SMPInstr.cpp

		//  not a call used as a branch within the function. A return instruction
		//  will usually cancel out the stack push that is implicit in the call, which 
		//  means that the function will have a net stack ptr delta of +4, which will
		//  cancel out the -4 value of the call instruction and set the delta to zero.
		//  However, this is not true in all cases, so we get the net stack ptr delta
		//  directly from the called function unless it is an unresolved indirect call,
		//  in which case we assume +4. !!!!****!!!! In the future, we could analyze
		//  the code around an unresolved indirect call to see if it seems to be
		//  removing items left on the stack by the callee.
#if 0
		// SPECIAL CASE: A jump used as a tail call will have a stack ptr effect that is equal
		//  to the net stack ptr effect of its target function, usually +4, whereas a jump
		//  would otherwise have a net stack ptr effect of 0.
#endif
		ea_t CalledFuncAddr = this->GetCallTarget();
		if ((BADADDR == CalledFuncAddr) || (0 == CalledFuncAddr)) {
			InstDelta = 0;
		}
		else { // We have a call target
			SMPFunction *CalleeFunc = this->GetBlock()->GetFunc()->GetProg()->FindFunction(CalledFuncAddr);
			sval_t AdjustmentDelta;
			if (CalleeFunc) {
				if (!CalleeFunc->HasSTARSStackPtrAnalysisCompleted()) {
					// Phase ordering issue in the call graph. A mutually recursive clique of functions has to
					//  be broken by starting processing somewhere, and all callees cannot be processed before
					//  we start. If we got our stack down to zero and then made a tail call, then we have to assume
					//  that the callee will use our return address, so we assume the default stack delta. If not a
					//  tail call, we ask our function to see if the information is available from IDA Pro analyses,
					//  or if it can be inferred from the fact that the call is followed by a stack adjustment.
					if (TailCall) {
						InstDelta = CALLING_CONVENTION_DEFAULT_FUNCTION_STACK_DELTA;
						SMP_msg("WARNING: Callee stack ptr analysis not yet performed at tail call inst %lx ; normal delta assumed\n",
							(unsigned long) this->GetAddr());
					}
					else {
						AdjustmentDelta = this->GetBlock()->GetFunc()->GetStackDeltaForCallee(CalledFuncAddr);
						InstDelta += AdjustmentDelta;
						SMP_msg("WARNING: Callee stack ptr analysis not yet performed at inst %lx ; stack adjustment used\n",
							(unsigned long) this->GetAddr());
					}
				}
				else if (!CalleeFunc->StackPtrAnalysisSucceeded()) {
					// Callee analyses were done, but they failed. In order to proceed, we have to assume
					//  the same situation as we just did in the case where analyses have not been performed.
					SMP_msg("WARNING: Callee stack ptr analysis failed at inst %lx ; normal delta assumed\n",
						(unsigned long) this->GetAddr());
					if (TailCall) {
						InstDelta = CALLING_CONVENTION_DEFAULT_FUNCTION_STACK_DELTA;
					}
					else {
						AdjustmentDelta = this->GetBlock()->GetFunc()->GetStackDeltaForCallee(this->GetAddr());
						InstDelta += AdjustmentDelta;
					}
				}
				else {
					// Callee's analyses have succeeded, so get delta straight from callee.
					InstDelta += CalleeFunc->GetNetStackPtrDelta();
				}
			}
			else {
#if 0
				SMP_msg("ERROR: SMPInstr::AnalyzeStackPointerDelta failed to find func at %lx in inst %lx\n",
					(unsigned long) CalledFuncAddr, (unsigned long) this->GetAddr());
				InstDelta = SMP_STACK_DELTA_ERROR_CODE;
#else
				SMP_msg("ERROR: SMPInstr::AnalyzeStackPointerDelta failed to find func at %lx in inst %lx\n",
					(unsigned long) CalledFuncAddr, (unsigned long) this->GetAddr());
				if (TailCall) {
					InstDelta = CALLING_CONVENTION_DEFAULT_FUNCTION_STACK_DELTA;
				}
				else {
					InstDelta = 0;
				}
#endif
			}
		}
	} // end CALL or INDIR_CALL or TailCall case
	else if (1 == InstDelta) { 
		// value of 1 is trigger to investigate the RTL for the 
		//  true value, which cannot be found simply by table lookup
		// In the special case of an x86 LEAVE instruction, the effect
		//  on the stack pointer is to deallocate the local frame size,
		//  plus pop the saved frame pointer into EBP. Helper functions
		//  need to know whether to look for this special case.
		bool IsLeaveInstr = this->MDIsLeaveInstr();
		InstDelta = this->RTL.TotalStackPointerAlteration(IsLeaveInstr, IncomingDelta, PreAllocDelta);
	}
	return InstDelta;
} // end of SMPInstr::AnalyzeStackPointerDelta()

// Total the stack adjustment bytes, as happens after a call to a function that leaves
//  outgoing args on the stack or swallows incoming args from the stack.
sval_t SMPInstr::FindStackAdjustment(void) {
	uint16 InstType = this->SMPcmd.itype;
	sval_t InstDelta = StackAlteration[InstType];

	if (1 == InstDelta) { 
		// value of 1 is trigger to investigate the RTL for the 
		//  true value, which cannot be found simply by table lookup
		// In the special case of an x86 LEAVE instruction, the effect
		//  on the stack pointer is to deallocate the local frame size,
		//  plus pop the saved frame pointer into EBP. Helper functions
		//  need to know whether to look for this special case.
		bool IsLeaveInstr = this->MDIsLeaveInstr();
		if (!IsLeaveInstr) {
			InstDelta = this->RTL.TotalStackPointerAlteration(IsLeaveInstr, 0, 0);
		}
		else {
			InstDelta = 0; // LEAVE is not the kind of instr we are looking for
		}
	}
	return InstDelta;
} // end of SMPInstr::FindStackAdjustment()

// Normalize stack operands to have a displacement from the stack pointer value on entry to the function,
//  rather than the current stack pointer value.
// UseFP indicates we are using a frame pointer in the function.
// FPDelta holds the stack delta (normalized) for the frame pointer.
// DefOp comes in with the operand to be normalized, and contains the normalized operand upon return.
// Return true if operand is a register or stack location, false otherwise (true => include in data flow analysis sets and SSA.)
bool SMPInstr::MDComputeNormalizedDataFlowOp(bool UseFP, sval_t FPDelta, op_t &DefOp) {
	if (o_reg == DefOp.type) {
		return true;
	}
	else if (MDIsStackAccessOpnd(DefOp, UseFP)) {
		op_t OldOp = DefOp;
		int SignedOffset = (int) DefOp.addr;
		sval_t NormalizedDelta;

		if (DefOp.hasSIB) {
			// We must deal with a potentially indexed memory expression. We want to
			//  normalize two different cases here: e.g. [esp+ebx+4] will become [esp+ebx-24]
			//  and [ebp+ebx-8] will become [esp+ebx-12] after normalization. A wrinkle
			//  on the second case is when the base register and index register are swapped
			//  in the SIB byte, and we make [ebx+ebp-4] into [esp+ebx-12], which involves
			//  correcting the index/base reg order in the SIB, because an index reg of ESP
			//  is the SIB encoding for "no index register" and we cannot leave it like that.
			int BaseReg = MD_STARS_sib_base(DefOp);
			int IndexReg = (int) MD_STARS_sib_index(DefOp);
			if (X86_STACK_POINTER_REG == IndexReg) // signifies no index register
				IndexReg = R_none;
			if (BaseReg == X86_STACK_POINTER_REG) {
				// We probably have an indexed ESP-relative operand.
				//  We leave the sib byte alone and normalize the offset.
				NormalizedDelta = this->GetStackPtrOffset() + (sval_t) SignedOffset;
			}
			else {
				// Must be EBP-relative.
				NormalizedDelta = FPDelta + (sval_t) SignedOffset;
				// Unfortunately, when we are dealing with a SIB byte in the opcode, we cannot
				//  just say DefOp.reg = MD_STACK_POINTER_REG to convert from the frame pointer
				//  to the stack pointer. Instead, we have to get into the nasty machine code
				//  level and change the SIB bits that specify either the base register or the
				//  index register, whichever one is the frame pointer.
				if (BaseReg == X86_FRAME_POINTER_REG) {
					// The three least significant bits of the SIB byte are the base register.
					//  They must contain a 5, which is the x86 value for register EBP, and we
					//  want to convert it to a 4, denoting register ESP. We can just zero out
					//  the least significant bit to accomplish that.
					DefOp.sib &= 0xfe;
				}
				else {
					// We sometimes have an instruction in which the frame pointer is used as
					//  the "index" register in the SIB byte, and the true index register is
					//  in the "base" register position in the SIB byte.
					assert(IndexReg == X86_FRAME_POINTER_REG);
					// The true index reg is in the lowest three bits, while the next three
					//  bits must contain a 5 (register EBP) and we want to make them a 4 (ESP).
					//  We must swap base and index regs as we normalize (see explanation above).
					char SIBtemp = DefOp.sib;
					char SIBindex = SIBtemp & 0x38;
					char SIBbase = SIBtemp & 0x07;
					assert ((SIBindex >> 3) == 5); // must be EBP
					SIBtemp &= 0xa0; // zero out lower 6 bits; upper 2 bits are scale factor - leave them alone
					SIBtemp &= (SIBbase << 3); // make old base reg (e.g. ebx) into a proper index reg
					SIBtemp |= 0x04; // make the new base reg be 4 (reg ESP)
					DefOp.sib = SIBtemp;
				}
				this->SetFPNormalizedToSP();
				// Add the stack pointer to the USE set for the instruction.
				this->MDAddRegUse(X86_STACK_POINTER_REG, false);
			}
		}

		else if (DefOp.reg == MD_FRAME_POINTER_REG) {
			// If FPDelta is -4 and SignedOffset is +8, then we have [ebp+8] as DefOp, and this
			//  is equivalent to [esp+4] where esp has its entry value, i.e. this would be the first incoming
			//  argument. If SignedOffset is -12, we have [ebp-12] as DefOp, and this is [esp-16] when
			//  normalized to the entry point value of the stack pointer. In both cases, we can see that the
			//  normalized stack delta is just FPDelta+SignedOffset.
			NormalizedDelta = FPDelta + (sval_t) SignedOffset;
			// Now, we simply convert the memory operand from EBP to ESP and replace the SignedOffset with the
			//  NormalizedDelta just computed.
			DefOp.reg = MD_STACK_POINTER_REG;
			this->SetFPNormalizedToSP();
			// Add the stack pointer to the USE set for the instruction.
			this->MDAddRegUse(DefOp.reg, false);
		}
		else {
			assert(DefOp.reg == MD_STACK_POINTER_REG);
			// We only need to adjust the offset to reflect the change in the stack pointer since the function
			//  was entered, e.g. [esp+4] is normalized to [esp-28] if the current esp value is 32 less than it
			//  was upon function entry. We get the value "-32" in that case from a member variable.
			NormalizedDelta = this->GetStackPtrOffset() + (sval_t) SignedOffset;
		}
		DefOp.addr = (ea_t) NormalizedDelta; // common to frame and stack pointer cases
		if ((o_phrase == DefOp.type) && (0 != NormalizedDelta)) {
			// mov [esp],eax has an [esp] operand of type o_phrase, because there is no
			//  displacement field. After normalization, it will have a displacement field, so
			//  it has become an operand like [esp-32] and is now type o_displ.
			DefOp.type = o_displ;
		}
		this->GetBlock()->GetFunc()->AddNormalizedStackOperand(OldOp, this->GetAddr(), DefOp);
		return true;
	}
	else {
		return false;
	}
} // end of SMPInstr::MDComputeNormalizedDataFlowOp()

// Normalize stack operands in all DEFs and USEs to have stack deltas relative to the function entry stack pointer.
// Return true if any stack DEFs or USEs were normalized.
bool SMPInstr::MDNormalizeStackOps(bool UseFP, sval_t FPDelta, bool Recomputing, sval_t DeltaIncrement) {
	bool StackOpFound = false;
	bool OpNormalized;
	bool UniqueDEFMemOp = true; // Does DEFMemOp not match any DEFs?
	bool UniqueUSEMemOp = true; // Does USEMemOp not match any USEs?
	bool UniqueLeaUSEMemOp = true; // Does LeaUSEMemOp not match any USEs?
	bool UniqueMoveSource = true; // Does MoveSource not match any USEs?
	set<DefOrUse, LessDefUse>::iterator DefIter, UseIter;
	list<pair<set<DefOrUse, LessDefUse>::iterator, op_t> > DefWorkList, UseWorkList;
	list<pair<set<DefOrUse, LessDefUse>::iterator, op_t> >::iterator WorkIter;
	op_t OldOp, NewOp;
	
	// Find all the DEFs that need changing, and put their iterators into a list.
	// Normalizing stack ops could change their sort order, hence we could skip over
	//  a DEF in the set by erasing a DEF and reinserting a normalized DEF, so we
	//  make all the changes after we iterate through the DEFS set.
	for (DefIter = this->GetFirstDef(); DefIter != this->GetLastDef(); ++DefIter) {
		OldOp = DefIter->GetOp();
		NewOp = OldOp;
		if ((o_reg != NewOp.type) && (o_imm != NewOp.type)) {
			if (Recomputing) {
				OpNormalized = this->MDRecomputeNormalizedDataFlowOp(DeltaIncrement, true, NewOp);
			}
			else {
				OpNormalized = this->MDComputeNormalizedDataFlowOp(UseFP, FPDelta, NewOp);
			}
			if (OpNormalized) {
				StackOpFound = true;
				if (IsEqOp(OldOp, this->DEFMemOp)) {
					UniqueDEFMemOp = false;
				}
				pair<set<DefOrUse, LessDefUse>::iterator, op_t> DefItem(DefIter, NewOp);
				DefWorkList.push_back(DefItem);
			}
		}
	}
	// Now go through the DEF worklist and change stack operands to normalized stack operands.
	for (WorkIter = DefWorkList.begin(); WorkIter != DefWorkList.end(); ++WorkIter) {
		DefIter = WorkIter->first;
		DefIter = this->Defs.SetOp(DefIter, WorkIter->second);
	}
	// Normalize op_t private data member DEFs.
	if (Recomputing) {
		OpNormalized = this->MDRecomputeNormalizedDataFlowOp(DeltaIncrement, UniqueDEFMemOp, this->DEFMemOp);
	}
	else {
		OpNormalized = this->MDComputeNormalizedDataFlowOp(UseFP, FPDelta, this->DEFMemOp);
	}

	// Find all USEs that need changing, and build a second work list.
	for (UseIter = this->GetFirstUse(); UseIter != this->GetLastUse(); ++UseIter) {
		OldOp = UseIter->GetOp();
		NewOp = OldOp;
		if ((o_reg != NewOp.type) && (o_imm != NewOp.type)) {
			if (Recomputing) {
				OpNormalized = this->MDRecomputeNormalizedDataFlowOp(DeltaIncrement, true, NewOp);
			}
			else {
				OpNormalized = this->MDComputeNormalizedDataFlowOp(UseFP, FPDelta, NewOp);
			}
			if (OpNormalized) {
				StackOpFound = true;
				if (IsEqOp(OldOp, this->USEMemOp)) {
					UniqueUSEMemOp = false;
				}
				if (IsEqOp(OldOp, this->GetLeaMemUseOp())) {
					UniqueLeaUSEMemOp = false;
				}
				if (IsEqOp(OldOp, this->MoveSource)) {
					UniqueMoveSource = false;
				}
				pair<set<DefOrUse, LessDefUse>::iterator, op_t> UseItem(UseIter, NewOp);
				UseWorkList.push_back(UseItem);
			}
		}
	}

	// Now go through the USE worklist and change stack operands to normalized stack operands.
	for (WorkIter = UseWorkList.begin(); WorkIter != UseWorkList.end(); ++WorkIter) {
		UseIter = WorkIter->first;
		UseIter = this->Uses.SetOp(UseIter, WorkIter->second);
	}
	// Normalize op_t private data member USEs.
	op_t TempLeaMemOp = this->GetLeaMemUseOp();
	if (Recomputing) {
		OpNormalized = this->MDRecomputeNormalizedDataFlowOp(DeltaIncrement, UniqueUSEMemOp, this->USEMemOp);
		OpNormalized = this->MDRecomputeNormalizedDataFlowOp(DeltaIncrement, UniqueLeaUSEMemOp, TempLeaMemOp);
		if (OpNormalized)
			this->SetLeaMemUseOp(TempLeaMemOp);
		OpNormalized = this->MDRecomputeNormalizedDataFlowOp(DeltaIncrement, UniqueMoveSource, this->MoveSource);
	}
	else {
		OpNormalized = this->MDComputeNormalizedDataFlowOp(UseFP, FPDelta, this->USEMemOp);
		OpNormalized = this->MDComputeNormalizedDataFlowOp(UseFP, FPDelta, TempLeaMemOp);
		if (OpNormalized)
			this->SetLeaMemUseOp(TempLeaMemOp);
		OpNormalized = this->MDComputeNormalizedDataFlowOp(UseFP, FPDelta, this->MoveSource);
	}
	// Declare victory.
	this->SetDefsNormalized();

	return StackOpFound;
} // end of SMPInstr::MDNormalizeStackOps()

// Renormalize SP-relative stack operands in functions that call alloca() by adding DeltaIncrement to their stack displacements.
// DefOp comes in with the operand to be renormalized, and contains the normalized operand upon return.
// Return true if operand is a register or stack location, false otherwise (true => include in data flow analysis sets and SSA.)
bool SMPInstr::MDRecomputeNormalizedDataFlowOp(sval_t DeltaIncrement, bool UpdateMaps, op_t &DefOp) {
	op_t OldOp = DefOp;
	if (o_reg == DefOp.type) {
		return true;
	}
	else if (MDIsStackAccessOpnd(DefOp, this->GetBlock()->GetFunc()->UsesFramePointer())) {
		if (this->HasFPNormalizedToSP()) {
			// FP-relative operands do no change in alloca() functions when the alloca()
			//  causes the SP to change.
			return true;
		}

		// The remaining cases are simple. The ESP-relative displacement is incremented by
		//  DeltaIncrement, regardless of the presence of a SIB byte.
		int SignedOffset = (int) DefOp.addr;
		sval_t NormalizedDelta = DeltaIncrement + (sval_t) SignedOffset;

		DefOp.addr = (ea_t) NormalizedDelta;

		if ((o_phrase == DefOp.type) && (0 != NormalizedDelta)) {
			// mov [esp],eax has an [esp] operand of type o_phrase, because there is no
			//  displacement field. After normalization, it will have a displacement field, so
			//  it has become an operand like [esp-32] and is now type o_displ.
			DefOp.type = o_displ;
		}

		if (UpdateMaps) { // We don't update maps for duplicate entries, e.g. USEMemOp, DEFMemOp, MoveSource
			this->GetBlock()->GetFunc()->AddNormalizedStackOperand(OldOp, this->GetAddr(), DefOp);
		}
		return true;
	}
	else {
		return false;
	}
} // end of SMPInstr::MDRecomputeNormalizedDataFlowOp()

// If NormOp is a normalized stack memory operand, unnormalize it.
void SMPInstr::MDGetUnnormalizedOp(op_t &NormOp) {
	sval_t SignedOffset;
	bool UseFP = this->GetBlock()->GetFunc()->UsesFramePointer();
	if (this->AreDefsNormalized() && MDIsStackAccessOpnd(NormOp, UseFP)) {
		if (this->HasFPNormalizedToSP()) {
			// Need to convert NormOp back to frame-pointer-relative address.
			if (NormOp.hasSIB) {
				// Convert base register from stack pointer back to frame pointer.
				NormOp.sib |= 0x01;
			}
			else {
				NormOp.reg = MD_FRAME_POINTER_REG;
			}
			SignedOffset = (sval_t) NormOp.addr;
			SignedOffset -= this->GetBlock()->GetFunc()->GetFramePtrStackDelta();
		}
		else {
			// NormOp should remain stack-pointer-relative address, but it
			//  should be a positive offset from the current stack pointer instead
			//  of a negative offset from the entry point of the function.
			SignedOffset = (sval_t) NormOp.addr;
			SignedOffset -= this->GetStackPtrOffset();
			assert((0 <= SignedOffset) || this->GetBlock()->GetFunc()->DoesStackFrameExtendPastStackTop());
		}
		NormOp.addr = (ea_t) SignedOffset;
	}
	return;
} // end of SMPInstr::MDGetUnnormalizedOp()

// Find USE-not-DEF operand that is not the flags register.
op_t SMPInstr::GetSourceOnlyOperand(void) {
	size_t OpNum;
	for (OpNum = 0; OpNum < UA_MAXOP; ++OpNum) {
		if (this->features & DefMacros[OpNum]) { // DEF
			;
		}
		else if (this->features & UseMacros[OpNum]) { // USE
			op_t CurrOp = this->SMPcmd.Operands[OpNum];
			if (!(CurrOp.is_reg(X86_FLAGS_REG))) {
				return CurrOp;
			}
		}
	}
	// It is expected that increment, decrement, and floating point stores
	//  will not have a USE-only operand. Increment and decrement have an
	//  operand that is both USEd and DEFed, while the floating point stack
	//  registers are implicit in most floating point opcodes. Also, exchange
	//  and exchange-and-add instructions have multiple DEF-and-USE operands.
	int TypeGroup = SMPTypeCategory[this->SMPcmd.itype];
	if ((TypeGroup != 2) && (TypeGroup != 4) && (TypeGroup != 9) && (TypeGroup != 12)
		&& (TypeGroup != 13)) {
		SMP_msg("ERROR: Could not find source only operand at %lx in %s\n",
			(unsigned long) this->address, DisAsmText.GetDisAsm(this->GetAddr()));
	}
	return InitOp;
} // end of SMPInstr::GetSourceOnlyOperand()

// Should apparent memory operands be ignored? e.g. lea opcode on x86
bool SMPInstr::MDIgnoreMemOps(void) {
	bool leaInst = (NN_lea == this->SMPcmd.itype);
	return leaInst;
}

// Find memory DEFs and USEs, store in DEFMemOp and USEMemOp
void SMPInstr::FindMemOps(void) {
	size_t OpNum;

	if (!(this->MDIgnoreMemOps())) {
		for (OpNum = 0; OpNum < UA_MAXOP; ++OpNum) {
			op_t TempOp = this->SMPcmd.Operands[OpNum];
			if ((TempOp.type >= o_mem) && (TempOp.type <= o_displ)) { // memory
				if (this->features & DefMacros[OpNum]) { // DEF
					if (this->DEFMemOp.type == o_void) { // only save first mem DEF
						this->DEFMemOp = TempOp;
					}
				}
				if (this->features & UseMacros[OpNum]) { // USE
					if (this->USEMemOp.type == o_void) { // only save first mem USE
						this->USEMemOp = TempOp;
					}
				}
			}
		} // end for (OpNum = 0; ...)
	}
	this->SetMemOpsFound();
	return;
} // end of SMPInstr::FindMemOps()

// Fix problems with the operands list in SMPcmd.
void SMPInstr::MDFixupIDAProOperandList(void) {
	// IDA Pro often takes the instruction imul eax,0x80 and creates the following operands and features bits:
	//  Opnd[0] = EAX, both DEF and USE
	//  Opnd[1] = EAX, just USE
	//  Opnd[2] = immediate, neither DEF nor USE
	//  Our RTL building keys in on the DEF/USE bits in features, so this looks like imul eax,eax to us.
	//  We want it to look like:
	//  Opnd[0] = EAX, both DEF and USE
	//  Opnd[1] = immediate, just USE
	if (NN_imul == this->SMPcmd.itype) {
		op_t Opnd2 = this->SMPcmd.Operands[2];
		if ((!(this->features & DefMacros[2]))
			&& (!(this->features & UseMacros[2]))) {
			if (o_void != Opnd2.type) {
				// We have a third operand that is neither DEF nor USE.
				SMP_msg("INFO: Fixing IMUL operand list at %lx\n", (unsigned long) this->GetAddr());
				this->Dump();
				// Two cases: Operands[0] == Operands[1], e.g. imul eax,Opnd2
				//  or else three-operand form: e.g. imul eax,ecx,Opnd2
				// For the three-operand form, make sure Opnd0 is DEF only, others
				//  are USE only. For the two-operand form, make sure Opnd0 is DEF and USE,
				//  Opnd1 is current Opnd2 and is USE only.
				op_t Opnd0 = this->SMPcmd.Operands[0];
				op_t Opnd1 = this->SMPcmd.Operands[1];
				if (IsEqOp(Opnd0, Opnd1)) {
					// No need for three-operand form.
					this->features |= DefMacros[0];
					this->features |= UseMacros[0];
					this->SMPcmd.Operands[1] = Opnd2;
					this->SMPcmd.Operands[2] = InitOp;
				}
				else { // Must have three-operand form.
					this->features |= UseMacros[2]; // set missing USE bit.
					this->features &= (~UseMacros[0]); // Ensure no USE of Opnd0.
				}
				this->Dump();
			}		
		}
	}
	return;
} // SMPInstr::MDFixupIDAProOperandList()

// Fill the Defs and Uses private data members.
void SMPInstr::BuildSMPDefUseLists(void) {
	size_t OpNum;
	bool DebugFlag = (0x8049b00 == this->GetAddr());
	bool WidthDoubler = this->MDDoublesWidth();

	this->Defs.clear();
	this->Uses.clear();

	// Start with the Defs.
	for (OpNum = 0; OpNum < UA_MAXOP; ++OpNum) {
		if (this->features & DefMacros[OpNum]) { // DEF
			op_t TempOp = this->SMPcmd.Operands[OpNum];
			if (WidthDoubler) {
				// Opcodes that sign-extend a byte to a word, or a word to a dword, 
				//  have only one operand. It is implicit, and it is the shorter USE.
				//  That means the DEF will have the same width as the USE, e.g. if
				//  we are sign-extending AX to EAX, the USE and DEF both be AX without
				//  a special fix. We fix this problem with the DEF operand now.
				if (TempOp.dtyp == dt_byte) {
					TempOp.dtyp = dt_word;
					TempOp.reg = MDCanonicalizeSubReg(TempOp.reg);
				}
				else if (TempOp.dtyp == dt_word) {
					TempOp.dtyp = dt_dword;
					TempOp.reg = MDCanonicalizeSubReg(TempOp.reg);
				}
				else if (TempOp.dtyp == dt_dword) {
					TempOp.dtyp = dt_qword;
				}
				else {
					SMP_msg("ERROR: Instruction operand %zu not 1,2, or 4 bytes at %lx dtyp: %d\n", 
						OpNum, (unsigned long) this->address, TempOp.dtyp);
				}
			}
			if (MDKnownOperandType(TempOp)) {
				if (DebugFlag) {
					SMP_msg("DEBUG: Setting DEF for: ");
					PrintOperand(TempOp);
					SMP_msg("\n");
				}
				this->Defs.SetRef(TempOp);
			}
		}
	} // end for (OpNum = 0; ...)

	if (this->IsRegClearIdiom()) {
		// Something like xor eax,eax clears eax but does not really
		//  use eax. It is the same as mov eax,0 and we don't want to
		//  extend the prior def-use chain for eax to this instruction
		//  by treating the instruction as xor eax,eax. Instead, we
		//  build the DEF and USE lists and RTL as if it were mov eax,0.
		op_t ImmOp = InitOp;
		ImmOp.type = o_imm;
		ImmOp.value = 0;
		ImmOp.dtyp = this->GetOperandDtypField();
		this->Uses.SetRef(ImmOp, NUMERIC);
		return;
	}

	// Now, do the Uses. Uses have special case operations, because
	//  any memory operand could have register uses in the addressing
	//  expression, and we must create Uses for those registers. For
	//  example:  mov eax,[ebx + esi*2 + 044Ch]
	//  This is a two-operand instruction with one def: eax. But
	//  there are three uses: [ebx + esi*2 + 044Ch], ebx, and esi.
	//  The first use is an op_t of type o_phrase (memory phrase),
	//  which can be copied from cmd.Operands[1]. Likewise, we just
	//  copy cmd.Operands[0] into the defs list. However, we must create
	//  op_t types for register ebx and register esi and append them
	//  to the Uses list. This is handled by the machine dependent
	//  method MDFixupDefUseLists().
	for (OpNum = 0; OpNum < UA_MAXOP; ++OpNum) {
		if (this->features & UseMacros[OpNum]) { // USE
			op_t TempOp = this->SMPcmd.Operands[OpNum];
			if (MDKnownOperandType(TempOp)) {
				if (DebugFlag) {
					SMP_msg("DEBUG: Setting USE for: ");
					PrintOperand(TempOp);
					SMP_msg("\n");
				}
				this->Uses.SetRef(TempOp);
			}
		}
	} // end for (OpNum = 0; ...)

	return;
} // end of SMPInstr::BuildSMPDefUseLists()

// Declare a branch/jump to be a tail call, clean up def/use lists.
void SMPInstr::SetTailCall(void) { 
	this->booleans1 |= INSTR_SET_TAIL_CALL;
	if (this->type == COND_BRANCH) {
		this->SetCondTailCall();
	}
	else {
		this->ResetCondTailCall();
	}
	this->CallTarget = this->FarBranchTarget;
	this->type = RETURN;
	this->GetBlock()->SetReturns(true);

	// We want to add the caller-saved registers to the USEs and DEFs lists
	this->MDAddRegDef(R_ax, false);
	this->MDAddRegDef(R_cx, false);
	this->MDAddRegDef(R_dx, false);
	this->MDAddRegUse(R_ax, false);
	this->MDAddRegUse(R_cx, false);
	this->MDAddRegUse(R_dx, false);
} // end of SMPInstr::SetTailCall()

// record original Lea instruction [pseudo-]memory operand.
void SMPInstr::SetLeaMemUseOp(op_t NewLeaOperand) {
	if (NULL != this->BasicBlock) {
		this->GetBlock()->GetFunc()->AddLeaOperand(this->GetAddr(), NewLeaOperand);
	}
	return;
}

// If DefReg is not already in the DEF list, add a DEF for it.
void SMPInstr::MDAddRegDef(ushort DefReg, bool Shown, SMPOperandType Type) {
	op_t TempDef = InitOp;
	TempDef.type = o_reg;
	TempDef.reg = DefReg;
	TempDef.dtyp = this->GetOperandDtypField();
	if (Shown)
		TempDef.set_showed();
	else
		TempDef.clr_showed();
	this->Defs.SetRef(TempDef, Type);
	return;
} // end of SMPInstr::MDAddRegDef()

// If UseReg is not already in the USE list, add a USE for it.
void SMPInstr::MDAddRegUse(ushort UseReg, bool Shown, SMPOperandType Type) {
	op_t TempUse = InitOp;
	TempUse.type = o_reg;
	TempUse.reg = UseReg;
	TempUse.dtyp = this->GetOperandDtypField();
	if (Shown)
		TempUse.set_showed();
	else
		TempUse.clr_showed();
	this->Uses.SetRef(TempUse, Type);
	return;
} // end of SMPInstr::MDAddRegUse()

// Perform machine dependent ad hoc fixes to the def and use lists.
//  For example, some multiply and divide instructions in x86 implicitly
//  use and/or define register EDX. For memory phrase examples, see comment
//  in BuildSMPDefUseLists().
void SMPInstr::MDFixupDefUseLists(void) {
	// First, handle the uses hidden in memory addressing modes. Note that we do not
	//  care whether we are dealing with a memory destination operand or source
	//  operand, because register USEs, not DEFs, happen within the addressing expressions.
	size_t OpNum;
	SMPOperandType RefType;
	unsigned short opcode = this->SMPcmd.itype;
	int BaseReg;
	int IndexReg;
	ushort ScaleFactor;
	ea_t displacement;
	bool UseFP = true;
	bool HasIndexReg = false;
	bool SingleAddressReg = false;
	bool leaInst = (NN_lea == opcode);
	bool DebugFlag = (this->GetAddr() == 0x8086177);
	if (DebugFlag) {
		SMP_msg("DEBUG: Fixing up DEF-USE lists for debug location\n");
		this->Dump();
	}

#if SMP_BASEREG_POINTER_TYPE
	// Some instructions are analyzed outside of any function or block when fixing up
	//  the IDB, so we have to assume the block and func pointers might be NULL.
	if ((NULL != this->BasicBlock) && (NULL != this->BasicBlock->GetFunc()))
		UseFP = this->BasicBlock->GetFunc()->UsesFramePointer();
#endif

	if (DebugFlag) {
		SMP_msg("DEBUG: UseFP = %d\n", UseFP);
	}

	for (OpNum = 0; OpNum < UA_MAXOP; ++OpNum) {
		op_t Opnd = SMPcmd.Operands[OpNum];
		if ((Opnd.type == o_phrase) || (Opnd.type == o_displ) || (Opnd.type == o_mem)) {
			MDExtractAddressFields(Opnd, BaseReg, IndexReg, ScaleFactor, displacement);
			SingleAddressReg = ((0 == displacement) 
				&& ((R_none == BaseReg) || (R_none == IndexReg)));
			if (R_none != IndexReg) { 
				op_t IndexOpnd = Opnd; // Init to current operand field values
				IndexOpnd.type = o_reg; // Change type and reg fields
				IndexOpnd.reg = (ushort) IndexReg;
				IndexOpnd.hasSIB = 0;
				IndexOpnd.set_showed();
				if (0 == ScaleFactor)
					this->Uses.SetRef(IndexOpnd);
				else { // scaling == shift ==> NUMERIC
					HasIndexReg = true;
					this->Uses.SetRef(IndexOpnd, NUMERIC);
				}
			}
			if (R_none != BaseReg) {
				op_t BaseOpnd = Opnd; // Init to current operand field values
				BaseOpnd.type = o_reg; // Change type and reg fields
				BaseOpnd.reg = (ushort) BaseReg;
				BaseOpnd.hasSIB = 0;
				BaseOpnd.set_showed();
				RefType = UNINIT;
#if SMP_BASEREG_POINTER_TYPE
				// R_sp and R_bp will get type STACKPTR in SMPInstr::SetImmedTypes().
				//  Other registers used as base registers should get their USEs as
				//  base registers typed as POINTER, which might get refined later
				//  to STACKPTR, GLOBALPTR, HEAPPTR, etc.
				// NOTE: the NN_lea opcode is often used without a true base register.
				//  E.g. lea eax,[eax+eax+5] is an x86 idiom for eax:=eax*2+5, which
				//  could not be done in one instruction without using the addressing
				//  modes of the machine to do the arithmetic. We don't want to set the
				//  USE of EAX to POINTER in this case, so we will conservatively skip
				//  all lea instructions here.
				// We cannot be sure that a register is truly a base register unless
				//  there is also an index register. E.g. with reg+displacement, we
				//  could have memaddr+indexreg or basereg+offset, depending on what
				//  the displacement is. The exception is if there is no offset and only
				//  one addressing register, e.g. mov eax,[ebx].
				if (BaseOpnd.is_reg(MD_STACK_POINTER_REG) || (UseFP && BaseOpnd.is_reg(MD_FRAME_POINTER_REG))
					|| leaInst || (!HasIndexReg && !SingleAddressReg)) {
					;
				}
				else {
					RefType = POINTER;
				}
#endif
				this->Uses.SetRef(BaseOpnd, RefType);
			} // end if R_none != BaseReg
		} // end if (o_phrase or o_displ operand)
	} // end for (all operands)

	// The lea (load effective address) instruction looks as if it has
	//  a memory USE:  lea ebx,[edx+esi]
	//  However, this instruction is really just: ebx := edx+esi
	//  Now that the above code has inserted the "addressing" registers
	//  into the USE list, we should remove the "memory USE".
	if (leaInst) {
		set<DefOrUse, LessDefUse>::iterator CurrUse;
		for (CurrUse = this->GetFirstUse(); CurrUse != this->GetLastUse(); ++CurrUse) {
			op_t UseOp = CurrUse->GetOp();
			if ((o_mem <= UseOp.type) && (o_displ >= UseOp.type)) {
				this->SetLeaMemUseOp(UseOp);
				this->EraseUse(CurrUse);
				this->USEMemOp = InitOp;
				break;
			}
		}
	}

	// Next, handle repeat prefices in the instructions. The Intel REPE/REPZ prefix
	//  is just the text printed for SCAS/CMPS instructions that have a REP prefix.
	//  Only two distinct prefix codes are actually defined: REP and REPNE/REPNZ, and
	//  REPNE/REPNZ only applies to SCAS and CMPS instructions.
	bool HasRepPrefix = (0 != (this->SMPcmd.auxpref & aux_rep));
	bool HasRepnePrefix = (0 != (this->SMPcmd.auxpref & aux_repne));
	if (HasRepPrefix && HasRepnePrefix)
		SMP_msg("REP and REPNE both present at %lx %s\n", (unsigned long) this->GetAddr(), DisAsmText.GetDisAsm(this->GetAddr()));
	if (HasRepPrefix || HasRepnePrefix) {
		// All repeating instructions use ECX as the countdown register.
		op_t BaseOpnd = InitOp;
		BaseOpnd.type = o_reg; // Change type and reg fields
		BaseOpnd.reg = R_cx;
		BaseOpnd.clr_showed();
		BaseOpnd.dtyp = this->GetOperandDtypField();
		this->Defs.SetRef(BaseOpnd, NUMERIC);
		this->Uses.SetRef(BaseOpnd, NUMERIC);
	}
	if ((opcode == NN_cmps) || (opcode == NN_scas) || (opcode == NN_movs) || (opcode == NN_stos)) {
		// ESI and EDI are USEd and DEFed to point to source and dest strings for CMPS/MOVS.
		//  Only EDI is involved with SCAS/STOS.
		op_t BaseOpnd = InitOp;
		BaseOpnd.type = o_reg; // Change type and reg fields
		BaseOpnd.clr_showed();
		BaseOpnd.dtyp = this->GetOperandDtypField();
		if ((opcode == NN_cmps) || (opcode == NN_movs)) {
			BaseOpnd.reg = R_si;
			this->Defs.SetRef(BaseOpnd, POINTER);
			this->Uses.SetRef(BaseOpnd, POINTER);
		}
		BaseOpnd.reg = R_di;
		this->Defs.SetRef(BaseOpnd, POINTER);
		this->Uses.SetRef(BaseOpnd, POINTER);
	}
	else if ((NN_loopw <= opcode) && (NN_loopqne >= opcode)) {
		op_t LoopCounterOp = InitOp;
		LoopCounterOp.type = o_reg;
		LoopCounterOp.reg = R_cx;
		LoopCounterOp.dtyp = this->GetOperandDtypField();
		this->Defs.SetRef(LoopCounterOp, NUMERIC);
		this->Uses.SetRef(LoopCounterOp, NUMERIC);
	}

	// Now, handle special instruction categories that have implicit operands.
	if (NN_cmpxchg == opcode) {
		// x86 Compare and Exchange conditionally sets EAX. We must keep data flow analysis
		//  sound by declaring that EAX is always a DEF.
		this->MDAddRegDef(R_ax, false);
	} // end if NN_cmpxchg
	else if (this->MDIsPopInstr() || this->MDIsPushInstr() || this->MDIsReturnInstr()) {
		// IDA does not include the stack pointer in the DEFs or USEs.
		this->MDAddRegDef(R_sp, false);
		this->MDAddRegUse(R_sp, false);
		if (!this->MDIsReturnInstr()) {
			// We always reference [esp+0] or [esp-4], so add it to the DEF or USE list.
			op_t StackOp = InitOp;
			StackOp.type = o_displ;
			StackOp.reg = R_sp;
			StackOp.dtyp = this->GetOperandDtypField();

			if (this->MDIsPopInstr()) {
				StackOp.addr = 0;  // [ESP+0]
				this->Uses.SetRef(StackOp);  // USE
			}
			else {
				StackOp.addr = (- ((ea_t) STARS_ISA_Bytewidth));  // [ESP-4]
				this->Defs.SetRef(StackOp); // DEF
			}
		}
	}
#if SMP_CALL_TRASHES_REGS
	else if ((this->type == CALL) || (this->type == INDIR_CALL) || this->IsTailCall()) {
		// We want to add the caller-saved registers to the USEs and DEFs lists
		this->MDAddRegDef(R_ax, false);
		this->MDAddRegDef(R_cx, false);
		this->MDAddRegDef(R_dx, false);
		this->MDAddRegUse(R_ax, false);
		this->MDAddRegUse(R_cx, false);
		this->MDAddRegUse(R_dx, false);
#if 1
		if (this->MDIsInterruptCall()) {
#endif
			this->MDAddRegDef(R_bx, false);
			this->MDAddRegUse(R_bx, false);
			this->MDAddRegDef(R_si, false);
			this->MDAddRegUse(R_si, false);
#if 1
		}
#endif
	}
#endif
	else if (this->MDIsEnterInstr() || this->MDIsLeaveInstr()) {
		// Entire function prologue or epilogue microcoded.
		this->MDAddRegDef(MD_STACK_POINTER_REG, false);
		this->MDAddRegUse(MD_STACK_POINTER_REG, false);
		this->MDAddRegDef(MD_FRAME_POINTER_REG, false);
		this->MDAddRegUse(MD_FRAME_POINTER_REG, false);
	}
	else if ((opcode == NN_maskmovq) || (opcode == NN_maskmovdqu)) {
		this->MDAddRegUse(R_di, false, POINTER);
	}
	else if (8 == this->GetOptType()) {
		// This category implicitly writes to EDX:EAX.
		this->MDAddRegDef(R_dx, false);
		this->MDAddRegDef(R_ax, false);
	} // end else if (8 == GetOptType)
	else if (7 == this->GetOptType()) {
		// Category 7 instructions sometimes write implicitly to EDX:EAX or DX:AX.
		//  DX is the same as EDX to IDA Pro (and SMP); ditto for EAX and AX.
		// DIV, IDIV, and MUL all have hidden EAX or AX operands (hidden in the IDA Pro
		//  sense, because they are not displayed in the disassembly text). For example:
		//  mul ebx means EDX:EAX <-- EAX*EBX, and mul bx means DX:AX <-- AX*BX. If the
		//  source operand is only 8 bits wide, there is room to hold the result in AX
		//  without using DX:  mul bl means AX <-- AL*BL.
		// IMUL has forms with a hidden EAX or AX operand and forms with no implicit
		//  operands:  imul ebx means EDX:EAX <-- EAX*EBX, but imul ebx,edx means that
		//  EBX*EDX gets truncated and the result placed in EBX (no hidden operands).
		for (OpNum = 0; OpNum < UA_MAXOP; ++OpNum) {
			op_t TempUse = this->SMPcmd.Operands[OpNum];
			if (!TempUse.showed()) { // hidden operand
				if (TempUse.is_reg(R_ax)) { // not R_al, so it is not 8 bits
					if ((NN_div == this->SMPcmd.itype) || (NN_idiv == this->SMPcmd.itype)) {
						this->MDAddRegUse(R_dx, false);
					}
					this->MDAddRegDef(R_ax, false);
					this->MDAddRegDef(R_dx, false);
				}
			}
		}
	} // end else if (7 == OptType)

#if 0
	// The floating point instructions in type categories 14 and 15 often USE and DEF
	//  the floating point register stack, e.g. pushing a value onto that stack is a
	//  massive copy downward of stack locations. We don't really care about the USE of
	//  the stack if the value being pushed came from elsewhere than the stack. For example,
	//  an "fld" opcode pushes its source onto the stack. We build RTLs with a simple
	//  move structure, but the RTL building can be fooled by seeing two "source" operands
	//  in the USE list.
	if ((14 == SMPTypeCategory[this->SMPcmd.itype])
		|| (15 == SMPTypeCategory[this->SMPcmd.itype])) {
	}
#endif

#if 0  // Not true for LOOP instructions that use only the ECX counter register.
	if (this->type == COND_BRANCH) {
		assert(SMPUsesFlags[opcode]);
	}
#endif
	// The return value register EAX is not quite like a caller-save or callee-save
	//  register (technically, it is caller-save). Within a callee, it might appear
	//  that EAX has become dead by the time a return instruction is reached, but
	//  the USE that would make it not dead is in the caller. To prevent type inference
	//  from mistakenly thinking that all USEs of EAX have been seen in the callee,
	//  we add EAX to the USE list for all return instructions, as well as for all
	//  tail calls, which are essentially returns in terms of data flow analysis.
	// This USE of EAX will always be of type UNINIT unless its DEF has a known type
	//  that propagates to it. Thus, it will prevent an invalid back inference of the
	//  DEF type from "all" USE types that are visible in the callee; even if they
	//  were all NUMERIC, this return USE will be UNINIT and inhibit the invalid
	//  type inference. EAX could be loaded with a pointer from memory, for example,
	//  and USEd only in a comparison instruction, making it falsely appear to be
	//  a NUMERIC, without this extra USE at the return instruction.
	// Because some of the library functions pass values around in EBX, EDI, etc.,
	//  we will add these general purpose registers to the USE list for returns
	//  in order to prevent erroneous analyses of dead registers or unused
	//  metadata.
	if ((this->type == RETURN) || this->IsTailCall()) {
		this->MDAddRegUse(R_ax, false);
		this->MDAddRegUse(R_bx, false);
		this->MDAddRegUse(R_cx, false);
		this->MDAddRegUse(R_dx, false);
		if (!UseFP)
			this->MDAddRegUse(MD_FRAME_POINTER_REG, false);
		this->MDAddRegUse(R_si, false);
		this->MDAddRegUse(R_di, false);
	}

	// Next, add the flags register to the DEFs and USEs for those instructions that
	//  are marked as defining or using flags.
	if (!this->IsDefsFlags() && SMPDefsFlags[opcode]) {
		this->MDAddRegDef(X86_FLAGS_REG, false);
		this->SetDefsFlags();
	}
	if (!this->IsUsesFlags() && SMPUsesFlags[opcode]) {
		this->MDAddRegUse(X86_FLAGS_REG, false);
		this->SetUsesFlags();
	}

#if 1
	if (this->IsNop()) {
		// Clear the DEFs and USEs for no-ops.
		//  These include machine idioms for no-ops, e.g. mov esi,esi
		//  or xchg ax,ax or lea esi,[esi].
		this->Defs.clear();
		this->Uses.clear();
		this->MoveSource = InitOp;
		this->DEFMemOp = InitOp;
		this->USEMemOp = InitOp;
		this->SetLeaMemUseOp(InitOp);
		this->OptType = 1;
	}
#endif

	if (DebugFlag) {
		SMP_msg("DEBUG after MDFixupDefUseLists:\n");
		this->Dump();
	}
	return;
} // end of SMPInstr::MDFixupDefUseLists()

// If we can definitely identify which part of the addressing expression
//  used in MemOp is the POINTER type, and it is not a STACKPTR or GLOBALPTR
//  immediate, set the USE type for that register to POINTER and return true.
//  If we can find definite NUMERIC addressing registers that are not already
//  typed as NUMERIC, set their USE types to NUMERIC and return true.
bool SMPInstr::MDFindPointerUse(op_t MemOp, bool UseFP) {
	bool changed = false;
	int BaseReg;
	int IndexReg;
	op_t BaseOp = InitOp;
	op_t IndexOp = InitOp;
	SMPOperandType BaseType = UNKNOWN;
	SMPOperandType IndexType = UNKNOWN;
	ushort ScaleFactor;
	ea_t offset;
	set<DefOrUse, LessDefUse>::iterator BaseIter;
	set<DefOrUse, LessDefUse>::iterator IndexIter;

	if (NN_lea == this->SMPcmd.itype)
		return false;  // lea instruction really has no memory operands
	if (NN_fnop == this->SMPcmd.itype)
		return false;  // SSA marker instruction

	MDExtractAddressFields(MemOp, BaseReg, IndexReg, ScaleFactor, offset);
	if (R_none != IndexReg) {
		IndexOp.type = o_reg;
		IndexOp.reg = MDCanonicalizeSubReg((ushort) IndexReg);
		IndexOp.dtyp = this->GetOperandDtypField(); // Canonical reg width
		IndexIter = this->FindUse(IndexOp);
		assert(IndexIter != this->GetLastUse());
		IndexType = IndexIter->GetType();
	}
	if (R_none != BaseReg) {
		BaseOp.type = o_reg;
		BaseOp.reg = MDCanonicalizeSubReg((ushort) BaseReg);
		BaseOp.dtyp = this->GetOperandDtypField(); // Canonical reg width