1 //===- X86ISelDAGToDAG.cpp - A DAG pattern matching inst selector for X86 -===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the Evan Cheng and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines a DAG pattern matching instruction selector for X86,
11 // converting from a legalized dag to a X86 dag.
13 //===----------------------------------------------------------------------===//
15 #define DEBUG_TYPE "isel"
17 #include "X86InstrBuilder.h"
18 #include "X86ISelLowering.h"
19 #include "X86RegisterInfo.h"
20 #include "X86Subtarget.h"
21 #include "X86TargetMachine.h"
22 #include "llvm/GlobalValue.h"
23 #include "llvm/Instructions.h"
24 #include "llvm/Intrinsics.h"
25 #include "llvm/Support/CFG.h"
26 #include "llvm/CodeGen/MachineConstantPool.h"
27 #include "llvm/CodeGen/MachineFunction.h"
28 #include "llvm/CodeGen/MachineFrameInfo.h"
29 #include "llvm/CodeGen/MachineInstrBuilder.h"
30 #include "llvm/CodeGen/SSARegMap.h"
31 #include "llvm/CodeGen/SelectionDAGISel.h"
32 #include "llvm/Target/TargetMachine.h"
33 #include "llvm/Support/Debug.h"
34 #include "llvm/Support/Visibility.h"
35 #include "llvm/ADT/Statistic.h"
41 //===----------------------------------------------------------------------===//
42 // Pattern Matcher Implementation
43 //===----------------------------------------------------------------------===//
46 /// X86ISelAddressMode - This corresponds to X86AddressMode, but uses
47 /// SDOperand's instead of register numbers for the leaves of the matched
49 struct X86ISelAddressMode {
55 struct { // This is really a union, discriminated by BaseType!
65 unsigned Align; // CP alignment.
68 : BaseType(RegBase), Scale(1), IndexReg(), Disp(0), GV(0),
76 NumFPKill("x86-codegen", "Number of FP_REG_KILL instructions added");
78 //===--------------------------------------------------------------------===//
79 /// ISel - X86 specific code to select X86 machine instructions for
80 /// SelectionDAG operations.
82 class VISIBILITY_HIDDEN X86DAGToDAGISel : public SelectionDAGISel {
83 /// ContainsFPCode - Every instruction we select that uses or defines a FP
84 /// register should set this to true.
87 /// X86Lowering - This object fully describes how to lower LLVM code to an
88 /// X86-specific SelectionDAG.
89 X86TargetLowering X86Lowering;
91 /// Subtarget - Keep a pointer to the X86Subtarget around so that we can
92 /// make the right decision when generating code for different targets.
93 const X86Subtarget *Subtarget;
95 unsigned GlobalBaseReg;
98 X86DAGToDAGISel(X86TargetMachine &TM)
99 : SelectionDAGISel(X86Lowering),
100 X86Lowering(*TM.getTargetLowering()),
101 Subtarget(&TM.getSubtarget<X86Subtarget>()),
102 DAGSize(0), TopOrder(NULL), IdToOrder(NULL),
103 RMRange(NULL), ReachibilityMatrix(NULL) {}
105 virtual bool runOnFunction(Function &Fn) {
106 // Make sure we re-emit a set of the global base reg if necessary
108 return SelectionDAGISel::runOnFunction(Fn);
111 virtual const char *getPassName() const {
112 return "X86 DAG->DAG Instruction Selection";
115 /// InstructionSelectBasicBlock - This callback is invoked by
116 /// SelectionDAGISel when it has created a SelectionDAG for us to codegen.
117 virtual void InstructionSelectBasicBlock(SelectionDAG &DAG);
119 virtual void EmitFunctionEntryCode(Function &Fn, MachineFunction &MF);
121 virtual bool IsFoldableBy(SDNode *N, SDNode *U);
123 // Include the pieces autogenerated from the target description.
124 #include "X86GenDAGISel.inc"
127 void DetermineTopologicalOrdering();
128 void DeterminReachibility(SDNode *f, SDNode *t);
130 void Select(SDOperand &Result, SDOperand N);
132 bool MatchAddress(SDOperand N, X86ISelAddressMode &AM, bool isRoot = true);
133 bool SelectAddr(SDOperand N, SDOperand &Base, SDOperand &Scale,
134 SDOperand &Index, SDOperand &Disp);
135 bool SelectLEAAddr(SDOperand N, SDOperand &Base, SDOperand &Scale,
136 SDOperand &Index, SDOperand &Disp);
137 bool TryFoldLoad(SDOperand P, SDOperand N,
138 SDOperand &Base, SDOperand &Scale,
139 SDOperand &Index, SDOperand &Disp);
140 /// SelectInlineAsmMemoryOperand - Implement addressing mode selection for
141 /// inline asm expressions.
142 virtual bool SelectInlineAsmMemoryOperand(const SDOperand &Op,
144 std::vector<SDOperand> &OutOps,
147 void EmitSpecialCodeForMain(MachineBasicBlock *BB, MachineFrameInfo *MFI);
149 inline void getAddressOperands(X86ISelAddressMode &AM, SDOperand &Base,
150 SDOperand &Scale, SDOperand &Index,
152 Base = (AM.BaseType == X86ISelAddressMode::FrameIndexBase) ?
153 CurDAG->getTargetFrameIndex(AM.Base.FrameIndex, MVT::i32) : AM.Base.Reg;
154 Scale = getI8Imm(AM.Scale);
156 Disp = AM.GV ? CurDAG->getTargetGlobalAddress(AM.GV, MVT::i32, AM.Disp)
158 CurDAG->getTargetConstantPool(AM.CP, MVT::i32, AM.Align, AM.Disp)
159 : getI32Imm(AM.Disp));
162 /// getI8Imm - Return a target constant with the specified value, of type
164 inline SDOperand getI8Imm(unsigned Imm) {
165 return CurDAG->getTargetConstant(Imm, MVT::i8);
168 /// getI16Imm - Return a target constant with the specified value, of type
170 inline SDOperand getI16Imm(unsigned Imm) {
171 return CurDAG->getTargetConstant(Imm, MVT::i16);
174 /// getI32Imm - Return a target constant with the specified value, of type
176 inline SDOperand getI32Imm(unsigned Imm) {
177 return CurDAG->getTargetConstant(Imm, MVT::i32);
180 /// getGlobalBaseReg - insert code into the entry mbb to materialize the PIC
181 /// base register. Return the virtual register that holds this value.
182 SDOperand getGlobalBaseReg();
184 /// DAGSize - Number of nodes in the DAG.
188 /// TopOrder - Topological ordering of all nodes in the DAG.
192 /// IdToOrder - Node id to topological order map.
196 /// RMRange - The range of reachibility information available for the
197 /// particular source node.
200 /// ReachibilityMatrix - A N x N matrix representing all pairs reachibility
201 /// information. One bit per potential edge.
202 unsigned char *ReachibilityMatrix;
204 inline void setReachable(SDNode *f, SDNode *t) {
205 unsigned Idx = f->getNodeId() * DAGSize + t->getNodeId();
206 ReachibilityMatrix[Idx / 8] |= 1 << (Idx % 8);
209 inline bool isReachable(SDNode *f, SDNode *t) {
210 unsigned Idx = f->getNodeId() * DAGSize + t->getNodeId();
211 return ReachibilityMatrix[Idx / 8] & (1 << (Idx % 8));
214 /// UnfoldableSet - An boolean array representing nodes which have been
215 /// folded into addressing modes and therefore should not be folded in
216 /// another operation.
217 unsigned char *UnfoldableSet;
219 inline void setUnfoldable(SDNode *N) {
220 unsigned Id = N->getNodeId();
221 UnfoldableSet[Id / 8] |= 1 << (Id % 8);
224 inline bool isUnfoldable(SDNode *N) {
225 unsigned Id = N->getNodeId();
226 return UnfoldableSet[Id / 8] & (1 << (Id % 8));
235 bool X86DAGToDAGISel::IsFoldableBy(SDNode *N, SDNode *U) {
236 // Is it already folded by SelectAddr / SelectLEAAddr?
240 // If U use can somehow reach N through another path then U can't fold N or
241 // it will create a cycle. e.g. In the following diagram, U can reach N
242 // through X. If N is foled into into U, then X is both a predecessor and
252 DeterminReachibility(U, N);
253 assert(isReachable(U, N) && "Attempting to fold a non-operand node?");
254 for (SDNode::op_iterator I = U->op_begin(), E = U->op_end(); I != E; ++I) {
256 if (P != N && isReachable(P, N))
262 /// DetermineTopologicalOrdering - Determine topological ordering of the nodes
264 void X86DAGToDAGISel::DetermineTopologicalOrdering() {
265 TopOrder = new SDNode*[DAGSize];
266 IdToOrder = new unsigned[DAGSize];
267 memset(IdToOrder, 0, DAGSize * sizeof(unsigned));
268 RMRange = new unsigned[DAGSize];
269 memset(RMRange, 0, DAGSize * sizeof(unsigned));
271 std::vector<unsigned> InDegree(DAGSize);
272 std::list<SDNode*> Sources;
273 for (SelectionDAG::allnodes_iterator I = CurDAG->allnodes_begin(),
274 E = CurDAG->allnodes_end(); I != E; ++I) {
276 unsigned Degree = N->use_size();
277 InDegree[N->getNodeId()] = Degree;
279 Sources.push_back(I);
283 while (!Sources.empty()) {
284 SDNode *N = Sources.front();
287 IdToOrder[N->getNodeId()] = Order;
289 for (SDNode::op_iterator I = N->op_begin(), E = N->op_end(); I != E; ++I) {
291 int PId = P->getNodeId();
292 unsigned Degree = InDegree[PId] - 1;
294 Sources.push_back(P);
295 InDegree[PId] = Degree;
300 void X86DAGToDAGISel::DeterminReachibility(SDNode *f, SDNode *t) {
301 if (!ReachibilityMatrix) {
302 unsigned RMSize = (DAGSize * DAGSize + 7) / 8;
303 ReachibilityMatrix = new unsigned char[RMSize];
304 memset(ReachibilityMatrix, 0, RMSize);
307 int Idf = f->getNodeId();
308 int Idt = t->getNodeId();
309 unsigned Orderf = IdToOrder[Idf];
310 unsigned Ordert = IdToOrder[Idt];
311 unsigned Range = RMRange[Idf];
317 for (unsigned i = Range; i < Ordert; ++i) {
318 SDNode *N = TopOrder[i];
320 // If N is a leaf node, there is nothing more to do.
321 if (N->getNumOperands() == 0)
324 for (unsigned i2 = Orderf; ; ++i2) {
325 SDNode *M = TopOrder[i2];
326 if (isReachable(M, N)) {
327 // Update reachibility from M to N's operands.
328 for (SDNode::op_iterator I = N->op_begin(), E = N->op_end(); I != E;++I)
329 setReachable(M, I->Val);
335 RMRange[Idf] = Ordert;
338 /// InstructionSelectBasicBlock - This callback is invoked by SelectionDAGISel
339 /// when it has created a SelectionDAG for us to codegen.
340 void X86DAGToDAGISel::InstructionSelectBasicBlock(SelectionDAG &DAG) {
342 MachineFunction::iterator FirstMBB = BB;
344 DAGSize = DAG.AssignNodeIds();
345 unsigned NumBytes = (DAGSize+7) / 8;
346 UnfoldableSet = new unsigned char[NumBytes];
347 memset(UnfoldableSet, 0, NumBytes);
349 DetermineTopologicalOrdering();
351 // Codegen the basic block.
353 DEBUG(std::cerr << "===== Instruction selection begins:\n");
356 DAG.setRoot(SelectRoot(DAG.getRoot()));
358 DEBUG(std::cerr << "===== Instruction selection ends:\n");
361 delete[] ReachibilityMatrix;
365 delete[] UnfoldableSet;
366 ReachibilityMatrix = NULL;
368 IdToOrder = RMRange = NULL;
369 UnfoldableSet = NULL;
373 DAG.RemoveDeadNodes();
375 // Emit machine code to BB.
376 ScheduleAndEmitDAG(DAG);
378 // If we are emitting FP stack code, scan the basic block to determine if this
379 // block defines any FP values. If so, put an FP_REG_KILL instruction before
380 // the terminator of the block.
381 if (!Subtarget->hasSSE2()) {
382 // Note that FP stack instructions *are* used in SSE code when returning
383 // values, but these are not live out of the basic block, so we don't need
384 // an FP_REG_KILL in this case either.
385 bool ContainsFPCode = false;
387 // Scan all of the machine instructions in these MBBs, checking for FP
389 MachineFunction::iterator MBBI = FirstMBB;
391 for (MachineBasicBlock::iterator I = MBBI->begin(), E = MBBI->end();
392 !ContainsFPCode && I != E; ++I) {
393 for (unsigned op = 0, e = I->getNumOperands(); op != e; ++op) {
394 if (I->getOperand(op).isRegister() && I->getOperand(op).isDef() &&
395 MRegisterInfo::isVirtualRegister(I->getOperand(op).getReg()) &&
396 RegMap->getRegClass(I->getOperand(0).getReg()) ==
397 X86::RFPRegisterClass) {
398 ContainsFPCode = true;
403 } while (!ContainsFPCode && &*(MBBI++) != BB);
405 // Check PHI nodes in successor blocks. These PHI's will be lowered to have
406 // a copy of the input value in this block.
407 if (!ContainsFPCode) {
408 // Final check, check LLVM BB's that are successors to the LLVM BB
409 // corresponding to BB for FP PHI nodes.
410 const BasicBlock *LLVMBB = BB->getBasicBlock();
412 for (succ_const_iterator SI = succ_begin(LLVMBB), E = succ_end(LLVMBB);
413 !ContainsFPCode && SI != E; ++SI) {
414 for (BasicBlock::const_iterator II = SI->begin();
415 (PN = dyn_cast<PHINode>(II)); ++II) {
416 if (PN->getType()->isFloatingPoint()) {
417 ContainsFPCode = true;
424 // Finally, if we found any FP code, emit the FP_REG_KILL instruction.
425 if (ContainsFPCode) {
426 BuildMI(*BB, BB->getFirstTerminator(), X86::FP_REG_KILL, 0);
432 /// EmitSpecialCodeForMain - Emit any code that needs to be executed only in
433 /// the main function.
434 void X86DAGToDAGISel::EmitSpecialCodeForMain(MachineBasicBlock *BB,
435 MachineFrameInfo *MFI) {
436 if (Subtarget->TargetType == X86Subtarget::isCygwin)
437 BuildMI(BB, X86::CALLpcrel32, 1).addExternalSymbol("__main");
439 // Switch the FPU to 64-bit precision mode for better compatibility and speed.
440 int CWFrameIdx = MFI->CreateStackObject(2, 2);
441 addFrameReference(BuildMI(BB, X86::FNSTCW16m, 4), CWFrameIdx);
443 // Set the high part to be 64-bit precision.
444 addFrameReference(BuildMI(BB, X86::MOV8mi, 5),
445 CWFrameIdx, 1).addImm(2);
447 // Reload the modified control word now.
448 addFrameReference(BuildMI(BB, X86::FLDCW16m, 4), CWFrameIdx);
451 void X86DAGToDAGISel::EmitFunctionEntryCode(Function &Fn, MachineFunction &MF) {
452 // If this is main, emit special code for main.
453 MachineBasicBlock *BB = MF.begin();
454 if (Fn.hasExternalLinkage() && Fn.getName() == "main")
455 EmitSpecialCodeForMain(BB, MF.getFrameInfo());
458 /// MatchAddress - Add the specified node to the specified addressing mode,
459 /// returning true if it cannot be done. This just pattern matches for the
461 bool X86DAGToDAGISel::MatchAddress(SDOperand N, X86ISelAddressMode &AM,
463 bool Available = false;
464 // If N has already been selected, reuse the result unless in some very
466 std::map<SDOperand, SDOperand>::iterator CGMI= CodeGenMap.find(N.getValue(0));
467 if (CGMI != CodeGenMap.end()) {
471 switch (N.getOpcode()) {
474 AM.Disp += cast<ConstantSDNode>(N)->getValue();
477 case X86ISD::Wrapper:
478 // If both base and index components have been picked, we can't fit
479 // the result available in the register in the addressing mode. Duplicate
480 // GlobalAddress or ConstantPool as displacement.
481 if (!Available || (AM.Base.Reg.Val && AM.IndexReg.Val)) {
482 if (ConstantPoolSDNode *CP =
483 dyn_cast<ConstantPoolSDNode>(N.getOperand(0))) {
486 AM.Align = CP->getAlignment();
487 AM.Disp += CP->getOffset();
490 } else if (GlobalAddressSDNode *G =
491 dyn_cast<GlobalAddressSDNode>(N.getOperand(0))) {
493 AM.GV = G->getGlobal();
494 AM.Disp += G->getOffset();
501 case ISD::FrameIndex:
502 if (AM.BaseType == X86ISelAddressMode::RegBase && AM.Base.Reg.Val == 0) {
503 AM.BaseType = X86ISelAddressMode::FrameIndexBase;
504 AM.Base.FrameIndex = cast<FrameIndexSDNode>(N)->getIndex();
510 if (!Available && AM.IndexReg.Val == 0 && AM.Scale == 1)
511 if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N.Val->getOperand(1))) {
512 unsigned Val = CN->getValue();
513 if (Val == 1 || Val == 2 || Val == 3) {
515 SDOperand ShVal = N.Val->getOperand(0);
517 // Okay, we know that we have a scale by now. However, if the scaled
518 // value is an add of something and a constant, we can fold the
519 // constant into the disp field here.
520 if (ShVal.Val->getOpcode() == ISD::ADD && ShVal.hasOneUse() &&
521 isa<ConstantSDNode>(ShVal.Val->getOperand(1))) {
522 AM.IndexReg = ShVal.Val->getOperand(0);
523 ConstantSDNode *AddVal =
524 cast<ConstantSDNode>(ShVal.Val->getOperand(1));
525 AM.Disp += AddVal->getValue() << Val;
535 // X*[3,5,9] -> X+X*[2,4,8]
537 AM.BaseType == X86ISelAddressMode::RegBase &&
538 AM.Base.Reg.Val == 0 &&
539 AM.IndexReg.Val == 0)
540 if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N.Val->getOperand(1)))
541 if (CN->getValue() == 3 || CN->getValue() == 5 || CN->getValue() == 9) {
542 AM.Scale = unsigned(CN->getValue())-1;
544 SDOperand MulVal = N.Val->getOperand(0);
547 // Okay, we know that we have a scale by now. However, if the scaled
548 // value is an add of something and a constant, we can fold the
549 // constant into the disp field here.
550 if (MulVal.Val->getOpcode() == ISD::ADD && MulVal.hasOneUse() &&
551 isa<ConstantSDNode>(MulVal.Val->getOperand(1))) {
552 Reg = MulVal.Val->getOperand(0);
553 ConstantSDNode *AddVal =
554 cast<ConstantSDNode>(MulVal.Val->getOperand(1));
555 AM.Disp += AddVal->getValue() * CN->getValue();
557 Reg = N.Val->getOperand(0);
560 AM.IndexReg = AM.Base.Reg = Reg;
567 X86ISelAddressMode Backup = AM;
568 if (!MatchAddress(N.Val->getOperand(0), AM, false) &&
569 !MatchAddress(N.Val->getOperand(1), AM, false))
572 if (!MatchAddress(N.Val->getOperand(1), AM, false) &&
573 !MatchAddress(N.Val->getOperand(0), AM, false))
582 X86ISelAddressMode Backup = AM;
583 // Look for (x << c1) | c2 where (c2 < c1)
584 ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N.Val->getOperand(0));
585 if (CN && !MatchAddress(N.Val->getOperand(1), AM, false)) {
586 if (AM.GV == NULL && AM.Disp == 0 && CN->getValue() < AM.Scale) {
587 AM.Disp = CN->getValue();
592 CN = dyn_cast<ConstantSDNode>(N.Val->getOperand(1));
593 if (CN && !MatchAddress(N.Val->getOperand(0), AM, false)) {
594 if (AM.GV == NULL && AM.Disp == 0 && CN->getValue() < AM.Scale) {
595 AM.Disp = CN->getValue();
605 // Is the base register already occupied?
606 if (AM.BaseType != X86ISelAddressMode::RegBase || AM.Base.Reg.Val) {
607 // If so, check to see if the scale index register is set.
608 if (AM.IndexReg.Val == 0) {
614 // Otherwise, we cannot select it.
618 // Default, generate it as a register.
619 AM.BaseType = X86ISelAddressMode::RegBase;
624 /// SelectAddr - returns true if it is able pattern match an addressing mode.
625 /// It returns the operands which make up the maximal addressing mode it can
626 /// match by reference.
627 bool X86DAGToDAGISel::SelectAddr(SDOperand N, SDOperand &Base, SDOperand &Scale,
628 SDOperand &Index, SDOperand &Disp) {
629 X86ISelAddressMode AM;
630 if (MatchAddress(N, AM))
633 if (AM.BaseType == X86ISelAddressMode::RegBase) {
634 if (!AM.Base.Reg.Val)
635 AM.Base.Reg = CurDAG->getRegister(0, MVT::i32);
638 if (!AM.IndexReg.Val)
639 AM.IndexReg = CurDAG->getRegister(0, MVT::i32);
641 getAddressOperands(AM, Base, Scale, Index, Disp);
643 int Id = Base.Val ? Base.Val->getNodeId() : -1;
645 setUnfoldable(Base.Val);
646 Id = Index.Val ? Index.Val->getNodeId() : -1;
648 setUnfoldable(Index.Val);
653 /// SelectLEAAddr - it calls SelectAddr and determines if the maximal addressing
654 /// mode it matches can be cost effectively emitted as an LEA instruction.
655 bool X86DAGToDAGISel::SelectLEAAddr(SDOperand N, SDOperand &Base,
657 SDOperand &Index, SDOperand &Disp) {
658 X86ISelAddressMode AM;
659 if (MatchAddress(N, AM))
662 unsigned Complexity = 0;
663 if (AM.BaseType == X86ISelAddressMode::RegBase)
667 AM.Base.Reg = CurDAG->getRegister(0, MVT::i32);
668 else if (AM.BaseType == X86ISelAddressMode::FrameIndexBase)
674 AM.IndexReg = CurDAG->getRegister(0, MVT::i32);
678 // Don't match just leal(,%reg,2). It's cheaper to do addl %reg, %reg
679 else if (AM.Scale > 1)
682 // FIXME: We are artificially lowering the criteria to turn ADD %reg, $GA
683 // to a LEA. This is determined with some expermentation but is by no means
684 // optimal (especially for code size consideration). LEA is nice because of
685 // its three-address nature. Tweak the cost function again when we can run
686 // convertToThreeAddress() at register allocation time.
690 if (AM.Disp && (AM.Base.Reg.Val || AM.IndexReg.Val))
693 if (Complexity > 2) {
694 getAddressOperands(AM, Base, Scale, Index, Disp);
698 int Id = Base.Val ? Base.Val->getNodeId() : -1;
700 setUnfoldable(Base.Val);
701 Id = Index.Val ? Index.Val->getNodeId() : -1;
703 setUnfoldable(Index.Val);
708 bool X86DAGToDAGISel::TryFoldLoad(SDOperand P, SDOperand N,
709 SDOperand &Base, SDOperand &Scale,
710 SDOperand &Index, SDOperand &Disp) {
711 if (N.getOpcode() == ISD::LOAD &&
713 !CodeGenMap.count(N.getValue(0)) &&
714 !IsFoldableBy(N.Val, P.Val))
715 return SelectAddr(N.getOperand(1), Base, Scale, Index, Disp);
719 static bool isRegister0(SDOperand Op) {
720 if (RegisterSDNode *R = dyn_cast<RegisterSDNode>(Op))
721 return (R->getReg() == 0);
725 /// getGlobalBaseReg - Output the instructions required to put the
726 /// base address to use for accessing globals into a register.
728 SDOperand X86DAGToDAGISel::getGlobalBaseReg() {
729 if (!GlobalBaseReg) {
730 // Insert the set of GlobalBaseReg into the first MBB of the function
731 MachineBasicBlock &FirstMBB = BB->getParent()->front();
732 MachineBasicBlock::iterator MBBI = FirstMBB.begin();
733 SSARegMap *RegMap = BB->getParent()->getSSARegMap();
734 // FIXME: when we get to LP64, we will need to create the appropriate
735 // type of register here.
736 GlobalBaseReg = RegMap->createVirtualRegister(X86::GR32RegisterClass);
737 BuildMI(FirstMBB, MBBI, X86::MovePCtoStack, 0);
738 BuildMI(FirstMBB, MBBI, X86::POP32r, 1, GlobalBaseReg);
740 return CurDAG->getRegister(GlobalBaseReg, MVT::i32);
743 static SDNode *FindCallStartFromCall(SDNode *Node) {
744 if (Node->getOpcode() == ISD::CALLSEQ_START) return Node;
745 assert(Node->getOperand(0).getValueType() == MVT::Other &&
746 "Node doesn't have a token chain argument!");
747 return FindCallStartFromCall(Node->getOperand(0).Val);
750 void X86DAGToDAGISel::Select(SDOperand &Result, SDOperand N) {
751 SDNode *Node = N.Val;
752 MVT::ValueType NVT = Node->getValueType(0);
754 unsigned Opcode = Node->getOpcode();
757 DEBUG(std::cerr << std::string(Indent, ' '));
758 DEBUG(std::cerr << "Selecting: ");
759 DEBUG(Node->dump(CurDAG));
760 DEBUG(std::cerr << "\n");
764 if (Opcode >= ISD::BUILTIN_OP_END && Opcode < X86ISD::FIRST_NUMBER) {
767 DEBUG(std::cerr << std::string(Indent-2, ' '));
768 DEBUG(std::cerr << "== ");
769 DEBUG(Node->dump(CurDAG));
770 DEBUG(std::cerr << "\n");
773 return; // Already selected.
776 std::map<SDOperand, SDOperand>::iterator CGMI = CodeGenMap.find(N);
777 if (CGMI != CodeGenMap.end()) {
778 Result = CGMI->second;
780 DEBUG(std::cerr << std::string(Indent-2, ' '));
781 DEBUG(std::cerr << "== ");
782 DEBUG(Result.Val->dump(CurDAG));
783 DEBUG(std::cerr << "\n");
791 case X86ISD::GlobalBaseReg:
792 Result = getGlobalBaseReg();
796 // Turn ADD X, c to MOV32ri X+c. This cannot be done with tblgen'd
797 // code and is matched first so to prevent it from being turned into
799 SDOperand N0 = N.getOperand(0);
800 SDOperand N1 = N.getOperand(1);
801 if (N.Val->getValueType(0) == MVT::i32 &&
802 N0.getOpcode() == X86ISD::Wrapper &&
803 N1.getOpcode() == ISD::Constant) {
804 unsigned Offset = (unsigned)cast<ConstantSDNode>(N1)->getValue();
806 // TODO: handle ExternalSymbolSDNode.
807 if (GlobalAddressSDNode *G =
808 dyn_cast<GlobalAddressSDNode>(N0.getOperand(0))) {
809 C = CurDAG->getTargetGlobalAddress(G->getGlobal(), MVT::i32,
810 G->getOffset() + Offset);
811 } else if (ConstantPoolSDNode *CP =
812 dyn_cast<ConstantPoolSDNode>(N0.getOperand(0))) {
813 C = CurDAG->getTargetConstantPool(CP->get(), MVT::i32,
815 CP->getOffset()+Offset);
819 if (N.Val->hasOneUse()) {
820 Result = CurDAG->SelectNodeTo(N.Val, X86::MOV32ri, MVT::i32, C);
822 SDNode *ResNode = CurDAG->getTargetNode(X86::MOV32ri, MVT::i32, C);
823 Result = CodeGenMap[N] = SDOperand(ResNode, 0);
829 // Other cases are handled by auto-generated code.
835 if (Opcode == ISD::MULHU)
837 default: assert(0 && "Unsupported VT!");
838 case MVT::i8: Opc = X86::MUL8r; MOpc = X86::MUL8m; break;
839 case MVT::i16: Opc = X86::MUL16r; MOpc = X86::MUL16m; break;
840 case MVT::i32: Opc = X86::MUL32r; MOpc = X86::MUL32m; break;
844 default: assert(0 && "Unsupported VT!");
845 case MVT::i8: Opc = X86::IMUL8r; MOpc = X86::IMUL8m; break;
846 case MVT::i16: Opc = X86::IMUL16r; MOpc = X86::IMUL16m; break;
847 case MVT::i32: Opc = X86::IMUL32r; MOpc = X86::IMUL32m; break;
850 unsigned LoReg, HiReg;
852 default: assert(0 && "Unsupported VT!");
853 case MVT::i8: LoReg = X86::AL; HiReg = X86::AH; break;
854 case MVT::i16: LoReg = X86::AX; HiReg = X86::DX; break;
855 case MVT::i32: LoReg = X86::EAX; HiReg = X86::EDX; break;
858 SDOperand N0 = Node->getOperand(0);
859 SDOperand N1 = Node->getOperand(1);
861 bool foldedLoad = false;
862 SDOperand Tmp0, Tmp1, Tmp2, Tmp3;
863 foldedLoad = TryFoldLoad(N, N1, Tmp0, Tmp1, Tmp2, Tmp3);
864 // MULHU and MULHS are commmutative
866 foldedLoad = TryFoldLoad(N, N0, Tmp0, Tmp1, Tmp2, Tmp3);
868 N0 = Node->getOperand(1);
869 N1 = Node->getOperand(0);
875 Select(Chain, N1.getOperand(0));
877 Chain = CurDAG->getEntryNode();
879 SDOperand InFlag(0, 0);
881 Chain = CurDAG->getCopyToReg(Chain, CurDAG->getRegister(LoReg, NVT),
883 InFlag = Chain.getValue(1);
891 CurDAG->getTargetNode(MOpc, MVT::Other, MVT::Flag, Tmp0, Tmp1,
892 Tmp2, Tmp3, Chain, InFlag);
893 Chain = SDOperand(CNode, 0);
894 InFlag = SDOperand(CNode, 1);
898 SDOperand(CurDAG->getTargetNode(Opc, MVT::Flag, N1, InFlag), 0);
901 Result = CurDAG->getCopyFromReg(Chain, HiReg, NVT, InFlag);
902 CodeGenMap[N.getValue(0)] = Result;
904 CodeGenMap[N1.getValue(1)] = Result.getValue(1);
905 AddHandleReplacement(N1.Val, 1, Result.Val, 1);
909 DEBUG(std::cerr << std::string(Indent-2, ' '));
910 DEBUG(std::cerr << "== ");
911 DEBUG(Result.Val->dump(CurDAG));
912 DEBUG(std::cerr << "\n");
922 bool isSigned = Opcode == ISD::SDIV || Opcode == ISD::SREM;
923 bool isDiv = Opcode == ISD::SDIV || Opcode == ISD::UDIV;
926 default: assert(0 && "Unsupported VT!");
927 case MVT::i8: Opc = X86::DIV8r; MOpc = X86::DIV8m; break;
928 case MVT::i16: Opc = X86::DIV16r; MOpc = X86::DIV16m; break;
929 case MVT::i32: Opc = X86::DIV32r; MOpc = X86::DIV32m; break;
933 default: assert(0 && "Unsupported VT!");
934 case MVT::i8: Opc = X86::IDIV8r; MOpc = X86::IDIV8m; break;
935 case MVT::i16: Opc = X86::IDIV16r; MOpc = X86::IDIV16m; break;
936 case MVT::i32: Opc = X86::IDIV32r; MOpc = X86::IDIV32m; break;
939 unsigned LoReg, HiReg;
940 unsigned ClrOpcode, SExtOpcode;
942 default: assert(0 && "Unsupported VT!");
944 LoReg = X86::AL; HiReg = X86::AH;
945 ClrOpcode = X86::MOV8r0;
946 SExtOpcode = X86::CBW;
949 LoReg = X86::AX; HiReg = X86::DX;
950 ClrOpcode = X86::MOV16r0;
951 SExtOpcode = X86::CWD;
954 LoReg = X86::EAX; HiReg = X86::EDX;
955 ClrOpcode = X86::MOV32r0;
956 SExtOpcode = X86::CDQ;
960 SDOperand N0 = Node->getOperand(0);
961 SDOperand N1 = Node->getOperand(1);
963 bool foldedLoad = false;
964 SDOperand Tmp0, Tmp1, Tmp2, Tmp3;
965 foldedLoad = TryFoldLoad(N, N1, Tmp0, Tmp1, Tmp2, Tmp3);
968 Select(Chain, N1.getOperand(0));
970 Chain = CurDAG->getEntryNode();
972 SDOperand InFlag(0, 0);
974 Chain = CurDAG->getCopyToReg(Chain, CurDAG->getRegister(LoReg, NVT),
976 InFlag = Chain.getValue(1);
979 // Sign extend the low part into the high part.
981 SDOperand(CurDAG->getTargetNode(SExtOpcode, MVT::Flag, InFlag), 0);
983 // Zero out the high part, effectively zero extending the input.
984 SDOperand ClrNode = SDOperand(CurDAG->getTargetNode(ClrOpcode, NVT), 0);
985 Chain = CurDAG->getCopyToReg(Chain, CurDAG->getRegister(HiReg, NVT),
987 InFlag = Chain.getValue(1);
996 CurDAG->getTargetNode(MOpc, MVT::Other, MVT::Flag, Tmp0, Tmp1,
997 Tmp2, Tmp3, Chain, InFlag);
998 Chain = SDOperand(CNode, 0);
999 InFlag = SDOperand(CNode, 1);
1003 SDOperand(CurDAG->getTargetNode(Opc, MVT::Flag, N1, InFlag), 0);
1006 Result = CurDAG->getCopyFromReg(Chain, isDiv ? LoReg : HiReg,
1008 CodeGenMap[N.getValue(0)] = Result;
1010 CodeGenMap[N1.getValue(1)] = Result.getValue(1);
1011 AddHandleReplacement(N1.Val, 1, Result.Val, 1);
1015 DEBUG(std::cerr << std::string(Indent-2, ' '));
1016 DEBUG(std::cerr << "== ");
1017 DEBUG(Result.Val->dump(CurDAG));
1018 DEBUG(std::cerr << "\n");
1024 case ISD::TRUNCATE: {
1025 if (NVT == MVT::i8) {
1028 switch (Node->getOperand(0).getValueType()) {
1029 default: assert(0 && "Unknown truncate!");
1031 Opc = X86::MOV16to16_;
1033 Opc2 = X86::TRUNC_GR16_GR8;
1036 Opc = X86::MOV32to32_;
1038 Opc2 = X86::TRUNC_GR32_GR8;
1042 SDOperand Tmp0, Tmp1;
1043 Select(Tmp0, Node->getOperand(0));
1044 Tmp1 = SDOperand(CurDAG->getTargetNode(Opc, VT, Tmp0), 0);
1045 Result = CodeGenMap[N] =
1046 SDOperand(CurDAG->getTargetNode(Opc2, NVT, Tmp1), 0);
1049 DEBUG(std::cerr << std::string(Indent-2, ' '));
1050 DEBUG(std::cerr << "== ");
1051 DEBUG(Result.Val->dump(CurDAG));
1052 DEBUG(std::cerr << "\n");
1062 SelectCode(Result, N);
1064 DEBUG(std::cerr << std::string(Indent-2, ' '));
1065 DEBUG(std::cerr << "=> ");
1066 DEBUG(Result.Val->dump(CurDAG));
1067 DEBUG(std::cerr << "\n");
1072 bool X86DAGToDAGISel::
1073 SelectInlineAsmMemoryOperand(const SDOperand &Op, char ConstraintCode,
1074 std::vector<SDOperand> &OutOps, SelectionDAG &DAG){
1075 SDOperand Op0, Op1, Op2, Op3;
1076 switch (ConstraintCode) {
1077 case 'o': // offsetable ??
1078 case 'v': // not offsetable ??
1079 default: return true;
1081 if (!SelectAddr(Op, Op0, Op1, Op2, Op3))
1087 Select(OutOps[0], Op0);
1088 Select(OutOps[1], Op1);
1089 Select(OutOps[2], Op2);
1090 Select(OutOps[3], Op3);
1094 /// createX86ISelDag - This pass converts a legalized DAG into a
1095 /// X86-specific DAG, ready for instruction scheduling.
1097 FunctionPass *llvm::createX86ISelDag(X86TargetMachine &TM) {
1098 return new X86DAGToDAGISel(TM);