X86RegisterInfo.cpp revision faf2671776da307af2ae73d94597d2934ef406b5 
1//===- X86RegisterInfo.cpp - X86 Register Information -----------*- C++ -*-===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file contains the X86 implementation of the TargetRegisterInfo class.
11// This file is responsible for the frame pointer elimination optimization
12// on X86.
13//
14//===----------------------------------------------------------------------===//
15
16#include "X86.h"
17#include "X86RegisterInfo.h"
18#include "X86InstrBuilder.h"
19#include "X86MachineFunctionInfo.h"
20#include "X86Subtarget.h"
21#include "X86TargetMachine.h"
22#include "llvm/Constants.h"
23#include "llvm/Function.h"
24#include "llvm/Type.h"
25#include "llvm/CodeGen/ValueTypes.h"
26#include "llvm/CodeGen/MachineInstrBuilder.h"
27#include "llvm/CodeGen/MachineFunction.h"
28#include "llvm/CodeGen/MachineFunctionPass.h"
29#include "llvm/CodeGen/MachineFrameInfo.h"
30#include "llvm/CodeGen/MachineLocation.h"
31#include "llvm/CodeGen/MachineModuleInfo.h"
32#include "llvm/CodeGen/MachineRegisterInfo.h"
33#include "llvm/Target/TargetAsmInfo.h"
34#include "llvm/Target/TargetFrameInfo.h"
35#include "llvm/Target/TargetInstrInfo.h"
36#include "llvm/Target/TargetMachine.h"
37#include "llvm/Target/TargetOptions.h"
38#include "llvm/ADT/BitVector.h"
39#include "llvm/ADT/STLExtras.h"
40#include "llvm/Support/Compiler.h"
41using namespace llvm;
42
43X86RegisterInfo::X86RegisterInfo(X86TargetMachine &tm,
44                                 const TargetInstrInfo &tii)
45  : X86GenRegisterInfo(tm.getSubtarget<X86Subtarget>().is64Bit() ?
46                         X86::ADJCALLSTACKDOWN64 :
47                         X86::ADJCALLSTACKDOWN32,
48                       tm.getSubtarget<X86Subtarget>().is64Bit() ?
49                         X86::ADJCALLSTACKUP64 :
50                         X86::ADJCALLSTACKUP32),
51    TM(tm), TII(tii) {
52  // Cache some information.
53  const X86Subtarget *Subtarget = &TM.getSubtarget<X86Subtarget>();
54  Is64Bit = Subtarget->is64Bit();
55  IsWin64 = Subtarget->isTargetWin64();
56  StackAlign = TM.getFrameInfo()->getStackAlignment();
57  if (Is64Bit) {
58    SlotSize = 8;
59    StackPtr = X86::RSP;
60    FramePtr = X86::RBP;
61  } else {
62    SlotSize = 4;
63    StackPtr = X86::ESP;
64    FramePtr = X86::EBP;
65  }
66}
67
68// getDwarfRegNum - This function maps LLVM register identifiers to the
69// Dwarf specific numbering, used in debug info and exception tables.
70
71int X86RegisterInfo::getDwarfRegNum(unsigned RegNo, bool isEH) const {
72  const X86Subtarget *Subtarget = &TM.getSubtarget<X86Subtarget>();
73  unsigned Flavour = DWARFFlavour::X86_64;
74  if (!Subtarget->is64Bit()) {
75    if (Subtarget->isTargetDarwin()) {
76      if (isEH)
77        Flavour = DWARFFlavour::X86_32_DarwinEH;
78      else
79        Flavour = DWARFFlavour::X86_32_Generic;
80    } else if (Subtarget->isTargetCygMing()) {
81      // Unsupported by now, just quick fallback
82      Flavour = DWARFFlavour::X86_32_Generic;
83    } else {
84      Flavour = DWARFFlavour::X86_32_Generic;
85    }
86  }
87
88  return X86GenRegisterInfo::getDwarfRegNumFull(RegNo, Flavour);
89}
90
91// getX86RegNum - This function maps LLVM register identifiers to their X86
92// specific numbering, which is used in various places encoding instructions.
93//
94unsigned X86RegisterInfo::getX86RegNum(unsigned RegNo) {
95  switch(RegNo) {
96  case X86::RAX: case X86::EAX: case X86::AX: case X86::AL: return N86::EAX;
97  case X86::RCX: case X86::ECX: case X86::CX: case X86::CL: return N86::ECX;
98  case X86::RDX: case X86::EDX: case X86::DX: case X86::DL: return N86::EDX;
99  case X86::RBX: case X86::EBX: case X86::BX: case X86::BL: return N86::EBX;
100  case X86::RSP: case X86::ESP: case X86::SP: case X86::SPL: case X86::AH:
101    return N86::ESP;
102  case X86::RBP: case X86::EBP: case X86::BP: case X86::BPL: case X86::CH:
103    return N86::EBP;
104  case X86::RSI: case X86::ESI: case X86::SI: case X86::SIL: case X86::DH:
105    return N86::ESI;
106  case X86::RDI: case X86::EDI: case X86::DI: case X86::DIL: case X86::BH:
107    return N86::EDI;
108
109  case X86::R8:  case X86::R8D:  case X86::R8W:  case X86::R8B:
110    return N86::EAX;
111  case X86::R9:  case X86::R9D:  case X86::R9W:  case X86::R9B:
112    return N86::ECX;
113  case X86::R10: case X86::R10D: case X86::R10W: case X86::R10B:
114    return N86::EDX;
115  case X86::R11: case X86::R11D: case X86::R11W: case X86::R11B:
116    return N86::EBX;
117  case X86::R12: case X86::R12D: case X86::R12W: case X86::R12B:
118    return N86::ESP;
119  case X86::R13: case X86::R13D: case X86::R13W: case X86::R13B:
120    return N86::EBP;
121  case X86::R14: case X86::R14D: case X86::R14W: case X86::R14B:
122    return N86::ESI;
123  case X86::R15: case X86::R15D: case X86::R15W: case X86::R15B:
124    return N86::EDI;
125
126  case X86::ST0: case X86::ST1: case X86::ST2: case X86::ST3:
127  case X86::ST4: case X86::ST5: case X86::ST6: case X86::ST7:
128    return RegNo-X86::ST0;
129
130  case X86::XMM0: case X86::XMM8: case X86::MM0:
131    return 0;
132  case X86::XMM1: case X86::XMM9: case X86::MM1:
133    return 1;
134  case X86::XMM2: case X86::XMM10: case X86::MM2:
135    return 2;
136  case X86::XMM3: case X86::XMM11: case X86::MM3:
137    return 3;
138  case X86::XMM4: case X86::XMM12: case X86::MM4:
139    return 4;
140  case X86::XMM5: case X86::XMM13: case X86::MM5:
141    return 5;
142  case X86::XMM6: case X86::XMM14: case X86::MM6:
143    return 6;
144  case X86::XMM7: case X86::XMM15: case X86::MM7:
145    return 7;
146
147  default:
148    assert(isVirtualRegister(RegNo) && "Unknown physical register!");
149    assert(0 && "Register allocator hasn't allocated reg correctly yet!");
150    return 0;
151  }
152}
153
154const TargetRegisterClass *X86RegisterInfo::getPointerRegClass() const {
155  const X86Subtarget *Subtarget = &TM.getSubtarget<X86Subtarget>();
156  if (Subtarget->is64Bit())
157    return &X86::GR64RegClass;
158  else
159    return &X86::GR32RegClass;
160}
161
162const TargetRegisterClass *
163X86RegisterInfo::getCrossCopyRegClass(const TargetRegisterClass *RC) const {
164  if (RC == &X86::CCRRegClass) {
165    if (Is64Bit)
166      return &X86::GR64RegClass;
167    else
168      return &X86::GR32RegClass;
169  }
170  return NULL;
171}
172
173const unsigned *
174X86RegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
175  bool callsEHReturn = false;
176
177  if (MF) {
178    const MachineFrameInfo *MFI = MF->getFrameInfo();
179    const MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
180    callsEHReturn = (MMI ? MMI->callsEHReturn() : false);
181  }
182
183  static const unsigned CalleeSavedRegs32Bit[] = {
184    X86::ESI, X86::EDI, X86::EBX, X86::EBP,  0
185  };
186
187  static const unsigned CalleeSavedRegs32EHRet[] = {
188    X86::EAX, X86::EDX, X86::ESI, X86::EDI, X86::EBX, X86::EBP,  0
189  };
190
191  static const unsigned CalleeSavedRegs64Bit[] = {
192    X86::RBX, X86::R12, X86::R13, X86::R14, X86::R15, X86::RBP, 0
193  };
194
195  static const unsigned CalleeSavedRegs64EHRet[] = {
196    X86::RAX, X86::RDX, X86::RBX, X86::R12,
197    X86::R13, X86::R14, X86::R15, X86::RBP, 0
198  };
199
200  static const unsigned CalleeSavedRegsWin64[] = {
201    X86::RBX,   X86::RBP,   X86::RDI,   X86::RSI,
202    X86::R12,   X86::R13,   X86::R14,   X86::R15,
203    X86::XMM6,  X86::XMM7,  X86::XMM8,  X86::XMM9,
204    X86::XMM10, X86::XMM11, X86::XMM12, X86::XMM13,
205    X86::XMM14, X86::XMM15, 0
206  };
207
208  if (Is64Bit) {
209    if (IsWin64)
210      return CalleeSavedRegsWin64;
211    else
212      return (callsEHReturn ? CalleeSavedRegs64EHRet : CalleeSavedRegs64Bit);
213  } else {
214    return (callsEHReturn ? CalleeSavedRegs32EHRet : CalleeSavedRegs32Bit);
215  }
216}
217
218const TargetRegisterClass* const*
219X86RegisterInfo::getCalleeSavedRegClasses(const MachineFunction *MF) const {
220  bool callsEHReturn = false;
221
222  if (MF) {
223    const MachineFrameInfo *MFI = MF->getFrameInfo();
224    const MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
225    callsEHReturn = (MMI ? MMI->callsEHReturn() : false);
226  }
227
228  static const TargetRegisterClass * const CalleeSavedRegClasses32Bit[] = {
229    &X86::GR32RegClass, &X86::GR32RegClass,
230    &X86::GR32RegClass, &X86::GR32RegClass,  0
231  };
232  static const TargetRegisterClass * const CalleeSavedRegClasses32EHRet[] = {
233    &X86::GR32RegClass, &X86::GR32RegClass,
234    &X86::GR32RegClass, &X86::GR32RegClass,
235    &X86::GR32RegClass, &X86::GR32RegClass,  0
236  };
237  static const TargetRegisterClass * const CalleeSavedRegClasses64Bit[] = {
238    &X86::GR64RegClass, &X86::GR64RegClass,
239    &X86::GR64RegClass, &X86::GR64RegClass,
240    &X86::GR64RegClass, &X86::GR64RegClass, 0
241  };
242  static const TargetRegisterClass * const CalleeSavedRegClasses64EHRet[] = {
243    &X86::GR64RegClass, &X86::GR64RegClass,
244    &X86::GR64RegClass, &X86::GR64RegClass,
245    &X86::GR64RegClass, &X86::GR64RegClass,
246    &X86::GR64RegClass, &X86::GR64RegClass, 0
247  };
248  static const TargetRegisterClass * const CalleeSavedRegClassesWin64[] = {
249    &X86::GR64RegClass,  &X86::GR64RegClass,
250    &X86::GR64RegClass,  &X86::GR64RegClass,
251    &X86::GR64RegClass,  &X86::GR64RegClass,
252    &X86::GR64RegClass,  &X86::GR64RegClass,
253    &X86::VR128RegClass, &X86::VR128RegClass,
254    &X86::VR128RegClass, &X86::VR128RegClass,
255    &X86::VR128RegClass, &X86::VR128RegClass,
256    &X86::VR128RegClass, &X86::VR128RegClass,
257    &X86::VR128RegClass, &X86::VR128RegClass, 0
258  };
259
260  if (Is64Bit) {
261    if (IsWin64)
262      return CalleeSavedRegClassesWin64;
263    else
264      return (callsEHReturn ?
265              CalleeSavedRegClasses64EHRet : CalleeSavedRegClasses64Bit);
266  } else {
267    return (callsEHReturn ?
268            CalleeSavedRegClasses32EHRet : CalleeSavedRegClasses32Bit);
269  }
270}
271
272BitVector X86RegisterInfo::getReservedRegs(const MachineFunction &MF) const {
273  BitVector Reserved(getNumRegs());
274  // Set the stack-pointer register and its aliases as reserved.
275  Reserved.set(X86::RSP);
276  Reserved.set(X86::ESP);
277  Reserved.set(X86::SP);
278  Reserved.set(X86::SPL);
279  // Set the frame-pointer register and its aliases as reserved if needed.
280  if (hasFP(MF)) {
281    Reserved.set(X86::RBP);
282    Reserved.set(X86::EBP);
283    Reserved.set(X86::BP);
284    Reserved.set(X86::BPL);
285  }
286  // Mark the x87 stack registers as reserved, since they don't
287  // behave normally with respect to liveness. We don't fully
288  // model the effects of x87 stack pushes and pops after
289  // stackification.
290  Reserved.set(X86::ST0);
291  Reserved.set(X86::ST1);
292  Reserved.set(X86::ST2);
293  Reserved.set(X86::ST3);
294  Reserved.set(X86::ST4);
295  Reserved.set(X86::ST5);
296  Reserved.set(X86::ST6);
297  Reserved.set(X86::ST7);
298  return Reserved;
299}
300
301//===----------------------------------------------------------------------===//
302// Stack Frame Processing methods
303//===----------------------------------------------------------------------===//
304
305static unsigned calculateMaxStackAlignment(const MachineFrameInfo *FFI) {
306  unsigned MaxAlign = 0;
307  for (int i = FFI->getObjectIndexBegin(),
308         e = FFI->getObjectIndexEnd(); i != e; ++i) {
309    if (FFI->isDeadObjectIndex(i))
310      continue;
311    unsigned Align = FFI->getObjectAlignment(i);
312    MaxAlign = std::max(MaxAlign, Align);
313  }
314
315  return MaxAlign;
316}
317
318// hasFP - Return true if the specified function should have a dedicated frame
319// pointer register.  This is true if the function has variable sized allocas or
320// if frame pointer elimination is disabled.
321//
322bool X86RegisterInfo::hasFP(const MachineFunction &MF) const {
323  const MachineFrameInfo *MFI = MF.getFrameInfo();
324  const MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
325
326  return (NoFramePointerElim ||
327          needsStackRealignment(MF) ||
328          MFI->hasVarSizedObjects() ||
329          MFI->isFrameAddressTaken() ||
330          MF.getInfo<X86MachineFunctionInfo>()->getForceFramePointer() ||
331          (MMI && MMI->callsUnwindInit()));
332}
333
334bool X86RegisterInfo::needsStackRealignment(const MachineFunction &MF) const {
335  const MachineFrameInfo *MFI = MF.getFrameInfo();
336
337  // FIXME: Currently we don't support stack realignment for functions with
338  // variable-sized allocas
339  return (RealignStack &&
340          (MFI->getMaxAlignment() > StackAlign &&
341           !MFI->hasVarSizedObjects()));
342}
343
344bool X86RegisterInfo::hasReservedCallFrame(MachineFunction &MF) const {
345  return !MF.getFrameInfo()->hasVarSizedObjects();
346}
347
348int
349X86RegisterInfo::getFrameIndexOffset(MachineFunction &MF, int FI) const {
350  int Offset = MF.getFrameInfo()->getObjectOffset(FI) + SlotSize;
351  uint64_t StackSize = MF.getFrameInfo()->getStackSize();
352
353  if (needsStackRealignment(MF)) {
354    if (FI < 0)
355      // Skip the saved EBP
356      Offset += SlotSize;
357    else {
358      unsigned Align = MF.getFrameInfo()->getObjectAlignment(FI);
359      assert( (-(Offset + StackSize)) % Align == 0);
360      Align = 0;
361      return Offset + StackSize;
362    }
363
364    // FIXME: Support tail calls
365  } else {
366    if (!hasFP(MF))
367      return Offset + StackSize;
368
369    // Skip the saved EBP
370    Offset += SlotSize;
371
372    // Skip the RETADDR move area
373    X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
374    int TailCallReturnAddrDelta = X86FI->getTCReturnAddrDelta();
375    if (TailCallReturnAddrDelta < 0) Offset -= TailCallReturnAddrDelta;
376  }
377
378  return Offset;
379}
380
381void X86RegisterInfo::
382eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
383                              MachineBasicBlock::iterator I) const {
384  if (!hasReservedCallFrame(MF)) {
385    // If the stack pointer can be changed after prologue, turn the
386    // adjcallstackup instruction into a 'sub ESP, <amt>' and the
387    // adjcallstackdown instruction into 'add ESP, <amt>'
388    // TODO: consider using push / pop instead of sub + store / add
389    MachineInstr *Old = I;
390    uint64_t Amount = Old->getOperand(0).getImm();
391    if (Amount != 0) {
392      // We need to keep the stack aligned properly.  To do this, we round the
393      // amount of space needed for the outgoing arguments up to the next
394      // alignment boundary.
395      Amount = (Amount+StackAlign-1)/StackAlign*StackAlign;
396
397      MachineInstr *New = 0;
398      if (Old->getOpcode() == getCallFrameSetupOpcode()) {
399        New = BuildMI(MF, Old->getDebugLoc(),
400                      TII.get(Is64Bit ? X86::SUB64ri32 : X86::SUB32ri),
401                      StackPtr).addReg(StackPtr).addImm(Amount);
402      } else {
403        assert(Old->getOpcode() == getCallFrameDestroyOpcode());
404        // factor out the amount the callee already popped.
405        uint64_t CalleeAmt = Old->getOperand(1).getImm();
406        Amount -= CalleeAmt;
407        if (Amount) {
408          unsigned Opc = (Amount < 128) ?
409            (Is64Bit ? X86::ADD64ri8 : X86::ADD32ri8) :
410            (Is64Bit ? X86::ADD64ri32 : X86::ADD32ri);
411          New = BuildMI(MF, Old->getDebugLoc(), TII.get(Opc), StackPtr)
412            .addReg(StackPtr).addImm(Amount);
413        }
414      }
415
416      if (New) {
417        // The EFLAGS implicit def is dead.
418        New->getOperand(3).setIsDead();
419
420        // Replace the pseudo instruction with a new instruction...
421        MBB.insert(I, New);
422      }
423    }
424  } else if (I->getOpcode() == getCallFrameDestroyOpcode()) {
425    // If we are performing frame pointer elimination and if the callee pops
426    // something off the stack pointer, add it back.  We do this until we have
427    // more advanced stack pointer tracking ability.
428    if (uint64_t CalleeAmt = I->getOperand(1).getImm()) {
429      unsigned Opc = (CalleeAmt < 128) ?
430        (Is64Bit ? X86::SUB64ri8 : X86::SUB32ri8) :
431        (Is64Bit ? X86::SUB64ri32 : X86::SUB32ri);
432      MachineInstr *Old = I;
433      MachineInstr *New =
434        BuildMI(MF, Old->getDebugLoc(), TII.get(Opc),
435                StackPtr).addReg(StackPtr).addImm(CalleeAmt);
436      // The EFLAGS implicit def is dead.
437      New->getOperand(3).setIsDead();
438
439      MBB.insert(I, New);
440    }
441  }
442
443  MBB.erase(I);
444}
445
446void X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
447                                          int SPAdj, RegScavenger *RS) const{
448  assert(SPAdj == 0 && "Unexpected");
449
450  unsigned i = 0;
451  MachineInstr &MI = *II;
452  MachineFunction &MF = *MI.getParent()->getParent();
453  while (!MI.getOperand(i).isFI()) {
454    ++i;
455    assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
456  }
457
458  int FrameIndex = MI.getOperand(i).getIndex();
459
460  unsigned BasePtr;
461  if (needsStackRealignment(MF))
462    BasePtr = (FrameIndex < 0 ? FramePtr : StackPtr);
463  else
464    BasePtr = (hasFP(MF) ? FramePtr : StackPtr);
465
466  // This must be part of a four operand memory reference.  Replace the
467  // FrameIndex with base register with EBP.  Add an offset to the offset.
468  MI.getOperand(i).ChangeToRegister(BasePtr, false);
469
470  // Now add the frame object offset to the offset from EBP.
471  if (MI.getOperand(i+3).isImm()) {
472    // Offset is a 32-bit integer.
473    int Offset = getFrameIndexOffset(MF, FrameIndex) +
474      (int)(MI.getOperand(i+3).getImm());
475
476     MI.getOperand(i+3).ChangeToImmediate(Offset);
477  } else {
478    // Offset is symbolic. This is extremely rare.
479    uint64_t Offset = getFrameIndexOffset(MF, FrameIndex) +
480                      (uint64_t)MI.getOperand(i+3).getOffset();
481    MI.getOperand(i+3).setOffset(Offset);
482  }
483}
484
485void
486X86RegisterInfo::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
487                                                      RegScavenger *RS) const {
488  MachineFrameInfo *FFI = MF.getFrameInfo();
489
490  // Calculate and set max stack object alignment early, so we can decide
491  // whether we will need stack realignment (and thus FP).
492  unsigned MaxAlign = std::max(FFI->getMaxAlignment(),
493                               calculateMaxStackAlignment(FFI));
494
495  FFI->setMaxAlignment(MaxAlign);
496}
497
498void
499X86RegisterInfo::processFunctionBeforeFrameFinalized(MachineFunction &MF) const{
500  X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
501  int32_t TailCallReturnAddrDelta = X86FI->getTCReturnAddrDelta();
502  if (TailCallReturnAddrDelta < 0) {
503    // create RETURNADDR area
504    //   arg
505    //   arg
506    //   RETADDR
507    //   { ...
508    //     RETADDR area
509    //     ...
510    //   }
511    //   [EBP]
512    MF.getFrameInfo()->
513      CreateFixedObject(-TailCallReturnAddrDelta,
514                        (-1*SlotSize)+TailCallReturnAddrDelta);
515  }
516  if (hasFP(MF)) {
517    assert((TailCallReturnAddrDelta <= 0) &&
518           "The Delta should always be zero or negative");
519    // Create a frame entry for the EBP register that must be saved.
520    int FrameIdx = MF.getFrameInfo()->CreateFixedObject(SlotSize,
521                                                        (int)SlotSize * -2+
522                                                       TailCallReturnAddrDelta);
523    assert(FrameIdx == MF.getFrameInfo()->getObjectIndexBegin() &&
524           "Slot for EBP register must be last in order to be found!");
525    FrameIdx = 0;
526  }
527}
528
529/// emitSPUpdate - Emit a series of instructions to increment / decrement the
530/// stack pointer by a constant value.
531static
532void emitSPUpdate(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
533                  unsigned StackPtr, int64_t NumBytes, bool Is64Bit,
534                  const TargetInstrInfo &TII) {
535  bool isSub = NumBytes < 0;
536  uint64_t Offset = isSub ? -NumBytes : NumBytes;
537  unsigned Opc = isSub
538    ? ((Offset < 128) ?
539       (Is64Bit ? X86::SUB64ri8 : X86::SUB32ri8) :
540       (Is64Bit ? X86::SUB64ri32 : X86::SUB32ri))
541    : ((Offset < 128) ?
542       (Is64Bit ? X86::ADD64ri8 : X86::ADD32ri8) :
543       (Is64Bit ? X86::ADD64ri32 : X86::ADD32ri));
544  uint64_t Chunk = (1LL << 31) - 1;
545  DebugLoc DL = (MBBI != MBB.end() ? MBBI->getDebugLoc() :
546                 DebugLoc::getUnknownLoc());
547
548  while (Offset) {
549    uint64_t ThisVal = (Offset > Chunk) ? Chunk : Offset;
550    MachineInstr *MI =
551      BuildMI(MBB, MBBI, DL, TII.get(Opc), StackPtr)
552         .addReg(StackPtr).addImm(ThisVal);
553    // The EFLAGS implicit def is dead.
554    MI->getOperand(3).setIsDead();
555    Offset -= ThisVal;
556  }
557}
558
559// mergeSPUpdatesUp - Merge two stack-manipulating instructions upper iterator.
560static
561void mergeSPUpdatesUp(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
562                      unsigned StackPtr, uint64_t *NumBytes = NULL) {
563  if (MBBI == MBB.begin()) return;
564
565  MachineBasicBlock::iterator PI = prior(MBBI);
566  unsigned Opc = PI->getOpcode();
567  if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
568       Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
569      PI->getOperand(0).getReg() == StackPtr) {
570    if (NumBytes)
571      *NumBytes += PI->getOperand(2).getImm();
572    MBB.erase(PI);
573  } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
574              Opc == X86::SUB32ri || Opc == X86::SUB32ri8) &&
575             PI->getOperand(0).getReg() == StackPtr) {
576    if (NumBytes)
577      *NumBytes -= PI->getOperand(2).getImm();
578    MBB.erase(PI);
579  }
580}
581
582// mergeSPUpdatesUp - Merge two stack-manipulating instructions lower iterator.
583static
584void mergeSPUpdatesDown(MachineBasicBlock &MBB,
585                        MachineBasicBlock::iterator &MBBI,
586                        unsigned StackPtr, uint64_t *NumBytes = NULL) {
587  return;
588
589  if (MBBI == MBB.end()) return;
590
591  MachineBasicBlock::iterator NI = next(MBBI);
592  if (NI == MBB.end()) return;
593
594  unsigned Opc = NI->getOpcode();
595  if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
596       Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
597      NI->getOperand(0).getReg() == StackPtr) {
598    if (NumBytes)
599      *NumBytes -= NI->getOperand(2).getImm();
600    MBB.erase(NI);
601    MBBI = NI;
602  } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
603              Opc == X86::SUB32ri || Opc == X86::SUB32ri8) &&
604             NI->getOperand(0).getReg() == StackPtr) {
605    if (NumBytes)
606      *NumBytes += NI->getOperand(2).getImm();
607    MBB.erase(NI);
608    MBBI = NI;
609  }
610}
611
612/// mergeSPUpdates - Checks the instruction before/after the passed
613/// instruction. If it is an ADD/SUB instruction it is deleted
614/// argument and the stack adjustment is returned as a positive value for ADD
615/// and a negative for SUB.
616static int mergeSPUpdates(MachineBasicBlock &MBB,
617                           MachineBasicBlock::iterator &MBBI,
618                           unsigned StackPtr,
619                           bool doMergeWithPrevious) {
620
621  if ((doMergeWithPrevious && MBBI == MBB.begin()) ||
622      (!doMergeWithPrevious && MBBI == MBB.end()))
623    return 0;
624
625  int Offset = 0;
626
627  MachineBasicBlock::iterator PI = doMergeWithPrevious ? prior(MBBI) : MBBI;
628  MachineBasicBlock::iterator NI = doMergeWithPrevious ? 0 : next(MBBI);
629  unsigned Opc = PI->getOpcode();
630  if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
631       Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
632      PI->getOperand(0).getReg() == StackPtr){
633    Offset += PI->getOperand(2).getImm();
634    MBB.erase(PI);
635    if (!doMergeWithPrevious) MBBI = NI;
636  } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
637              Opc == X86::SUB32ri || Opc == X86::SUB32ri8) &&
638             PI->getOperand(0).getReg() == StackPtr) {
639    Offset -= PI->getOperand(2).getImm();
640    MBB.erase(PI);
641    if (!doMergeWithPrevious) MBBI = NI;
642  }
643
644  return Offset;
645}
646
647void X86RegisterInfo::emitFrameMoves(MachineFunction &MF,
648                                     unsigned FrameLabelId,
649                                     unsigned ReadyLabelId) const {
650  MachineFrameInfo *MFI = MF.getFrameInfo();
651  MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
652  if (!MMI)
653    return;
654
655  uint64_t StackSize = MFI->getStackSize();
656  std::vector<MachineMove> &Moves = MMI->getFrameMoves();
657  const TargetData *TD = MF.getTarget().getTargetData();
658
659  // Calculate amount of bytes used for return address storing
660  int stackGrowth =
661    (MF.getTarget().getFrameInfo()->getStackGrowthDirection() ==
662     TargetFrameInfo::StackGrowsUp ?
663     TD->getPointerSize() : -TD->getPointerSize());
664
665  if (StackSize) {
666    // Show update of SP.
667    if (hasFP(MF)) {
668      // Adjust SP
669      MachineLocation SPDst(MachineLocation::VirtualFP);
670      MachineLocation SPSrc(MachineLocation::VirtualFP, 2*stackGrowth);
671      Moves.push_back(MachineMove(FrameLabelId, SPDst, SPSrc));
672    } else {
673      MachineLocation SPDst(MachineLocation::VirtualFP);
674      MachineLocation SPSrc(MachineLocation::VirtualFP,
675                            -StackSize+stackGrowth);
676      Moves.push_back(MachineMove(FrameLabelId, SPDst, SPSrc));
677    }
678  } else {
679    //FIXME: Verify & implement for FP
680    MachineLocation SPDst(StackPtr);
681    MachineLocation SPSrc(StackPtr, stackGrowth);
682    Moves.push_back(MachineMove(FrameLabelId, SPDst, SPSrc));
683  }
684
685  // Add callee saved registers to move list.
686  const std::vector<CalleeSavedInfo> &CSI = MFI->getCalleeSavedInfo();
687
688  // FIXME: This is dirty hack. The code itself is pretty mess right now.
689  // It should be rewritten from scratch and generalized sometimes.
690
691  // Determine maximum offset (minumum due to stack growth)
692  int64_t MaxOffset = 0;
693  for (unsigned I = 0, E = CSI.size(); I!=E; ++I)
694    MaxOffset = std::min(MaxOffset,
695                         MFI->getObjectOffset(CSI[I].getFrameIdx()));
696
697  // Calculate offsets
698  int64_t saveAreaOffset = (hasFP(MF) ? 3 : 2)*stackGrowth;
699  for (unsigned I = 0, E = CSI.size(); I!=E; ++I) {
700    int64_t Offset = MFI->getObjectOffset(CSI[I].getFrameIdx());
701    unsigned Reg = CSI[I].getReg();
702    Offset = (MaxOffset-Offset+saveAreaOffset);
703    MachineLocation CSDst(MachineLocation::VirtualFP, Offset);
704    MachineLocation CSSrc(Reg);
705    Moves.push_back(MachineMove(FrameLabelId, CSDst, CSSrc));
706  }
707
708  if (hasFP(MF)) {
709    // Save FP
710    MachineLocation FPDst(MachineLocation::VirtualFP, 2*stackGrowth);
711    MachineLocation FPSrc(FramePtr);
712    Moves.push_back(MachineMove(ReadyLabelId, FPDst, FPSrc));
713  }
714
715  MachineLocation FPDst(hasFP(MF) ? FramePtr : StackPtr);
716  MachineLocation FPSrc(MachineLocation::VirtualFP);
717  Moves.push_back(MachineMove(ReadyLabelId, FPDst, FPSrc));
718}
719
720
721void X86RegisterInfo::emitPrologue(MachineFunction &MF) const {
722  MachineBasicBlock &MBB = MF.front();   // Prolog goes in entry BB
723  MachineFrameInfo *MFI = MF.getFrameInfo();
724  const Function* Fn = MF.getFunction();
725  const X86Subtarget* Subtarget = &MF.getTarget().getSubtarget<X86Subtarget>();
726  MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
727  X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
728  MachineBasicBlock::iterator MBBI = MBB.begin();
729  bool needsFrameMoves = (MMI && MMI->hasDebugInfo()) ||
730                          !Fn->doesNotThrow() ||
731                          UnwindTablesMandatory;
732  DebugLoc DL = (MBBI != MBB.end() ? MBBI->getDebugLoc() :
733                 DebugLoc::getUnknownLoc());
734
735  // Prepare for frame info.
736  unsigned FrameLabelId = 0;
737
738  // Get the number of bytes to allocate from the FrameInfo.
739  uint64_t StackSize = MFI->getStackSize();
740
741  // Get desired stack alignment
742  uint64_t MaxAlign  = MFI->getMaxAlignment();
743
744  // Add RETADDR move area to callee saved frame size.
745  int TailCallReturnAddrDelta = X86FI->getTCReturnAddrDelta();
746  if (TailCallReturnAddrDelta < 0)
747    X86FI->setCalleeSavedFrameSize(
748          X86FI->getCalleeSavedFrameSize() +(-TailCallReturnAddrDelta));
749
750  // If this is x86-64 and the Red Zone is not disabled, if we are a leaf
751  // function, and use up to 128 bytes of stack space, don't have a frame
752  // pointer, calls, or dynamic alloca then we do not need to adjust the
753  // stack pointer (we fit in the Red Zone).
754  bool DisableRedZone = Fn->hasFnAttr(Attribute::NoRedZone);
755  if (Is64Bit && !DisableRedZone &&
756      !needsStackRealignment(MF) &&
757      !MFI->hasVarSizedObjects() &&                // No dynamic alloca.
758      !MFI->hasCalls() &&                          // No calls.
759      !Subtarget->isTargetWin64()) {               // Win64 has no Red Zone
760    uint64_t MinSize = X86FI->getCalleeSavedFrameSize();
761    if (hasFP(MF)) MinSize += SlotSize;
762    StackSize = std::max(MinSize,
763                         StackSize > 128 ? StackSize - 128 : 0);
764    MFI->setStackSize(StackSize);
765  }
766
767  // Insert stack pointer adjustment for later moving of return addr.  Only
768  // applies to tail call optimized functions where the callee argument stack
769  // size is bigger than the callers.
770  if (TailCallReturnAddrDelta < 0) {
771    MachineInstr *MI =
772      BuildMI(MBB, MBBI, DL, TII.get(Is64Bit? X86::SUB64ri32 : X86::SUB32ri),
773              StackPtr).addReg(StackPtr).addImm(-TailCallReturnAddrDelta);
774    // The EFLAGS implicit def is dead.
775    MI->getOperand(3).setIsDead();
776  }
777
778  uint64_t NumBytes = 0;
779  if (hasFP(MF)) {
780    // Calculate required stack adjustment
781    uint64_t FrameSize = StackSize - SlotSize;
782    if (needsStackRealignment(MF))
783      FrameSize = (FrameSize + MaxAlign - 1)/MaxAlign*MaxAlign;
784
785    NumBytes = FrameSize - X86FI->getCalleeSavedFrameSize();
786
787    // Get the offset of the stack slot for the EBP register... which is
788    // guaranteed to be the last slot by processFunctionBeforeFrameFinalized.
789    // Update the frame offset adjustment.
790    MFI->setOffsetAdjustment(-NumBytes);
791
792    // Save EBP into the appropriate stack slot...
793    BuildMI(MBB, MBBI, DL, TII.get(Is64Bit ? X86::PUSH64r : X86::PUSH32r))
794      .addReg(FramePtr, RegState::Kill);
795
796    if (needsFrameMoves) {
797      // Mark effective beginning of when frame pointer becomes valid.
798      FrameLabelId = MMI->NextLabelID();
799      BuildMI(MBB, MBBI, DL, TII.get(X86::DBG_LABEL)).addImm(FrameLabelId);
800    }
801
802    // Update EBP with the new base value...
803    BuildMI(MBB, MBBI, DL,
804            TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr), FramePtr)
805        .addReg(StackPtr);
806
807    // Mark the FramePtr as live-in in every block except the entry.
808    for (MachineFunction::iterator I = next(MF.begin()), E = MF.end();
809         I != E; ++I)
810      I->addLiveIn(FramePtr);
811
812    // Realign stack
813    if (needsStackRealignment(MF)) {
814      MachineInstr *MI =
815        BuildMI(MBB, MBBI, DL,
816                TII.get(Is64Bit ? X86::AND64ri32 : X86::AND32ri),
817                StackPtr).addReg(StackPtr).addImm(-MaxAlign);
818      // The EFLAGS implicit def is dead.
819      MI->getOperand(3).setIsDead();
820    }
821  } else {
822    NumBytes = StackSize - X86FI->getCalleeSavedFrameSize();
823  }
824
825  unsigned ReadyLabelId = 0;
826  if (needsFrameMoves) {
827    // Mark effective beginning of when frame pointer is ready.
828    ReadyLabelId = MMI->NextLabelID();
829    BuildMI(MBB, MBBI, DL, TII.get(X86::DBG_LABEL)).addImm(ReadyLabelId);
830  }
831
832  // Skip the callee-saved push instructions.
833  while (MBBI != MBB.end() &&
834         (MBBI->getOpcode() == X86::PUSH32r ||
835          MBBI->getOpcode() == X86::PUSH64r))
836    ++MBBI;
837
838  if (MBBI != MBB.end())
839    DL = MBBI->getDebugLoc();
840
841  if (NumBytes) {   // adjust stack pointer: ESP -= numbytes
842    if (NumBytes >= 4096 && Subtarget->isTargetCygMing()) {
843      // Check, whether EAX is livein for this function
844      bool isEAXAlive = false;
845      for (MachineRegisterInfo::livein_iterator
846           II = MF.getRegInfo().livein_begin(),
847           EE = MF.getRegInfo().livein_end(); (II != EE) && !isEAXAlive; ++II) {
848        unsigned Reg = II->first;
849        isEAXAlive = (Reg == X86::EAX || Reg == X86::AX ||
850                      Reg == X86::AH || Reg == X86::AL);
851      }
852
853      // Function prologue calls _alloca to probe the stack when allocating
854      // more than 4k bytes in one go. Touching the stack at 4K increments is
855      // necessary to ensure that the guard pages used by the OS virtual memory
856      // manager are allocated in correct sequence.
857      if (!isEAXAlive) {
858        BuildMI(MBB, MBBI, DL, TII.get(X86::MOV32ri), X86::EAX)
859          .addImm(NumBytes);
860        BuildMI(MBB, MBBI, DL, TII.get(X86::CALLpcrel32))
861          .addExternalSymbol("_alloca");
862      } else {
863        // Save EAX
864        BuildMI(MBB, MBBI, DL, TII.get(X86::PUSH32r))
865          .addReg(X86::EAX, RegState::Kill);
866        // Allocate NumBytes-4 bytes on stack. We'll also use 4 already
867        // allocated bytes for EAX.
868        BuildMI(MBB, MBBI, DL,
869                TII.get(X86::MOV32ri), X86::EAX).addImm(NumBytes-4);
870        BuildMI(MBB, MBBI, DL, TII.get(X86::CALLpcrel32))
871          .addExternalSymbol("_alloca");
872        // Restore EAX
873        MachineInstr *MI = addRegOffset(BuildMI(MF, DL, TII.get(X86::MOV32rm),
874                                                X86::EAX),
875                                        StackPtr, false, NumBytes-4);
876        MBB.insert(MBBI, MI);
877      }
878    } else {
879      // If there is an SUB32ri of ESP immediately before this instruction,
880      // merge the two. This can be the case when tail call elimination is
881      // enabled and the callee has more arguments then the caller.
882      NumBytes -= mergeSPUpdates(MBB, MBBI, StackPtr, true);
883      // If there is an ADD32ri or SUB32ri of ESP immediately after this
884      // instruction, merge the two instructions.
885      mergeSPUpdatesDown(MBB, MBBI, StackPtr, &NumBytes);
886
887      if (NumBytes)
888        emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit, TII);
889    }
890  }
891
892  if (needsFrameMoves)
893    emitFrameMoves(MF, FrameLabelId, ReadyLabelId);
894}
895
896void X86RegisterInfo::emitEpilogue(MachineFunction &MF,
897                                   MachineBasicBlock &MBB) const {
898  const MachineFrameInfo *MFI = MF.getFrameInfo();
899  X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
900  MachineBasicBlock::iterator MBBI = prior(MBB.end());
901  unsigned RetOpcode = MBBI->getOpcode();
902  DebugLoc DL = MBBI->getDebugLoc();
903
904  switch (RetOpcode) {
905  case X86::RET:
906  case X86::RETI:
907  case X86::TCRETURNdi:
908  case X86::TCRETURNri:
909  case X86::TCRETURNri64:
910  case X86::TCRETURNdi64:
911  case X86::EH_RETURN:
912  case X86::EH_RETURN64:
913  case X86::TAILJMPd:
914  case X86::TAILJMPr:
915  case X86::TAILJMPm: break;  // These are ok
916  default:
917    assert(0 && "Can only insert epilog into returning blocks");
918  }
919
920  // Get the number of bytes to allocate from the FrameInfo
921  uint64_t StackSize = MFI->getStackSize();
922  uint64_t MaxAlign  = MFI->getMaxAlignment();
923  unsigned CSSize = X86FI->getCalleeSavedFrameSize();
924  uint64_t NumBytes = 0;
925
926  if (hasFP(MF)) {
927    // Calculate required stack adjustment
928    uint64_t FrameSize = StackSize - SlotSize;
929    if (needsStackRealignment(MF))
930      FrameSize = (FrameSize + MaxAlign - 1)/MaxAlign*MaxAlign;
931
932    NumBytes = FrameSize - CSSize;
933
934    // pop EBP.
935    BuildMI(MBB, MBBI, DL,
936            TII.get(Is64Bit ? X86::POP64r : X86::POP32r), FramePtr);
937  } else {
938    NumBytes = StackSize - CSSize;
939  }
940
941  // Skip the callee-saved pop instructions.
942  MachineBasicBlock::iterator LastCSPop = MBBI;
943  while (MBBI != MBB.begin()) {
944    MachineBasicBlock::iterator PI = prior(MBBI);
945    unsigned Opc = PI->getOpcode();
946    if (Opc != X86::POP32r && Opc != X86::POP64r &&
947        !PI->getDesc().isTerminator())
948      break;
949    --MBBI;
950  }
951
952  DL = MBBI->getDebugLoc();
953
954  // If there is an ADD32ri or SUB32ri of ESP immediately before this
955  // instruction, merge the two instructions.
956  if (NumBytes || MFI->hasVarSizedObjects())
957    mergeSPUpdatesUp(MBB, MBBI, StackPtr, &NumBytes);
958
959  // If dynamic alloca is used, then reset esp to point to the last callee-saved
960  // slot before popping them off! Same applies for the case, when stack was
961  // realigned
962  if (needsStackRealignment(MF)) {
963    // We cannot use LEA here, because stack pointer was realigned. We need to
964    // deallocate local frame back
965    if (CSSize) {
966      emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, TII);
967      MBBI = prior(LastCSPop);
968    }
969
970    BuildMI(MBB, MBBI, DL,
971            TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr),
972            StackPtr).addReg(FramePtr);
973  } else if (MFI->hasVarSizedObjects()) {
974    if (CSSize) {
975      unsigned Opc = Is64Bit ? X86::LEA64r : X86::LEA32r;
976      MachineInstr *MI = addLeaRegOffset(BuildMI(MF, DL, TII.get(Opc), StackPtr),
977                                         FramePtr, false, -CSSize);
978      MBB.insert(MBBI, MI);
979    } else
980      BuildMI(MBB, MBBI, DL, TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr),
981              StackPtr).addReg(FramePtr);
982
983  } else {
984    // adjust stack pointer back: ESP += numbytes
985    if (NumBytes)
986      emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, TII);
987  }
988
989  // We're returning from function via eh_return.
990  if (RetOpcode == X86::EH_RETURN || RetOpcode == X86::EH_RETURN64) {
991    MBBI = prior(MBB.end());
992    MachineOperand &DestAddr  = MBBI->getOperand(0);
993    assert(DestAddr.isReg() && "Offset should be in register!");
994    BuildMI(MBB, MBBI, DL,
995            TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr),
996            StackPtr).addReg(DestAddr.getReg());
997  // Tail call return: adjust the stack pointer and jump to callee
998  } else if (RetOpcode == X86::TCRETURNri || RetOpcode == X86::TCRETURNdi ||
999             RetOpcode== X86::TCRETURNri64 || RetOpcode == X86::TCRETURNdi64) {
1000    MBBI = prior(MBB.end());
1001    MachineOperand &JumpTarget = MBBI->getOperand(0);
1002    MachineOperand &StackAdjust = MBBI->getOperand(1);
1003    assert(StackAdjust.isImm() && "Expecting immediate value.");
1004
1005    // Adjust stack pointer.
1006    int StackAdj = StackAdjust.getImm();
1007    int MaxTCDelta = X86FI->getTCReturnAddrDelta();
1008    int Offset = 0;
1009    assert(MaxTCDelta <= 0 && "MaxTCDelta should never be positive");
1010    // Incoporate the retaddr area.
1011    Offset = StackAdj-MaxTCDelta;
1012    assert(Offset >= 0 && "Offset should never be negative");
1013
1014    if (Offset) {
1015      // Check for possible merge with preceeding ADD instruction.
1016      Offset += mergeSPUpdates(MBB, MBBI, StackPtr, true);
1017      emitSPUpdate(MBB, MBBI, StackPtr, Offset, Is64Bit, TII);
1018    }
1019
1020    // Jump to label or value in register.
1021    if (RetOpcode == X86::TCRETURNdi|| RetOpcode == X86::TCRETURNdi64)
1022      BuildMI(MBB, MBBI, DL, TII.get(X86::TAILJMPd)).
1023        addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset());
1024    else if (RetOpcode== X86::TCRETURNri64)
1025      BuildMI(MBB, MBBI, DL, TII.get(X86::TAILJMPr64), JumpTarget.getReg());
1026    else
1027       BuildMI(MBB, MBBI, DL, TII.get(X86::TAILJMPr), JumpTarget.getReg());
1028
1029    // Delete the pseudo instruction TCRETURN.
1030    MBB.erase(MBBI);
1031  } else if ((RetOpcode == X86::RET || RetOpcode == X86::RETI) &&
1032             (X86FI->getTCReturnAddrDelta() < 0)) {
1033    // Add the return addr area delta back since we are not tail calling.
1034    int delta = -1*X86FI->getTCReturnAddrDelta();
1035    MBBI = prior(MBB.end());
1036    // Check for possible merge with preceeding ADD instruction.
1037    delta += mergeSPUpdates(MBB, MBBI, StackPtr, true);
1038    emitSPUpdate(MBB, MBBI, StackPtr, delta, Is64Bit, TII);
1039  }
1040}
1041
1042unsigned X86RegisterInfo::getRARegister() const {
1043  if (Is64Bit)
1044    return X86::RIP;  // Should have dwarf #16
1045  else
1046    return X86::EIP;  // Should have dwarf #8
1047}
1048
1049unsigned X86RegisterInfo::getFrameRegister(MachineFunction &MF) const {
1050  return hasFP(MF) ? FramePtr : StackPtr;
1051}
1052
1053void X86RegisterInfo::getInitialFrameState(std::vector<MachineMove> &Moves)
1054                                                                         const {
1055  // Calculate amount of bytes used for return address storing
1056  int stackGrowth = (Is64Bit ? -8 : -4);
1057
1058  // Initial state of the frame pointer is esp+4.
1059  MachineLocation Dst(MachineLocation::VirtualFP);
1060  MachineLocation Src(StackPtr, stackGrowth);
1061  Moves.push_back(MachineMove(0, Dst, Src));
1062
1063  // Add return address to move list
1064  MachineLocation CSDst(StackPtr, stackGrowth);
1065  MachineLocation CSSrc(getRARegister());
1066  Moves.push_back(MachineMove(0, CSDst, CSSrc));
1067}
1068
1069unsigned X86RegisterInfo::getEHExceptionRegister() const {
1070  assert(0 && "What is the exception register");
1071  return 0;
1072}
1073
1074unsigned X86RegisterInfo::getEHHandlerRegister() const {
1075  assert(0 && "What is the exception handler register");
1076  return 0;
1077}
1078
1079namespace llvm {
1080unsigned getX86SubSuperRegister(unsigned Reg, MVT VT, bool High) {
1081  switch (VT.getSimpleVT()) {
1082  default: return Reg;
1083  case MVT::i8:
1084    if (High) {
1085      switch (Reg) {
1086      default: return 0;
1087      case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
1088        return X86::AH;
1089      case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
1090        return X86::DH;
1091      case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
1092        return X86::CH;
1093      case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
1094        return X86::BH;
1095      }
1096    } else {
1097      switch (Reg) {
1098      default: return 0;
1099      case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
1100        return X86::AL;
1101      case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
1102        return X86::DL;
1103      case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
1104        return X86::CL;
1105      case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
1106        return X86::BL;
1107      case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
1108        return X86::SIL;
1109      case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
1110        return X86::DIL;
1111      case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
1112        return X86::BPL;
1113      case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
1114        return X86::SPL;
1115      case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
1116        return X86::R8B;
1117      case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
1118        return X86::R9B;
1119      case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
1120        return X86::R10B;
1121      case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
1122        return X86::R11B;
1123      case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
1124        return X86::R12B;
1125      case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
1126        return X86::R13B;
1127      case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
1128        return X86::R14B;
1129      case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
1130        return X86::R15B;
1131      }
1132    }
1133  case MVT::i16:
1134    switch (Reg) {
1135    default: return Reg;
1136    case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
1137      return X86::AX;
1138    case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
1139      return X86::DX;
1140    case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
1141      return X86::CX;
1142    case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
1143      return X86::BX;
1144    case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
1145      return X86::SI;
1146    case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
1147      return X86::DI;
1148    case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
1149      return X86::BP;
1150    case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
1151      return X86::SP;
1152    case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
1153      return X86::R8W;
1154    case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
1155      return X86::R9W;
1156    case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
1157      return X86::R10W;
1158    case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
1159      return X86::R11W;
1160    case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
1161      return X86::R12W;
1162    case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
1163      return X86::R13W;
1164    case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
1165      return X86::R14W;
1166    case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
1167      return X86::R15W;
1168    }
1169  case MVT::i32:
1170    switch (Reg) {
1171    default: return Reg;
1172    case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
1173      return X86::EAX;
1174    case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
1175      return X86::EDX;
1176    case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
1177      return X86::ECX;
1178    case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
1179      return X86::EBX;
1180    case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
1181      return X86::ESI;
1182    case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
1183      return X86::EDI;
1184    case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
1185      return X86::EBP;
1186    case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
1187      return X86::ESP;
1188    case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
1189      return X86::R8D;
1190    case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
1191      return X86::R9D;
1192    case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
1193      return X86::R10D;
1194    case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
1195      return X86::R11D;
1196    case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
1197      return X86::R12D;
1198    case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
1199      return X86::R13D;
1200    case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
1201      return X86::R14D;
1202    case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
1203      return X86::R15D;
1204    }
1205  case MVT::i64:
1206    switch (Reg) {
1207    default: return Reg;
1208    case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
1209      return X86::RAX;
1210    case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
1211      return X86::RDX;
1212    case X86::CH: case X86::CL: case X86::CX: case X86::ECX: case X86::RCX:
1213      return X86::RCX;
1214    case X86::BH: case X86::BL: case X86::BX: case X86::EBX: case X86::RBX:
1215      return X86::RBX;
1216    case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
1217      return X86::RSI;
1218    case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
1219      return X86::RDI;
1220    case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
1221      return X86::RBP;
1222    case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
1223      return X86::RSP;
1224    case X86::R8B: case X86::R8W: case X86::R8D: case X86::R8:
1225      return X86::R8;
1226    case X86::R9B: case X86::R9W: case X86::R9D: case X86::R9:
1227      return X86::R9;
1228    case X86::R10B: case X86::R10W: case X86::R10D: case X86::R10:
1229      return X86::R10;
1230    case X86::R11B: case X86::R11W: case X86::R11D: case X86::R11:
1231      return X86::R11;
1232    case X86::R12B: case X86::R12W: case X86::R12D: case X86::R12:
1233      return X86::R12;
1234    case X86::R13B: case X86::R13W: case X86::R13D: case X86::R13:
1235      return X86::R13;
1236    case X86::R14B: case X86::R14W: case X86::R14D: case X86::R14:
1237      return X86::R14;
1238    case X86::R15B: case X86::R15W: case X86::R15D: case X86::R15:
1239      return X86::R15;
1240    }
1241  }
1242
1243  return Reg;
1244}
1245}
1246
1247#include "X86GenRegisterInfo.inc"
1248
1249namespace {
1250  struct VISIBILITY_HIDDEN MSAC : public MachineFunctionPass {
1251    static char ID;
1252    MSAC() : MachineFunctionPass(&ID) {}
1253
1254    virtual bool runOnMachineFunction(MachineFunction &MF) {
1255      MachineFrameInfo *FFI = MF.getFrameInfo();
1256      MachineRegisterInfo &RI = MF.getRegInfo();
1257
1258      // Calculate max stack alignment of all already allocated stack objects.
1259      unsigned MaxAlign = calculateMaxStackAlignment(FFI);
1260
1261      // Be over-conservative: scan over all vreg defs and find, whether vector
1262      // registers are used. If yes - there is probability, that vector register
1263      // will be spilled and thus stack needs to be aligned properly.
1264      for (unsigned RegNum = TargetRegisterInfo::FirstVirtualRegister;
1265           RegNum < RI.getLastVirtReg(); ++RegNum)
1266        MaxAlign = std::max(MaxAlign, RI.getRegClass(RegNum)->getAlignment());
1267
1268      FFI->setMaxAlignment(MaxAlign);
1269
1270      return false;
1271    }
1272
1273    virtual const char *getPassName() const {
1274      return "X86 Maximal Stack Alignment Calculator";
1275    }
1276  };
1277
1278  char MSAC::ID = 0;
1279}
1280
1281FunctionPass*
1282llvm::createX86MaxStackAlignmentCalculatorPass() { return new MSAC(); }
1283