Multiinstance jit

src/ARMJIT.cpp

@@ -30,6 +30,7 @@
 #include "ARMJIT_Internal.h"
 #include "ARMJIT_Memory.h"
 #include "ARMJIT_Compiler.h"
+#include "ARMJIT_Global.h"
 
 #include "ARMInterpreter_ALU.h"
 #include "ARMInterpreter_LoadStore.h"
@@ -467,6 +468,16 @@ InterpreterFunc InterpretTHUMB[ARMInstrInfo::tk_Count] =
 };
 #undef F
 
+ARMJIT::ARMJIT(melonDS::NDS& nds, std::optional<JITArgs> jit) noexcept : 
+        NDS(nds),
+        Memory(nds),
+        JITCompiler(nds),
+        MaxBlockSize(jit.has_value() ? std::clamp(jit->MaxBlockSize, 1u, 32u) : 32),
+        LiteralOptimizations(jit.has_value() ? jit->LiteralOptimizations : false),
+        BranchOptimizations(jit.has_value() ? jit->BranchOptimizations : false),
+        FastMemory((jit.has_value() ? jit->FastMemory : false) && ARMJIT_Memory::IsFastMemSupported())
+{}
+
 void ARMJIT::RetireJitBlock(JitBlock* block) noexcept
 {
     auto it = RestoreCandidates.find(block->InstrHash);
@@ -483,6 +494,7 @@ void ARMJIT::RetireJitBlock(JitBlock* block) noexcept
 
 void ARMJIT::SetJITArgs(JITArgs args) noexcept
 {
+    args.FastMemory = args.FastMemory && ARMJIT_Memory::IsFastMemSupported();
     args.MaxBlockSize = std::clamp(args.MaxBlockSize, 1u, 32u);
 
     if (MaxBlockSize != args.MaxBlockSize
@@ -499,36 +511,22 @@ void ARMJIT::SetJITArgs(JITArgs args) noexcept
 
 void ARMJIT::SetMaxBlockSize(int size) noexcept
 {
-    size = std::clamp(size, 1, 32);
-
-    if (size != MaxBlockSize)
-        ResetBlockCache();
-
-    MaxBlockSize = size;
+    SetJITArgs(JITArgs{static_cast<unsigned>(size), LiteralOptimizations, LiteralOptimizations, FastMemory});
 }
 
 void ARMJIT::SetLiteralOptimizations(bool enabled) noexcept
 {
-    if (LiteralOptimizations != enabled)
-        ResetBlockCache();
-
-    LiteralOptimizations = enabled;
+    SetJITArgs(JITArgs{static_cast<unsigned>(MaxBlockSize), enabled, BranchOptimizations, FastMemory});
 }
 
 void ARMJIT::SetBranchOptimizations(bool enabled) noexcept
 {
-    if (BranchOptimizations != enabled)
-        ResetBlockCache();
-
-    BranchOptimizations = enabled;
+    SetJITArgs(JITArgs{static_cast<unsigned>(MaxBlockSize), LiteralOptimizations, enabled, FastMemory});
 }
 
 void ARMJIT::SetFastMemory(bool enabled) noexcept
 {
-    if (FastMemory != enabled)
-        ResetBlockCache();
-
-    FastMemory = enabled;
+    SetJITArgs(JITArgs{static_cast<unsigned>(MaxBlockSize), LiteralOptimizations, BranchOptimizations, enabled});
 }
 
 void ARMJIT::CompileBlock(ARM* cpu) noexcept
@@ -918,7 +916,7 @@ void ARMJIT::CompileBlock(ARM* cpu) noexcept
 
         AddressRange* region = CodeMemRegions[addressRanges[j] >> 27];
 
-        if (!PageContainsCode(&region[(addressRanges[j] & 0x7FFF000) / 512]))
+        if (!PageContainsCode(&region[(addressRanges[j] & 0x7FFF000 & ~(Memory.PageSize - 1)) / 512], Memory.PageSize))
             Memory.SetCodeProtection(addressRanges[j] >> 27, addressRanges[j] & 0x7FFFFFF, true);
 
         AddressRange* range = &region[(addressRanges[j] & 0x7FFFFFF) / 512];
@@ -971,7 +969,7 @@ void ARMJIT::InvalidateByAddr(u32 localAddr) noexcept
         range->Blocks.Remove(i);
 
         if (range->Blocks.Length == 0
-            && !PageContainsCode(&region[(localAddr & 0x7FFF000) / 512]))
+            && !PageContainsCode(&region[(localAddr & 0x7FFF000 & ~(Memory.PageSize - 1)) / 512], Memory.PageSize))
         {
             Memory.SetCodeProtection(localAddr >> 27, localAddr & 0x7FFFFFF, false);
         }
@@ -1005,7 +1003,7 @@ void ARMJIT::InvalidateByAddr(u32 localAddr) noexcept
 
                 if (otherRange->Blocks.Length == 0)
                 {
-                    if (!PageContainsCode(&otherRegion[(addr & 0x7FFF000) / 512]))
+                    if (!PageContainsCode(&otherRegion[(addr & 0x7FFF000 & ~(Memory.PageSize - 1)) / 512], Memory.PageSize))
                         Memory.SetCodeProtection(addr >> 27, addr & 0x7FFFFFF, false);
 
                     otherRange->Code = 0;

src/ARMJIT.h

@@ -44,15 +44,7 @@ class JitBlock;
 class ARMJIT
 {
 public:
-    ARMJIT(melonDS::NDS& nds, std::optional<JITArgs> jit) noexcept :
-        NDS(nds),
-        Memory(nds),
-        JITCompiler(nds),
-        MaxBlockSize(jit.has_value() ? std::clamp(jit->MaxBlockSize, 1u, 32u) : 32),
-        LiteralOptimizations(jit.has_value() ? jit->LiteralOptimizations : false),
-        BranchOptimizations(jit.has_value() ? jit->BranchOptimizations : false),
-        FastMemory(jit.has_value() ? jit->FastMemory : false)
-    {}
+    ARMJIT(melonDS::NDS& nds, std::optional<JITArgs> jit) noexcept;
     ~ARMJIT() noexcept;
     void InvalidateByAddr(u32) noexcept;
     void CheckAndInvalidateWVRAM(int) noexcept;
@@ -80,6 +72,7 @@ class ARMJIT
     bool LiteralOptimizations = false;
     bool BranchOptimizations = false;
     bool FastMemory = false;
+
 public:
     melonDS::NDS& NDS;
     TinyVector<u32> InvalidLiterals {};

src/ARMJIT_A64/ARMJIT_Compiler.cpp

@@ -22,17 +22,7 @@
 #include "../ARMInterpreter.h"
 #include "../ARMJIT.h"
 #include "../NDS.h"
-
-#if defined(__SWITCH__)
-#include <switch.h>
-
-extern char __start__;
-#elif defined(_WIN32)
-#include <windows.h>
-#else
-#include <sys/mman.h>
-#include <unistd.h>
-#endif
+#include "../ARMJIT_Global.h"
 
 #include <stdlib.h>
 
@@ -66,11 +56,6 @@ const int RegisterCache<Compiler, ARM64Reg>::NativeRegsAvailable = 15;
 
 const BitSet32 CallerSavedPushRegs({W8, W9, W10, W11, W12, W13, W14, W15});
 
-const int JitMemSize = 16 * 1024 * 1024;
-#ifndef __SWITCH__
-u8 JitMem[JitMemSize];
-#endif
-
 void Compiler::MovePC()
 {
     ADD(MapReg(15), MapReg(15), Thumb ? 2 : 4);
@@ -260,29 +245,12 @@ Compiler::Compiler(melonDS::NDS& nds) : Arm64Gen::ARM64XEmitter(), NDS(nds)
     SetCodeBase((u8*)JitRWStart, (u8*)JitRXStart);
     JitMemMainSize = JitMemSize;
 #else
-    #ifdef _WIN32
-        SYSTEM_INFO sysInfo;
-        GetSystemInfo(&sysInfo);
-
-        u64 pageSize = (u64)sysInfo.dwPageSize;
-    #else
-        u64 pageSize = sysconf(_SC_PAGE_SIZE);
-    #endif
-    u8* pageAligned = (u8*)(((u64)JitMem & ~(pageSize - 1)) + pageSize);
-    u64 alignedSize = (((u64)JitMem + sizeof(JitMem)) & ~(pageSize - 1)) - (u64)pageAligned;
-
-    #if defined(_WIN32)
-        DWORD dummy;
-        VirtualProtect(pageAligned, alignedSize, PAGE_EXECUTE_READWRITE, &dummy);
-    #elif defined(__APPLE__)
-        pageAligned = (u8*)mmap(NULL, 1024*1024*16, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS | MAP_JIT,-1, 0);
-        nds.JIT.JitEnableWrite();
-    #else
-        mprotect(pageAligned, alignedSize, PROT_EXEC | PROT_READ | PROT_WRITE);
-    #endif
-
-    SetCodeBase(pageAligned, pageAligned);
-    JitMemMainSize = alignedSize;
+    ARMJIT_Global::Init();
+
+    CodeMemBase = ARMJIT_Global::AllocateCodeMem();
+
+    SetCodeBase(reinterpret_cast<u8*>(CodeMemBase), reinterpret_cast<u8*>(CodeMemBase));
+    JitMemMainSize = ARMJIT_Global::CodeMemorySliceSize;
 #endif
     SetCodePtr(0);
 
@@ -493,6 +461,9 @@ Compiler::~Compiler()
         free(JitRWBase);
     }
 #endif
+
+    ARMJIT_Global::FreeCodeMem(CodeMemBase);
+    ARMJIT_Global::DeInit();
 }
 
 void Compiler::LoadCycles()

src/ARMJIT_A64/ARMJIT_Compiler.h

@@ -275,6 +275,7 @@ class Compiler : public Arm64Gen::ARM64XEmitter
     void* JitRWStart;
     void* JitRXStart;
 #endif
+    void* CodeMemBase;
 
     void* ReadBanked, *WriteBanked;
 

src/ARMJIT_Global.cpp

@@ -0,0 +1,118 @@
+#include "ARMJIT_Global.h"
+#include "ARMJIT_Memory.h"
+
+#ifdef _WIN32
+#include <windows.h>
+#else
+#include <sys/mman.h>
+#include <unistd.h>
+#endif
+
+#include <stdio.h>
+#include <stdint.h>
+
+#include <mutex>
+
+namespace melonDS
+{
+
+namespace ARMJIT_Global
+{
+
+std::mutex globalMutex;
+
+#ifndef __APPLE__
+static constexpr size_t NumCodeMemSlices = 4;
+static constexpr size_t CodeMemoryAlignedSize = NumCodeMemSlices * CodeMemorySliceSize;
+
+// I haven't heard of pages larger than 16 KB
+u8 CodeMemory[CodeMemoryAlignedSize + 16*1024];
+
+u32 AvailableCodeMemSlices = (1 << NumCodeMemSlices) - 1;
+
+u8* GetAlignedCodeMemoryStart()
+{
+    return reinterpret_cast<u8*>((reinterpret_cast<intptr_t>(CodeMemory) + (16*1024-1)) & ~static_cast<intptr_t>(16*1024-1));
+}
+#endif
+
+int RefCounter = 0;
+
+void* AllocateCodeMem()
+{
+    std::lock_guard guard(globalMutex);
+
+#ifndef __APPLE__
+    if (AvailableCodeMemSlices)
+    {
+        int slice = __builtin_ctz(AvailableCodeMemSlices);
+        AvailableCodeMemSlices &= ~(1 << slice);
+        //printf("allocating slice %d\n", slice);
+        return &GetAlignedCodeMemoryStart()[slice * CodeMemorySliceSize];
+    }
+#endif
+
+    // allocate
+#ifdef _WIN32
+    return VirtualAlloc(nullptr, CodeMemorySliceSize, MEM_RESERVE|MEM_COMMIT, PAGE_EXECUTE_READWRITE);
+#else
+    //printf("mmaping...\n");
+    return mmap(nullptr, CodeMemorySliceSize, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+#endif
+}
+
+void FreeCodeMem(void* codeMem)
+{
+    std::lock_guard guard(globalMutex);
+
+    for (int i = 0; i < NumCodeMemSlices; i++)
+    {
+        if (codeMem == &GetAlignedCodeMemoryStart()[CodeMemorySliceSize * i])
+        {
+            //printf("freeing slice\n");
+            AvailableCodeMemSlices |= 1 << i;
+            return;
+        }
+    }
+
+#ifdef _WIN32
+    VirtualFree(codeMem, CodeMemorySliceSize, MEM_RELEASE|MEM_DECOMMIT);
+#else
+    munmap(codeMem, CodeMemorySliceSize);
+#endif
+}
+
+void Init()
+{
+    std::lock_guard guard(globalMutex);
+
+    RefCounter++;
+    if (RefCounter == 1)
+    {
+        #ifdef _WIN32
+            DWORD dummy;
+            VirtualProtect(GetAlignedCodeMemoryStart(), CodeMemoryAlignedSize, PAGE_EXECUTE_READWRITE, &dummy);
+        #elif defined(__APPLE__)
+            // Apple always uses dynamic allocation
+        #else
+            mprotect(GetAlignedCodeMemoryStart(), CodeMemoryAlignedSize, PROT_EXEC | PROT_READ | PROT_WRITE);
+        #endif
+
+        ARMJIT_Memory::RegisterFaultHandler();
+    }
+}
+
+void DeInit()
+{
+    std::lock_guard guard(globalMutex);
+
+    RefCounter--;
+    if (RefCounter == 0)
+    {
+        ARMJIT_Memory::UnregisterFaultHandler();
+    }
+}
+
+}
+
+}

src/ARMJIT_Global.h

@@ -0,0 +1,44 @@
+/*
+    Copyright 2016-2024 melonDS team
+
+    This file is part of melonDS.
+
+    melonDS is free software: you can redistribute it and/or modify it under
+    the terms of the GNU General Public License as published by the Free
+    Software Foundation, either version 3 of the License, or (at your option)
+    any later version.
+
+    melonDS is distributed in the hope that it will be useful, but WITHOUT ANY
+    WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+    FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with melonDS. If not, see http://www.gnu.org/licenses/.
+*/
+
+#ifndef ARMJIT_GLOBAL_H
+#define ARMJIT_GLOBAL_H
+
+#include "types.h"
+
+#include <stdlib.h>
+
+namespace melonDS
+{
+
+namespace ARMJIT_Global
+{
+
+static constexpr size_t CodeMemorySliceSize = 1024*1024*32;
+
+void Init();
+void DeInit();
+
+void* AllocateCodeMem();
+void FreeCodeMem(void* codeMem);
+
+}
+
+}
+
+#endif

src/ARMJIT_Internal.h

@@ -85,9 +85,9 @@ typedef void (*InterpreterFunc)(ARM* cpu);
 extern InterpreterFunc InterpretARM[];
 extern InterpreterFunc InterpretTHUMB[];
 
-inline bool PageContainsCode(const AddressRange* range)
+inline bool PageContainsCode(const AddressRange* range, u32 pageSize)
 {
-    for (int i = 0; i < 8; i++)
+    for (int i = 0; i < pageSize / 512; i++)
     {
         if (range[i].Blocks.Length > 0)
             return true;