Merge remote-tracking branch 'remotes/origin/master' into dsi_camera

2020-12-04 18:28:15 +01:00 · 2020-12-04 18:28:15 +01:00 · 129018a662
parent 6aad429383 42e083960e
commit 129018a662
51 changed files with 1772 additions and 760 deletions
--- a/.github/workflows/build-macos.yml
+++ b/.github/workflows/build-macos.yml
@ -0,0 +1,39 @@
+name: CMake Build (macOS x86-64)
+
+on:
+  push:
+    branches:
+      - master
+  pull_request:
+    branches:
+      - master
+
+env:
+  BUILD_TYPE: Release
+
+jobs:
+  build:
+
+    runs-on: macos-latest
+
+    steps:
+    - uses: actions/checkout@v1
+    - name: Install dependencies
+      working-directory: ${{runner.workspace}}
+      run: |
+        brew install cmake sdl2 qt5 libslirp
+    - name: Create build environment
+      run: mkdir ${{runner.workspace}}/build
+    - name: Configure
+      working-directory: ${{runner.workspace}}/build
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DQt5_DIR=$(brew --prefix qt5)/lib/cmake/Qt5
+    - name: Make
+      working-directory: ${{runner.workspace}}/build
+      run: |
+        make -j$(sysctl -n hw.ncpu)
+        mkdir dist
+        cp -r melonDS.app dist
+    - uses: actions/upload-artifact@v1
+      with:
+        name: melonDS.app
+        path: ${{runner.workspace}}/build/dist
--- a/.github/workflows/build-ubuntu-aarch64.yml
+++ b/.github/workflows/build-ubuntu-aarch64.yml
@ -29,6 +29,7 @@ jobs:
      shell: bash
      working-directory: ${{runner.workspace}}
      run: |
+        sudo rm -f /etc/apt/sources.list.d/kubernetes.list
        sudo dpkg --add-architecture arm64
        sudo sh -c "sed \"s|^deb \([a-z\.:/]*\) \([a-z\-]*\) \(.*\)$|deb [arch=amd64] \1 \2 \3\ndeb [arch=arm64] http://ports.ubuntu.com/ubuntu-ports \2 \3|\" /etc/apt/sources.list > /etc/apt/sources.list.new"
        sudo rm /etc/apt/sources.list
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@ -1,20 +1,30 @@
 cmake_minimum_required(VERSION 3.13)

+include(CheckSymbolExists)
+include(CheckLibraryExists)
+
 cmake_policy(VERSION 3.13)
 if (POLICY CMP0076)
    cmake_policy(SET CMP0076 NEW)
 endif()

-set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_OSX_DEPLOYMENT_TARGET "10.9" CACHE STRING "Minimum OS X deployment version")
+
+project(melonDS CXX)
+
+set(CMAKE_C_STANDARD 11)
+set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})

-project(melonDS)
-
-if (NOT CMAKE_BUILD_TYPE)
-  set(CMAKE_BUILD_TYPE Release)
+check_library_exists(m pow "" LIBM)
+if(LIBM)
+    link_libraries(m)
+endif()
+
+if (NOT CMAKE_BUILD_TYPE)
+    set(CMAKE_BUILD_TYPE Release)
 endif()

-include(CheckSymbolExists)
 function(detect_architecture symbol arch)
    if (NOT DEFINED ARCHITECTURE)
        set(CMAKE_REQUIRED_QUIET 1)
@ -65,8 +75,22 @@ if (CMAKE_BUILD_TYPE STREQUAL Release)
 	add_link_options(-s)
 endif()

-add_compile_options(-fno-pic)
-add_link_options(-no-pie)
+if (ENABLE_LTO)
+	add_compile_options(-flto -fPIC)
+	add_link_options(-flto -fuse-linker-plugin -pie)
+endif()
+
+if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+	set(CMAKE_AR "gcc-ar")
+	set(CMAKE_RANLIB "gcc-ranlib")
+elseif (CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+	find_program(LLD NAMES ld.lld ld64.lld lld-link)
+	if (NOT LLD STREQUAL "LLD-NOTFOUND")
+		add_link_options(-fuse-ld=lld)
+	endif()
+	set(CMAKE_AR "llvm-ar")
+	set(CMAKE_RANLIB "llvm-ranlib")
+endif()

 option(BUILD_QT_SDL "Build Qt/SDL frontend" ON)

@ -78,4 +102,4 @@ add_subdirectory(src)

 if (BUILD_QT_SDL)
 	add_subdirectory(src/frontend/qt_sdl)
-endif()
+endif()
--- a/README.md
+++ b/README.md
@ -38,7 +38,7 @@ As for the rest, the interface should be pretty straightforward. If you have a q
 * Install dependencies:

 ```sh
-sudo apt-get install cmake libgtk-3-dev libcurl4-gnutls-dev libpcap0.8-dev libsdl2-dev qtbase5-dev qtdeclarative5-dev libslirp-dev
+sudo apt-get install cmake libcurl4-gnutls-dev libpcap0.8-dev libsdl2-dev qtbase5-dev qtdeclarative5-dev libslirp-dev
 ```

 * Compile:
@ -84,6 +84,21 @@ If everything went well, melonDS and the libraries it needs should now be in the
   ```
 If everything went well, melonDS should now be in the `dist` folder.

+### macOS:
+1. Install the [Homebrew Package Manager](https://brew.sh)
+2. Install dependencies: `brew install git pkg-config cmake sdl2 qt5 libslirp`
+3. Compile:
+   ```zsh
+   git clone https://github.com/Arisotura/melonDS.git
+   cd melonDS
+   mkdir build && cd build
+   cmake .. -DQt5_DIR=$(brew --prefix qt5)/lib/cmake/Qt5
+   make -j$(sysctl -n hw.ncpu)
+   mkdir dist && cp -r melonDS.app dist
+   ```
+If everything went well, melonDS.app should now be in the `dist` folder.
+
+   
 ## TODO LIST

 * DSi emulation
--- a/melonDS.icns
+++ b/melonDS.icns
--- a/melonDS.plist
+++ b/melonDS.plist
@ -0,0 +1,24 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple Computer//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+    <key>CFBundleExecutable</key>
+    <string>melonDS</string>
+    <key>CFBundleIconFile</key>
+    <string>melonDS.icns</string>
+    <key>CFBundleIdentifier</key>
+    <string>net.kuribo64.melonDS</string>
+    <key>CFBundleDevelopmentRegion</key>
+    <string>English</string>
+    <key>CFBundlePackageType</key>
+    <string>APPL</string>
+    <key>CFBundleVersion</key>
+    <string>0.9</string>
+    <key>CFBundleShortVersionString</key>
+    <string>0.9</string> 
+    <key>NSHumanReadableCopyright</key>
+    <string>Licensed under GPLv3</string>
+    <key>NSHighResolutionCapable</key>
+    <true/>
+</dict>
+</plist>
--- a/src/ARMJIT.cpp
+++ b/src/ARMJIT.cpp
@ -176,7 +176,7 @@ T SlowRead9(u32 addr, ARMv5* cpu)
 }

 template <typename T, int ConsoleType>
-void SlowWrite9(u32 addr, ARMv5* cpu, T val)
+void SlowWrite9(u32 addr, ARMv5* cpu, u32 val)
 {
    addr &= ~(sizeof(T) - 1);

@ -224,7 +224,7 @@ T SlowRead7(u32 addr)
 }

 template <typename T, int ConsoleType>
-void SlowWrite7(u32 addr, T val)
+void SlowWrite7(u32 addr, u32 val)
 {
    addr &= ~(sizeof(T) - 1);

@ -266,16 +266,16 @@ void SlowBlockTransfer7(u32 addr, u64* data, u32 num)

 #define INSTANTIATE_SLOWMEM(consoleType) \
    template void SlowWrite9<u32, consoleType>(u32, ARMv5*, u32); \
-    template void SlowWrite9<u16, consoleType>(u32, ARMv5*, u16); \
-    template void SlowWrite9<u8, consoleType>(u32, ARMv5*, u8); \
+    template void SlowWrite9<u16, consoleType>(u32, ARMv5*, u32); \
+    template void SlowWrite9<u8, consoleType>(u32, ARMv5*, u32); \
    \
    template u32 SlowRead9<u32, consoleType>(u32, ARMv5*); \
    template u16 SlowRead9<u16, consoleType>(u32, ARMv5*); \
    template u8 SlowRead9<u8, consoleType>(u32, ARMv5*); \
    \
    template void SlowWrite7<u32, consoleType>(u32, u32); \
-    template void SlowWrite7<u16, consoleType>(u32, u16); \
-    template void SlowWrite7<u8, consoleType>(u32, u8); \
+    template void SlowWrite7<u16, consoleType>(u32, u32); \
+    template void SlowWrite7<u8, consoleType>(u32, u32); \
    \
    template u32 SlowRead7<u32, consoleType>(u32); \
    template u16 SlowRead7<u16, consoleType>(u32); \
@ -298,6 +298,7 @@ void Init()

 void DeInit()
 {
+    ResetBlockCache();
    ARMJIT_Memory::DeInit();

    delete JITCompiler;
@ -594,7 +595,8 @@ void CompileBlock(ARM* cpu)
    u32 r15 = cpu->R[15];

    u32 addressRanges[Config::JIT_MaxBlockSize];
-    u32 addressMasks[Config::JIT_MaxBlockSize] = {0};
+    u32 addressMasks[Config::JIT_MaxBlockSize];
+    memset(addressMasks, 0, Config::JIT_MaxBlockSize * sizeof(u32));
    u32 numAddressRanges = 0;

    u32 numLiterals = 0;
@ -1116,6 +1118,7 @@ void ResetBlockCache()
            range->Blocks.Clear();
            range->Code = 0;
        }
+        delete block;
    }
    JitBlocks9.clear();
    JitBlocks7.clear();
--- a/src/ARMJIT_A64/ARMJIT_Compiler.cpp
+++ b/src/ARMJIT_A64/ARMJIT_Compiler.cpp
@ -68,6 +68,11 @@ void Compiler::A_Comp_MRS()
        MOV(rd, RCPSR);
 }

+void UpdateModeTrampoline(ARM* arm, u32 oldmode, u32 newmode)
+{
+    arm->UpdateMode(oldmode, newmode);
+}
+
 void Compiler::A_Comp_MSR()
 {
    Comp_AddCycles_C();
@ -139,7 +144,7 @@ void Compiler::A_Comp_MSR()

            PushRegs(true);

-            QuickCallFunction(X3, (void*)&ARM::UpdateMode);
+            QuickCallFunction(X3, (void*)&UpdateModeTrampoline);
        
            PopRegs(true);
        }
@ -915,4 +920,4 @@ void Compiler::Comp_AddCycles_CD()
        ConstantCycles += cycles;
 }

-}
+}
--- a/src/ARMJIT_A64/ARMJIT_Compiler.h
+++ b/src/ARMJIT_A64/ARMJIT_Compiler.h
@ -187,6 +187,7 @@ public:
    void Comp_RegShiftReg(int op, bool S, Op2& op2, Arm64Gen::ARM64Reg rs);

    bool Comp_MemLoadLiteral(int size, bool signExtend, int rd, u32 addr);
+
    enum
    {
        memop_Writeback = 1 << 0,
@ -213,8 +214,8 @@ public:
        return (u8*)entry - GetRXBase();
    }

-    bool IsJITFault(u64 pc);
-    s64 RewriteMemAccess(u64 pc);
+    bool IsJITFault(u8* pc);
+    u8* RewriteMemAccess(u8* pc);

    void SwapCodeRegion()
    {
--- a/src/ARMJIT_A64/ARMJIT_Linkage.S
+++ b/src/ARMJIT_A64/ARMJIT_Linkage.S
--- a/src/ARMJIT_A64/ARMJIT_LoadStore.cpp
+++ b/src/ARMJIT_A64/ARMJIT_LoadStore.cpp
@ -9,37 +9,34 @@ using namespace Arm64Gen;
 namespace ARMJIT
 {

-bool Compiler::IsJITFault(u64 pc)
+bool Compiler::IsJITFault(u8* pc)
 {
-    return pc >= (u64)GetRXBase() && pc - (u64)GetRXBase() < (JitMemMainSize + JitMemSecondarySize);
+    return (u64)pc >= (u64)GetRXBase() && (u64)pc - (u64)GetRXBase() < (JitMemMainSize + JitMemSecondarySize);
 }

-s64 Compiler::RewriteMemAccess(u64 pc)
+u8* Compiler::RewriteMemAccess(u8* pc)
 {
-    ptrdiff_t pcOffset = pc - (u64)GetRXBase();
+    ptrdiff_t pcOffset = pc - GetRXBase();

    auto it = LoadStorePatches.find(pcOffset);

    if (it != LoadStorePatches.end())
    {
        LoadStorePatch patch = it->second;
+        LoadStorePatches.erase(it);

        ptrdiff_t curCodeOffset = GetCodeOffset();

        SetCodePtrUnsafe(pcOffset + patch.PatchOffset);

        BL(patch.PatchFunc);
-
        for (int i = 0; i < patch.PatchSize / 4 - 1; i++)
            HINT(HINT_NOP);
-
        FlushIcacheSection((u8*)pc + patch.PatchOffset, (u8*)GetRXPtr());

        SetCodePtrUnsafe(curCodeOffset);

-        LoadStorePatches.erase(it);
-
-        return patch.PatchOffset;
+        return pc + (ptrdiff_t)patch.PatchOffset;
    }
    printf("this is a JIT bug! %08x\n", __builtin_bswap32(*(u32*)pc));
    abort();
@ -192,7 +189,7 @@ void Compiler::Comp_MemAccess(int rd, int rn, Op2 offset, int size, int flags)
        else
        {
            LDRGeneric(size, flags & memop_SignExtend, rdMapped, size > 8 ? X1 : X0, X7);
-            if (size == 32)
+            if (size == 32 && !addrIsStatic)
            {
                UBFIZ(W0, W0, 3, 2);
                RORV(rdMapped, rdMapped, W0);
--- a/src/ARMJIT_Internal.h
+++ b/src/ARMJIT_Internal.h
@ -216,9 +216,9 @@ template <u32 Num>
 void LinkBlock(ARM* cpu, u32 codeOffset);

 template <typename T, int ConsoleType> T SlowRead9(u32 addr, ARMv5* cpu);
-template <typename T, int ConsoleType> void SlowWrite9(u32 addr, ARMv5* cpu, T val);
+template <typename T, int ConsoleType> void SlowWrite9(u32 addr, ARMv5* cpu, u32 val);
 template <typename T, int ConsoleType> T SlowRead7(u32 addr);
-template <typename T, int ConsoleType> void SlowWrite7(u32 addr, T val);
+template <typename T, int ConsoleType> void SlowWrite7(u32 addr, u32 val);

 template <bool Write, int ConsoleType> void SlowBlockTransfer9(u32 addr, u64* data, u32 num, ARMv5* cpu);
 template <bool Write, int ConsoleType> void SlowBlockTransfer7(u32 addr, u64* data, u32 num);
--- a/src/ARMJIT_Memory.cpp
+++ b/src/ARMJIT_Memory.cpp
@ -10,6 +10,12 @@
 #include <signal.h>
 #endif

+#if defined(__ANDROID__)
+#include <dlfcn.h>
+#include <linux/ashmem.h>
+#include <sys/ioctl.h>
+#endif
+
 #include "ARMJIT_Memory.h"

 #include "ARMJIT_Internal.h"
@ -22,7 +28,9 @@
 #include "NDSCart.h"
 #include "SPU.h"

+#ifndef __APPLE__
 #include <malloc.h>
+#endif

 /*
    We're handling fastmem here.
@ -40,7 +48,8 @@
    We handle this by only mapping those regions which are actually
    used and by praying the games don't go wild.

-    Beware, this file is full of platform specific code.
+    Beware, this file is full of platform specific code and copied
+    from Dolphin, so enjoy the copied comments!

 */

@ -49,12 +58,16 @@ namespace ARMJIT_Memory
 struct FaultDescription
 {
    u32 EmulatedFaultAddr;
-    u64 FaultPC;
+    u8* FaultPC;
 };

-bool FaultHandler(FaultDescription* faultDesc, s32& offset);
+bool FaultHandler(FaultDescription& faultDesc);
 }

+#if defined(__ANDROID__)
+#define ASHMEM_DEVICE "/dev/ashmem"
+#endif
+
 #if defined(__SWITCH__)
 // with LTO the symbols seem to be not properly overriden
 // if they're somewhere else
@ -75,7 +88,7 @@ void __libnx_exception_handler(ThreadExceptionDump* ctx)
    ARMJIT_Memory::FaultDescription desc;
    u8* curArea = (u8*)(NDS::CurCPU == 0 ? ARMJIT_Memory::FastMem9Start : ARMJIT_Memory::FastMem7Start);
    desc.EmulatedFaultAddr = (u8*)ctx->far.x - curArea;
-    desc.FaultPC = ctx->pc.x;
+    desc.FaultPC = (u8*)ctx->pc.x;

    u64 integerRegisters[33];
    memcpy(integerRegisters, &ctx->cpu_gprs[0].x, 8*29);
@ -84,10 +97,9 @@ void __libnx_exception_handler(ThreadExceptionDump* ctx)
    integerRegisters[31] = ctx->sp.x;
    integerRegisters[32] = ctx->pc.x;

-    s32 offset;
-    if (ARMJIT_Memory::FaultHandler(&desc, offset))
+    if (ARMJIT_Memory::FaultHandler(desc, offset))
    {
-        integerRegisters[32] += offset;
+        integerRegisters[32] = (u64)desc.FaultPC;

        ARM_RestoreContext(integerRegisters);	
    }
@ -117,12 +129,11 @@ static LONG ExceptionHandler(EXCEPTION_POINTERS* exceptionInfo)
    ARMJIT_Memory::FaultDescription desc;
    u8* curArea = (u8*)(NDS::CurCPU == 0 ? ARMJIT_Memory::FastMem9Start : ARMJIT_Memory::FastMem7Start);
    desc.EmulatedFaultAddr = (u8*)exceptionInfo->ExceptionRecord->ExceptionInformation[1] - curArea;
-    desc.FaultPC = exceptionInfo->ContextRecord->Rip;
+    desc.FaultPC = (u8*)exceptionInfo->ContextRecord->Rip;

-    s32 offset = 0;
-    if (ARMJIT_Memory::FaultHandler(&desc, offset))
+    if (ARMJIT_Memory::FaultHandler(desc))
    {
-        exceptionInfo->ContextRecord->Rip += offset;
+        exceptionInfo->ContextRecord->Rip = (u64)desc.FaultPC;
        return EXCEPTION_CONTINUE_EXECUTION;
    }

@ -131,50 +142,75 @@ static LONG ExceptionHandler(EXCEPTION_POINTERS* exceptionInfo)

 #else

-struct sigaction NewSa;
-struct sigaction OldSa;
+static struct sigaction OldSaSegv;
+static struct sigaction OldSaBus;

 static void SigsegvHandler(int sig, siginfo_t* info, void* rawContext)
 {
+    if (sig != SIGSEGV && sig != SIGBUS)
+    {
+        // We are not interested in other signals - handle it as usual.
+        return;
+    }
+    if (info->si_code != SEGV_MAPERR && info->si_code != SEGV_ACCERR)
+    {
+        // Huh? Return.
+        return;
+    }
+
    ucontext_t* context = (ucontext_t*)rawContext;
-    
+
    ARMJIT_Memory::FaultDescription desc;
    u8* curArea = (u8*)(NDS::CurCPU == 0 ? ARMJIT_Memory::FastMem9Start : ARMJIT_Memory::FastMem7Start);
 #ifdef __x86_64__
    desc.EmulatedFaultAddr = (u8*)info->si_addr - curArea;
-    desc.FaultPC = context->uc_mcontext.gregs[REG_RIP];
+    #ifdef __APPLE__
+        desc.FaultPC = (u8*)context->uc_mcontext->__ss.__rip;
+    #else
+        desc.FaultPC = (u8*)context->uc_mcontext.gregs[REG_RIP];
+    #endif
+   
 #else
    desc.EmulatedFaultAddr = (u8*)context->uc_mcontext.fault_address - curArea;
-    desc.FaultPC = context->uc_mcontext.pc;
+    desc.FaultPC = (u8*)context->uc_mcontext.pc;
 #endif

-    s32 offset = 0;
-    if (ARMJIT_Memory::FaultHandler(&desc, offset))
+    if (ARMJIT_Memory::FaultHandler(desc))
    {
 #ifdef __x86_64__
-        context->uc_mcontext.gregs[REG_RIP] += offset;
+        #ifdef __APPLE__
+            context->uc_mcontext->__ss.__rip = (u64)desc.FaultPC;
+        #else
+            context->uc_mcontext.gregs[REG_RIP] = (u64)desc.FaultPC;
+        #endif
 #else
-        context->uc_mcontext.pc += offset;
+        context->uc_mcontext.pc = (u64)desc.FaultPC;
 #endif
        return;
    }

-    if (OldSa.sa_flags & SA_SIGINFO)
+    struct sigaction* oldSa;
+    if (sig == SIGSEGV)
+      oldSa = &OldSaSegv;
+    else
+      oldSa = &OldSaBus;
+
+    if (oldSa->sa_flags & SA_SIGINFO)
    {
-      OldSa.sa_sigaction(sig, info, rawContext);
+      oldSa->sa_sigaction(sig, info, rawContext);
      return;
    }
-    if (OldSa.sa_handler == SIG_DFL)
+    if (oldSa->sa_handler == SIG_DFL)
    {
      signal(sig, SIG_DFL);
      return;
    }
-    if (OldSa.sa_handler == SIG_IGN)
+    if (oldSa->sa_handler == SIG_IGN)
    {
      // Ignore signal
      return;
    }
-    OldSa.sa_handler(sig);
+    oldSa->sa_handler(sig);
 }

 #endif
@ -231,7 +267,7 @@ enum
 {
    memstate_Unmapped,
    memstate_MappedRW,
-    // on switch this is unmapped as well
+    // on Switch this is unmapped as well
    memstate_MappedProtected,
 };

@ -314,14 +350,16 @@ struct Mapping

    void Unmap(int region)
    {
+        u32 dtcmStart = NDS::ARM9->DTCMBase;
+        u32 dtcmSize = NDS::ARM9->DTCMSize;
        bool skipDTCM = Num == 0 && region != memregion_DTCM;
        u8* statuses = Num == 0 ? MappingStatus9 : MappingStatus7;
        u32 offset = 0;
        while (offset < Size)
        {
-            if (skipDTCM && Addr + offset == NDS::ARM9->DTCMBase)
+            if (skipDTCM && Addr + offset == dtcmStart)
            {
-                offset += NDS::ARM9->DTCMSize;
+                offset += dtcmSize;
            }
            else
            {
@ -329,7 +367,7 @@ struct Mapping
                u8 status = statuses[(Addr + offset) >> 12];
                while (statuses[(Addr + offset) >> 12] == status
                    && offset < Size
-                    && (!skipDTCM || Addr + offset != NDS::ARM9->DTCMBase))
+                    && (!skipDTCM || Addr + offset != dtcmStart))
                {
                    assert(statuses[(Addr + offset) >> 12] != memstate_Unmapped);
                    statuses[(Addr + offset) >> 12] = memstate_Unmapped;
@ -347,9 +385,33 @@ struct Mapping
 #endif
            }
        }
+
 #ifndef __SWITCH__
-        bool succeded = UnmapFromRange(Addr, Num, OffsetsPerRegion[region] + LocalOffset, Size);
-        assert(succeded);
+#ifndef _WIN32
+        u32 dtcmEnd = dtcmStart + dtcmSize;
+        if (Num == 0
+            && dtcmEnd >= Addr
+            && dtcmStart < Addr + Size)
+        {
+            bool success;
+            if (dtcmStart > Addr)
+            {
+                success = UnmapFromRange(Addr, 0, OffsetsPerRegion[region] + LocalOffset, dtcmStart - Addr);
+                assert(success);
+            }
+            if (dtcmEnd < Addr + Size)
+            {
+                u32 offset = dtcmStart - Addr + dtcmSize;
+                success = UnmapFromRange(dtcmEnd, 0, OffsetsPerRegion[region] + LocalOffset + offset, Size - offset);
+                assert(success);
+            }
+        }
+        else
+#endif
+        {
+            bool succeded = UnmapFromRange(Addr, Num, OffsetsPerRegion[region] + LocalOffset, Size);
+            assert(succeded);
+        }
 #endif
    }
 };
@ -418,10 +480,10 @@ void RemapDTCM(u32 newBase, u32 newSize)

            printf("unmapping %d %x %x %x %x\n", region, mapping.Addr, mapping.Size, mapping.Num, mapping.LocalOffset);

-            bool oldOverlap = NDS::ARM9->DTCMSize > 0 && !(oldDTCMBase >= end || oldDTCBEnd <= start);
-            bool newOverlap = newSize > 0 && !(newBase >= end || newEnd <= start);
+            bool overlap = (NDS::ARM9->DTCMSize > 0 && oldDTCMBase < end && oldDTCBEnd > start)
+                || (newSize > 0 && newBase < end && newEnd > start);

-            if (mapping.Num == 0 && (oldOverlap || newOverlap))
+            if (mapping.Num == 0 && overlap)
            {
                mapping.Unmap(region);
                Mappings[region].Remove(i);
@ -445,8 +507,8 @@ void RemapNWRAM(int num)
    for (int i = 0; i < Mappings[memregion_SharedWRAM].Length;)
    {
        Mapping& mapping = Mappings[memregion_SharedWRAM][i];
-        if (!(DSi::NWRAMStart[mapping.Num][num] >= mapping.Addr + mapping.Size
-            || DSi::NWRAMEnd[mapping.Num][num] < mapping.Addr))
+        if (DSi::NWRAMStart[mapping.Num][num] < mapping.Addr + mapping.Size
+            && DSi::NWRAMEnd[mapping.Num][num] > mapping.Addr)
        {
            mapping.Unmap(memregion_SharedWRAM);
            Mappings[memregion_SharedWRAM].Remove(i);
@ -469,7 +531,7 @@ void RemapSWRAM()
    for (int i = 0; i < Mappings[memregion_WRAM7].Length;)
    {
        Mapping& mapping = Mappings[memregion_WRAM7][i];
-        if (mapping.Addr + mapping.Size < 0x03800000)
+        if (mapping.Addr + mapping.Size <= 0x03800000)
        {
            mapping.Unmap(memregion_WRAM7);
            Mappings[memregion_WRAM7].Remove(i);
@ -501,26 +563,53 @@ bool MapAtAddress(u32 addr)
        return false;

    u8* states = num == 0 ? MappingStatus9 : MappingStatus7;
-    printf("trying to create mapping %x, %x %x %d %d\n", mirrorStart, mirrorSize, memoryOffset, region, num);
+    printf("mapping mirror %x, %x %x %d %d\n", mirrorStart, mirrorSize, memoryOffset, region, num);
    bool isExecutable = ARMJIT::CodeMemRegions[region];

+    u32 dtcmStart = NDS::ARM9->DTCMBase;
+    u32 dtcmSize = NDS::ARM9->DTCMSize;
+    u32 dtcmEnd = dtcmStart + dtcmSize;
 #ifndef __SWITCH__
-    bool succeded = MapIntoRange(mirrorStart, num, OffsetsPerRegion[region] + memoryOffset, mirrorSize);
-    assert(succeded);
+#ifndef _WIN32
+    if (num == 0
+        && dtcmEnd >= mirrorStart
+        && dtcmStart < mirrorStart + mirrorSize)
+    {
+        bool success;
+        if (dtcmStart > mirrorStart)
+        {
+            success = MapIntoRange(mirrorStart, 0, OffsetsPerRegion[region] + memoryOffset, dtcmStart - mirrorStart);
+            assert(success);
+        }
+        if (dtcmEnd < mirrorStart + mirrorSize)
+        {
+            u32 offset = dtcmStart - mirrorStart + dtcmSize;
+            success = MapIntoRange(dtcmEnd, 0, OffsetsPerRegion[region] + memoryOffset + offset, mirrorSize - offset);
+            assert(success);
+        }
+    }
+    else
+#endif
+    {
+        bool succeded = MapIntoRange(mirrorStart, num, OffsetsPerRegion[region] + memoryOffset, mirrorSize);
+        assert(succeded);
+    }
 #endif

    ARMJIT::AddressRange* range = ARMJIT::CodeMemRegions[region] + memoryOffset / 512;

    // this overcomplicated piece of code basically just finds whole pieces of code memory
-    // which can be mapped
+    // which can be mapped/protected
    u32 offset = 0;	
    bool skipDTCM = num == 0 && region != memregion_DTCM;
    while (offset < mirrorSize)
    {
-        if (skipDTCM && mirrorStart + offset == NDS::ARM9->DTCMBase)
+        if (skipDTCM && mirrorStart + offset == dtcmStart)
        {
-            SetCodeProtectionRange(NDS::ARM9->DTCMBase, NDS::ARM9->DTCMSize, 0, 0);
-            offset += NDS::ARM9->DTCMSize;
+#ifdef _WIN32
+            SetCodeProtectionRange(dtcmStart, dtcmSize, 0, 0);
+#endif
+            offset += dtcmSize;
        }
        else
        {
@ -557,26 +646,25 @@ bool MapAtAddress(u32 addr)
    Mapping mapping{mirrorStart, mirrorSize, memoryOffset, num};
    Mappings[region].Add(mapping);

-    printf("mapped mirror at %08x-%08x\n", mirrorStart, mirrorStart + mirrorSize - 1);
+    //printf("mapped mirror at %08x-%08x\n", mirrorStart, mirrorStart + mirrorSize - 1);

    return true;
 }

-bool FaultHandler(FaultDescription* faultDesc, s32& offset)
+bool FaultHandler(FaultDescription& faultDesc)
 {
-    if (ARMJIT::JITCompiler->IsJITFault(faultDesc->FaultPC))
+    if (ARMJIT::JITCompiler->IsJITFault(faultDesc.FaultPC))
    {
        bool rewriteToSlowPath = true;

-        u32 addr = faultDesc->EmulatedFaultAddr;
+        u8* memStatus = NDS::CurCPU == 0 ? MappingStatus9 : MappingStatus7;

-        if ((NDS::CurCPU == 0 ? MappingStatus9 : MappingStatus7)[addr >> 12] == memstate_Unmapped)
-            rewriteToSlowPath = !MapAtAddress(faultDesc->EmulatedFaultAddr);
+        if (memStatus[faultDesc.EmulatedFaultAddr >> 12] == memstate_Unmapped)
+            rewriteToSlowPath = !MapAtAddress(faultDesc.EmulatedFaultAddr);

        if (rewriteToSlowPath)
-        {
-            offset = ARMJIT::JITCompiler->RewriteMemAccess(faultDesc->FaultPC);
-        }
+            faultDesc.FaultPC = ARMJIT::JITCompiler->RewriteMemAccess(faultDesc.FaultPC);
+
        return true;
    }
    return false;
@ -624,22 +712,52 @@ void Init()

    u8* basePtr = MemoryBase;
 #else
-    FastMem9Start = mmap(NULL, AddrSpaceSize, PROT_NONE, MAP_ANON | MAP_PRIVATE, -1, 0);
-    FastMem7Start = mmap(NULL, AddrSpaceSize, PROT_NONE, MAP_ANON | MAP_PRIVATE, -1, 0);
+    // this used to be allocated with three different mmaps
+    // The idea was to give the OS more freedom where to position the buffers, 
+    // but something was bad about this so instead we take this vmem eating monster
+    // which seems to work better.
+    MemoryBase = (u8*)mmap(NULL, AddrSpaceSize*4, PROT_NONE, MAP_ANON | MAP_PRIVATE, -1, 0);
+    munmap(MemoryBase, AddrSpaceSize*4);
+    FastMem9Start = MemoryBase;
+    FastMem7Start = MemoryBase + AddrSpaceSize;
+    MemoryBase = MemoryBase + AddrSpaceSize*2;

-    MemoryBase = (u8*)mmap(NULL, MemoryTotalSize, PROT_NONE, MAP_ANON | MAP_PRIVATE, -1, 0);
+#if defined(__ANDROID__)
+    static void* libandroid = dlopen("libandroid.so", RTLD_LAZY | RTLD_LOCAL);
+    using type_ASharedMemory_create = int(*)(const char* name, size_t size);
+    static void* symbol = dlsym(libandroid, "ASharedMemory_create");
+    static auto shared_memory_create = reinterpret_cast<type_ASharedMemory_create>(symbol);

+    if (shared_memory_create)
+    {
+        MemoryFile = shared_memory_create("melondsfastmem", MemoryTotalSize);
+    }
+    else
+    {
+        int fd = open(ASHMEM_DEVICE, O_RDWR);
+        ioctl(fd, ASHMEM_SET_NAME, "melondsfastmem");
+        ioctl(fd, ASHMEM_SET_SIZE, MemoryTotalSize);
+        MemoryFile = fd;
+    }
+#elif defined(__APPLE__)
+    char* fastmemPidName = new char[snprintf(NULL, 0, "melondsfastmem%d", getpid()) + 1];
+    sprintf(fastmemPidName, "melondsfastmem%d", getpid());
+    MemoryFile = shm_open(fastmemPidName, O_RDWR|O_CREAT, 0600);
+    delete[] fastmemPidName;
+#else
    MemoryFile = memfd_create("melondsfastmem", 0);
+#endif
    ftruncate(MemoryFile, MemoryTotalSize);

-    NewSa.sa_flags = SA_SIGINFO;
-    sigemptyset(&NewSa.sa_mask);
-    NewSa.sa_sigaction = SigsegvHandler;
-    sigaction(SIGSEGV, &NewSa, &OldSa);
-
-    munmap(MemoryBase, MemoryTotalSize);
-    munmap(FastMem9Start, AddrSpaceSize);
-    munmap(FastMem7Start, AddrSpaceSize);
+    struct sigaction sa;
+    sa.sa_handler = nullptr;
+    sa.sa_sigaction = &SigsegvHandler;
+    sa.sa_flags = SA_SIGINFO;
+    sigemptyset(&sa.sa_mask);
+    sigaction(SIGSEGV, &sa, &OldSaSegv);
+#ifdef __APPLE__
+    sigaction(SIGBUS, &sa, &OldSaBus);
+#endif

    mmap(MemoryBase, MemoryTotalSize, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, MemoryFile, 0);

@ -657,17 +775,30 @@ void Init()
 void DeInit()
 {
 #if defined(__SWITCH__)
-    virtmemFree(FastMem9Start, 0x100000000);
-    virtmemFree(FastMem7Start, 0x100000000);
+    virtmemFree(FastMem9Start, AddrSpaceSize);
+    virtmemFree(FastMem7Start, AddrSpaceSize);

    svcUnmapProcessCodeMemory(envGetOwnProcessHandle(), (u64)MemoryBaseCodeMem, (u64)MemoryBase, MemoryTotalSize);
    virtmemFree(MemoryBaseCodeMem, MemoryTotalSize);
    free(MemoryBase);
+#elif defined(__APPLE__)
+    char* fastmemPidName = new char[snprintf(NULL, 0, "melondsfastmem%d", getpid()) + 1];
+    sprintf(fastmemPidName, "melondsfastmem%d", getpid());
+    shm_unlink(fastmemPidName);
+    delete[] fastmemPidName;
 #elif defined(_WIN32)
    assert(UnmapViewOfFile(MemoryBase));
    CloseHandle(MemoryFile);

    RemoveVectoredExceptionHandler(ExceptionHandlerHandle);
+#else
+    sigaction(SIGSEGV, &OldSaSegv, nullptr);
+#ifdef __APPLE__
+    sigaction(SIGBUS, &OldSaBus, nullptr);
+#endif
+
+    munmap(MemoryBase, MemoryTotalSize);
+    close(MemoryFile);
 #endif
 }

@ -997,9 +1128,11 @@ int ClassifyAddress7(u32 addr)
        case 0x06000000:
        case 0x06800000:
            return memregion_VWRAM;
+
+        default:
+            return memregion_Other;
        }
    }
-    return memregion_Other;
 }

 void WifiWrite32(u32 addr, u32 val)
@ -1176,4 +1309,4 @@ void* GetFuncForAddr(ARM* cpu, u32 addr, bool store, int size)
    return NULL;
 }

-}
+}
--- a/src/ARMJIT_x64/ARMJIT_Branch.cpp
+++ b/src/ARMJIT_x64/ARMJIT_Branch.cpp
@ -130,6 +130,16 @@ void Compiler::Comp_JumpTo(u32 addr, bool forceNonConstantCycles)
        ADD(32, MDisp(RCPU, offsetof(ARM, Cycles)), Imm8(cycles));
 }

+void ARMv4JumpToTrampoline(ARMv4* arm, u32 addr, bool restorecpsr)
+{
+    arm->JumpTo(addr, restorecpsr);
+}
+
+void ARMv5JumpToTrampoline(ARMv5* arm, u32 addr, bool restorecpsr)
+{
+    arm->JumpTo(addr, restorecpsr);
+}
+
 void Compiler::Comp_JumpTo(Gen::X64Reg addr, bool restoreCPSR)
 {
    IrregularCycles = true;
@ -146,9 +156,9 @@ void Compiler::Comp_JumpTo(Gen::X64Reg addr, bool restoreCPSR)
    else
        MOV(32, R(ABI_PARAM3), Imm32(true)); // what a waste
    if (Num == 0)
-        CALL((void*)&ARMv5::JumpTo);
+        CALL((void*)&ARMv5JumpToTrampoline);
    else
-        CALL((void*)&ARMv4::JumpTo);
+        CALL((void*)&ARMv4JumpToTrampoline);

    PopRegs(restoreCPSR);

@ -269,4 +279,4 @@ void Compiler::T_Comp_BL_Merged()
    Comp_JumpTo(target);
 }

-}
+}
--- a/src/ARMJIT_x64/ARMJIT_Compiler.cpp
+++ b/src/ARMJIT_x64/ARMJIT_Compiler.cpp
@ -101,6 +101,11 @@ void Compiler::A_Comp_MRS()
        MOV(32, rd, R(RCPSR));
 }

+void UpdateModeTrampoline(ARM* arm, u32 oldmode, u32 newmode)
+{
+    arm->UpdateMode(oldmode, newmode);
+}
+
 void Compiler::A_Comp_MSR()
 {
    Comp_AddCycles_C();
@ -185,7 +190,7 @@ void Compiler::A_Comp_MSR()
            MOV(32, R(ABI_PARAM3), R(RCPSR));
            MOV(32, R(ABI_PARAM2), R(RSCRATCH3));
            MOV(64, R(ABI_PARAM1), R(RCPU));
-            CALL((void*)&ARM::UpdateMode);
+            CALL((void*)&UpdateModeTrampoline);

            PopRegs(true);
        }
@ -216,6 +221,8 @@ Compiler::Compiler()
    #ifdef _WIN32
        DWORD dummy;
        VirtualProtect(pageAligned, alignedSize, PAGE_EXECUTE_READWRITE, &dummy);
+    #elif defined(__APPLE__)
+        pageAligned = (u8*)mmap(NULL, 1024*1024*32, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS ,-1, 0);
    #else
        mprotect(pageAligned, alignedSize, PROT_EXEC | PROT_READ | PROT_WRITE);
    #endif
@ -340,7 +347,7 @@ Compiler::Compiler()
                    ABI_PushRegistersAndAdjustStack(CallerSavedPushRegs, 8);
                    if (consoleType == 0)
                    {
-                        switch ((8 << size) |  num)
+                        switch ((8 << size) | num)
                        {
                        case 32: ABI_CallFunction(SlowWrite9<u32, 0>); break;
                        case 33: ABI_CallFunction(SlowWrite7<u32, 0>); break;
@ -352,7 +359,7 @@ Compiler::Compiler()
                    }
                    else
                    {
-                        switch ((8 << size) |  num)
+                        switch ((8 << size) | num)
                        {
                        case 32: ABI_CallFunction(SlowWrite9<u32, 1>); break;
                        case 33: ABI_CallFunction(SlowWrite7<u32, 1>); break;
@ -375,7 +382,7 @@ Compiler::Compiler()
                        ABI_PushRegistersAndAdjustStack(CallerSavedPushRegs, 8);
                        if (consoleType == 0)
                        {
-                            switch ((8 << size) |  num)
+                            switch ((8 << size) | num)
                            {
                            case 32: ABI_CallFunction(SlowRead9<u32, 0>); break;
                            case 33: ABI_CallFunction(SlowRead7<u32, 0>); break;
@ -387,7 +394,7 @@ Compiler::Compiler()
                        }
                        else
                        {
-                            switch ((8 << size) |  num)
+                            switch ((8 << size) | num)
                            {
                            case 32: ABI_CallFunction(SlowRead9<u32, 1>); break;
                            case 33: ABI_CallFunction(SlowRead7<u32, 1>); break;
@ -612,9 +619,9 @@ void Compiler::Reset()
    LoadStorePatches.clear();
 }

-bool Compiler::IsJITFault(u64 addr)
+bool Compiler::IsJITFault(u8* addr)
 {
-    return addr >= (u64)CodeMemory && addr < (u64)CodeMemory + sizeof(CodeMemory);
+    return (u64)addr >= (u64)ResetStart && (u64)addr < (u64)ResetStart + CodeMemSize;
 }

 void Compiler::Comp_SpecialBranchBehaviour(bool taken)
@ -896,5 +903,4 @@ void Compiler::Comp_AddCycles_CD()
    else
        ConstantCycles += cycles;
 }
-
-}
+}
--- a/src/ARMJIT_x64/ARMJIT_Compiler.h
+++ b/src/ARMJIT_x64/ARMJIT_Compiler.h
@ -208,9 +208,9 @@ public:
        SetCodePtr(FarCode);
    }

-    bool IsJITFault(u64 addr);
+    bool IsJITFault(u8* addr);

-    s32 RewriteMemAccess(u64 pc);
+    u8* RewriteMemAccess(u8* pc);

    u8* FarCode;
    u8* NearCode;
--- a/src/ARMJIT_x64/ARMJIT_Linkage.S
+++ b/src/ARMJIT_x64/ARMJIT_Linkage.S
@ -29,8 +29,13 @@

 .p2align 4,,15

+#ifdef __APPLE__
+.global _ARM_Dispatch
+_ARM_Dispatch:
+#else
 .global ARM_Dispatch
 ARM_Dispatch:
+#endif
 #ifdef WIN64
    push rdi
    push rsi
@ -54,8 +59,13 @@ ARM_Dispatch:

 .p2align 4,,15

+#ifdef __APPLE__
+.global _ARM_Ret
+_ARM_Ret:
+#else
 .global ARM_Ret
 ARM_Ret:
+#endif
    mov [RCPU + ARM_CPSR_offset], RCPSR

 #ifdef WIN64
--- a/src/ARMJIT_x64/ARMJIT_LoadStore.cpp
+++ b/src/ARMJIT_x64/ARMJIT_LoadStore.cpp
@ -15,28 +15,24 @@ int squeezePointer(T* ptr)
    return truncated;
 }

-s32 Compiler::RewriteMemAccess(u64 pc)
+u8* Compiler::RewriteMemAccess(u8* pc)
 {
-    auto it = LoadStorePatches.find((u8*)pc);
+    auto it = LoadStorePatches.find(pc);
    if (it != LoadStorePatches.end())
    {
        LoadStorePatch patch = it->second;
        LoadStorePatches.erase(it);

-        u8* curCodePtr = GetWritableCodePtr();
-        u8* rewritePtr = (u8*)pc + (ptrdiff_t)patch.Offset;
-        SetCodePtr(rewritePtr);
+        //printf("rewriting memory access %p %d %d\n", (u8*)pc-ResetStart, patch.Offset, patch.Size);

-        CALL(patch.PatchFunc);
-        u32 remainingSize = patch.Size - (GetWritableCodePtr() - rewritePtr);
+        XEmitter emitter(pc + (ptrdiff_t)patch.Offset);
+        emitter.CALL(patch.PatchFunc);
+        ptrdiff_t remainingSize = (ptrdiff_t)patch.Size - 5;
+        assert(remainingSize >= 0);
        if (remainingSize > 0)
-            NOP(remainingSize);
+            emitter.NOP(remainingSize);

-        //printf("rewriting memory access %p %d %d\n", patch.PatchFunc, patch.Offset, patch.Size);
-
-        SetCodePtr(curCodePtr);
-
-        return patch.Offset;
+        return pc + (ptrdiff_t)patch.Offset;
    }

    printf("this is a JIT bug %llx\n", pc);
@ -192,6 +188,7 @@ void Compiler::Comp_MemAccess(int rd, int rn, const Op2& op2, int size, int flag
        u8* memopStart = GetWritableCodePtr();
        LoadStorePatch patch;

+        assert(rdMapped.GetSimpleReg() >= 0 && rdMapped.GetSimpleReg() < 16);
        patch.PatchFunc = flags & memop_Store
            ? PatchedStoreFuncs[NDS::ConsoleType][Num][__builtin_ctz(size) - 3][rdMapped.GetSimpleReg()]
            : PatchedLoadFuncs[NDS::ConsoleType][Num][__builtin_ctz(size) - 3][!!(flags & memop_SignExtend)][rdMapped.GetSimpleReg()];
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@ -80,9 +80,8 @@ if (ENABLE_JIT)
 			ARMJIT_x64/ARMJIT_LoadStore.cpp
 			ARMJIT_x64/ARMJIT_Branch.cpp

-			ARMJIT_x64/ARMJIT_Linkage.s
+			ARMJIT_x64/ARMJIT_Linkage.S
 		)
-		set_source_files_properties(ARMJIT_x64/ARMJIT_Linkage.s PROPERTIES COMPILE_FLAGS "-x assembler-with-cpp")
 	endif()
 	if (ARCHITECTURE STREQUAL ARM64)
 		target_sources(core PRIVATE
@ -94,16 +93,22 @@ if (ENABLE_JIT)
 			ARMJIT_A64/ARMJIT_LoadStore.cpp
 			ARMJIT_A64/ARMJIT_Branch.cpp

-			ARMJIT_A64/ARMJIT_Linkage.s
+			ARMJIT_A64/ARMJIT_Linkage.S
 		)
-		set_source_files_properties(ARMJIT_A64/ARMJIT_Linkage.s PROPERTIES COMPILE_FLAGS "-x assembler-with-cpp")
 	endif()
 endif()

+if (APPLE)
+        target_include_directories(core PUBLIC /usr/local/include)
+        target_link_directories(core PUBLIC /usr/local/lib)
+endif()
+
 if (ENABLE_OGLRENDERER)
 	if (WIN32)
 		target_link_libraries(core ole32 comctl32 ws2_32 opengl32)
-	else()
+  elseif (APPLE)
+    target_link_libraries(core "-framework OpenGL")
+  else()
 		target_link_libraries(core GL EGL)
 	endif()
 else()
--- a/src/Config.cpp
+++ b/src/Config.cpp
@ -73,7 +73,11 @@ ConfigEntry ConfigFile[] =
    {"JIT_MaxBlockSize", 0, &JIT_MaxBlockSize, 32, NULL, 0},
    {"JIT_BranchOptimisations", 0, &JIT_BranchOptimisations, 1, NULL, 0},
    {"JIT_LiteralOptimisations", 0, &JIT_LiteralOptimisations, 1, NULL, 0},
-    {"JIT_FastMemory", 0, &JIT_FastMemory, 1, NULL, 0},
+    #ifdef __APPLE__
+        {"JIT_FastMemory", 0, &JIT_FastMemory, 0, NULL, 0},
+    #else
+        {"JIT_FastMemory", 0, &JIT_FastMemory, 1, NULL, 0},
+    #endif
 #endif

    {"", -1, NULL, 0, NULL, 0}
--- a/src/DMA.cpp
+++ b/src/DMA.cpp
@ -77,21 +77,6 @@ void DMA::Reset()

    Running = false;
    InProgress = false;
-
-    if (NDS::ConsoleType == 1)
-    {
-        BusRead16 = (CPU==0) ? DSi::ARM9Read16 : DSi::ARM7Read16;
-        BusRead32 = (CPU==0) ? DSi::ARM9Read32 : DSi::ARM7Read32;
-        BusWrite16 = (CPU==0) ? DSi::ARM9Write16 : DSi::ARM7Write16;
-        BusWrite32 = (CPU==0) ? DSi::ARM9Write32 : DSi::ARM7Write32;
-    }
-    else
-    {
-        BusRead16 = (CPU==0) ? NDS::ARM9Read16 : NDS::ARM7Read16;
-        BusRead32 = (CPU==0) ? NDS::ARM9Read32 : NDS::ARM7Read32;
-        BusWrite16 = (CPU==0) ? NDS::ARM9Write16 : NDS::ARM7Write16;
-        BusWrite32 = (CPU==0) ? NDS::ARM9Write32 : NDS::ARM7Write32;
-    }
 }

 void DMA::DoSavestate(Savestate* file)
@ -198,13 +183,7 @@ void DMA::Start()
    NDS::StopCPU(CPU, 1<<Num);
 }

-void DMA::Run()
-{
-    if (!Running) return;
-    if (CPU == 0) return Run9();
-    else          return Run7();
-}
-
+template <int ConsoleType>
 void DMA::Run9()
 {
    if (NDS::ARM9Timestamp >= NDS::ARM9Target) return;
@ -242,7 +221,10 @@ void DMA::Run9()
        {
            NDS::ARM9Timestamp += (unitcycles << NDS::ARM9ClockShift);

-            BusWrite16(CurDstAddr, BusRead16(CurSrcAddr));
+            if (ConsoleType == 1)
+                DSi::ARM9Write16(CurDstAddr, DSi::ARM9Read16(CurSrcAddr));
+            else
+                NDS::ARM9Write16(CurDstAddr, NDS::ARM9Read16(CurSrcAddr));

            CurSrcAddr += SrcAddrInc<<1;
            CurDstAddr += DstAddrInc<<1;
@ -278,7 +260,10 @@ void DMA::Run9()
        {
            NDS::ARM9Timestamp += (unitcycles << NDS::ARM9ClockShift);

-            BusWrite32(CurDstAddr, BusRead32(CurSrcAddr));
+            if (ConsoleType == 1)
+                DSi::ARM9Write32(CurDstAddr, DSi::ARM9Read32(CurSrcAddr));
+            else
+                NDS::ARM9Write32(CurDstAddr, NDS::ARM9Read32(CurSrcAddr));

            CurSrcAddr += SrcAddrInc<<2;
            CurDstAddr += DstAddrInc<<2;
@ -317,6 +302,7 @@ void DMA::Run9()
    NDS::ResumeCPU(0, 1<<Num);
 }

+template <int ConsoleType>
 void DMA::Run7()
 {
    if (NDS::ARM7Timestamp >= NDS::ARM7Target) return;
@ -354,7 +340,10 @@ void DMA::Run7()
        {
            NDS::ARM7Timestamp += unitcycles;

-            BusWrite16(CurDstAddr, BusRead16(CurSrcAddr));
+            if (ConsoleType == 1)
+                DSi::ARM7Write16(CurDstAddr, DSi::ARM7Read16(CurSrcAddr));
+            else
+                NDS::ARM7Write16(CurDstAddr, NDS::ARM7Read16(CurSrcAddr));

            CurSrcAddr += SrcAddrInc<<1;
            CurDstAddr += DstAddrInc<<1;
@ -390,7 +379,10 @@ void DMA::Run7()
        {
            NDS::ARM7Timestamp += unitcycles;

-            BusWrite32(CurDstAddr, BusRead32(CurSrcAddr));
+            if (ConsoleType == 1)
+                DSi::ARM7Write32(CurDstAddr, DSi::ARM7Read32(CurSrcAddr));
+            else
+                NDS::ARM7Write32(CurDstAddr, NDS::ARM7Read32(CurSrcAddr));

            CurSrcAddr += SrcAddrInc<<2;
            CurDstAddr += DstAddrInc<<2;
@ -425,3 +417,14 @@ void DMA::Run7()
    InProgress = false;
    NDS::ResumeCPU(1, 1<<Num);
 }
+
+template <int ConsoleType>
+void DMA::Run()
+{
+    if (!Running) return;
+    if (CPU == 0) return Run9<ConsoleType>();
+    else          return Run7<ConsoleType>();
+}
+
+template void DMA::Run<0>();
+template void DMA::Run<1>();
--- a/src/DMA.h
+++ b/src/DMA.h
@ -34,9 +34,12 @@ public:
    void WriteCnt(u32 val);
    void Start();

+    template <int ConsoleType>
    void Run();

+    template <int ConsoleType>
    void Run9();
+    template <int ConsoleType>
    void Run7();

    bool IsInMode(u32 mode)
@ -86,11 +89,6 @@ private:
    bool Stall;

    bool IsGXFIFODMA;
-
-    u16 (*BusRead16)(u32 addr);
-    u32 (*BusRead32)(u32 addr);
-    void (*BusWrite16)(u32 addr, u16 val);
-    void (*BusWrite32)(u32 addr, u32 val);
 };

 #endif
--- a/src/DSi.cpp
+++ b/src/DSi.cpp
@ -543,15 +543,15 @@ void MapNWRAM_A(u32 num, u8 val)
        return;
    }

-#ifdef JIT_ENABLED
-    ARMJIT_Memory::RemapNWRAM(0);
-#endif
-
    int mbkn = 0, mbks = 8*num;

    u8 oldval = (MBK[0][mbkn] >> mbks) & 0xFF;
    if (oldval == val) return;

+#ifdef JIT_ENABLED
+    ARMJIT_Memory::RemapNWRAM(0);
+#endif
+
    MBK[0][mbkn] &= ~(0xFF << mbks);
    MBK[0][mbkn] |= (val << mbks);
    MBK[1][mbkn] = MBK[0][mbkn];
@ -578,15 +578,15 @@ void MapNWRAM_B(u32 num, u8 val)
        return;
    }

-#ifdef JIT_ENABLED
-    ARMJIT_Memory::RemapNWRAM(1);
-#endif
-
    int mbkn = 1+(num>>2), mbks = 8*(num&3);

    u8 oldval = (MBK[0][mbkn] >> mbks) & 0xFF;
    if (oldval == val) return;

+#ifdef JIT_ENABLED
+    ARMJIT_Memory::RemapNWRAM(1);
+#endif
+
    MBK[0][mbkn] &= ~(0xFF << mbks);
    MBK[0][mbkn] |= (val << mbks);
    MBK[1][mbkn] = MBK[0][mbkn];
@ -617,15 +617,15 @@ void MapNWRAM_C(u32 num, u8 val)
        return;
    }

-#ifdef JIT_ENABLED
-    ARMJIT_Memory::RemapNWRAM(2);
-#endif
-
    int mbkn = 3+(num>>2), mbks = 8*(num&3);

    u8 oldval = (MBK[0][mbkn] >> mbks) & 0xFF;
    if (oldval == val) return;

+#ifdef JIT_ENABLED
+    ARMJIT_Memory::RemapNWRAM(2);
+#endif
+
    MBK[0][mbkn] &= ~(0xFF << mbks);
    MBK[0][mbkn] |= (val << mbks);
    MBK[1][mbkn] = MBK[0][mbkn];
--- a/src/GPU.cpp
+++ b/src/GPU.cpp
@ -49,8 +49,8 @@ u8 VRAM_F[ 16*1024];
 u8 VRAM_G[ 16*1024];
 u8 VRAM_H[ 32*1024];
 u8 VRAM_I[ 16*1024];
-u8* VRAM[9]     = {VRAM_A,  VRAM_B,  VRAM_C,  VRAM_D,  VRAM_E, VRAM_F, VRAM_G, VRAM_H, VRAM_I};
-u32 VRAMMask[9] = {0x1FFFF, 0x1FFFF, 0x1FFFF, 0x1FFFF, 0xFFFF, 0x3FFF, 0x3FFF, 0x7FFF, 0x3FFF};
+u8* const VRAM[9]     = {VRAM_A,  VRAM_B,  VRAM_C,  VRAM_D,  VRAM_E, VRAM_F, VRAM_G, VRAM_H, VRAM_I};
+u32 const VRAMMask[9] = {0x1FFFF, 0x1FFFF, 0x1FFFF, 0x1FFFF, 0xFFFF, 0x3FFF, 0x3FFF, 0x7FFF, 0x3FFF};

 u8 VRAMCNT[9];
 u8 VRAMSTAT;
@ -85,6 +85,62 @@ bool Accelerated;
 GPU2D* GPU2D_A;
 GPU2D* GPU2D_B;

+/*
+    VRAM invalidation tracking
+
+    - we want to know when a VRAM region used for graphics changed
+    - for some regions unmapping is mandatory to modify them (Texture, TexPal and ExtPal) and
+        we don't want to completely invalidate them every time they're unmapped and remapped
+
+    For this reason we don't track the dirtyness per mapping region, but instead per VRAM bank
+    with VRAMDirty. Writes to LCDC go directly into VRAMDirty, while writes via other mapping regions
+    like BG or OBJ are first tracked in VRAMWritten_* and need to be flushed using SyncDirtyFlags.
+
+    This is more or less a description of VRAMTrackingSet::DeriveState
+        Each time before the memory is read two things could have happened
+        to each 16kb piece (16kb is the smallest unit in which mappings can
+        be made thus also the size VRAMMap_* use):
+            - this piece was remapped compared to last time we checked,
+                which means this location in memory is invalid.
+            - this piece wasn't remapped, which means we need to check whether
+                it was changed. This can be archived by checking VRAMDirty.
+                VRAMDirty need to be reset for the respective VRAM bank.
+*/
+
+VRAMTrackingSet<512*1024, 16*1024> VRAMDirty_ABG;
+VRAMTrackingSet<256*1024, 16*1024> VRAMDirty_AOBJ;
+VRAMTrackingSet<128*1024, 16*1024> VRAMDirty_BBG;
+VRAMTrackingSet<128*1024, 16*1024> VRAMDirty_BOBJ;
+
+VRAMTrackingSet<32*1024, 8*1024> VRAMDirty_ABGExtPal;
+VRAMTrackingSet<32*1024, 8*1024> VRAMDirty_BBGExtPal;
+VRAMTrackingSet<8*1024, 8*1024> VRAMDirty_AOBJExtPal;
+VRAMTrackingSet<8*1024, 8*1024> VRAMDirty_BOBJExtPal;
+
+VRAMTrackingSet<512*1024, 128*1024> VRAMDirty_Texture;
+VRAMTrackingSet<128*1024, 16*1024> VRAMDirty_TexPal;
+
+
+NonStupidBitField<512*1024/VRAMDirtyGranularity> VRAMWritten_ABG;
+NonStupidBitField<256*1024/VRAMDirtyGranularity> VRAMWritten_AOBJ;
+NonStupidBitField<128*1024/VRAMDirtyGranularity> VRAMWritten_BBG;
+NonStupidBitField<128*1024/VRAMDirtyGranularity> VRAMWritten_BOBJ;
+NonStupidBitField<256*1024/VRAMDirtyGranularity> VRAMWritten_ARM7;
+
+NonStupidBitField<128*1024/VRAMDirtyGranularity> VRAMDirty[9];
+
+u8 VRAMFlat_ABG[512*1024];
+u8 VRAMFlat_BBG[128*1024];
+u8 VRAMFlat_AOBJ[256*1024];
+u8 VRAMFlat_BOBJ[128*1024];
+
+u8 VRAMFlat_ABGExtPal[32*1024];
+u8 VRAMFlat_BBGExtPal[32*1024];
+u8 VRAMFlat_AOBJExtPal[8*1024];
+u8 VRAMFlat_BOBJExtPal[8*1024];
+
+u8 VRAMFlat_Texture[512*1024];
+u8 VRAMFlat_TexPal[128*1024];

 bool Init()
 {
@ -113,6 +169,30 @@ void DeInit()
    if (Framebuffer[1][1]) delete[] Framebuffer[1][1];
 }

+void ResetVRAMCache()
+{
+    for (int i = 0; i < 9; i++)
+        VRAMDirty[i] = NonStupidBitField<128*1024/VRAMDirtyGranularity>();
+
+    VRAMDirty_ABG.Reset();
+    VRAMDirty_BBG.Reset();
+    VRAMDirty_AOBJ.Reset();
+    VRAMDirty_BOBJ.Reset();
+    VRAMDirty_ABGExtPal.Reset();
+    VRAMDirty_BBGExtPal.Reset();
+    VRAMDirty_AOBJExtPal.Reset();
+    VRAMDirty_BOBJExtPal.Reset();
+
+    memset(VRAMFlat_ABG, 0, sizeof(VRAMFlat_ABG));
+    memset(VRAMFlat_BBG, 0, sizeof(VRAMFlat_BBG));
+    memset(VRAMFlat_AOBJ, 0, sizeof(VRAMFlat_AOBJ));
+    memset(VRAMFlat_BOBJ, 0, sizeof(VRAMFlat_BOBJ));
+    memset(VRAMFlat_ABGExtPal, 0, sizeof(VRAMFlat_ABGExtPal));
+    memset(VRAMFlat_BBGExtPal, 0, sizeof(VRAMFlat_BBGExtPal));
+    memset(VRAMFlat_AOBJExtPal, 0, sizeof(VRAMFlat_AOBJExtPal));
+    memset(VRAMFlat_BOBJExtPal, 0, sizeof(VRAMFlat_BOBJExtPal));
+}
+
 void Reset()
 {
    VCount = 0;
@ -186,6 +266,8 @@ void Reset()
    GPU2D_B->SetFramebuffer(Framebuffer[backbuf][0]);

    ResetRenderer();
+
+    ResetVRAMCache();
 }

 void Stop()
@ -261,6 +343,8 @@ void DoSavestate(Savestate* file)
    GPU2D_A->DoSavestate(file);
    GPU2D_B->DoSavestate(file);
    GPU3D::DoSavestate(file);
+
+    ResetVRAMCache();
 }

 void AssignFramebuffers()
@ -411,18 +495,8 @@ void SetRenderSettings(int renderer, RenderSettings& settings)

 u8* GetUniqueBankPtr(u32 mask, u32 offset)
 {
-    if (!mask) return NULL;
-
-    int num = 0;
-    if (!(mask & 0xFF)) { mask >>= 8; num += 8; }
-    else
-    {
-        if (!(mask & 0xF)) { mask >>= 4; num += 4; }
-        if (!(mask & 0x3)) { mask >>= 2; num += 2; }
-        if (!(mask & 0x1)) { mask >>= 1; num += 1; }
-    }
-    if (mask != 1) return NULL;
-
+    if (!mask || (mask & (mask - 1)) != 0) return NULL;
+    int num = __builtin_ctz(mask);
    return &VRAM[num][offset & VRAMMask[num]];
 }

@ -606,8 +680,6 @@ void MapVRAM_E(u32 bank, u8 cnt)

        case 4: // ABG ext palette
            UNMAP_RANGE(ABGExtPal, 0, 4);
-            GPU2D_A->BGExtPalDirty(0);
-            GPU2D_A->BGExtPalDirty(2);
            break;
        }
    }
@ -634,8 +706,6 @@ void MapVRAM_E(u32 bank, u8 cnt)

        case 4: // ABG ext palette
            MAP_RANGE(ABGExtPal, 0, 4);
-            GPU2D_A->BGExtPalDirty(0);
-            GPU2D_A->BGExtPalDirty(2);
            break;
        }
    }
@ -687,12 +757,10 @@ void MapVRAM_FG(u32 bank, u8 cnt)
        case 4: // ABG ext palette
            VRAMMap_ABGExtPal[((oldofs & 0x1) << 1)] &= ~bankmask;
            VRAMMap_ABGExtPal[((oldofs & 0x1) << 1) + 1] &= ~bankmask;
-            GPU2D_A->BGExtPalDirty((oldofs & 0x1) << 1);
            break;

        case 5: // AOBJ ext palette
            VRAMMap_AOBJExtPal &= ~bankmask;
-            GPU2D_A->OBJExtPalDirty();
            break;
        }
    }
@ -732,12 +800,10 @@ void MapVRAM_FG(u32 bank, u8 cnt)
        case 4: // ABG ext palette
            VRAMMap_ABGExtPal[((ofs & 0x1) << 1)] |= bankmask;
            VRAMMap_ABGExtPal[((ofs & 0x1) << 1) + 1] |= bankmask;
-            GPU2D_A->BGExtPalDirty((ofs & 0x1) << 1);
            break;

        case 5: // AOBJ ext palette
            VRAMMap_AOBJExtPal |= bankmask;
-            GPU2D_A->OBJExtPalDirty();
            break;
        }
    }
@ -773,8 +839,6 @@ void MapVRAM_H(u32 bank, u8 cnt)

        case 2: // BBG ext palette
            UNMAP_RANGE(BBGExtPal, 0, 4);
-            GPU2D_B->BGExtPalDirty(0);
-            GPU2D_B->BGExtPalDirty(2);
            break;
        }
    }
@ -800,8 +864,6 @@ void MapVRAM_H(u32 bank, u8 cnt)

        case 2: // BBG ext palette
            MAP_RANGE(BBGExtPal, 0, 4);
-            GPU2D_B->BGExtPalDirty(0);
-            GPU2D_B->BGExtPalDirty(2);
            break;
        }
    }
@ -841,7 +903,6 @@ void MapVRAM_I(u32 bank, u8 cnt)

        case 3: // BOBJ ext palette
            VRAMMap_BOBJExtPal &= ~bankmask;
-            GPU2D_B->OBJExtPalDirty();
            break;
        }
    }
@ -871,7 +932,6 @@ void MapVRAM_I(u32 bank, u8 cnt)

        case 3: // BOBJ ext palette
            VRAMMap_BOBJExtPal |= bankmask;
-            GPU2D_B->OBJExtPalDirty();
            break;
        }
    }
@ -937,6 +997,8 @@ void StartHBlank(u32 line)
    DispStat[0] |= (1<<1);
    DispStat[1] |= (1<<1);

+    SyncDirtyFlags();
+
    if (VCount < 192)
    {
        // draw
@ -1096,4 +1158,224 @@ void SetVCount(u16 val)
    NextVCount = val;
 }

+template <u32 Size, u32 MappingGranularity>
+NonStupidBitField<Size/VRAMDirtyGranularity> VRAMTrackingSet<Size, MappingGranularity>::DeriveState(u32* currentMappings)
+{
+    NonStupidBitField<Size/VRAMDirtyGranularity> result;
+    u16 banksToBeZeroed = 0;
+    for (u32 i = 0; i < Size / MappingGranularity; i++)
+    {
+        if (currentMappings[i] != Mapping[i])
+        {
+            result |= NonStupidBitField<Size/VRAMDirtyGranularity>(i*VRAMBitsPerMapping, VRAMBitsPerMapping);
+            banksToBeZeroed |= currentMappings[i];
+            Mapping[i] = currentMappings[i];
+        }
+        else
+        {
+            u32 mapping = Mapping[i];
+
+            banksToBeZeroed |= mapping;
+
+            while (mapping != 0)
+            {
+                u32 num = __builtin_ctz(mapping);
+                mapping &= ~(1 << num);
+
+                // hack for **speed**
+                // this could probably be done less ugly but then we would rely
+                // on the compiler for vectorisation
+                static_assert(VRAMDirtyGranularity == 512);
+                if (MappingGranularity == 16*1024)
+                {
+                    u32 dirty = ((u32*)VRAMDirty[num].Data)[i & (VRAMMask[num] >> 14)];
+                    ((u32*)result.Data)[i] |= dirty;
+                }
+                else if (MappingGranularity == 8*1024)
+                {
+                    u16 dirty = ((u16*)VRAMDirty[num].Data)[i & (VRAMMask[num] >> 13)];
+                    ((u16*)result.Data)[i] |= dirty;
+                }
+                else if (MappingGranularity == 128*1024)
+                {
+                    ((u64*)result.Data)[i * 4 + 0] |= ((u64*)VRAMDirty[num].Data)[0];
+                    ((u64*)result.Data)[i * 4 + 1] |= ((u64*)VRAMDirty[num].Data)[1];
+                    ((u64*)result.Data)[i * 4 + 2] |= ((u64*)VRAMDirty[num].Data)[2];
+                    ((u64*)result.Data)[i * 4 + 3] |= ((u64*)VRAMDirty[num].Data)[3];
+                }
+                else
+                {
+                    // welp
+                    abort();
+                }
+            }
+        }
+    }
+
+    while (banksToBeZeroed != 0)
+    {
+        u32 num = __builtin_ctz(banksToBeZeroed);
+        banksToBeZeroed &= ~(1 << num);
+        memset(VRAMDirty[num].Data, 0, sizeof(VRAMDirty[num].Data));
+    }
+
+    return result;
 }
+
+template NonStupidBitField<32*1024/VRAMDirtyGranularity> VRAMTrackingSet<32*1024, 8*1024>::DeriveState(u32*);
+template NonStupidBitField<8*1024/VRAMDirtyGranularity> VRAMTrackingSet<8*1024, 8*1024>::DeriveState(u32*);
+template NonStupidBitField<512*1024/VRAMDirtyGranularity> VRAMTrackingSet<512*1024, 128*1024>::DeriveState(u32*);
+template NonStupidBitField<128*1024/VRAMDirtyGranularity> VRAMTrackingSet<128*1024, 16*1024>::DeriveState(u32*);
+template NonStupidBitField<256*1024/VRAMDirtyGranularity> VRAMTrackingSet<256*1024, 16*1024>::DeriveState(u32*);
+template NonStupidBitField<512*1024/VRAMDirtyGranularity> VRAMTrackingSet<512*1024, 16*1024>::DeriveState(u32*);
+
+template <u32 Size>
+void SyncDirtyFlags(u32* mappings, NonStupidBitField<Size>& writtenFlags)
+{
+    const u32 VRAMWrittenBitsPer16KB = 16*1024/VRAMDirtyGranularity;
+
+    for (typename NonStupidBitField<Size>::Iterator it = writtenFlags.Begin(); it != writtenFlags.End(); it++)
+    {
+        u32 mapping = mappings[*it / VRAMWrittenBitsPer16KB];
+        while (mapping != 0)
+        {
+            u32 num = __builtin_ctz(mapping);
+
+            VRAMDirty[num][*it & (VRAMMask[num] / VRAMDirtyGranularity)] = true;
+
+            mapping &= ~(1 << num);
+        }
+    }
+    memset(writtenFlags.Data, 0, sizeof(writtenFlags.Data));
+}
+
+void SyncDirtyFlags()
+{
+    SyncDirtyFlags(VRAMMap_ABG, VRAMWritten_ABG);
+    SyncDirtyFlags(VRAMMap_AOBJ, VRAMWritten_AOBJ);
+    SyncDirtyFlags(VRAMMap_BBG, VRAMWritten_BBG);
+    SyncDirtyFlags(VRAMMap_BOBJ, VRAMWritten_BOBJ);
+    SyncDirtyFlags(VRAMMap_ARM7, VRAMWritten_ARM7);
+}
+
+template <u32 MappingGranularity, u32 Size>
+inline bool CopyLinearVRAM(u8* flat, u32* mappings, NonStupidBitField<Size>& dirty, u64 (*slowAccess)(u32 addr))
+{
+    const u32 VRAMBitsPerMapping = MappingGranularity / VRAMDirtyGranularity;
+
+    bool change = false;
+
+    typename NonStupidBitField<Size>::Iterator it = dirty.Begin();
+    while (it != dirty.End())
+    {
+        u32 offset = *it * VRAMDirtyGranularity;
+        u8* dst = flat + offset;
+        u8* fastAccess = GetUniqueBankPtr(mappings[*it / VRAMBitsPerMapping], offset);
+        if (fastAccess)
+        {
+            memcpy(dst, fastAccess, VRAMDirtyGranularity);
+        }
+        else
+        {
+            for (u32 i = 0; i < VRAMDirtyGranularity; i += 8)
+                *(u64*)&dst[i] = slowAccess(offset + i);
+        }
+        change = true;
+        it++;
+    }
+    return change;
+}
+
+bool MakeVRAMFlat_TextureCoherent(NonStupidBitField<512*1024/VRAMDirtyGranularity>& dirty)
+{
+    return CopyLinearVRAM<128*1024>(VRAMFlat_Texture, VRAMMap_Texture, dirty, ReadVRAM_Texture<u64>);
+}
+bool MakeVRAMFlat_TexPalCoherent(NonStupidBitField<128*1024/VRAMDirtyGranularity>& dirty)
+{
+    return CopyLinearVRAM<16*1024>(VRAMFlat_TexPal, VRAMMap_TexPal, dirty, ReadVRAM_TexPal<u64>);
+}
+
+bool MakeVRAMFlat_ABGCoherent(NonStupidBitField<512*1024/VRAMDirtyGranularity>& dirty)
+{
+    return CopyLinearVRAM<16*1024>(VRAMFlat_ABG, VRAMMap_ABG, dirty, ReadVRAM_ABG<u64>);
+}
+bool MakeVRAMFlat_BBGCoherent(NonStupidBitField<128*1024/VRAMDirtyGranularity>& dirty)
+{
+    return CopyLinearVRAM<16*1024>(VRAMFlat_BBG, VRAMMap_BBG, dirty, ReadVRAM_BBG<u64>);
+}
+
+bool MakeVRAMFlat_AOBJCoherent(NonStupidBitField<256*1024/VRAMDirtyGranularity>& dirty)
+{
+    return CopyLinearVRAM<16*1024>(VRAMFlat_AOBJ, VRAMMap_AOBJ, dirty, ReadVRAM_AOBJ<u64>);
+}
+bool MakeVRAMFlat_BOBJCoherent(NonStupidBitField<128*1024/VRAMDirtyGranularity>& dirty)
+{
+    return CopyLinearVRAM<16*1024>(VRAMFlat_BOBJ, VRAMMap_BOBJ, dirty, ReadVRAM_BOBJ<u64>);
+}
+
+template<typename T>
+T ReadVRAM_ABGExtPal(u32 addr)
+{
+    u32 mask = VRAMMap_ABGExtPal[(addr >> 13) & 0x3];
+
+    T ret = 0;
+    if (mask & (1<<4)) ret |= *(T*)&VRAM_E[addr & 0x7FFF];
+    if (mask & (1<<5)) ret |= *(T*)&VRAM_F[addr & 0x3FFF];
+    if (mask & (1<<6)) ret |= *(T*)&VRAM_G[addr & 0x3FFF];
+
+    return ret;
+}
+
+template<typename T>
+T ReadVRAM_BBGExtPal(u32 addr)
+{
+    u32 mask = VRAMMap_BBGExtPal[(addr >> 13) & 0x3];
+
+    T ret = 0;
+    if (mask & (1<<7)) ret |= *(T*)&VRAM_H[addr & 0x7FFF];
+
+    return ret;
+}
+
+template<typename T>
+T ReadVRAM_AOBJExtPal(u32 addr)
+{
+    u32 mask = VRAMMap_AOBJExtPal;
+
+    T ret = 0;
+    if (mask & (1<<4)) ret |= *(T*)&VRAM_F[addr & 0x1FFF];
+    if (mask & (1<<5)) ret |= *(T*)&VRAM_G[addr & 0x1FFF];
+
+    return ret;
+}
+
+template<typename T>
+T ReadVRAM_BOBJExtPal(u32 addr)
+{
+    u32 mask = VRAMMap_BOBJExtPal;
+
+    T ret = 0;
+    if (mask & (1<<8)) ret |= *(T*)&VRAM_I[addr & 0x1FFF];
+
+    return ret;
+}
+
+bool MakeVRAMFlat_ABGExtPalCoherent(NonStupidBitField<32*1024/VRAMDirtyGranularity>& dirty)
+{
+    return CopyLinearVRAM<8*1024>(VRAMFlat_ABGExtPal, VRAMMap_ABGExtPal, dirty, ReadVRAM_ABGExtPal<u64>);
+}
+bool MakeVRAMFlat_BBGExtPalCoherent(NonStupidBitField<32*1024/VRAMDirtyGranularity>& dirty)
+{
+    return CopyLinearVRAM<8*1024>(VRAMFlat_BBGExtPal, VRAMMap_BBGExtPal, dirty, ReadVRAM_BBGExtPal<u64>);
+}
+
+bool MakeVRAMFlat_AOBJExtPalCoherent(NonStupidBitField<8*1024/VRAMDirtyGranularity>& dirty)
+{
+    return CopyLinearVRAM<8*1024>(VRAMFlat_AOBJExtPal, &VRAMMap_AOBJExtPal, dirty, ReadVRAM_AOBJExtPal<u64>);
+}
+bool MakeVRAMFlat_BOBJExtPalCoherent(NonStupidBitField<8*1024/VRAMDirtyGranularity>& dirty)
+{
+    return CopyLinearVRAM<8*1024>(VRAMFlat_BOBJExtPal, &VRAMMap_BOBJExtPal, dirty, ReadVRAM_BOBJExtPal<u64>);
+}
+
+}
--- a/src/GPU.h
+++ b/src/GPU.h
@ -20,6 +20,7 @@
 #define GPU_H

 #include "GPU2D.h"
+#include "NonStupidBitfield.h"

 namespace GPU
 {
@ -45,7 +46,7 @@ extern u8 VRAM_G[ 16*1024];
 extern u8 VRAM_H[ 32*1024];
 extern u8 VRAM_I[ 16*1024];

-extern u8* VRAM[9];
+extern u8* const VRAM[9];

 extern u32 VRAMMap_LCDC;
 extern u32 VRAMMap_ABG[0x20];
@ -73,6 +74,73 @@ extern GPU2D* GPU2D_B;

 extern int Renderer;

+const u32 VRAMDirtyGranularity = 512;
+
+extern NonStupidBitField<512*1024/VRAMDirtyGranularity> VRAMWritten_ABG;
+extern NonStupidBitField<256*1024/VRAMDirtyGranularity> VRAMWritten_AOBJ;
+extern NonStupidBitField<128*1024/VRAMDirtyGranularity> VRAMWritten_BBG;
+extern NonStupidBitField<128*1024/VRAMDirtyGranularity> VRAMWritten_BOBJ;
+extern NonStupidBitField<256*1024/VRAMDirtyGranularity> VRAMWritten_ARM7;
+
+extern NonStupidBitField<128*1024/VRAMDirtyGranularity> VRAMDirty[9];
+
+template <u32 Size, u32 MappingGranularity>
+struct VRAMTrackingSet
+{
+    u16 Mapping[Size / MappingGranularity];
+
+    const u32 VRAMBitsPerMapping = MappingGranularity / VRAMDirtyGranularity;
+
+    void Reset()
+    {
+        memset(Mapping, 0, sizeof(Mapping));
+    }
+    NonStupidBitField<Size/VRAMDirtyGranularity> DeriveState(u32* currentMappings);
+};
+
+extern VRAMTrackingSet<512*1024, 16*1024> VRAMDirty_ABG;
+extern VRAMTrackingSet<256*1024, 16*1024> VRAMDirty_AOBJ;
+extern VRAMTrackingSet<128*1024, 16*1024> VRAMDirty_BBG;
+extern VRAMTrackingSet<128*1024, 16*1024> VRAMDirty_BOBJ;
+
+extern VRAMTrackingSet<32*1024, 8*1024> VRAMDirty_ABGExtPal;
+extern VRAMTrackingSet<32*1024, 8*1024> VRAMDirty_BBGExtPal;
+extern VRAMTrackingSet<8*1024, 8*1024> VRAMDirty_AOBJExtPal;
+extern VRAMTrackingSet<8*1024, 8*1024> VRAMDirty_BOBJExtPal;
+
+extern VRAMTrackingSet<512*1024, 128*1024> VRAMDirty_Texture;
+extern VRAMTrackingSet<128*1024, 16*1024> VRAMDirty_TexPal;
+
+extern u8 VRAMFlat_ABG[512*1024];
+extern u8 VRAMFlat_BBG[128*1024];
+extern u8 VRAMFlat_AOBJ[256*1024];
+extern u8 VRAMFlat_BOBJ[128*1024];
+
+extern u8 VRAMFlat_ABGExtPal[32*1024];
+extern u8 VRAMFlat_BBGExtPal[32*1024];
+
+extern u8 VRAMFlat_AOBJExtPal[8*1024];
+extern u8 VRAMFlat_BOBJExtPal[8*1024];
+
+extern u8 VRAMFlat_Texture[512*1024];
+extern u8 VRAMFlat_TexPal[128*1024];
+
+bool MakeVRAMFlat_ABGCoherent(NonStupidBitField<512*1024/VRAMDirtyGranularity>& dirty);
+bool MakeVRAMFlat_BBGCoherent(NonStupidBitField<128*1024/VRAMDirtyGranularity>& dirty);
+
+bool MakeVRAMFlat_AOBJCoherent(NonStupidBitField<256*1024/VRAMDirtyGranularity>& dirty);
+bool MakeVRAMFlat_BOBJCoherent(NonStupidBitField<128*1024/VRAMDirtyGranularity>& dirty);
+
+bool MakeVRAMFlat_ABGExtPalCoherent(NonStupidBitField<32*1024/VRAMDirtyGranularity>& dirty);
+bool MakeVRAMFlat_BBGExtPalCoherent(NonStupidBitField<32*1024/VRAMDirtyGranularity>& dirty);
+
+bool MakeVRAMFlat_AOBJExtPalCoherent(NonStupidBitField<8*1024/VRAMDirtyGranularity>& dirty);
+bool MakeVRAMFlat_BOBJExtPalCoherent(NonStupidBitField<8*1024/VRAMDirtyGranularity>& dirty);
+
+bool MakeVRAMFlat_TextureCoherent(NonStupidBitField<512*1024/VRAMDirtyGranularity>& dirty);
+bool MakeVRAMFlat_TexPalCoherent(NonStupidBitField<128*1024/VRAMDirtyGranularity>& dirty);
+
+void SyncDirtyFlags();

 typedef struct
 {
@ -233,7 +301,11 @@ void WriteVRAM_LCDC(u32 addr, T val)
    default: return;
    }

-    if (VRAMMap_LCDC & (1<<bank)) *(T*)&VRAM[bank][addr] = val;
+    if (VRAMMap_LCDC & (1<<bank))
+    {
+        *(T*)&VRAM[bank][addr] = val;
+        VRAMDirty[bank][addr / VRAMDirtyGranularity] = true;
+    }
 }


@ -262,6 +334,8 @@ void WriteVRAM_ABG(u32 addr, T val)
 {
    u32 mask = VRAMMap_ABG[(addr >> 14) & 0x1F];

+    VRAMWritten_ABG[(addr & 0x7FFFF) / VRAMDirtyGranularity] = true;
+
    if (mask & (1<<0)) *(T*)&VRAM_A[addr & 0x1FFFF] = val;
    if (mask & (1<<1)) *(T*)&VRAM_B[addr & 0x1FFFF] = val;
    if (mask & (1<<2)) *(T*)&VRAM_C[addr & 0x1FFFF] = val;
@ -295,6 +369,8 @@ void WriteVRAM_AOBJ(u32 addr, T val)
 {
    u32 mask = VRAMMap_AOBJ[(addr >> 14) & 0xF];

+    VRAMWritten_AOBJ[(addr & 0x3FFFF) / VRAMDirtyGranularity] = true;
+
    if (mask & (1<<0)) *(T*)&VRAM_A[addr & 0x1FFFF] = val;
    if (mask & (1<<1)) *(T*)&VRAM_B[addr & 0x1FFFF] = val;
    if (mask & (1<<4)) *(T*)&VRAM_E[addr & 0xFFFF] = val;
@ -324,6 +400,8 @@ void WriteVRAM_BBG(u32 addr, T val)
 {
    u32 mask = VRAMMap_BBG[(addr >> 14) & 0x7];

+    VRAMWritten_BBG[(addr & 0x1FFFF) / VRAMDirtyGranularity] = true;
+
    if (mask & (1<<2)) *(T*)&VRAM_C[addr & 0x1FFFF] = val;
    if (mask & (1<<7)) *(T*)&VRAM_H[addr & 0x7FFF] = val;
    if (mask & (1<<8)) *(T*)&VRAM_I[addr & 0x3FFF] = val;
@ -350,11 +428,12 @@ void WriteVRAM_BOBJ(u32 addr, T val)
 {
    u32 mask = VRAMMap_BOBJ[(addr >> 14) & 0x7];

+    VRAMWritten_BOBJ[(addr & 0x1FFFF) / VRAMDirtyGranularity] = true;
+
    if (mask & (1<<3)) *(T*)&VRAM_D[addr & 0x1FFFF] = val;
    if (mask & (1<<8)) *(T*)&VRAM_I[addr & 0x3FFF] = val;
 }

-
 template<typename T>
 T ReadVRAM_ARM7(u32 addr)
 {
@ -372,6 +451,8 @@ void WriteVRAM_ARM7(u32 addr, T val)
 {
    u32 mask = VRAMMap_ARM7[(addr >> 17) & 0x1];

+    VRAMWritten_ARM7[(addr & 0x1FFFF) / VRAMDirtyGranularity] = true;
+
    if (mask & (1<<2)) *(T*)&VRAM_C[addr & 0x1FFFF] = val;
    if (mask & (1<<3)) *(T*)&VRAM_D[addr & 0x1FFFF] = val;
 }
--- a/src/GPU2D.cpp
+++ b/src/GPU2D.cpp
@ -148,12 +148,6 @@ void GPU2D::Reset()
    CaptureCnt = 0;

    MasterBrightness = 0;
-
-    BGExtPalStatus[0] = 0;
-    BGExtPalStatus[1] = 0;
-    BGExtPalStatus[2] = 0;
-    BGExtPalStatus[3] = 0;
-    OBJExtPalStatus = 0;
 }

 void GPU2D::DoSavestate(Savestate* file)
@ -208,13 +202,6 @@ void GPU2D::DoSavestate(Savestate* file)

    if (!file->Saving)
    {
-        // refresh those
-        BGExtPalStatus[0] = 0;
-        BGExtPalStatus[1] = 0;
-        BGExtPalStatus[2] = 0;
-        BGExtPalStatus[3] = 0;
-        OBJExtPalStatus = 0;
-
        CurBGXMosaicTable = MosaicTable[BGMosaicSize[0]];
        CurOBJXMosaicTable = MosaicTable[OBJMosaicSize[0]];
    }
@ -228,9 +215,6 @@ void GPU2D::SetFramebuffer(u32* buf)
 void GPU2D::SetRenderSettings(bool accel)
 {
    Accelerated = accel;
-
-    if (Accelerated) DrawPixel = DrawPixel_Accel;
-    else             DrawPixel = DrawPixel_Normal;
 }


@ -761,6 +745,25 @@ void GPU2D::DrawScanline(u32 line)
    int n3dline = line;
    line = GPU::VCount;

+    if (Num == 0)
+    {
+        auto bgDirty = GPU::VRAMDirty_ABG.DeriveState(GPU::VRAMMap_ABG);
+        GPU::MakeVRAMFlat_ABGCoherent(bgDirty);
+        auto bgExtPalDirty = GPU::VRAMDirty_ABGExtPal.DeriveState(GPU::VRAMMap_ABGExtPal);
+        GPU::MakeVRAMFlat_ABGExtPalCoherent(bgExtPalDirty);
+        auto objExtPalDirty = GPU::VRAMDirty_AOBJExtPal.DeriveState(&GPU::VRAMMap_AOBJExtPal);
+        GPU::MakeVRAMFlat_AOBJExtPalCoherent(objExtPalDirty);
+    }
+    else
+    {
+        auto bgDirty = GPU::VRAMDirty_BBG.DeriveState(GPU::VRAMMap_BBG);
+        GPU::MakeVRAMFlat_BBGCoherent(bgDirty);
+        auto bgExtPalDirty = GPU::VRAMDirty_BBGExtPal.DeriveState(GPU::VRAMMap_BBGExtPal);
+        GPU::MakeVRAMFlat_BBGExtPalCoherent(bgExtPalDirty);
+        auto objExtPalDirty = GPU::VRAMDirty_BOBJExtPal.DeriveState(&GPU::VRAMMap_BOBJExtPal);
+        GPU::MakeVRAMFlat_BOBJExtPalCoherent(objExtPalDirty);
+    }
+
    bool forceblank = false;

    // scanlines that end up outside of the GPU drawing range
@ -973,6 +976,9 @@ void GPU2D::DoCapture(u32 line, u32 width)
    u16* dst = (u16*)GPU::VRAM[dstvram];
    u32 dstaddr = (((CaptureCnt >> 18) & 0x3) << 14) + (line * width);

+    static_assert(GPU::VRAMDirtyGranularity == 512);
+    GPU::VRAMDirty[dstvram][(dstaddr & 0x1FFFF) / GPU::VRAMDirtyGranularity] = true;
+
    // TODO: handle 3D in accelerated mode!!

    u32* srcA;
@ -1191,85 +1197,20 @@ void GPU2D::SampleFIFO(u32 offset, u32 num)
    }
 }

-
-void GPU2D::BGExtPalDirty(u32 base)
-{
-    BGExtPalStatus[base] = 0;
-    BGExtPalStatus[base+1] = 0;
-}
-
-void GPU2D::OBJExtPalDirty()
-{
-    OBJExtPalStatus = 0;
-}
-
-
 u16* GPU2D::GetBGExtPal(u32 slot, u32 pal)
 {
-    u16* dst = &BGExtPalCache[slot][pal << 8];
-
-    if (!(BGExtPalStatus[slot] & (1<<pal)))
-    {
-        if (Num)
-        {
-            if (GPU::VRAMMap_BBGExtPal[slot] & (1<<7))
-                memcpy(dst, &GPU::VRAM_H[(slot << 13) + (pal << 9)], 256*2);
-            else
-                memset(dst, 0, 256*2);
-        }
-        else
-        {
-            memset(dst, 0, 256*2);
-
-            if (GPU::VRAMMap_ABGExtPal[slot] & (1<<4))
-                for (int i = 0; i < 256; i+=2)
-                    *(u32*)&dst[i] |= *(u32*)&GPU::VRAM_E[(slot << 13) + (pal << 9) + (i << 1)];
-
-            if (GPU::VRAMMap_ABGExtPal[slot] & (1<<5))
-                for (int i = 0; i < 256; i+=2)
-                    *(u32*)&dst[i] |= *(u32*)&GPU::VRAM_F[((slot&1) << 13) + (pal << 9) + (i << 1)];
-
-            if (GPU::VRAMMap_ABGExtPal[slot] & (1<<6))
-                for (int i = 0; i < 256; i+=2)
-                    *(u32*)&dst[i] |= *(u32*)&GPU::VRAM_G[((slot&1) << 13) + (pal << 9) + (i << 1)];
-        }
-
-        BGExtPalStatus[slot] |= (1<<pal);
-    }
-
-    return dst;
+    const u32 PaletteSize = 256 * 2;
+    const u32 SlotSize = PaletteSize * 16;
+    return (u16*)&(Num == 0
+         ? GPU::VRAMFlat_ABGExtPal
+         : GPU::VRAMFlat_BBGExtPal)[slot * SlotSize + pal * PaletteSize];
 }

 u16* GPU2D::GetOBJExtPal()
 {
-    u16* dst = OBJExtPalCache;
-
-    if (!OBJExtPalStatus)
-    {
-        if (Num)
-        {
-            if (GPU::VRAMMap_BOBJExtPal & (1<<8))
-                memcpy(dst, &GPU::VRAM_I[0], 16*256*2);
-            else
-                memset(dst, 0, 16*256*2);
-        }
-        else
-        {
-            memset(dst, 0, 16*256*2);
-
-            if (GPU::VRAMMap_AOBJExtPal & (1<<5))
-                for (int i = 0; i < 16*256; i+=2)
-                    *(u32*)&dst[i] |= *(u32*)&GPU::VRAM_F[i << 1];
-
-            if (GPU::VRAMMap_AOBJExtPal & (1<<6))
-                for (int i = 0; i < 16*256; i+=2)
-                    *(u32*)&dst[i] |= *(u32*)&GPU::VRAM_G[i << 1];
-        }
-
-        OBJExtPalStatus = 1;
-    }
-
-    return dst;
+    return Num == 0
+         ? (u16*)GPU::VRAMFlat_AOBJExtPal
+         : (u16*)GPU::VRAMFlat_BOBJExtPal;
 }


@ -1330,10 +1271,36 @@ void GPU2D::CalculateWindowMask(u32 line)


 #define DoDrawBG(type, line, num) \
-    { if ((BGCnt[num] & 0x0040) && (BGMosaicSize[0] > 0)) DrawBG_##type<true>(line, num); else DrawBG_##type<false>(line, num); }
+    { \
+        if ((BGCnt[num] & 0x0040) && (BGMosaicSize[0] > 0)) \
+        { \
+            if (Accelerated) DrawBG_##type<true, DrawPixel_Accel>(line, num); \
+            else DrawBG_##type<true, DrawPixel_Normal>(line, num); \
+        } \
+        else \
+        { \
+            if (Accelerated) DrawBG_##type<false, DrawPixel_Accel>(line, num); \
+            else DrawBG_##type<false, DrawPixel_Normal>(line, num); \
+        } \
+    }

 #define DoDrawBG_Large(line) \
-    { if ((BGCnt[2] & 0x0040) && (BGMosaicSize[0] > 0)) DrawBG_Large<true>(line); else DrawBG_Large<false>(line); }
+    do \
+    { \
+        if ((BGCnt[2] & 0x0040) && (BGMosaicSize[0] > 0)) \
+        { \
+            if (Accelerated) DrawBG_Large<true, DrawPixel_Accel>(line); \
+            else DrawBG_Large<true, DrawPixel_Normal>(line); \
+        } \
+        else \
+        { \
+            if (Accelerated) DrawBG_Large<false, DrawPixel_Accel>(line); \
+            else DrawBG_Large<false, DrawPixel_Normal>(line); \
+        } \
+    } while (false)
+
+#define DoInterleaveSprites(prio) \
+    if (Accelerated) InterleaveSprites<DrawPixel_Accel>(prio); else InterleaveSprites<DrawPixel_Normal>(prio);

 template<u32 bgmode>
 void GPU2D::DrawScanlineBGMode(u32 line)
@ -1382,7 +1349,7 @@ void GPU2D::DrawScanlineBGMode(u32 line)
            }
        }
        if ((DispCnt & 0x1000) && NumSprites)
-            InterleaveSprites(0x40000 | (i<<16));
+            DoInterleaveSprites(0x40000 | (i<<16));
    }
 }

@ -1394,7 +1361,7 @@ void GPU2D::DrawScanlineBGMode6(u32 line)
        {
            if (DispCnt & 0x0400)
            {
-                DoDrawBG_Large(line)
+                DoDrawBG_Large(line);
            }
        }
        if ((BGCnt[0] & 0x3) == i)
@ -1406,7 +1373,7 @@ void GPU2D::DrawScanlineBGMode6(u32 line)
            }
        }
        if ((DispCnt & 0x1000) && NumSprites)
-            InterleaveSprites(0x40000 | (i<<16));
+            DoInterleaveSprites(0x40000 | (i<<16))
    }
 }

@ -1434,7 +1401,7 @@ void GPU2D::DrawScanlineBGMode7(u32 line)
            }
        }
        if ((DispCnt & 0x1000) && NumSprites)
-            InterleaveSprites(0x40000 | (i<<16));
+            DoInterleaveSprites(0x40000 | (i<<16))
    }
 }

@ -1674,7 +1641,21 @@ void GPU2D::DrawBG_3D()
    }
 }

-template<bool mosaic>
+void GetBGVRAM(u32 num, u8*& data, u32& mask)
+{
+    if (num == 0)
+    {
+        data = GPU::VRAMFlat_ABG;
+        mask = 0x7FFFF;
+    }
+    else
+    {
+        data = GPU::VRAMFlat_BBG;
+        mask = 0x1FFFF;
+    }
+}
+
+template<bool mosaic, GPU2D::DrawPixel drawPixel>
 void GPU2D::DrawBG_Text(u32 line, u32 bgnum)
 {
    u16 bgcnt = BGCnt[bgnum];
@ -1697,17 +1678,20 @@ void GPU2D::DrawBG_Text(u32 line, u32 bgnum)
    extpal = (DispCnt & 0x40000000);
    if (extpal) extpalslot = ((bgnum<2) && (bgcnt&0x2000)) ? (2+bgnum) : bgnum;

+    u8* bgvram;
+    u32 bgvrammask;
+    GetBGVRAM(Num, bgvram, bgvrammask);
    if (Num)
    {
-        tilesetaddr = 0x06200000 + ((bgcnt & 0x003C) << 12);
-        tilemapaddr = 0x06200000 + ((bgcnt & 0x1F00) << 3);
+        tilesetaddr = ((bgcnt & 0x003C) << 12);
+        tilemapaddr = ((bgcnt & 0x1F00) << 3);

        pal = (u16*)&GPU::Palette[0x400];
    }
    else
    {
-        tilesetaddr = 0x06000000 + ((DispCnt & 0x07000000) >> 8) + ((bgcnt & 0x003C) << 12);
-        tilemapaddr = 0x06000000 + ((DispCnt & 0x38000000) >> 11) + ((bgcnt & 0x1F00) << 3);
+        tilesetaddr = ((DispCnt & 0x07000000) >> 8) + ((bgcnt & 0x003C) << 12);
+        tilemapaddr = ((DispCnt & 0x38000000) >> 11) + ((bgcnt & 0x1F00) << 3);

        pal = (u16*)&GPU::Palette[0];
    }
@ -1735,7 +1719,7 @@ void GPU2D::DrawBG_Text(u32 line, u32 bgnum)
        // preload shit as needed
        if ((xoff & 0x7) || mosaic)
        {
-            curtile = GPU::ReadVRAM_BG<u16>(tilemapaddr + ((xoff & 0xF8) >> 2) + ((xoff & widexmask) << 3));
+            curtile = *(u16*)&bgvram[(tilemapaddr + ((xoff & 0xF8) >> 2) + ((xoff & widexmask) << 3)) & bgvrammask];

            if (extpal) curpal = GetBGExtPal(extpalslot, curtile>>12);
            else        curpal = pal;
@ -1756,7 +1740,7 @@ void GPU2D::DrawBG_Text(u32 line, u32 bgnum)
                (mosaic && ((xpos >> 3) != (lastxpos >> 3))))
            {
                // load a new tile
-                curtile = GPU::ReadVRAM_BG<u16>(tilemapaddr + ((xpos & 0xF8) >> 2) + ((xpos & widexmask) << 3));
+                curtile = *(u16*)&bgvram[(tilemapaddr + ((xpos & 0xF8) >> 2) + ((xpos & widexmask) << 3)) & bgvrammask];

                if (extpal) curpal = GetBGExtPal(extpalslot, curtile>>12);
                else        curpal = pal;
@ -1771,10 +1755,10 @@ void GPU2D::DrawBG_Text(u32 line, u32 bgnum)
            if (WindowMask[i] & (1<<bgnum))
            {
                u32 tilexoff = (curtile & 0x0400) ? (7-(xpos&0x7)) : (xpos&0x7);
-                color = GPU::ReadVRAM_BG<u8>(pixelsaddr + tilexoff);
+                color = bgvram[(pixelsaddr + tilexoff) & bgvrammask];

                if (color)
-                    DrawPixel(&BGOBJLine[i], curpal[color], 0x01000000<<bgnum);
+                    drawPixel(&BGOBJLine[i], curpal[color], 0x01000000<<bgnum);
            }

            xoff++;
@ -1787,7 +1771,7 @@ void GPU2D::DrawBG_Text(u32 line, u32 bgnum)
        // preload shit as needed
        if ((xoff & 0x7) || mosaic)
        {
-            curtile = GPU::ReadVRAM_BG<u16>(tilemapaddr + ((xoff & 0xF8) >> 2) + ((xoff & widexmask) << 3));
+            curtile = *(u16*)&bgvram[((tilemapaddr + ((xoff & 0xF8) >> 2) + ((xoff & widexmask) << 3))) & bgvrammask];
            curpal = pal + ((curtile & 0xF000) >> 8);
            pixelsaddr = tilesetaddr + ((curtile & 0x03FF) << 5)
                                     + (((curtile & 0x0800) ? (7-(yoff&0x7)) : (yoff&0x7)) << 2);
@ -1805,7 +1789,7 @@ void GPU2D::DrawBG_Text(u32 line, u32 bgnum)
                (mosaic && ((xpos >> 3) != (lastxpos >> 3))))
            {
                // load a new tile
-                curtile = GPU::ReadVRAM_BG<u16>(tilemapaddr + ((xpos & 0xF8) >> 2) + ((xpos & widexmask) << 3));
+                curtile = *(u16*)&bgvram[(tilemapaddr + ((xpos & 0xF8) >> 2) + ((xpos & widexmask) << 3)) & bgvrammask];
                curpal = pal + ((curtile & 0xF000) >> 8);
                pixelsaddr = tilesetaddr + ((curtile & 0x03FF) << 5)
                                         + (((curtile & 0x0800) ? (7-(yoff&0x7)) : (yoff&0x7)) << 2);
@ -1819,15 +1803,15 @@ void GPU2D::DrawBG_Text(u32 line, u32 bgnum)
                u32 tilexoff = (curtile & 0x0400) ? (7-(xpos&0x7)) : (xpos&0x7);
                if (tilexoff & 0x1)
                {
-                    color = GPU::ReadVRAM_BG<u8>(pixelsaddr + (tilexoff >> 1)) >> 4;
+                    color = bgvram[(pixelsaddr + (tilexoff >> 1)) & bgvrammask] >> 4;
                }
                else
                {
-                    color = GPU::ReadVRAM_BG<u8>(pixelsaddr + (tilexoff >> 1)) & 0x0F;
+                    color = bgvram[(pixelsaddr + (tilexoff >> 1)) & bgvrammask] & 0x0F;
                }

                if (color)
-                    DrawPixel(&BGOBJLine[i], curpal[color], 0x01000000<<bgnum);
+                    drawPixel(&BGOBJLine[i], curpal[color], 0x01000000<<bgnum);
            }

            xoff++;
@ -1835,7 +1819,7 @@ void GPU2D::DrawBG_Text(u32 line, u32 bgnum)
    }
 }

-template<bool mosaic>
+template<bool mosaic, GPU2D::DrawPixel drawPixel>
 void GPU2D::DrawBG_Affine(u32 line, u32 bgnum)
 {
    u16 bgcnt = BGCnt[bgnum];
@ -1872,17 +1856,20 @@ void GPU2D::DrawBG_Affine(u32 line, u32 bgnum)
        rotY -= (BGMosaicY * rotD);
    }

+    u8* bgvram;
+    u32 bgvrammask;
+
    if (Num)
    {
-        tilesetaddr = 0x06200000 + ((bgcnt & 0x003C) << 12);
-        tilemapaddr = 0x06200000 + ((bgcnt & 0x1F00) << 3);
+        tilesetaddr = ((bgcnt & 0x003C) << 12);
+        tilemapaddr = ((bgcnt & 0x1F00) << 3);

        pal = (u16*)&GPU::Palette[0x400];
    }
    else
    {
-        tilesetaddr = 0x06000000 + ((DispCnt & 0x07000000) >> 8) + ((bgcnt & 0x003C) << 12);
-        tilemapaddr = 0x06000000 + ((DispCnt & 0x38000000) >> 11) + ((bgcnt & 0x1F00) << 3);
+        tilesetaddr = ((DispCnt & 0x07000000) >> 8) + ((bgcnt & 0x003C) << 12);
+        tilemapaddr = ((DispCnt & 0x38000000) >> 11) + ((bgcnt & 0x1F00) << 3);

        pal = (u16*)&GPU::Palette[0];
    }
@ -1911,16 +1898,16 @@ void GPU2D::DrawBG_Affine(u32 line, u32 bgnum)

            if ((!((finalX|finalY) & overflowmask)))
            {
-                curtile = GPU::ReadVRAM_BG<u8>(tilemapaddr + ((((finalY & coordmask) >> 11) << yshift) + ((finalX & coordmask) >> 11)));
+                curtile = bgvram[(tilemapaddr + ((((finalY & coordmask) >> 11) << yshift) + ((finalX & coordmask) >> 11))) & bgvrammask];

                // draw pixel
                u32 tilexoff = (finalX >> 8) & 0x7;
                u32 tileyoff = (finalY >> 8) & 0x7;

-                color = GPU::ReadVRAM_BG<u8>(tilesetaddr + (curtile << 6) + (tileyoff << 3) + tilexoff);
+                color = bgvram[(tilesetaddr + (curtile << 6) + (tileyoff << 3) + tilexoff) & bgvrammask];

                if (color)
-                    DrawPixel(&BGOBJLine[i], pal[color], 0x01000000<<bgnum);
+                    drawPixel(&BGOBJLine[i], pal[color], 0x01000000<<bgnum);
            }
        }

@ -1932,7 +1919,7 @@ void GPU2D::DrawBG_Affine(u32 line, u32 bgnum)
    BGYRefInternal[bgnum-2] += rotD;
 }

-template<bool mosaic>
+template<bool mosaic, GPU2D::DrawPixel drawPixel>
 void GPU2D::DrawBG_Extended(u32 line, u32 bgnum)
 {
    u16 bgcnt = BGCnt[bgnum];
@ -1941,6 +1928,10 @@ void GPU2D::DrawBG_Extended(u32 line, u32 bgnum)
    u16* pal;
    u32 extpal;

+    u8* bgvram;
+    u32 bgvrammask;
+    GetBGVRAM(Num, bgvram, bgvrammask);
+
    extpal = (DispCnt & 0x40000000);

    s16 rotA = BGRotA[bgnum-2];
@ -1984,8 +1975,8 @@ void GPU2D::DrawBG_Extended(u32 line, u32 bgnum)
            ofymask = ~ymask;
        }

-        if (Num) tilemapaddr = 0x06200000 + ((bgcnt & 0x1F00) << 6);
-        else     tilemapaddr = 0x06000000 + ((bgcnt & 0x1F00) << 6);
+        if (Num) tilemapaddr = ((bgcnt & 0x1F00) << 6);
+        else     tilemapaddr = ((bgcnt & 0x1F00) << 6);

        if (bgcnt & 0x0004)
        {
@ -2012,10 +2003,10 @@ void GPU2D::DrawBG_Extended(u32 line, u32 bgnum)

                    if (!(finalX & ofxmask) && !(finalY & ofymask))
                    {
-                        color = GPU::ReadVRAM_BG<u16>(tilemapaddr + (((((finalY & ymask) >> 8) << yshift) + ((finalX & xmask) >> 8)) << 1));
+                        color = *(u16*)&bgvram[(tilemapaddr + (((((finalY & ymask) >> 8) << yshift) + ((finalX & xmask) >> 8)) << 1)) & bgvrammask];

                        if (color & 0x8000)
-                            DrawPixel(&BGOBJLine[i], color, 0x01000000<<bgnum);
+                            drawPixel(&BGOBJLine[i], color, 0x01000000<<bgnum);
                    }
                }

@ -2051,10 +2042,10 @@ void GPU2D::DrawBG_Extended(u32 line, u32 bgnum)

                    if (!(finalX & ofxmask) && !(finalY & ofymask))
                    {
-                        color = GPU::ReadVRAM_BG<u8>(tilemapaddr + (((finalY & ymask) >> 8) << yshift) + ((finalX & xmask) >> 8));
+                        color = bgvram[(tilemapaddr + (((finalY & ymask) >> 8) << yshift) + ((finalX & xmask) >> 8)) & bgvrammask];

                        if (color)
-                            DrawPixel(&BGOBJLine[i], pal[color], 0x01000000<<bgnum);
+                            drawPixel(&BGOBJLine[i], pal[color], 0x01000000<<bgnum);
                    }
                }

@ -2083,15 +2074,15 @@ void GPU2D::DrawBG_Extended(u32 line, u32 bgnum)

        if (Num)
        {
-            tilesetaddr = 0x06200000 + ((bgcnt & 0x003C) << 12);
-            tilemapaddr = 0x06200000 + ((bgcnt & 0x1F00) << 3);
+            tilesetaddr = ((bgcnt & 0x003C) << 12);
+            tilemapaddr = ((bgcnt & 0x1F00) << 3);

            pal = (u16*)&GPU::Palette[0x400];
        }
        else
        {
-            tilesetaddr = 0x06000000 + ((DispCnt & 0x07000000) >> 8) + ((bgcnt & 0x003C) << 12);
-            tilemapaddr = 0x06000000 + ((DispCnt & 0x38000000) >> 11) + ((bgcnt & 0x1F00) << 3);
+            tilesetaddr = ((DispCnt & 0x07000000) >> 8) + ((bgcnt & 0x003C) << 12);
+            tilemapaddr = ((DispCnt & 0x38000000) >> 11) + ((bgcnt & 0x1F00) << 3);

            pal = (u16*)&GPU::Palette[0];
        }
@ -2121,7 +2112,7 @@ void GPU2D::DrawBG_Extended(u32 line, u32 bgnum)

                if ((!((finalX|finalY) & overflowmask)))
                {
-                    curtile = GPU::ReadVRAM_BG<u16>(tilemapaddr + (((((finalY & coordmask) >> 11) << yshift) + ((finalX & coordmask) >> 11)) << 1));
+                    curtile = *(u16*)&bgvram[(tilemapaddr + (((((finalY & coordmask) >> 11) << yshift) + ((finalX & coordmask) >> 11)) << 1)) & bgvrammask];

                    if (extpal) curpal = GetBGExtPal(bgnum, curtile>>12);
                    else        curpal = pal;
@ -2133,10 +2124,10 @@ void GPU2D::DrawBG_Extended(u32 line, u32 bgnum)
                    if (curtile & 0x0400) tilexoff = 7-tilexoff;
                    if (curtile & 0x0800) tileyoff = 7-tileyoff;

-                    color = GPU::ReadVRAM_BG<u8>(tilesetaddr + ((curtile & 0x03FF) << 6) + (tileyoff << 3) + tilexoff);
+                    color = bgvram[(tilesetaddr + ((curtile & 0x03FF) << 6) + (tileyoff << 3) + tilexoff) & bgvrammask];

                    if (color)
-                        DrawPixel(&BGOBJLine[i], curpal[color], 0x01000000<<bgnum);
+                        drawPixel(&BGOBJLine[i], curpal[color], 0x01000000<<bgnum);
                }
            }

@ -2149,7 +2140,7 @@ void GPU2D::DrawBG_Extended(u32 line, u32 bgnum)
    BGYRefInternal[bgnum-2] += rotD;
 }

-template<bool mosaic>
+template<bool mosaic, GPU2D::DrawPixel drawPixel>
 void GPU2D::DrawBG_Large(u32 line) // BG is always BG2
 {
    u16 bgcnt = BGCnt[2];
@ -2199,8 +2190,9 @@ void GPU2D::DrawBG_Large(u32 line) // BG is always BG2
        rotY -= (BGMosaicY * rotD);
    }

-    if (Num) tilemapaddr = 0x06200000;
-    else     tilemapaddr = 0x06000000;
+    u8* bgvram;
+    u32 bgvrammask;
+    GetBGVRAM(Num, bgvram, bgvrammask);

    // 256-color bitmap

@ -2228,10 +2220,10 @@ void GPU2D::DrawBG_Large(u32 line) // BG is always BG2

            if (!(finalX & ofxmask) && !(finalY & ofymask))
            {
-                color = GPU::ReadVRAM_BG<u8>(tilemapaddr + (((finalY & ymask) >> 8) << yshift) + ((finalX & xmask) >> 8));
+                color = bgvram[(tilemapaddr + (((finalY & ymask) >> 8) << yshift) + ((finalX & xmask) >> 8)) & bgvrammask];

                if (color)
-                    DrawPixel(&BGOBJLine[i], pal[color], 0x01000000<<2);
+                    drawPixel(&BGOBJLine[i], pal[color], 0x01000000<<2);
            }
        }

@ -2274,6 +2266,7 @@ void GPU2D::ApplySpriteMosaicX()
    }
 }

+template <GPU2D::DrawPixel drawPixel>
 void GPU2D::InterleaveSprites(u32 prio)
 {
    u16* pal = (u16*)&GPU::Palette[Num ? 0x600 : 0x200];
@ -2297,7 +2290,7 @@ void GPU2D::InterleaveSprites(u32 prio)
            else
                color = extpal[pixel & 0xFFF];

-            DrawPixel(&BGOBJLine[i], color, pixel & 0xFF000000);
+            drawPixel(&BGOBJLine[i], color, pixel & 0xFF000000);
        }
    }
    else
@ -2317,11 +2310,25 @@ void GPU2D::InterleaveSprites(u32 prio)
            else
                color = pal[pixel & 0xFF];

-            DrawPixel(&BGOBJLine[i], color, pixel & 0xFF000000);
+            drawPixel(&BGOBJLine[i], color, pixel & 0xFF000000);
        }
    }
 }

+void GetOBJVRAM(u32 num, u8*& data, u32& mask)
+{
+    if (num == 0)
+    {
+        data = GPU::VRAMFlat_AOBJ;
+        mask = 0x3FFFF;
+    }
+    else
+    {
+        data = GPU::VRAMFlat_BOBJ;
+        mask = 0x1FFFF;
+    }
+}
+
 #define DoDrawSprite(type, ...) \
    if (iswin) \
    { \
@ -2346,6 +2353,17 @@ void GPU2D::DrawSprites(u32 line)
        OBJMosaicYCount = 0;
    }

+    if (Num == 0)
+    {
+        auto objDirty = GPU::VRAMDirty_AOBJ.DeriveState(GPU::VRAMMap_AOBJ);
+        GPU::MakeVRAMFlat_AOBJCoherent(objDirty);
+    }
+    else
+    {
+        auto objDirty = GPU::VRAMDirty_BOBJ.DeriveState(GPU::VRAMMap_BOBJ);
+        GPU::MakeVRAMFlat_BOBJCoherent(objDirty);
+    }
+
    NumSprites = 0;
    memset(OBJLine, 0, 256*4);
    memset(OBJWindow, 0, 256);
@ -2458,6 +2476,10 @@ void GPU2D::DrawSprite_Rotscale(u32 num, u32 boundwidth, u32 boundheight, u32 wi

    u32 ytilefactor;

+    u8* objvram;
+    u32 objvrammask;
+    GetOBJVRAM(Num, objvram, objvrammask);
+
    s32 centerX = boundwidth >> 1;
    s32 centerY = boundheight >> 1;

@ -2501,6 +2523,7 @@ void GPU2D::DrawSprite_Rotscale(u32 num, u32 boundwidth, u32 boundheight, u32 wi

        pixelattr |= (0xC0000000 | (alpha << 24));

+        u32 pixelsaddr;
        if (DispCnt & 0x40)
        {
            if (DispCnt & 0x20)
@ -2512,7 +2535,7 @@ void GPU2D::DrawSprite_Rotscale(u32 num, u32 boundwidth, u32 boundheight, u32 wi
            }
            else
            {
-                tilenum <<= (7 + ((DispCnt >> 22) & 0x1));
+                pixelsaddr = tilenum << (7 + ((DispCnt >> 22) & 0x1));
                ytilefactor = ((width >> 8) * 2);
            }
        }
@ -2520,23 +2543,21 @@ void GPU2D::DrawSprite_Rotscale(u32 num, u32 boundwidth, u32 boundheight, u32 wi
        {
            if (DispCnt & 0x20)
            {
-                tilenum = ((tilenum & 0x01F) << 4) + ((tilenum & 0x3E0) << 7);
+                pixelsaddr = ((tilenum & 0x01F) << 4) + ((tilenum & 0x3E0) << 7);
                ytilefactor = (256 * 2);
            }
            else
            {
-                tilenum = ((tilenum & 0x00F) << 4) + ((tilenum & 0x3F0) << 7);
+                pixelsaddr = ((tilenum & 0x00F) << 4) + ((tilenum & 0x3F0) << 7);
                ytilefactor = (128 * 2);
            }
        }

-        u32 pixelsaddr = (Num ? 0x06600000 : 0x06400000) + tilenum;
-
        for (; xoff < boundwidth;)
        {
            if ((u32)rotX < width && (u32)rotY < height)
            {
-                color = GPU::ReadVRAM_OBJ<u16>(pixelsaddr + ((rotY >> 8) * ytilefactor) + ((rotX >> 8) << 1));
+                color = *(u16*)&objvram[(pixelsaddr + ((rotY >> 8) * ytilefactor) + ((rotX >> 8) << 1)) & objvrammask];

                if (color & 0x8000)
                {
@ -2561,9 +2582,10 @@ void GPU2D::DrawSprite_Rotscale(u32 num, u32 boundwidth, u32 boundheight, u32 wi
    }
    else
    {
+        u32 pixelsaddr = tilenum;
        if (DispCnt & 0x10)
        {
-            tilenum <<= ((DispCnt >> 20) & 0x3);
+            pixelsaddr <<= ((DispCnt >> 20) & 0x3);
            ytilefactor = (width >> 11) << ((attrib[0] & 0x2000) ? 1:0);
        }
        else
@ -2574,12 +2596,12 @@ void GPU2D::DrawSprite_Rotscale(u32 num, u32 boundwidth, u32 boundheight, u32 wi
        if (spritemode == 1) pixelattr |= 0x80000000;
        else                 pixelattr |= 0x10000000;

+        ytilefactor <<= 5;
+        pixelsaddr <<= 5;
+
        if (attrib[0] & 0x2000)
        {
            // 256-color
-            tilenum <<= 5;
-            ytilefactor <<= 5;
-            u32 pixelsaddr = (Num ? 0x06600000 : 0x06400000) + tilenum;

            if (!window)
            {
@ -2593,7 +2615,7 @@ void GPU2D::DrawSprite_Rotscale(u32 num, u32 boundwidth, u32 boundheight, u32 wi
            {
                if ((u32)rotX < width && (u32)rotY < height)
                {
-                    color = GPU::ReadVRAM_OBJ<u8>(pixelsaddr + ((rotY>>11)*ytilefactor) + ((rotY&0x700)>>5) + ((rotX>>11)*64) + ((rotX&0x700)>>8));
+                    color = objvram[(pixelsaddr + ((rotY>>11)*ytilefactor) + ((rotY&0x700)>>5) + ((rotX>>11)*64) + ((rotX&0x700)>>8)) & objvrammask];

                    if (color)
                    {
@ -2619,10 +2641,6 @@ void GPU2D::DrawSprite_Rotscale(u32 num, u32 boundwidth, u32 boundheight, u32 wi
        else
        {
            // 16-color
-            tilenum <<= 5;
-            ytilefactor <<= 5;
-            u32 pixelsaddr = (Num ? 0x06600000 : 0x06400000) + tilenum;
-
            if (!window)
            {
                pixelattr |= 0x1000;
@ -2633,7 +2651,7 @@ void GPU2D::DrawSprite_Rotscale(u32 num, u32 boundwidth, u32 boundheight, u32 wi
            {
                if ((u32)rotX < width && (u32)rotY < height)
                {
-                    color = GPU::ReadVRAM_OBJ<u8>(pixelsaddr + ((rotY>>11)*ytilefactor) + ((rotY&0x700)>>6) + ((rotX>>11)*32) + ((rotX&0x700)>>9));
+                    color = objvram[(pixelsaddr + ((rotY>>11)*ytilefactor) + ((rotY&0x700)>>6) + ((rotX>>11)*32) + ((rotX&0x700)>>9)) & objvrammask];
                    if (rotX & 0x100)
                        color >>= 4;
                    else
@ -2681,6 +2699,10 @@ void GPU2D::DrawSprite_Normal(u32 num, u32 width, u32 height, s32 xpos, s32 ypos
        pixelattr |= 0x100000;
    }

+    u8* objvram;
+    u32 objvrammask;
+    GetOBJVRAM(Num, objvram, objvrammask);
+
    // yflip
    if (attrib[1] & 0x2000)
        ypos = height-1 - ypos;
@ -2711,6 +2733,7 @@ void GPU2D::DrawSprite_Normal(u32 num, u32 width, u32 height, s32 xpos, s32 ypos

        pixelattr |= (0xC0000000 | (alpha << 24));

+        u32 pixelsaddr = tilenum;
        if (DispCnt & 0x40)
        {
            if (DispCnt & 0x20)
@ -2722,25 +2745,24 @@ void GPU2D::DrawSprite_Normal(u32 num, u32 width, u32 height, s32 xpos, s32 ypos
            }
            else
            {
-                tilenum <<= (7 + ((DispCnt >> 22) & 0x1));
-                tilenum += (ypos * width * 2);
+                pixelsaddr <<= (7 + ((DispCnt >> 22) & 0x1));
+                pixelsaddr += (ypos * width * 2);
            }
        }
        else
        {
            if (DispCnt & 0x20)
            {
-                tilenum = ((tilenum & 0x01F) << 4) + ((tilenum & 0x3E0) << 7);
-                tilenum += (ypos * 256 * 2);
+                pixelsaddr = ((tilenum & 0x01F) << 4) + ((tilenum & 0x3E0) << 7);
+                pixelsaddr += (ypos * 256 * 2);
            }
            else
            {
-                tilenum = ((tilenum & 0x00F) << 4) + ((tilenum & 0x3F0) << 7);
-                tilenum += (ypos * 128 * 2);
+                pixelsaddr = ((tilenum & 0x00F) << 4) + ((tilenum & 0x3F0) << 7);
+                pixelsaddr += (ypos * 128 * 2);
            }
        }

-        u32 pixelsaddr = (Num ? 0x06600000 : 0x06400000) + tilenum;
        s32 pixelstride;

        if (attrib[1] & 0x1000) // xflip
@ -2757,7 +2779,7 @@ void GPU2D::DrawSprite_Normal(u32 num, u32 width, u32 height, s32 xpos, s32 ypos

        for (; xoff < xend;)
        {
-            color = GPU::ReadVRAM_OBJ<u16>(pixelsaddr);
+            color = *(u16*)&objvram[pixelsaddr & objvrammask];

            pixelsaddr += pixelstride;

@ -2781,14 +2803,15 @@ void GPU2D::DrawSprite_Normal(u32 num, u32 width, u32 height, s32 xpos, s32 ypos
    }
    else
    {
+        u32 pixelsaddr = tilenum;
        if (DispCnt & 0x10)
        {
-            tilenum <<= ((DispCnt >> 20) & 0x3);
-            tilenum += ((ypos >> 3) * (width >> 3)) << ((attrib[0] & 0x2000) ? 1:0);
+            pixelsaddr <<= ((DispCnt >> 20) & 0x3);
+            pixelsaddr += ((ypos >> 3) * (width >> 3)) << ((attrib[0] & 0x2000) ? 1:0);
        }
        else
        {
-            tilenum += ((ypos >> 3) * 0x20);
+            pixelsaddr += ((ypos >> 3) * 0x20);
        }

        if (spritemode == 1) pixelattr |= 0x80000000;
@ -2797,8 +2820,7 @@ void GPU2D::DrawSprite_Normal(u32 num, u32 width, u32 height, s32 xpos, s32 ypos
        if (attrib[0] & 0x2000)
        {
            // 256-color
-            tilenum <<= 5;
-            u32 pixelsaddr = (Num ? 0x06600000 : 0x06400000) + tilenum;
+            pixelsaddr <<= 5;
            pixelsaddr += ((ypos & 0x7) << 3);
            s32 pixelstride;

@ -2827,7 +2849,7 @@ void GPU2D::DrawSprite_Normal(u32 num, u32 width, u32 height, s32 xpos, s32 ypos

            for (; xoff < xend;)
            {
-                color = GPU::ReadVRAM_OBJ<u8>(pixelsaddr);
+                color = objvram[pixelsaddr];

                pixelsaddr += pixelstride;

@ -2853,8 +2875,7 @@ void GPU2D::DrawSprite_Normal(u32 num, u32 width, u32 height, s32 xpos, s32 ypos
        else
        {
            // 16-color
-            tilenum <<= 5;
-            u32 pixelsaddr = (Num ? 0x06600000 : 0x06400000) + tilenum;
+            pixelsaddr <<= 5;
            pixelsaddr += ((ypos & 0x7) << 2);
            s32 pixelstride;

@ -2886,13 +2907,13 @@ void GPU2D::DrawSprite_Normal(u32 num, u32 width, u32 height, s32 xpos, s32 ypos
            {
                if (attrib[1] & 0x1000)
                {
-                    if (xoff & 0x1) { color = GPU::ReadVRAM_OBJ<u8>(pixelsaddr) & 0x0F; pixelsaddr--; }
-                    else              color = GPU::ReadVRAM_OBJ<u8>(pixelsaddr) >> 4;
+                    if (xoff & 0x1) { color = objvram[pixelsaddr & objvrammask] & 0x0F; pixelsaddr--; }
+                    else              color = objvram[pixelsaddr & objvrammask] >> 4;
                }
                else
                {
-                    if (xoff & 0x1) { color = GPU::ReadVRAM_OBJ<u8>(pixelsaddr) >> 4; pixelsaddr++; }
-                    else              color = GPU::ReadVRAM_OBJ<u8>(pixelsaddr) & 0x0F;
+                    if (xoff & 0x1) { color = objvram[pixelsaddr & objvrammask] >> 4; pixelsaddr++; }
+                    else              color = objvram[pixelsaddr & objvrammask] & 0x0F;
                }

                if (color)
--- a/src/GPU2D.h
+++ b/src/GPU2D.h
@ -59,9 +59,6 @@ public:

    void CheckWindows(u32 line);

-    void BGExtPalDirty(u32 base);
-    void OBJExtPalDirty();
-
    u16* GetBGExtPal(u32 slot, u32 pal);
    u16* GetOBJExtPal();

@ -128,9 +125,6 @@ private:
    u16 MasterBrightness;

    u16 BGExtPalCache[4][16*256];
-    u16 OBJExtPalCache[16*256];
-    u32 BGExtPalStatus[4];
-    u32 OBJExtPalStatus;

    u32 ColorBlend4(u32 val1, u32 val2, u32 eva, u32 evb);
    u32 ColorBlend5(u32 val1, u32 val2);
@ -147,15 +141,17 @@ private:

    static void DrawPixel_Normal(u32* dst, u16 color, u32 flag);
    static void DrawPixel_Accel(u32* dst, u16 color, u32 flag);
-    void (*DrawPixel)(u32* dst, u16 color, u32 flag);
+    
+    typedef void (*DrawPixel)(u32* dst, u16 color, u32 flag);

    void DrawBG_3D();
-    template<bool mosaic> void DrawBG_Text(u32 line, u32 bgnum);
-    template<bool mosaic> void DrawBG_Affine(u32 line, u32 bgnum);
-    template<bool mosaic> void DrawBG_Extended(u32 line, u32 bgnum);
-    template<bool mosaic> void DrawBG_Large(u32 line);
+    template<bool mosaic, DrawPixel drawPixel> void DrawBG_Text(u32 line, u32 bgnum);
+    template<bool mosaic, DrawPixel drawPixel> void DrawBG_Affine(u32 line, u32 bgnum);
+    template<bool mosaic, DrawPixel drawPixel> void DrawBG_Extended(u32 line, u32 bgnum);
+    template<bool mosaic, DrawPixel drawPixel> void DrawBG_Large(u32 line);

    void ApplySpriteMosaicX();
+    template<DrawPixel drawPixel>
    void InterleaveSprites(u32 prio);
    template<bool window> void DrawSprite_Rotscale(u32 num, u32 boundwidth, u32 boundheight, u32 width, u32 height, s32 xpos, s32 ypos);
    template<bool window> void DrawSprite_Normal(u32 num, u32 width, u32 height, s32 xpos, s32 ypos);
--- a/src/GPU3D.cpp
+++ b/src/GPU3D.cpp
@ -179,6 +179,8 @@ u8 RenderFogDensityTable[34];

 u32 RenderClearAttr1, RenderClearAttr2;

+bool RenderFrameIdentical;
+
 u32 ZeroDotWLimit;

 u32 GXStat;
@ -2491,6 +2493,19 @@ void VBlank()
                }

                RenderNumPolygons = NumPolygons;
+                RenderFrameIdentical = false;
+            }
+            else
+            {
+                RenderFrameIdentical = RenderDispCnt == DispCnt
+                    && RenderAlphaRef == AlphaRef
+                    && RenderClearAttr1 == ClearAttr1
+                    && RenderClearAttr2 == ClearAttr2
+                    && RenderFogColor == FogColor
+                    && RenderFogOffset == FogOffset * 0x200
+                    && memcmp(RenderEdgeTable, EdgeTable, 8*2) == 0
+                    && memcmp(RenderFogDensityTable + 1, FogDensityTable, 32) == 0
+                    && memcmp(RenderToonTable, ToonTable, 32*2) == 0;
            }

            RenderDispCnt = DispCnt;
--- a/src/GPU3D.h
+++ b/src/GPU3D.h
@ -87,6 +87,8 @@ extern u8 RenderFogDensityTable[34];

 extern u32 RenderClearAttr1, RenderClearAttr2;

+extern bool RenderFrameIdentical;
+
 extern std::array<Polygon*,2048> RenderPolygonRAM;
 extern u32 RenderNumPolygons;

--- a/src/GPU3D_OpenGL.cpp
+++ b/src/GPU3D_OpenGL.cpp
@ -74,11 +74,11 @@ typedef struct
    Polygon* PolyData;

    u32 NumIndices;
-    u16* Indices;
+    u32 IndicesOffset;
    GLuint PrimType;

    u32 NumEdgeIndices;
-    u16* EdgeIndices;
+    u32 EdgeIndicesOffset;

    u32 RenderKey;

@ -107,7 +107,11 @@ u32 VertexBuffer[10240 * 7];
 u32 NumVertices;

 GLuint VertexArrayID;
+GLuint IndexBufferID;
 u16 IndexBuffer[2048 * 40];
+u32 NumIndices, NumEdgeIndices;
+
+const u32 EdgeIndicesOffset = 2048 * 30;

 GLuint TexMemID;
 GLuint TexPalMemID;
@ -320,6 +324,9 @@ bool Init()
    glEnableVertexAttribArray(3); // attrib
    glVertexAttribIPointer(3, 3, GL_UNSIGNED_INT, 7*4, (void*)(4*4));

+    glGenBuffers(1, &IndexBufferID);
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, IndexBufferID);
+    glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(IndexBuffer), NULL, GL_DYNAMIC_DRAW);

    glGenFramebuffers(4, &FramebufferID[0]);
    glBindFramebuffer(GL_FRAMEBUFFER, FramebufferID[0]);
@ -563,15 +570,15 @@ void BuildPolygons(RendererPolygon* polygons, int npolys)
    u32* vptr = &VertexBuffer[0];
    u32 vidx = 0;

-    u16* iptr = &IndexBuffer[0];
-    u16* eiptr = &IndexBuffer[2048*30];
+    u32 iidx = 0;
+    u32 eidx = EdgeIndicesOffset;

    for (int i = 0; i < npolys; i++)
    {
        RendererPolygon* rp = &polygons[i];
        Polygon* poly = rp->PolyData;

-        rp->Indices = iptr;
+        rp->IndicesOffset = iidx;
        rp->NumIndices = 0;

        u32 vidx_first = vidx;
@ -606,7 +613,7 @@ void BuildPolygons(RendererPolygon* polygons, int npolys)

                vptr = SetupVertex(poly, j, vtx, vtxattr, vptr);

-                *iptr++ = vidx;
+                IndexBuffer[iidx++] = vidx;
                rp->NumIndices++;

                vidx++;
@ -627,9 +634,9 @@ void BuildPolygons(RendererPolygon* polygons, int npolys)
            }

            // build a triangle
-            *iptr++ = vidx_first;
-            *iptr++ = vidx - 2;
-            *iptr++ = vidx - 1;
+            IndexBuffer[iidx++] = vidx_first;
+            IndexBuffer[iidx++] = vidx - 2;
+            IndexBuffer[iidx++] = vidx - 1;
            rp->NumIndices += 3;
        }
        else // quad, pentagon, etc
@ -649,9 +656,9 @@ void BuildPolygons(RendererPolygon* polygons, int npolys)
                    if (j >= 2)
                    {
                        // build a triangle
-                        *iptr++ = vidx_first;
-                        *iptr++ = vidx - 1;
-                        *iptr++ = vidx;
+                        IndexBuffer[iidx++] = vidx_first;
+                        IndexBuffer[iidx++] = vidx - 1;
+                        IndexBuffer[iidx++] = vidx;
                        rp->NumIndices += 3;
                    }

@ -743,46 +750,48 @@ void BuildPolygons(RendererPolygon* polygons, int npolys)
                    if (j >= 1)
                    {
                        // build a triangle
-                        *iptr++ = vidx_first;
-                        *iptr++ = vidx - 1;
-                        *iptr++ = vidx;
+                        IndexBuffer[iidx++] = vidx_first;
+                        IndexBuffer[iidx++] = vidx - 1;
+                        IndexBuffer[iidx++] = vidx;
                        rp->NumIndices += 3;
                    }

                    vidx++;
                }

-                *iptr++ = vidx_first;
-                *iptr++ = vidx - 1;
-                *iptr++ = vidx_first + 1;
+                IndexBuffer[iidx++] = vidx_first;
+                IndexBuffer[iidx++] = vidx - 1;
+                IndexBuffer[iidx++] = vidx_first + 1;
                rp->NumIndices += 3;
            }
        }

-        rp->EdgeIndices = eiptr;
+        rp->EdgeIndicesOffset = eidx;
        rp->NumEdgeIndices = 0;

        u32 vidx_cur = vidx_first;
        for (int j = 1; j < poly->NumVertices; j++)
        {
-            *eiptr++ = vidx_cur;
-            *eiptr++ = vidx_cur + 1;
+            IndexBuffer[eidx++] = vidx_cur;
+            IndexBuffer[eidx++] = vidx_cur + 1;
            vidx_cur++;
            rp->NumEdgeIndices += 2;
        }
-        *eiptr++ = vidx_cur;
-        *eiptr++ = vidx_first;
+        IndexBuffer[eidx++] = vidx_cur;
+        IndexBuffer[eidx++] = vidx_first;
        rp->NumEdgeIndices += 2;
    }

    NumVertices = vidx;
+    NumIndices = iidx;
+    NumEdgeIndices = eidx - EdgeIndicesOffset;
 }

 void RenderSinglePolygon(int i)
 {
    RendererPolygon* rp = &PolygonList[i];

-    glDrawElements(rp->PrimType, rp->NumIndices, GL_UNSIGNED_SHORT, rp->Indices);
+    glDrawElements(rp->PrimType, rp->NumIndices, GL_UNSIGNED_SHORT, (void*)(uintptr_t)(rp->IndicesOffset * 2));
 }

 int RenderPolygonBatch(int i)
@ -803,7 +812,7 @@ int RenderPolygonBatch(int i)
        numindices += cur_rp->NumIndices;
    }

-    glDrawElements(primtype, numindices, GL_UNSIGNED_SHORT, rp->Indices);
+    glDrawElements(primtype, numindices, GL_UNSIGNED_SHORT, (void*)(uintptr_t)(rp->IndicesOffset * 2));
    return numpolys;
 }

@ -823,7 +832,7 @@ int RenderPolygonEdgeBatch(int i)
        numindices += cur_rp->NumEdgeIndices;
    }

-    glDrawElements(GL_LINES, numindices, GL_UNSIGNED_SHORT, rp->EdgeIndices);
+    glDrawElements(GL_LINES, numindices, GL_UNSIGNED_SHORT, (void*)(uintptr_t)(rp->EdgeIndicesOffset * 2));
    return numpolys;
 }

@ -1320,6 +1329,11 @@ void RenderFrame()
        glBindBuffer(GL_ARRAY_BUFFER, VertexBufferID);
        glBufferSubData(GL_ARRAY_BUFFER, 0, NumVertices*7*4, VertexBuffer);

+        // bind to access the index buffer
+        glBindVertexArray(VertexArrayID);
+        glBufferSubData(GL_ELEMENT_ARRAY_BUFFER, 0, NumIndices * 2, IndexBuffer);
+        glBufferSubData(GL_ELEMENT_ARRAY_BUFFER, EdgeIndicesOffset * 2, NumEdgeIndices * 2, IndexBuffer + EdgeIndicesOffset);
+
        RenderSceneChunk(0, 192);
    }

--- a/src/GPU3D_Soft.cpp
+++ b/src/GPU3D_Soft.cpp
@ -58,15 +58,17 @@ bool PrevIsShadowMask;

 bool Enabled;

+bool FrameIdentical;
+
 // threading

 bool Threaded;
-void* RenderThread;
+Platform::Thread* RenderThread;
 bool RenderThreadRunning;
 bool RenderThreadRendering;
-void* Sema_RenderStart;
-void* Sema_RenderDone;
-void* Sema_ScanlineCount;
+Platform::Semaphore* Sema_RenderStart;
+Platform::Semaphore* Sema_RenderDone;
+Platform::Semaphore* Sema_ScanlineCount;

 void RenderThreadFunc();

@ -550,6 +552,16 @@ typedef struct

 RendererPolygon PolygonList[2048];

+template <typename T>
+inline T ReadVRAM_Texture(u32 addr)
+{
+    return *(T*)&GPU::VRAMFlat_Texture[addr & 0x7FFFF];
+}
+template <typename T>
+inline T ReadVRAM_TexPal(u32 addr)
+{
+    return *(T*)&GPU::VRAMFlat_TexPal[addr & 0x1FFFF];
+}

 void TextureLookup(u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha)
 {
@ -606,10 +618,10 @@ void TextureLookup(u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha
    case 1: // A3I5
        {
            vramaddr += ((t * width) + s);
-            u8 pixel = GPU::ReadVRAM_Texture<u8>(vramaddr);
+            u8 pixel = ReadVRAM_Texture<u8>(vramaddr);

            texpal <<= 4;
-            *color = GPU::ReadVRAM_TexPal<u16>(texpal + ((pixel&0x1F)<<1));
+            *color = ReadVRAM_TexPal<u16>(texpal + ((pixel&0x1F)<<1));
            *alpha = ((pixel >> 3) & 0x1C) + (pixel >> 6);
        }
        break;
@ -617,12 +629,12 @@ void TextureLookup(u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha
    case 2: // 4-color
        {
            vramaddr += (((t * width) + s) >> 2);
-            u8 pixel = GPU::ReadVRAM_Texture<u8>(vramaddr);
+            u8 pixel = ReadVRAM_Texture<u8>(vramaddr);
            pixel >>= ((s & 0x3) << 1);
            pixel &= 0x3;

            texpal <<= 3;
-            *color = GPU::ReadVRAM_TexPal<u16>(texpal + (pixel<<1));
+            *color = ReadVRAM_TexPal<u16>(texpal + (pixel<<1));
            *alpha = (pixel==0) ? alpha0 : 31;
        }
        break;
@ -630,12 +642,12 @@ void TextureLookup(u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha
    case 3: // 16-color
        {
            vramaddr += (((t * width) + s) >> 1);
-            u8 pixel = GPU::ReadVRAM_Texture<u8>(vramaddr);
+            u8 pixel = ReadVRAM_Texture<u8>(vramaddr);
            if (s & 0x1) pixel >>= 4;
            else         pixel &= 0xF;

            texpal <<= 4;
-            *color = GPU::ReadVRAM_TexPal<u16>(texpal + (pixel<<1));
+            *color = ReadVRAM_TexPal<u16>(texpal + (pixel<<1));
            *alpha = (pixel==0) ? alpha0 : 31;
        }
        break;
@ -643,10 +655,10 @@ void TextureLookup(u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha
    case 4: // 256-color
        {
            vramaddr += ((t * width) + s);
-            u8 pixel = GPU::ReadVRAM_Texture<u8>(vramaddr);
+            u8 pixel = ReadVRAM_Texture<u8>(vramaddr);

            texpal <<= 4;
-            *color = GPU::ReadVRAM_TexPal<u16>(texpal + (pixel<<1));
+            *color = ReadVRAM_TexPal<u16>(texpal + (pixel<<1));
            *alpha = (pixel==0) ? alpha0 : 31;
        }
        break;
@ -660,30 +672,30 @@ void TextureLookup(u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha
            if (vramaddr >= 0x40000)
                slot1addr += 0x10000;

-            u8 val = GPU::ReadVRAM_Texture<u8>(vramaddr);
+            u8 val = ReadVRAM_Texture<u8>(vramaddr);
            val >>= (2 * (s & 0x3));

-            u16 palinfo = GPU::ReadVRAM_Texture<u16>(slot1addr);
+            u16 palinfo = ReadVRAM_Texture<u16>(slot1addr);
            u32 paloffset = (palinfo & 0x3FFF) << 2;
            texpal <<= 4;

            switch (val & 0x3)
            {
            case 0:
-                *color = GPU::ReadVRAM_TexPal<u16>(texpal + paloffset);
+                *color = ReadVRAM_TexPal<u16>(texpal + paloffset);
                *alpha = 31;
                break;

            case 1:
-                *color = GPU::ReadVRAM_TexPal<u16>(texpal + paloffset + 2);
+                *color = ReadVRAM_TexPal<u16>(texpal + paloffset + 2);
                *alpha = 31;
                break;

            case 2:
                if ((palinfo >> 14) == 1)
                {
-                    u16 color0 = GPU::ReadVRAM_TexPal<u16>(texpal + paloffset);
-                    u16 color1 = GPU::ReadVRAM_TexPal<u16>(texpal + paloffset + 2);
+                    u16 color0 = ReadVRAM_TexPal<u16>(texpal + paloffset);
+                    u16 color1 = ReadVRAM_TexPal<u16>(texpal + paloffset + 2);

                    u32 r0 = color0 & 0x001F;
                    u32 g0 = color0 & 0x03E0;
@ -700,8 +712,8 @@ void TextureLookup(u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha
                }
                else if ((palinfo >> 14) == 3)
                {
-                    u16 color0 = GPU::ReadVRAM_TexPal<u16>(texpal + paloffset);
-                    u16 color1 = GPU::ReadVRAM_TexPal<u16>(texpal + paloffset + 2);
+                    u16 color0 = ReadVRAM_TexPal<u16>(texpal + paloffset);
+                    u16 color1 = ReadVRAM_TexPal<u16>(texpal + paloffset + 2);

                    u32 r0 = color0 & 0x001F;
                    u32 g0 = color0 & 0x03E0;
@ -717,20 +729,20 @@ void TextureLookup(u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha
                    *color = r | g | b;
                }
                else
-                    *color = GPU::ReadVRAM_TexPal<u16>(texpal + paloffset + 4);
+                    *color = ReadVRAM_TexPal<u16>(texpal + paloffset + 4);
                *alpha = 31;
                break;

            case 3:
                if ((palinfo >> 14) == 2)
                {
-                    *color = GPU::ReadVRAM_TexPal<u16>(texpal + paloffset + 6);
+                    *color = ReadVRAM_TexPal<u16>(texpal + paloffset + 6);
                    *alpha = 31;
                }
                else if ((palinfo >> 14) == 3)
                {
-                    u16 color0 = GPU::ReadVRAM_TexPal<u16>(texpal + paloffset);
-                    u16 color1 = GPU::ReadVRAM_TexPal<u16>(texpal + paloffset + 2);
+                    u16 color0 = ReadVRAM_TexPal<u16>(texpal + paloffset);
+                    u16 color1 = ReadVRAM_TexPal<u16>(texpal + paloffset + 2);

                    u32 r0 = color0 & 0x001F;
                    u32 g0 = color0 & 0x03E0;
@ -759,10 +771,10 @@ void TextureLookup(u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha
    case 6: // A5I3
        {
            vramaddr += ((t * width) + s);
-            u8 pixel = GPU::ReadVRAM_Texture<u8>(vramaddr);
+            u8 pixel = ReadVRAM_Texture<u8>(vramaddr);

            texpal <<= 4;
-            *color = GPU::ReadVRAM_TexPal<u16>(texpal + ((pixel&0x7)<<1));
+            *color = ReadVRAM_TexPal<u16>(texpal + ((pixel&0x7)<<1));
            *alpha = (pixel >> 3);
        }
        break;
@ -770,7 +782,7 @@ void TextureLookup(u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha
    case 7: // direct color
        {
            vramaddr += (((t * width) + s) << 1);
-            *color = GPU::ReadVRAM_Texture<u16>(vramaddr);
+            *color = ReadVRAM_Texture<u16>(vramaddr);
            *alpha = (*color & 0x8000) ? 31 : 0;
        }
        break;
@ -2007,8 +2019,8 @@ void ClearBuffers()
        {
            for (int x = 0; x < 256; x++)
            {
-                u16 val2 = GPU::ReadVRAM_Texture<u16>(0x40000 + (yoff << 9) + (xoff << 1));
-                u16 val3 = GPU::ReadVRAM_Texture<u16>(0x60000 + (yoff << 9) + (xoff << 1));
+                u16 val2 = ReadVRAM_Texture<u16>(0x40000 + (yoff << 9) + (xoff << 1));
+                u16 val3 = ReadVRAM_Texture<u16>(0x60000 + (yoff << 9) + (xoff << 1));

                // TODO: confirm color conversion
                u32 r = (val2 << 1) & 0x3E; if (r) r++;
@ -2088,11 +2100,19 @@ void VCount144()

 void RenderFrame()
 {
+    auto textureDirty = GPU::VRAMDirty_Texture.DeriveState(GPU::VRAMMap_Texture);
+    auto texPalDirty = GPU::VRAMDirty_TexPal.DeriveState(GPU::VRAMMap_TexPal);
+
+    bool textureChanged = GPU::MakeVRAMFlat_TextureCoherent(textureDirty);
+    bool texPalChanged = GPU::MakeVRAMFlat_TexPalCoherent(texPalDirty);
+
+    FrameIdentical = !(textureChanged || texPalChanged) && RenderFrameIdentical;
+
    if (RenderThreadRunning)
    {
        Platform::Semaphore_Post(Sema_RenderStart);
    }
-    else
+    else if (!FrameIdentical)
    {
        ClearBuffers();
        RenderPolygons(false, &RenderPolygonRAM[0], RenderNumPolygons);
@ -2107,8 +2127,15 @@ void RenderThreadFunc()
        if (!RenderThreadRunning) return;

        RenderThreadRendering = true;
-        ClearBuffers();
-        RenderPolygons(true, &RenderPolygonRAM[0], RenderNumPolygons);
+        if (FrameIdentical)
+        {
+            Platform::Semaphore_Post(Sema_ScanlineCount, 192);
+        }
+        else
+        {
+            ClearBuffers();
+            RenderPolygons(true, &RenderPolygonRAM[0], RenderNumPolygons);
+        }

        Platform::Semaphore_Post(Sema_RenderDone);
        RenderThreadRendering = false;
--- a/src/NDS.cpp
+++ b/src/NDS.cpp
@ -210,13 +210,13 @@ bool Init()

 void DeInit()
 {
-    delete ARM9;
-    delete ARM7;
-
 #ifdef JIT_ENABLED
    ARMJIT::DeInit();
 #endif

+    delete ARM9;
+    delete ARM7;
+
    for (int i = 0; i < 8; i++)
        delete DMAs[i];

@ -908,7 +908,7 @@ void RunSystem(u64 timestamp)
    }
 }

-template <bool EnableJIT>
+template <bool EnableJIT, int ConsoleType>
 u32 RunFrame()
 {
    FrameStartTimestamp = SysTimestamp;
@ -934,10 +934,10 @@ u32 RunFrame()
        }
        else if (CPUStop & 0x0FFF)
        {
-            DMAs[0]->Run();
-            if (!(CPUStop & 0x80000000)) DMAs[1]->Run();
-            if (!(CPUStop & 0x80000000)) DMAs[2]->Run();
-            if (!(CPUStop & 0x80000000)) DMAs[3]->Run();
+            DMAs[0]->Run<ConsoleType>();
+            if (!(CPUStop & 0x80000000)) DMAs[1]->Run<ConsoleType>();
+            if (!(CPUStop & 0x80000000)) DMAs[2]->Run<ConsoleType>();
+            if (!(CPUStop & 0x80000000)) DMAs[3]->Run<ConsoleType>();
            if (ConsoleType == 1) DSi::RunNDMAs(0);
        }
        else
@ -962,10 +962,10 @@ u32 RunFrame()

            if (CPUStop & 0x0FFF0000)
            {
-                DMAs[4]->Run();
-                DMAs[5]->Run();
-                DMAs[6]->Run();
-                DMAs[7]->Run();
+                DMAs[4]->Run<ConsoleType>();
+                DMAs[5]->Run<ConsoleType>();
+                DMAs[6]->Run<ConsoleType>();
+                DMAs[7]->Run<ConsoleType>();
                if (ConsoleType == 1) DSi::RunNDMAs(1);
            }
            else
@ -999,6 +999,9 @@ u32 RunFrame()
           ARM7Timestamp-SysTimestamp,
           GPU3D::Timestamp-SysTimestamp);
 #endif
+    SPU::TransferOutput();
+
+    NDSCart::FlushSRAMFile();

    NumFrames++;

@ -1009,10 +1012,14 @@ u32 RunFrame()
 {
 #ifdef JIT_ENABLED
    if (Config::JIT_Enable)
-        return RunFrame<true>();
+        return NDS::ConsoleType == 1
+            ? RunFrame<true, 1>()
+            : RunFrame<true, 0>();
    else
 #endif
-        return RunFrame<false>();
+        return NDS::ConsoleType == 1
+            ? RunFrame<false, 1>()
+            : RunFrame<false, 0>();
 }

 void Reschedule(u64 target)
@ -3130,6 +3137,10 @@ void ARM9IOWrite8(u32 addr, u8 val)
        NDSCart::WriteSPIData(val);
        return;

+    case 0x04000188:
+        ARM9IOWrite32(addr, val | (val << 8) | (val << 16) | (val << 24));
+        return;
+
    case 0x040001A8: NDSCart::ROMCommand[0] = val; return;
    case 0x040001A9: NDSCart::ROMCommand[1] = val; return;
    case 0x040001AA: NDSCart::ROMCommand[2] = val; return;
@ -3246,6 +3257,10 @@ void ARM9IOWrite16(u32 addr, u16 val)
        IPCFIFOCnt9 = val & 0x8404;
        return;

+    case 0x04000188:
+        ARM9IOWrite32(addr, val | (val << 16)); 
+        return;
+
    case 0x040001A0:
        if (!(ExMemCnt[0] & (1<<11))) NDSCart::WriteSPICnt(val);
        return;
@ -3733,6 +3748,10 @@ void ARM7IOWrite8(u32 addr, u8 val)

    case 0x04000138: RTC::Write(val, true); return;

+    case 0x04000188:
+        ARM7IOWrite32(addr, val | (val << 8) | (val << 16) | (val << 24));
+        return;
+
    case 0x040001A0:
        if (ExMemCnt[0] & (1<<11))
        {
@ -3841,6 +3860,10 @@ void ARM7IOWrite16(u32 addr, u16 val)
        IPCFIFOCnt7 = val & 0x8404;
        return;

+    case 0x04000188:
+        ARM7IOWrite32(addr, val | (val << 16));
+        return;
+
    case 0x040001A0:
        if (ExMemCnt[0] & (1<<11))
            NDSCart::WriteSPICnt(val);
--- a/src/NDSCart.cpp
+++ b/src/NDSCart.cpp
@ -37,6 +37,7 @@ u8* SRAM;
 u32 SRAMLength;

 char SRAMPath[1024];
+bool SRAMFileDirty;

 void (*WriteFunc)(u8 val, bool islast);

@ -445,14 +446,21 @@ void Write(u8 val, u32 hold)
        break;
    }

-    if (islast && (CurCmd == 0x02 || CurCmd == 0x0A) && (SRAMLength > 0))
+    SRAMFileDirty |= islast && (CurCmd == 0x02 || CurCmd == 0x0A) && (SRAMLength > 0);
+}
+
+void FlushSRAMFile()
+{
+    if (!SRAMFileDirty)
+        return;
+
+    SRAMFileDirty = false;
+
+    FILE* f = Platform::OpenFile(SRAMPath, "wb");
+    if (f)
    {
-        FILE* f = Platform::OpenFile(SRAMPath, "wb");
-        if (f)
-        {
-            fwrite(SRAM, SRAMLength, 1, f);
-            fclose(f);
-        }
+        fwrite(SRAM, SRAMLength, 1, f);
+        fclose(f);
    }
 }

@ -1034,6 +1042,11 @@ void RelocateSave(const char* path, bool write)
    NDSCart_SRAM::RelocateSave(path, write);
 }

+void FlushSRAMFile()
+{
+    NDSCart_SRAM::FlushSRAMFile();
+}
+
 int ImportSRAM(const u8* data, u32 length)
 {
    memcpy(NDSCart_SRAM::SRAM, data, std::min(length, NDSCart_SRAM::SRAMLength));
--- a/src/NDSCart.h
+++ b/src/NDSCart.h
@ -46,6 +46,9 @@ void DoSavestate(Savestate* file);

 void DecryptSecureArea(u8* out);
 bool LoadROM(const char* path, const char* sram, bool direct);
+
+void FlushSRAMFile();
+
 void RelocateSave(const char* path, bool write);

 int ImportSRAM(const u8* data, u32 length);
--- a/src/NonStupidBitfield.h
+++ b/src/NonStupidBitfield.h
@ -0,0 +1,149 @@
+#ifndef NONSTUPIDBITFIELD_H
+#define NONSTUPIDBITFIELD_H
+
+#include "types.h"
+
+#include <memory.h>
+
+#include <initializer_list>
+#include <algorithm>
+
+// like std::bitset but less stupid and optimised for 
+// our use case (keeping track of memory invalidations)
+
+template <u32 Size>
+struct NonStupidBitField
+{
+    static_assert((Size % 8) == 0, "bitfield size must be a multiple of 8");
+    static const u32 DataLength = Size / 8;
+    u8 Data[DataLength];
+
+    struct Ref
+    {
+        NonStupidBitField<Size>& BitField;
+        u32 Idx;
+
+        operator bool()
+        {
+            return BitField.Data[Idx >> 3] & (1 << (Idx & 0x7));
+        }
+
+        Ref& operator=(bool set)
+        {
+            BitField.Data[Idx >> 3] &= ~(1 << (Idx & 0x7));
+            BitField.Data[Idx >> 3] |= ((u8)set << (Idx & 0x7));
+            return *this;
+        }
+    };
+
+    struct Iterator
+    {
+        NonStupidBitField<Size>& BitField;
+        u32 DataIdx;
+        u32 BitIdx;
+        u64 RemainingBits;
+
+        u32 operator*() { return DataIdx * 8 + BitIdx; }
+
+        bool operator==(const Iterator& other) { return other.DataIdx == DataIdx; }
+        bool operator!=(const Iterator& other) { return other.DataIdx != DataIdx; }
+
+        template <typename T>
+        void Next()
+        {
+            while (RemainingBits == 0 && DataIdx < DataLength)
+            {
+                DataIdx += sizeof(T);
+                RemainingBits = *(T*)&BitField.Data[DataIdx];
+            }
+
+            BitIdx = __builtin_ctzll(RemainingBits);
+            RemainingBits &= ~(1ULL << BitIdx);
+        }
+
+        Iterator operator++(int)
+        {
+            Iterator prev(*this);
+            ++*this;
+            return prev;
+        }
+
+        Iterator& operator++()
+        {
+            if ((DataLength % 8) == 0)
+                Next<u64>();
+            else if ((DataLength % 4) == 0)
+                Next<u32>();
+            else if ((DataLength % 2) == 0)
+                Next<u16>();
+            else
+                Next<u8>();
+
+            return *this;
+        }
+    };
+
+    NonStupidBitField(u32 start, u32 size)
+    {
+        memset(Data, 0, sizeof(Data));
+
+        if (size == 0)
+            return;
+
+        u32 roundedStartBit = (start + 7) & ~7;
+        u32 roundedEndBit = (start + size) & ~7;
+        if (roundedStartBit != roundedEndBit)
+            memset(Data + roundedStartBit / 8, 0xFF, (roundedEndBit - roundedStartBit) / 8);
+
+        if (start & 0x7)
+            Data[start >> 3] = 0xFF << (start & 0x7);
+        if ((start + size) & 0x7)
+            Data[(start + size) >> 3] = 0xFF >> ((start + size) & 0x7);
+    }
+
+    NonStupidBitField()
+    {
+        memset(Data, 0, sizeof(Data));
+    }
+
+    Iterator End()
+    {
+        return Iterator{*this, DataLength, 0, 0};
+    }
+    Iterator Begin()
+    {
+        if ((DataLength % 8) == 0)
+            return ++Iterator{*this, 0, 0, *(u64*)Data};
+        else if ((DataLength % 4) == 0)
+            return ++Iterator{*this, 0, 0, *(u32*)Data};
+        else if ((DataLength % 2) == 0)
+            return ++Iterator{*this, 0, 0, *(u16*)Data};
+        else
+            return ++Iterator{*this, 0, 0, *Data};
+    }
+
+    Ref operator[](u32 idx)
+    {
+        return Ref{*this, idx};
+    }
+
+    NonStupidBitField& operator|=(const NonStupidBitField<Size>& other)
+    {
+        for (u32 i = 0; i < DataLength; i++)
+        {
+            Data[i] |= other.Data[i];
+        }
+        return *this;
+    }
+    NonStupidBitField& operator&=(const NonStupidBitField<Size>& other)
+    {
+        for (u32 i = 0; i < DataLength; i++)
+        {
+            Data[i] &= other.Data[i];
+        }
+        return *this;
+    }
+};
+
+
+#endif
--- a/src/OpenGLSupport.h
+++ b/src/OpenGLSupport.h
@ -23,8 +23,13 @@
 #include <string.h>

 // TODO: different includes for each platform
-#include <GL/gl.h>
-#include <GL/glext.h>
+#ifdef __APPLE__
+    #include <OpenGL/gl3.h>
+    #include <OpenGL/gl3ext.h>
+#else
+    #include <GL/gl.h>
+    #include <GL/glext.h>
+#endif

 #include "Platform.h"

@ -61,6 +66,11 @@

 #endif

+#ifdef __APPLE__
+
+#define DO_PROCLIST(func)
+
+#else

 #define DO_PROCLIST(func) \
    DO_PROCLIST_1_3(func) \
@ -128,6 +138,7 @@
     \
    func(GLGETSTRINGI, glGetStringi); \

+#endif

 namespace OpenGL
 {
--- a/src/Platform.h
+++ b/src/Platform.h
@ -67,15 +67,24 @@ inline bool LocalFileExists(const char* name)
    return true;
 }

-void* Thread_Create(void (*func)());
-void Thread_Free(void* thread);
-void Thread_Wait(void* thread);
+struct Thread;
+Thread* Thread_Create(void (*func)());
+void Thread_Free(Thread* thread);
+void Thread_Wait(Thread* thread);

-void* Semaphore_Create();
-void Semaphore_Free(void* sema);
-void Semaphore_Reset(void* sema);
-void Semaphore_Wait(void* sema);
-void Semaphore_Post(void* sema);
+struct Semaphore;
+Semaphore* Semaphore_Create();
+void Semaphore_Free(Semaphore* sema);
+void Semaphore_Reset(Semaphore* sema);
+void Semaphore_Wait(Semaphore* sema);
+void Semaphore_Post(Semaphore* sema, int count = 1);
+
+struct Mutex;
+Mutex* Mutex_Create();
+void Mutex_Free(Mutex* mutex);
+void Mutex_Lock(Mutex* mutex);
+void Mutex_Unlock(Mutex* mutex);
+bool Mutex_TryLock(Mutex* mutex);

 void* GL_GetProcAddress(const char* proc);

--- a/src/SPU.cpp
+++ b/src/SPU.cpp
@ -18,6 +18,7 @@

 #include <stdio.h>
 #include <string.h>
+#include "Platform.h"
 #include "NDS.h"
 #include "DSi.h"
 #include "SPU.h"
@ -61,13 +62,15 @@ const s16 PSGTable[8][8] =
    {-0x7FFF, -0x7FFF, -0x7FFF, -0x7FFF, -0x7FFF, -0x7FFF, -0x7FFF, -0x7FFF}
 };

-const u32 kSamplesPerRun = 1;
+const u32 OutputBufferSize = 2*2048;
+s16 OutputBackbuffer[2 * OutputBufferSize];
+u32 OutputBackbufferWritePosition;

-const u32 OutputBufferSize = 2*1024;
-s16 OutputBuffer[2 * OutputBufferSize];
-volatile u32 OutputReadOffset;
-volatile u32 OutputWriteOffset;
+s16 OutputFrontBuffer[2 * OutputBufferSize];
+u32 OutputFrontBufferWritePosition;
+u32 OutputFrontBufferReadPosition;

+Platform::Mutex* AudioLock;

 u16 Cnt;
 u8 MasterVolume;
@ -85,6 +88,8 @@ bool Init()
    Capture[0] = new CaptureUnit(0);
    Capture[1] = new CaptureUnit(1);

+    AudioLock = Platform::Mutex_Create();
+
    return true;
 }

@ -95,6 +100,8 @@ void DeInit()

    delete Capture[0];
    delete Capture[1];
+
+    Platform::Mutex_Free(AudioLock);
 }

 void Reset()
@ -111,15 +118,18 @@ void Reset()
    Capture[0]->Reset();
    Capture[1]->Reset();

-    NDS::ScheduleEvent(NDS::Event_SPU, true, 1024*kSamplesPerRun, Mix, kSamplesPerRun);
+    NDS::ScheduleEvent(NDS::Event_SPU, true, 1024, Mix, 0);
 }

 void Stop()
 {
-    memset(OutputBuffer, 0, 2*OutputBufferSize*2);
+    Platform::Mutex_Lock(AudioLock);
+    memset(OutputFrontBuffer, 0, 2*OutputBufferSize*2);

-    OutputReadOffset = 0;
-    OutputWriteOffset = 0;
+    OutputBackbufferWritePosition = 0;
+    OutputFrontBufferReadPosition = 0;
+    OutputFrontBufferWritePosition = 0;
+    Platform::Mutex_Unlock(AudioLock);
 }

 void DoSavestate(Savestate* file)
@ -416,11 +426,11 @@ void Channel::NextSample_Noise()
 }

 template<u32 type>
-void Channel::Run(s32* buf, u32 samples)
+s32 Channel::Run()
 {
-    if (!(Cnt & (1<<31))) return;
+    if (!(Cnt & (1<<31))) return 0;

-    if ((type < 3) && ((Length+LoopPos) < 16)) return;
+    if ((type < 3) && ((Length+LoopPos) < 16)) return 0;

    if (KeyOn)
    {
@ -428,45 +438,32 @@ void Channel::Run(s32* buf, u32 samples)
        KeyOn = false;
    }

-    for (u32 s = 0; s < samples; s++)
+    Timer += 512; // 1 sample = 512 cycles at 16MHz
+
+    while (Timer >> 16)
    {
-        Timer += 512; // 1 sample = 512 cycles at 16MHz
+        Timer = TimerReload + (Timer - 0x10000);

-        while (Timer >> 16)
+        switch (type)
        {
-            Timer = TimerReload + (Timer - 0x10000);
-
-            switch (type)
-            {
-            case 0: NextSample_PCM8(); break;
-            case 1: NextSample_PCM16(); break;
-            case 2: NextSample_ADPCM(); break;
-            case 3: NextSample_PSG(); break;
-            case 4: NextSample_Noise(); break;
-            }
+        case 0: NextSample_PCM8(); break;
+        case 1: NextSample_PCM16(); break;
+        case 2: NextSample_ADPCM(); break;
+        case 3: NextSample_PSG(); break;
+        case 4: NextSample_Noise(); break;
        }
-
-        s32 val = (s32)CurSample;
-        val <<= VolumeShift;
-        val *= Volume;
-        buf[s] = val;
-
-        if (!(Cnt & (1<<31))) break;
    }
+
+    s32 val = (s32)CurSample;
+    val <<= VolumeShift;
+    val *= Volume;
+    return val;
 }

-void Channel::PanOutput(s32* inbuf, u32 samples, s32* leftbuf, s32* rightbuf)
+void Channel::PanOutput(s32 in, s32& left, s32& right)
 {
-    for (u32 s = 0; s < samples; s++)
-    {
-        s32 val = (s32)inbuf[s];
-
-        s32 l = ((s64)val * (128-Pan)) >> 10;
-        s32 r = ((s64)val * Pan) >> 10;
-
-        leftbuf[s] += l;
-        rightbuf[s] += r;
-    }
+    left += ((s64)in * (128-Pan)) >> 10;
+    right += ((s64)in * Pan) >> 10;
 }


@ -602,39 +599,31 @@ void CaptureUnit::Run(s32 sample)
 }


-void Mix(u32 samples)
+void Mix(u32 dummy)
 {
-    s32 channelbuf[32];
-    s32 leftbuf[32], rightbuf[32];
-    s32 ch0buf[32], ch1buf[32], ch2buf[32], ch3buf[32];
-    s32 leftoutput[32], rightoutput[32];
-
-    for (u32 s = 0; s < samples; s++)
-    {
-        leftbuf[s] = 0; rightbuf[s] = 0;
-        leftoutput[s] = 0; rightoutput[s] = 0;
-    }
+    s32 left = 0, right = 0;
+    s32 leftoutput = 0, rightoutput = 0;

    if (Cnt & (1<<15))
    {
-        Channels[0]->DoRun(ch0buf, samples);
-        Channels[1]->DoRun(ch1buf, samples);
-        Channels[2]->DoRun(ch2buf, samples);
-        Channels[3]->DoRun(ch3buf, samples);
+        s32 ch0 = Channels[0]->DoRun();
+        s32 ch1 = Channels[1]->DoRun();
+        s32 ch2 = Channels[2]->DoRun();
+        s32 ch3 = Channels[3]->DoRun();

        // TODO: addition from capture registers
-        Channels[0]->PanOutput(ch0buf, samples, leftbuf, rightbuf);
-        Channels[2]->PanOutput(ch2buf, samples, leftbuf, rightbuf);
+        Channels[0]->PanOutput(ch0, left, right);
+        Channels[2]->PanOutput(ch2, left, right);

-        if (!(Cnt & (1<<12))) Channels[1]->PanOutput(ch1buf, samples, leftbuf, rightbuf);
-        if (!(Cnt & (1<<13))) Channels[3]->PanOutput(ch3buf, samples, leftbuf, rightbuf);
+        if (!(Cnt & (1<<12))) Channels[1]->PanOutput(ch1, left, right);
+        if (!(Cnt & (1<<13))) Channels[3]->PanOutput(ch3, left, right);

        for (int i = 4; i < 16; i++)
        {
            Channel* chan = Channels[i];

-            chan->DoRun(channelbuf, samples);
-            chan->PanOutput(channelbuf, samples, leftbuf, rightbuf);
+            s32 channel = chan->DoRun();
+            chan->PanOutput(channel, left, right);
        }

        // sound capture
@ -642,32 +631,24 @@ void Mix(u32 samples)

        if (Capture[0]->Cnt & (1<<7))
        {
-            for (u32 s = 0; s < samples; s++)
-            {
-                s32 val = leftbuf[s];
+            s32 val = left;

-                val >>= 8;
-                if      (val < -0x8000) val = -0x8000;
-                else if (val > 0x7FFF)  val = 0x7FFF;
+            val >>= 8;
+            if      (val < -0x8000) val = -0x8000;
+            else if (val > 0x7FFF)  val = 0x7FFF;

-                Capture[0]->Run(val);
-                if (!(Capture[0]->Cnt & (1<<7))) break;
-            }
+            Capture[0]->Run(val);
        }

        if (Capture[1]->Cnt & (1<<7))
        {
-            for (u32 s = 0; s < samples; s++)
-            {
-                s32 val = rightbuf[s];
+            s32 val = right;

-                val >>= 8;
-                if      (val < -0x8000) val = -0x8000;
-                else if (val > 0x7FFF)  val = 0x7FFF;
+            val >>= 8;
+            if      (val < -0x8000) val = -0x8000;
+            else if (val > 0x7FFF)  val = 0x7FFF;

-                Capture[1]->Run(val);
-                if (!(Capture[1]->Cnt & (1<<7))) break;
-            }
+            Capture[1]->Run(val);
        }

        // final output
@ -675,31 +656,25 @@ void Mix(u32 samples)
        switch (Cnt & 0x0300)
        {
        case 0x0000: // left mixer
-            {
-                for (u32 s = 0; s < samples; s++)
-                    leftoutput[s] = leftbuf[s];
-            }
+            leftoutput = left;
            break;
        case 0x0100: // channel 1
            {
                s32 pan = 128 - Channels[1]->Pan;
-                for (u32 s = 0; s < samples; s++)
-                    leftoutput[s] = ((s64)ch1buf[s] * pan) >> 10;
+                leftoutput = ((s64)ch1 * pan) >> 10;
            }
            break;
        case 0x0200: // channel 3
            {
                s32 pan = 128 - Channels[3]->Pan;
-                for (u32 s = 0; s < samples; s++)
-                    leftoutput[s] = ((s64)ch3buf[s] * pan) >> 10;
+                leftoutput = ((s64)ch3 * pan) >> 10;
            }
            break;
        case 0x0300: // channel 1+3
            {
                s32 pan1 = 128 - Channels[1]->Pan;
                s32 pan3 = 128 - Channels[3]->Pan;
-                for (u32 s = 0; s < samples; s++)
-                    leftoutput[s] = (((s64)ch1buf[s] * pan1) >> 10) + (((s64)ch3buf[s] * pan3) >> 10);
+                leftoutput = (((s64)ch1 * pan1) >> 10) + (((s64)ch3 * pan3) >> 10);
            }
            break;
        }
@ -707,105 +682,122 @@ void Mix(u32 samples)
        switch (Cnt & 0x0C00)
        {
        case 0x0000: // right mixer
-            {
-                for (u32 s = 0; s < samples; s++)
-                    rightoutput[s] = rightbuf[s];
-            }
+            rightoutput = right;
            break;
        case 0x0400: // channel 1
            {
                s32 pan = Channels[1]->Pan;
-                for (u32 s = 0; s < samples; s++)
-                    rightoutput[s] = ((s64)ch1buf[s] * pan) >> 10;
+                rightoutput = ((s64)ch1 * pan) >> 10;
            }
            break;
        case 0x0800: // channel 3
            {
                s32 pan = Channels[3]->Pan;
-                for (u32 s = 0; s < samples; s++)
-                    rightoutput[s] = ((s64)ch3buf[s] * pan) >> 10;
+                rightoutput = ((s64)ch3 * pan) >> 10;
            }
            break;
        case 0x0C00: // channel 1+3
            {
                s32 pan1 = Channels[1]->Pan;
                s32 pan3 = Channels[3]->Pan;
-                for (u32 s = 0; s < samples; s++)
-                    rightoutput[s] = (((s64)ch1buf[s] * pan1) >> 10) + (((s64)ch3buf[s] * pan3) >> 10);
+                rightoutput = (((s64)ch1 * pan1) >> 10) + (((s64)ch3 * pan3) >> 10);
            }
            break;
        }
    }

-    for (u32 s = 0; s < samples; s++)
-    {
-        s32 l = leftoutput[s];
-        s32 r = rightoutput[s];
+    leftoutput = ((s64)leftoutput * MasterVolume) >> 7;
+    rightoutput = ((s64)rightoutput * MasterVolume) >> 7;

-        l = ((s64)l * MasterVolume) >> 7;
-        r = ((s64)r * MasterVolume) >> 7;
+    leftoutput >>= 8;
+    if      (leftoutput < -0x8000) leftoutput = -0x8000;
+    else if (leftoutput > 0x7FFF)  leftoutput = 0x7FFF;
+    rightoutput >>= 8;
+    if      (rightoutput < -0x8000) rightoutput = -0x8000;
+    else if (rightoutput > 0x7FFF)  rightoutput = 0x7FFF;

-        l >>= 8;
-        if      (l < -0x8000) l = -0x8000;
-        else if (l > 0x7FFF)  l = 0x7FFF;
-        r >>= 8;
-        if      (r < -0x8000) r = -0x8000;
-        else if (r > 0x7FFF)  r = 0x7FFF;
+    // OutputBufferFrame can never get full because it's
+    // transfered to OutputBuffer at the end of the frame
+    OutputBackbuffer[OutputBackbufferWritePosition    ] = leftoutput >> 1;
+    OutputBackbuffer[OutputBackbufferWritePosition + 1] = rightoutput >> 1;
+    OutputBackbufferWritePosition += 2;

-        OutputBuffer[OutputWriteOffset    ] = l >> 1;
-        OutputBuffer[OutputWriteOffset + 1] = r >> 1;
-        OutputWriteOffset += 2;
-        OutputWriteOffset &= ((2*OutputBufferSize)-1);
-        if (OutputWriteOffset == OutputReadOffset)
-        {
-            //printf("!! SOUND FIFO OVERFLOW %d\n", OutputWriteOffset>>1);
-            // advance the read position too, to avoid losing the entire FIFO
-            OutputReadOffset += 2;
-            OutputReadOffset &= ((2*OutputBufferSize)-1);
-        }
-    }
-
-    NDS::ScheduleEvent(NDS::Event_SPU, true, 1024*kSamplesPerRun, Mix, kSamplesPerRun);
+    NDS::ScheduleEvent(NDS::Event_SPU, true, 1024, Mix, 0);
 }

+void TransferOutput()
+{
+    Platform::Mutex_Lock(AudioLock);
+    for (u32 i = 0; i < OutputBackbufferWritePosition; i += 2)
+    {
+        OutputFrontBuffer[OutputFrontBufferWritePosition    ] = OutputBackbuffer[i   ];
+        OutputFrontBuffer[OutputFrontBufferWritePosition + 1] = OutputBackbuffer[i + 1];
+
+        OutputFrontBufferWritePosition += 2;
+        OutputFrontBufferWritePosition &= OutputBufferSize*2-1;
+        if (OutputFrontBufferWritePosition == OutputFrontBufferReadPosition)
+        {
+            // advance the read position too, to avoid losing the entire FIFO
+            OutputFrontBufferReadPosition += 2;
+            OutputFrontBufferReadPosition &= OutputBufferSize*2-1;
+        }
+    }
+    OutputBackbufferWritePosition = 0;
+    Platform::Mutex_Unlock(AudioLock);
+}

 void TrimOutput()
 {
+    Platform::Mutex_Lock(AudioLock);
    const int halflimit = (OutputBufferSize / 2);

-    int readpos = OutputWriteOffset - (halflimit*2);
+    int readpos = OutputFrontBufferWritePosition - (halflimit*2);
    if (readpos < 0) readpos += (OutputBufferSize*2);

-    OutputReadOffset = readpos;
+    OutputFrontBufferReadPosition = readpos;
+    Platform::Mutex_Unlock(AudioLock);
 }

 void DrainOutput()
 {
-    OutputReadOffset = 0;
-    OutputWriteOffset = 0;
+    Platform::Mutex_Lock(AudioLock);
+    OutputFrontBufferWritePosition = 0;
+    OutputFrontBufferReadPosition = 0;
+    Platform::Mutex_Unlock(AudioLock);
 }

 void InitOutput()
 {
-    memset(OutputBuffer, 0, 2*OutputBufferSize*2);
-    OutputReadOffset = 0;
-    OutputWriteOffset = OutputBufferSize;
+    Platform::Mutex_Lock(AudioLock);
+    memset(OutputBackbuffer, 0, 2*OutputBufferSize*2);
+    memset(OutputFrontBuffer, 0, 2*OutputBufferSize*2);
+    OutputFrontBufferReadPosition = 0;
+    OutputFrontBufferWritePosition = 0;
+    Platform::Mutex_Unlock(AudioLock);
 }

 int GetOutputSize()
 {
+    Platform::Mutex_Lock(AudioLock);
+
    int ret;
-    if (OutputWriteOffset >= OutputReadOffset)
-        ret = OutputWriteOffset - OutputReadOffset;
+    if (OutputFrontBufferWritePosition >= OutputFrontBufferReadPosition)
+        ret = OutputFrontBufferWritePosition - OutputFrontBufferReadPosition;
    else
-        ret = (OutputBufferSize*2) - OutputReadOffset + OutputWriteOffset;
+        ret = (OutputBufferSize*2) - OutputFrontBufferReadPosition + OutputFrontBufferWritePosition;

    ret >>= 1;
+
+    Platform::Mutex_Unlock(AudioLock);
    return ret;
 }

 void Sync(bool wait)
 {
+    // this function is currently not used anywhere
+    // depending on the usage context the thread safety measures could be made
+    // a lot faster
+
    // sync to audio output in case the core is running too fast
    // * wait=true: wait until enough audio data has been played
    // * wait=false: merely skip some audio data to avoid a FIFO overflow
@ -819,32 +811,42 @@ void Sync(bool wait)
    }
    else if (GetOutputSize() > halflimit)
    {
-        int readpos = OutputWriteOffset - (halflimit*2);
+        Platform::Mutex_Lock(AudioLock);
+
+        int readpos = OutputFrontBufferWritePosition - (halflimit*2);
        if (readpos < 0) readpos += (OutputBufferSize*2);

-        OutputReadOffset = readpos;
+        OutputFrontBufferReadPosition = readpos;
+
+        Platform::Mutex_Unlock(AudioLock);
    }
 }

 int ReadOutput(s16* data, int samples)
 {
-    if (OutputReadOffset == OutputWriteOffset)
+    Platform::Mutex_Lock(AudioLock);
+    if (OutputFrontBufferReadPosition == OutputFrontBufferWritePosition)
+    {
+        Platform::Mutex_Unlock(AudioLock);
        return 0;
+    }

    for (int i = 0; i < samples; i++)
    {
-        *data++ = OutputBuffer[OutputReadOffset];
-        *data++ = OutputBuffer[OutputReadOffset + 1];
+        *data++ = OutputFrontBuffer[OutputFrontBufferReadPosition];
+        *data++ = OutputFrontBuffer[OutputFrontBufferReadPosition + 1];

-        //if (OutputReadOffset != OutputWriteOffset)
+        OutputFrontBufferReadPosition += 2;
+        OutputFrontBufferReadPosition &= ((2*OutputBufferSize)-1);
+
+        if (OutputFrontBufferWritePosition == OutputFrontBufferReadPosition)
        {
-            OutputReadOffset += 2;
-            OutputReadOffset &= ((2*OutputBufferSize)-1);
-        }
-        if (OutputReadOffset == OutputWriteOffset)
+            Platform::Mutex_Unlock(AudioLock);
            return i+1;
+        }
    }

+    Platform::Mutex_Unlock(AudioLock);
    return samples;
 }

--- a/src/SPU.h
+++ b/src/SPU.h
@ -33,7 +33,7 @@ void DoSavestate(Savestate* file);

 void SetBias(u16 bias);

-void Mix(u32 samples);
+void Mix(u32 dummy);

 void TrimOutput();
 void DrainOutput();
@ -41,6 +41,7 @@ void InitOutput();
 int GetOutputSize();
 void Sync(bool wait);
 int ReadOutput(s16* data, int samples);
+void TransferOutput();

 u8 Read8(u32 addr);
 u16 Read16(u32 addr);
@ -123,26 +124,24 @@ public:
    void NextSample_PSG();
    void NextSample_Noise();

-    template<u32 type> void Run(s32* buf, u32 samples);
+    template<u32 type> s32 Run();

-    void DoRun(s32* buf, u32 samples)
+    s32 DoRun()
    {
-        for (u32 s = 0; s < samples; s++)
-            buf[s] = 0;
-
        switch ((Cnt >> 29) & 0x3)
        {
-        case 0: Run<0>(buf, samples); break;
-        case 1: Run<1>(buf, samples); break;
-        case 2: Run<2>(buf, samples); break;
+        case 0: return Run<0>(); break;
+        case 1: return Run<1>(); break;
+        case 2: return Run<2>(); break;
        case 3:
-            if      (Num >= 14) Run<4>(buf, samples);
-            else if (Num >= 8)  Run<3>(buf, samples);
-            break;
+            if      (Num >= 14) return Run<4>();
+            else if (Num >= 8)  return Run<3>();
+        default:
+            return 0;
        }
    }

-    void PanOutput(s32* inbuf, u32 samples, s32* leftbuf, s32* rightbuf);
+    void PanOutput(s32 in, s32& left, s32& right);

 private:
    u32 (*BusRead32)(u32 addr);
--- a/src/frontend/SharedConfig.h
+++ b/src/frontend/SharedConfig.h
@ -0,0 +1,13 @@
+#ifndef SHAREDCONFIG_H
+#define SHAREDCONFIG_H
+
+namespace Config
+{
+
+extern int ConsoleType;
+extern int DirectBoot;
+extern int SavestateRelocSRAM;
+
+}
+
+#endif
--- a/src/frontend/Util_ROM.cpp
+++ b/src/frontend/Util_ROM.cpp
@ -21,7 +21,7 @@

 #include "FrontendUtil.h"
 #include "Config.h"
-#include "qt_sdl/PlatformConfig.h" // FIXME!!!
+#include "SharedConfig.h"
 #include "Platform.h"

 #include "NDS.h"
--- a/src/frontend/qt_sdl/CMakeLists.txt
+++ b/src/frontend/qt_sdl/CMakeLists.txt
@ -95,6 +95,19 @@ if (PORTABLE)
    add_definitions(-DPORTABLE)
 endif()

+if (APPLE)
+    set_target_properties(melonDS PROPERTIES
+                         MACOSX_BUNDLE true
+                         MACOSX_BUNDLE_INFO_PLIST ${CMAKE_SOURCE_DIR}/melonDS.plist
+                         OUTPUT_NAME melonDS
+                         )
+
+    # Copy icon into the bundle
+    target_sources(melonDS PRIVATE "${CMAKE_SOURCE_DIR}/melonDS.icns")
+    set_source_files_properties("${CMAKE_SOURCE_DIR}/melonDS.icns" PROPERTIES MACOSX_PACKAGE_LOCATION Resources)
+
+endif()
+
 install(FILES ../../../net.kuribo64.melonDS.desktop DESTINATION ${CMAKE_INSTALL_PREFIX}/share/applications)
 install(FILES ../../../icon/melon_16x16.png DESTINATION ${CMAKE_INSTALL_PREFIX}/share/icons/hicolor/16x16/apps RENAME net.kuribo64.melonDS.png)
 install(FILES ../../../icon/melon_32x32.png DESTINATION ${CMAKE_INSTALL_PREFIX}/share/icons/hicolor/32x32/apps RENAME net.kuribo64.melonDS.png)
@ -102,4 +115,4 @@ install(FILES ../../../icon/melon_48x48.png DESTINATION ${CMAKE_INSTALL_PREFIX}/
 install(FILES ../../../icon/melon_64x64.png DESTINATION ${CMAKE_INSTALL_PREFIX}/share/icons/hicolor/64x64/apps RENAME net.kuribo64.melonDS.png)
 install(FILES ../../../icon/melon_128x128.png DESTINATION ${CMAKE_INSTALL_PREFIX}/share/icons/hicolor/128x128/apps RENAME net.kuribo64.melonDS.png)
 install(FILES ../../../icon/melon_256x256.png DESTINATION ${CMAKE_INSTALL_PREFIX}/share/icons/hicolor/256x256/apps RENAME net.kuribo64.melonDS.png)
-install(TARGETS melonDS RUNTIME DESTINATION ${CMAKE_INSTALL_PREFIX}/bin)
+install(TARGETS melonDS BUNDLE DESTINATION ${CMAKE_BINARY_DIR} RUNTIME DESTINATION ${CMAKE_INSTALL_PREFIX}/bin)
--- a/src/frontend/qt_sdl/EmuSettingsDialog.cpp
+++ b/src/frontend/qt_sdl/EmuSettingsDialog.cpp
@ -65,6 +65,9 @@ EmuSettingsDialog::EmuSettingsDialog(QWidget* parent) : QDialog(parent), ui(new
    ui->chkJITBranchOptimisations->setChecked(Config::JIT_BranchOptimisations != 0);
    ui->chkJITLiteralOptimisations->setChecked(Config::JIT_LiteralOptimisations != 0);
    ui->chkJITFastMemory->setChecked(Config::JIT_FastMemory != 0);
+    #ifdef __APPLE__
+        ui->chkJITFastMemory->setDisabled(true);
+    #endif
    ui->spnJITMaximumBlockSize->setValue(Config::JIT_MaxBlockSize);
 #else
    ui->chkEnableJIT->setDisabled(true);
@ -329,6 +332,8 @@ void EmuSettingsDialog::on_chkEnableJIT_toggled()
    bool disabled = !ui->chkEnableJIT->isChecked();
    ui->chkJITBranchOptimisations->setDisabled(disabled);
    ui->chkJITLiteralOptimisations->setDisabled(disabled);
-    ui->chkJITFastMemory->setDisabled(disabled);
+    #ifndef __APPLE__
+        ui->chkJITFastMemory->setDisabled(disabled);
+    #endif
    ui->spnJITMaximumBlockSize->setDisabled(disabled);
 }
--- a/src/frontend/qt_sdl/InputConfigDialog.cpp
+++ b/src/frontend/qt_sdl/InputConfigDialog.cpp
@ -216,6 +216,7 @@ KeyMapButton::KeyMapButton(int* mapping, bool hotkey) : QPushButton()

    setCheckable(true);
    setText(mappingText());
+    setFocusPolicy(Qt::StrongFocus); //Fixes binding keys in macOS

    connect(this, &KeyMapButton::clicked, this, &KeyMapButton::onClick);
 }
--- a/src/frontend/qt_sdl/LAN_PCap.cpp
+++ b/src/frontend/qt_sdl/LAN_PCap.cpp
@ -33,7 +33,11 @@
 	#include <sys/types.h>
 	#include <ifaddrs.h>
 	#include <netinet/in.h>
-	#include <linux/if_packet.h>
+        #ifdef __linux__
+            #include <linux/if_packet.h>
+        #else
+	    #include <net/if_dl.h>
+        #endif
 #endif


@ -66,6 +70,9 @@ const char* PCapLibNames[] =
 #ifdef __WIN32__
    // TODO: name for npcap in non-WinPCap mode
    "wpcap.dll",
+#elif defined(__APPLE__)
+    "libpcap.A.dylib",
+    "libpcap.dylib",
 #else
    // Linux lib names
    "libpcap.so.1",
@ -276,6 +283,7 @@ bool Init(bool open_adapter)
                struct sockaddr_in* sa = (sockaddr_in*)curaddr->ifa_addr;
                memcpy(adata->IP_v4, &sa->sin_addr, 4);
            }
+            #ifdef __linux__
            else if (af == AF_PACKET)
            {
                struct sockaddr_ll* sa = (sockaddr_ll*)curaddr->ifa_addr;
@ -284,7 +292,16 @@ bool Init(bool open_adapter)
                else
                    memcpy(adata->MAC, sa->sll_addr, 6);
            }
-
+            #else
+            else if (af == AF_LINK)
+            {
+                struct sockaddr_dl* sa = (sockaddr_dl*)curaddr->ifa_addr;
+                if (sa->sdl_alen != 6)
+                    printf("weird MAC length %d for %s\n", sa->sdl_alen, curaddr->ifa_name);
+                else
+                    memcpy(adata->MAC, LLADDR(sa), 6);
+            }
+            #endif
            curaddr = curaddr->ifa_next;
        }
    }
--- a/src/frontend/qt_sdl/Platform.cpp
+++ b/src/frontend/qt_sdl/Platform.cpp
@ -23,6 +23,7 @@
 #include <QDir>
 #include <QThread>
 #include <QSemaphore>
+#include <QMutex>
 #include <QOpenGLContext>

 #include "Platform.h"
@ -187,53 +188,77 @@ FILE* OpenLocalFile(const char* path, const char* mode)
    return OpenFile(fullpath.toUtf8(), mode, mode[0] != 'w');
 }

-void* Thread_Create(void (* func)())
+Thread* Thread_Create(void (* func)())
 {
    QThread* t = QThread::create(func);
    t->start();
-    return (void*) t;
+    return (Thread*) t;
 }

-void Thread_Free(void* thread)
+void Thread_Free(Thread* thread)
 {
    QThread* t = (QThread*) thread;
    t->terminate();
    delete t;
 }

-void Thread_Wait(void* thread)
+void Thread_Wait(Thread* thread)
 {
    ((QThread*) thread)->wait();
 }


-void* Semaphore_Create()
+Semaphore* Semaphore_Create()
 {
-    return new QSemaphore();
+    return (Semaphore*)new QSemaphore();
 }

-void Semaphore_Free(void* sema)
+void Semaphore_Free(Semaphore* sema)
 {
    delete (QSemaphore*) sema;
 }

-void Semaphore_Reset(void* sema)
+void Semaphore_Reset(Semaphore* sema)
 {
    QSemaphore* s = (QSemaphore*) sema;

    s->acquire(s->available());
 }

-void Semaphore_Wait(void* sema)
+void Semaphore_Wait(Semaphore* sema)
 {
    ((QSemaphore*) sema)->acquire();
 }

-void Semaphore_Post(void* sema)
+void Semaphore_Post(Semaphore* sema, int count)
 {
-    ((QSemaphore*) sema)->release();
+    ((QSemaphore*) sema)->release(count);
 }

+Mutex* Mutex_Create()
+{
+    return (Mutex*)new QMutex();
+}
+
+void Mutex_Free(Mutex* mutex)
+{
+    delete (QMutex*) mutex;
+}
+
+void Mutex_Lock(Mutex* mutex)
+{
+    ((QMutex*) mutex)->lock();
+}
+
+void Mutex_Unlock(Mutex* mutex)
+{
+    ((QMutex*) mutex)->unlock();
+}
+
+bool Mutex_TryLock(Mutex* mutex)
+{
+    return ((QMutex*) mutex)->try_lock();
+}

 void* GL_GetProcAddress(const char* proc)
 {
--- a/src/frontend/qt_sdl/PlatformConfig.cpp
+++ b/src/frontend/qt_sdl/PlatformConfig.cpp
@ -120,7 +120,7 @@ ConfigEntry PlatformConfigFile[] =
    {"HKJoy_Reset",               0, &HKJoyMapping[HK_Reset],               -1, NULL, 0},
    {"HKJoy_FastForward",         0, &HKJoyMapping[HK_FastForward],         -1, NULL, 0},
    {"HKJoy_FastForwardToggle",   0, &HKJoyMapping[HK_FastForwardToggle],   -1, NULL, 0},
-    {"HKJoy_FastForwardToggle",   0, &HKJoyMapping[HK_FullscreenToggle],    -1, NULL, 0},
+    {"HKJoy_FullscreenToggle",    0, &HKJoyMapping[HK_FullscreenToggle],    -1, NULL, 0},
    {"HKJoy_SolarSensorDecrease", 0, &HKJoyMapping[HK_SolarSensorDecrease], -1, NULL, 0},
    {"HKJoy_SolarSensorIncrease", 0, &HKJoyMapping[HK_SolarSensorIncrease], -1, NULL, 0},

--- a/src/frontend/qt_sdl/WifiSettingsDialog.cpp
+++ b/src/frontend/qt_sdl/WifiSettingsDialog.cpp
@ -54,7 +54,7 @@ WifiSettingsDialog::WifiSettingsDialog(QWidget* parent) : QDialog(parent), ui(ne
    LAN_Socket::Init();
    haspcap = LAN_PCap::Init(false);

-    ui->cbDirectMode->setText("Direct mode (requires " PCAP_NAME " and ethernet connection)");
+    ui->rbDirectMode->setText("Direct mode (requires " PCAP_NAME " and ethernet connection)");

    ui->cbBindAnyAddr->setChecked(Config::SocketBindAnyAddr != 0);
    ui->cbRandomizeMAC->setChecked(Config::RandomizeMAC != 0);
@ -71,8 +71,9 @@ WifiSettingsDialog::WifiSettingsDialog(QWidget* parent) : QDialog(parent), ui(ne
    }
    ui->cbxDirectAdapter->setCurrentIndex(sel);

-    ui->cbDirectMode->setChecked(Config::DirectLAN != 0);
-    if (!haspcap) ui->cbDirectMode->setEnabled(false);
+    ui->rbDirectMode->setChecked(Config::DirectLAN != 0);
+    ui->rbIndirectMode->setChecked(Config::DirectLAN == 0);
+    if (!haspcap) ui->rbDirectMode->setEnabled(false);

    updateAdapterControls();
 }
@ -101,7 +102,7 @@ void WifiSettingsDialog::done(int r)

        Config::SocketBindAnyAddr = ui->cbBindAnyAddr->isChecked() ? 1:0;
        Config::RandomizeMAC = randommac;
-        Config::DirectLAN = ui->cbDirectMode->isChecked() ? 1:0;
+        Config::DirectLAN = ui->rbDirectMode->isChecked() ? 1:0;

        int sel = ui->cbxDirectAdapter->currentIndex();
        if (sel < 0 || sel >= LAN_PCap::NumAdapters) sel = 0;
@ -125,11 +126,14 @@ void WifiSettingsDialog::done(int r)
    closeDlg();
 }

-void WifiSettingsDialog::on_cbDirectMode_stateChanged(int state)
+void WifiSettingsDialog::on_rbDirectMode_clicked()
+{
+    updateAdapterControls();
+}
+void WifiSettingsDialog::on_rbIndirectMode_clicked()
 {
    updateAdapterControls();
 }
-
 void WifiSettingsDialog::on_cbxDirectAdapter_currentIndexChanged(int sel)
 {
    if (!haspcap) return;
@ -153,7 +157,7 @@ void WifiSettingsDialog::on_cbxDirectAdapter_currentIndexChanged(int sel)

 void WifiSettingsDialog::updateAdapterControls()
 {
-    bool enable = haspcap && ui->cbDirectMode->isChecked();
+    bool enable = haspcap && ui->rbDirectMode->isChecked();

    ui->cbxDirectAdapter->setEnabled(enable);
    ui->lblAdapterMAC->setEnabled(enable);
--- a/src/frontend/qt_sdl/WifiSettingsDialog.h
+++ b/src/frontend/qt_sdl/WifiSettingsDialog.h
@ -55,7 +55,8 @@ public:
 private slots:
    void done(int r);

-    void on_cbDirectMode_stateChanged(int state);
+    void on_rbDirectMode_clicked();
+    void on_rbIndirectMode_clicked();
    void on_cbxDirectAdapter_currentIndexChanged(int sel);

 private:
--- a/src/frontend/qt_sdl/WifiSettingsDialog.ui
+++ b/src/frontend/qt_sdl/WifiSettingsDialog.ui
@ -6,8 +6,8 @@
   <rect>
    <x>0</x>
    <y>0</y>
-    <width>479</width>
-    <height>240</height>
+    <width>572</width>
+    <height>296</height>
   </rect>
  </property>
  <property name="sizePolicy">
@ -58,15 +58,81 @@
      <string>Online</string>
     </property>
     <layout class="QGridLayout" name="gridLayout_2">
-      <item row="2" column="0">
-       <widget class="QLabel" name="label_2">
+      <item row="3" column="0" rowspan="3" colspan="2">
+       <widget class="QGroupBox" name="groupBox_3">
+        <property name="title">
+         <string>Direct Mode Settings</string>
+        </property>
+        <layout class="QGridLayout" name="gridLayout_3">
+         <item row="0" column="0">
+          <widget class="QLabel" name="label">
+           <property name="text">
+            <string>Network adapter:</string>
+           </property>
+          </widget>
+         </item>
+         <item row="0" column="1">
+          <widget class="QComboBox" name="cbxDirectAdapter">
+           <property name="sizePolicy">
+            <sizepolicy hsizetype="Expanding" vsizetype="Fixed">
+             <horstretch>0</horstretch>
+             <verstretch>0</verstretch>
+            </sizepolicy>
+           </property>
+           <property name="minimumSize">
+            <size>
+             <width>300</width>
+             <height>0</height>
+            </size>
+           </property>
+           <property name="whatsThis">
+            <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;Selects the network adapter through which to route network traffic under direct mode.&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
+           </property>
+          </widget>
+         </item>
+         <item row="1" column="0">
+          <widget class="QLabel" name="label_2">
+           <property name="text">
+            <string>MAC address:</string>
+           </property>
+          </widget>
+         </item>
+         <item row="1" column="1">
+          <widget class="QLabel" name="lblAdapterMAC">
+           <property name="text">
+            <string>[PLACEHOLDER]</string>
+           </property>
+          </widget>
+         </item>
+         <item row="2" column="0">
+          <widget class="QLabel" name="label_3">
+           <property name="text">
+            <string>IP address:</string>
+           </property>
+          </widget>
+         </item>
+         <item row="2" column="1">
+          <widget class="QLabel" name="lblAdapterIP">
+           <property name="text">
+            <string>[PLACEHOLDER]</string>
+           </property>
+          </widget>
+         </item>
+        </layout>
+       </widget>
+      </item>
+      <item row="1" column="0">
+       <widget class="QRadioButton" name="rbIndirectMode">
+        <property name="whatsThis">
+         <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;Indirect mode uses libslirp. It requires no extra setup and is easy to use.&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
+        </property>
        <property name="text">
-         <string>MAC address:</string>
+         <string>Indirect Mode (uses libslirp, recommended)</string>
        </property>
       </widget>
      </item>
-      <item row="0" column="0" colspan="2">
-       <widget class="QCheckBox" name="cbDirectMode">
+      <item row="2" column="0">
+       <widget class="QRadioButton" name="rbDirectMode">
        <property name="whatsThis">
         <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;Direct mode directly routes network traffic to the host network. It is the most reliable, but requires an ethernet connection.&lt;/p&gt;&lt;p&gt;&lt;br/&gt;&lt;/p&gt;&lt;p&gt;Non-direct mode uses a layer of emulation to get around this, but is more prone to problems.&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
        </property>
@ -75,53 +141,6 @@
        </property>
       </widget>
      </item>
-      <item row="1" column="1">
-       <widget class="QComboBox" name="cbxDirectAdapter">
-        <property name="sizePolicy">
-         <sizepolicy hsizetype="Expanding" vsizetype="Fixed">
-          <horstretch>0</horstretch>
-          <verstretch>0</verstretch>
-         </sizepolicy>
-        </property>
-        <property name="minimumSize">
-         <size>
-          <width>350</width>
-          <height>0</height>
-         </size>
-        </property>
-        <property name="whatsThis">
-         <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;Selects the network adapter through which to route network traffic under direct mode.&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
-        </property>
-       </widget>
-      </item>
-      <item row="1" column="0">
-       <widget class="QLabel" name="label">
-        <property name="text">
-         <string>Network adapter:</string>
-        </property>
-       </widget>
-      </item>
-      <item row="3" column="0">
-       <widget class="QLabel" name="label_3">
-        <property name="text">
-         <string>IP address:</string>
-        </property>
-       </widget>
-      </item>
-      <item row="2" column="1">
-       <widget class="QLabel" name="lblAdapterMAC">
-        <property name="text">
-         <string>[PLACEHOLDER]</string>
-        </property>
-       </widget>
-      </item>
-      <item row="3" column="1">
-       <widget class="QLabel" name="lblAdapterIP">
-        <property name="text">
-         <string>[PLACEHOLDER]</string>
-        </property>
-       </widget>
-      </item>
     </layout>
    </widget>
   </item>
--- a/src/frontend/qt_sdl/main.cpp
+++ b/src/frontend/qt_sdl/main.cpp
@ -355,10 +355,10 @@ void EmuThread::run()
    Input::Init();

    u32 nframes = 0;
-    u32 starttick = SDL_GetTicks();
-    u32 lasttick = starttick;
-    u32 lastmeasuretick = lasttick;
-    u32 fpslimitcount = 0;
+    double perfCountsSec = 1.0 / SDL_GetPerformanceFrequency();
+    double lastTime = SDL_GetPerformanceCounter() * perfCountsSec;
+    double frameLimitError = 0.0;
+    double lastMeasureTime = lastTime;

    char melontitle[100];

@ -492,49 +492,43 @@ void EmuThread::run()
                SDL_UnlockMutex(audioSyncLock);
            }

-            float framerate = (1000.0f * nlines) / (60.0f * 263.0f);
+            double frametimeStep = nlines / (60.0 * 263.0);

            {
-                u32 curtick = SDL_GetTicks();
-                u32 delay = curtick - lasttick;
-
                bool limitfps = Config::LimitFPS && !fastforward;
-                if (limitfps)
-                {
-                    float wantedtickF = starttick + (framerate * (fpslimitcount+1));
-                    u32 wantedtick = (u32)ceil(wantedtickF);
-                    if (curtick < wantedtick) SDL_Delay(wantedtick - curtick);

-                    lasttick = SDL_GetTicks();
-                    fpslimitcount++;
-                    if ((abs(wantedtickF - (float)wantedtick) < 0.001312) || (fpslimitcount > 60))
-                    {
-                        fpslimitcount = 0;
-                        starttick = lasttick;
-                    }
-                }
-                else
+                double practicalFramelimit = limitfps ? frametimeStep : 1.0 / 1000.0;
+
+                double curtime = SDL_GetPerformanceCounter() * perfCountsSec;
+
+                frameLimitError += practicalFramelimit - (curtime - lastTime);
+                if (frameLimitError < -practicalFramelimit)
+                    frameLimitError = -practicalFramelimit;
+                if (frameLimitError > practicalFramelimit)
+                    frameLimitError = practicalFramelimit;
+                
+                if (round(frameLimitError * 1000.0) > 0.0)
                {
-                    if (delay < 1) SDL_Delay(1);
-                    lasttick = SDL_GetTicks();
+                    SDL_Delay(round(frameLimitError * 1000.0));
+                    double timeBeforeSleep = curtime;
+                    curtime = SDL_GetPerformanceCounter() * perfCountsSec;
+                    frameLimitError -= curtime - timeBeforeSleep;
                }
+
+                lastTime = curtime;
            }

            nframes++;
            if (nframes >= 30)
            {
-                u32 tick = SDL_GetTicks();
-                u32 diff = tick - lastmeasuretick;
-                lastmeasuretick = tick;
+                double time = SDL_GetPerformanceCounter() * perfCountsSec;
+                double dt = time - lastMeasureTime;
+                lastMeasureTime = time;

-                u32 fps;
-                if (diff < 1) fps = 77777;
-                else fps = (nframes * 1000) / diff;
+                u32 fps = round(nframes / dt);
                nframes = 0;

-                float fpstarget;
-                if (framerate < 1) fpstarget = 999;
-                else fpstarget = 1000.0f/framerate;
+                float fpstarget = 1.0/frametimeStep;

                sprintf(melontitle, "[%d/%.0f] melonDS " MELONDS_VERSION, fps, fpstarget);
                changeWindowTitle(melontitle);
@ -544,10 +538,8 @@ void EmuThread::run()
        {
            // paused
            nframes = 0;
-            lasttick = SDL_GetTicks();
-            starttick = lasttick;
-            lastmeasuretick = lasttick;
-            fpslimitcount = 0;
+            lastTime = SDL_GetPerformanceCounter() * perfCountsSec;
+            lastMeasureTime = lastTime;

            emit windowUpdate();

@ -1354,7 +1346,7 @@ void MainWindow::dragEnterEvent(QDragEnterEvent* event)
    QString filename = urls.at(0).toLocalFile();
    QString ext = filename.right(3);

-    if (ext == "nds" || ext == "srl" || ext == "dsi" || (ext == "gba" && RunningSomething))
+    if (ext == "nds" || ext == "srl" || ext == "dsi" || ext == "gba")
        event->acceptProposedAction();
 }