ppsspp/Core/MIPS/x86/Jit.cpp

// Copyright (c) 2012- PPSSPP Project.

// This program 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, version 2.0 or later versions.

// This program 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 2.0 for more details.

// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/

// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.

#include "ppsspp_config.h"
#if PPSSPP_ARCH(X86) || PPSSPP_ARCH(AMD64)

#include <algorithm>
#include <iterator>

#include "Common/Math/math_util.h"
#include "Common/Profiler/Profiler.h"

#include "Common/Serialize/Serializer.h"
#include "Common/Serialize/SerializeFuncs.h"
#include "Core/Core.h"
#include "Core/MemMap.h"
#include "Core/System.h"
#include "Core/CoreTiming.h"
#include "Core/Config.h"
#include "Core/Reporting.h"
#include "Core/Debugger/SymbolMap.h"
#include "Core/MIPS/MIPS.h"
#include "Core/MIPS/MIPSAnalyst.h"
#include "Core/MIPS/MIPSCodeUtils.h"
#include "Core/MIPS/MIPSInt.h"
#include "Core/MIPS/MIPSTables.h"
#include "Core/HLE/ReplaceTables.h"

#include "RegCache.h"
#include "Jit.h"

#include "Core/Debugger/Breakpoints.h"

namespace MIPSComp
{
using namespace Gen;

const bool USE_JIT_MISSMAP = false;
static std::map<std::string, u32> notJitOps;

template<typename A, typename B>
std::pair<B,A> flip_pair(const std::pair<A,B> &p) {
	return std::pair<B, A>(p.second, p.first);
}

// This is called when Jit hits a breakpoint.  Returns 1 when hit.
u32 JitBreakpoint(uint32_t addr)
{
	// Should we skip this breakpoint?
	if (CBreakPoints::CheckSkipFirst() == currentMIPS->pc || CBreakPoints::CheckSkipFirst() == addr)
		return 0;

	BreakAction result = CBreakPoints::ExecBreakPoint(addr);
	if ((result & BREAK_ACTION_PAUSE) == 0)
		return 0;

	// There's probably a better place for this.
	if (USE_JIT_MISSMAP) {
		std::map<u32, std::string> notJitSorted;
		std::transform(notJitOps.begin(), notJitOps.end(), std::inserter(notJitSorted, notJitSorted.begin()), flip_pair<std::string, u32>);

		std::string message;
		char temp[256];
		int remaining = 15;
		for (auto it = notJitSorted.rbegin(), end = notJitSorted.rend(); it != end && --remaining >= 0; ++it)
		{
			snprintf(temp, 256, " (%d), ", it->first);
			message += it->second + temp;
		}

		if (message.size() > 2)
			message.resize(message.size() - 2);

		NOTICE_LOG(JIT, "Top ops compiled to interpreter: %s", message.c_str());
	}

	return 1;
}

static u32 JitMemCheck(u32 addr, u32 pc) {
	// Should we skip this breakpoint?
	if (CBreakPoints::CheckSkipFirst() == currentMIPS->pc)
		return 0;

	// Did we already hit one?
	if (coreState != CORE_RUNNING && coreState != CORE_NEXTFRAME)
		return 1;

	// Note: pc may be the delay slot.
	CBreakPoints::ExecOpMemCheck(addr, pc);
	return coreState == CORE_RUNNING || coreState == CORE_NEXTFRAME ? 0 : 1;
}

static void JitLogMiss(MIPSOpcode op)
{
	if (USE_JIT_MISSMAP)
		notJitOps[MIPSGetName(op)]++;

	MIPSInterpretFunc func = MIPSGetInterpretFunc(op);
	func(op);
}

#ifdef _MSC_VER
// JitBlockCache doesn't use this, just stores it.
#pragma warning(disable:4355)
#endif
Jit::Jit(MIPSState *mipsState)
		: blocks(mipsState, this), mips_(mipsState) {
	blocks.Init();
	gpr.SetEmitter(this);
	fpr.SetEmitter(this);
	AllocCodeSpace(1024 * 1024 * 16);
	GenerateFixedCode(jo);

	safeMemFuncs.Init(&thunks);

	js.startDefaultPrefix = mips_->HasDefaultPrefix();

	// The debugger sets this so that "go" on a breakpoint will actually... go.
	// But if they reset, we can end up hitting it by mistake, since it's based on PC and ticks.
	CBreakPoints::SetSkipFirst(0);
}

Jit::~Jit() {
}

void Jit::DoState(PointerWrap &p) {
	auto s = p.Section("Jit", 1, 2);
	if (!s)
		return;

	Do(p, js.startDefaultPrefix);
	if (p.mode == PointerWrap::MODE_READ && !js.startDefaultPrefix) {
		WARN_LOG(CPU, "Jit: An uneaten prefix was previously detected. Jitting in unknown-prefix mode.");
	}
	if (s >= 2) {
		Do(p, js.hasSetRounding);
		if (p.mode == PointerWrap::MODE_READ) {
			js.lastSetRounding = 0;
		}
	} else {
		js.hasSetRounding = 1;
	}

	// The debugger sets this so that "go" on a breakpoint will actually... go.
	// But if they load a state, we can end up hitting it by mistake, since it's based on PC and ticks.
	CBreakPoints::SetSkipFirst(0);
}

void Jit::UpdateFCR31() {
}

void Jit::GetStateAndFlushAll(RegCacheState &state) {
	gpr.GetState(state.gpr);
	fpr.GetState(state.fpr);
	FlushAll();
}

void Jit::RestoreState(const RegCacheState& state) {
	gpr.RestoreState(state.gpr);
	fpr.RestoreState(state.fpr);
}

void Jit::FlushAll() {
	gpr.Flush();
	fpr.Flush();
	FlushPrefixV();
}

void Jit::FlushPrefixV() {
	if (js.startDefaultPrefix && !js.blockWrotePrefixes && js.HasNoPrefix()) {
		// They started default, we never modified in memory, and they're default now.
		// No reason to modify memory.  This is common at end of blocks.  Just clear dirty.
		js.prefixSFlag = (JitState::PrefixState)(js.prefixSFlag & ~JitState::PREFIX_DIRTY);
		js.prefixTFlag = (JitState::PrefixState)(js.prefixTFlag & ~JitState::PREFIX_DIRTY);
		js.prefixDFlag = (JitState::PrefixState)(js.prefixDFlag & ~JitState::PREFIX_DIRTY);
		return;
	}

	if ((js.prefixSFlag & JitState::PREFIX_DIRTY) != 0) {
		MOV(32, MIPSSTATE_VAR(vfpuCtrl[VFPU_CTRL_SPREFIX]), Imm32(js.prefixS));
		js.prefixSFlag = (JitState::PrefixState) (js.prefixSFlag & ~JitState::PREFIX_DIRTY);
	}

	if ((js.prefixTFlag & JitState::PREFIX_DIRTY) != 0) {
		MOV(32, MIPSSTATE_VAR(vfpuCtrl[VFPU_CTRL_TPREFIX]), Imm32(js.prefixT));
		js.prefixTFlag = (JitState::PrefixState) (js.prefixTFlag & ~JitState::PREFIX_DIRTY);
	}

	if ((js.prefixDFlag & JitState::PREFIX_DIRTY) != 0) {
		MOV(32, MIPSSTATE_VAR(vfpuCtrl[VFPU_CTRL_DPREFIX]), Imm32(js.prefixD));
		js.prefixDFlag = (JitState::PrefixState) (js.prefixDFlag & ~JitState::PREFIX_DIRTY);
	}

	// If we got here, we must've written prefixes to memory in this block.
	js.blockWrotePrefixes = true;
}

void Jit::WriteDowncount(int offset) {
	const int downcount = js.downcountAmount + offset;
	SUB(32, MIPSSTATE_VAR(downcount), downcount > 127 ? Imm32(downcount) : Imm8(downcount));
}

void Jit::RestoreRoundingMode(bool force) {
	// If the game has never set an interesting rounding mode, we can safely skip this.
	if (force || js.hasSetRounding) {
		CALL(restoreRoundingMode);
	}
}

void Jit::ApplyRoundingMode(bool force) {
	// If the game has never set an interesting rounding mode, we can safely skip this.
	if (force || js.hasSetRounding) {
		CALL(applyRoundingMode);
	}
}

void Jit::UpdateRoundingMode(u32 fcr31) {
	// We must set js.hasSetRounding at compile time, or this block will use the wrong rounding mode.
	// The fcr31 parameter is -1 when not known at compile time, so we just assume it was changed.
	if (fcr31 & 0x01000003) {
		js.hasSetRounding = true;
	}
}

void Jit::ClearCache()
{
	blocks.Clear();
	ClearCodeSpace(0);
	GenerateFixedCode(jo);
}

void Jit::SaveFlags() {
	PUSHF();
#if PPSSPP_ARCH(AMD64)
	// On X64, the above misaligns the stack. However there might be a cheaper solution than this.
	POP(64, R(EAX));
	MOV(64, MIPSSTATE_VAR(saved_flags), R(EAX));
#endif
}

void Jit::LoadFlags() {
#if PPSSPP_ARCH(AMD64)
	MOV(64, R(EAX), MIPSSTATE_VAR(saved_flags));
	PUSH(64, R(EAX));
#endif
	POPF();
}

void Jit::CompileDelaySlot(int flags, RegCacheState *state) {
	// Need to offset the downcount which was already incremented for the branch + delay slot.
	CheckJitBreakpoint(GetCompilerPC() + 4, -2);

	if (flags & DELAYSLOT_SAFE)
		SaveFlags(); // preserve flag around the delay slot!

	js.inDelaySlot = true;
	MIPSOpcode op = GetOffsetInstruction(1);
	MIPSCompileOp(op, this);
	js.inDelaySlot = false;

	if (flags & DELAYSLOT_FLUSH) {
		if (state != NULL)
			GetStateAndFlushAll(*state);
		else
			FlushAll();
	}
	if (flags & DELAYSLOT_SAFE)
		LoadFlags(); // restore flag!
}

void Jit::EatInstruction(MIPSOpcode op) {
	MIPSInfo info = MIPSGetInfo(op);
	if (info & DELAYSLOT) {
		ERROR_LOG_REPORT_ONCE(ateDelaySlot, JIT, "Ate a branch op.");
	}
	if (js.inDelaySlot) {
		ERROR_LOG_REPORT_ONCE(ateInDelaySlot, JIT, "Ate an instruction inside a delay slot.");
	}

	CheckJitBreakpoint(GetCompilerPC() + 4, 0);
	js.numInstructions++;
	js.compilerPC += 4;
	js.downcountAmount += MIPSGetInstructionCycleEstimate(op);
}

void Jit::Compile(u32 em_address) {
	PROFILE_THIS_SCOPE("jitc");
	if (GetSpaceLeft() < 0x10000 || blocks.IsFull()) {
		ClearCache();
	}

	if (!Memory::IsValidAddress(em_address) || (em_address & 3) != 0) {
		Core_ExecException(em_address, em_address, ExecExceptionType::JUMP);
		return;
	}

	// Sometimes we compile fairly large blocks, although it's uncommon.
	BeginWrite(JitBlockCache::MAX_BLOCK_INSTRUCTIONS * 16);

	int block_num = blocks.AllocateBlock(em_address);
	JitBlock *b = blocks.GetBlock(block_num);
	DoJit(em_address, b);
	_assert_msg_(b->originalAddress == em_address, "original %08x != em_address %08x (block %d)", b->originalAddress, em_address, b->blockNum);
	blocks.FinalizeBlock(block_num, jo.enableBlocklink);

	EndWrite();

	bool cleanSlate = false;

	if (js.hasSetRounding && !js.lastSetRounding) {
		WARN_LOG(JIT, "Detected rounding mode usage, rebuilding jit with checks");
		// Won't loop, since hasSetRounding is only ever set to 1.
		js.lastSetRounding = js.hasSetRounding;
		cleanSlate = true;
	}

	// Drat.  The VFPU hit an uneaten prefix at the end of a block.
	if (js.startDefaultPrefix && js.MayHavePrefix()) {
		WARN_LOG_REPORT(JIT, "An uneaten prefix at end of block: %08x", GetCompilerPC() - 4);
		js.LogPrefix();

		// Let's try that one more time.  We won't get back here because we toggled the value.
		js.startDefaultPrefix = false;
		cleanSlate = true;
	}

	if (cleanSlate) {
		// Our assumptions are all wrong so it's clean-slate time.
		ClearCache();
		Compile(em_address);
	}
}

void Jit::RunLoopUntil(u64 globalticks) {
	PROFILE_THIS_SCOPE("jit");
	((void (*)())enterDispatcher)();
}

u32 Jit::GetCompilerPC() {
	return js.compilerPC;
}

MIPSOpcode Jit::GetOffsetInstruction(int offset) {
	return Memory::Read_Instruction(GetCompilerPC() + 4 * offset);
}

const u8 *Jit::DoJit(u32 em_address, JitBlock *b) {
	js.cancel = false;
	js.blockStart = em_address;
	js.compilerPC = em_address;
	js.lastContinuedPC = 0;
	js.initialBlockSize = 0;
	js.nextExit = 0;
	js.downcountAmount = 0;
	js.curBlock = b;
	js.compiling = true;
	js.inDelaySlot = false;
	js.blockWrotePrefixes = false;
	js.afterOp = JitState::AFTER_NONE;
	js.PrefixStart();

	// We add a check before the block, used when entering from a linked block.
	b->checkedEntry = GetCodePtr();
	// Downcount flag check. The last block decremented downcounter, and the flag should still be available.
	FixupBranch skip = J_CC(CC_NS);
	MOV(32, MIPSSTATE_VAR(pc), Imm32(js.blockStart));
	JMP(outerLoop, true);  // downcount hit zero - go advance.
	SetJumpTarget(skip);

	b->normalEntry = GetCodePtr();

	MIPSAnalyst::AnalysisResults analysis = MIPSAnalyst::Analyze(em_address);

	gpr.Start(mips_, &js, &jo, analysis);
	fpr.Start(mips_, &js, &jo, analysis, RipAccessible(&mips_->v[0]));

	js.numInstructions = 0;
	while (js.compiling) {
		// Jit breakpoints are quite fast, so let's do them in release too.
		CheckJitBreakpoint(GetCompilerPC(), 0);

		MIPSOpcode inst = Memory::Read_Opcode_JIT(GetCompilerPC());
		js.downcountAmount += MIPSGetInstructionCycleEstimate(inst);

		MIPSCompileOp(inst, this);

		if (js.afterOp & JitState::AFTER_CORE_STATE) {
			// CORE_RUNNING is <= CORE_NEXTFRAME.
			if (RipAccessible((const void *)&coreState)) {
				CMP(32, M(&coreState), Imm32(CORE_NEXTFRAME));  // rip accessible
			} else {
				MOV(PTRBITS, R(RAX), ImmPtr((const void *)&coreState));
				CMP(32, MatR(RAX), Imm32(CORE_NEXTFRAME));
			}
			FixupBranch skipCheck = J_CC(CC_LE, true);
			// All cases of AFTER_CORE_STATE should update PC.  We don't update here.
			RegCacheState state;
			GetStateAndFlushAll(state);
			WriteSyscallExit();

			SetJumpTarget(skipCheck);
			// If we didn't jump, we can keep our regs as they were.
			RestoreState(state);

			js.afterOp = JitState::AFTER_NONE;
		}

		js.compilerPC += 4;
		js.numInstructions++;

		if (jo.Disabled(JitDisable::REGALLOC_GPR)) {
			gpr.Flush();
		}
		if (jo.Disabled(JitDisable::REGALLOC_FPR)) {
			fpr.Flush();
			FlushPrefixV();
		}

		// Safety check, in case we get a bunch of really large jit ops without a lot of branching.
		if (GetSpaceLeft() < 0x800 || js.numInstructions >= JitBlockCache::MAX_BLOCK_INSTRUCTIONS) {
			FlushAll();
			WriteExit(GetCompilerPC(), js.nextExit++);
			js.compiling = false;
		}
	}

	b->codeSize = (u32)(GetCodePtr() - b->normalEntry);
	NOP();
	AlignCode4();
	if (js.lastContinuedPC == 0) {
		b->originalSize = js.numInstructions;
	} else {
		// We continued at least once.  Add the last proxy and set the originalSize correctly.
		blocks.ProxyBlock(js.blockStart, js.lastContinuedPC, (GetCompilerPC() - js.lastContinuedPC) / sizeof(u32), GetCodePtr());
		b->originalSize = js.initialBlockSize;
	}
	return b->normalEntry;
}

void Jit::AddContinuedBlock(u32 dest) {
	// The first block is the root block.  When we continue, we create proxy blocks after that.
	if (js.lastContinuedPC == 0)
		js.initialBlockSize = js.numInstructions;
	else
		blocks.ProxyBlock(js.blockStart, js.lastContinuedPC, (GetCompilerPC() - js.lastContinuedPC) / sizeof(u32), GetCodePtr());
	js.lastContinuedPC = dest;
}

bool Jit::DescribeCodePtr(const u8 *ptr, std::string &name) {
	if (ptr == applyRoundingMode)
		name = "applyRoundingMode";
	else if (ptr == dispatcher)
		name = "dispatcher";
	else if (ptr == dispatcherInEAXNoCheck)
		name = "dispatcher (PC in EAX)";
	else if (ptr == dispatcherNoCheck)
		name = "dispatcherNoCheck";
	else if (ptr == dispatcherCheckCoreState)
		name = "dispatcherCheckCoreState";
	else if (ptr == enterDispatcher)
		name = "enterDispatcher";
	else if (ptr == restoreRoundingMode)
		name = "restoreRoundingMode";
	else if (ptr == crashHandler)
		name = "crashHandler";
	else {
		u32 jitAddr = blocks.GetAddressFromBlockPtr(ptr);

		// Returns 0 when it's valid, but unknown.
		if (jitAddr == 0) {
			name = "UnknownOrDeletedBlock";
		} else if (jitAddr != (u32)-1) {
			char temp[1024];
			const std::string label = g_symbolMap ? g_symbolMap->GetDescription(jitAddr) : "";
			if (!label.empty())
				snprintf(temp, sizeof(temp), "%08x_%s", jitAddr, label.c_str());
			else
				snprintf(temp, sizeof(temp), "%08x", jitAddr);
			name = temp;
		} else if (IsInSpace(ptr)) {
			if (ptr < endOfPregeneratedCode) {
				name = "PreGenCode";
			} else {
				name = "Unknown";
			}
		} else if (thunks.IsInSpace(ptr)) {
			name = "Thunk";
		} else if (safeMemFuncs.IsInSpace(ptr)) {
			name = "JitSafeMem";
		} else {
			// Not anywhere in jit, then.
			return false;
		}
	}
	// If we got here, one of the above cases matched.
	return true;
}

void Jit::Comp_RunBlock(MIPSOpcode op) {
	// This shouldn't be necessary, the dispatcher should catch us before we get here.
	ERROR_LOG(JIT, "Comp_RunBlock");
}

void Jit::LinkBlock(u8 *exitPoint, const u8 *checkedEntry) {
	if (PlatformIsWXExclusive()) {
		ProtectMemoryPages(exitPoint, 32, MEM_PROT_READ | MEM_PROT_WRITE);
	}
	XEmitter emit(exitPoint);
	// Okay, this is a bit ugly, but we check here if it already has a JMP.
	// That means it doesn't have a full exit to pad with INT 3.
	bool prelinked = *emit.GetCodePointer() == 0xE9;
	emit.JMP(checkedEntry, true);
	if (!prelinked) {
		ptrdiff_t actualSize = emit.GetWritableCodePtr() - exitPoint;
		int pad = JitBlockCache::GetBlockExitSize() - (int)actualSize;
		for (int i = 0; i < pad; ++i) {
			emit.INT3();
		}
	}
	if (PlatformIsWXExclusive()) {
		ProtectMemoryPages(exitPoint, 32, MEM_PROT_READ | MEM_PROT_EXEC);
	}
}

void Jit::UnlinkBlock(u8 *checkedEntry, u32 originalAddress) {
	if (PlatformIsWXExclusive()) {
		ProtectMemoryPages(checkedEntry, 16, MEM_PROT_READ | MEM_PROT_WRITE);
	}
	// Send anyone who tries to run this block back to the dispatcher.
	// Not entirely ideal, but .. pretty good.
	// Spurious entrances from previously linked blocks can only come through checkedEntry
	XEmitter emit(checkedEntry);
	emit.MOV(32, MIPSSTATE_VAR(pc), Imm32(originalAddress));
	emit.JMP(MIPSComp::jit->GetDispatcher(), true);
	if (PlatformIsWXExclusive()) {
		ProtectMemoryPages(checkedEntry, 16, MEM_PROT_READ | MEM_PROT_EXEC);
	}
}

bool Jit::ReplaceJalTo(u32 dest) {
	const ReplacementTableEntry *entry = nullptr;
	u32 funcSize = 0;
	if (!CanReplaceJalTo(dest, &entry, &funcSize)) {
		return false;
	}

	// Warning - this might be bad if the code at the destination changes...
	if (entry->flags & REPFLAG_ALLOWINLINE) {
		// Jackpot! Just do it, no flushing. The code will be entirely inlined.

		// First, compile the delay slot. It's unconditional so no issues.
		CompileDelaySlot(DELAYSLOT_NICE);
		// Technically, we should write the unused return address to RA, but meh.
		MIPSReplaceFunc repl = entry->jitReplaceFunc;
		int cycles = (this->*repl)();
		js.downcountAmount += cycles;
	} else {
		gpr.SetImm(MIPS_REG_RA, GetCompilerPC() + 8);
		CompileDelaySlot(DELAYSLOT_NICE);
		FlushAll();
		MOV(32, MIPSSTATE_VAR(pc), Imm32(GetCompilerPC()));
		RestoreRoundingMode();
		ABI_CallFunction(entry->replaceFunc);
		SUB(32, MIPSSTATE_VAR(downcount), R(EAX));
		ApplyRoundingMode();
	}

	js.compilerPC += 4;
	// No writing exits, keep going!

	if (CBreakPoints::HasMemChecks()) {
		// We could modify coreState, so we need to write PC and check.
		// Otherwise, PC may end up on the jal.  We add 4 to skip the delay slot.
		MOV(32, MIPSSTATE_VAR(pc), Imm32(GetCompilerPC() + 4));
		js.afterOp |= JitState::AFTER_CORE_STATE;
	}

	// Add a trigger so that if the inlined code changes, we invalidate this block.
	blocks.ProxyBlock(js.blockStart, dest, funcSize / sizeof(u32), GetCodePtr());
	return true;
}

void Jit::Comp_ReplacementFunc(MIPSOpcode op) {
	// We get here if we execute the first instruction of a replaced function. This means
	// that we do need to return to RA.

	// Inlined function calls (caught in jal) are handled differently.

	int index = op.encoding & MIPS_EMUHACK_VALUE_MASK;

	const ReplacementTableEntry *entry = GetReplacementFunc(index);
	if (!entry) {
		ERROR_LOG_REPORT_ONCE(replFunc, HLE, "Invalid replacement op %08x at %08x", op.encoding, js.compilerPC);
		return;
	}

	u32 funcSize = g_symbolMap->GetFunctionSize(GetCompilerPC());
	bool disabled = (entry->flags & REPFLAG_DISABLED) != 0;
	if (!disabled && funcSize != SymbolMap::INVALID_ADDRESS && funcSize > sizeof(u32)) {
		// We don't need to disable hooks, the code will still run.
		if ((entry->flags & (REPFLAG_HOOKENTER | REPFLAG_HOOKEXIT)) == 0) {
			// Any breakpoint at the func entry was already tripped, so we can still run the replacement.
			// That's a common case - just to see how often the replacement hits.
			disabled = CBreakPoints::RangeContainsBreakPoint(GetCompilerPC() + sizeof(u32), funcSize - sizeof(u32));
		}
	}

	// Hack for old savestates: Avoid stack overflow (MIPSCompileOp/CompReplacementFunc)
	// Not sure about the cause.
	Memory::Opcode origInstruction = Memory::Read_Instruction(GetCompilerPC(), true);
	if (origInstruction.encoding == op.encoding) {
		ERROR_LOG(HLE, "Replacement broken (savestate problem?): %08x at %08x", op.encoding, GetCompilerPC());
		return;
	}

	if (disabled) {
		MIPSCompileOp(origInstruction, this);
	} else if (entry->jitReplaceFunc) {
		MIPSReplaceFunc repl = entry->jitReplaceFunc;
		int cycles = (this->*repl)();

		if (entry->flags & (REPFLAG_HOOKENTER | REPFLAG_HOOKEXIT)) {
			// Compile the original instruction at this address.  We ignore cycles for hooks.
			MIPSCompileOp(origInstruction, this);
		} else {
			FlushAll();
			MOV(32, R(ECX), MIPSSTATE_VAR(r[MIPS_REG_RA]));
			js.downcountAmount += cycles;
			WriteExitDestInReg(ECX);
			js.compiling = false;
		}
	} else if (entry->replaceFunc) {
		FlushAll();

		// Standard function call, nothing fancy.
		// The function returns the number of cycles it took in EAX.
		MOV(32, MIPSSTATE_VAR(pc), Imm32(GetCompilerPC()));
		RestoreRoundingMode();
		ABI_CallFunction(entry->replaceFunc);

		if (entry->flags & (REPFLAG_HOOKENTER | REPFLAG_HOOKEXIT)) {
			// Compile the original instruction at this address.  We ignore cycles for hooks.
			ApplyRoundingMode();
			MIPSCompileOp(Memory::Read_Instruction(GetCompilerPC(), true), this);
		} else {
			CMP(32, R(EAX), Imm32(0));
			FixupBranch positive = J_CC(CC_GE);

			MOV(32, R(ECX), MIPSSTATE_VAR(pc));
			ADD(32, MIPSSTATE_VAR(downcount), R(EAX));
			FixupBranch done = J();

			SetJumpTarget(positive);
			MOV(32, R(ECX), MIPSSTATE_VAR(r[MIPS_REG_RA]));
			SUB(32, MIPSSTATE_VAR(downcount), R(EAX));

			SetJumpTarget(done);
			ApplyRoundingMode();
			// Need to set flags again, ApplyRoundingMode destroyed them (and EAX.)
			SUB(32, MIPSSTATE_VAR(downcount), Imm8(0));
			WriteExitDestInReg(ECX);
			js.compiling = false;
		}
	} else {
		ERROR_LOG(HLE, "Replacement function %s has neither jit nor regular impl", entry->name);
	}
}

void Jit::Comp_Generic(MIPSOpcode op) {
	FlushAll();
	MIPSInterpretFunc func = MIPSGetInterpretFunc(op);
	_dbg_assert_msg_((MIPSGetInfo(op) & DELAYSLOT) == 0, "Cannot use interpreter for branch ops.");

	if (func)
	{
		// TODO: Maybe we'd be better off keeping the rounding mode within interp?
		RestoreRoundingMode();
		MOV(32, MIPSSTATE_VAR(pc), Imm32(GetCompilerPC()));
		if (USE_JIT_MISSMAP)
			ABI_CallFunctionC(&JitLogMiss, op.encoding);
		else
			ABI_CallFunctionC(func, op.encoding);
		ApplyRoundingMode();
	}
	else
		ERROR_LOG_REPORT(JIT, "Trying to compile instruction %08x that can't be interpreted", op.encoding);

	const MIPSInfo info = MIPSGetInfo(op);
	if ((info & IS_VFPU) != 0 && (info & VFPU_NO_PREFIX) == 0)
	{
		// If it does eat them, it'll happen in MIPSCompileOp().
		if ((info & OUT_EAT_PREFIX) == 0)
			js.PrefixUnknown();

		// Even if DISABLE'd, we want to set this flag so we overwrite.
		if ((info & OUT_VFPU_PREFIX) != 0)
			js.blockWrotePrefixes = true;
	}
}

static void HitInvalidBranch(uint32_t dest) {
	Core_ExecException(dest, currentMIPS->pc, ExecExceptionType::JUMP);
}

void Jit::WriteExit(u32 destination, int exit_num) {
	_assert_msg_(exit_num < MAX_JIT_BLOCK_EXITS, "Expected a valid exit_num. dest=%08x", destination);

	if (!Memory::IsValidAddress(destination) || (destination & 3) != 0) {
		ERROR_LOG_REPORT(JIT, "Trying to write block exit to illegal destination %08x: pc = %08x", destination, currentMIPS->pc);
		MOV(32, MIPSSTATE_VAR(pc), Imm32(GetCompilerPC()));
		ABI_CallFunctionC(&HitInvalidBranch, destination);
		js.afterOp |= JitState::AFTER_CORE_STATE;
	}
	// If we need to verify coreState, we may not jump yet.
	if (js.afterOp & JitState::AFTER_CORE_STATE) {
		// CORE_RUNNING is <= CORE_NEXTFRAME.
		if (RipAccessible((const void *)&coreState)) {
			CMP(32, M(&coreState), Imm32(CORE_NEXTFRAME));  // rip accessible
		} else {
			MOV(PTRBITS, R(RAX), ImmPtr((const void *)&coreState));
			CMP(32, MatR(RAX), Imm32(CORE_NEXTFRAME));
		}
		FixupBranch skipCheck = J_CC(CC_LE);
		// All cases of AFTER_CORE_STATE should update PC.  We don't update here.
		WriteSyscallExit();
		SetJumpTarget(skipCheck);
	}

	WriteDowncount();

	//If nobody has taken care of this yet (this can be removed when all branches are done)
	JitBlock *b = js.curBlock;
	b->exitAddress[exit_num] = destination;
	b->exitPtrs[exit_num] = GetWritableCodePtr();

	// Link opportunity!
	int block = blocks.GetBlockNumberFromStartAddress(destination);
	if (block >= 0 && jo.enableBlocklink) {
		// It exists! Joy of joy!
		JMP(blocks.GetBlock(block)->checkedEntry, true);
		b->linkStatus[exit_num] = true;
	} else {
		// No blocklinking.
		MOV(32, MIPSSTATE_VAR(pc), Imm32(destination));
		JMP(dispatcher, true);

		// Normally, exits are 15 bytes (MOV + &pc + dest + JMP + dest) on 64 or 32 bit.
		// But just in case we somehow optimized, pad.
		ptrdiff_t actualSize = GetWritableCodePtr() - b->exitPtrs[exit_num];
		int pad = JitBlockCache::GetBlockExitSize() - (int)actualSize;
		for (int i = 0; i < pad; ++i) {
			INT3();
		}
	}
}

static u32 IsValidJumpTarget(uint32_t addr) {
	if (Memory::IsValidAddress(addr) && (addr & 3) == 0)
		return 1;
	return 0;
}

static void HitInvalidJumpReg(uint32_t source) {
	Core_ExecException(currentMIPS->pc, source, ExecExceptionType::JUMP);
	currentMIPS->pc = source + 8;
}

void Jit::WriteExitDestInReg(X64Reg reg) {
	// If we need to verify coreState, we may not jump yet.
	if (js.afterOp & JitState::AFTER_CORE_STATE) {
		// CORE_RUNNING is <= CORE_NEXTFRAME.
		if (RipAccessible((const void *)&coreState)) {
			CMP(32, M(&coreState), Imm32(CORE_NEXTFRAME));  // rip accessible
		} else {
			X64Reg temp = reg == RAX ? RDX : RAX;
			MOV(PTRBITS, R(temp), ImmPtr((const void *)&coreState));
			CMP(32, MatR(temp), Imm32(CORE_NEXTFRAME));
		}
		FixupBranch skipCheck = J_CC(CC_LE);
		// All cases of AFTER_CORE_STATE should update PC.  We don't update here.
		WriteSyscallExit();
		SetJumpTarget(skipCheck);
	}

	MOV(32, MIPSSTATE_VAR(pc), R(reg));
	WriteDowncount();

	// Validate the jump to avoid a crash?
	if (!g_Config.bFastMemory) {
		CMP(32, R(reg), Imm32(PSP_GetKernelMemoryBase()));
		FixupBranch tooLow = J_CC(CC_B);
		CMP(32, R(reg), Imm32(PSP_GetUserMemoryEnd()));
		FixupBranch tooHigh = J_CC(CC_AE);

		// Need to set neg flag again.
		SUB(32, MIPSSTATE_VAR(downcount), Imm8(0));
		if (reg == EAX)
			J_CC(CC_NS, dispatcherInEAXNoCheck, true);
		JMP(dispatcher, true);

		SetJumpTarget(tooLow);
		SetJumpTarget(tooHigh);

		ABI_CallFunctionA((const void *)&IsValidJumpTarget, R(reg));

		// If we're ignoring, coreState didn't trip - so trip it now.
		CMP(32, R(EAX), Imm32(0));
		FixupBranch skip = J_CC(CC_NE);
		ABI_CallFunctionC(&HitInvalidJumpReg, GetCompilerPC());
		SetJumpTarget(skip);

		SUB(32, MIPSSTATE_VAR(downcount), Imm8(0));
		JMP(dispatcherCheckCoreState, true);
	} else if (reg == EAX) {
		J_CC(CC_NS, dispatcherInEAXNoCheck, true);
		JMP(dispatcher, true);
	} else {
		JMP(dispatcher, true);
	}
}

void Jit::WriteSyscallExit() {
	WriteDowncount();
	JMP(dispatcherCheckCoreState, true);
}

bool Jit::CheckJitBreakpoint(u32 addr, int downcountOffset) {
	if (CBreakPoints::IsAddressBreakPoint(addr)) {
		SaveFlags();
		FlushAll();
		MOV(32, MIPSSTATE_VAR(pc), Imm32(GetCompilerPC()));
		RestoreRoundingMode();
		ABI_CallFunctionC(&JitBreakpoint, addr);

		// If 0, the conditional breakpoint wasn't taken.
		CMP(32, R(EAX), Imm32(0));
		FixupBranch skip = J_CC(CC_Z);
		WriteDowncount(downcountOffset);
		ApplyRoundingMode();
		// Just to fix the stack.
		LoadFlags();
		JMP(dispatcherCheckCoreState, true);
		SetJumpTarget(skip);

		ApplyRoundingMode();
		LoadFlags();
		return true;
	}

	return false;
}

void Jit::CheckMemoryBreakpoint(int instructionOffset, MIPSGPReg rs, int offset) {
	if (!CBreakPoints::HasMemChecks())
		return;

	int totalInstructionOffset = instructionOffset + (js.inDelaySlot ? 1 : 0);
	uint32_t checkedPC = GetCompilerPC() + totalInstructionOffset * 4;
	int size = MIPSAnalyst::OpMemoryAccessSize(checkedPC);
	bool isWrite = MIPSAnalyst::IsOpMemoryWrite(checkedPC);

	// 0 because we normally execute before increasing.
	int downcountOffset = js.inDelaySlot ? -2 : -1;
	// TODO: In likely branches, downcount will be incorrect.  This might make resume fail.
	if (js.downcountAmount + downcountOffset < 0) {
		downcountOffset = 0;
	}

	if (gpr.IsImm(rs)) {
		uint32_t iaddr = gpr.GetImm(rs) + offset;
		MemCheck check;
		if (CBreakPoints::GetMemCheckInRange(iaddr, size, &check)) {
			if (!(check.cond & MEMCHECK_READ) && !isWrite)
				return;
			if (!(check.cond & MEMCHECK_WRITE) && isWrite)
				return;

			// We need to flush, or conditions and log expressions will see old register values.
			FlushAll();

			MOV(32, MIPSSTATE_VAR(pc), Imm32(GetCompilerPC()));
			CallProtectedFunction(&JitMemCheck, iaddr, checkedPC);

			CMP(32, R(RAX), Imm32(0));
			FixupBranch skipCheck = J_CC(CC_E);
			WriteDowncount(downcountOffset);
			JMP(dispatcherCheckCoreState, true);

			SetJumpTarget(skipCheck);
		}
	} else {
		const auto memchecks = CBreakPoints::GetMemCheckRanges(isWrite);
		bool possible = !memchecks.empty();
		if (!possible)
			return;

		gpr.Lock(rs);
		gpr.MapReg(rs, true, false);
		LEA(32, RAX, MDisp(gpr.RX(rs), offset));
		gpr.UnlockAll();

		// We need to flush, or conditions and log expressions will see old register values.
		FlushAll();

		std::vector<FixupBranch> hitChecks;
		hitChecks.reserve(memchecks.size());
		for (auto it = memchecks.begin(), end = memchecks.end(); it != end; ++it) {
			if (it->end != 0) {
				CMP(32, R(RAX), Imm32(it->start - size));
				FixupBranch skipNext = J_CC(CC_BE);

				CMP(32, R(RAX), Imm32(it->end));
				hitChecks.push_back(J_CC(CC_B, true));

				SetJumpTarget(skipNext);
			} else {
				CMP(32, R(RAX), Imm32(it->start));
				hitChecks.push_back(J_CC(CC_E, true));
			}
		}

		FixupBranch noHits = J(true);

		// Okay, now land any hit here.
		for (auto &fixup : hitChecks)
			SetJumpTarget(fixup);
		hitChecks.clear();

		MOV(32, MIPSSTATE_VAR(pc), Imm32(GetCompilerPC()));
		CallProtectedFunction(&JitMemCheck, R(RAX), checkedPC);

		CMP(32, R(RAX), Imm32(0));
		FixupBranch skipCheck = J_CC(CC_E);
		WriteDowncount(downcountOffset);
		JMP(dispatcherCheckCoreState, true);

		SetJumpTarget(skipCheck);
		SetJumpTarget(noHits);
	}
}

void Jit::CallProtectedFunction(const void *func, const OpArg &arg1) {
	// We don't regcache RCX, so the below is safe (and also faster, maybe branch prediction?)
	ABI_CallFunctionA(thunks.ProtectFunction(func, 1), arg1);
}

void Jit::CallProtectedFunction(const void *func, const OpArg &arg1, const OpArg &arg2) {
	// We don't regcache RCX/RDX, so the below is safe (and also faster, maybe branch prediction?)
	ABI_CallFunctionAA(thunks.ProtectFunction(func, 2), arg1, arg2);
}

void Jit::CallProtectedFunction(const void *func, const u32 arg1, const u32 arg2) {
	// We don't regcache RCX/RDX, so the below is safe (and also faster, maybe branch prediction?)
	ABI_CallFunctionCC(thunks.ProtectFunction(func, 2), arg1, arg2);
}

void Jit::CallProtectedFunction(const void *func, const OpArg &arg1, const u32 arg2) {
	// We don't regcache RCX/RDX, so the below is safe (and also faster, maybe branch prediction?)
	ABI_CallFunctionAC(thunks.ProtectFunction(func, 2), arg1, arg2);
}

void Jit::Comp_DoNothing(MIPSOpcode op) { }

MIPSOpcode Jit::GetOriginalOp(MIPSOpcode op) {
	JitBlockCache *bc = GetBlockCache();
	int block_num = bc->GetBlockNumberFromEmuHackOp(op, true);
	if (block_num >= 0) {
		return bc->GetOriginalFirstOp(block_num);
	} else {
		return op;
	}
}

} // namespace

#endif // PPSSPP_ARCH(X86) || PPSSPP_ARCH(AMD64)