Multiinstance jit (#2201)

* works on Linux x64
still needs to be fixed for everything else

* use lots of PROT_NONE memory to reliably reserve virtual address space

* multi instance fastmem on Linux

* Windows

* blarg

* disable fastmem if the page size is not 4kb

* fix fast mem dialog option

* make aarch64 work as well

* fastmem 16kb pages support

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();
+    }
+}
+
+}
+
+}