Merge pull request #7602 from CrystalGamma/stop-g_jit

stop using g_jit outside of JitInterface

Source/Core/Core/PowerPC/Jit64/JitAsm.cpp

@@ -18,7 +18,7 @@
 
 using namespace Gen;
 
-Jit64AsmRoutineManager::Jit64AsmRoutineManager(JitBase& jit) : m_jit{jit}
+Jit64AsmRoutineManager::Jit64AsmRoutineManager(Jitx86Base& jit) : m_jit{jit}, CommonAsmRoutines(jit)
 {
 }
 

Source/Core/Core/PowerPC/Jit64/JitAsm.h

@@ -35,7 +35,7 @@ public:
   // want to ensure this number is big enough.
   static constexpr size_t CODE_SIZE = 16384;
 
-  explicit Jit64AsmRoutineManager(JitBase& jit);
+  explicit Jit64AsmRoutineManager(Jitx86Base& jit);
 
   void Init(u8* stack_top);
 

Source/Core/Core/PowerPC/Jit64Common/EmuCodeBlock.cpp

@@ -54,12 +54,13 @@ void EmuCodeBlock::MemoryExceptionCheck()
   // load/store, the trampoline generator will have stashed the exception
   // handler (that we previously generated after the fastmem instruction) in
   // trampolineExceptionHandler.
-  if (g_jit->js.generatingTrampoline)
+  auto& js = m_jit.js;
+  if (js.generatingTrampoline)
   {
-    if (g_jit->js.trampolineExceptionHandler)
+    if (js.trampolineExceptionHandler)
     {
       TEST(32, PPCSTATE(Exceptions), Gen::Imm32(EXCEPTION_DSI));
-      J_CC(CC_NZ, g_jit->js.trampolineExceptionHandler);
+      J_CC(CC_NZ, js.trampolineExceptionHandler);
     }
     return;
   }
@@ -67,11 +68,11 @@ void EmuCodeBlock::MemoryExceptionCheck()
   // If memcheck (ie: MMU) mode is enabled and we haven't generated an
   // exception handler for this instruction yet, we will generate an
   // exception check.
-  if (g_jit->jo.memcheck && !g_jit->js.fastmemLoadStore && !g_jit->js.fixupExceptionHandler)
+  if (m_jit.jo.memcheck && !js.fastmemLoadStore && !js.fixupExceptionHandler)
   {
     TEST(32, PPCSTATE(Exceptions), Gen::Imm32(EXCEPTION_DSI));
-    g_jit->js.exceptionHandler = J_CC(Gen::CC_NZ, true);
-    g_jit->js.fixupExceptionHandler = true;
+    js.exceptionHandler = J_CC(Gen::CC_NZ, true);
+    js.fixupExceptionHandler = true;
   }
 }
 
@@ -318,8 +319,9 @@ void EmuCodeBlock::SafeLoadToReg(X64Reg reg_value, const Gen::OpArg& opAddress,
 {
   bool slowmem = (flags & SAFE_LOADSTORE_FORCE_SLOWMEM) != 0;
 
+  auto& js = m_jit.js;
   registersInUse[reg_value] = false;
-  if (g_jit->jo.fastmem && !(flags & (SAFE_LOADSTORE_NO_FASTMEM | SAFE_LOADSTORE_NO_UPDATE_PC)) &&
+  if (m_jit.jo.fastmem && !(flags & (SAFE_LOADSTORE_NO_FASTMEM | SAFE_LOADSTORE_NO_UPDATE_PC)) &&
       !slowmem)
   {
     u8* backpatchStart = GetWritableCodePtr();
@@ -327,7 +329,7 @@ void EmuCodeBlock::SafeLoadToReg(X64Reg reg_value, const Gen::OpArg& opAddress,
     bool offsetAddedToAddress =
         UnsafeLoadToReg(reg_value, opAddress, accessSize, offset, signExtend, &mov);
     TrampolineInfo& info = m_back_patch_info[mov.address];
-    info.pc = g_jit->js.compilerPC;
+    info.pc = js.compilerPC;
     info.nonAtomicSwapStoreSrc = mov.nonAtomicSwapStore ? mov.nonAtomicSwapStoreSrc : INVALID_REG;
     info.start = backpatchStart;
     info.read = true;
@@ -346,7 +348,7 @@ void EmuCodeBlock::SafeLoadToReg(X64Reg reg_value, const Gen::OpArg& opAddress,
     }
     info.len = static_cast<u32>(GetCodePtr() - info.start);
 
-    g_jit->js.fastmemLoadStore = mov.address;
+    js.fastmemLoadStore = mov.address;
     return;
   }
 
@@ -384,7 +386,7 @@ void EmuCodeBlock::SafeLoadToReg(X64Reg reg_value, const Gen::OpArg& opAddress,
   // Invalid for calls from Jit64AsmCommon routines
   if (!(flags & SAFE_LOADSTORE_NO_UPDATE_PC))
   {
-    MOV(32, PPCSTATE(pc), Imm32(g_jit->js.compilerPC));
+    MOV(32, PPCSTATE(pc), Imm32(js.compilerPC));
   }
 
   size_t rsp_alignment = (flags & SAFE_LOADSTORE_NO_PROLOG) ? 8 : 0;
@@ -448,7 +450,7 @@ void EmuCodeBlock::SafeLoadToRegImmediate(X64Reg reg_value, u32 address, int acc
   }
 
   // Helps external systems know which instruction triggered the read.
-  MOV(32, PPCSTATE(pc), Imm32(g_jit->js.compilerPC));
+  MOV(32, PPCSTATE(pc), Imm32(m_jit.js.compilerPC));
 
   // Fall back to general-case code.
   ABI_PushRegistersAndAdjustStack(registersInUse, 0);
@@ -490,14 +492,15 @@ void EmuCodeBlock::SafeWriteRegToReg(OpArg reg_value, X64Reg reg_addr, int acces
   // set the correct immediate format
   reg_value = FixImmediate(accessSize, reg_value);
 
-  if (g_jit->jo.fastmem && !(flags & (SAFE_LOADSTORE_NO_FASTMEM | SAFE_LOADSTORE_NO_UPDATE_PC)) &&
+  auto& js = m_jit.js;
+  if (m_jit.jo.fastmem && !(flags & (SAFE_LOADSTORE_NO_FASTMEM | SAFE_LOADSTORE_NO_UPDATE_PC)) &&
       !slowmem)
   {
     u8* backpatchStart = GetWritableCodePtr();
     MovInfo mov;
     UnsafeWriteRegToReg(reg_value, reg_addr, accessSize, offset, swap, &mov);
     TrampolineInfo& info = m_back_patch_info[mov.address];
-    info.pc = g_jit->js.compilerPC;
+    info.pc = js.compilerPC;
     info.nonAtomicSwapStoreSrc = mov.nonAtomicSwapStore ? mov.nonAtomicSwapStoreSrc : INVALID_REG;
     info.start = backpatchStart;
     info.read = false;
@@ -515,7 +518,7 @@ void EmuCodeBlock::SafeWriteRegToReg(OpArg reg_value, X64Reg reg_addr, int acces
     }
     info.len = static_cast<u32>(GetCodePtr() - info.start);
 
-    g_jit->js.fastmemLoadStore = mov.address;
+    js.fastmemLoadStore = mov.address;
 
     return;
   }
@@ -551,7 +554,7 @@ void EmuCodeBlock::SafeWriteRegToReg(OpArg reg_value, X64Reg reg_addr, int acces
   // Invalid for calls from Jit64AsmCommon routines
   if (!(flags & SAFE_LOADSTORE_NO_UPDATE_PC))
   {
-    MOV(32, PPCSTATE(pc), Imm32(g_jit->js.compilerPC));
+    MOV(32, PPCSTATE(pc), Imm32(js.compilerPC));
   }
 
   size_t rsp_alignment = (flags & SAFE_LOADSTORE_NO_PROLOG) ? 8 : 0;
@@ -617,7 +620,7 @@ bool EmuCodeBlock::WriteToConstAddress(int accessSize, OpArg arg, u32 address,
 
   // If we already know the address through constant folding, we can do some
   // fun tricks...
-  if (g_jit->jo.optimizeGatherPipe && PowerPC::IsOptimizableGatherPipeWrite(address))
+  if (m_jit.jo.optimizeGatherPipe && PowerPC::IsOptimizableGatherPipeWrite(address))
   {
     X64Reg arg_reg = RSCRATCH;
 
@@ -634,7 +637,7 @@ bool EmuCodeBlock::WriteToConstAddress(int accessSize, OpArg arg, u32 address,
     ADD(64, R(RSCRATCH2), Imm8(accessSize >> 3));
     MOV(64, PPCSTATE(gather_pipe_ptr), R(RSCRATCH2));
 
-    g_jit->js.fifoBytesSinceCheck += accessSize >> 3;
+    m_jit.js.fifoBytesSinceCheck += accessSize >> 3;
     return false;
   }
   else if (PowerPC::IsOptimizableRAMAddress(address))
@@ -645,7 +648,7 @@ bool EmuCodeBlock::WriteToConstAddress(int accessSize, OpArg arg, u32 address,
   else
   {
     // Helps external systems know which instruction triggered the write
-    MOV(32, PPCSTATE(pc), Imm32(g_jit->js.compilerPC));
+    MOV(32, PPCSTATE(pc), Imm32(m_jit.js.compilerPC));
 
     ABI_PushRegistersAndAdjustStack(registersInUse, 0);
     switch (accessSize)
@@ -724,7 +727,7 @@ void EmuCodeBlock::ForceSinglePrecision(X64Reg output, const OpArg& input, bool
                                         bool duplicate)
 {
   // Most games don't need these. Zelda requires it though - some platforms get stuck without them.
-  if (g_jit->jo.accurateSinglePrecision)
+  if (m_jit.jo.accurateSinglePrecision)
   {
     if (packed)
     {
@@ -840,7 +843,7 @@ alignas(16) static const u64 psRoundBit[2] = {0x8000000, 0x8000000};
 // It needs a temp, so let the caller pass that in.
 void EmuCodeBlock::Force25BitPrecision(X64Reg output, const OpArg& input, X64Reg tmp)
 {
-  if (g_jit->jo.accurateSinglePrecision)
+  if (m_jit.jo.accurateSinglePrecision)
   {
     // mantissa = (mantissa & ~0xFFFFFFF) + ((mantissa & (1ULL << 27)) << 1);
     if (input.IsSimpleReg() && cpu_info.bAVX)

Source/Core/Core/PowerPC/Jit64Common/EmuCodeBlock.h

@@ -19,10 +19,13 @@ namespace MMIO
 class Mapping;
 }
 
+class Jitx86Base;
+
 // Like XCodeBlock but has some utilities for memory access.
 class EmuCodeBlock : public Gen::X64CodeBlock
 {
 public:
+  explicit EmuCodeBlock(Jitx86Base& jit) : m_jit{jit} {}
   void MemoryExceptionCheck();
 
   // Simple functions to switch between near and far code emitting
@@ -125,6 +128,7 @@ public:
   void Clear();
 
 protected:
+  Jitx86Base& m_jit;
   ConstantPool m_const_pool;
   FarCodeCache m_far_code;
   u8* m_near_code;  // Backed up when we switch to far code.

Source/Core/Core/PowerPC/Jit64Common/Jit64AsmCommon.cpp

@@ -505,7 +505,7 @@ void QuantizedMemoryRoutines::GenQuantizedLoad(bool single, EQuantizeType type,
 
   bool extend = single && (type == QUANTIZE_S8 || type == QUANTIZE_S16);
 
-  if (g_jit->jo.memcheck)
+  if (m_jit.jo.memcheck)
   {
     BitSet32 regsToSave = QUANTIZED_REGS_TO_SAVE_LOAD;
     int flags = isInline ? 0 :
@@ -632,7 +632,7 @@ void QuantizedMemoryRoutines::GenQuantizedLoadFloat(bool single, bool isInline)
   int size = single ? 32 : 64;
   bool extend = false;
 
-  if (g_jit->jo.memcheck)
+  if (m_jit.jo.memcheck)
   {
     BitSet32 regsToSave = QUANTIZED_REGS_TO_SAVE;
     int flags = isInline ? 0 :
@@ -643,7 +643,7 @@ void QuantizedMemoryRoutines::GenQuantizedLoadFloat(bool single, bool isInline)
 
   if (single)
   {
-    if (g_jit->jo.memcheck)
+    if (m_jit.jo.memcheck)
     {
       MOVD_xmm(XMM0, R(RSCRATCH_EXTRA));
     }
@@ -668,7 +668,7 @@ void QuantizedMemoryRoutines::GenQuantizedLoadFloat(bool single, bool isInline)
     // for a good reason, or merely because no game does this.
     // If we find something that actually does do this, maybe this should be changed. How
     // much of a performance hit would it be?
-    if (g_jit->jo.memcheck)
+    if (m_jit.jo.memcheck)
     {
       ROL(64, R(RSCRATCH_EXTRA), Imm8(32));
       MOVQ_xmm(XMM0, R(RSCRATCH_EXTRA));

Source/Core/Core/PowerPC/Jit64Common/Jit64AsmCommon.h

@@ -13,6 +13,7 @@ enum EQuantizeType : u32;
 class QuantizedMemoryRoutines : public EmuCodeBlock
 {
 public:
+  explicit QuantizedMemoryRoutines(Jitx86Base& jit) : EmuCodeBlock(jit) {}
   void GenQuantizedLoad(bool single, EQuantizeType type, int quantize);
   void GenQuantizedStore(bool single, EQuantizeType type, int quantize);
 
@@ -24,6 +25,7 @@ private:
 class CommonAsmRoutines : public CommonAsmRoutinesBase, public QuantizedMemoryRoutines
 {
 public:
+  explicit CommonAsmRoutines(Jitx86Base& jit) : QuantizedMemoryRoutines(jit) {}
   void GenFrsqrte();
   void GenFres();
   void GenMfcr();

Source/Core/Core/PowerPC/Jit64Common/Jit64Base.h

@@ -32,14 +32,15 @@ constexpr size_t CODE_SIZE = 1024 * 1024 * 32;
 
 class Jitx86Base : public JitBase, public QuantizedMemoryRoutines
 {
+public:
+  Jitx86Base() : QuantizedMemoryRoutines(*this) {}
+  JitBlockCache* GetBlockCache() override { return &blocks; }
+  bool HandleFault(uintptr_t access_address, SContext* ctx) override;
+
 protected:
   bool BackPatch(u32 emAddress, SContext* ctx);
   JitBlockCache blocks{*this};
-  TrampolineCache trampolines;
-
-public:
-  JitBlockCache* GetBlockCache() override { return &blocks; }
-  bool HandleFault(uintptr_t access_address, SContext* ctx) override;
+  TrampolineCache trampolines{*this};
 };
 
 void LogGeneratedX86(size_t size, const PPCAnalyst::CodeBuffer& code_buffer, const u8* normalEntry,

Source/Core/Core/PowerPC/Jit64Common/TrampolineCache.h

@@ -24,6 +24,7 @@ class TrampolineCache : public EmuCodeBlock
   const u8* GenerateWriteTrampoline(const TrampolineInfo& info);
 
 public:
+  explicit TrampolineCache(Jitx86Base& jit) : EmuCodeBlock(jit) {}
   const u8* GenerateTrampoline(const TrampolineInfo& info);
   void ClearCodeSpace();
 };

Source/Core/Core/PowerPC/JitCommon/JitBase.cpp

@@ -10,8 +10,6 @@
 #include "Core/PowerPC/PPCAnalyst.h"
 #include "Core/PowerPC/PowerPC.h"
 
-JitBase* g_jit;
-
 const u8* JitBase::Dispatch(JitBase& jit)
 {
   return jit.GetBlockCache()->Dispatch();

Source/Core/Core/PowerPC/JitCommon/JitBase.h

@@ -39,10 +39,6 @@
 #define JITDISABLE(setting)                                                                        \
   FALLBACK_IF(SConfig::GetInstance().bJITOff || SConfig::GetInstance().setting)
 
-class JitBase;
-
-extern JitBase* g_jit;
-
 class JitBase : public CPUCoreBase
 {
 protected:

Source/Core/Core/PowerPC/JitInterface.cpp

@@ -40,6 +40,11 @@
 
 namespace JitInterface
 {
+static JitBase* g_jit = nullptr;
+void SetJit(JitBase* jit)
+{
+  g_jit = jit;
+}
 void DoState(PointerWrap& p)
 {
   if (g_jit && p.GetMode() == PointerWrap::MODE_READ)

Source/Core/Core/PowerPC/JitInterface.h

@@ -11,6 +11,7 @@
 
 class CPUCoreBase;
 class PointerWrap;
+class JitBase;
 
 namespace PowerPC
 {
@@ -65,5 +66,8 @@ void InvalidateICache(u32 address, u32 size, bool forced);
 
 void CompileExceptionCheck(ExceptionType type);
 
+/// used for the page fault unit test, don't use outside of tests!
+void SetJit(JitBase* jit);
+
 void Shutdown();
 }

Source/UnitTests/Core/PageFaultTest.cpp

@@ -8,6 +8,7 @@
 #include "Common/Timer.h"
 #include "Core/MemTools.h"
 #include "Core/PowerPC/JitCommon/JitBase.h"
+#include "Core/PowerPC/JitInterface.h"
 
 // include order is important
 #include <gtest/gtest.h>  // NOLINT
@@ -56,7 +57,7 @@ TEST(PageFault, PageFault)
   Common::WriteProtectMemory(data, PAGE_GRAN, false);
 
   PageFaultFakeJit pfjit;
-  g_jit = &pfjit;
+  JitInterface::SetJit(&pfjit);
   pfjit.m_data = data;
 
   auto start = std::chrono::high_resolution_clock::now();
@@ -67,7 +68,7 @@ TEST(PageFault, PageFault)
   ((unsigned long long)std::chrono::duration_cast<std::chrono::nanoseconds>(diff).count())
 
   EMM::UninstallExceptionHandler();
-  g_jit = nullptr;
+  JitInterface::SetJit(nullptr);
 
   printf("page fault timing:\n");
   printf("start->HandleFault     %llu ns\n", AS_NS(pfjit.m_pre_unprotect_time - start));

Source/UnitTests/Core/PowerPC/Jit64Common/Frsqrte.cpp

@@ -10,8 +10,8 @@
 #include "Common/FloatUtils.h"
 #include "Common/x64ABI.h"
 #include "Core/PowerPC/Gekko.h"
+#include "Core/PowerPC/Jit64/Jit.h"
 #include "Core/PowerPC/Jit64Common/Jit64AsmCommon.h"
-#include "Core/PowerPC/Jit64Common/Jit64Base.h"
 #include "Core/PowerPC/Jit64Common/Jit64PowerPCState.h"
 
 #include <gtest/gtest.h>
@@ -19,7 +19,7 @@
 class TestCommonAsmRoutines : public CommonAsmRoutines
 {
 public:
-  TestCommonAsmRoutines()
+  TestCommonAsmRoutines() : CommonAsmRoutines(jit)
   {
     using namespace Gen;
 
@@ -49,6 +49,7 @@ public:
   }
 
   u64 (*wrapped_frsqrte)(u64, UReg_FPSCR&);
+  Jit64 jit;
 };
 
 TEST(Jit64, Frsqrte)