIndex);
-		}
-	}
-	if (list_modifiedRegisters.count > 0)
-	{
-		// remove all registers from the list that are modified by non-ADD/SUB instructions
-		// todo: We should also cover the case where ADD+SUB on the same register cancel the effect out
-		PPCImlOptimizerUsedRegisters_t registersUsed;
-		for (sint32 t = 0; t < imlSegment->imlListCount; t++)
-		{
-			if (imlSegment->imlList[t].type == PPCREC_IML_TYPE_R_S32 && (imlSegment->imlList[t].operation == PPCREC_IML_OP_ADD || imlSegment->imlList[t].operation == PPCREC_IML_OP_SUB))
-				continue;
-			PPCRecompiler_checkRegisterUsage(NULL, imlSegment->imlList + t, &registersUsed);
-			if(registersUsed.writtenNamedReg1 < 0)
-				continue;
-			list_modifiedRegisters.remove(registersUsed.writtenNamedReg1);
-		}
-		if (list_modifiedRegisters.count > 0)
-		{
-			return true;
-		}
-	}
-	return false;
-}
-
-/*
-* Returns true if the imlInstruction can overwrite CR (depending on value of ->crRegister)
-*/
-bool PPCRecompilerImlAnalyzer_canTypeWriteCR(PPCRecImlInstruction_t* imlInstruction)
-{
-	if (imlInstruction->type == PPCREC_IML_TYPE_R_R)
-		return true;
-	if (imlInstruction->type == PPCREC_IML_TYPE_R_R_R)
-		return true;
-	if (imlInstruction->type == PPCREC_IML_TYPE_R_R_S32)
-		return true;
-	if (imlInstruction->type == PPCREC_IML_TYPE_R_S32)
-		return true;
-	if (imlInstruction->type == PPCREC_IML_TYPE_FPR_R_R)
-		return true;
-	if (imlInstruction->type == PPCREC_IML_TYPE_FPR_R_R_R)
-		return true;
-	if (imlInstruction->type == PPCREC_IML_TYPE_FPR_R_R_R_R)
-		return true;
-	if (imlInstruction->type == PPCREC_IML_TYPE_FPR_R)
-		return true;
-	return false;
-}
-
-void PPCRecompilerImlAnalyzer_getCRTracking(PPCRecImlInstruction_t* imlInstruction, PPCRecCRTracking_t* crTracking)
-{
-	crTracking->readCRBits = 0;
-	crTracking->writtenCRBits = 0;
-	if (imlInstruction->type == PPCREC_IML_TYPE_CJUMP)
-	{
-		if (imlInstruction->op_conditionalJump.condition != PPCREC_JUMP_CONDITION_NONE)
-		{
-			uint32 crBitFlag = 1 << (imlInstruction->op_conditionalJump.crRegisterIndex * 4 + imlInstruction->op_conditionalJump.crBitIndex);
-			crTracking->readCRBits = (crBitFlag);
-		}
-	}
-	else if (imlInstruction->type == PPCREC_IML_TYPE_CONDITIONAL_R_S32)
-	{
-		uint32 crBitFlag = 1 << (imlInstruction->op_conditional_r_s32.crRegisterIndex * 4 + imlInstruction->op_conditional_r_s32.crBitIndex);
-		crTracking->readCRBits = crBitFlag;
-	}
-	else if (imlInstruction->type == PPCREC_IML_TYPE_R_S32 && imlInstruction->operation == PPCREC_IML_OP_MFCR)
-	{
-		crTracking->readCRBits = 0xFFFFFFFF;
-	}
-	else if (imlInstruction->type == PPCREC_IML_TYPE_R_S32 && imlInstruction->operation == PPCREC_IML_OP_MTCRF)
-	{
-		crTracking->writtenCRBits |= ppc_MTCRFMaskToCRBitMask((uint32)imlInstruction->op_r_immS32.immS32);
-	}
-	else if (imlInstruction->type == PPCREC_IML_TYPE_CR)
-	{
-		if (imlInstruction->operation == PPCREC_IML_OP_CR_CLEAR ||
-			imlInstruction->operation == PPCREC_IML_OP_CR_SET)
-		{
-			uint32 crBitFlag = 1 << (imlInstruction->op_cr.crD);
-			crTracking->writtenCRBits = crBitFlag;
-		}
-		else if (imlInstruction->operation == PPCREC_IML_OP_CR_OR ||
-			imlInstruction->operation == PPCREC_IML_OP_CR_ORC ||
-			imlInstruction->operation == PPCREC_IML_OP_CR_AND || 
-			imlInstruction->operation == PPCREC_IML_OP_CR_ANDC)
-		{
-			uint32 crBitFlag = 1 << (imlInstruction->op_cr.crD);
-			crTracking->writtenCRBits = crBitFlag;
-			crBitFlag = 1 << (imlInstruction->op_cr.crA);
-			crTracking->readCRBits = crBitFlag;
-			crBitFlag = 1 << (imlInstruction->op_cr.crB);
-			crTracking->readCRBits |= crBitFlag;
-		}
-		else
-			assert_dbg();
-	}
-	else if (PPCRecompilerImlAnalyzer_canTypeWriteCR(imlInstruction) && imlInstruction->crRegister >= 0 && imlInstruction->crRegister <= 7)
-	{
-		crTracking->writtenCRBits |= (0xF << (imlInstruction->crRegister * 4));
-	}
-	else if ((imlInstruction->type == PPCREC_IML_TYPE_STORE || imlInstruction->type == PPCREC_IML_TYPE_STORE_INDEXED) && imlInstruction->op_storeLoad.copyWidth == PPC_REC_STORE_STWCX_MARKER)
-	{
-		// overwrites CR0
-		crTracking->writtenCRBits |= (0xF << 0);
-	}
-}

src/Cafe/HW/Espresso/Recompiler/PPCRecompilerImlRanges.cpp

@@ -1,399 +0,0 @@
-#include "PPCRecompiler.h"
-#include "PPCRecompilerIml.h"
-#include "PPCRecompilerX64.h"
-#include "PPCRecompilerImlRanges.h"
-#include "util/helpers/MemoryPool.h"
-
-void PPCRecRARange_addLink_perVirtualGPR(raLivenessSubrange_t** root, raLivenessSubrange_t* subrange)
-{
-#ifdef CEMU_DEBUG_ASSERT
-	if ((*root) && (*root)->range->virtualRegister != subrange->range->virtualRegister)
-		assert_dbg();
-#endif
-	subrange->link_sameVirtualRegisterGPR.next = *root;
-	if (*root)
-		(*root)->link_sameVirtualRegisterGPR.prev = subrange;
-	subrange->link_sameVirtualRegisterGPR.prev = nullptr;
-	*root = subrange;
-}
-
-void PPCRecRARange_addLink_allSubrangesGPR(raLivenessSubrange_t** root, raLivenessSubrange_t* subrange)
-{
-	subrange->link_segmentSubrangesGPR.next = *root;
-	if (*root)
-		(*root)->link_segmentSubrangesGPR.prev = subrange;
-	subrange->link_segmentSubrangesGPR.prev = nullptr;
-	*root = subrange;
-}
-
-void PPCRecRARange_removeLink_perVirtualGPR(raLivenessSubrange_t** root, raLivenessSubrange_t* subrange)
-{
-	raLivenessSubrange_t* tempPrev = subrange->link_sameVirtualRegisterGPR.prev;
-	if (subrange->link_sameVirtualRegisterGPR.prev)
-		subrange->link_sameVirtualRegisterGPR.prev->link_sameVirtualRegisterGPR.next = subrange->link_sameVirtualRegisterGPR.next;
-	else
-		(*root) = subrange->link_sameVirtualRegisterGPR.next;
-	if (subrange->link_sameVirtualRegisterGPR.next)
-		subrange->link_sameVirtualRegisterGPR.next->link_sameVirtualRegisterGPR.prev = tempPrev;
-#ifdef CEMU_DEBUG_ASSERT
-	subrange->link_sameVirtualRegisterGPR.prev = (raLivenessSubrange_t*)1;
-	subrange->link_sameVirtualRegisterGPR.next = (raLivenessSubrange_t*)1;
-#endif
-}
-
-void PPCRecRARange_removeLink_allSubrangesGPR(raLivenessSubrange_t** root, raLivenessSubrange_t* subrange)
-{
-	raLivenessSubrange_t* tempPrev = subrange->link_segmentSubrangesGPR.prev;
-	if (subrange->link_segmentSubrangesGPR.prev)
-		subrange->link_segmentSubrangesGPR.prev->link_segmentSubrangesGPR.next = subrange->link_segmentSubrangesGPR.next;
-	else
-		(*root) = subrange->link_segmentSubrangesGPR.next;
-	if (subrange->link_segmentSubrangesGPR.next)
-		subrange->link_segmentSubrangesGPR.next->link_segmentSubrangesGPR.prev = tempPrev;
-#ifdef CEMU_DEBUG_ASSERT
-	subrange->link_segmentSubrangesGPR.prev = (raLivenessSubrange_t*)1;
-	subrange->link_segmentSubrangesGPR.next = (raLivenessSubrange_t*)1;
-#endif
-}
-
-MemoryPoolPermanentObjects<raLivenessRange_t> memPool_livenessRange(4096);
-MemoryPoolPermanentObjects<raLivenessSubrange_t> memPool_livenessSubrange(4096);
-
-raLivenessRange_t* PPCRecRA_createRangeBase(ppcImlGenContext_t* ppcImlGenContext, uint32 virtualRegister, uint32 name)
-{
-	raLivenessRange_t* livenessRange = memPool_livenessRange.acquireObj();
-	livenessRange->list_subranges.resize(0);
-	livenessRange->virtualRegister = virtualRegister;
-	livenessRange->name = name;
-	livenessRange->physicalRegister = -1;
-	ppcImlGenContext->raInfo.list_ranges.push_back(livenessRange);
-	return livenessRange;
-}
-
-raLivenessSubrange_t* PPCRecRA_createSubrange(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange_t* range, PPCRecImlSegment_t* imlSegment, sint32 startIndex, sint32 endIndex)
-{
-	raLivenessSubrange_t* livenessSubrange = memPool_livenessSubrange.acquireObj();
-	livenessSubrange->list_locations.resize(0);
-	livenessSubrange->range = range;
-	livenessSubrange->imlSegment = imlSegment;
-	PPCRecompilerIml_setSegmentPoint(&livenessSubrange->start, imlSegment, startIndex);
-	PPCRecompilerIml_setSegmentPoint(&livenessSubrange->end, imlSegment, endIndex);
-	// default values
-	livenessSubrange->hasStore = false;
-	livenessSubrange->hasStoreDelayed = false;
-	livenessSubrange->lastIterationIndex = 0;
-	livenessSubrange->subrangeBranchNotTaken = nullptr;
-	livenessSubrange->subrangeBranchTaken = nullptr;
-	livenessSubrange->_noLoad = false;
-	// add to range
-	range->list_subranges.push_back(livenessSubrange);
-	// add to segment
-	PPCRecRARange_addLink_perVirtualGPR(&(imlSegment->raInfo.linkedList_perVirtualGPR[range->virtualRegister]), livenessSubrange);
-	PPCRecRARange_addLink_allSubrangesGPR(&imlSegment->raInfo.linkedList_allSubranges, livenessSubrange);
-	return livenessSubrange;
-}
-
-void _unlinkSubrange(raLivenessSubrange_t* subrange)
-{
-	PPCRecImlSegment_t* imlSegment = subrange->imlSegment;
-	PPCRecRARange_removeLink_perVirtualGPR(&imlSegment->raInfo.linkedList_perVirtualGPR[subrange->range->virtualRegister], subrange);
-	PPCRecRARange_removeLink_allSubrangesGPR(&imlSegment->raInfo.linkedList_allSubranges, subrange);
-}
-
-void PPCRecRA_deleteSubrange(ppcImlGenContext_t* ppcImlGenContext, raLivenessSubrange_t* subrange)
-{
-	_unlinkSubrange(subrange);
-	subrange->range->list_subranges.erase(std::find(subrange->range->list_subranges.begin(), subrange->range->list_subranges.end(), subrange));
-	subrange->list_locations.clear();
-	PPCRecompilerIml_removeSegmentPoint(&subrange->start);
-	PPCRecompilerIml_removeSegmentPoint(&subrange->end);
-	memPool_livenessSubrange.releaseObj(subrange);
-}
-
-void _PPCRecRA_deleteSubrangeNoUnlinkFromRange(ppcImlGenContext_t* ppcImlGenContext, raLivenessSubrange_t* subrange)
-{
-	_unlinkSubrange(subrange);
-	PPCRecompilerIml_removeSegmentPoint(&subrange->start);
-	PPCRecompilerIml_removeSegmentPoint(&subrange->end);
-	memPool_livenessSubrange.releaseObj(subrange);
-}
-
-void PPCRecRA_deleteRange(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange_t* range)
-{
-	for (auto& subrange : range->list_subranges)
-	{
-		_PPCRecRA_deleteSubrangeNoUnlinkFromRange(ppcImlGenContext, subrange);
-	}
-	ppcImlGenContext->raInfo.list_ranges.erase(std::find(ppcImlGenContext->raInfo.list_ranges.begin(), ppcImlGenContext->raInfo.list_ranges.end(), range));
-	memPool_livenessRange.releaseObj(range);
-}
-
-void PPCRecRA_deleteRangeNoUnlink(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange_t* range)
-{
-	for (auto& subrange : range->list_subranges)
-	{
-		_PPCRecRA_deleteSubrangeNoUnlinkFromRange(ppcImlGenContext, subrange);
-	}
-	memPool_livenessRange.releaseObj(range);
-}
-
-void PPCRecRA_deleteAllRanges(ppcImlGenContext_t* ppcImlGenContext)
-{
-	for(auto& range : ppcImlGenContext->raInfo.list_ranges)
-	{
-		PPCRecRA_deleteRangeNoUnlink(ppcImlGenContext, range);
-	}
-	ppcImlGenContext->raInfo.list_ranges.clear();
-}
-
-void PPCRecRA_mergeRanges(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange_t* range, raLivenessRange_t* absorbedRange)
-{
-	cemu_assert_debug(range != absorbedRange);
-	cemu_assert_debug(range->virtualRegister == absorbedRange->virtualRegister);
-	// move all subranges from absorbedRange to range
-	for (auto& subrange : absorbedRange->list_subranges)
-	{
-		range->list_subranges.push_back(subrange);
-		subrange->range = range;
-	}
-	absorbedRange->list_subranges.clear();
-	PPCRecRA_deleteRange(ppcImlGenContext, absorbedRange);
-}
-
-void PPCRecRA_mergeSubranges(ppcImlGenContext_t* ppcImlGenContext, raLivenessSubrange_t* subrange, raLivenessSubrange_t* absorbedSubrange)
-{
-#ifdef CEMU_DEBUG_ASSERT
-	PPCRecRA_debugValidateSubrange(subrange);
-	PPCRecRA_debugValidateSubrange(absorbedSubrange);
-	if (subrange->imlSegment != absorbedSubrange->imlSegment)
-		assert_dbg();
-	if (subrange->end.index > absorbedSubrange->start.index)
-		assert_dbg();
-	if (subrange->subrangeBranchTaken || subrange->subrangeBranchNotTaken)
-		assert_dbg();
-	if (subrange == absorbedSubrange)
-		assert_dbg();
-#endif
-	subrange->subrangeBranchTaken = absorbedSubrange->subrangeBranchTaken;
-	subrange->subrangeBranchNotTaken = absorbedSubrange->subrangeBranchNotTaken;
-
-	// merge usage locations
-	for (auto& location : absorbedSubrange->list_locations)
-	{
-		subrange->list_locations.push_back(location);
-	}
-	absorbedSubrange->list_locations.clear();
-
-	subrange->end.index = absorbedSubrange->end.index;
-
-	PPCRecRA_debugValidateSubrange(subrange);
-
-	PPCRecRA_deleteSubrange(ppcImlGenContext, absorbedSubrange);
-}
-
-// remove all inter-segment connections from the range and split it into local ranges (also removes empty ranges)
-void PPCRecRA_explodeRange(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange_t* range)
-{
-	if (range->list_subranges.size() == 1)
-		assert_dbg();
-	for (auto& subrange : range->list_subranges)
-	{
-		if (subrange->list_locations.empty())
-			continue;
-		raLivenessRange_t* newRange = PPCRecRA_createRangeBase(ppcImlGenContext, range->virtualRegister, range->name);
-		raLivenessSubrange_t* newSubrange = PPCRecRA_createSubrange(ppcImlGenContext, newRange, subrange->imlSegment, subrange->list_locations.data()[0].index, subrange->list_locations.data()[subrange->list_locations.size() - 1].index + 1);
-		// copy locations
-		for (auto& location : subrange->list_locations)
-		{
-			newSubrange->list_locations.push_back(location);
-		}
-	}
-	// remove original range
-	PPCRecRA_deleteRange(ppcImlGenContext, range);
-}
-
-#ifdef CEMU_DEBUG_ASSERT
-void PPCRecRA_debugValidateSubrange(raLivenessSubrange_t* subrange)
-{
-	// validate subrange
-	if (subrange->subrangeBranchTaken && subrange->subrangeBranchTaken->imlSegment != subrange->imlSegment->nextSegmentBranchTaken)
-		assert_dbg();
-	if (subrange->subrangeBranchNotTaken && subrange->subrangeBranchNotTaken->imlSegment != subrange->imlSegment->nextSegmentBranchNotTaken)
-		assert_dbg();
-}
-#else
-void PPCRecRA_debugValidateSubrange(raLivenessSubrange_t* subrange) {}
-#endif
-
-// split subrange at the given index
-// After the split there will be two ranges/subranges:
-// head -> subrange is shortned to end at splitIndex
-// tail -> a new subrange that reaches from splitIndex to the end of the original subrange
-// if head has a physical register assigned it will not carry over to tail
-// The return value is the tail subrange
-// If trimToHole is true, the end of the head subrange and the start of the tail subrange will be moved to fit the locations
-// Ranges that begin at RA_INTER_RANGE_START are allowed and can be split
-raLivenessSubrange_t* PPCRecRA_splitLocalSubrange(ppcImlGenContext_t* ppcImlGenContext, raLivenessSubrange_t* subrange, sint32 splitIndex, bool trimToHole)
-{
-	// validation
-#ifdef CEMU_DEBUG_ASSERT
-	if (subrange->end.index == RA_INTER_RANGE_END || subrange->end.index == RA_INTER_RANGE_START)
-		assert_dbg();
-	if (subrange->start.index >= splitIndex)
-		assert_dbg();
-	if (subrange->end.index <= splitIndex)
-		assert_dbg();
-#endif
-	// create tail
-	raLivenessRange_t* tailRange = PPCRecRA_createRangeBase(ppcImlGenContext, subrange->range->virtualRegister, subrange->range->name);
-	raLivenessSubrange_t* tailSubrange = PPCRecRA_createSubrange(ppcImlGenContext, tailRange, subrange->imlSegment, splitIndex, subrange->end.index);
-	// copy locations
-	for (auto& location : subrange->list_locations)
-	{
-		if (location.index >= splitIndex)
-			tailSubrange->list_locations.push_back(location);
-	}
-	// remove tail locations from head
-	for (sint32 i = 0; i < subrange->list_locations.size(); i++)
-	{
-		raLivenessLocation_t* location = subrange->list_locations.data() + i;
-		if (location->index >= splitIndex)
-		{
-			subrange->list_locations.resize(i);
-			break;
-		}
-	}
-	// adjust start/end
-	if (trimToHole)
-	{
-		if (subrange->list_locations.empty())
-		{
-			subrange->end.index = subrange->start.index+1;
-		}
-		else
-		{
-			subrange->end.index = subrange->list_locations.back().index + 1;
-		}
-		if (tailSubrange->list_locations.empty())
-		{
-			assert_dbg(); // should not happen? (In this case we can just avoid generating a tail at all)
-		}
-		else
-		{
-			tailSubrange->start.index = tailSubrange->list_locations.front().index;
-		}
-	}
-	return tailSubrange;
-}
-
-void PPCRecRA_updateOrAddSubrangeLocation(raLivenessSubrange_t* subrange, sint32 index, bool isRead, bool isWrite)
-{
-	if (subrange->list_locations.empty())
-	{
-		subrange->list_locations.emplace_back(index, isRead, isWrite);
-		return;
-	}
-	raLivenessLocation_t* lastLocation = subrange->list_locations.data() + (subrange->list_locations.size() - 1);
-	cemu_assert_debug(lastLocation->index <= index);
-	if (lastLocation->index == index)
-	{
-		// update
-		lastLocation->isRead = lastLocation->isRead || isRead;
-		lastLocation->isWrite = lastLocation->isWrite || isWrite;
-		return;
-	}
-	// add new
-	subrange->list_locations.emplace_back(index, isRead, isWrite);
-}
-
-sint32 PPCRecRARange_getReadWriteCost(PPCRecImlSegment_t* imlSegment)
-{
-	sint32 v = imlSegment->loopDepth + 1;
-	v *= 5;
-	return v*v; // 25, 100, 225, 400
-}
-
-// calculate cost of entire range
-// ignores data flow and does not detect avoidable reads/stores
-sint32 PPCRecRARange_estimateCost(raLivenessRange_t* range)
-{
-	sint32 cost = 0;
-
-	// todo - this algorithm isn't accurate. If we have 10 parallel branches with a load each then the actual cost is still only that of one branch (plus minimal extra cost for generating more code). 
-
-	// currently we calculate the cost based on the most expensive entry/exit point
-
-	sint32 mostExpensiveRead = 0;
-	sint32 mostExpensiveWrite = 0;
-	sint32 readCount = 0;
-	sint32 writeCount = 0;
-
-	for (auto& subrange : range->list_subranges)
-	{
-		if (subrange->start.index != RA_INTER_RANGE_START)
-		{
-			//cost += PPCRecRARange_getReadWriteCost(subrange->imlSegment);
-			mostExpensiveRead = std::max(mostExpensiveRead, PPCRecRARange_getReadWriteCost(subrange->imlSegment));
-			readCount++;
-		}
-		if (subrange->end.index != RA_INTER_RANGE_END)
-		{
-			//cost += PPCRecRARange_getReadWriteCost(subrange->imlSegment);
-			mostExpensiveWrite = std::max(mostExpensiveWrite, PPCRecRARange_getReadWriteCost(subrange->imlSegment));
-			writeCount++;
-		}
-	}
-	cost = mostExpensiveRead + mostExpensiveWrite;
-	cost = cost + (readCount + writeCount) / 10;
-	return cost;
-}
-
-// calculate cost of range that it would have after calling PPCRecRA_explodeRange() on it
-sint32 PPCRecRARange_estimateAdditionalCostAfterRangeExplode(raLivenessRange_t* range)
-{
-	sint32 cost = -PPCRecRARange_estimateCost(range);
-	for (auto& subrange : range->list_subranges)
-	{
-		if (subrange->list_locations.empty())
-			continue;
-		cost += PPCRecRARange_getReadWriteCost(subrange->imlSegment) * 2; // we assume a read and a store
-	}
-	return cost;
-}
-
-sint32 PPCRecRARange_estimateAdditionalCostAfterSplit(raLivenessSubrange_t* subrange, sint32 splitIndex)
-{
-	// validation
-#ifdef CEMU_DEBUG_ASSERT
-	if (subrange->end.index == RA_INTER_RANGE_END)
-		assert_dbg();
-#endif
-
-	sint32 cost = 0;
-	// find split position in location list
-	if (subrange->list_locations.empty())
-	{
-		assert_dbg(); // should not happen?
-		return 0;
-	}
-	if (splitIndex <= subrange->list_locations.front().index)
-		return 0;
-	if (splitIndex > subrange->list_locations.back().index)
-		return 0;
-
-	// todo - determine exact cost of split subranges
-
-	cost += PPCRecRARange_getReadWriteCost(subrange->imlSegment) * 2; // currently we assume that the additional region will require a read and a store
-
-	//for (sint32 f = 0; f < subrange->list_locations.size(); f++)
-	//{
-	//	raLivenessLocation_t* location = subrange->list_locations.data() + f;
-	//	if (location->index >= splitIndex)
-	//	{
-	//		...
-	//		return cost;
-	//	}
-	//}
-
-	return cost;
-}

src/Cafe/HW/Espresso/Recompiler/PPCRecompilerImlRanges.h

@@ -1,27 +0,0 @@
-#pragma once
-
-raLivenessRange_t* PPCRecRA_createRangeBase(ppcImlGenContext_t* ppcImlGenContext, uint32 virtualRegister, uint32 name);
-raLivenessSubrange_t* PPCRecRA_createSubrange(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange_t* range, PPCRecImlSegment_t* imlSegment, sint32 startIndex, sint32 endIndex);
-void PPCRecRA_deleteSubrange(ppcImlGenContext_t* ppcImlGenContext, raLivenessSubrange_t* subrange);
-void PPCRecRA_deleteRange(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange_t* range);
-void PPCRecRA_deleteAllRanges(ppcImlGenContext_t* ppcImlGenContext);
-
-void PPCRecRA_mergeRanges(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange_t* range, raLivenessRange_t* absorbedRange);
-void PPCRecRA_explodeRange(ppcImlGenContext_t* ppcImlGenContext, raLivenessRange_t* range);
-
-void PPCRecRA_mergeSubranges(ppcImlGenContext_t* ppcImlGenContext, raLivenessSubrange_t* subrange, raLivenessSubrange_t* absorbedSubrange);
-
-raLivenessSubrange_t* PPCRecRA_splitLocalSubrange(ppcImlGenContext_t* ppcImlGenContext, raLivenessSubrange_t* subrange, sint32 splitIndex, bool trimToHole = false);
-
-void PPCRecRA_updateOrAddSubrangeLocation(raLivenessSubrange_t* subrange, sint32 index, bool isRead, bool isWrite);
-void PPCRecRA_debugValidateSubrange(raLivenessSubrange_t* subrange);
-
-// cost estimation
-sint32 PPCRecRARange_getReadWriteCost(PPCRecImlSegment_t* imlSegment);
-sint32 PPCRecRARange_estimateCost(raLivenessRange_t* range);
-sint32 PPCRecRARange_estimateAdditionalCostAfterRangeExplode(raLivenessRange_t* range);
-sint32 PPCRecRARange_estimateAdditionalCostAfterSplit(raLivenessSubrange_t* subrange, sint32 splitIndex);
-
-// special values to mark the index of ranges that reach across the segment border
-#define RA_INTER_RANGE_START	(-1)
-#define RA_INTER_RANGE_END		(0x70000000)

src/Cafe/HW/Espresso/Recompiler/PPCRecompilerImlRegisterAllocator2.cpp

@@ -1,414 +0,0 @@
-#include "PPCRecompiler.h"
-#include "PPCRecompilerIml.h"
-#include "PPCRecompilerX64.h"
-#include "PPCRecompilerImlRanges.h"
-#include <queue>
-
-bool _isRangeDefined(PPCRecImlSegment_t* imlSegment, sint32 vGPR)
-{
-	return (imlSegment->raDistances.reg[vGPR].usageStart != INT_MAX);
-}
-
-void PPCRecRA_calculateSegmentMinMaxRanges(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlSegment_t* imlSegment)
-{
-	for (sint32 i = 0; i < PPC_REC_MAX_VIRTUAL_GPR; i++)
-	{
-		imlSegment->raDistances.reg[i].usageStart = INT_MAX;
-		imlSegment->raDistances.reg[i].usageEnd = INT_MIN;
-	}
-	// scan instructions for usage range
-	sint32 index = 0;
-	PPCImlOptimizerUsedRegisters_t gprTracking;
-	while (index < imlSegment->imlListCount)
-	{
-		// end loop at suffix instruction
-		if (PPCRecompiler_isSuffixInstruction(imlSegment->imlList + index))
-			break;
-		// get accessed GPRs
-		PPCRecompiler_checkRegisterUsage(NULL, imlSegment->imlList + index, &gprTracking);
-		for (sint32 t = 0; t < 4; t++)
-		{
-			sint32 virtualRegister = gprTracking.gpr[t];
-			if (virtualRegister < 0)
-				continue;
-			cemu_assert_debug(virtualRegister < PPC_REC_MAX_VIRTUAL_GPR);
-			imlSegment->raDistances.reg[virtualRegister].usageStart = std::min(imlSegment->raDistances.reg[virtualRegister].usageStart, index); // index before/at instruction
-			imlSegment->raDistances.reg[virtualRegister].usageEnd = std::max(imlSegment->raDistances.reg[virtualRegister].usageEnd, index+1); // index after instruction
-		}
-		// next instruction
-		index++;
-	}
-}
-
-void PPCRecRA_calculateLivenessRangesV2(ppcImlGenContext_t* ppcImlGenContext)
-{
-	// for each register calculate min/max index of usage range within each segment
-	for (sint32 s = 0; s < ppcImlGenContext->segmentListCount; s++)
-	{
-		PPCRecRA_calculateSegmentMinMaxRanges(ppcImlGenContext, ppcImlGenContext->segmentList[s]);
-	}
-}
-
-raLivenessSubrange_t* PPCRecRA_convertToMappedRanges(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlSegment_t* imlSegment, sint32 vGPR, raLivenessRange_t* range)
-{
-	if (imlSegment->raDistances.isProcessed[vGPR])
-	{
-		// return already existing segment
-		return imlSegment->raInfo.linkedList_perVirtualGPR[vGPR];
-	}
-	imlSegment->raDistances.isProcessed[vGPR] = true;
-	if (_isRangeDefined(imlSegment, vGPR) == false)
-		return nullptr;
-	// create subrange
-	cemu_assert_debug(imlSegment->raInfo.linkedList_perVirtualGPR[vGPR] == nullptr);
-	raLivenessSubrange_t* subrange = PPCRecRA_createSubrange(ppcImlGenContext, range, imlSegment, imlSegment->raDistances.reg[vGPR].usageStart, imlSegment->raDistances.reg[vGPR].usageEnd);
-	// traverse forward
-	if (imlSegment->raDistances.reg[vGPR].usageEnd == RA_INTER_RANGE_END)
-	{
-		if (imlSegment->nextSegmentBranchTaken && imlSegment->nextSegmentBranchTaken->raDistances.reg[vGPR].usageStart == RA_INTER_RANGE_START)
-		{
-			subrange->subrangeBranchTaken = PPCRecRA_convertToMappedRanges(ppcImlGenContext, imlSegment->nextSegmentBranchTaken, vGPR, range);
-			cemu_assert_debug(subrange->subrangeBranchTaken->start.index == RA_INTER_RANGE_START);
-		}
-		if (imlSegment->nextSegmentBranchNotTaken && imlSegment->nextSegmentBranchNotTaken->raDistances.reg[vGPR].usageStart == RA_INTER_RANGE_START)
-		{
-			subrange->subrangeBranchNotTaken = PPCRecRA_convertToMappedRanges(ppcImlGenContext, imlSegment->nextSegmentBranchNotTaken, vGPR, range);
-			cemu_assert_debug(subrange->subrangeBranchNotTaken->start.index == RA_INTER_RANGE_START);
-		}
-	}
-	// traverse backward
-	if (imlSegment->raDistances.reg[vGPR].usageStart == RA_INTER_RANGE_START)
-	{
-		for (auto& it : imlSegment->list_prevSegments)
-		{
-			if (it->raDistances.reg[vGPR].usageEnd == RA_INTER_RANGE_END)
-				PPCRecRA_convertToMappedRanges(ppcImlGenContext, it, vGPR, range);
-		}
-	}
-	// return subrange
-	return subrange;
-}
-
-void PPCRecRA_createSegmentLivenessRanges(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlSegment_t* imlSegment)
-{
-	for (sint32 i = 0; i < PPC_REC_MAX_VIRTUAL_GPR; i++)
-	{
-		if( _isRangeDefined(imlSegment, i) == false )
-			continue;
-		if( imlSegment->raDistances.isProcessed[i])
-			continue;
-		raLivenessRange_t* range = PPCRecRA_createRangeBase(ppcImlGenContext, i, ppcImlGenContext->mappedRegister[i]);
-		PPCRecRA_convertToMappedRanges(ppcImlGenContext, imlSegment, i, range);
-	}
-	// create lookup table of ranges
-	raLivenessSubrange_t* vGPR2Subrange[PPC_REC_MAX_VIRTUAL_GPR];
-	for (sint32 i = 0; i < PPC_REC_MAX_VIRTUAL_GPR; i++)
-	{
-		vGPR2Subrange[i] = imlSegment->raInfo.linkedList_perVirtualGPR[i];
-#ifdef CEMU_DEBUG_ASSERT
-		if (vGPR2Subrange[i] && vGPR2Subrange[i]->link_sameVirtualRegisterGPR.next != nullptr)
-			assert_dbg();
-#endif
-	}
-	// parse instructions and convert to locations
-	sint32 index = 0;
-	PPCImlOptimizerUsedRegisters_t gprTracking;
-	while (index < imlSegment->imlListCount)
-	{
-		// end loop at suffix instruction
-		if (PPCRecompiler_isSuffixInstruction(imlSegment->imlList + index))
-			break;
-		// get accessed GPRs
-		PPCRecompiler_checkRegisterUsage(NULL, imlSegment->imlList + index, &gprTracking);
-		// handle accessed GPR
-		for (sint32 t = 0; t < 4; t++)
-		{
-			sint32 virtualRegister = gprTracking.gpr[t];
-			if (virtualRegister < 0)
-				continue;
-			bool isWrite = (t == 3);
-			// add location
-			PPCRecRA_updateOrAddSubrangeLocation(vGPR2Subrange[virtualRegister], index, isWrite == false, isWrite);
-#ifdef CEMU_DEBUG_ASSERT
-			if (index < vGPR2Subrange[virtualRegister]->start.index)
-				assert_dbg();
-			if (index+1 > vGPR2Subrange[virtualRegister]->end.index)
-				assert_dbg();
-#endif
-		}
-		// next instruction
-		index++;
-	}
-}
-
-void PPCRecRA_extendRangeToEndOfSegment(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlSegment_t* imlSegment, sint32 vGPR)
-{
-	if (_isRangeDefined(imlSegment, vGPR) == false)
-	{
-		imlSegment->raDistances.reg[vGPR].usageStart = RA_INTER_RANGE_END;
-		imlSegment->raDistances.reg[vGPR].usageEnd = RA_INTER_RANGE_END;
-		return;
-	}
-	imlSegment->raDistances.reg[vGPR].usageEnd = RA_INTER_RANGE_END;
-}
-
-void PPCRecRA_extendRangeToBeginningOfSegment(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlSegment_t* imlSegment, sint32 vGPR)
-{
-	if (_isRangeDefined(imlSegment, vGPR) == false)
-	{
-		imlSegment->raDistances.reg[vGPR].usageStart = RA_INTER_RANGE_START;
-		imlSegment->raDistances.reg[vGPR].usageEnd = RA_INTER_RANGE_START;
-	}
-	else
-	{
-		imlSegment->raDistances.reg[vGPR].usageStart = RA_INTER_RANGE_START;
-	}
-	// propagate backwards
-	for (auto& it : imlSegment->list_prevSegments)
-	{
-		PPCRecRA_extendRangeToEndOfSegment(ppcImlGenContext, it, vGPR);
-	}
-}
-
-void _PPCRecRA_connectRanges(ppcImlGenContext_t* ppcImlGenContext, sint32 vGPR, PPCRecImlSegment_t** route, sint32 routeDepth)
-{
-#ifdef CEMU_DEBUG_ASSERT
-	if (routeDepth < 2)
-		assert_dbg();
-#endif
-	// extend starting range to end of segment
-	PPCRecRA_extendRangeToEndOfSegment(ppcImlGenContext, route[0], vGPR);
-	// extend all the connecting segments in both directions
-	for (sint32 i = 1; i < (routeDepth - 1); i++)
-	{
-		PPCRecRA_extendRangeToEndOfSegment(ppcImlGenContext, route[i], vGPR);
-		PPCRecRA_extendRangeToBeginningOfSegment(ppcImlGenContext, route[i], vGPR);
-	}
-	// extend the final segment towards the beginning
-	PPCRecRA_extendRangeToBeginningOfSegment(ppcImlGenContext, route[routeDepth-1], vGPR);
-}
-
-void _PPCRecRA_checkAndTryExtendRange(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlSegment_t* currentSegment, sint32 vGPR, sint32 distanceLeft, PPCRecImlSegment_t** route, sint32 routeDepth)
-{
-	if (routeDepth >= 64)
-	{
-		cemuLog_logDebug(LogType::Force, "Recompiler RA route maximum depth exceeded for function 0x{:08x}", ppcImlGenContext->functionRef->ppcAddress);
-		return;
-	}
-	route[routeDepth] = currentSegment;
-	if (currentSegment->raDistances.reg[vGPR].usageStart == INT_MAX)
-	{
-		// measure distance to end of segment
-		distanceLeft -= currentSegment->imlListCount;
-		if (distanceLeft > 0)
-		{
-			if (currentSegment->nextSegmentBranchNotTaken)
-				_PPCRecRA_checkAndTryExtendRange(ppcImlGenContext, currentSegment->nextSegmentBranchNotTaken, vGPR, distanceLeft, route, routeDepth + 1);
-			if (currentSegment->nextSegmentBranchTaken)
-				_PPCRecRA_checkAndTryExtendRange(ppcImlGenContext, currentSegment->nextSegmentBranchTaken, vGPR, distanceLeft, route, routeDepth + 1);
-		}
-		return;
-	}
-	else
-	{
-		// measure distance to range
-		if (currentSegment->raDistances.reg[vGPR].usageStart == RA_INTER_RANGE_END)
-		{
-			if (distanceLeft < currentSegment->imlListCount)
-				return; // range too far away
-		}
-		else if (currentSegment->raDistances.reg[vGPR].usageStart != RA_INTER_RANGE_START && currentSegment->raDistances.reg[vGPR].usageStart > distanceLeft)
-			return; // out of range
-		// found close range -> connect ranges
-		_PPCRecRA_connectRanges(ppcImlGenContext, vGPR, route, routeDepth + 1);
-	}
-}
-
-void PPCRecRA_checkAndTryExtendRange(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlSegment_t* currentSegment, sint32 vGPR)
-{
-#ifdef CEMU_DEBUG_ASSERT
-	if (currentSegment->raDistances.reg[vGPR].usageEnd < 0)
-		assert_dbg();
-#endif
-	// count instructions to end of initial segment
-	if (currentSegment->raDistances.reg[vGPR].usageEnd == RA_INTER_RANGE_START)
-		assert_dbg();
-	sint32 instructionsUntilEndOfSeg;
-	if (currentSegment->raDistances.reg[vGPR].usageEnd == RA_INTER_RANGE_END)
-		instructionsUntilEndOfSeg = 0;
-	else
-		instructionsUntilEndOfSeg = currentSegment->imlListCount - currentSegment->raDistances.reg[vGPR].usageEnd;
-
-#ifdef CEMU_DEBUG_ASSERT
-	if (instructionsUntilEndOfSeg < 0)
-		assert_dbg();
-#endif
-	sint32 remainingScanDist = 45 - instructionsUntilEndOfSeg;
-	if (remainingScanDist <= 0)
-		return; // can't reach end
-
-	// also dont forget: Extending is easier if we allow 'non symetric' branches. E.g. register range one enters one branch
-	PPCRecImlSegment_t* route[64];
-	route[0] = currentSegment;
-	if (currentSegment->nextSegmentBranchNotTaken)
-	{
-		_PPCRecRA_checkAndTryExtendRange(ppcImlGenContext, currentSegment->nextSegmentBranchNotTaken, vGPR, remainingScanDist, route, 1);
-	}
-	if (currentSegment->nextSegmentBranchTaken)
-	{
-		_PPCRecRA_checkAndTryExtendRange(ppcImlGenContext, currentSegment->nextSegmentBranchTaken, vGPR, remainingScanDist, route, 1);
-	}
-}
-
-void PPCRecRA_mergeCloseRangesForSegmentV2(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlSegment_t* imlSegment)
-{
-	for (sint32 i = 0; i < PPC_REC_MAX_VIRTUAL_GPR; i++) // todo: Use dynamic maximum or list of used vGPRs so we can avoid parsing empty entries
-	{
-		if(imlSegment->raDistances.reg[i].usageStart == INT_MAX)
-			continue; // not used
-		// check and extend if possible
-		PPCRecRA_checkAndTryExtendRange(ppcImlGenContext, imlSegment, i);
-	}
-#ifdef CEMU_DEBUG_ASSERT
-	if (imlSegment->list_prevSegments.empty() == false && imlSegment->isEnterable)
-		assert_dbg();
-	if ((imlSegment->nextSegmentBranchNotTaken != nullptr || imlSegment->nextSegmentBranchTaken != nullptr) && imlSegment->nextSegmentIsUncertain)
-		assert_dbg();
-#endif
-}
-
-void PPCRecRA_followFlowAndExtendRanges(ppcImlGenContext_t* ppcImlGenContext, PPCRecImlSegment_t* imlSegment)
-{
-	std::vector<PPCRecImlSegment_t*> list_segments;
-	list_segments.reserve(1000);
-	sint32 index = 0;
-	imlSegment->raRangeExtendProcessed = true;
-	list_segments.push_back(imlSegment);
-	while (index < list_segments.size())
-	{
-		PPCRecImlSegment_t* currentSegment = list_segments[index];
-		PPCRecRA_mergeCloseRangesForSegmentV2(ppcImlGenContext, currentSegment);
-		// follow flow
-		if (currentSegment->nextSegmentBranchNotTaken && currentSegment->nextSegmentBranchNotTaken->raRangeExtendProcessed == false)
-		{
-			currentSegment->nextSegmentBranchNotTaken->raRangeExtendProcessed = true;
-			list_segments.push_back(currentSegment->nextSegmentBranchNotTaken);
-		}
-		if (currentSegment->nextSegmentBranchTaken && currentSegment->nextSegmentBranchTaken->raRangeExtendProcessed == false)
-		{
-			currentSegment->nextSegmentBranchTaken->raRangeExtendProcessed = true;
-			list_segments.push_back(currentSegment->nextSegmentBranchTaken);
-		}
-		index++;
-	}
-}
-
-void PPCRecRA_mergeCloseRangesV2(ppcImlGenContext_t* ppcImlGenContext)
-{
-	for (sint32 s = 0; s < ppcImlGenContext->segmentListCount; s++)
-	{
-		PPCRecImlSegment_t* imlSegment = ppcImlGenContext->segmentList[s];
-		if (imlSegment->list_prevSegments.empty())
-		{
-			if (imlSegment->raRangeExtendProcessed)
-				assert_dbg(); // should not happen
-			PPCRecRA_followFlowAndExtendRanges(ppcImlGenContext, imlSegment);
-		}
-	}
-}
-
-void PPCRecRA_extendRangesOutOfLoopsV2(ppcImlGenContext_t* ppcImlGenContext)
-{
-	for (sint32 s = 0; s < ppcImlGenContext->segmentListCount; s++)
-	{
-		PPCRecImlSegment_t* imlSegment = ppcImlGenContext->segmentList[s];
-		auto localLoopDepth = imlSegment->loopDepth;
-		if( localLoopDepth <= 0 )
-			continue; // not inside a loop
-		// look for loop exit
-		bool hasLoopExit = false;
-		if (imlSegment->nextSegmentBranchTaken && imlSegment->nextSegmentBranchTaken->loopDepth < localLoopDepth)
-		{
-			hasLoopExit = true;
-		}
-		if (imlSegment->nextSegmentBranchNotTaken && imlSegment->nextSegmentBranchNotTaken->loopDepth < localLoopDepth)
-		{
-			hasLoopExit = true;
-		}
-		if(hasLoopExit == false)
-			continue;
-
-		// extend looping ranges into all exits (this allows the data flow analyzer to move stores out of the loop)
-		for (sint32 i = 0; i < PPC_REC_MAX_VIRTUAL_GPR; i++) // todo: Use dynamic maximum or list of used vGPRs so we can avoid parsing empty entries
-		{
-			if (imlSegment->raDistances.reg[i].usageEnd != RA_INTER_RANGE_END)
-				continue; // range not set or does not reach end of segment
-			if(imlSegment->nextSegmentBranchTaken)
-				PPCRecRA_extendRangeToBeginningOfSegment(ppcImlGenContext, imlSegment->nextSegmentBranchTaken, i);
-			if(imlSegment->nextSegmentBranchNotTaken)
-				PPCRecRA_extendRangeToBeginningOfSegment(ppcImlGenContext, imlSegment->nextSegmentBranchNotTaken, i);
-		}
-	}
-}
-
-void PPCRecRA_processFlowAndCalculateLivenessRangesV2(ppcImlGenContext_t* ppcImlGenContext)
-{
-	// merge close ranges
-	PPCRecRA_mergeCloseRangesV2(ppcImlGenContext);
-	// extra pass to move register stores out of loops
-	PPCRecRA_extendRangesOutOfLoopsV2(ppcImlGenContext);
-	// calculate liveness ranges
-	for (sint32 s = 0; s < ppcImlGenContext->segmentListCount; s++)
-	{
-		PPCRecImlSegment_t* imlSegment = ppcImlGenContext->segmentList[s];
-		PPCRecRA_createSegmentLivenessRanges(ppcImlGenContext, imlSegment);
-	}
-}
-
-void PPCRecRA_analyzeSubrangeDataDependencyV2(raLivenessSubrange_t* subrange)
-{
-	bool isRead = false;
-	bool isWritten = false;
-	bool isOverwritten = false;
-	for (auto& location : subrange->list_locations)
-	{
-		if (location.isRead)
-		{
-			isRead = true;
-		}
-		if (location.isWrite)
-		{
-			if (isRead == false)
-				isOverwritten = true;
-			isWritten = true;
-		}
-	}
-	subrange->_noLoad = isOverwritten;
-	subrange->hasStore = isWritten;
-
-	if (subrange->start.index == RA_INTER_RANGE_START)
-		subrange->_noLoad = true;
-}
-
-void _analyzeRangeDataFlow(raLivenessSubrange_t* subrange);
-
-void PPCRecRA_analyzeRangeDataFlowV2(ppcImlGenContext_t* ppcImlGenContext)
-{
-	// this function is called after _assignRegisters(), which means that all ranges are already final and wont change anymore
-	// first do a per-subrange pass
-	for (auto& range : ppcImlGenContext->raInfo.list_ranges)
-	{
-		for (auto& subrange : range->list_subranges)
-		{
-			PPCRecRA_analyzeSubrangeDataDependencyV2(subrange);
-		}
-	}
-	// then do a second pass where we scan along subrange flow
-	for (auto& range : ppcImlGenContext->raInfo.list_ranges)
-	{
-		for (auto& subrange : range->list_subranges) // todo - traversing this backwards should be faster and yield better results due to the nature of the algorithm
-		{
-			_analyzeRangeDataFlow(subrange);
-		}
-	}
-}

src/Cafe/HW/Espresso/Recompiler/PPCRecompilerIntermediate.cpp

@@ -1,173 +1,26 @@
 #include "PPCRecompiler.h"
 #include "PPCRecompilerIml.h"
 
-PPCRecImlSegment_t* PPCRecompiler_getSegmentByPPCJumpAddress(ppcImlGenContext_t* ppcImlGenContext, uint32 ppcOffset)
-{
-	for(sint32 s=0; s<ppcImlGenContext->segmentListCount; s++)
-	{
-		if( ppcImlGenContext->segmentList[s]->isJumpDestination && ppcImlGenContext->segmentList[s]->jumpDestinationPPCAddress == ppcOffset )
-		{
-			return ppcImlGenContext->segmentList[s];
-		}
-	}
-	debug_printf("PPCRecompiler_getSegmentByPPCJumpAddress(): Unable to find segment (ppcOffset 0x%08x)\n", ppcOffset);
-	return NULL;
-}
-
-void PPCRecompilerIml_setLinkBranchNotTaken(PPCRecImlSegment_t* imlSegmentSrc, PPCRecImlSegment_t* imlSegmentDst)
-{
-	// make sure segments aren't already linked
-	if (imlSegmentSrc->nextSegmentBranchNotTaken == imlSegmentDst)
-		return;
-	// add as next segment for source
-	if (imlSegmentSrc->nextSegmentBranchNotTaken != NULL)
-		assert_dbg();
-	imlSegmentSrc->nextSegmentBranchNotTaken = imlSegmentDst;
-	// add as previous segment for destination
-	imlSegmentDst->list_prevSegments.push_back(imlSegmentSrc);
-}
-
-void PPCRecompilerIml_setLinkBranchTaken(PPCRecImlSegment_t* imlSegmentSrc, PPCRecImlSegment_t* imlSegmentDst)
-{
-	// make sure segments aren't already linked
-	if (imlSegmentSrc->nextSegmentBranchTaken == imlSegmentDst)
-		return;
-	// add as next segment for source
-	if (imlSegmentSrc->nextSegmentBranchTaken != NULL)
-		assert_dbg();
-	imlSegmentSrc->nextSegmentBranchTaken = imlSegmentDst;
-	// add as previous segment for destination
-	imlSegmentDst->list_prevSegments.push_back(imlSegmentSrc);
-}
-
-void PPCRecompilerIML_removeLink(PPCRecImlSegment_t* imlSegmentSrc, PPCRecImlSegment_t* imlSegmentDst)
-{
-	if (imlSegmentSrc->nextSegmentBranchNotTaken == imlSegmentDst)
-	{
-		imlSegmentSrc->nextSegmentBranchNotTaken = NULL;
-	}
-	else if (imlSegmentSrc->nextSegmentBranchTaken == imlSegmentDst)
-	{
-		imlSegmentSrc->nextSegmentBranchTaken = NULL;
-	}
-	else
-		assert_dbg();
-
-	bool matchFound = false;
-	for (sint32 i = 0; i < imlSegmentDst->list_prevSegments.size(); i++)
-	{
-		if (imlSegmentDst->list_prevSegments[i] == imlSegmentSrc)
-		{
-			imlSegmentDst->list_prevSegments.erase(imlSegmentDst->list_prevSegments.begin()+i);
-			matchFound = true;
-			break;
-		}
-	}
-	if (matchFound == false)
-		assert_dbg();
-}
-
-/*
- * Replaces all links to segment orig with linkts to segment new
- */
-void PPCRecompilerIML_relinkInputSegment(PPCRecImlSegment_t* imlSegmentOrig, PPCRecImlSegment_t* imlSegmentNew)
-{
-	while (imlSegmentOrig->list_prevSegments.size() != 0)
-	{
-		PPCRecImlSegment_t* prevSegment = imlSegmentOrig->list_prevSegments[0];
-		if (prevSegment->nextSegmentBranchNotTaken == imlSegmentOrig)
-		{
-			PPCRecompilerIML_removeLink(prevSegment, imlSegmentOrig);
-			PPCRecompilerIml_setLinkBranchNotTaken(prevSegment, imlSegmentNew);
-		}
-		else if (prevSegment->nextSegmentBranchTaken == imlSegmentOrig)
-		{
-			PPCRecompilerIML_removeLink(prevSegment, imlSegmentOrig);
-			PPCRecompilerIml_setLinkBranchTaken(prevSegment, imlSegmentNew);
-		}
-		else
-		{
-			assert_dbg();
-		}
-	}
-}
-
-void PPCRecompilerIML_linkSegments(ppcImlGenContext_t* ppcImlGenContext)
-{
-	for(sint32 s=0; s<ppcImlGenContext->segmentListCount; s++)
-	{
-		PPCRecImlSegment_t* imlSegment = ppcImlGenContext->segmentList[s];
-
-		bool isLastSegment = (s+1)>=ppcImlGenContext->segmentListCount;
-		PPCRecImlSegment_t* nextSegment = isLastSegment?NULL:ppcImlGenContext->segmentList[s+1];
-		// handle empty segment
-		if( imlSegment->imlListCount == 0 )
-		{
-			if (isLastSegment == false)
-				PPCRecompilerIml_setLinkBranchNotTaken(imlSegment, ppcImlGenContext->segmentList[s+1]); // continue execution to next segment
-			else
-				imlSegment->nextSegmentIsUncertain = true;
-			continue;
-		}
-		// check last instruction of segment
-		PPCRecImlInstruction_t* imlInstruction = imlSegment->imlList+(imlSegment->imlListCount-1);
-		if( imlInstruction->type == PPCREC_IML_TYPE_CJUMP || imlInstruction->type == PPCREC_IML_TYPE_CJUMP_CYCLE_CHECK )
-		{
-			// find destination segment by ppc jump address
-			PPCRecImlSegment_t* jumpDestSegment = PPCRecompiler_getSegmentByPPCJumpAddress(ppcImlGenContext, imlInstruction->op_conditionalJump.jumpmarkAddress);
-			if( jumpDestSegment )
-			{
-				if (imlInstruction->op_conditionalJump.condition != PPCREC_JUMP_CONDITION_NONE)
-					PPCRecompilerIml_setLinkBranchNotTaken(imlSegment, nextSegment);
-				PPCRecompilerIml_setLinkBranchTaken(imlSegment, jumpDestSegment);
-			}
-			else
-			{
-				imlSegment->nextSegmentIsUncertain = true;
-			}
-		}
-		else if( imlInstruction->type == PPCREC_IML_TYPE_MACRO )
-		{
-			// currently we assume that the next segment is unknown for all macros
-			imlSegment->nextSegmentIsUncertain = true;
-		}
-		else
-		{
-			// all other instruction types do not branch
-			//imlSegment->nextSegment[0] = nextSegment;
-			PPCRecompilerIml_setLinkBranchNotTaken(imlSegment, nextSegment);
-			//imlSegment->nextSegmentIsUncertain = true;
-		}
-	}
-}
-
 void PPCRecompilerIML_isolateEnterableSegments(ppcImlGenContext_t* ppcImlGenContext)
 {
-	sint32 initialSegmentCount = ppcImlGenContext->segmentListCount;
-	for (sint32 i = 0; i < ppcImlGenContext->segmentListCount; i++)
+	size_t initialSegmentCount = ppcImlGenContext->segmentList2.size();
+	for (size_t i = 0; i < initialSegmentCount; i++)
 	{
-		PPCRecImlSegment_t* imlSegment = ppcImlGenContext->segmentList[i];
+		IMLSegment* imlSegment = ppcImlGenContext->segmentList2[i];
 		if (imlSegment->list_prevSegments.empty() == false && imlSegment->isEnterable)
 		{
 			// spawn new segment at end
-			PPCRecompilerIml_insertSegments(ppcImlGenContext, ppcImlGenContext->segmentListCount, 1);
-			PPCRecImlSegment_t* entrySegment = ppcImlGenContext->segmentList[ppcImlGenContext->segmentListCount-1];
+			PPCRecompilerIml_insertSegments(ppcImlGenContext, ppcImlGenContext->segmentList2.size(), 1);
+			IMLSegment* entrySegment = ppcImlGenContext->segmentList2[ppcImlGenContext->segmentList2.size()-1];
 			entrySegment->isEnterable = true;
 			entrySegment->enterPPCAddress = imlSegment->enterPPCAddress;
 			// create jump instruction
 			PPCRecompiler_pushBackIMLInstructions(entrySegment, 0, 1);
-			PPCRecompilerImlGen_generateNewInstruction_jumpSegment(ppcImlGenContext, entrySegment->imlList + 0);
-			PPCRecompilerIml_setLinkBranchTaken(entrySegment, imlSegment);
+			entrySegment->imlList.data()[0].make_jump();
+			IMLSegment_SetLinkBranchTaken(entrySegment, imlSegment);
 			// remove enterable flag from original segment
 			imlSegment->isEnterable = false;
 			imlSegment->enterPPCAddress = 0;
 		}
 	}
-}
-
-PPCRecImlInstruction_t* PPCRecompilerIML_getLastInstruction(PPCRecImlSegment_t* imlSegment)
-{
-	if (imlSegment->imlListCount == 0)
-		return nullptr;
-	return imlSegment->imlList + (imlSegment->imlListCount - 1);
-}
+}

src/Cafe/HW/Latte/Core/Latte.h

@@ -52,7 +52,7 @@ struct LatteGPUState_t
 	uint32 gx2InitCalled; // incremented every time GX2Init() is called
 	// OpenGL control
 	uint32 glVendor; // GLVENDOR_*
-	bool alwaysDisplayDRC = false;
+	bool isDRCPrimary = false;
 	// temporary (replace with proper solution later)
 	bool tvBufferUsesSRGB;
 	bool drcBufferUsesSRGB;

src/Cafe/HW/Latte/Core/LatteCommandProcessor.cpp

@@ -141,6 +141,14 @@ private:
 
 void LatteCP_processCommandBuffer(DrawPassContext& drawPassCtx);
 
+// called whenever the GPU runs out of commands or hits a wait condition (semaphores, HLE waits)
+void LatteCP_signalEnterWait()
+{
+	// based on the assumption that games won't do a rugpull and swap out buffer data in the middle of an uninterrupted sequence of drawcalls,
+	// we only flush caches when the GPU goes idle or has to wait for any operation
+	LatteIndices_invalidateAll();
+}
+
 /*
 * Read a U32 from the command buffer
 * If no data is available then wait in a busy loop
@@ -466,6 +474,8 @@ LatteCMDPtr LatteCP_itWaitRegMem(LatteCMDPtr cmd, uint32 nWords)
 	const uint32 GPU7_WAIT_MEM_OP_GREATER = 6;
 	const uint32 GPU7_WAIT_MEM_OP_NEVER = 7;
 
+	LatteCP_signalEnterWait();
+
 	bool stalls = false;
 	if ((word0 & 0x10) != 0)
 	{
@@ -594,6 +604,7 @@ LatteCMDPtr LatteCP_itMemSemaphore(LatteCMDPtr cmd, uint32 nWords)
 	else if(SEM_SIGNAL == 7)
 	{
 		// wait
+		LatteCP_signalEnterWait();
 		size_t loopCount = 0;
 		while (true)
 		{
@@ -788,7 +799,7 @@ LatteCMDPtr LatteCP_itHLESampleTimer(LatteCMDPtr cmd, uint32 nWords)
 {
 	cemu_assert_debug(nWords == 1);
 	MPTR timerMPTR = (MPTR)LatteReadCMD();
-	memory_writeU64(timerMPTR, coreinit::coreinit_getTimerTick());
+	memory_writeU64(timerMPTR, coreinit::OSGetSystemTime());
 	return cmd;
 }
 
@@ -1305,11 +1316,13 @@ void LatteCP_processCommandBuffer(DrawPassContext& drawPassCtx)
 				}
 				case IT_HLE_TRIGGER_SCANBUFFER_SWAP:
 				{
+					LatteCP_signalEnterWait();
 					LatteCP_itHLESwapScanBuffer(cmdData, nWords);
 					break;
 				}
 				case IT_HLE_WAIT_FOR_FLIP:
 				{
+					LatteCP_signalEnterWait();
 					LatteCP_itHLEWaitForFlip(cmdData, nWords);
 					break;
 				}
@@ -1594,12 +1607,14 @@ void LatteCP_ProcessRingbuffer()
 			}
 			case IT_HLE_TRIGGER_SCANBUFFER_SWAP:
 			{
+				LatteCP_signalEnterWait();
 				LatteCP_itHLESwapScanBuffer(cmd, nWords);
 				timerRecheck += CP_TIMER_RECHECK / 64;
 				break;
 			}
 			case IT_HLE_WAIT_FOR_FLIP:
 			{
+				LatteCP_signalEnterWait();
 				LatteCP_itHLEWaitForFlip(cmd, nWords);
 				timerRecheck += CP_TIMER_RECHECK / 1;
 				break;

src/Cafe/HW/Latte/Core/LatteIndices.cpp

@@ -1,6 +1,7 @@
 #include "Cafe/HW/Latte/Core/LatteConst.h"
 #include "Cafe/HW/Latte/Renderer/Renderer.h"
 #include "Cafe/HW/Latte/ISA/RegDefines.h"
+#include "Cafe/HW/Latte/Core/LattePerformanceMonitor.h"
 #include "Common/cpu_features.h"
 
 #if defined(ARCH_X86_64) && defined(__GNUC__)
@@ -9,32 +10,53 @@
 
 struct  
 {
-	const void* lastPtr;
-	uint32 lastCount;
-	LattePrimitiveMode lastPrimitiveMode;
-	LatteIndexType lastIndexType;
-	// output
-	uint32 indexMin;
-	uint32 indexMax;
-	Renderer::INDEX_TYPE renderIndexType;
-	uint32 outputCount;
-	uint32 indexBufferOffset;
-	uint32 indexBufferIndex;
+	struct CacheEntry
+	{
+		// input data
+		const void* lastPtr;
+		uint32 lastCount;
+		LattePrimitiveMode lastPrimitiveMode;
+		LatteIndexType lastIndexType;
+		uint64 lastUsed;
+		// output
+		uint32 indexMin;
+		uint32 indexMax;
+		Renderer::INDEX_TYPE renderIndexType;
+		uint32 outputCount;
+		Renderer::IndexAllocation indexAllocation;
+	};
+	std::array<CacheEntry, 8> entry;
+	uint64 currentUsageCounter{0};
 }LatteIndexCache{};
 
 void LatteIndices_invalidate(const void* memPtr, uint32 size)
 {
-	if (LatteIndexCache.lastPtr >= memPtr && (LatteIndexCache.lastPtr < ((uint8*)memPtr + size)) )
+	for(auto& entry : LatteIndexCache.entry)
 	{
-		LatteIndexCache.lastPtr = nullptr;
-		LatteIndexCache.lastCount = 0;
+		if (entry.lastPtr >= memPtr && (entry.lastPtr < ((uint8*)memPtr + size)) )
+		{
+			if(entry.lastPtr != nullptr)
+				g_renderer->indexData_releaseIndexMemory(entry.indexAllocation);
+			entry.lastPtr = nullptr;
+			entry.lastCount = 0;
+		}
 	}
 }
 
 void LatteIndices_invalidateAll()
 {
-	LatteIndexCache.lastPtr = nullptr;
-	LatteIndexCache.lastCount = 0;
+	for(auto& entry : LatteIndexCache.entry)
+	{
+		if (entry.lastPtr != nullptr)
+			g_renderer->indexData_releaseIndexMemory(entry.indexAllocation);
+		entry.lastPtr = nullptr;
+		entry.lastCount = 0;
+	}
+}
+
+uint64 LatteIndices_GetNextUsageIndex()
+{
+	return LatteIndexCache.currentUsageCounter++;
 }
 
 uint32 LatteIndices_calculateIndexOutputSize(LattePrimitiveMode primitiveMode, LatteIndexType indexType, uint32 count)
@@ -532,7 +554,7 @@ void LatteIndices_alternativeCalculateIndexMinMax(const void* indexData, LatteIn
 	}
 }
 
-void LatteIndices_decode(const void* indexData, LatteIndexType indexType, uint32 count, LattePrimitiveMode primitiveMode, uint32& indexMin, uint32& indexMax, Renderer::INDEX_TYPE& renderIndexType, uint32& outputCount, uint32& indexBufferOffset, uint32& indexBufferIndex)
+void LatteIndices_decode(const void* indexData, LatteIndexType indexType, uint32 count, LattePrimitiveMode primitiveMode, uint32& indexMin, uint32& indexMax, Renderer::INDEX_TYPE& renderIndexType, uint32& outputCount, Renderer::IndexAllocation& indexAllocation)
 {
 	// what this should do:
 	// [x] use fast SIMD-based index decoding
@@ -542,17 +564,18 @@ void LatteIndices_decode(const void* indexData, LatteIndexType indexType, uint32
 	// [ ] better cache implementation, allow to cache across frames
 
 	// reuse from cache if data didn't change
-	if (LatteIndexCache.lastPtr == indexData &&
-		LatteIndexCache.lastCount == count &&
-		LatteIndexCache.lastPrimitiveMode == primitiveMode &&
-		LatteIndexCache.lastIndexType == indexType)
+	auto cacheEntry = std::find_if(LatteIndexCache.entry.begin(), LatteIndexCache.entry.end(), [indexData, count, primitiveMode, indexType](const auto& entry)
 	{
-		indexMin = LatteIndexCache.indexMin;
-		indexMax = LatteIndexCache.indexMax;
-		renderIndexType = LatteIndexCache.renderIndexType;
-		outputCount = LatteIndexCache.outputCount;
-		indexBufferOffset = LatteIndexCache.indexBufferOffset;
-		indexBufferIndex = LatteIndexCache.indexBufferIndex;
+		return entry.lastPtr == indexData && entry.lastCount == count && entry.lastPrimitiveMode == primitiveMode && entry.lastIndexType == indexType;
+	});
+	if (cacheEntry != LatteIndexCache.entry.end())
+	{
+		indexMin = cacheEntry->indexMin;
+		indexMax = cacheEntry->indexMax;
+		renderIndexType = cacheEntry->renderIndexType;
+		outputCount = cacheEntry->outputCount;
+		indexAllocation = cacheEntry->indexAllocation;
+		cacheEntry->lastUsed = LatteIndices_GetNextUsageIndex();
 		return;
 	}
 
@@ -576,10 +599,12 @@ void LatteIndices_decode(const void* indexData, LatteIndexType indexType, uint32
 		indexMin = 0;
 		indexMax = std::max(count, 1u)-1;
 		renderIndexType = Renderer::INDEX_TYPE::NONE;
+		indexAllocation = {};
 		return; // no indices
 	}
 	// query index buffer from renderer
-	void* indexOutputPtr = g_renderer->indexData_reserveIndexMemory(indexOutputSize, indexBufferOffset, indexBufferIndex);
+	indexAllocation = g_renderer->indexData_reserveIndexMemory(indexOutputSize);
+	void* indexOutputPtr = indexAllocation.mem;
 
 	// decode indices
 	indexMin = std::numeric_limits<uint32>::max();
@@ -704,16 +729,25 @@ void LatteIndices_decode(const void* indexData, LatteIndexType indexType, uint32
 		// recalculate index range but filter out primitive restart index
 		LatteIndices_alternativeCalculateIndexMinMax(indexData, indexType, count, indexMin, indexMax);
 	}
-	g_renderer->indexData_uploadIndexMemory(indexBufferOffset, indexOutputSize);
+	g_renderer->indexData_uploadIndexMemory(indexAllocation);
+	performanceMonitor.cycle[performanceMonitor.cycleIndex].indexDataUploaded += indexOutputSize;
+	// get least recently used cache entry
+	auto lruEntry = std::min_element(LatteIndexCache.entry.begin(), LatteIndexCache.entry.end(), [](const auto& a, const auto& b)
+	{
+		return a.lastUsed < b.lastUsed;
+	});
+	// invalidate previous allocation
+	if(lruEntry->lastPtr != nullptr)
+		g_renderer->indexData_releaseIndexMemory(lruEntry->indexAllocation);
 	// update cache
-	LatteIndexCache.lastPtr = indexData;
-	LatteIndexCache.lastCount = count;
-	LatteIndexCache.lastPrimitiveMode = primitiveMode;
-	LatteIndexCache.lastIndexType = indexType;
-	LatteIndexCache.indexMin = indexMin;
-	LatteIndexCache.indexMax = indexMax;
-	LatteIndexCache.renderIndexType = renderIndexType;
-	LatteIndexCache.outputCount = outputCount;
-	LatteIndexCache.indexBufferOffset = indexBufferOffset;
-	LatteIndexCache.indexBufferIndex = indexBufferIndex;
+	lruEntry->lastPtr = indexData;
+	lruEntry->lastCount = count;
+	lruEntry->lastPrimitiveMode = primitiveMode;
+	lruEntry->lastIndexType = indexType;
+	lruEntry->indexMin = indexMin;
+	lruEntry->indexMax = indexMax;
+	lruEntry->renderIndexType = renderIndexType;
+	lruEntry->outputCount = outputCount;
+	lruEntry->indexAllocation = indexAllocation;
+	lruEntry->lastUsed = LatteIndices_GetNextUsageIndex();
 }

src/Cafe/HW/Latte/Core/LatteIndices.h

@@ -4,4 +4,4 @@
 
 void LatteIndices_invalidate(const void* memPtr, uint32 size);
 void LatteIndices_invalidateAll();
-void LatteIndices_decode(const void* indexData, LatteIndexType indexType, uint32 count, LattePrimitiveMode primitiveMode, uint32& indexMin, uint32& indexMax, Renderer::INDEX_TYPE& renderIndexType, uint32& outputCount, uint32& indexBufferOffset, uint32& indexBufferIndex);
+void LatteIndices_decode(const void* indexData, LatteIndexType indexType, uint32 count, LattePrimitiveMode primitiveMode, uint32& indexMin, uint32& indexMax, Renderer::INDEX_TYPE& renderIndexType, uint32& outputCount, Renderer::IndexAllocation& indexAllocation);

src/Cafe/HW/Latte/Core/LatteOverlay.cpp

@@ -107,7 +107,13 @@ void LatteOverlay_renderOverlay(ImVec2& position, ImVec2& pivot, sint32 directio
 				ImGui::Text("VRAM: %dMB / %dMB", g_state.vramUsage, g_state.vramTotal);
 
 			if (config.overlay.debug)
+			{
+				// general debug info
+				ImGui::Text("--- Debug info ---");
+				ImGui::Text("IndexUploadPerFrame: %dKB", (performanceMonitor.stats.indexDataUploadPerFrame+1023)/1024);
+				// backend specific info
 				g_renderer->AppendOverlayDebugInfo();
+			}
 
 			position.y += (ImGui::GetWindowSize().y + 10.0f) * direction;
 		}

src/Cafe/HW/Latte/Core/LattePerformanceMonitor.cpp

@@ -74,7 +74,6 @@ void LattePerformanceMonitor_frameEnd()
 		uniformBankDataUploadedPerFrame /= 1024ULL;
 		uint32 uniformBankCountUploadedPerFrame = (uint32)(uniformBankUploadedCount / (uint64)elapsedFrames);
 		uint64 indexDataUploadPerFrame = (indexDataUploaded / (uint64)elapsedFrames);
-		indexDataUploadPerFrame /= 1024ULL;
 
 		double fps = (double)elapsedFrames2S * 1000.0 / (double)totalElapsedTimeFPS;
 		uint32 shaderBindsPerFrame = shaderBindCounter / elapsedFrames;
@@ -82,7 +81,7 @@ void LattePerformanceMonitor_frameEnd()
 		uint32 rlps = (uint32)((uint64)recompilerLeaveCount * 1000ULL / (uint64)totalElapsedTime);
 		uint32 tlps = (uint32)((uint64)threadLeaveCount * 1000ULL / (uint64)totalElapsedTime);
 		// set stats
-
+		performanceMonitor.stats.indexDataUploadPerFrame = indexDataUploadPerFrame;
 		// next counter cycle
 		sint32 nextCycleIndex = (performanceMonitor.cycleIndex + 1) % PERFORMANCE_MONITOR_TRACK_CYCLES;
 		performanceMonitor.cycle[nextCycleIndex].drawCallCounter = 0;

src/Cafe/HW/Latte/Core/LattePerformanceMonitor.h

@@ -124,6 +124,7 @@ typedef struct
 		LattePerfStatCounter numGraphicPipelines;
 		LattePerfStatCounter numImages;
 		LattePerfStatCounter numImageViews;
+		LattePerfStatCounter numSamplers;
 		LattePerfStatCounter numRenderPass;
 		LattePerfStatCounter numFramebuffer;
 
@@ -131,6 +132,12 @@ typedef struct
 		LattePerfStatCounter numDrawBarriersPerFrame;
 		LattePerfStatCounter numBeginRenderpassPerFrame;
 	}vk;
+
+	// calculated stats (per frame)
+	struct
+	{
+		uint32 indexDataUploadPerFrame;
+	}stats;
 }performanceMonitor_t;
 
 extern performanceMonitor_t performanceMonitor;

src/Cafe/HW/Latte/Core/LatteRenderTarget.cpp

@@ -11,7 +11,6 @@
 #include "Cafe/HW/Latte/Core/LattePerformanceMonitor.h"
 #include "Cafe/GraphicPack/GraphicPack2.h"
 #include "config/ActiveSettings.h"
-#include "Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h"
 #include "gui/guiWrapper.h"
 #include "Cafe/OS/libs/erreula/erreula.h"
 #include "input/InputManager.h"
@@ -933,13 +932,6 @@ void LatteRenderTarget_copyToBackbuffer(LatteTextureView* textureView, bool isPa
 	if (shader == nullptr)
 	{
 		sint32 scaling_filter = downscaling ? GetConfig().downscale_filter : GetConfig().upscale_filter;
-		
-		if (g_renderer->GetType() == RendererAPI::Vulkan)
-		{
-			// force linear or nearest neighbor filter
-			if(scaling_filter != kLinearFilter && scaling_filter != kNearestNeighborFilter)
-				scaling_filter = kLinearFilter;
-		}
 
 		if (scaling_filter == kLinearFilter)
 		{
@@ -957,7 +949,7 @@ void LatteRenderTarget_copyToBackbuffer(LatteTextureView* textureView, bool isPa
 			else
 				shader = RendererOutputShader::s_bicubic_shader;
 
-			filter = LatteTextureView::MagFilter::kNearestNeighbor;
+			filter = LatteTextureView::MagFilter::kLinear;
 		}
 		else if (scaling_filter == kBicubicHermiteFilter)
 		{
@@ -989,8 +981,6 @@ void LatteRenderTarget_copyToBackbuffer(LatteTextureView* textureView, bool isPa
 	g_renderer->ImguiEnd();
 }
 
-bool ctrlTabHotkeyPressed = false;
-
 void LatteRenderTarget_itHLECopyColorBufferToScanBuffer(MPTR colorBufferPtr, uint32 colorBufferWidth, uint32 colorBufferHeight, uint32 colorBufferSliceIndex, uint32 colorBufferFormat, uint32 colorBufferPitch, Latte::E_HWTILEMODE colorBufferTilemode, uint32 colorBufferSwizzle, uint32 renderTarget)
 {
 	cemu_assert_debug(colorBufferSliceIndex == 0); // todo - support for non-zero slice
@@ -1000,38 +990,31 @@ void LatteRenderTarget_itHLECopyColorBufferToScanBuffer(MPTR colorBufferPtr, uin
 		return;
 	}
 
+	auto getVPADScreenActive = [](size_t n) -> std::pair<bool, bool> {
+		auto controller = InputManager::instance().get_vpad_controller(n);
+		if (!controller)
+			return {false,false};
+		auto pressed = controller->is_screen_active();
+		auto toggle = controller->is_screen_active_toggle();
+		return {pressed && !toggle, pressed && toggle};
+	};
+
 	const bool tabPressed = gui_isKeyDown(PlatformKeyCodes::TAB);
 	const bool ctrlPressed = gui_isKeyDown(PlatformKeyCodes::LCONTROL);
+	const auto [vpad0Active, vpad0Toggle] = getVPADScreenActive(0);
+	const auto [vpad1Active, vpad1Toggle] = getVPADScreenActive(1);
 
-	bool showDRC = swkbd_hasKeyboardInputHook() == false && tabPressed;
-	bool& alwaysDisplayDRC = LatteGPUState.alwaysDisplayDRC;
+	const bool altScreenRequested = (!ctrlPressed && tabPressed) || vpad0Active || vpad1Active;
+	const bool togglePressed = (ctrlPressed && tabPressed) || vpad0Toggle || vpad1Toggle;
+	static bool togglePressedLast = false;
 
-	if (ctrlPressed && tabPressed)
-	{
-		if (ctrlTabHotkeyPressed == false)
-		{
-			alwaysDisplayDRC = !alwaysDisplayDRC;
-			ctrlTabHotkeyPressed = true;
-		}
-	}
-	else
-		ctrlTabHotkeyPressed = false;
+	bool& isDRCPrimary = LatteGPUState.isDRCPrimary;
 
-	if (alwaysDisplayDRC)
-		showDRC = !tabPressed;
+	if(togglePressed && !togglePressedLast)
+		isDRCPrimary = !isDRCPrimary;
+	togglePressedLast = togglePressed;
 
-	if (!showDRC)
-	{
-		auto controller = InputManager::instance().get_vpad_controller(0);
-		if (controller && controller->is_screen_active())
-			showDRC = true;
-		if (!showDRC)
-		{
-			controller = InputManager::instance().get_vpad_controller(1);
-			if (controller && controller->is_screen_active())
-				showDRC = true;	
-		}
-	}
+	bool showDRC = swkbd_hasKeyboardInputHook() == false && (isDRCPrimary ^ altScreenRequested);
 
 	if ((renderTarget & RENDER_TARGET_DRC) && g_renderer->IsPadWindowActive())
 		LatteRenderTarget_copyToBackbuffer(texView, true);

src/Cafe/HW/Latte/Core/LatteShader.cpp

@@ -451,9 +451,8 @@ void LatteShader_DumpShader(uint64 baseHash, uint64 auxHash, LatteDecompilerShad
 		suffix = "gs";
 	else if (shader->shaderType == LatteConst::ShaderType::Pixel)
 		suffix = "ps";
-	fs::path dumpPath = "dump/shaders";
-	dumpPath /= fmt::format("{:016x}_{:016x}_{}.txt", baseHash, auxHash, suffix);
-	FileStream* fs = FileStream::createFile2(dumpPath);
+
+	FileStream* fs = FileStream::createFile2(ActiveSettings::GetUserDataPath("dump/shaders/{:016x}_{:016x}_{}.txt", baseHash, auxHash, suffix));
 	if (fs)
 	{
 		if (shader->strBuf_shaderSource)
@@ -479,9 +478,8 @@ void LatteShader_DumpRawShader(uint64 baseHash, uint64 auxHash, uint32 type, uin
 		suffix = "copy";
 	else if (type == SHADER_DUMP_TYPE_COMPUTE)
 		suffix = "compute";
-	fs::path dumpPath = "dump/shaders";
-	dumpPath /= fmt::format("{:016x}_{:016x}_{}.bin", baseHash, auxHash, suffix);
-	FileStream* fs = FileStream::createFile2(dumpPath);
+
+	FileStream* fs = FileStream::createFile2(ActiveSettings::GetUserDataPath("dump/shaders/{:016x}_{:016x}_{}.bin", baseHash, auxHash, suffix));
 	if (fs)
 	{
 		fs->writeData(programCode, programLen);

src/Cafe/HW/Latte/Core/LatteShaderCache.cpp

@@ -25,6 +25,9 @@
 #include "util/helpers/Serializer.h"
 
 #include <wx/msgdlg.h>
+#include <audio/IAudioAPI.h>
+#include <util/bootSound/BootSoundReader.h>
+#include <thread>
 
 #if BOOST_OS_WINDOWS
 #include <psapi.h>
@@ -155,6 +158,118 @@ bool LoadTGAFile(const std::vector<uint8>& buffer, TGAFILE *tgaFile)
 	return true;
 }
 
+class BootSoundPlayer
+{
+  public:
+	BootSoundPlayer() = default;
+	~BootSoundPlayer()
+	{
+		m_stopRequested = true;
+	}
+
+	void StartSound()
+	{
+		if (!m_bootSndPlayThread.joinable())
+		{
+			m_fadeOutRequested = false;
+			m_stopRequested = false;
+			m_bootSndPlayThread = std::thread{[this]() {
+				StreamBootSound();
+			}};
+		}
+	}
+
+	void FadeOutSound()
+	{
+		m_fadeOutRequested = true;
+	}
+
+	void ApplyFadeOutEffect(std::span<sint16> samples, uint64& fadeOutSample, uint64 fadeOutDuration)
+	{
+		for (size_t i = 0; i < samples.size(); i += 2)
+		{
+			const float decibel = (float)fadeOutSample / fadeOutDuration * -60.0f;
+			const float volumeFactor = pow(10, decibel / 20);
+			samples[i] *= volumeFactor;
+			samples[i + 1] *= volumeFactor;
+			fadeOutSample++;
+		}
+	}
+
+	void StreamBootSound()
+	{
+		SetThreadName("bootsnd");
+		constexpr sint32 sampleRate = 48'000;
+		constexpr sint32 bitsPerSample = 16;
+		constexpr sint32 samplesPerBlock = sampleRate / 10; // block is 1/10th of a second
+		constexpr sint32 nChannels = 2;
+		static_assert(bitsPerSample % 8 == 0, "bits per sample is not a multiple of 8");
+
+		AudioAPIPtr bootSndAudioDev;
+
+		try
+		{
+			bootSndAudioDev = IAudioAPI::CreateDeviceFromConfig(true, sampleRate, nChannels, samplesPerBlock, bitsPerSample);
+			if(!bootSndAudioDev)
+				return;
+		}
+		catch (const std::runtime_error& ex)
+		{
+			cemuLog_log(LogType::Force, "Failed to initialise audio device for bootup sound");
+			return;
+		}
+		bootSndAudioDev->SetAudioDelayOverride(4);
+		bootSndAudioDev->Play();
+
+		std::string sndPath = fmt::format("{}/meta/{}", CafeSystem::GetMlcStoragePath(CafeSystem::GetForegroundTitleId()), "bootSound.btsnd");
+		sint32 fscStatus = FSC_STATUS_UNDEFINED;
+
+		if(!fsc_doesFileExist(sndPath.c_str()))
+			return;
+
+		FSCVirtualFile* bootSndFileHandle = fsc_open(sndPath.c_str(), FSC_ACCESS_FLAG::OPEN_FILE | FSC_ACCESS_FLAG::READ_PERMISSION, &fscStatus);
+		if(!bootSndFileHandle)
+		{
+			cemuLog_log(LogType::Force, "failed to open bootSound.btsnd");
+			return;
+		}
+
+		constexpr sint32 audioBlockSize = samplesPerBlock * (bitsPerSample/8) * nChannels;
+		BootSoundReader bootSndFileReader(bootSndFileHandle, audioBlockSize);
+
+		uint64 fadeOutSample = 0; // track how far into the fadeout
+		constexpr uint64 fadeOutDuration = sampleRate * 2; // fadeout should last 2 seconds
+		while(fadeOutSample < fadeOutDuration && !m_stopRequested)
+		{
+			while (bootSndAudioDev->NeedAdditionalBlocks())
+			{
+				sint16* data = bootSndFileReader.getSamples();
+				if(data == nullptr)
+				{
+					// break outer loop
+					m_stopRequested = true;
+					break;
+				}
+				if(m_fadeOutRequested)
+					ApplyFadeOutEffect({data, samplesPerBlock * nChannels}, fadeOutSample, fadeOutDuration);
+
+				bootSndAudioDev->FeedBlock(data);
+			}
+			// sleep for the duration of a single block
+			std::this_thread::sleep_for(std::chrono::milliseconds(samplesPerBlock / (sampleRate/ 1'000)));
+		}
+
+		if(bootSndFileHandle)
+			fsc_close(bootSndFileHandle);
+	}
+
+  private:
+	std::thread m_bootSndPlayThread;
+	std::atomic_bool m_fadeOutRequested = false;
+	std::atomic_bool m_stopRequested = false;
+};
+static BootSoundPlayer g_bootSndPlayer;
+
 void LatteShaderCache_finish()
 {
 	if (g_renderer->GetType() == RendererAPI::Vulkan)
@@ -299,6 +414,9 @@ void LatteShaderCache_Load()
 	loadBackgroundTexture(true, g_shaderCacheLoaderState.textureTVId);
 	loadBackgroundTexture(false, g_shaderCacheLoaderState.textureDRCId);
 
+	if(GetConfig().play_boot_sound)
+		g_bootSndPlayer.StartSound();
+
 	sint32 numLoadedShaders = 0;
 	uint32 loadIndex = 0;
 
@@ -365,6 +483,11 @@ void LatteShaderCache_Load()
 		g_renderer->DeleteTexture(g_shaderCacheLoaderState.textureTVId);
 	if (g_shaderCacheLoaderState.textureDRCId)
 		g_renderer->DeleteTexture(g_shaderCacheLoaderState.textureDRCId);
+
+	g_bootSndPlayer.FadeOutSound();
+
+	if(Latte_GetStopSignal())
+		LatteThread_Exit();
 }
 
 void LatteShaderCache_ShowProgress(const std::function <bool(void)>& loadUpdateFunc, bool isPipelines)
@@ -505,8 +628,6 @@ void LatteShaderCache_LoadVulkanPipelineCache(uint64 cacheTitleId)
 	g_shaderCacheLoaderState.loadedPipelines = 0;
 	LatteShaderCache_ShowProgress(LatteShaderCache_updatePipelineLoadingProgress, true);
 	pipelineCache.EndLoading();
-    if(Latte_GetStopSignal())
-        LatteThread_Exit();
 }
 
 bool LatteShaderCache_updatePipelineLoadingProgress()
@@ -805,4 +926,4 @@ void LatteShaderCache_handleDeprecatedCacheFiles(fs::path pathGeneric, fs::path
 			fs::remove(pathGenericPre1_25_0, ec);
 		}
 	}
-}
+}

src/Cafe/HW/Latte/Core/LatteThread.cpp

@@ -235,6 +235,8 @@ void Latte_Start()
 void Latte_Stop()
 {
 	std::unique_lock _lock(sLatteThreadStateMutex);
+	if (!sLatteThreadRunning)
+		return;
 	sLatteThreadRunning = false;
 	_lock.unlock();
 	sLatteThread.join();
@@ -257,6 +259,7 @@ void LatteThread_Exit()
     LatteSHRC_UnloadAll();
     // close disk cache
     LatteShaderCache_Close();
+	RendererOutputShader::ShutdownStatic();
     // destroy renderer but make sure that g_renderer remains valid until the destructor has finished
 	if (g_renderer)
 	{

src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.cpp

@@ -370,6 +370,8 @@ bool LatteDecompiler_IsALUTransInstruction(bool isOP3, uint32 opcode)
 		opcode == ALU_OP2_INST_LSHR_INT ||
 		opcode == ALU_OP2_INST_MAX_INT ||
 		opcode == ALU_OP2_INST_MIN_INT ||
+		opcode == ALU_OP2_INST_MAX_UINT ||
+		opcode == ALU_OP2_INST_MIN_UINT ||
 		opcode == ALU_OP2_INST_MOVA_FLOOR ||
 		opcode == ALU_OP2_INST_MOVA_INT ||
 		opcode == ALU_OP2_INST_SETE_DX10 ||

src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerAnalyzer.cpp

@@ -140,6 +140,8 @@ bool _isIntegerInstruction(const LatteDecompilerALUInstruction& aluInstruction)
 		case ALU_OP2_INST_SUB_INT:
 		case ALU_OP2_INST_MAX_INT:
 		case ALU_OP2_INST_MIN_INT:
+		case ALU_OP2_INST_MAX_UINT:
+		case ALU_OP2_INST_MIN_UINT:
 		case ALU_OP2_INST_SETE_INT:
 		case ALU_OP2_INST_SETGT_INT:
 		case ALU_OP2_INST_SETGE_INT:

src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitGLSL.cpp

@@ -1415,19 +1415,23 @@ void _emitALUOP2InstructionCode(LatteDecompilerShaderContext* shaderContext, Lat
 	}
 	else if( aluInstruction->opcode == ALU_OP2_INST_ADD_INT )
 		_emitALUOperationBinary<LATTE_DECOMPILER_DTYPE_SIGNED_INT>(shaderContext, aluInstruction, " + ");
-	else if( aluInstruction->opcode == ALU_OP2_INST_MAX_INT || aluInstruction->opcode == ALU_OP2_INST_MIN_INT )
+	else if( aluInstruction->opcode == ALU_OP2_INST_MAX_INT || aluInstruction->opcode == ALU_OP2_INST_MIN_INT ||
+			 aluInstruction->opcode == ALU_OP2_INST_MAX_UINT || aluInstruction->opcode == ALU_OP2_INST_MIN_UINT)
 	{
 		// not verified
+		bool isUnsigned = aluInstruction->opcode == ALU_OP2_INST_MAX_UINT || aluInstruction->opcode == ALU_OP2_INST_MIN_UINT;
+		auto opType = isUnsigned ? LATTE_DECOMPILER_DTYPE_UNSIGNED_INT : LATTE_DECOMPILER_DTYPE_SIGNED_INT;
 		_emitInstructionOutputVariableName(shaderContext, aluInstruction);
-		if( aluInstruction->opcode == ALU_OP2_INST_MAX_INT )
-			src->add(" = max(");
+		src->add(" = ");
+		_emitTypeConversionPrefix(shaderContext, opType, outputType);
+		if( aluInstruction->opcode == ALU_OP2_INST_MAX_INT || aluInstruction->opcode == ALU_OP2_INST_MAX_UINT )
+			src->add("max(");
 		else
-			src->add(" = min(");
-		_emitTypeConversionPrefix(shaderContext, LATTE_DECOMPILER_DTYPE_SIGNED_INT, outputType);
-		_emitOperandInputCode(shaderContext, aluInstruction, 0, LATTE_DECOMPILER_DTYPE_SIGNED_INT);
+			src->add("min(");
+		_emitOperandInputCode(shaderContext, aluInstruction, 0, opType);
 		src->add(", ");
-		_emitOperandInputCode(shaderContext, aluInstruction, 1, LATTE_DECOMPILER_DTYPE_SIGNED_INT);
-		_emitTypeConversionSuffix(shaderContext, LATTE_DECOMPILER_DTYPE_SIGNED_INT, outputType);
+		_emitOperandInputCode(shaderContext, aluInstruction, 1, opType);
+		_emitTypeConversionSuffix(shaderContext, opType, outputType);
 		src->add(");" _CRLF);
 	}
 	else if( aluInstruction->opcode == ALU_OP2_INST_SUB_INT )

src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerInstructions.h

@@ -60,6 +60,8 @@
 #define ALU_OP2_INST_SUB_INT		(0x035) // integer instruction
 #define ALU_OP2_INST_MAX_INT		(0x036) // integer instruction
 #define ALU_OP2_INST_MIN_INT		(0x037) // integer instruction
+#define ALU_OP2_INST_MAX_UINT		(0x038) // integer instruction
+#define ALU_OP2_INST_MIN_UINT		(0x039) // integer instruction
 #define ALU_OP2_INST_SETE_INT		(0x03A) // integer instruction
 #define ALU_OP2_INST_SETGT_INT		(0x03B) // integer instruction
 #define ALU_OP2_INST_SETGE_INT		(0x03C) // integer instruction

src/Cafe/HW/Latte/Renderer/OpenGL/OpenGLRenderer.cpp

@@ -570,13 +570,10 @@ void OpenGLRenderer::DrawBackbufferQuad(LatteTextureView* texView, RendererOutpu
 		g_renderer->ClearColorbuffer(padView);
 	}
 
-	sint32 effectiveWidth, effectiveHeight;
-	texView->baseTexture->GetEffectiveSize(effectiveWidth, effectiveHeight, 0);
-
 	shader_unbind(RendererShader::ShaderType::kGeometry);
 	shader_bind(shader->GetVertexShader());
 	shader_bind(shader->GetFragmentShader());
-	shader->SetUniformParameters(*texView, { effectiveWidth, effectiveHeight }, { imageWidth, imageHeight });
+	shader->SetUniformParameters(*texView, {imageWidth, imageHeight});
 
 	// set viewport
 	glViewportIndexedf(0, imageX, imageY, imageWidth, imageHeight);
@@ -584,6 +581,12 @@ void OpenGLRenderer::DrawBackbufferQuad(LatteTextureView* texView, RendererOutpu
 	LatteTextureViewGL* texViewGL = (LatteTextureViewGL*)texView;
 	texture_bindAndActivate(texView, 0);
 
+	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+	texViewGL->samplerState.clampS = texViewGL->samplerState.clampT = 0xFF;
+
+	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, useLinearTexFilter ? GL_LINEAR : GL_NEAREST);
+	texViewGL->samplerState.filterMin = 0xFFFFFFFF;
 	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, useLinearTexFilter ? GL_LINEAR : GL_NEAREST);
 	texViewGL->samplerState.filterMag = 0xFFFFFFFF;
 

src/Cafe/HW/Latte/Renderer/OpenGL/OpenGLRenderer.h

@@ -102,16 +102,21 @@ public:
 	static void SetAttributeArrayState(uint32 index, bool isEnabled, sint32 aluDivisor);
 	static void SetArrayElementBuffer(GLuint arrayElementBuffer);
 
-	// index
-	void* indexData_reserveIndexMemory(uint32 size, uint32& offset, uint32& bufferIndex) override
+	// index (not used by OpenGL renderer yet)
+	IndexAllocation indexData_reserveIndexMemory(uint32 size) override
 	{
-		assert_dbg();
-		return nullptr;
+		cemu_assert_unimplemented();
+		return {};
 	}
 
-	void indexData_uploadIndexMemory(uint32 offset, uint32 size) override
+	void indexData_releaseIndexMemory(IndexAllocation& allocation) override
 	{
-		assert_dbg();
+		cemu_assert_unimplemented();
+	}
+
+	void indexData_uploadIndexMemory(IndexAllocation& allocation) override
+	{
+		cemu_assert_unimplemented();
 	}
 
 	// uniform

src/Cafe/HW/Latte/Renderer/Renderer.h

@@ -138,8 +138,15 @@ public:
 	virtual void draw_endSequence() = 0;
 
 	// index
-	virtual void* indexData_reserveIndexMemory(uint32 size, uint32& offset, uint32& bufferIndex) = 0;
-	virtual void indexData_uploadIndexMemory(uint32 offset, uint32 size) = 0;
+	struct IndexAllocation
+	{
+		void* mem; // pointer to index data inside buffer
+		void* rendererInternal; // for renderer use
+	};
+
+	virtual IndexAllocation indexData_reserveIndexMemory(uint32 size) = 0;
+	virtual void indexData_releaseIndexMemory(IndexAllocation& allocation) = 0;
+	virtual void indexData_uploadIndexMemory(IndexAllocation& allocation) = 0;
 
 	// occlusion queries
 	virtual LatteQueryObject* occlusionQuery_create() = 0;

src/Cafe/HW/Latte/Renderer/RendererOuputShader.cpp

@@ -2,18 +2,7 @@
 #include "Cafe/HW/Latte/Renderer/OpenGL/OpenGLRenderer.h"
 
 const std::string RendererOutputShader::s_copy_shader_source =
-R"(#version 420
-
-#ifdef VULKAN
-layout(location = 0) in vec2 passUV;
-layout(binding = 0) uniform sampler2D textureSrc;
-layout(location = 0) out vec4 colorOut0;
-#else
-in vec2 passUV;
-layout(binding=0) uniform sampler2D textureSrc;
-layout(location = 0) out vec4 colorOut0;
-#endif
-
+R"(
 void main()
 {
 	colorOut0 = vec4(texture(textureSrc, passUV).rgb,1.0);
@@ -22,20 +11,6 @@ void main()
 
 const std::string RendererOutputShader::s_bicubic_shader_source =
 R"(
-#version 420
-
-#ifdef VULKAN
-layout(location = 0) in vec2 passUV;
-layout(binding = 0)  uniform sampler2D textureSrc;
-layout(binding = 1)  uniform vec2 textureSrcResolution;
-layout(location = 0) out vec4 colorOut0;
-#else
-in vec2 passUV;
-layout(binding=0) uniform sampler2D textureSrc;
-uniform vec2 textureSrcResolution;
-layout(location = 0) out vec4 colorOut0;
-#endif
-
 vec4 cubic(float x)
 {
 	float x2 = x * x;
@@ -48,24 +23,23 @@ vec4 cubic(float x)
 	return w / 6.0;
 }
 
-vec4 bcFilter(vec2 texcoord, vec2 texscale)
+vec4 bcFilter(vec2 uv, vec4 texelSize)
 {
-	float fx = fract(texcoord.x);
-	float fy = fract(texcoord.y);
-	texcoord.x -= fx;
-	texcoord.y -= fy;
+	vec2 pixel = uv*texelSize.zw - 0.5;
+	vec2 pixelFrac = fract(pixel);
+	vec2 pixelInt = pixel - pixelFrac;
 
-	vec4 xcubic = cubic(fx);
-	vec4 ycubic = cubic(fy);
+	vec4 xcubic = cubic(pixelFrac.x);
+	vec4 ycubic = cubic(pixelFrac.y);
 
-	vec4 c = vec4(texcoord.x - 0.5, texcoord.x + 1.5, texcoord.y - 0.5, texcoord.y + 1.5);
+	vec4 c = vec4(pixelInt.x - 0.5, pixelInt.x + 1.5, pixelInt.y - 0.5, pixelInt.y + 1.5);
 	vec4 s = vec4(xcubic.x + xcubic.y, xcubic.z + xcubic.w, ycubic.x + ycubic.y, ycubic.z + ycubic.w);
 	vec4 offset = c + vec4(xcubic.y, xcubic.w, ycubic.y, ycubic.w) / s;
 
-	vec4 sample0 = texture(textureSrc, vec2(offset.x, offset.z) * texscale);
-	vec4 sample1 = texture(textureSrc, vec2(offset.y, offset.z) * texscale);
-	vec4 sample2 = texture(textureSrc, vec2(offset.x, offset.w) * texscale);
-	vec4 sample3 = texture(textureSrc, vec2(offset.y, offset.w) * texscale);
+	vec4 sample0 = texture(textureSrc, vec2(offset.x, offset.z) * texelSize.xy);
+	vec4 sample1 = texture(textureSrc, vec2(offset.y, offset.z) * texelSize.xy);
+	vec4 sample2 = texture(textureSrc, vec2(offset.x, offset.w) * texelSize.xy);
+	vec4 sample3 = texture(textureSrc, vec2(offset.y, offset.w) * texelSize.xy);
 
 	float sx = s.x / (s.x + s.y);
 	float sy = s.z / (s.z + s.w);
@@ -76,20 +50,13 @@ vec4 bcFilter(vec2 texcoord, vec2 texscale)
 }
 
 void main(){
-	colorOut0 = vec4(bcFilter(passUV*textureSrcResolution, vec2(1.0,1.0)/textureSrcResolution).rgb,1.0);
+	vec4 texelSize = vec4( 1.0 / textureSrcResolution.xy, textureSrcResolution.xy);
+	colorOut0 = vec4(bcFilter(passUV, texelSize).rgb,1.0);
 }
 )";
 
 const std::string RendererOutputShader::s_hermite_shader_source =
-R"(#version 420
-
-in vec4 gl_FragCoord;	
-in vec2 passUV;
-layout(binding=0) uniform sampler2D textureSrc;
-uniform vec2 textureSrcResolution;
-uniform vec2 outputResolution;
-layout(location = 0) out vec4 colorOut0;
-
+R"(
 // https://www.shadertoy.com/view/MllSzX
 
 vec3 CubicHermite (vec3 A, vec3 B, vec3 C, vec3 D, float t)
@@ -111,7 +78,7 @@ vec3 BicubicHermiteTexture(vec2 uv, vec4 texelSize)
 	vec2 frac = fract(pixel);	
     pixel = floor(pixel) / texelSize.zw - vec2(texelSize.xy/2.0);
 	
-	vec4 doubleSize = texelSize*texelSize;
+	vec4 doubleSize = texelSize*2.0;
 
 	vec3 C00 = texture(textureSrc, pixel + vec2(-texelSize.x ,-texelSize.y)).rgb;
     vec3 C10 = texture(textureSrc, pixel + vec2( 0.0        ,-texelSize.y)).rgb;
@@ -142,15 +109,17 @@ vec3 BicubicHermiteTexture(vec2 uv, vec4 texelSize)
 }
 
 void main(){
-	vec4 texelSize = vec4( 1.0 / outputResolution.xy, outputResolution.xy);
+	vec4 texelSize = vec4( 1.0 / textureSrcResolution.xy, textureSrcResolution.xy);
 	colorOut0 = vec4(BicubicHermiteTexture(passUV, texelSize), 1.0);
 }
 )";
 
 RendererOutputShader::RendererOutputShader(const std::string& vertex_source, const std::string& fragment_source)
 {
-	m_vertex_shader = g_renderer->shader_create(RendererShader::ShaderType::kVertex, 0, 0, vertex_source, false, false);
-	m_fragment_shader = g_renderer->shader_create(RendererShader::ShaderType::kFragment, 0, 0, fragment_source, false, false);
+	auto finalFragmentSrc = PrependFragmentPreamble(fragment_source);
+
+	m_vertex_shader.reset(g_renderer->shader_create(RendererShader::ShaderType::kVertex, 0, 0, vertex_source, false, false));
+	m_fragment_shader.reset(g_renderer->shader_create(RendererShader::ShaderType::kFragment, 0, 0, finalFragmentSrc, false, false));
 
 	m_vertex_shader->PreponeCompilation(true);
 	m_fragment_shader->PreponeCompilation(true);
@@ -163,74 +132,45 @@ RendererOutputShader::RendererOutputShader(const std::string& vertex_source, con
 
 	if (g_renderer->GetType() == RendererAPI::OpenGL)
 	{
-		m_attributes[0].m_loc_texture_src_resolution = m_vertex_shader->GetUniformLocation("textureSrcResolution");
-		m_attributes[0].m_loc_input_resolution = m_vertex_shader->GetUniformLocation("inputResolution");
-		m_attributes[0].m_loc_output_resolution = m_vertex_shader->GetUniformLocation("outputResolution");
+		m_uniformLocations[0].m_loc_textureSrcResolution = m_vertex_shader->GetUniformLocation("textureSrcResolution");
+		m_uniformLocations[0].m_loc_nativeResolution = m_vertex_shader->GetUniformLocation("nativeResolution");
+		m_uniformLocations[0].m_loc_outputResolution = m_vertex_shader->GetUniformLocation("outputResolution");
 
-		m_attributes[1].m_loc_texture_src_resolution = m_fragment_shader->GetUniformLocation("textureSrcResolution");
-		m_attributes[1].m_loc_input_resolution = m_fragment_shader->GetUniformLocation("inputResolution");
-		m_attributes[1].m_loc_output_resolution = m_fragment_shader->GetUniformLocation("outputResolution");
+		m_uniformLocations[1].m_loc_textureSrcResolution = m_fragment_shader->GetUniformLocation("textureSrcResolution");
+		m_uniformLocations[1].m_loc_nativeResolution = m_fragment_shader->GetUniformLocation("nativeResolution");
+		m_uniformLocations[1].m_loc_outputResolution = m_fragment_shader->GetUniformLocation("outputResolution");
 	}
-	else
-	{
-		cemuLog_logDebug(LogType::Force, "RendererOutputShader() - todo for Vulkan");
-		m_attributes[0].m_loc_texture_src_resolution = -1;
-		m_attributes[0].m_loc_input_resolution = -1;
-		m_attributes[0].m_loc_output_resolution = -1;
-
-		m_attributes[1].m_loc_texture_src_resolution = -1;
-		m_attributes[1].m_loc_input_resolution = -1;
-		m_attributes[1].m_loc_output_resolution = -1;
-	}
-
 }
 
-void RendererOutputShader::SetUniformParameters(const LatteTextureView& texture_view, const Vector2i& input_res, const Vector2i& output_res) const
+void RendererOutputShader::SetUniformParameters(const LatteTextureView& texture_view, const Vector2i& output_res) const
 {
-	float res[2];
-	// vertex shader
-	if (m_attributes[0].m_loc_texture_src_resolution != -1)
-	{ 
-		res[0] = (float)texture_view.baseTexture->width;
-		res[1] = (float)texture_view.baseTexture->height;
-		m_vertex_shader->SetUniform2fv(m_attributes[0].m_loc_texture_src_resolution, res, 1);
-	}
+	sint32 effectiveWidth, effectiveHeight;
+	texture_view.baseTexture->GetEffectiveSize(effectiveWidth, effectiveHeight, 0);
+	auto setUniforms = [&](RendererShader* shader, const UniformLocations& locations){
+	  float res[2];
+	  if (locations.m_loc_textureSrcResolution != -1)
+	  {
+		  res[0] = (float)effectiveWidth;
+		  res[1] = (float)effectiveHeight;
+		  shader->SetUniform2fv(locations.m_loc_textureSrcResolution, res, 1);
+	  }
 
-	if (m_attributes[0].m_loc_input_resolution != -1)
-	{
-		res[0] = (float)input_res.x;
-		res[1] = (float)input_res.y;
-		m_vertex_shader->SetUniform2fv(m_attributes[0].m_loc_input_resolution, res, 1);
-	}
+	  if (locations.m_loc_nativeResolution != -1)
+	  {
+		  res[0] = (float)texture_view.baseTexture->width;
+		  res[1] = (float)texture_view.baseTexture->height;
+		  shader->SetUniform2fv(locations.m_loc_nativeResolution, res, 1);
+	  }
 
-	if (m_attributes[0].m_loc_output_resolution != -1)
-	{
-		res[0] = (float)output_res.x;
-		res[1] = (float)output_res.y;
-		m_vertex_shader->SetUniform2fv(m_attributes[0].m_loc_output_resolution, res, 1);
-	}
-
-	// fragment shader
-	if (m_attributes[1].m_loc_texture_src_resolution != -1)
-	{
-		res[0] = (float)texture_view.baseTexture->width;
-		res[1] = (float)texture_view.baseTexture->height;
-		m_fragment_shader->SetUniform2fv(m_attributes[1].m_loc_texture_src_resolution, res, 1);
-	}
-
-	if (m_attributes[1].m_loc_input_resolution != -1)
-	{
-		res[0] = (float)input_res.x;
-		res[1] = (float)input_res.y;
-		m_fragment_shader->SetUniform2fv(m_attributes[1].m_loc_input_resolution, res, 1);
-	}
-
-	if (m_attributes[1].m_loc_output_resolution != -1)
-	{
-		res[0] = (float)output_res.x;
-		res[1] = (float)output_res.y;
-		m_fragment_shader->SetUniform2fv(m_attributes[1].m_loc_output_resolution, res, 1);
-	}
+	  if (locations.m_loc_outputResolution != -1)
+	  {
+		  res[0] = (float)output_res.x;
+		  res[1] = (float)output_res.y;
+		  shader->SetUniform2fv(locations.m_loc_outputResolution, res, 1);
+	  }
+	};
+	setUniforms(m_vertex_shader.get(), m_uniformLocations[0]);
+	setUniforms(m_fragment_shader.get(), m_uniformLocations[1]);
 }
 
 RendererOutputShader* RendererOutputShader::s_copy_shader;
@@ -247,8 +187,8 @@ std::string RendererOutputShader::GetOpenGlVertexSource(bool render_upside_down)
 	// vertex shader
 	std::ostringstream vertex_source;
 		vertex_source <<
-			R"(#version 400
-out vec2 passUV;
+			R"(#version 420
+layout(location = 0) smooth out vec2 passUV;
 
 out gl_PerVertex 
 { 
@@ -341,6 +281,27 @@ void main(){
 )";
 		return vertex_source.str();
 }
+
+std::string RendererOutputShader::PrependFragmentPreamble(const std::string& shaderSrc)
+{
+	return R"(#version 430
+#ifdef VULKAN
+layout(push_constant) uniform pc {
+	vec2 textureSrcResolution;
+	vec2 nativeResolution;
+	vec2 outputResolution;
+};
+#else
+uniform vec2 textureSrcResolution;
+uniform vec2 nativeResolution;
+uniform vec2 outputResolution;
+#endif
+
+layout(location = 0) smooth in vec2 passUV;
+layout(binding = 0) uniform sampler2D textureSrc;
+layout(location = 0) out vec4 colorOut0;
+)" + shaderSrc;
+}
 void RendererOutputShader::InitializeStatic()
 {
 	std::string vertex_source, vertex_source_ud;
@@ -349,28 +310,30 @@ void RendererOutputShader::InitializeStatic()
 	{
 		vertex_source = GetOpenGlVertexSource(false);
 		vertex_source_ud = GetOpenGlVertexSource(true);
-
-		s_copy_shader = new RendererOutputShader(vertex_source, s_copy_shader_source);
-		s_copy_shader_ud = new RendererOutputShader(vertex_source_ud, s_copy_shader_source);
-
-		s_bicubic_shader = new RendererOutputShader(vertex_source, s_bicubic_shader_source);
-		s_bicubic_shader_ud = new RendererOutputShader(vertex_source_ud, s_bicubic_shader_source);
-
-		s_hermit_shader = new RendererOutputShader(vertex_source, s_hermite_shader_source);
-		s_hermit_shader_ud = new RendererOutputShader(vertex_source_ud, s_hermite_shader_source);
 	}
 	else
 	{
 		vertex_source = GetVulkanVertexSource(false);
 		vertex_source_ud = GetVulkanVertexSource(true);
-
-		s_copy_shader = new RendererOutputShader(vertex_source, s_copy_shader_source);
-		s_copy_shader_ud = new RendererOutputShader(vertex_source_ud, s_copy_shader_source);
-
-	/*	s_bicubic_shader = new RendererOutputShader(vertex_source, s_bicubic_shader_source); TODO
-		s_bicubic_shader_ud = new RendererOutputShader(vertex_source_ud, s_bicubic_shader_source);
-
-		s_hermit_shader = new RendererOutputShader(vertex_source, s_hermite_shader_source);
-		s_hermit_shader_ud = new RendererOutputShader(vertex_source_ud, s_hermite_shader_source);*/
 	}
+	s_copy_shader = new RendererOutputShader(vertex_source, s_copy_shader_source);
+	s_copy_shader_ud = new RendererOutputShader(vertex_source_ud, s_copy_shader_source);
+
+	s_bicubic_shader = new RendererOutputShader(vertex_source, s_bicubic_shader_source);
+	s_bicubic_shader_ud = new RendererOutputShader(vertex_source_ud, s_bicubic_shader_source);
+
+	s_hermit_shader = new RendererOutputShader(vertex_source, s_hermite_shader_source);
+	s_hermit_shader_ud = new RendererOutputShader(vertex_source_ud, s_hermite_shader_source);
+}
+
+void RendererOutputShader::ShutdownStatic()
+{
+	delete s_copy_shader;
+	delete s_copy_shader_ud;
+
+	delete s_bicubic_shader;
+	delete s_bicubic_shader_ud;
+
+	delete s_hermit_shader;
+	delete s_hermit_shader_ud;
 }

src/Cafe/HW/Latte/Renderer/RendererOuputShader.h

@@ -17,19 +17,20 @@ public:
 	RendererOutputShader(const std::string& vertex_source, const std::string& fragment_source);
 	virtual ~RendererOutputShader() = default;
 
-	void SetUniformParameters(const LatteTextureView& texture_view, const Vector2i& input_res, const Vector2i& output_res) const;
+	void SetUniformParameters(const LatteTextureView& texture_view, const Vector2i& output_res) const;
 
 	RendererShader* GetVertexShader() const
 	{
-		return m_vertex_shader;
+		return m_vertex_shader.get();
 	}
 
 	RendererShader* GetFragmentShader() const
 	{
-		return m_fragment_shader;
+		return m_fragment_shader.get();
 	}
 
 	static void InitializeStatic();
+	static void ShutdownStatic();
 
 	static RendererOutputShader* s_copy_shader;
 	static RendererOutputShader* s_copy_shader_ud;
@@ -43,16 +44,18 @@ public:
 	static std::string GetVulkanVertexSource(bool render_upside_down);
 	static std::string GetOpenGlVertexSource(bool render_upside_down);
 
-protected:
-	RendererShader* m_vertex_shader;
-	RendererShader* m_fragment_shader;
+	static std::string PrependFragmentPreamble(const std::string& shaderSrc);
 
-	struct
+protected:
+	std::unique_ptr<RendererShader> m_vertex_shader;
+	std::unique_ptr<RendererShader> m_fragment_shader;
+
+	struct UniformLocations
 	{
-		sint32 m_loc_texture_src_resolution = -1;
-		sint32 m_loc_input_resolution = -1;
-		sint32 m_loc_output_resolution = -1;
-	} m_attributes[2]{};
+		sint32 m_loc_textureSrcResolution = -1;
+		sint32 m_loc_nativeResolution = -1;
+		sint32 m_loc_outputResolution = -1;
+	} m_uniformLocations[2]{};
 
 private:
 	static const std::string s_copy_shader_source;

src/Cafe/HW/Latte/Renderer/Vulkan/LatteTextureViewVk.cpp

@@ -202,6 +202,13 @@ VkSampler LatteTextureViewVk::GetDefaultTextureSampler(bool useLinearTexFilter)
 	VkSamplerCreateInfo samplerInfo{};
 	samplerInfo.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO;
 
+	// emulate OpenGL minFilters
+	// see note under: https://docs.vulkan.org/spec/latest/chapters/samplers.html#VkSamplerCreateInfo
+	// if maxLod = 0 then magnification is always performed
+	samplerInfo.mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST;
+	samplerInfo.minLod = 0.0f;
+	samplerInfo.maxLod = 0.25f;
+
 	if (useLinearTexFilter)
 	{
 		samplerInfo.magFilter = VK_FILTER_LINEAR;
@@ -212,6 +219,9 @@ VkSampler LatteTextureViewVk::GetDefaultTextureSampler(bool useLinearTexFilter)
 		samplerInfo.magFilter = VK_FILTER_NEAREST;
 		samplerInfo.minFilter = VK_FILTER_NEAREST;
 	}
+	samplerInfo.addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
+	samplerInfo.addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
+	samplerInfo.addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
 
 	if (vkCreateSampler(m_device, &samplerInfo, nullptr, &sampler) != VK_SUCCESS)
 	{

src/Cafe/HW/Latte/Renderer/Vulkan/RendererShaderVk.cpp

@@ -211,6 +211,9 @@ RendererShaderVk::~RendererShaderVk()
 {
 	while (!list_pipelineInfo.empty())
 		delete list_pipelineInfo[0];
+
+	VkDevice vkDev = VulkanRenderer::GetInstance()->GetLogicalDevice();
+	vkDestroyShaderModule(vkDev, m_shader_module, nullptr);
 }
 
 void RendererShaderVk::Init()

src/Cafe/HW/Latte/Renderer/Vulkan/SwapchainInfoVk.cpp

@@ -60,7 +60,7 @@ void SwapchainInfoVk::Create()
 	VkAttachmentDescription colorAttachment = {};
 	colorAttachment.format = m_surfaceFormat.format;
 	colorAttachment.samples = VK_SAMPLE_COUNT_1_BIT;
-	colorAttachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
+	colorAttachment.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 	colorAttachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
 	colorAttachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 	colorAttachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
@@ -146,8 +146,17 @@ void SwapchainInfoVk::Create()
 			UnrecoverableError("Failed to create semaphore for swapchain acquire");
 	}
 
+	VkFenceCreateInfo fenceInfo = {};
+	fenceInfo.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO;
+	fenceInfo.flags = VK_FENCE_CREATE_SIGNALED_BIT;
+	result = vkCreateFence(m_logicalDevice, &fenceInfo, nullptr, &m_imageAvailableFence);
+	if (result != VK_SUCCESS)
+		UnrecoverableError("Failed to create fence for swapchain");
+
 	m_acquireIndex = 0;
 	hasDefinedSwapchainImage = false;
+
+	m_queueDepth = 0;
 }
 
 void SwapchainInfoVk::Cleanup()
@@ -177,6 +186,12 @@ void SwapchainInfoVk::Cleanup()
 	m_swapchainFramebuffers.clear();
 
 
+	if (m_imageAvailableFence)
+	{
+		WaitAvailableFence();
+		vkDestroyFence(m_logicalDevice, m_imageAvailableFence, nullptr);
+		m_imageAvailableFence = nullptr;
+	}
 	if (m_swapchain)
 	{
 		vkDestroySwapchainKHR(m_logicalDevice, m_swapchain, nullptr);
@@ -189,6 +204,18 @@ bool SwapchainInfoVk::IsValid() const
 	return m_swapchain && !m_acquireSemaphores.empty();
 }
 
+void SwapchainInfoVk::WaitAvailableFence()
+{
+	if(m_awaitableFence != VK_NULL_HANDLE)
+		vkWaitForFences(m_logicalDevice, 1, &m_awaitableFence, VK_TRUE, UINT64_MAX);
+	m_awaitableFence = VK_NULL_HANDLE;
+}
+
+void SwapchainInfoVk::ResetAvailableFence() const
+{
+	vkResetFences(m_logicalDevice, 1, &m_imageAvailableFence);
+}
+
 VkSemaphore SwapchainInfoVk::ConsumeAcquireSemaphore()
 {
 	VkSemaphore ret = m_currentSemaphore;
@@ -198,8 +225,10 @@ VkSemaphore SwapchainInfoVk::ConsumeAcquireSemaphore()
 
 bool SwapchainInfoVk::AcquireImage()
 {
+	ResetAvailableFence();
+
 	VkSemaphore acquireSemaphore = m_acquireSemaphores[m_acquireIndex];
-	VkResult result = vkAcquireNextImageKHR(m_logicalDevice, m_swapchain, 1'000'000'000, acquireSemaphore, nullptr, &swapchainImageIndex);
+	VkResult result = vkAcquireNextImageKHR(m_logicalDevice, m_swapchain, 1'000'000'000, acquireSemaphore, m_imageAvailableFence, &swapchainImageIndex);
 	if (result == VK_ERROR_OUT_OF_DATE_KHR || result == VK_SUBOPTIMAL_KHR)
 		m_shouldRecreate = true;
 	if (result == VK_TIMEOUT)
@@ -216,6 +245,7 @@ bool SwapchainInfoVk::AcquireImage()
 		return false;
 	}
 	m_currentSemaphore = acquireSemaphore;
+	m_awaitableFence = m_imageAvailableFence;
 	m_acquireIndex = (m_acquireIndex + 1) % m_swapchainImages.size();
 
 	return true;
@@ -319,6 +349,7 @@ VkExtent2D SwapchainInfoVk::ChooseSwapExtent(const VkSurfaceCapabilitiesKHR& cap
 
 VkPresentModeKHR SwapchainInfoVk::ChoosePresentMode(const std::vector<VkPresentModeKHR>& modes)
 {
+	m_maxQueued = 0;
 	const auto vsyncState = (VSync)GetConfig().vsync.GetValue();
 	if (vsyncState == VSync::MAILBOX)
 	{
@@ -345,6 +376,7 @@ VkPresentModeKHR SwapchainInfoVk::ChoosePresentMode(const std::vector<VkPresentM
 		return VK_PRESENT_MODE_FIFO_KHR;
 	}
 
+	m_maxQueued = 1;
 	return VK_PRESENT_MODE_FIFO_KHR;
 }
 

src/Cafe/HW/Latte/Renderer/Vulkan/SwapchainInfoVk.h

@@ -26,6 +26,9 @@ struct SwapchainInfoVk
 
 	bool IsValid() const;
 
+	void WaitAvailableFence();
+	void ResetAvailableFence() const;
+
 	bool AcquireImage();
 	// retrieve semaphore of last acquire for submitting a wait operation
 	// only one wait operation must be submitted per acquire (which submits a single signal operation)
@@ -67,7 +70,11 @@ struct SwapchainInfoVk
 	VkSurfaceFormatKHR m_surfaceFormat{};
 	VkSwapchainKHR m_swapchain{};
 	Vector2i m_desiredExtent{};
+	VkExtent2D m_actualExtent{};
 	uint32 swapchainImageIndex = (uint32)-1;
+	uint64 m_presentId = 1;
+	uint64 m_queueDepth = 0; // number of frames with pending presentation requests
+	uint64 m_maxQueued = 0; // the maximum number of frames with presentation requests.
 
 
 	// swapchain image ringbuffer (indexed by swapchainImageIndex)
@@ -81,8 +88,9 @@ struct SwapchainInfoVk
 private:
 	uint32 m_acquireIndex = 0;
 	std::vector<VkSemaphore> m_acquireSemaphores; // indexed by m_acquireIndex
+	VkFence m_imageAvailableFence{};
+	VkFence m_awaitableFence = VK_NULL_HANDLE;
 	VkSemaphore m_currentSemaphore = VK_NULL_HANDLE;
 
 	std::array<uint32, 2> m_swapchainQueueFamilyIndices;
-	VkExtent2D m_actualExtent{};
 };

src/Cafe/HW/Latte/Renderer/Vulkan/TextureReadbackVk.cpp

@@ -22,7 +22,7 @@ uint32 LatteTextureReadbackInfoVk::GetImageSize(LatteTextureView* textureView)
 		cemu_assert(textureFormat == VK_FORMAT_R8G8B8A8_UNORM);
 		return baseTexture->width * baseTexture->height * 4;
 	}
-	else if (textureView->format == Latte::E_GX2SURFFMT::R8_UNORM)
+	else if (textureView->format == Latte::E_GX2SURFFMT::R8_UNORM )
 	{
 		cemu_assert(textureFormat == VK_FORMAT_R8_UNORM);
 		return baseTexture->width * baseTexture->height * 1;
@@ -79,6 +79,13 @@ uint32 LatteTextureReadbackInfoVk::GetImageSize(LatteTextureView* textureView)
 		// todo - if driver does not support VK_FORMAT_D24_UNORM_S8_UINT this is represented as VK_FORMAT_D32_SFLOAT_S8_UINT which is 8 bytes
 		return baseTexture->width * baseTexture->height * 4;
 	}
+	else if (textureView->format == Latte::E_GX2SURFFMT::R5_G6_B5_UNORM )
+	{
+		if(textureFormat == VK_FORMAT_R5G6B5_UNORM_PACK16){
+			return baseTexture->width * baseTexture->height * 2;
+		}	
+		return 0;
+	}
 	else
 	{
 		cemuLog_log(LogType::Force, "Unsupported texture readback format {:04x}", (uint32)textureView->format);

src/Cafe/HW/Latte/Renderer/Vulkan/VKRBase.h

@@ -19,7 +19,7 @@ public:
 
 	virtual ~VKRMoveableRefCounter()
 	{
-		cemu_assert_debug(refCount == 0);
+		cemu_assert_debug(m_refCount == 0);
 
 		// remove references
 #ifdef CEMU_DEBUG_ASSERT
@@ -30,7 +30,11 @@ public:
 		}
 #endif
 		for (auto itr : refs)
-			itr->ref->refCount--;
+		{
+			itr->ref->m_refCount--;
+			if (itr->ref->m_refCount == 0)
+				itr->ref->RefCountReachedZero();
+		}
 		refs.clear();
 		delete selfRef;
 		selfRef = nullptr;
@@ -41,8 +45,8 @@ public:
 	VKRMoveableRefCounter(VKRMoveableRefCounter&& rhs) noexcept
 	{
 		this->refs = std::move(rhs.refs);
-		this->refCount = rhs.refCount;
-		rhs.refCount = 0;
+		this->m_refCount = rhs.m_refCount;
+		rhs.m_refCount = 0;
 		this->selfRef = rhs.selfRef;
 		rhs.selfRef = nullptr;
 		this->selfRef->ref = this;
@@ -57,7 +61,7 @@ public:
 	void addRef(VKRMoveableRefCounter* refTarget)
 	{
 		this->refs.emplace_back(refTarget->selfRef);
-		refTarget->refCount++;
+		refTarget->m_refCount++;
 
 #ifdef CEMU_DEBUG_ASSERT
 		// add reverse ref
@@ -68,16 +72,23 @@ public:
 	// methods to directly increment/decrement ref counter (for situations where no external object is available)
 	void incRef()
 	{
-		this->refCount++;
+		m_refCount++;
 	}
 
 	void decRef()
 	{
-		this->refCount--;
+		m_refCount--;
+		if (m_refCount == 0)
+			RefCountReachedZero();
 	}
 
 protected:
-	int refCount{};
+  	virtual void RefCountReachedZero()
+	{
+		// does nothing by default
+	}
+
+	int m_refCount{};
 private:
 	VKRMoveableRefCounterRef* selfRef;
 	std::vector<VKRMoveableRefCounterRef*> refs;
@@ -88,7 +99,7 @@ private:
 	void moveObj(VKRMoveableRefCounter&& rhs)
 	{
 		this->refs = std::move(rhs.refs);
-		this->refCount = rhs.refCount;
+		this->m_refCount = rhs.m_refCount;
 		this->selfRef = rhs.selfRef;
 		this->selfRef->ref = this;
 	}
@@ -131,6 +142,25 @@ public:
 	VkSampler m_textureDefaultSampler[2] = { VK_NULL_HANDLE, VK_NULL_HANDLE }; // relict from LatteTextureViewVk, get rid of it eventually
 };
 
+
+class VKRObjectSampler : public VKRDestructibleObject
+{
+  public:
+	VKRObjectSampler(VkSamplerCreateInfo* samplerInfo);
+	~VKRObjectSampler() override;
+
+	static VKRObjectSampler* GetOrCreateSampler(VkSamplerCreateInfo* samplerInfo);
+	static void DestroyCache();
+
+	void RefCountReachedZero() override; // sampler objects are destroyed when not referenced anymore
+
+	VkSampler GetSampler() const { return m_sampler; }
+  private:
+	static std::unordered_map<uint64, VKRObjectSampler*> s_samplerCache;
+	VkSampler m_sampler{ VK_NULL_HANDLE };
+	uint64 m_hash;
+};
+
 class VKRObjectRenderPass : public VKRDestructibleObject
 {
 public:
@@ -191,11 +221,14 @@ public:
 	VKRObjectPipeline();
 	~VKRObjectPipeline() override;
 
-	void setPipeline(VkPipeline newPipeline);
+	void SetPipeline(VkPipeline newPipeline);
+	VkPipeline GetPipeline() const { return m_pipeline; }
 
-	VkPipeline pipeline = VK_NULL_HANDLE;
-	VkDescriptorSetLayout vertexDSL = VK_NULL_HANDLE, pixelDSL = VK_NULL_HANDLE, geometryDSL = VK_NULL_HANDLE;
-	VkPipelineLayout pipeline_layout = VK_NULL_HANDLE;
+	VkDescriptorSetLayout m_vertexDSL = VK_NULL_HANDLE, m_pixelDSL = VK_NULL_HANDLE, m_geometryDSL = VK_NULL_HANDLE;
+	VkPipelineLayout m_pipelineLayout = VK_NULL_HANDLE;
+
+private:
+	VkPipeline m_pipeline = VK_NULL_HANDLE;
 };
 
 class VKRObjectDescriptorSet : public VKRDestructibleObject

src/Cafe/HW/Latte/Renderer/Vulkan/VKRMemoryManager.cpp

@@ -4,6 +4,14 @@
 
 /* VKRSynchronizedMemoryBuffer */
 
+VKRSynchronizedRingAllocator::~VKRSynchronizedRingAllocator()
+{
+	for(auto& buf : m_buffers)
+	{
+		m_vkrMemMgr->DeleteBuffer(buf.vk_buffer, buf.vk_mem);
+	}
+}
+
 void VKRSynchronizedRingAllocator::addUploadBufferSyncPoint(AllocatorBuffer_t& buffer, uint32 offset)
 {
 	auto cmdBufferId = m_vkr->GetCurrentCommandBufferId();
@@ -23,11 +31,11 @@ void VKRSynchronizedRingAllocator::allocateAdditionalUploadBuffer(uint32 sizeReq
 	AllocatorBuffer_t newBuffer{};
 	newBuffer.writeIndex = 0;
 	newBuffer.basePtr = nullptr;
-	if (m_bufferType == BUFFER_TYPE::STAGING)
+	if (m_bufferType == VKR_BUFFER_TYPE::STAGING)
 		m_vkrMemMgr->CreateBuffer(bufferAllocSize, VK_BUFFER_USAGE_TRANSFER_SRC_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT, newBuffer.vk_buffer, newBuffer.vk_mem);
-	else if (m_bufferType == BUFFER_TYPE::INDEX)
+	else if (m_bufferType == VKR_BUFFER_TYPE::INDEX)
 		m_vkrMemMgr->CreateBuffer(bufferAllocSize, VK_BUFFER_USAGE_INDEX_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT, newBuffer.vk_buffer, newBuffer.vk_mem);
-	else if (m_bufferType == BUFFER_TYPE::STRIDE)
+	else if (m_bufferType == VKR_BUFFER_TYPE::STRIDE)
 		m_vkrMemMgr->CreateBuffer(bufferAllocSize, VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT, newBuffer.vk_buffer, newBuffer.vk_mem);
 	else
 		cemu_assert_debug(false);
@@ -53,7 +61,7 @@ VKRSynchronizedRingAllocator::AllocatorReservation_t VKRSynchronizedRingAllocato
 		uint32 distanceToSyncPoint;
 		if (!itr.queue_syncPoints.empty())
 		{
-			if(itr.queue_syncPoints.front().offset < itr.writeIndex)
+			if (itr.queue_syncPoints.front().offset < itr.writeIndex)
 				distanceToSyncPoint = 0xFFFFFFFF;
 			else
 				distanceToSyncPoint = itr.queue_syncPoints.front().offset - itr.writeIndex;
@@ -100,7 +108,7 @@ VKRSynchronizedRingAllocator::AllocatorReservation_t VKRSynchronizedRingAllocato
 
 void VKRSynchronizedRingAllocator::FlushReservation(AllocatorReservation_t& uploadReservation)
 {
-	cemu_assert_debug(m_bufferType == BUFFER_TYPE::STAGING); // only the staging buffer isn't coherent
+	cemu_assert_debug(m_bufferType == VKR_BUFFER_TYPE::STAGING); // only the staging buffer isn't coherent
 	// todo - use nonCoherentAtomSize for flush size (instead of hardcoded constant)
 	VkMappedMemoryRange flushedRange{};
 	flushedRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE;
@@ -167,15 +175,88 @@ void VKRSynchronizedRingAllocator::GetStats(uint32& numBuffers, size_t& totalBuf
 	}
 }
 
+/* VKRSynchronizedHeapAllocator */
+
+VKRSynchronizedHeapAllocator::VKRSynchronizedHeapAllocator(class VKRMemoryManager* vkMemoryManager, VKR_BUFFER_TYPE bufferType, size_t minimumBufferAllocSize)
+	: m_vkrMemMgr(vkMemoryManager), m_chunkedHeap(bufferType, minimumBufferAllocSize) {};
+
+VKRSynchronizedHeapAllocator::AllocatorReservation* VKRSynchronizedHeapAllocator::AllocateBufferMemory(uint32 size, uint32 alignment)
+{
+	CHAddr addr = m_chunkedHeap.alloc(size, alignment);
+	m_activeAllocations.emplace_back(addr);
+	AllocatorReservation* res = m_poolAllocatorReservation.allocObj();
+	res->bufferIndex = addr.chunkIndex;
+	res->bufferOffset = addr.offset;
+	res->size = size;
+	res->memPtr = m_chunkedHeap.GetChunkPtr(addr.chunkIndex) + addr.offset;
+	m_chunkedHeap.GetChunkVkMemInfo(addr.chunkIndex, res->vkBuffer, res->vkMem);
+	return res;
+}
+
+void VKRSynchronizedHeapAllocator::FreeReservation(AllocatorReservation* uploadReservation)
+{
+	// put the allocation on a delayed release queue for the current command buffer
+	uint64 currentCommandBufferId = VulkanRenderer::GetInstance()->GetCurrentCommandBufferId();
+	auto it = std::find_if(m_activeAllocations.begin(), m_activeAllocations.end(), [&uploadReservation](const TrackedAllocation& allocation) { return allocation.allocation.chunkIndex == uploadReservation->bufferIndex && allocation.allocation.offset == uploadReservation->bufferOffset; });
+	cemu_assert_debug(it != m_activeAllocations.end());
+	m_releaseQueue[currentCommandBufferId].emplace_back(it->allocation);
+	m_activeAllocations.erase(it);
+	m_poolAllocatorReservation.freeObj(uploadReservation);
+}
+
+void VKRSynchronizedHeapAllocator::FlushReservation(AllocatorReservation* uploadReservation)
+{
+	if (m_chunkedHeap.RequiresFlush(uploadReservation->bufferIndex))
+	{
+		VkMappedMemoryRange flushedRange{};
+		flushedRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE;
+		flushedRange.memory = uploadReservation->vkMem;
+		flushedRange.offset = uploadReservation->bufferOffset;
+		flushedRange.size = uploadReservation->size;
+		vkFlushMappedMemoryRanges(VulkanRenderer::GetInstance()->GetLogicalDevice(), 1, &flushedRange);
+	}
+}
+
+void VKRSynchronizedHeapAllocator::CleanupBuffer(uint64 latestFinishedCommandBufferId)
+{
+	auto it = m_releaseQueue.begin();
+	while (it != m_releaseQueue.end())
+	{
+		if (it->first <= latestFinishedCommandBufferId)
+		{
+			// release allocations
+			for(auto& addr : it->second)
+				m_chunkedHeap.free(addr);
+			it = m_releaseQueue.erase(it);
+			continue;
+		}
+		it++;
+	}
+}
+
+void VKRSynchronizedHeapAllocator::GetStats(uint32& numBuffers, size_t& totalBufferSize, size_t& freeBufferSize) const
+{
+	m_chunkedHeap.GetStats(numBuffers, totalBufferSize, freeBufferSize);
+}
+
 /* VkTextureChunkedHeap */
 
+VkTextureChunkedHeap::~VkTextureChunkedHeap()
+{
+	VkDevice device = VulkanRenderer::GetInstance()->GetLogicalDevice();
+	for (auto& i : m_list_chunkInfo)
+	{
+		vkFreeMemory(device, i.mem, nullptr);
+	}
+}
+
 uint32 VkTextureChunkedHeap::allocateNewChunk(uint32 chunkIndex, uint32 minimumAllocationSize)
 {
 	cemu_assert_debug(m_list_chunkInfo.size() == chunkIndex);
 	m_list_chunkInfo.resize(m_list_chunkInfo.size() + 1);
 
 	// pad minimumAllocationSize to 32KB alignment
-	minimumAllocationSize = (minimumAllocationSize + (32*1024-1)) & ~(32 * 1024 - 1);
+	minimumAllocationSize = (minimumAllocationSize + (32 * 1024 - 1)) & ~(32 * 1024 - 1);
 
 	uint32 allocationSize = 1024 * 1024 * 128;
 	if (chunkIndex == 0)
@@ -189,8 +270,7 @@ uint32 VkTextureChunkedHeap::allocateNewChunk(uint32 chunkIndex, uint32 minimumA
 	std::vector<uint32> deviceLocalMemoryTypeIndices = m_vkrMemoryManager->FindMemoryTypes(m_typeFilter, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
 	std::vector<uint32> hostLocalMemoryTypeIndices = m_vkrMemoryManager->FindMemoryTypes(m_typeFilter, 0);
 	// remove device local memory types from host local vector
-	auto pred = [&deviceLocalMemoryTypeIndices](const uint32& v) ->bool
-	{
+	auto pred = [&deviceLocalMemoryTypeIndices](const uint32& v) -> bool {
 		return std::find(deviceLocalMemoryTypeIndices.begin(), deviceLocalMemoryTypeIndices.end(), v) != deviceLocalMemoryTypeIndices.end();
 	};
 	hostLocalMemoryTypeIndices.erase(std::remove_if(hostLocalMemoryTypeIndices.begin(), hostLocalMemoryTypeIndices.end(), pred), hostLocalMemoryTypeIndices.end());
@@ -206,7 +286,7 @@ uint32 VkTextureChunkedHeap::allocateNewChunk(uint32 chunkIndex, uint32 minimumA
 			allocInfo.memoryTypeIndex = memType;
 
 			VkDeviceMemory imageMemory;
-			VkResult r = vkAllocateMemory(m_device, &allocInfo, nullptr, &imageMemory);
+			VkResult r = vkAllocateMemory(VulkanRenderer::GetInstance()->GetLogicalDevice(), &allocInfo, nullptr, &imageMemory);
 			if (r != VK_SUCCESS)
 				continue;
 			m_list_chunkInfo[chunkIndex].mem = imageMemory;
@@ -221,7 +301,7 @@ uint32 VkTextureChunkedHeap::allocateNewChunk(uint32 chunkIndex, uint32 minimumA
 			allocInfo.memoryTypeIndex = memType;
 
 			VkDeviceMemory imageMemory;
-			VkResult r = vkAllocateMemory(m_device, &allocInfo, nullptr, &imageMemory);
+			VkResult r = vkAllocateMemory(VulkanRenderer::GetInstance()->GetLogicalDevice(), &allocInfo, nullptr, &imageMemory);
 			if (r != VK_SUCCESS)
 				continue;
 			m_list_chunkInfo[chunkIndex].mem = imageMemory;
@@ -238,28 +318,76 @@ uint32 VkTextureChunkedHeap::allocateNewChunk(uint32 chunkIndex, uint32 minimumA
 	return 0;
 }
 
-uint32_t VKRMemoryManager::FindMemoryType(uint32_t typeFilter, VkMemoryPropertyFlags properties) const
-{
-	VkPhysicalDeviceMemoryProperties memProperties;
-	vkGetPhysicalDeviceMemoryProperties(m_vkr->GetPhysicalDevice(), &memProperties);
+/* VkBufferChunkedHeap */
 
-	for (uint32 i = 0; i < memProperties.memoryTypeCount; i++)
+VKRBuffer* VKRBuffer::Create(VKR_BUFFER_TYPE bufferType, size_t bufferSize, VkMemoryPropertyFlags properties)
+{
+	auto* memMgr = VulkanRenderer::GetInstance()->GetMemoryManager();
+	VkBuffer buffer;
+	VkDeviceMemory bufferMemory;
+	bool allocSuccess;
+	if (bufferType == VKR_BUFFER_TYPE::STAGING)
+		allocSuccess = memMgr->CreateBuffer(bufferSize, VK_BUFFER_USAGE_TRANSFER_SRC_BIT, properties, buffer, bufferMemory);
+	else if (bufferType == VKR_BUFFER_TYPE::INDEX)
+		allocSuccess = memMgr->CreateBuffer(bufferSize, VK_BUFFER_USAGE_INDEX_BUFFER_BIT, properties, buffer, bufferMemory);
+	else if (bufferType == VKR_BUFFER_TYPE::STRIDE)
+		allocSuccess = memMgr->CreateBuffer(bufferSize, VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, properties, buffer, bufferMemory);
+	else
+		cemu_assert_debug(false);
+	if (!allocSuccess)
+		return nullptr;
+
+	VKRBuffer* bufferObj = new VKRBuffer(buffer, bufferMemory);
+	// if host visible, then map buffer
+	void* data = nullptr;
+	if (properties & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT)
 	{
-		if ((typeFilter & (1 << i)) != 0 && (memProperties.memoryTypes[i].propertyFlags & properties) == properties)
-			return i;
+		vkMapMemory(VulkanRenderer::GetInstance()->GetLogicalDevice(), bufferMemory, 0, bufferSize, 0, &data);
+		bufferObj->m_requiresFlush = !HAS_FLAG(properties, VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
 	}
-	m_vkr->UnrecoverableError(fmt::format("failed to find suitable memory type ({0:#08x} {1:#08x})", typeFilter, properties).c_str());
-	return 0;
+	bufferObj->m_mappedMemory = (uint8*)data;
+	return bufferObj;
 }
 
-bool VKRMemoryManager::FindMemoryType2(uint32 typeFilter, VkMemoryPropertyFlags properties, uint32& memoryIndex) const
+VKRBuffer::~VKRBuffer()
+{
+	if (m_mappedMemory)
+		vkUnmapMemory(VulkanRenderer::GetInstance()->GetLogicalDevice(), m_bufferMemory);
+	if (m_bufferMemory != VK_NULL_HANDLE)
+		vkFreeMemory(VulkanRenderer::GetInstance()->GetLogicalDevice(), m_bufferMemory, nullptr);
+	if (m_buffer != VK_NULL_HANDLE)
+		vkDestroyBuffer(VulkanRenderer::GetInstance()->GetLogicalDevice(), m_buffer, nullptr);
+}
+
+VkBufferChunkedHeap::~VkBufferChunkedHeap()
+{
+	for (auto& chunk : m_chunkBuffers)
+		delete chunk;
+}
+
+uint32 VkBufferChunkedHeap::allocateNewChunk(uint32 chunkIndex, uint32 minimumAllocationSize)
+{
+	size_t allocationSize = std::max<size_t>(m_minimumBufferAllocationSize, minimumAllocationSize);
+	VKRBuffer* buffer = VKRBuffer::Create(m_bufferType, allocationSize, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
+	if(!buffer)
+		buffer = VKRBuffer::Create(m_bufferType, allocationSize, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT);
+	if(!buffer)
+		VulkanRenderer::GetInstance()->UnrecoverableError("Failed to allocate buffer memory for VkBufferChunkedHeap");
+	cemu_assert_debug(buffer);
+	cemu_assert_debug(m_chunkBuffers.size() == chunkIndex);
+	m_chunkBuffers.emplace_back(buffer);
+	// todo - VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT might be worth it?
+	return allocationSize;
+}
+
+bool VKRMemoryManager::FindMemoryType(uint32 typeFilter, VkMemoryPropertyFlags properties, uint32& memoryIndex) const
 {
 	VkPhysicalDeviceMemoryProperties memProperties;
 	vkGetPhysicalDeviceMemoryProperties(m_vkr->GetPhysicalDevice(), &memProperties);
 
 	for (uint32_t i = 0; i < memProperties.memoryTypeCount; i++)
 	{
-		if (typeFilter & (1 << i) && memProperties.memoryTypes[i].propertyFlags == properties)
+		if (typeFilter & (1 << i) && (memProperties.memoryTypes[i].propertyFlags & properties) == properties)
 		{
 			memoryIndex = i;
 			return true;
@@ -330,31 +458,7 @@ size_t VKRMemoryManager::GetTotalMemoryForBufferType(VkBufferUsageFlags usage, V
 	return total;
 }
 
-void VKRMemoryManager::CreateBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkBuffer& buffer, VkDeviceMemory& bufferMemory) const
-{
-	VkBufferCreateInfo bufferInfo{};
-	bufferInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
-	bufferInfo.usage = usage;
-	bufferInfo.size = size;
-	bufferInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
-	if (vkCreateBuffer(m_vkr->GetLogicalDevice(), &bufferInfo, nullptr, &buffer) != VK_SUCCESS)
-		m_vkr->UnrecoverableError("Failed to create buffer");
-
-	VkMemoryRequirements memRequirements;
-	vkGetBufferMemoryRequirements(m_vkr->GetLogicalDevice(), buffer, &memRequirements);
-
-	VkMemoryAllocateInfo allocInfo{};
-	allocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
-	allocInfo.allocationSize = memRequirements.size;
-	allocInfo.memoryTypeIndex = FindMemoryType(memRequirements.memoryTypeBits, properties);
-
-	if (vkAllocateMemory(m_vkr->GetLogicalDevice(), &allocInfo, nullptr, &bufferMemory) != VK_SUCCESS)
-		m_vkr->UnrecoverableError("Failed to allocate buffer memory");
-	if (vkBindBufferMemory(m_vkr->GetLogicalDevice(), buffer, bufferMemory, 0) != VK_SUCCESS)
-		m_vkr->UnrecoverableError("Failed to bind buffer memory");
-}
-
-bool VKRMemoryManager::CreateBuffer2(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkBuffer& buffer, VkDeviceMemory& bufferMemory) const
+bool VKRMemoryManager::CreateBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkBuffer& buffer, VkDeviceMemory& bufferMemory) const
 {
 	VkBufferCreateInfo bufferInfo{};
 	bufferInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
@@ -363,7 +467,7 @@ bool VKRMemoryManager::CreateBuffer2(VkDeviceSize size, VkBufferUsageFlags usage
 	bufferInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
 	if (vkCreateBuffer(m_vkr->GetLogicalDevice(), &bufferInfo, nullptr, &buffer) != VK_SUCCESS)
 	{
-		cemuLog_log(LogType::Force, "Failed to create buffer (CreateBuffer2)");
+		cemuLog_log(LogType::Force, "Failed to create buffer (CreateBuffer)");
 		return false;
 	}
 
@@ -373,7 +477,7 @@ bool VKRMemoryManager::CreateBuffer2(VkDeviceSize size, VkBufferUsageFlags usage
 	VkMemoryAllocateInfo allocInfo{};
 	allocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
 	allocInfo.allocationSize = memRequirements.size;
-	if (!FindMemoryType2(memRequirements.memoryTypeBits, properties, allocInfo.memoryTypeIndex))
+	if (!FindMemoryType(memRequirements.memoryTypeBits, properties, allocInfo.memoryTypeIndex))
 	{
 		vkDestroyBuffer(m_vkr->GetLogicalDevice(), buffer, nullptr);
 		return false;
@@ -386,7 +490,7 @@ bool VKRMemoryManager::CreateBuffer2(VkDeviceSize size, VkBufferUsageFlags usage
 	if (vkBindBufferMemory(m_vkr->GetLogicalDevice(), buffer, bufferMemory, 0) != VK_SUCCESS)
 	{
 		vkDestroyBuffer(m_vkr->GetLogicalDevice(), buffer, nullptr);
-		cemuLog_log(LogType::Force, "Failed to bind buffer (CreateBuffer2)");
+		cemuLog_log(LogType::Force, "Failed to bind buffer (CreateBuffer)");
 		return false;
 	}
 	return true;
@@ -408,7 +512,7 @@ bool VKRMemoryManager::CreateBufferFromHostMemory(void* hostPointer, VkDeviceSiz
 
 	if (vkCreateBuffer(m_vkr->GetLogicalDevice(), &bufferInfo, nullptr, &buffer) != VK_SUCCESS)
 	{
-		cemuLog_log(LogType::Force, "Failed to create buffer (CreateBuffer2)");
+		cemuLog_log(LogType::Force, "Failed to create buffer (CreateBuffer)");
 		return false;
 	}
 
@@ -423,13 +527,13 @@ bool VKRMemoryManager::CreateBufferFromHostMemory(void* hostPointer, VkDeviceSiz
 	importHostMem.sType = VK_STRUCTURE_TYPE_IMPORT_MEMORY_HOST_POINTER_INFO_EXT;
 	importHostMem.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT;
 	importHostMem.pHostPointer = hostPointer;
-	// VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT or 
+	// VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT or
 	// VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_MAPPED_FOREIGN_MEMORY_BIT_EXT
 	// whats the difference ?
 
 	allocInfo.pNext = &importHostMem;
 
-	if (!FindMemoryType2(memRequirements.memoryTypeBits, properties, allocInfo.memoryTypeIndex))
+	if (!FindMemoryType(memRequirements.memoryTypeBits, properties, allocInfo.memoryTypeIndex))
 	{
 		vkDestroyBuffer(m_vkr->GetLogicalDevice(), buffer, nullptr);
 		return false;
@@ -469,11 +573,11 @@ VkImageMemAllocation* VKRMemoryManager::imageMemoryAllocate(VkImage image)
 	auto it = map_textureHeap.find(typeFilter);
 	if (it == map_textureHeap.end())
 	{
-		texHeap = new VkTextureChunkedHeap(this, typeFilter, m_vkr->GetLogicalDevice());
+		texHeap = new VkTextureChunkedHeap(this, typeFilter);
 		map_textureHeap.emplace(typeFilter, texHeap);
 	}
 	else
-		texHeap = it->second;
+		texHeap = it->second.get();
 
 	// alloc mem from heap
 	uint32 allocationSize = (uint32)memRequirements.size;

src/Cafe/HW/Latte/Renderer/Vulkan/VKRMemoryManager.h

@@ -2,6 +2,36 @@
 #include "Cafe/HW/Latte/Renderer/Renderer.h"
 #include "Cafe/HW/Latte/Renderer/Vulkan/VulkanAPI.h"
 #include "util/ChunkedHeap/ChunkedHeap.h"
+#include "util/helpers/MemoryPool.h"
+
+enum class VKR_BUFFER_TYPE
+{
+	STAGING, // staging upload buffer
+	INDEX, // buffer for index data
+	STRIDE, // buffer for stride-adjusted vertex data
+};
+
+class VKRBuffer
+{
+  public:
+	static VKRBuffer* Create(VKR_BUFFER_TYPE bufferType, size_t bufferSize, VkMemoryPropertyFlags properties);
+	~VKRBuffer();
+
+	VkBuffer GetVkBuffer() const { return m_buffer; }
+	VkDeviceMemory GetVkBufferMemory() const { return m_bufferMemory; }
+
+	uint8* GetPtr() const { return m_mappedMemory; }
+
+	bool RequiresFlush() const { return m_requiresFlush; }
+
+  private:
+	VKRBuffer(VkBuffer buffer, VkDeviceMemory bufferMem) : m_buffer(buffer), m_bufferMemory(bufferMem) { };
+
+	VkBuffer m_buffer;
+	VkDeviceMemory m_bufferMemory;
+	uint8* m_mappedMemory;
+	bool m_requiresFlush{false};
+};
 
 struct VkImageMemAllocation
 {
@@ -14,18 +44,17 @@ struct VkImageMemAllocation
 	uint32 getAllocationSize() { return allocationSize; }
 };
 
-class VkTextureChunkedHeap : private ChunkedHeap
+class VkTextureChunkedHeap : private ChunkedHeap<>
 {
 public:
-	VkTextureChunkedHeap(class VKRMemoryManager* memoryManager, uint32 typeFilter, VkDevice device) : m_vkrMemoryManager(memoryManager), m_typeFilter(typeFilter), m_device(device) { };
+	VkTextureChunkedHeap(class VKRMemoryManager* memoryManager, uint32 typeFilter) : m_vkrMemoryManager(memoryManager), m_typeFilter(typeFilter) { };
+	~VkTextureChunkedHeap();
 
 	struct ChunkInfo
 	{
 		VkDeviceMemory mem;
 	};
 
-	uint32 allocateNewChunk(uint32 chunkIndex, uint32 minimumAllocationSize) override;
-
 	CHAddr allocMem(uint32 size, uint32 alignment)
 	{
 		if (alignment < 4)
@@ -43,11 +72,6 @@ public:
 		this->free(addr);
 	}
 
-	void setDevice(VkDevice dev)
-	{
-		m_device = dev;
-	}
-
 	VkDeviceMemory getChunkMem(uint32 index)
 	{
 		if (index >= m_list_chunkInfo.size())
@@ -57,29 +81,75 @@ public:
 
 	void getStatistics(uint32& totalHeapSize, uint32& allocatedBytes) const
 	{
-		totalHeapSize = numHeapBytes;
-		allocatedBytes = numAllocatedBytes;
+		totalHeapSize = m_numHeapBytes;
+		allocatedBytes = m_numAllocatedBytes;
 	}
 
-	VkDevice m_device;
+  private:
+	uint32 allocateNewChunk(uint32 chunkIndex, uint32 minimumAllocationSize) override;
+
 	uint32 m_typeFilter{ 0xFFFFFFFF };
 	class VKRMemoryManager* m_vkrMemoryManager;
 	std::vector<ChunkInfo> m_list_chunkInfo;
 };
 
+class VkBufferChunkedHeap : private ChunkedHeap<>
+{
+  public:
+	VkBufferChunkedHeap(VKR_BUFFER_TYPE bufferType, size_t minimumBufferAllocationSize) : m_bufferType(bufferType), m_minimumBufferAllocationSize(minimumBufferAllocationSize) { };
+	~VkBufferChunkedHeap();
+
+	using ChunkedHeap::alloc;
+	using ChunkedHeap::free;
+
+	uint8* GetChunkPtr(uint32 index) const
+	{
+		if (index >= m_chunkBuffers.size())
+			return nullptr;
+		return m_chunkBuffers[index]->GetPtr();
+	}
+
+	void GetChunkVkMemInfo(uint32 index, VkBuffer& buffer, VkDeviceMemory& mem)
+	{
+		if (index >= m_chunkBuffers.size())
+		{
+			buffer = VK_NULL_HANDLE;
+			mem = VK_NULL_HANDLE;
+			return;
+		}
+		buffer = m_chunkBuffers[index]->GetVkBuffer();
+		mem = m_chunkBuffers[index]->GetVkBufferMemory();
+	}
+
+	void GetStats(uint32& numBuffers, size_t& totalBufferSize, size_t& freeBufferSize) const
+	{
+		numBuffers = m_chunkBuffers.size();
+		totalBufferSize = m_numHeapBytes;
+		freeBufferSize = m_numHeapBytes - m_numAllocatedBytes;
+	}
+
+	bool RequiresFlush(uint32 index) const
+	{
+		if (index >= m_chunkBuffers.size())
+			return false;
+		return m_chunkBuffers[index]->RequiresFlush();
+	}
+
+  private:
+	uint32 allocateNewChunk(uint32 chunkIndex, uint32 minimumAllocationSize) override;
+
+	VKR_BUFFER_TYPE m_bufferType;
+	std::vector<VKRBuffer*> m_chunkBuffers;
+	size_t m_minimumBufferAllocationSize;
+};
+
 // a circular ring-buffer which tracks and releases memory per command-buffer
 class VKRSynchronizedRingAllocator
 {
 public:
-	enum class BUFFER_TYPE
-	{
-		STAGING, // staging upload buffer
-		INDEX, // buffer for index data
-		STRIDE, // buffer for stride-adjusted vertex data
-	};
-
-	VKRSynchronizedRingAllocator(class VulkanRenderer* vkRenderer, class VKRMemoryManager* vkMemoryManager, BUFFER_TYPE bufferType, uint32 minimumBufferAllocSize) : m_vkr(vkRenderer), m_vkrMemMgr(vkMemoryManager), m_bufferType(bufferType), m_minimumBufferAllocSize(minimumBufferAllocSize) {};
+	VKRSynchronizedRingAllocator(class VulkanRenderer* vkRenderer, class VKRMemoryManager* vkMemoryManager, VKR_BUFFER_TYPE bufferType, uint32 minimumBufferAllocSize) : m_vkr(vkRenderer), m_vkrMemMgr(vkMemoryManager), m_bufferType(bufferType), m_minimumBufferAllocSize(minimumBufferAllocSize) {};
 	VKRSynchronizedRingAllocator(const VKRSynchronizedRingAllocator&) = delete; // disallow copy
+	~VKRSynchronizedRingAllocator();
 
 	struct BufferSyncPoint_t
 	{
@@ -126,13 +196,53 @@ private:
 
 	const class VulkanRenderer* m_vkr;
 	const class VKRMemoryManager* m_vkrMemMgr;
-	const BUFFER_TYPE m_bufferType;
+	const VKR_BUFFER_TYPE m_bufferType;
 	const uint32 m_minimumBufferAllocSize;
 
 	std::vector<AllocatorBuffer_t> m_buffers;
 
 };
 
+// heap style allocator with released memory being freed after the current command buffer finishes
+class VKRSynchronizedHeapAllocator
+{
+	struct TrackedAllocation
+	{
+		TrackedAllocation(CHAddr allocation) : allocation(allocation) {};
+		CHAddr allocation;
+	};
+
+  public:
+	VKRSynchronizedHeapAllocator(class VKRMemoryManager* vkMemoryManager, VKR_BUFFER_TYPE bufferType, size_t minimumBufferAllocSize);
+	VKRSynchronizedHeapAllocator(const VKRSynchronizedHeapAllocator&) = delete; // disallow copy
+
+	struct AllocatorReservation
+	{
+		VkBuffer vkBuffer;
+		VkDeviceMemory vkMem;
+		uint8* memPtr;
+		uint32 bufferOffset;
+		uint32 size;
+		uint32 bufferIndex;
+	};
+
+	AllocatorReservation* AllocateBufferMemory(uint32 size, uint32 alignment);
+	void FreeReservation(AllocatorReservation* uploadReservation);
+	void FlushReservation(AllocatorReservation* uploadReservation);
+
+	void CleanupBuffer(uint64 latestFinishedCommandBufferId);
+
+	void GetStats(uint32& numBuffers, size_t& totalBufferSize, size_t& freeBufferSize) const;
+  private:
+	const class VKRMemoryManager* m_vkrMemMgr;
+	VkBufferChunkedHeap m_chunkedHeap;
+	// allocations
+	std::vector<TrackedAllocation> m_activeAllocations;
+	MemoryPool<AllocatorReservation> m_poolAllocatorReservation{32};
+	// release queue
+	std::unordered_map<uint64, std::vector<CHAddr>> m_releaseQueue;
+};
+
 void LatteIndices_invalidateAll();
 
 class VKRMemoryManager
@@ -140,15 +250,15 @@ class VKRMemoryManager
 	friend class VKRSynchronizedRingAllocator;
 public:
 	VKRMemoryManager(class VulkanRenderer* renderer) :
-			m_stagingBuffer(renderer, this, VKRSynchronizedRingAllocator::BUFFER_TYPE::STAGING, 32u * 1024 * 1024),
-			m_indexBuffer(renderer, this, VKRSynchronizedRingAllocator::BUFFER_TYPE::INDEX, 4u * 1024 * 1024),
-			m_vertexStrideMetalBuffer(renderer, this, VKRSynchronizedRingAllocator::BUFFER_TYPE::STRIDE, 4u * 1024 * 1024)
+			m_stagingBuffer(renderer, this, VKR_BUFFER_TYPE::STAGING, 32u * 1024 * 1024),
+			m_indexBuffer(this, VKR_BUFFER_TYPE::INDEX, 4u * 1024 * 1024),
+			m_vertexStrideMetalBuffer(renderer, this, VKR_BUFFER_TYPE::STRIDE, 4u * 1024 * 1024)
 	{
 		m_vkr = renderer;
 	}
 
 	// texture memory management
-	std::unordered_map<uint32, VkTextureChunkedHeap*> map_textureHeap; // one heap per memory type
+	std::unordered_map<uint32, std::unique_ptr<VkTextureChunkedHeap>> map_textureHeap; // one heap per memory type
 	std::vector<uint8> m_textureUploadBuffer;
 
 	// texture upload buffer
@@ -167,7 +277,7 @@ public:
 	}
 
 	VKRSynchronizedRingAllocator& getStagingAllocator() { return m_stagingBuffer; }; // allocator for texture/attribute/uniform uploads
-	VKRSynchronizedRingAllocator& getIndexAllocator() { return m_indexBuffer; }; // allocator for index data
+	VKRSynchronizedHeapAllocator& GetIndexAllocator() { return m_indexBuffer; }; // allocator for index data
 	VKRSynchronizedRingAllocator& getMetalStrideWorkaroundAllocator() { return m_vertexStrideMetalBuffer; }; // allocator for stride-adjusted vertex data
 
 	void cleanupBuffers(uint64 latestFinishedCommandBufferId)
@@ -178,9 +288,7 @@ public:
 		m_vertexStrideMetalBuffer.CleanupBuffer(latestFinishedCommandBufferId);
 	}
 
-	// memory helpers
-	uint32_t FindMemoryType(uint32_t typeFilter, VkMemoryPropertyFlags properties) const;
-	bool FindMemoryType2(uint32 typeFilter, VkMemoryPropertyFlags properties, uint32& memoryIndex) const; // searches for exact properties. Can gracefully fail without throwing exception (returns false)
+	bool FindMemoryType(uint32 typeFilter, VkMemoryPropertyFlags properties, uint32& memoryIndex) const; // searches for exact properties. Can gracefully fail without throwing exception (returns false)
 	std::vector<uint32> FindMemoryTypes(uint32_t typeFilter, VkMemoryPropertyFlags properties) const;
 
 	// image memory allocation
@@ -190,8 +298,7 @@ public:
 	// buffer management
 	size_t GetTotalMemoryForBufferType(VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, size_t minimumBufferSize = 16 * 1024 * 1024);
 
-	void CreateBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkBuffer& buffer, VkDeviceMemory& bufferMemory) const;
-	bool CreateBuffer2(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkBuffer& buffer, VkDeviceMemory& bufferMemory) const; // same as CreateBuffer but doesn't throw exception on failure
+	bool CreateBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkBuffer& buffer, VkDeviceMemory& bufferMemory) const; // same as CreateBuffer but doesn't throw exception on failure
 	bool CreateBufferFromHostMemory(void* hostPointer, VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkBuffer& buffer, VkDeviceMemory& bufferMemory) const;
 
 	void DeleteBuffer(VkBuffer& buffer, VkDeviceMemory& deviceMem) const;
@@ -202,6 +309,6 @@ public:
 	private:
 		class VulkanRenderer* m_vkr;
 		VKRSynchronizedRingAllocator m_stagingBuffer;
-		VKRSynchronizedRingAllocator m_indexBuffer;
+		VKRSynchronizedHeapAllocator m_indexBuffer;
 		VKRSynchronizedRingAllocator m_vertexStrideMetalBuffer;
 };

src/Cafe/HW/Latte/Renderer/Vulkan/VKRPipelineInfo.cpp

@@ -26,7 +26,6 @@ PipelineInfo::PipelineInfo(uint64 minimalStateHash, uint64 pipelineHash, LatteFe
 
 	// init VKRObjPipeline
 	m_vkrObjPipeline = new VKRObjectPipeline();
-	m_vkrObjPipeline->pipeline = VK_NULL_HANDLE;
 
 	// track dependency with shaders
 	if (vertexShaderVk)

src/Cafe/HW/Latte/Renderer/Vulkan/VulkanAPI.h

@@ -165,6 +165,7 @@ VKFUNC_DEVICE(vkCmdDraw);
 VKFUNC_DEVICE(vkCmdCopyBufferToImage);
 VKFUNC_DEVICE(vkCmdCopyImageToBuffer);
 VKFUNC_DEVICE(vkCmdClearColorImage);
+VKFUNC_DEVICE(vkCmdClearAttachments);
 VKFUNC_DEVICE(vkCmdBindIndexBuffer);
 VKFUNC_DEVICE(vkCmdBindVertexBuffers);
 VKFUNC_DEVICE(vkCmdDrawIndexed);
@@ -188,6 +189,9 @@ VKFUNC_DEVICE(vkCmdPipelineBarrier2KHR);
 VKFUNC_DEVICE(vkCmdBeginRenderingKHR);
 VKFUNC_DEVICE(vkCmdEndRenderingKHR);
 
+// khr_present_wait
+VKFUNC_DEVICE(vkWaitForPresentKHR);
+
 // transform feedback extension
 VKFUNC_DEVICE(vkCmdBindTransformFeedbackBuffersEXT);
 VKFUNC_DEVICE(vkCmdBeginTransformFeedbackEXT);
@@ -195,6 +199,7 @@ VKFUNC_DEVICE(vkCmdEndTransformFeedbackEXT);
 
 // query
 VKFUNC_DEVICE(vkCreateQueryPool);
+VKFUNC_DEVICE(vkDestroyQueryPool);
 VKFUNC_DEVICE(vkCmdResetQueryPool);
 VKFUNC_DEVICE(vkCmdBeginQuery);
 VKFUNC_DEVICE(vkCmdEndQuery);
@@ -233,6 +238,7 @@ VKFUNC_DEVICE(vkAllocateDescriptorSets);
 VKFUNC_DEVICE(vkFreeDescriptorSets);
 VKFUNC_DEVICE(vkUpdateDescriptorSets);
 VKFUNC_DEVICE(vkCreateDescriptorPool);
+VKFUNC_DEVICE(vkDestroyDescriptorPool);
 VKFUNC_DEVICE(vkDestroyDescriptorSetLayout);
 
 #undef VKFUNC_INIT

src/Cafe/HW/Latte/Renderer/Vulkan/VulkanPipelineCompiler.cpp

@@ -558,8 +558,8 @@ void PipelineCompiler::InitRasterizerState(const LatteContextRegister& latteRegi
 	rasterizerExt.flags = 0;
 
 	rasterizer.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO;
-	rasterizer.pNext = &rasterizerExt;
 	rasterizer.rasterizerDiscardEnable = LatteGPUState.contextNew.PA_CL_CLIP_CNTL.get_DX_RASTERIZATION_KILL();
+	rasterizer.pNext = VulkanRenderer::GetInstance()->m_featureControl.deviceExtensions.depth_clip_enable ? &rasterizerExt : nullptr;
 	// GX2SetSpecialState(0, true) workaround
 	if (!LatteGPUState.contextNew.PA_CL_VTE_CNTL.get_VPORT_X_OFFSET_ENA())
 		rasterizer.rasterizerDiscardEnable = false;
@@ -730,7 +730,7 @@ void PipelineCompiler::InitDescriptorSetLayouts(VulkanRenderer* vkRenderer, Pipe
 	{
 		cemu_assert_debug(descriptorSetLayoutCount == 0);
 		CreateDescriptorSetLayout(vkRenderer, vertexShader, descriptorSetLayout[descriptorSetLayoutCount], vkrPipelineInfo);
-		vkObjPipeline->vertexDSL = descriptorSetLayout[descriptorSetLayoutCount];
+		vkObjPipeline->m_vertexDSL = descriptorSetLayout[descriptorSetLayoutCount];
 		descriptorSetLayoutCount++;
 	}
 
@@ -738,7 +738,7 @@ void PipelineCompiler::InitDescriptorSetLayouts(VulkanRenderer* vkRenderer, Pipe
 	{
 		cemu_assert_debug(descriptorSetLayoutCount == 1);
 		CreateDescriptorSetLayout(vkRenderer, pixelShader, descriptorSetLayout[descriptorSetLayoutCount], vkrPipelineInfo);
-		vkObjPipeline->pixelDSL = descriptorSetLayout[descriptorSetLayoutCount];
+		vkObjPipeline->m_pixelDSL = descriptorSetLayout[descriptorSetLayoutCount];
 		descriptorSetLayoutCount++;
 	}
 	else if (geometryShader)
@@ -757,7 +757,7 @@ void PipelineCompiler::InitDescriptorSetLayouts(VulkanRenderer* vkRenderer, Pipe
 	{
 		cemu_assert_debug(descriptorSetLayoutCount == 2);
 		CreateDescriptorSetLayout(vkRenderer, geometryShader, descriptorSetLayout[descriptorSetLayoutCount], vkrPipelineInfo);
-		vkObjPipeline->geometryDSL = descriptorSetLayout[descriptorSetLayoutCount];
+		vkObjPipeline->m_geometryDSL = descriptorSetLayout[descriptorSetLayoutCount];
 		descriptorSetLayoutCount++;
 	}
 }
@@ -918,7 +918,7 @@ bool PipelineCompiler::InitFromCurrentGPUState(PipelineInfo* pipelineInfo, const
 	pipelineLayoutInfo.pPushConstantRanges = nullptr;
 	pipelineLayoutInfo.pushConstantRangeCount = 0;
 
-	VkResult result = vkCreatePipelineLayout(vkRenderer->m_logicalDevice, &pipelineLayoutInfo, nullptr, &m_pipeline_layout);
+	VkResult result = vkCreatePipelineLayout(vkRenderer->m_logicalDevice, &pipelineLayoutInfo, nullptr, &m_pipelineLayout);
 	if (result != VK_SUCCESS)
 	{
 		cemuLog_log(LogType::Force, "Failed to create pipeline layout: {}", result);
@@ -936,7 +936,7 @@ bool PipelineCompiler::InitFromCurrentGPUState(PipelineInfo* pipelineInfo, const
 
 	// ##########################################################################################################################################
 
-	pipelineInfo->m_vkrObjPipeline->pipeline_layout = m_pipeline_layout;
+	pipelineInfo->m_vkrObjPipeline->m_pipelineLayout = m_pipelineLayout;
 
 	// increment ref counter for vkrObjPipeline and renderpass object to make sure they dont get released while we are using them
 	m_vkrObjPipeline->incRef();
@@ -989,7 +989,7 @@ bool PipelineCompiler::Compile(bool forceCompile, bool isRenderThread, bool show
 	pipelineInfo.pRasterizationState = &rasterizer;
 	pipelineInfo.pMultisampleState = &multisampling;
 	pipelineInfo.pColorBlendState = &colorBlending;
-	pipelineInfo.layout = m_pipeline_layout;
+	pipelineInfo.layout = m_pipelineLayout;
 	pipelineInfo.renderPass = m_renderPassObj->m_renderPass;
 	pipelineInfo.pDepthStencilState = &depthStencilState;
 	pipelineInfo.subpass = 0;
@@ -1037,7 +1037,7 @@ bool PipelineCompiler::Compile(bool forceCompile, bool isRenderThread, bool show
 	}
 	else if (result == VK_SUCCESS)
 	{
-		m_vkrObjPipeline->setPipeline(pipeline);
+		m_vkrObjPipeline->SetPipeline(pipeline);
 	}
 	else
 	{

src/Cafe/HW/Latte/Renderer/Vulkan/VulkanPipelineCompiler.h

@@ -41,7 +41,7 @@ public:
 	bool InitFromCurrentGPUState(PipelineInfo* pipelineInfo, const LatteContextRegister& latteRegister, VKRObjectRenderPass* renderPassObj);
 	void TrackAsCached(uint64 baseHash, uint64 pipelineStateHash); // stores pipeline to permanent cache if not yet cached. Must be called synchronously from render thread due to dependency on GPU state
 
-	VkPipelineLayout m_pipeline_layout;
+	VkPipelineLayout m_pipelineLayout;
 	VKRObjectRenderPass* m_renderPassObj{};
 
 	/* shader stages */

src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.cpp

@@ -47,7 +47,10 @@ const  std::vector<const char*> kOptionalDeviceExtensions =
 	VK_EXT_FILTER_CUBIC_EXTENSION_NAME, // not supported by any device yet
 	VK_EXT_EXTERNAL_MEMORY_HOST_EXTENSION_NAME,
 	VK_KHR_SYNCHRONIZATION_2_EXTENSION_NAME,
-	VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME
+	VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME,
+	VK_KHR_PRESENT_WAIT_EXTENSION_NAME,
+	VK_KHR_PRESENT_ID_EXTENSION_NAME,
+	VK_EXT_DEPTH_CLIP_ENABLE_EXTENSION_NAME
 };
 
 const std::vector<const char*> kRequiredDeviceExtensions =
@@ -80,8 +83,6 @@ VKAPI_ATTR VkBool32 VKAPI_CALL DebugUtilsCallback(VkDebugUtilsMessageSeverityFla
 	if (strstr(pCallbackData->pMessage, "Number of currently valid sampler objects is not less than the maximum allowed"))
 		return VK_FALSE;
 
-	assert_dbg();
-
 #endif
 
 	cemuLog_log(LogType::Force, (char*)pCallbackData->pMessage);
@@ -252,12 +253,24 @@ void VulkanRenderer::GetDeviceFeatures()
 	pcc.pNext = prevStruct;
 	prevStruct = &pcc;
 
+	VkPhysicalDevicePresentIdFeaturesKHR pidf{};
+	pidf.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENT_ID_FEATURES_KHR;
+	pidf.pNext = prevStruct;
+	prevStruct = &pidf;
+
+	VkPhysicalDevicePresentWaitFeaturesKHR pwf{};
+	pwf.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENT_WAIT_FEATURES_KHR;
+	pwf.pNext = prevStruct;
+	prevStruct = &pwf;
+
 	VkPhysicalDeviceFeatures2 physicalDeviceFeatures2{};
 	physicalDeviceFeatures2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
 	physicalDeviceFeatures2.pNext = prevStruct;
 
 	vkGetPhysicalDeviceFeatures2(m_physicalDevice, &physicalDeviceFeatures2);
 
+	cemuLog_log(LogType::Force, "Vulkan: present_wait extension: {}", (pwf.presentWait && pidf.presentId) ? "supported" : "unsupported");
+
 	/* Get Vulkan device properties and limits */
 	VkPhysicalDeviceFloatControlsPropertiesKHR pfcp{};
 	prevStruct = nullptr;
@@ -300,7 +313,10 @@ void VulkanRenderer::GetDeviceFeatures()
 			cemuLog_log(LogType::Force, "VK_EXT_custom_border_color not supported. Cannot emulate arbitrary border color");
 		}
 	}
-
+	if (!m_featureControl.deviceExtensions.depth_clip_enable)
+	{
+		cemuLog_log(LogType::Force, "VK_EXT_depth_clip_enable not supported");
+	}
 	// get limits
 	m_featureControl.limits.minUniformBufferOffsetAlignment = std::max(prop2.properties.limits.minUniformBufferOffsetAlignment, (VkDeviceSize)4);
 	m_featureControl.limits.nonCoherentAtomSize = std::max(prop2.properties.limits.nonCoherentAtomSize, (VkDeviceSize)4);
@@ -425,7 +441,7 @@ VulkanRenderer::VulkanRenderer()
 	GetDeviceFeatures();
 
 	// init memory manager
-	memoryManager = new VKRMemoryManager(this);
+	memoryManager.reset(new VKRMemoryManager(this));
 
 	try
 	{
@@ -490,6 +506,24 @@ VulkanRenderer::VulkanRenderer()
 		customBorderColorFeature.customBorderColors = VK_TRUE;
 		customBorderColorFeature.customBorderColorWithoutFormat = VK_TRUE;
 	}
+	// enable VK_KHR_present_id
+	VkPhysicalDevicePresentIdFeaturesKHR presentIdFeature{};
+	if(m_featureControl.deviceExtensions.present_wait)
+	{
+		presentIdFeature.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENT_ID_FEATURES_KHR;
+		presentIdFeature.pNext = deviceExtensionFeatures;
+		deviceExtensionFeatures = &presentIdFeature;
+		presentIdFeature.presentId = VK_TRUE;
+	}
+	// enable VK_KHR_present_wait
+	VkPhysicalDevicePresentWaitFeaturesKHR presentWaitFeature{};
+	if(m_featureControl.deviceExtensions.present_wait)
+	{
+		presentWaitFeature.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENT_WAIT_FEATURES_KHR;
+		presentWaitFeature.pNext = deviceExtensionFeatures;
+		deviceExtensionFeatures = &presentWaitFeature;
+		presentWaitFeature.presentWait = VK_TRUE;
+	}
 
 	std::vector<const char*> used_extensions;
 	VkDeviceCreateInfo createInfo = CreateDeviceCreateInfo(queueCreateInfos, deviceFeatures, deviceExtensionFeatures, used_extensions);
@@ -545,15 +579,15 @@ VulkanRenderer::VulkanRenderer()
 	void* bufferPtr;
 	// init ringbuffer for uniform vars
 	m_uniformVarBufferMemoryIsCoherent = false;
-	if (memoryManager->CreateBuffer2(UNIFORMVAR_RINGBUFFER_SIZE, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT, m_uniformVarBuffer, m_uniformVarBufferMemory))
+	if (memoryManager->CreateBuffer(UNIFORMVAR_RINGBUFFER_SIZE, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT, m_uniformVarBuffer, m_uniformVarBufferMemory))
 		m_uniformVarBufferMemoryIsCoherent = true;
-	else if (memoryManager->CreateBuffer2(UNIFORMVAR_RINGBUFFER_SIZE, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT | VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, m_uniformVarBuffer, m_uniformVarBufferMemory))
+	else if (memoryManager->CreateBuffer(UNIFORMVAR_RINGBUFFER_SIZE, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT | VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, m_uniformVarBuffer, m_uniformVarBufferMemory))
 		m_uniformVarBufferMemoryIsCoherent = true; // unified memory
-	else if (memoryManager->CreateBuffer2(UNIFORMVAR_RINGBUFFER_SIZE, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT, m_uniformVarBuffer, m_uniformVarBufferMemory))
+	else if (memoryManager->CreateBuffer(UNIFORMVAR_RINGBUFFER_SIZE, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT, m_uniformVarBuffer, m_uniformVarBufferMemory))
 		m_uniformVarBufferMemoryIsCoherent = true;
 	else
 	{
-		memoryManager->CreateBuffer2(UNIFORMVAR_RINGBUFFER_SIZE, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT, m_uniformVarBuffer, m_uniformVarBufferMemory);
+		memoryManager->CreateBuffer(UNIFORMVAR_RINGBUFFER_SIZE, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT, m_uniformVarBuffer, m_uniformVarBufferMemory);
 	}
 
 	if (!m_uniformVarBufferMemoryIsCoherent)
@@ -596,6 +630,31 @@ VulkanRenderer::~VulkanRenderer()
 	m_pipeline_cache_semaphore.notify();
 	m_pipeline_cache_save_thread.join();
 
+	vkDestroyPipelineCache(m_logicalDevice, m_pipeline_cache, nullptr);
+
+	if(!m_backbufferBlitDescriptorSetCache.empty())
+	{
+		std::vector<VkDescriptorSet> freeVector;
+		freeVector.reserve(m_backbufferBlitDescriptorSetCache.size());
+		std::transform(m_backbufferBlitDescriptorSetCache.begin(), m_backbufferBlitDescriptorSetCache.end(), std::back_inserter(freeVector), [](auto& i) {
+		  return i.second;
+		});
+		vkFreeDescriptorSets(m_logicalDevice, m_descriptorPool, freeVector.size(), freeVector.data());
+	}
+
+	vkDestroyDescriptorPool(m_logicalDevice, m_descriptorPool, nullptr);
+
+	for(auto& i : m_backbufferBlitPipelineCache)
+	{
+		vkDestroyPipeline(m_logicalDevice, i.second, nullptr);
+	}
+	m_backbufferBlitPipelineCache = {};
+
+	if(m_occlusionQueries.queryPool != VK_NULL_HANDLE)
+		vkDestroyQueryPool(m_logicalDevice, m_occlusionQueries.queryPool, nullptr);
+
+	vkDestroyDescriptorSetLayout(m_logicalDevice, m_swapchainDescriptorSetLayout, nullptr);
+
 	// shut down imgui
 	ImGui_ImplVulkan_Shutdown();
 
@@ -608,10 +667,6 @@ VulkanRenderer::~VulkanRenderer()
 	memoryManager->DeleteBuffer(m_xfbRingBuffer, m_xfbRingBufferMemory);
 	memoryManager->DeleteBuffer(m_occlusionQueries.bufferQueryResults, m_occlusionQueries.memoryQueryResults);
 	memoryManager->DeleteBuffer(m_bufferCache, m_bufferCacheMemory);
-	// texture memory
-	// todo
-	// upload buffers
-	// todo
 
 	m_padSwapchainInfo = nullptr;
 	m_mainSwapchainInfo = nullptr;
@@ -634,12 +689,20 @@ VulkanRenderer::~VulkanRenderer()
 		it = VK_NULL_HANDLE;
 	}
 
+	for(auto& sem : m_commandBufferSemaphores)
+	{
+		vkDestroySemaphore(m_logicalDevice, sem, nullptr);
+		sem = VK_NULL_HANDLE;
+	}
+
 	if (m_pipelineLayout != VK_NULL_HANDLE)
 		vkDestroyPipelineLayout(m_logicalDevice, m_pipelineLayout, nullptr);
 
 	if (m_commandPool != VK_NULL_HANDLE)
 		vkDestroyCommandPool(m_logicalDevice, m_commandPool, nullptr);
 
+	VKRObjectSampler::DestroyCache();
+
 	// destroy debug callback
 	if (m_debugCallback)
 	{
@@ -647,6 +710,12 @@ VulkanRenderer::~VulkanRenderer()
 		vkDestroyDebugUtilsMessengerEXT(m_instance, m_debugCallback, nullptr);
 	}
 
+	while(!m_destructionQueue.empty())
+		ProcessDestructionQueue();
+
+	// destroy memory manager
+	memoryManager.reset();
+
 	// destroy instance, devices
 	if (m_instance != VK_NULL_HANDLE)
 	{
@@ -658,9 +727,6 @@ VulkanRenderer::~VulkanRenderer()
 		vkDestroyInstance(m_instance, nullptr);
 	}
 
-	// destroy memory manager
-	delete memoryManager;
-
 	// crashes?
 	//glslang::FinalizeProcess();
 }
@@ -791,7 +857,14 @@ void VulkanRenderer::HandleScreenshotRequest(LatteTextureView* texView, bool pad
 			VkMemoryAllocateInfo allocInfo{};
 			allocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
 			allocInfo.allocationSize = memRequirements.size;
-			allocInfo.memoryTypeIndex = memoryManager->FindMemoryType(memRequirements.memoryTypeBits, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
+			uint32 memIndex;
+			bool foundMemory = memoryManager->FindMemoryType(memRequirements.memoryTypeBits, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, memIndex);
+			if(!foundMemory)
+			{
+				cemuLog_log(LogType::Force, "Screenshot request failed due to incompatible vulkan memory types.");
+				return;
+			}
+			allocInfo.memoryTypeIndex = memIndex;
 
 			if (vkAllocateMemory(m_logicalDevice, &allocInfo, nullptr, &imageMemory) != VK_SUCCESS)
 			{
@@ -1047,6 +1120,13 @@ VkDeviceCreateInfo VulkanRenderer::CreateDeviceCreateInfo(const std::vector<VkDe
 		used_extensions.emplace_back(VK_KHR_DYNAMIC_RENDERING_EXTENSION_NAME);
 	if (m_featureControl.deviceExtensions.shader_float_controls)
 		used_extensions.emplace_back(VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME);
+	if (m_featureControl.deviceExtensions.depth_clip_enable)
+		used_extensions.emplace_back(VK_EXT_DEPTH_CLIP_ENABLE_EXTENSION_NAME);
+	if (m_featureControl.deviceExtensions.present_wait)
+	{
+		used_extensions.emplace_back(VK_KHR_PRESENT_ID_EXTENSION_NAME);
+		used_extensions.emplace_back(VK_KHR_PRESENT_WAIT_EXTENSION_NAME);
+	}
 
 	VkDeviceCreateInfo createInfo{};
 	createInfo.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO;
@@ -1143,7 +1223,9 @@ bool VulkanRenderer::CheckDeviceExtensionSupport(const VkPhysicalDevice device,
 	info.deviceExtensions.synchronization2 = isExtensionAvailable(VK_KHR_SYNCHRONIZATION_2_EXTENSION_NAME);
 	info.deviceExtensions.shader_float_controls = isExtensionAvailable(VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME);
 	info.deviceExtensions.dynamic_rendering = false; // isExtensionAvailable(VK_KHR_DYNAMIC_RENDERING_EXTENSION_NAME);
+	info.deviceExtensions.depth_clip_enable = isExtensionAvailable(VK_EXT_DEPTH_CLIP_ENABLE_EXTENSION_NAME);
 	// dynamic rendering doesn't provide any benefits for us right now. Driver implementations are very unoptimized as of Feb 2022
+	info.deviceExtensions.present_wait = isExtensionAvailable(VK_KHR_PRESENT_WAIT_EXTENSION_NAME) && isExtensionAvailable(VK_KHR_PRESENT_ID_EXTENSION_NAME);
 
 	// check for framedebuggers
 	info.debugMarkersSupported = false;
@@ -1513,37 +1595,35 @@ void VulkanRenderer::DeleteNullObjects()
 
 void VulkanRenderer::ImguiInit()
 {
-	if (m_imguiRenderPass == VK_NULL_HANDLE)
-	{
-		// TODO: renderpass swapchain format may change between srgb and rgb -> need reinit
-		VkAttachmentDescription colorAttachment = {};
-		colorAttachment.format = m_mainSwapchainInfo->m_surfaceFormat.format;
-		colorAttachment.samples = VK_SAMPLE_COUNT_1_BIT;
-		colorAttachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
-		colorAttachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
-		colorAttachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
-		colorAttachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
-		colorAttachment.initialLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
-		colorAttachment.finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
+	VkRenderPass prevRenderPass = m_imguiRenderPass;
 
-		VkAttachmentReference colorAttachmentRef = {};
-		colorAttachmentRef.attachment = 0;
-		colorAttachmentRef.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
-		VkSubpassDescription subpass = {};
-		subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
-		subpass.colorAttachmentCount = 1;
-		subpass.pColorAttachments = &colorAttachmentRef;
+	VkAttachmentDescription colorAttachment = {};
+	colorAttachment.format = m_mainSwapchainInfo->m_surfaceFormat.format;
+	colorAttachment.samples = VK_SAMPLE_COUNT_1_BIT;
+	colorAttachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
+	colorAttachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
+	colorAttachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
+	colorAttachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
+	colorAttachment.initialLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
+	colorAttachment.finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
 
-		VkRenderPassCreateInfo renderPassInfo = {};
-		renderPassInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
-		renderPassInfo.attachmentCount = 1;
-		renderPassInfo.pAttachments = &colorAttachment;
-		renderPassInfo.subpassCount = 1;
-		renderPassInfo.pSubpasses = &subpass;
-		const auto result = vkCreateRenderPass(m_logicalDevice, &renderPassInfo, nullptr, &m_imguiRenderPass);
-		if (result != VK_SUCCESS)
-			throw VkException(result, "can't create imgui renderpass");
-	}
+	VkAttachmentReference colorAttachmentRef = {};
+	colorAttachmentRef.attachment = 0;
+	colorAttachmentRef.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
+	VkSubpassDescription subpass = {};
+	subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
+	subpass.colorAttachmentCount = 1;
+	subpass.pColorAttachments = &colorAttachmentRef;
+
+	VkRenderPassCreateInfo renderPassInfo = {};
+	renderPassInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
+	renderPassInfo.attachmentCount = 1;
+	renderPassInfo.pAttachments = &colorAttachment;
+	renderPassInfo.subpassCount = 1;
+	renderPassInfo.pSubpasses = &subpass;
+	const auto result = vkCreateRenderPass(m_logicalDevice, &renderPassInfo, nullptr, &m_imguiRenderPass);
+	if (result != VK_SUCCESS)
+		throw VkException(result, "can't create imgui renderpass");
 
 	ImGui_ImplVulkan_InitInfo info{};
 	info.Instance = m_instance;
@@ -1557,6 +1637,9 @@ void VulkanRenderer::ImguiInit()
 	info.ImageCount = info.MinImageCount;
 
 	ImGui_ImplVulkan_Init(&info, m_imguiRenderPass);
+
+	if (prevRenderPass != VK_NULL_HANDLE)
+		vkDestroyRenderPass(GetLogicalDevice(), prevRenderPass, nullptr);
 }
 
 void VulkanRenderer::Initialize()
@@ -1569,6 +1652,7 @@ void VulkanRenderer::Initialize()
 
 void VulkanRenderer::Shutdown()
 {
+	DeleteFontTextures();
 	Renderer::Shutdown();
 	SubmitCommandBuffer();
 	WaitDeviceIdle();
@@ -1769,7 +1853,6 @@ void VulkanRenderer::ImguiEnd()
 	vkCmdEndRenderPass(m_state.currentCommandBuffer);
 }
 
-std::vector<LatteTextureVk*> g_imgui_textures; // TODO manage better
 ImTextureID VulkanRenderer::GenerateTexture(const std::vector<uint8>& data, const Vector2i& size)
 {
 	try
@@ -1799,6 +1882,7 @@ void VulkanRenderer::DeleteTexture(ImTextureID id)
 
 void VulkanRenderer::DeleteFontTextures()
 {
+	WaitDeviceIdle();
 	ImGui_ImplVulkan_DestroyFontsTexture();
 }
 
@@ -1837,7 +1921,7 @@ void VulkanRenderer::InitFirstCommandBuffer()
 	vkResetFences(m_logicalDevice, 1, &m_cmd_buffer_fences[m_commandBufferIndex]);
 	VkCommandBufferBeginInfo beginInfo{};
 	beginInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
-	beginInfo.flags = VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT;
+	beginInfo.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
 	vkBeginCommandBuffer(m_state.currentCommandBuffer, &beginInfo);
 
 	vkCmdSetViewport(m_state.currentCommandBuffer, 0, 1, &m_state.currentViewport);
@@ -1855,6 +1939,7 @@ void VulkanRenderer::ProcessFinishedCommandBuffers()
 		if (fenceStatus == VK_SUCCESS)
 		{
 			ProcessDestructionQueue();
+			m_uniformVarBufferReadIndex = m_cmdBufferUniformRingbufIndices[m_commandBufferSyncIndex];
 			m_commandBufferSyncIndex = (m_commandBufferSyncIndex + 1) % m_commandBuffers.size();
 			memoryManager->cleanupBuffers(m_countCommandBufferFinished);
 			m_countCommandBufferFinished++;
@@ -1948,6 +2033,7 @@ void VulkanRenderer::SubmitCommandBuffer(VkSemaphore signalSemaphore, VkSemaphor
 		cemuLog_logDebug(LogType::Force, "Vulkan: Waiting for available command buffer...");
 		WaitForNextFinishedCommandBuffer();
 	}
+	m_cmdBufferUniformRingbufIndices[nextCmdBufferIndex] = m_cmdBufferUniformRingbufIndices[m_commandBufferIndex];
 	m_commandBufferIndex = nextCmdBufferIndex;
 
 
@@ -1957,7 +2043,7 @@ void VulkanRenderer::SubmitCommandBuffer(VkSemaphore signalSemaphore, VkSemaphor
 
 	VkCommandBufferBeginInfo beginInfo{};
 	beginInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
-	beginInfo.flags = VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT;
+	beginInfo.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
 	vkBeginCommandBuffer(m_state.currentCommandBuffer, &beginInfo);
 
 	// make sure some states are set for this command buffer
@@ -2478,9 +2564,8 @@ VkPipeline VulkanRenderer::backbufferBlit_createGraphicsPipeline(VkDescriptorSet
 	hash += (uint64)(chainInfo.m_usesSRGB);
 	hash += ((uint64)padView) << 1;
 
-	static std::unordered_map<uint64, VkPipeline> s_pipeline_cache;
-	const auto it = s_pipeline_cache.find(hash);
-	if (it != s_pipeline_cache.cend())
+	const auto it = m_backbufferBlitPipelineCache.find(hash);
+	if (it != m_backbufferBlitPipelineCache.cend())
 		return it->second;
 
 	std::vector<VkPipelineShaderStageCreateInfo> shaderStages;
@@ -2542,10 +2627,18 @@ VkPipeline VulkanRenderer::backbufferBlit_createGraphicsPipeline(VkDescriptorSet
 	colorBlending.blendConstants[2] = 0.0f;
 	colorBlending.blendConstants[3] = 0.0f;
 
+	VkPushConstantRange pushConstantRange{
+		.stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT,
+		.offset = 0,
+		.size = 3 * sizeof(float) * 2 // 3 vec2's
+	};
+
 	VkPipelineLayoutCreateInfo pipelineLayoutInfo{};
 	pipelineLayoutInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
 	pipelineLayoutInfo.setLayoutCount = 1;
 	pipelineLayoutInfo.pSetLayouts = &descriptorLayout;
+	pipelineLayoutInfo.pushConstantRangeCount = 1;
+	pipelineLayoutInfo.pPushConstantRanges = &pushConstantRange;
 
 	VkResult result = vkCreatePipelineLayout(m_logicalDevice, &pipelineLayoutInfo, nullptr, &m_pipelineLayout);
 	if (result != VK_SUCCESS)
@@ -2576,7 +2669,7 @@ VkPipeline VulkanRenderer::backbufferBlit_createGraphicsPipeline(VkDescriptorSet
 		throw std::runtime_error(fmt::format("Failed to create graphics pipeline: {}", result));
 	}
 
-	s_pipeline_cache[hash] = pipeline;
+	m_backbufferBlitPipelineCache[hash] = pipeline;
 	m_pipeline_cache_semaphore.notify();
 
 	return pipeline;
@@ -2695,11 +2788,21 @@ void VulkanRenderer::SwapBuffer(bool mainWindow)
 		ClearColorImageRaw(chainInfo.m_swapchainImages[chainInfo.swapchainImageIndex], 0, 0, clearColor, VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_PRESENT_SRC_KHR);
 	}
 
+	const size_t currentFrameCmdBufferID = GetCurrentCommandBufferId();
+
 	VkSemaphore presentSemaphore = chainInfo.m_presentSemaphores[chainInfo.swapchainImageIndex];
 	SubmitCommandBuffer(presentSemaphore); // submit all command and signal semaphore
 
 	cemu_assert_debug(m_numSubmittedCmdBuffers > 0);
 
+	// wait for the previous frame to finish rendering
+	WaitCommandBufferFinished(m_commandBufferIDOfPrevFrame);
+	m_commandBufferIDOfPrevFrame = currentFrameCmdBufferID;
+
+	chainInfo.WaitAvailableFence();
+
+	VkPresentIdKHR presentId = {};
+
 	VkPresentInfoKHR presentInfo = {};
 	presentInfo.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR;
 	presentInfo.swapchainCount = 1;
@@ -2709,6 +2812,24 @@ void VulkanRenderer::SwapBuffer(bool mainWindow)
 	presentInfo.waitSemaphoreCount = 1;
 	presentInfo.pWaitSemaphores = &presentSemaphore;
 
+	// if present_wait is available and enabled, add frame markers to present requests
+	// and limit the number of queued present operations
+	if (m_featureControl.deviceExtensions.present_wait && chainInfo.m_maxQueued > 0)
+	{
+		presentId.sType = VK_STRUCTURE_TYPE_PRESENT_ID_KHR;
+		presentId.swapchainCount = 1;
+		presentId.pPresentIds = &chainInfo.m_presentId;
+
+		presentInfo.pNext = &presentId;
+
+		if(chainInfo.m_queueDepth >= chainInfo.m_maxQueued)
+		{
+			uint64 waitFrameId = chainInfo.m_presentId - chainInfo.m_queueDepth;
+			vkWaitForPresentKHR(m_logicalDevice, chainInfo.m_swapchain, waitFrameId, 40'000'000);
+			chainInfo.m_queueDepth--;
+		}
+	}
+
 	VkResult result = vkQueuePresentKHR(m_presentQueue, &presentInfo);
 	if (result < 0 && result != VK_ERROR_OUT_OF_DATE_KHR)
 	{
@@ -2717,6 +2838,12 @@ void VulkanRenderer::SwapBuffer(bool mainWindow)
 	if(result == VK_ERROR_OUT_OF_DATE_KHR || result == VK_SUBOPTIMAL_KHR)
 		chainInfo.m_shouldRecreate = true;
 
+	if(result >= 0)
+	{
+		chainInfo.m_queueDepth++;
+		chainInfo.m_presentId++;
+	}
+
 	chainInfo.hasDefinedSwapchainImage = false;
 
 	chainInfo.swapchainImageIndex = -1;
@@ -2839,9 +2966,6 @@ void VulkanRenderer::DrawBackbufferQuad(LatteTextureView* texView, RendererOutpu
 	LatteTextureViewVk* texViewVk = (LatteTextureViewVk*)texView;
 	draw_endRenderPass();
 
-	if (clearBackground)
-		ClearColorbuffer(padView);
-
 	// barrier for input texture
 	VkMemoryBarrier memoryBarrier{};
 	memoryBarrier.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER;
@@ -2878,11 +3002,40 @@ void VulkanRenderer::DrawBackbufferQuad(LatteTextureView* texView, RendererOutpu
 
 	vkCmdBeginRenderPass(m_state.currentCommandBuffer, &renderPassInfo, VK_SUBPASS_CONTENTS_INLINE);
 
+	if (clearBackground)
+	{
+		VkClearAttachment clearAttachment{};
+		clearAttachment.clearValue = {0,0,0,0};
+		clearAttachment.colorAttachment = 0;
+		clearAttachment.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
+		VkClearRect clearExtent = {{{0,0},chainInfo.m_actualExtent}, 0, 1};
+		vkCmdClearAttachments(m_state.currentCommandBuffer, 1, &clearAttachment, 1, &clearExtent);
+	}
+
 	vkCmdBindPipeline(m_state.currentCommandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
 	m_state.currentPipeline = pipeline;
 
 	vkCmdBindDescriptorSets(m_state.currentCommandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, m_pipelineLayout, 0, 1, &descriptSet, 0, nullptr);
 
+	// update push constants
+	Vector2f pushData[3];
+
+	// textureSrcResolution
+	sint32 effectiveWidth, effectiveHeight;
+	texView->baseTexture->GetEffectiveSize(effectiveWidth, effectiveHeight, 0);
+	pushData[0] = {(float)effectiveWidth, (float)effectiveHeight};
+
+	// nativeResolution
+	pushData[1] = {
+		(float)texViewVk->baseTexture->width,
+		(float)texViewVk->baseTexture->height,
+	};
+
+	// outputResolution
+	pushData[2] = {(float)imageWidth,(float)imageHeight};
+
+	vkCmdPushConstants(m_state.currentCommandBuffer, m_pipelineLayout, VK_SHADER_STAGE_FRAGMENT_BIT, 0, sizeof(float) * 2 * 3, &pushData);
+
 	vkCmdDraw(m_state.currentCommandBuffer, 6, 1, 0, 0);
 
 	vkCmdEndRenderPass(m_state.currentCommandBuffer);
@@ -2923,9 +3076,8 @@ VkDescriptorSet VulkanRenderer::backbufferBlit_createDescriptorSet(VkDescriptorS
 	hash += (uint64)texViewVk->GetViewRGBA();
 	hash += (uint64)texViewVk->GetDefaultTextureSampler(useLinearTexFilter);
 
-	static std::unordered_map<uint64, VkDescriptorSet> s_set_cache;
-	const auto it = s_set_cache.find(hash);
-	if (it != s_set_cache.cend())
+	const auto it = m_backbufferBlitDescriptorSetCache.find(hash);
+	if (it != m_backbufferBlitDescriptorSetCache.cend())
 		return it->second;
 
 	VkDescriptorSetAllocateInfo allocInfo = {};
@@ -2956,7 +3108,7 @@ VkDescriptorSet VulkanRenderer::backbufferBlit_createDescriptorSet(VkDescriptorS
 	vkUpdateDescriptorSets(m_logicalDevice, 1, &descriptorWrites, 0, nullptr);
 	performanceMonitor.vk.numDescriptorSamplerTextures.increment();
 
-	s_set_cache[hash] = result;
+	m_backbufferBlitDescriptorSetCache[hash] = result;
 	return result;
 }
 
@@ -3089,7 +3241,8 @@ VkDescriptorSetInfo::~VkDescriptorSetInfo()
 	performanceMonitor.vk.numDescriptorDynUniformBuffers.decrement(statsNumDynUniformBuffers);
 	performanceMonitor.vk.numDescriptorStorageBuffers.decrement(statsNumStorageBuffers);
 
-	VulkanRenderer::GetInstance()->ReleaseDestructibleObject(m_vkObjDescriptorSet);
+	auto renderer = VulkanRenderer::GetInstance();
+	renderer->ReleaseDestructibleObject(m_vkObjDescriptorSet);
 	m_vkObjDescriptorSet = nullptr;
 }
 
@@ -3491,13 +3644,13 @@ void VulkanRenderer::buffer_bindUniformBuffer(LatteConst::ShaderType shaderType,
 	switch (shaderType)
 	{
 	case LatteConst::ShaderType::Vertex:
-		dynamicOffsetInfo.shaderUB[VulkanRendererConst::SHADER_STAGE_INDEX_VERTEX].unformBufferOffset[bufferIndex] = offset;
+		dynamicOffsetInfo.shaderUB[VulkanRendererConst::SHADER_STAGE_INDEX_VERTEX].uniformBufferOffset[bufferIndex] = offset;
 		break;
 	case LatteConst::ShaderType::Geometry:
-		dynamicOffsetInfo.shaderUB[VulkanRendererConst::SHADER_STAGE_INDEX_GEOMETRY].unformBufferOffset[bufferIndex] = offset;
+		dynamicOffsetInfo.shaderUB[VulkanRendererConst::SHADER_STAGE_INDEX_GEOMETRY].uniformBufferOffset[bufferIndex] = offset;
 		break;
 	case LatteConst::ShaderType::Pixel:
-		dynamicOffsetInfo.shaderUB[VulkanRendererConst::SHADER_STAGE_INDEX_FRAGMENT].unformBufferOffset[bufferIndex] = offset;
+		dynamicOffsetInfo.shaderUB[VulkanRendererConst::SHADER_STAGE_INDEX_FRAGMENT].uniformBufferOffset[bufferIndex] = offset;
 		break;
 	default:
 		cemu_assert_debug(false);
@@ -3599,7 +3752,7 @@ void VulkanRenderer::bufferCache_copyStreamoutToMainBuffer(uint32 srcOffset, uin
 
 void VulkanRenderer::AppendOverlayDebugInfo()
 {
-	ImGui::Text("--- Vulkan info ---");
+	ImGui::Text("--- Vulkan debug info ---");
 	ImGui::Text("GfxPipelines   %u", performanceMonitor.vk.numGraphicPipelines.get());
 	ImGui::Text("DescriptorSets %u", performanceMonitor.vk.numDescriptorSets.get());
 	ImGui::Text("DS ImgSamplers %u", performanceMonitor.vk.numDescriptorSamplerTextures.get());
@@ -3607,6 +3760,7 @@ void VulkanRenderer::AppendOverlayDebugInfo()
 	ImGui::Text("DS StorageBuf  %u", performanceMonitor.vk.numDescriptorStorageBuffers.get());
 	ImGui::Text("Images         %u", performanceMonitor.vk.numImages.get());
 	ImGui::Text("ImageView      %u", performanceMonitor.vk.numImageViews.get());
+	ImGui::Text("ImageSampler   %u", performanceMonitor.vk.numSamplers.get());
 	ImGui::Text("RenderPass     %u", performanceMonitor.vk.numRenderPass.get());
 	ImGui::Text("Framebuffer    %u", performanceMonitor.vk.numFramebuffer.get());
 	m_spinlockDestructionQueue.lock();
@@ -3616,7 +3770,7 @@ void VulkanRenderer::AppendOverlayDebugInfo()
 
 	ImGui::Text("BeginRP/f      %u", performanceMonitor.vk.numBeginRenderpassPerFrame.get());
 	ImGui::Text("Barriers/f     %u", performanceMonitor.vk.numDrawBarriersPerFrame.get());
-	ImGui::Text("--- Cache info ---");
+	ImGui::Text("--- Cache debug info ---");
 
 	uint32 bufferCacheHeapSize = 0;
 	uint32 bufferCacheAllocationSize = 0;
@@ -3636,7 +3790,7 @@ void VulkanRenderer::AppendOverlayDebugInfo()
 	ImGui::SameLine(60.0f);
 	ImGui::Text("%06uKB / %06uKB Buffers: %u", ((uint32)(totalSize - freeSize) + 1023) / 1024, ((uint32)totalSize + 1023) / 1024, (uint32)numBuffers);
 
-	memoryManager->getIndexAllocator().GetStats(numBuffers, totalSize, freeSize);
+	memoryManager->GetIndexAllocator().GetStats(numBuffers, totalSize, freeSize);
 	ImGui::Text("Index");
 	ImGui::SameLine(60.0f);
 	ImGui::Text("%06uKB / %06uKB Buffers: %u", ((uint32)(totalSize - freeSize) + 1023) / 1024, ((uint32)totalSize + 1023) / 1024, (uint32)numBuffers);
@@ -3652,7 +3806,7 @@ void VKRDestructibleObject::flagForCurrentCommandBuffer()
 
 bool VKRDestructibleObject::canDestroy()
 {
-	if (refCount > 0)
+	if (m_refCount > 0)
 		return false;
 	return VulkanRenderer::GetInstance()->HasCommandBufferFinished(m_lastCmdBufferId);
 }
@@ -3693,6 +3847,111 @@ VKRObjectTextureView::~VKRObjectTextureView()
 	performanceMonitor.vk.numImageViews.decrement();
 }
 
+static uint64 CalcHashSamplerCreateInfo(const VkSamplerCreateInfo& info)
+{
+	uint64 h = 0xcbf29ce484222325ULL;
+	auto fnvHashCombine = [](uint64_t &h, auto val) {
+		using T = decltype(val);
+		static_assert(sizeof(T) <= 8);
+		uint64_t val64 = 0;
+		std::memcpy(&val64, &val, sizeof(val));
+		h ^= val64;
+		h *= 0x100000001b3ULL;
+	};
+	cemu_assert_debug(info.sType == VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO);
+	fnvHashCombine(h, info.flags);
+	fnvHashCombine(h, info.magFilter);
+	fnvHashCombine(h, info.minFilter);
+	fnvHashCombine(h, info.mipmapMode);
+	fnvHashCombine(h, info.addressModeU);
+	fnvHashCombine(h, info.addressModeV);
+	fnvHashCombine(h, info.addressModeW);
+	fnvHashCombine(h, info.mipLodBias);
+	fnvHashCombine(h, info.anisotropyEnable);
+	if(info.anisotropyEnable == VK_TRUE)
+		fnvHashCombine(h, info.maxAnisotropy);
+	fnvHashCombine(h, info.compareEnable);
+	if(info.compareEnable == VK_TRUE)
+		fnvHashCombine(h, info.compareOp);
+	fnvHashCombine(h, info.minLod);
+	fnvHashCombine(h, info.maxLod);
+	fnvHashCombine(h, info.borderColor);
+	fnvHashCombine(h, info.unnormalizedCoordinates);
+	// handle custom border color
+	VkBaseOutStructure* ext = (VkBaseOutStructure*)info.pNext;
+	while(ext)
+	{
+		if(ext->sType == VK_STRUCTURE_TYPE_SAMPLER_CUSTOM_BORDER_COLOR_CREATE_INFO_EXT)
+		{
+			auto* extInfo = (VkSamplerCustomBorderColorCreateInfoEXT*)ext;
+			fnvHashCombine(h, extInfo->customBorderColor.uint32[0]);
+			fnvHashCombine(h, extInfo->customBorderColor.uint32[1]);
+			fnvHashCombine(h, extInfo->customBorderColor.uint32[2]);
+			fnvHashCombine(h, extInfo->customBorderColor.uint32[3]);
+		}
+		else
+		{
+			cemu_assert_unimplemented();
+		}
+		ext = ext->pNext;
+	}
+	return h;
+}
+
+std::unordered_map<uint64, VKRObjectSampler*> VKRObjectSampler::s_samplerCache;
+
+VKRObjectSampler::VKRObjectSampler(VkSamplerCreateInfo* samplerInfo)
+{
+	auto* vulkanRenderer = VulkanRenderer::GetInstance();
+	if (vkCreateSampler(vulkanRenderer->GetLogicalDevice(), samplerInfo, nullptr, &m_sampler) != VK_SUCCESS)
+		vulkanRenderer->UnrecoverableError("Failed to create texture sampler");
+	performanceMonitor.vk.numSamplers.increment();
+	m_hash = CalcHashSamplerCreateInfo(*samplerInfo);
+}
+
+VKRObjectSampler::~VKRObjectSampler()
+{
+	vkDestroySampler(VulkanRenderer::GetInstance()->GetLogicalDevice(), m_sampler, nullptr);
+	performanceMonitor.vk.numSamplers.decrement();
+	// remove from cache
+	auto it = s_samplerCache.find(m_hash);
+	if(it != s_samplerCache.end())
+		s_samplerCache.erase(it);
+}
+
+void VKRObjectSampler::RefCountReachedZero()
+{
+	VulkanRenderer::GetInstance()->ReleaseDestructibleObject(this);
+}
+
+VKRObjectSampler* VKRObjectSampler::GetOrCreateSampler(VkSamplerCreateInfo* samplerInfo)
+{
+	auto* vulkanRenderer = VulkanRenderer::GetInstance();
+	uint64 hash = CalcHashSamplerCreateInfo(*samplerInfo);
+	auto it = s_samplerCache.find(hash);
+	if (it != s_samplerCache.end())
+	{
+		auto* sampler = it->second;
+		return sampler;
+	}
+	auto* sampler = new VKRObjectSampler(samplerInfo);
+	s_samplerCache[hash] = sampler;
+	return sampler;
+}
+
+void VKRObjectSampler::DestroyCache()
+{
+	// assuming all other objects which depend on vkSampler are destroyed, this cache should also have been emptied already
+	// but just to be sure lets still clear the cache
+	cemu_assert_debug(s_samplerCache.empty());
+	for(auto& sampler : s_samplerCache)
+	{
+		cemu_assert_debug(sampler.second->m_refCount == 0);
+		delete sampler.second;
+	}
+	s_samplerCache.clear();
+}
+
 VKRObjectRenderPass::VKRObjectRenderPass(AttachmentInfo_t& attachmentInfo, sint32 colorAttachmentCount)
 {
 	// generate helper hash for pipeline state
@@ -3860,33 +4119,36 @@ VKRObjectFramebuffer::~VKRObjectFramebuffer()
 
 VKRObjectPipeline::VKRObjectPipeline()
 {
-	// todo
 }
 
-void VKRObjectPipeline::setPipeline(VkPipeline newPipeline)
+void VKRObjectPipeline::SetPipeline(VkPipeline newPipeline)
 {
-	cemu_assert_debug(pipeline == VK_NULL_HANDLE);
-	pipeline = newPipeline;
-	if(newPipeline != VK_NULL_HANDLE)
+	if (m_pipeline == newPipeline)
+		return;
+	cemu_assert_debug(m_pipeline == VK_NULL_HANDLE); // replacing an already assigned pipeline is not intended
+	if(m_pipeline == VK_NULL_HANDLE && newPipeline != VK_NULL_HANDLE)
 		performanceMonitor.vk.numGraphicPipelines.increment();
+	else if(m_pipeline != VK_NULL_HANDLE && newPipeline == VK_NULL_HANDLE)
+		performanceMonitor.vk.numGraphicPipelines.decrement();
+	m_pipeline = newPipeline;
 }
 
 VKRObjectPipeline::~VKRObjectPipeline()
 {
 	auto vkr = VulkanRenderer::GetInstance();
-	if (pipeline != VK_NULL_HANDLE)
+	if (m_pipeline != VK_NULL_HANDLE)
 	{
-		vkDestroyPipeline(vkr->GetLogicalDevice(), pipeline, nullptr);
+		vkDestroyPipeline(vkr->GetLogicalDevice(), m_pipeline, nullptr);
 		performanceMonitor.vk.numGraphicPipelines.decrement();
 	}
-	if (vertexDSL != VK_NULL_HANDLE)
-		vkDestroyDescriptorSetLayout(vkr->GetLogicalDevice(), vertexDSL, nullptr);
-	if (pixelDSL != VK_NULL_HANDLE)
-		vkDestroyDescriptorSetLayout(vkr->GetLogicalDevice(), pixelDSL, nullptr);
-	if (geometryDSL != VK_NULL_HANDLE)
-		vkDestroyDescriptorSetLayout(vkr->GetLogicalDevice(), geometryDSL, nullptr);
-	if (pipeline_layout != VK_NULL_HANDLE)
-		vkDestroyPipelineLayout(vkr->GetLogicalDevice(), pipeline_layout, nullptr);
+	if (m_vertexDSL != VK_NULL_HANDLE)
+		vkDestroyDescriptorSetLayout(vkr->GetLogicalDevice(), m_vertexDSL, nullptr);
+	if (m_pixelDSL != VK_NULL_HANDLE)
+		vkDestroyDescriptorSetLayout(vkr->GetLogicalDevice(), m_pixelDSL, nullptr);
+	if (m_geometryDSL != VK_NULL_HANDLE)
+		vkDestroyDescriptorSetLayout(vkr->GetLogicalDevice(), m_geometryDSL, nullptr);
+	if (m_pipelineLayout != VK_NULL_HANDLE)
+		vkDestroyPipelineLayout(vkr->GetLogicalDevice(), m_pipelineLayout, nullptr);
 }
 
 VKRObjectDescriptorSet::VKRObjectDescriptorSet()

src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h

@@ -137,8 +137,8 @@ class VulkanRenderer : public Renderer
 public:
 
 	// memory management
-	VKRMemoryManager* memoryManager{};
-	VKRMemoryManager* GetMemoryManager() const { return memoryManager; };
+	std::unique_ptr<VKRMemoryManager> memoryManager;
+	VKRMemoryManager* GetMemoryManager() const { return memoryManager.get(); };
 
 	VkSupportedFormatInfo_t m_supportedFormatInfo;
 
@@ -328,8 +328,9 @@ public:
 
 	RendererShader* shader_create(RendererShader::ShaderType type, uint64 baseHash, uint64 auxHash, const std::string& source, bool isGameShader, bool isGfxPackShader) override;
 
-	void* indexData_reserveIndexMemory(uint32 size, uint32& offset, uint32& bufferIndex) override;
-	void indexData_uploadIndexMemory(uint32 offset, uint32 size) override;
+	IndexAllocation indexData_reserveIndexMemory(uint32 size) override;
+	void indexData_releaseIndexMemory(IndexAllocation& allocation) override;
+	void indexData_uploadIndexMemory(IndexAllocation& allocation) override;
 
 	// externally callable
 	void GetTextureFormatInfoVK(Latte::E_GX2SURFFMT format, bool isDepth, Latte::E_DIM dim, sint32 width, sint32 height, FormatInfoVK* formatInfoOut);
@@ -450,6 +451,8 @@ private:
 			bool synchronization2 = false; // VK_KHR_synchronization2
 			bool dynamic_rendering = false; // VK_KHR_dynamic_rendering
 			bool shader_float_controls = false; // VK_KHR_shader_float_controls
+			bool present_wait = false; // VK_KHR_present_wait
+			bool depth_clip_enable = false; // VK_EXT_depth_clip_enable
 		}deviceExtensions;
 
 		struct
@@ -457,7 +460,7 @@ private:
 			bool shaderRoundingModeRTEFloat32{ false };
 		}shaderFloatControls; // from VK_KHR_shader_float_controls
 
-		struct  
+		struct
 		{
 			bool debug_utils = false; // VK_EXT_DEBUG_UTILS
 		}instanceExtensions;
@@ -581,6 +584,8 @@ private:
 	std::shared_mutex m_pipeline_cache_save_mutex;
 	std::thread m_pipeline_cache_save_thread;
 	VkPipelineCache m_pipeline_cache{ nullptr };
+	std::unordered_map<uint64, VkPipeline> m_backbufferBlitPipelineCache;
+	std::unordered_map<uint64, VkDescriptorSet> m_backbufferBlitDescriptorSetCache;
 	VkPipelineLayout m_pipelineLayout{nullptr};
 	VkCommandPool m_commandPool{ nullptr };
 	
@@ -590,6 +595,7 @@ private:
 	bool m_uniformVarBufferMemoryIsCoherent{false};
 	uint8* m_uniformVarBufferPtr = nullptr;
 	uint32 m_uniformVarBufferWriteIndex = 0;
+	uint32 m_uniformVarBufferReadIndex = 0;
 
 	// transform feedback ringbuffer
 	VkBuffer m_xfbRingBuffer = VK_NULL_HANDLE;
@@ -635,6 +641,8 @@ private:
 
 	size_t m_commandBufferIndex = 0; // current buffer being filled
 	size_t m_commandBufferSyncIndex = 0; // latest buffer that finished execution (updated on submit)
+	size_t m_commandBufferIDOfPrevFrame = 0;
+	std::array<size_t, kCommandBufferPoolSize> m_cmdBufferUniformRingbufIndices {}; // index in the uniform ringbuffer
 	std::array<VkFence, kCommandBufferPoolSize> m_cmd_buffer_fences;
 	std::array<VkCommandBuffer, kCommandBufferPoolSize> m_commandBuffers;
 	std::array<VkSemaphore, kCommandBufferPoolSize> m_commandBufferSemaphores;
@@ -657,7 +665,7 @@ private:
 		uint32 uniformVarBufferOffset[VulkanRendererConst::SHADER_STAGE_INDEX_COUNT];
 		struct  
 		{
-			uint32 unformBufferOffset[LATTE_NUM_MAX_UNIFORM_BUFFERS];
+			uint32 uniformBufferOffset[LATTE_NUM_MAX_UNIFORM_BUFFERS];
 		}shaderUB[VulkanRendererConst::SHADER_STAGE_INDEX_COUNT];
 	}dynamicOffsetInfo{};
 
@@ -855,7 +863,7 @@ private:
 		memBarrier.pNext = nullptr;
 
 		VkPipelineStageFlags srcStages = VK_PIPELINE_STAGE_TRANSFER_BIT;
-		VkPipelineStageFlags dstStages = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
+		VkPipelineStageFlags dstStages = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT;
 
 		memBarrier.srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
 		memBarrier.dstAccessMask = 0;