Replace slow LEA instructions in X86

[lsaba]

Summary

According to Intel's Optimization Reference Manual for SNB+:
" For LEA instructions with three source operands and some specific situations, instruction latency has increased to 3 cycles, and must
  dispatch via port 1:
- LEA that has all three source operands: base, index, and offset
- LEA that uses base and index registers where the base is EBP, RBP,or R13
- LEA that uses RIP relative addressing mode
- LEA that uses 16-bit addressing mode "

This patch currently handles the first 2 cases only.

[skatkov]

Could you please also add some negative tests when you cannot do such transformation?
For example, involving eflags.

[aqjune]

We report our observation regarding this issue.
We observed 10% speed up for the Queens benchmark in Nightly Test by simply swapping r13 and r14.

Here are more details.

LLVM & Clang: 4.0 (official release)
Queens.c: SingleSource/Benchmarks/Stanford/Queens.c in Nightly Tests.
Queens.s: clang -O3 -S Queen.c
Queens.swap.s: obtained by simply swapping r13 and r14 in Queen.s

We compiled "Queen.s" and "Queens.swap.s" using clang 4.0 (and gcc 5.4.0 too) and observed that Queens.swap.s is 10% faster than Queens.s in the following machine.

CPU: Intel(R) Core(TM) i7 CPU 870 @ 2.93GHz
OS: Ubuntu 16.04
We tested the same on several other machines and observed consistent speed up between 2 ~ 10 % depending on machines.

Thanks.

Queens.c3 KBDownload

Queens.s14 KBDownload

Queens.swap.s14 KBDownload

result.txt9 KBDownload

[skatkov]

In D32277#731870, @aqjune wrote:

We report our observation regarding this issue.
...

How does your comment relate to this patch?

[zvi]

lib/Target/X86/X86FixupLEAs.cpp

110 ↗

(On Diff #95903)

Method name should start with a lowercase character

200 ↗

(On Diff #95903)

Maybe pass a MachineBasicBlock &MBB instead?

201 ↗

(On Diff #95903)

Itr -> InsertBefore

203 ↗

(On Diff #95903)

What is the benefit of dumping every instruction added? If you do keep the print, please make it more descriptive.
I know this was there before you moved the code, but still.

204 ↗

(On Diff #95903)

return NewMI

299 ↗

(On Diff #95903)

This function and others below can take const MachineInstr& arguments.

318 ↗

(On Diff #95903)

Does it make more sense to check if isLEA() before calling this function and document that MI is a LEA instruction?

367 ↗

(On Diff #95903)

Can you avoid the need for this function by using TII::copyPhysReg?

569 ↗

(On Diff #95903)

Is it safe to erase while iterating over instructions? Maybe erase the instructions after you visited all the interesting instructions?

578 ↗

(On Diff #95903)

You don't need to create an instruction and then insert it using your Insert function. There is already a MachineInstrBuilder::BuildMI overload that does that

[efriedma]

lib/Target/X86/X86FixupLEAs.cpp

296 ↗

(On Diff #95903)

Do you need to check for R13D?

[aqjune]

@skatkov Well, the assemblies also included leal with r13 register as well. :)

102     leal    8(%r14), %eax

->

102     leal    8(%r13), %eax

The performance gap may be due to the instruction, but I'm not sure. (actually, converting r14 to r13 increased performance in this case, but I have no idea what's happening inside CPU..)

[skatkov]

Interesting... it would be nice to know why :)

Anyway I still believe that some negative tests should also be added here. For example for the sequence
cmp
lea
jb

lea cannot be converted to add due to add implicitly kills the eflags and defines it as well while lea does not do it.

[RKSimon]

lib/Target/X86/X86FixupLEAs.cpp

633 ↗

(On Diff #95903)

Is it a good idea to directly associate a general (and very vague....) feature bit with a specific set of targets like this?

[RKSimon]

D32352 is looking at more aggressive conversion of IMUL to multiple LEA instructions.

[lsaba]

In D32277#733146, @aqjune wrote:

@skatkov Well, the assemblies also included leal with r13 register as well. :)

102     leal    8(%r14), %eax

->

102     leal    8(%r13), %eax

The performance gap may be due to the instruction, but I'm not sure. (actually, converting r14 to r13 increased performance in this case, but I have no idea what's happening inside CPU..)

I double checked these two instruction and performed a test of my own, the r13 version is not slower, looks like the problem is somewhere else in the code

lib/Target/X86/X86FixupLEAs.cpp

633 ↗

(On Diff #95903)

I am not sure I understand the comment, are you referring to the usage of the SlowLEA feature specifically? or to limiting the optimization to a set of targets using a feature in general? Do you have other suggestions?

[lsaba]

In D32277#735719, @RKSimon wrote:

D32352 is looking at more aggressive conversion of IMUL to multiple LEA instructions.

Thanks for notifying, the patch does not contain any 3-Operand LEAs as far as I can see, the only case we need to be careful with is for 2-Operand LEA with RBP/R13/EBP as a base register, since this is determined only after RA, I am thinking it's better to let my patch fix those cases rather than preventing that patch from running on the problematic targets

[RKSimon]

In D32277#736509, @lsaba wrote:

In D32277#735719, @RKSimon wrote:

D32352 is looking at more aggressive conversion of IMUL to multiple LEA instructions.

Thanks for notifying, the patch does not contain any 3-Operand LEAs as far as I can see, the only case we need to be careful with is for 2-Operand LEA with RBP/R13/EBP as a base register, since this is determined only after RA, I am thinking it's better to let my patch fix those cases rather than preventing that patch from running on the problematic targets

Sorry for being pedantic about the naming, but for AMD architectures the 'slowlea' cases (whether it uses the ALU or AGU pipe) include scale != 1 (even for 2-ops), but it doesn't seem to be noticeably affected by RBP/R13/EBP. Hence my interest in PR32326 to try and make it easier to discriminate.

[lsaba]

In D32277#736540, @RKSimon wrote:

In D32277#736509, @lsaba wrote:

In D32277#735719, @RKSimon wrote:

D32352 is looking at more aggressive conversion of IMUL to multiple LEA instructions.

Thanks for notifying, the patch does not contain any 3-Operand LEAs as far as I can see, the only case we need to be careful with is for 2-Operand LEA with RBP/R13/EBP as a base register, since this is determined only after RA, I am thinking it's better to let my patch fix those cases rather than preventing that patch from running on the problematic targets

Sorry for being pedantic about the naming, but for AMD architectures the 'slowlea' cases (whether it uses the ALU or AGU pipe) include scale != 1 (even for 2-ops), but it doesn't seem to be noticeably affected by RBP/R13/EBP. Hence my interest in PR32326 to try and make it easier to discriminate.

will separating this feature form the existing slowLEA feature which is used in SLM (and giving it another name) make it less confusing?

[RKSimon]

In D32277#736594, @lsaba wrote:

In D32277#736540, @RKSimon wrote:

In D32277#736509, @lsaba wrote:

In D32277#735719, @RKSimon wrote:

D32352 is looking at more aggressive conversion of IMUL to multiple LEA instructions.

Thanks for notifying, the patch does not contain any 3-Operand LEAs as far as I can see, the only case we need to be careful with is for 2-Operand LEA with RBP/R13/EBP as a base register, since this is determined only after RA, I am thinking it's better to let my patch fix those cases rather than preventing that patch from running on the problematic targets

Sorry for being pedantic about the naming, but for AMD architectures the 'slowlea' cases (whether it uses the ALU or AGU pipe) include scale != 1 (even for 2-ops), but it doesn't seem to be noticeably affected by RBP/R13/EBP. Hence my interest in PR32326 to try and make it easier to discriminate.

will separating this feature form the existing slowLEA feature which is used in SLM (and giving it another name) make it less confusing?

Yes please, we need to discriminate between different slow LEA behaviours and separate features is probably the most straightforward way to do it.

[lsaba]

Updated patch with ZVi's comments

[lsaba]

In D32277#731728, @skatkov wrote:

Could you please also add some negative tests when you cannot do such transformation?
For example, involving eflags.

I added negative tests

lib/Target/X86/X86FixupLEAs.cpp

110 ↗

(On Diff #95903)

I removed this function

200 ↗

(On Diff #95903)

I removed this function

201 ↗

(On Diff #95903)

I removed this function

203 ↗

(On Diff #95903)

Keeping it since it's helpful, the description is at the beginning of the function

296 ↗

(On Diff #95903)

there are no LEA instructions in 64bit which take 32-bit register operands, so we shouldn't run into this case

[lsaba]

In D32277#736652, @RKSimon wrote:

In D32277#736594, @lsaba wrote:

In D32277#736540, @RKSimon wrote:

In D32277#736509, @lsaba wrote:

In D32277#735719, @RKSimon wrote:

D32352 is looking at more aggressive conversion of IMUL to multiple LEA instructions.

Thanks for notifying, the patch does not contain any 3-Operand LEAs as far as I can see, the only case we need to be careful with is for 2-Operand LEA with RBP/R13/EBP as a base register, since this is determined only after RA, I am thinking it's better to let my patch fix those cases rather than preventing that patch from running on the problematic targets

Sorry for being pedantic about the naming, but for AMD architectures the 'slowlea' cases (whether it uses the ALU or AGU pipe) include scale != 1 (even for 2-ops), but it doesn't seem to be noticeably affected by RBP/R13/EBP. Hence my interest in PR32326 to try and make it easier to discriminate.

will separating this feature form the existing slowLEA feature which is used in SLM (and giving it another name) make it less confusing?

Yes please, we need to discriminate between different slow LEA behaviours and separate features is probably the most straightforward way to do it.

separated into 2 features

[skatkov]

I have not got the conclusion: R13 is a bad register or not?
According to what I see in the comment it is not but patch still consider it is?
Could you please clarify this?

[lsaba]

In D32277#739023, @skatkov wrote:

I have not got the conclusion: R13 is a bad register or not?
According to what I see in the comment it is not but patch still consider it is?
Could you please clarify this?

according to the optimization guide: "LEA that uses base and index registers where the base is EBP, RBP,or R13" so R13 is bad when there is an index register as well, in aqjune's example the instruction does not have an index register (leal 8(%r13), %eax )

[skatkov]

Thanks

[zvi]

lib/Target/X86/X86FixupLEAs.cpp

479 ↗

(On Diff #96731)

No need to pass the iterator by reference. In fact, you could instead pass MachineInstr &MI and then you won't need to define 'MI' below.

483 ↗

(On Diff #96731)

Please capitalize

490 ↗

(On Diff #96731)

This bunch of MachineOperands can all be const

514 ↗

(On Diff #96731)

You can avoid this early declaration and assignment to nullptr by defining variables local to the scope of their use.
At this point, it is known the function will never return nullptr, so the assignment here is not to a default return value.

596 ↗

(On Diff #96731)

Now that you added the feature, maybe rename 'processInstructionForSNBPlus' to something like 'processforSlow3OpLEA'?

lib/Target/X86/X86Subtarget.h

253 ↗

(On Diff #96731)

Please specify in the comment all cases to which this feature applies.

[lsaba]

remove redundant variable

[zvi]

lib/Target/X86/X86FixupLEAs.cpp

285 ↗

(On Diff #98100)

These helpers can be improved by avoiding the operand index magic by passing the operands of interest as arguments.
The '"heavylifting" of identifying these operands happens at the beginning of 'processInstrForSlow3OpLEA'.

isRegOperand(const MachineOperand &Op)
hasInefficientLEABaseReg(const MachineOperand &Base,const MachineOperand& Index)
hasLEAOffset(const MachineOperand &Offset)
isThreeOperandsLEA(const MachineOperand &Base, const MachineOperand &Index, const MachineOperand &Offset)

481 ↗

(On Diff #98100)

Maybe rename to LEAOpcode?

516 ↗

(On Diff #98100)

with one or two (for the 3-op LEA case) add instructions?

[zvi]

lib/Target/X86/X86FixupLEAs.cpp

312 ↗

(On Diff #98134)

Sorry for not pointing this out earlier, but here is another index-magic case we can avoid by passing Offset to this function, this function can possibly be changed to

getADDFromLEA(int LEAOpcode, const MachineOperand &Offset, bool IsImm)

497 ↗

(On Diff #98134)

getOperand(5) - > Segment

[lsaba]

Updated with Zvi's comments

[zvi]

LGTM with some minor comments.
Thanks, Lama!

lib/Target/X86/X86FixupLEAs.cpp

308 ↗

(On Diff #98154)

This variable can be dropped if the case blocks will be terminated with returns. e.g. case X86::LEA16r:: return X86::ADD16rr;

327 ↗

(On Diff #98154)

Same as above

[lsaba]

lib/Target/X86/X86FixupLEAs.cpp

312 ↗

(On Diff #98134)

Decided to split this into 2 functions as well

llvm/trunk/lib/Target/X86/X86.td

 // instructions, which should be avoided in favor of a MOV + register CALL/PUSH.
 def FeatureCallRegIndirect : SubtargetFeature<"call-reg-indirect",
                                      "CallRegIndirect", "true",
                                      "Call register indirect">;
 def FeatureLEAUsesAG : SubtargetFeature<"lea-uses-ag", "LEAUsesAG", "true",
                                    "LEA instruction needs inputs at AG stage">;
 def FeatureSlowLEA : SubtargetFeature<"slow-lea", "SlowLEA", "true",
                                    "LEA instruction with certain arguments is slow">;
+def FeatureSlow3OpsLEA : SubtargetFeature<"slow-3ops-lea", "Slow3OpsLEA", "true",
+                                   "LEA instruction with 3 ops or certain registers is slow">;
 def FeatureSlowIncDec : SubtargetFeature<"slow-incdec", "SlowIncDec", "true",
                                    "INC and DEC instructions are slower than ADD and SUB">;
 def FeatureSoftFloat
     : SubtargetFeature<"soft-float", "UseSoftFloat", "true",
                        "Use software floating point features.">;
 // On some X86 processors, there is no performance hazard to writing only the
 // lower parts of a YMM or ZMM register without clearing the upper part.
 def FeatureFastPartialYMMorZMMWrite
def SNBFeatures : ProcessorFeatures<[], [
   FeatureCMPXCHG16B,
   FeaturePOPCNT,
   FeatureAES,
   FeatureSlowDivide64,
   FeaturePCLMUL,
   FeatureXSAVE,
   FeatureXSAVEOPT,
   FeatureLAHFSAHF,
+  FeatureSlow3OpsLEA,
   FeatureFastScalarFSQRT,
   FeatureFastSHLDRotate
 ]>;
 
 class SandyBridgeProc<string Name> : ProcModel<Name, SandyBridgeModel,
                                                SNBFeatures.Value, [
   FeatureSlowBTMem,
   FeatureSlowUAMem32

llvm/trunk/lib/Target/X86/X86FixupLEAs.cpp

 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetInstrInfo.h"
 using namespace llvm;
 
-#define DEBUG_TYPE "x86-fixup-LEAs"
+namespace llvm {
+void initializeFixupLEAPassPass(PassRegistry &);
+}
+
+#define FIXUPLEA_DESC "X86 LEA Fixup"
+#define FIXUPLEA_NAME "x86-fixup-LEAs"
+
+#define DEBUG_TYPE FIXUPLEA_NAME
 
 STATISTIC(NumLEAs, "Number of LEA instructions created");
 
 namespace {
 class FixupLEAPass : public MachineFunctionPass {
   enum RegUsageState { RU_NotUsed, RU_Write, RU_Read };
-  static char ID;
+
   /// \brief Loop over all of the instructions in the basic block
   /// replacing applicable instructions with LEA instructions,
   /// where appropriate.
   bool processBasicBlock(MachineFunction &MF, MachineFunction::iterator MFI);
 
-  StringRef getPassName() const override { return "X86 LEA Fixup"; }
 
   /// \brief Given a machine register, look for the instruction
   /// which writes it in the current basic block. If found,
   /// try to replace it with an equivalent LEA instruction.
   /// If replacement succeeds, then also process the newly created
   /// instruction.
   void seekLEAFixup(MachineOperand &p, MachineBasicBlock::iterator &I,
                     MachineFunction::iterator MFI);
 
   /// \brief Given a memory access or LEA instruction
   /// whose address mode uses a base and/or index register, look for
   /// an opportunity to replace the instruction which sets the base or index
   /// register with an equivalent LEA instruction.
   void processInstruction(MachineBasicBlock::iterator &I,
                           MachineFunction::iterator MFI);
 
   /// \brief Given a LEA instruction which is unprofitable
   /// on Silvermont try to replace it with an equivalent ADD instruction
   void processInstructionForSLM(MachineBasicBlock::iterator &I,
                                 MachineFunction::iterator MFI);
 
+
+  /// \brief Given a LEA instruction which is unprofitable
+  /// on SNB+ try to replace it with other instructions.
+  /// According to Intel's Optimization Reference Manual:
+  /// " For LEA instructions with three source operands and some specific
+  ///   situations, instruction latency has increased to 3 cycles, and must
+  ///   dispatch via port 1:
+  /// - LEA that has all three source operands: base, index, and offset
+  /// - LEA that uses base and index registers where the base is EBP, RBP,
+  ///   or R13
+  /// - LEA that uses RIP relative addressing mode
+  /// - LEA that uses 16-bit addressing mode "
+  /// This function currently handles the first 2 cases only.
+  MachineInstr *processInstrForSlow3OpLEA(MachineInstr &MI,
+                                          MachineFunction::iterator MFI);
+
   /// \brief Look for LEAs that add 1 to reg or subtract 1 from reg
   /// and convert them to INC or DEC respectively.
   bool fixupIncDec(MachineBasicBlock::iterator &I,
                    MachineFunction::iterator MFI) const;
 
   /// \brief Determine if an instruction references a machine register
   /// and, if so, whether it reads or writes the register.
   RegUsageState usesRegister(MachineOperand &p, MachineBasicBlock::iterator I);
 
   /// \brief Step backwards through a basic block, looking
   /// for an instruction which writes a register within
   /// a maximum of INSTR_DISTANCE_THRESHOLD instruction latency cycles.
   MachineBasicBlock::iterator searchBackwards(MachineOperand &p,
                                               MachineBasicBlock::iterator &I,
                                               MachineFunction::iterator MFI);
 
   /// \brief if an instruction can be converted to an
   /// equivalent LEA, insert the new instruction into the basic block
   /// and return a pointer to it. Otherwise, return zero.
   MachineInstr *postRAConvertToLEA(MachineFunction::iterator &MFI,
                                    MachineBasicBlock::iterator &MBBI) const;
 
 public:
-  FixupLEAPass() : MachineFunctionPass(ID) {}
+  static char ID;
+
+  StringRef getPassName() const override { return FIXUPLEA_DESC; }
+
+  FixupLEAPass() : MachineFunctionPass(ID) {
+    initializeFixupLEAPassPass(*PassRegistry::getPassRegistry());
+  }
 
   /// \brief Loop over all of the basic blocks,
   /// replacing instructions by equivalent LEA instructions
   /// if needed and when possible.
   bool runOnMachineFunction(MachineFunction &MF) override;
 
   // This pass runs after regalloc and doesn't support VReg operands.
   MachineFunctionProperties getRequiredProperties() const override {
     return MachineFunctionProperties().set(
         MachineFunctionProperties::Property::NoVRegs);
   }
 
 private:
   MachineFunction *MF;
   const X86InstrInfo *TII; // Machine instruction info.
   bool OptIncDec;
   bool OptLEA;
 };
-char FixupLEAPass::ID = 0;
 }
 
+char FixupLEAPass::ID = 0;
+
+INITIALIZE_PASS(FixupLEAPass, FIXUPLEA_NAME, FIXUPLEA_DESC, false, false)
+
 MachineInstr *
 FixupLEAPass::postRAConvertToLEA(MachineFunction::iterator &MFI,
                                  MachineBasicBlock::iterator &MBBI) const {
   MachineInstr &MI = *MBBI;
   switch (MI.getOpcode()) {
   case X86::MOV32rr:
   case X86::MOV64rr: {
     const MachineOperand &Src = MI.getOperand(1);

 
 bool FixupLEAPass::runOnMachineFunction(MachineFunction &Func) {
   if (skipFunction(*Func.getFunction()))
     return false;
 
   MF = &Func;
   const X86Subtarget &ST = Func.getSubtarget<X86Subtarget>();
   OptIncDec = !ST.slowIncDec() || Func.getFunction()->optForMinSize();
-  OptLEA = ST.LEAusesAG() || ST.slowLEA();
+  OptLEA = ST.LEAusesAG() || ST.slowLEA() || ST.slow3OpsLEA();
 
   if (!OptLEA && !OptIncDec)
     return false;
 
   TII = ST.getInstrInfo();
 
   DEBUG(dbgs() << "Start X86FixupLEAs\n";);
   // Process all basic blocks.
while (Found && I != CurInst) {
     }
     InstrDistance += TII->getInstrLatency(
         MF->getSubtarget().getInstrItineraryData(), *CurInst);
     Found = getPreviousInstr(CurInst, MFI);
   }
   return MachineBasicBlock::iterator();
 }
 
-static inline bool isLEA(const int opcode) {
-  return opcode == X86::LEA16r || opcode == X86::LEA32r ||
-         opcode == X86::LEA64r || opcode == X86::LEA64_32r;
+static inline bool isLEA(const int Opcode) {
+  return Opcode == X86::LEA16r || Opcode == X86::LEA32r ||
+         Opcode == X86::LEA64r || Opcode == X86::LEA64_32r;
+}
+
+static inline bool isInefficientLEAReg(unsigned int Reg) {
+  return Reg == X86::EBP || Reg == X86::RBP || Reg == X86::R13;
+}
+
+static inline bool isRegOperand(const MachineOperand &Op) {
+  return Op.isReg() && Op.getReg() != X86::NoRegister;
+}
+/// hasIneffecientLEARegs - LEA that uses base and index registers
+/// where the base is EBP, RBP, or R13
+static inline bool hasInefficientLEABaseReg(const MachineOperand &Base,
+                                            const MachineOperand &Index) {
+  return Base.isReg() && isInefficientLEAReg(Base.getReg()) &&
+         isRegOperand(Index);
+}
+
+static inline bool hasLEAOffset(const MachineOperand &Offset) {
+  return (Offset.isImm() && Offset.getImm() != 0) || Offset.isGlobal();
+}
+
+// LEA instruction that has all three operands: offset, base and index
+static inline bool isThreeOperandsLEA(const MachineOperand &Base,
+                                      const MachineOperand &Index,
+                                      const MachineOperand &Offset) {
+  return isRegOperand(Base) && isRegOperand(Index) && hasLEAOffset(Offset);
+}
+
+static inline int getADDrrFromLEA(int LEAOpcode) {
+  switch (LEAOpcode) {
+  default:
+    llvm_unreachable("Unexpected LEA instruction");
+  case X86::LEA16r:
+    return X86::ADD16rr;
+  case X86::LEA32r:
+    return X86::ADD32rr;
+  case X86::LEA64_32r:
+  case X86::LEA64r:
+    return X86::ADD64rr;
+  }
+}
+
+static inline int getADDriFromLEA(int LEAOpcode, const MachineOperand &Offset) {
+  bool IsInt8 = Offset.isImm() && isInt<8>(Offset.getImm());
+  switch (LEAOpcode) {
+  default:
+    llvm_unreachable("Unexpected LEA instruction");
+  case X86::LEA16r:
+    return IsInt8 ? X86::ADD16ri8 : X86::ADD16ri;
+  case X86::LEA32r:
+  case X86::LEA64_32r:
+    return IsInt8 ? X86::ADD32ri8 : X86::ADD32ri;
+  case X86::LEA64r:
+    return IsInt8 ? X86::ADD64ri8 : X86::ADD64ri32;
+  }
 }
 
 /// isLEASimpleIncOrDec - Does this LEA have one these forms:
 /// lea  %reg, 1(%reg)
 /// lea  %reg, -1(%reg)
 static inline bool isLEASimpleIncOrDec(MachineInstr &LEA) {
   unsigned SrcReg = LEA.getOperand(1 + X86::AddrBaseReg).getReg();
   unsigned DstReg = LEA.getOperand(0).getReg();
if (NewMI) {
       processInstruction(J, MFI);
     }
   }
 }
 
 void FixupLEAPass::processInstructionForSLM(MachineBasicBlock::iterator &I,
                                             MachineFunction::iterator MFI) {
   MachineInstr &MI = *I;
-  const int opcode = MI.getOpcode();
-  if (!isLEA(opcode))
+  const int Opcode = MI.getOpcode();
+  if (!isLEA(Opcode))
     return;
   if (MI.getOperand(5).getReg() != 0 || !MI.getOperand(4).isImm() ||
       !TII->isSafeToClobberEFLAGS(*MFI, I))
     return;
   const unsigned DstR = MI.getOperand(0).getReg();
   const unsigned SrcR1 = MI.getOperand(1).getReg();
   const unsigned SrcR2 = MI.getOperand(3).getReg();
   if ((SrcR1 == 0 || SrcR1 != DstR) && (SrcR2 == 0 || SrcR2 != DstR))
     return;
   if (MI.getOperand(2).getImm() > 1)
     return;
-  int addrr_opcode, addri_opcode;
-  switch (opcode) {
-  default:
-    llvm_unreachable("Unexpected LEA instruction");
-  case X86::LEA16r:
-    addrr_opcode = X86::ADD16rr;
-    addri_opcode = X86::ADD16ri;
-    break;
-  case X86::LEA32r:
-    addrr_opcode = X86::ADD32rr;
-    addri_opcode = X86::ADD32ri;
-    break;
-  case X86::LEA64_32r:
-  case X86::LEA64r:
-    addrr_opcode = X86::ADD64rr;
-    addri_opcode = X86::ADD64ri32;
-    break;
-  }
   DEBUG(dbgs() << "FixLEA: Candidate to replace:"; I->dump(););
   DEBUG(dbgs() << "FixLEA: Replaced by: ";);
   MachineInstr *NewMI = nullptr;
-  const MachineOperand &Dst = MI.getOperand(0);
   // Make ADD instruction for two registers writing to LEA's destination
   if (SrcR1 != 0 && SrcR2 != 0) {
-    const MachineOperand &Src1 = MI.getOperand(SrcR1 == DstR ? 1 : 3);
-    const MachineOperand &Src2 = MI.getOperand(SrcR1 == DstR ? 3 : 1);
-    NewMI = BuildMI(*MF, MI.getDebugLoc(), TII->get(addrr_opcode))
-                .add(Dst)
-                .add(Src1)
-                .add(Src2);
-    MFI->insert(I, NewMI);
+    const MCInstrDesc &ADDrr = TII->get(getADDrrFromLEA(Opcode));
+    const MachineOperand &Src = MI.getOperand(SrcR1 == DstR ? 3 : 1);
+    NewMI =
+        BuildMI(*MFI, I, MI.getDebugLoc(), ADDrr, DstR).addReg(DstR).add(Src);
     DEBUG(NewMI->dump(););
   }
   // Make ADD instruction for immediate
   if (MI.getOperand(4).getImm() != 0) {
+    const MCInstrDesc &ADDri =
+        TII->get(getADDriFromLEA(Opcode, MI.getOperand(4)));
     const MachineOperand &SrcR = MI.getOperand(SrcR1 == DstR ? 1 : 3);
-    NewMI = BuildMI(*MF, MI.getDebugLoc(), TII->get(addri_opcode))
-                .add(Dst)
+    NewMI = BuildMI(*MFI, I, MI.getDebugLoc(), ADDri, DstR)
                 .add(SrcR)
                 .addImm(MI.getOperand(4).getImm());
-    MFI->insert(I, NewMI);
     DEBUG(NewMI->dump(););
   }
   if (NewMI) {
     MFI->erase(I);
-    I = static_cast<MachineBasicBlock::iterator>(NewMI);
+    I = NewMI;
+  }
+}
+
+MachineInstr *
+FixupLEAPass::processInstrForSlow3OpLEA(MachineInstr &MI,
+                                        MachineFunction::iterator MFI) {
+
+  const int LEAOpcode = MI.getOpcode();
+  if (!isLEA(LEAOpcode))
+    return nullptr;
+
+  const MachineOperand &Dst = MI.getOperand(0);
+  const MachineOperand &Base = MI.getOperand(1);
+  const MachineOperand &Scale = MI.getOperand(2);
+  const MachineOperand &Index = MI.getOperand(3);
+  const MachineOperand &Offset = MI.getOperand(4);
+  const MachineOperand &Segment = MI.getOperand(5);
+
+  if (!(isThreeOperandsLEA(Base, Index, Offset) ||
+        hasInefficientLEABaseReg(Base, Index)) ||
+      !TII->isSafeToClobberEFLAGS(*MFI, MI) ||
+      Segment.getReg() != X86::NoRegister)
+    return nullptr;
+
+  unsigned int DstR = Dst.getReg();
+  unsigned int BaseR = Base.getReg();
+  unsigned int IndexR = Index.getReg();
+  unsigned SSDstR =
+      (LEAOpcode == X86::LEA64_32r) ? getX86SubSuperRegister(DstR, 64) : DstR;
+  bool IsScale1 = Scale.getImm() == 1;
+  bool IsInefficientBase = isInefficientLEAReg(BaseR);
+  bool IsInefficientIndex = isInefficientLEAReg(IndexR);
+
+  // Skip these cases since it takes more than 2 instructions
+  // to replace the LEA instruction.
+  if (IsInefficientBase && SSDstR == BaseR && !IsScale1)
+    return nullptr;
+  if (LEAOpcode == X86::LEA64_32r && IsInefficientBase &&
+      (IsInefficientIndex || !IsScale1))
+    return nullptr;
+
+  const DebugLoc DL = MI.getDebugLoc();
+  const MCInstrDesc &ADDrr = TII->get(getADDrrFromLEA(LEAOpcode));
+  const MCInstrDesc &ADDri = TII->get(getADDriFromLEA(LEAOpcode, Offset));
+
+  DEBUG(dbgs() << "FixLEA: Candidate to replace:"; MI.dump(););
+  DEBUG(dbgs() << "FixLEA: Replaced by: ";);
+
+  // First try to replace LEA with one or two (for the 3-op LEA case)
+  // add instructions:
+  // 1.lea (%base,%index,1), %base => add %index,%base
+  // 2.lea (%base,%index,1), %index => add %base,%index
+  if (IsScale1 && (DstR == BaseR || DstR == IndexR)) {
+    const MachineOperand &Src = DstR == BaseR ? Index : Base;
+    MachineInstr *NewMI =
+        BuildMI(*MFI, MI, DL, ADDrr, DstR).addReg(DstR).add(Src);
+    DEBUG(NewMI->dump(););
+    // Create ADD instruction for the Offset in case of 3-Ops LEA.
+    if (hasLEAOffset(Offset)) {
+      NewMI = BuildMI(*MFI, MI, DL, ADDri, DstR).addReg(DstR).add(Offset);
+      DEBUG(NewMI->dump(););
+    }
+    return NewMI;
+  }
+  // If the base is inefficient try switching the index and base operands,
+  // otherwise just break the 3-Ops LEA inst into 2-Ops LEA + ADD instruction:
+  // lea offset(%base,%index,scale),%dst =>
+  // lea (%base,%index,scale); add offset,%dst
+  if (!IsInefficientBase || (!IsInefficientIndex && IsScale1)) {
+    MachineInstr *NewMI = BuildMI(*MFI, MI, DL, TII->get(LEAOpcode))
+                              .add(Dst)
+                              .add(IsInefficientBase ? Index : Base)
+                              .add(Scale)
+                              .add(IsInefficientBase ? Base : Index)
+                              .addImm(0)
+                              .add(Segment);
+    DEBUG(NewMI->dump(););
+    // Create ADD instruction for the Offset in case of 3-Ops LEA.
+    if (hasLEAOffset(Offset)) {
+      NewMI = BuildMI(*MFI, MI, DL, ADDri, DstR).addReg(DstR).add(Offset);
+      DEBUG(NewMI->dump(););
+    }
+    return NewMI;
+  }
+  // Handle the rest of the cases with inefficient base register:
+  assert(SSDstR != BaseR && "SSDstR == BaseR should be handled already!");
+  assert(IsInefficientBase && "efficient base should be handled already!");
+
+  // lea (%base,%index,1), %dst => mov %base,%dst; add %index,%dst
+  if (IsScale1 && !hasLEAOffset(Offset)) {
+    TII->copyPhysReg(*MFI, MI, DL, DstR, BaseR, Base.isKill());
+    DEBUG(MI.getPrevNode()->dump(););
+
+    MachineInstr *NewMI =
+        BuildMI(*MFI, MI, DL, ADDrr, DstR).addReg(DstR).add(Index);
+    DEBUG(NewMI->dump(););
+    return NewMI;
   }
+  // lea offset(%base,%index,scale), %dst =>
+  // lea offset( ,%index,scale), %dst; add %base,%dst
+  MachineInstr *NewMI = BuildMI(*MFI, MI, DL, TII->get(LEAOpcode))
+                            .add(Dst)
+                            .addReg(0)
+                            .add(Scale)
+                            .add(Index)
+                            .add(Offset)
+                            .add(Segment);
+  DEBUG(NewMI->dump(););
+
+  NewMI = BuildMI(*MFI, MI, DL, ADDrr, DstR).addReg(DstR).add(Base);
+  DEBUG(NewMI->dump(););
+  return NewMI;
 }
 
 bool FixupLEAPass::processBasicBlock(MachineFunction &MF,
                                      MachineFunction::iterator MFI) {
 
   for (MachineBasicBlock::iterator I = MFI->begin(); I != MFI->end(); ++I) {
     if (OptIncDec)
       if (fixupIncDec(I, MFI))
         continue;
 
     if (OptLEA) {
       if (MF.getSubtarget<X86Subtarget>().isSLM())
         processInstructionForSLM(I, MFI);
-      else
+
+      else {
+        if (MF.getSubtarget<X86Subtarget>().slow3OpsLEA()) {
+          if (auto *NewMI = processInstrForSlow3OpLEA(*I, MFI)) {
+            MFI->erase(I);
+            I = NewMI;
+          }
+        } else
           processInstruction(I, MFI);
       }
     }
+  }
   return false;
 }

llvm/trunk/lib/Target/X86/X86Subtarget.h

protected:
 
   /// True if the LEA instruction inputs have to be ready at address generation
   /// (AG) time.
   bool LEAUsesAG;
 
   /// True if the LEA instruction with certain arguments is slow
   bool SlowLEA;
 
+  /// True if the LEA instruction has all three source operands: base, index,
+  /// and offset or if the LEA instruction uses base and index registers where
+  /// the base is EBP, RBP,or R13
+  bool Slow3OpsLEA;
+
   /// True if INC and DEC instructions are slow when writing to flags
   bool SlowIncDec;
 
   /// Processor has AVX-512 PreFetch Instructions
   bool HasPFI;
 
   /// Processor has AVX-512 Exponential and Reciprocal Instructions
   bool HasERI;
public:
   bool hasFastSHLDRotate() const { return HasFastSHLDRotate; }
   bool hasERMSB() const { return HasERMSB; }
   bool hasSlowDivide32() const { return HasSlowDivide32; }
   bool hasSlowDivide64() const { return HasSlowDivide64; }
   bool padShortFunctions() const { return PadShortFunctions; }
   bool callRegIndirect() const { return CallRegIndirect; }
   bool LEAusesAG() const { return LEAUsesAG; }
   bool slowLEA() const { return SlowLEA; }
+  bool slow3OpsLEA() const { return Slow3OpsLEA; }
   bool slowIncDec() const { return SlowIncDec; }
   bool hasCDI() const { return HasCDI; }
   bool hasPFI() const { return HasPFI; }
   bool hasERI() const { return HasERI; }
   bool hasDQI() const { return HasDQI; }
   bool hasBWI() const { return HasBWI; }
   bool hasVLX() const { return HasVLX; }
   bool hasPKU() const { return HasPKU; }

llvm/trunk/lib/Target/X86/X86TargetMachine.cpp

 
 static cl::opt<bool> EnableMachineCombinerPass("x86-machine-combiner",
                                cl::desc("Enable the machine combiner pass"),
                                cl::init(true), cl::Hidden);
 
 namespace llvm {
 
 void initializeWinEHStatePassPass(PassRegistry &);
+void initializeFixupLEAPassPass(PassRegistry &);
 void initializeX86ExecutionDepsFixPass(PassRegistry &);
 
 } // end namespace llvm
 
 extern "C" void LLVMInitializeX86Target() {
   // Register the target.
   RegisterTargetMachine<X86TargetMachine> X(getTheX86_32Target());
   RegisterTargetMachine<X86TargetMachine> Y(getTheX86_64Target());
 
   PassRegistry &PR = *PassRegistry::getPassRegistry();
   initializeGlobalISel(PR);
   initializeWinEHStatePassPass(PR);
   initializeFixupBWInstPassPass(PR);
   initializeEvexToVexInstPassPass(PR);
+  initializeFixupLEAPassPass(PR);
   initializeX86ExecutionDepsFixPass(PR);
 }
 
 static std::unique_ptr<TargetLoweringObjectFile> createTLOF(const Triple &TT) {
   if (TT.isOSBinFormatMachO()) {
     if (TT.getArch() == Triple::x86_64)
       return llvm::make_unique<X86_64MachoTargetObjectFile>();
     return llvm::make_unique<TargetLoweringObjectFileMachO>();

llvm/trunk/test/CodeGen/X86/leaFixup32.mir

+# RUN: llc -run-pass x86-fixup-LEAs -mcpu=corei7-avx -o - %s | FileCheck %s
+--- |
+  ; ModuleID = 'test/CodeGen/X86/fixup-lea.ll'
+  source_filename = "test/CodeGen/X86/fixup-lea.ll"
+  target datalayout = "e-m:e-p:32:32-f64:32:64-f80:32-n8:16:32-S128"
+  target triple = "i386"
+  ;generated using: llc -stop-after x86-pad-short-functions fixup-lea.ll > leaFinxup32.mir
+
+  ;test2add_32: 3 operands LEA32r that can be replaced with 2 add instructions
+  ; where ADD32ri8 is chosen
+  define i32 @test2add_32() {
+    ret i32 0
+  }
+
+  ;test2add_ebp_32: 3 operands LEA32r that can be replaced with 2 add instructions
+  ; where the base is rbp/r13/ebp register
+  define i32 @test2add_ebp_32() {
+    ret i32 0
+  }
+
+  ;test1add_ebp_32: 2 operands LEA32r where base register is ebp and can be replaced
+  ; with an add instruction
+  define i32 @test1add_ebp_32() {
+    ret i32 0
+  }
+
+  ;testleaadd_32: 3 operands LEA32r that can be replaced with 1 lea 1 add instructions
+  define i32 @testleaadd_32() {
+    ret i32 0
+  }
+
+  ;testleaadd_ebp_32: 3 operands LEA32r that can be replaced with 1 lea 1 add instructions
+  ; where the base is ebp register
+  define i32 @testleaadd_ebp_32() {
+    ret i32 0
+  }
+
+  ;test1lea_ebp_32: 2 operands LEA32r wher base register is rbp/r13/ebp and can be replaced
+  ; with a lea instruction
+  define i32 @test1lea_ebp_32() {
+    ret i32 0
+  }
+ 
+  ;test2addi32_32: 3 operands LEA32r that can be replaced with 2 add instructions where ADD32ri32
+  ; is chosen
+  define i32 @test2addi32_32() {
+    ret i32 0
+  }
+ 
+  ;test1mov1add_ebp_32: 2 operands LEA32r that can be replaced with 1 add 1 mov instructions
+  ; where the base is rbp/r13/ebp register
+  define i32 @test1mov1add_ebp_32() {
+    ret i32 0
+  }
+
+  ;testleaadd_ebp_index_32: 3 operands LEA32r that can be replaced with 1 lea 1 add instructions
+  ; where the base and the index are ebp register and there is offset
+  define i32 @testleaadd_ebp_index_32() {
+    ret i32 0
+  }
+
+  ;testleaadd_ebp_index2_32: 3 operands LEA32r that can be replaced with 1 lea 1 add instructions
+  ; where the base and the index are ebp register and there is scale
+  define i32 @testleaadd_ebp_index2_32() {
+    ret i32 0
+  }
+  
+  ;test_skip_opt_32: 3 operands LEA32r that can not be replaced with 2 instructions
+  define i32 @test_skip_opt_32() {
+    ret i32 0
+  }
+
+  ;test_skip_eflags_32: LEA32r that cannot be replaced since its not safe to clobber eflags
+  define i32 @test_skip_eflags_32() {
+    ret i32 0
+  }
+
+...
+---
+name:            test2add_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%eax' }
+  - { reg: '%ebp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp
+    ; CHECK: %eax = ADD32rr %eax, killed %ebp
+    ; CHECK: %eax = ADD32ri8 %eax, -5
+ 
+    %eax = LEA32r killed %eax, 1, killed %ebp, -5, _
+    RETQ %eax
+
+...
+---
+name:            test2add_ebp_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%eax' }
+  - { reg: '%ebp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp
+    ; CHECK: %ebp = ADD32rr %ebp, killed %eax
+    ; CHECK: %ebp = ADD32ri8 %ebp, -5
+ 
+    %ebp = LEA32r killed %ebp, 1, killed %eax, -5, _
+    RETQ %ebp
+
+...
+---
+name:            test1add_ebp_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%eax' }
+  - { reg: '%ebp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp
+    ; CHECK: %ebp = ADD32rr %ebp, killed %eax
+ 
+    %ebp = LEA32r killed %ebp, 1, killed %eax, 0, _
+    RETQ %ebp
+
+...
+---
+name:            testleaadd_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%eax' }
+  - { reg: '%ebp' }
+  - { reg: '%ebx' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp, %esi
+    ; CHECK: %ebx = LEA32r killed %eax, 1, killed %ebp, 0
+    ; CHECK: %ebx = ADD32ri8 %ebx, -5
+ 
+    %ebx = LEA32r killed %eax, 1, killed %ebp, -5, _
+    RETQ %ebx
+
+...
+---
+name:            testleaadd_ebp_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%eax' }
+  - { reg: '%ebp' }
+  - { reg: '%ebx' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp
+    ; CHECK: %ebx = LEA32r killed %eax, 1, killed %ebp, 0, _
+    ; CHECK: %ebx = ADD32ri8  %ebx, -5
+ 
+    %ebx = LEA32r killed %ebp, 1, killed %eax, -5, _
+    RETQ %ebx
+
+...
+---
+name:            test1lea_ebp_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%eax' }
+  - { reg: '%ebp' }
+  - { reg: '%ebx' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp
+    ; CHECK: %ebx = LEA32r killed %eax, 1, killed %ebp, 0, _
+ 
+    %ebx = LEA32r killed %ebp, 1, killed %eax, 0, _
+    RETQ %ebx
+
+...
+---
+name:            test2addi32_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%eax' }
+  - { reg: '%ebp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp
+    ; CHECK: %eax = ADD32rr %eax, killed %ebp
+    ; CHECK: %eax = ADD32ri %eax, 129
+ 
+    %eax = LEA32r killed %eax, 1, killed %ebp, 129, _
+    RETQ %eax
+
+...
+---
+name:            test1mov1add_ebp_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%eax' }
+  - { reg: '%ebp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp, %ebx
+    ; CHECK: %ebx = MOV32rr killed %ebp
+    ; CHECK: %ebx = ADD32rr %ebx, killed %ebp
+ 
+    %ebx = LEA32r killed %ebp, 1, killed %ebp, 0, _
+    RETQ %ebx
+
+...
+---
+name:            testleaadd_ebp_index_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%ebx' }
+  - { reg: '%ebp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp, %ebx
+    ; CHECK: %ebx = LEA32r _, 1, killed %ebp, 5, _
+    ; CHECK: %ebx = ADD32rr %ebx, killed %ebp
+ 
+    %ebx = LEA32r killed %ebp, 1, killed %ebp, 5, _
+    RETQ %ebx
+
+...
+---
+name:            testleaadd_ebp_index2_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%ebx' }
+  - { reg: '%ebp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp, %ebx
+    ; CHECK: %ebx = LEA32r _, 4, killed %ebp, 5, _
+    ; CHECK: %ebx = ADD32rr %ebx, killed %ebp
+ 
+    %ebx = LEA32r killed %ebp, 4, killed %ebp, 5, _
+    RETQ %ebx
+
+...
+---
+name:            test_skip_opt_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%ebx' }
+  - { reg: '%ebp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp, %ebx
+    ; CHECK: %ebp = LEA32r killed %ebp, 4, killed %ebp, 0, _
+ 
+    %ebp = LEA32r killed %ebp, 4, killed %ebp, 0, _
+    RETQ %ebp
+
+...
+---
+name:            test_skip_eflags_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%ebp' }
+  - { reg: '%eax' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp, %ebx
+    ; CHECK: %ebx = LEA32r killed %eax, 4, killed %eax, 5, _
+    ; CHECK: %ebp = LEA32r killed %ebx, 4, killed %ebx, 0, _
+    ; CHECK: %ebp = ADD32ri8 %ebp, 5
+   
+    CMP32rr   %eax, killed %ebx, implicit-def %eflags
+    %ebx = LEA32r killed %eax, 4, killed %eax, 5, _
+    JE_1 %bb.1, implicit %eflags
+    RETQ %ebx
+  bb.1:
+    liveins: %eax, %ebp, %ebx
+    %ebp = LEA32r killed %ebx, 4, killed %ebx, 5, _
+    RETQ %ebp
+
+...
+
+
+

llvm/trunk/test/CodeGen/X86/leaFixup64.mir

+# RUN: llc -run-pass x86-fixup-LEAs -mcpu=corei7-avx -o - %s | FileCheck %s
+--- |
+  ; ModuleID = 'lea-2.ll'
+  source_filename = "lea-2.ll"
+  target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+  ;generated using: llc -stop-after x86-pad-short-functions lea-2.ll > leaFinxup64.mir
+
+  ;testleaadd_64_32_1: 3 operands LEA64_32r cannot be replaced with 2 add instructions
+  ; but can be replaced with 1 lea + 1 add
+  define i32 @testleaadd_64_32_1() {
+    ret i32 0
+  }
+
+  ;testleaadd_rbp_64_32_1: 3 operands LEA64_32r cannot be replaced with 2 add instructions
+  ; where the base is rbp/r13/ebp register but it can be replaced with 1 lea + 1 add
+  define i32 @testleaadd_rbp_64_32_1() {
+    ret i32 0
+  }
+
+  ;test1lea_rbp_64_32_1: 2 operands LEA64_32r where base register is rbp/r13/ebp and can not
+  ; be replaced with an add instruction but can be replaced with 1 lea instruction
+  define i32 @test1lea_rbp_64_32_1() {
+    ret i32 0
+  }
+
+  ;test2add_64: 3 operands LEA64r that can be replaced with 2 add instructions
+  define i32 @test2add_64() {
+    ret i32 0
+  }
+
+  ;test2add_rbp_64: 3 operands LEA64r that can be replaced with 2 add instructions
+  ; where the base is rbp/r13/ebp register
+  define i32 @test2add_rbp_64() {
+    ret i32 0
+  }
+
+  ;test1add_rbp_64: 2 operands LEA64r where base register is rbp/r13/ebp and can be replaced
+  ; with an add instruction
+  define i32 @test1add_rbp_64() {
+    ret i32 0
+  }
+
+  ;testleaadd_64_32: 3 operands LEA64_32r that can be replaced with 1 lea 1 add instructions
+  define i32 @testleaadd_64_32() {
+    ret i32 0
+  }
+
+  ;testleaadd_rbp_64_32: 3 operands LEA64_32r that can be replaced with 1 lea 1 add instructions
+  ; where the base is rbp/r13/ebp register
+  define i32 @testleaadd_rbp_64_32() {
+    ret i32 0
+  }
+
+  ;test1lea_rbp_64_32: 2 operands LEA64_32r where base register is rbp/r13/ebp and can be replaced
+  ; with a lea instruction
+  define i32 @test1lea_rbp_64_32() {
+    ret i32 0
+  }
+
+  ;testleaadd_64: 3 operands LEA64r that can be replaced with 1 lea 1 add instructions
+  define i32 @testleaadd_64() {
+    ret i32 0
+  }
+
+  ;testleaadd_rbp_64: 3 operands LEA64r that can be replaced with 1 lea 1 add instructions
+  ; where the base is rbp/r13/ebp register
+  define i32 @testleaadd_rbp_64() {
+    ret i32 0
+  }
+
+  ;test1lea_rbp_64: 2 operands LEA64r wher base register is rbp/r13/ebp and can be replaced
+  ; with a lea instruction
+  define i32 @test1lea_rbp_64() {
+    ret i32 0
+  }
+
+  ;test8: dst = base & scale!=1, can't optimize
+  define i32 @test8() {
+      ret i32 0
+  }
+ 
+  ;testleaaddi32_64_32: 3 operands LEA64_32r that can be replaced with 1 lea + 1 add instructions where
+  ; ADD64ri32 is chosen
+  define i32 @testleaaddi32_64_32() {
+    ret i32 0
+  }
+ 
+  ;test1mov1add_rbp_64_32: 2 operands LEA64_32r cannot be replaced with 1 add 1 mov instructions
+  ; where the base is rbp/r13/ebp register
+  define i32 @test1mov1add_rbp_64_32() {
+    ret i32 0
+  }
+
+  ;testleaadd_rbp_index_64_32: 3 operands LEA64_32r that cannot replaced with 1 lea 1 add instructions
+  ; where the base and the index are ebp register and there is offset
+  define i32 @testleaadd_rbp_index_64_32() {
+    ret i32 0
+  }
+
+  ;testleaadd_rbp_index2_64_32: 3 operands LEA64_32r that cannot replaced with 1 lea 1 add instructions
+  ; where the base and the index are ebp register and there is scale
+  define i32 @testleaadd_rbp_index2_64_32() {
+    ret i32 0
+  }
+ 
+  ;test2addi32_64: 3 operands LEA64r that can be replaced with 2 add instructions where ADD64ri32
+  ; is chosen
+  define i32 @test2addi32_64() {
+    ret i32 0
+  }
+ 
+  ;test1mov1add_rbp_64: 2 operands LEA64r that can be replaced with 1 add 1 mov instructions
+  ; where the base is rbp/r13/ebp register
+  define i32 @test1mov1add_rbp_64() {
+    ret i32 0
+  }
+
+  ;testleaadd_rbp_index_64: 3 operands LEA64r that can be replaced with 1 lea 1 add instructions
+  ; where the base and the index are ebp register and there is offset
+  define i32 @testleaadd_rbp_index_64() {
+    ret i32 0
+  }
+
+  ;testleaadd_rbp_index2_64: 3 operands LEA64r that can be replaced with 1 lea 1 add instructions
+  ; where the base and the index are ebp register and there is scale
+  define i32 @testleaadd_rbp_index2_64() {
+    ret i32 0
+  }
+
+  ;test_skip_opt_64: 3 operands LEA64r that can not be replaced with 2 instructions
+  define i32 @test_skip_opt_64() {
+    ret i32 0
+  }
+
+  ;test_skip_eflags_64: LEA64r that cannot be replaced since its not safe to clobber eflags
+  define i32 @test_skip_eflags_64() {
+    ret i32 0
+  }
+
+  ;test_skip_opt_64_32: 3 operands LEA64_32r that can not be replaced with 2 instructions
+  define i32 @test_skip_opt_64_32() {
+    ret i32 0
+  }
+
+  ;test_skip_eflags_64_32: LEA64_32r that cannot be replaced since its not safe to clobber eflags
+  define i32 @test_skip_eflags_64_32() {
+    ret i32 0
+  }
+
+
+...
+---
+name:            testleaadd_64_32_1
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %eax = LEA64_32r killed %rax, 1, killed %rbp, 0
+    ; CHECK: %eax = ADD32ri8 %eax, -5
+ 
+    %eax = LEA64_32r killed %rax, 1, killed %rbp, -5, _
+    RETQ %eax
+
+...
+---
+name:            testleaadd_rbp_64_32_1
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %ebp = LEA64_32r killed %rax, 1,  killed %rbp, 0
+    ; CHECK: %ebp = ADD32ri8 %ebp, -5
+ 
+    %ebp = LEA64_32r killed %rbp, 1, killed %rax, -5, _
+    RETQ %ebp
+
+...
+---
+name:            test1lea_rbp_64_32_1
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %ebp = LEA64_32r killed %rax, 1, killed %rbp, 0
+ 
+    %ebp = LEA64_32r killed %rbp, 1, killed %rax, 0, _
+    RETQ %ebp
+
+...
+---
+name:            test2add_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %rax = ADD64rr %rax, killed %rbp
+    ; CHECK: %rax = ADD64ri8 %rax, -5
+ 
+    %rax = LEA64r killed %rax, 1, killed %rbp, -5, _
+    RETQ %eax
+
+...
+---
+name:            test2add_rbp_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %rbp = ADD64rr %rbp, killed %rax
+    ; CHECK: %rbp = ADD64ri8 %rbp, -5
+ 
+    %rbp = LEA64r killed %rbp, 1, killed %rax, -5, _
+    RETQ %ebp
+
+...
+---
+name:            test1add_rbp_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %rbp = ADD64rr %rbp, killed %rax
+ 
+    %rbp = LEA64r killed %rbp, 1, killed %rax, 0, _
+    RETQ %ebp
+
+...
+---
+name:            testleaadd_64_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+  - { reg: '%rbx' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %ebx = LEA64_32r killed %rax, 1, killed %rbp, 0, _
+    ; CHECK: %ebx = ADD32ri8 %ebx, -5
+ 
+    %ebx = LEA64_32r killed %rax, 1, killed %rbp, -5, _
+    RETQ %ebx
+
+...
+---
+name:            testleaadd_rbp_64_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+  - { reg: '%rbx' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %ebx = LEA64_32r killed %rax, 1, killed %rbp, 0, _
+    ; CHECK: %ebx = ADD32ri8 %ebx, -5
+ 
+    %ebx = LEA64_32r killed %rbp, 1, killed %rax, -5, _
+    RETQ %ebx
+
+...
+---
+name:            test1lea_rbp_64_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+  - { reg: '%rbx' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %ebx = LEA64_32r killed %rax, 1, killed %rbp, 0, _
+ 
+    %ebx = LEA64_32r killed %rbp, 1, killed %rax, 0, _
+    RETQ %ebx
+
+...
+---
+name:            testleaadd_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+  - { reg: '%rbx' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %rbx = LEA64r killed %rax, 1, killed %rbp, 0, _
+    ; CHECK: %rbx = ADD64ri8 %rbx, -5
+ 
+    %rbx = LEA64r killed %rax, 1, killed %rbp, -5, _
+    RETQ %ebx
+
+...
+---
+name:            testleaadd_rbp_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+  - { reg: '%rbx' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %rbx = LEA64r killed %rax, 1, killed %rbp, 0, _
+    ; CHECK: %rbx = ADD64ri8 %rbx, -5
+ 
+    %rbx = LEA64r killed %rbp, 1, killed %rax, -5, _
+    RETQ %ebx
+
+...
+---
+name:            test1lea_rbp_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+  - { reg: '%rbx' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %rbx = LEA64r killed %rax, 1, killed %rbp, 0, _
+ 
+    %rbx = LEA64r killed %rbp, 1, killed %rax, 0, _
+    RETQ %ebx
+
+...
+---
+name:            test8
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rdi' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rdi, %rbp
+    ; CHECK:  %r12 = LEA64r _, 2, killed %r13, 5, _
+    ; CHECK:  %r12 = ADD64rr %r12, killed %rbp
+    %rbp = KILL %rbp, implicit-def %rbp
+    %r13 = KILL %rdi, implicit-def %r13
+    %r12 = LEA64r killed %rbp, 2, killed %r13, 5, _
+    RETQ %r12
+
+...
+---
+name:            testleaaddi32_64_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %eax = LEA64_32r killed %rax, 1, killed %rbp, 0
+    ; CHECK: %eax = ADD32ri %eax, 129
+ 
+    %eax = LEA64_32r killed %rax, 1, killed %rbp, 129, _
+    RETQ %eax
+
+...
+---
+name:            test1mov1add_rbp_64_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp, %rbx
+    ; CHECK: %ebx = LEA64_32r killed %rbp, 1, killed %rbp, 0, _
+
+    %ebx = LEA64_32r killed %rbp, 1, killed %rbp, 0, _
+    RETQ %ebx
+
+...
+---
+name:            testleaadd_rbp_index_64_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rbx' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp, %rbx
+    ; CHECK: %ebx = LEA64_32r killed %rbp, 1, killed %rbp, 5, _
+ 
+    %ebx = LEA64_32r killed %rbp, 1, killed %rbp, 5, _
+    RETQ %ebx
+
+...
+---
+name:            testleaadd_rbp_index2_64_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rbx' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %eax, %ebp, %ebx
+    ; CHECK: %ebx = LEA64_32r killed %rbp, 4, killed %rbp, 5, _
+ 
+    %ebx = LEA64_32r killed %rbp, 4, killed %rbp, 5, _
+    RETQ %ebx
+
+...
+---
+name:            test2addi32_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp
+    ; CHECK: %rax = ADD64rr %rax, killed %rbp
+    ; CHECK: %rax = ADD64ri32 %rax, 129
+ 
+    %rax = LEA64r killed %rax, 1, killed %rbp, 129, _
+    RETQ %eax
+
+...
+---
+name:            test1mov1add_rbp_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rax' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp, %rbx
+    ; CHECK: %rbx = MOV64rr killed %rbp
+    ; CHECK: %rbx = ADD64rr %rbx, killed %rbp
+ 
+    %rbx = LEA64r killed %rbp, 1, killed %rbp, 0, _
+    RETQ %ebx
+
+...
+---
+name:            testleaadd_rbp_index_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rbx' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp, %rbx
+    ; CHECK: %rbx = LEA64r _, 1, killed %rbp, 5, _
+    ; CHECK: %rbx = ADD64rr %rbx, killed %rbp
+ 
+    %rbx = LEA64r killed %rbp, 1, killed %rbp, 5, _
+    RETQ %ebx
+
+...
+---
+name:            testleaadd_rbp_index2_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rbx' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp, %rbx
+    ; CHECK: %rbx = LEA64r _, 4, killed %rbp, 5, _
+    ; CHECK: %rbx = ADD64rr %rbx, killed %rbp
+ 
+    %rbx = LEA64r killed %rbp, 4, killed %rbp, 5, _
+    RETQ %ebx
+
+...
+---
+name:            test_skip_opt_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rbx' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp, %rbx
+    ; CHECK: %rbp = LEA64r killed %rbp, 4, killed %rbp, 0, _
+ 
+    %rbp = LEA64r killed %rbp, 4, killed %rbp, 0, _
+    RETQ %ebp
+
+...
+---
+name:            test_skip_eflags_64
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rbp' }
+  - { reg: '%rax' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp, %rbx
+    ; CHECK: %rbx = LEA64r killed %rax, 4, killed %rax, 5, _
+    ; CHECK: %rbp = LEA64r killed %rbx, 4, killed %rbx, 0, _
+    ; CHECK: %rbp = ADD64ri8 %rbp, 5
+   
+    CMP64rr   %rax, killed %rbx, implicit-def %eflags
+    %rbx = LEA64r killed %rax, 4, killed %rax, 5, _
+    JE_1 %bb.1, implicit %eflags
+    RETQ %ebx
+  bb.1:
+    liveins: %rax, %rbp, %rbx
+    %rbp = LEA64r killed %rbx, 4, killed %rbx, 5, _
+    RETQ %ebp
+
+...
+---
+name:            test_skip_opt_64_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rbx' }
+  - { reg: '%rbp' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp, %rbx
+    ; CHECK: %ebp = LEA64_32r killed %rbp, 4, killed %rbp, 0, _
+ 
+    %ebp = LEA64_32r killed %rbp, 4, killed %rbp, 0, _
+    RETQ %ebp
+
+...
+---
+name:            test_skip_eflags_64_32
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+tracksRegLiveness: true
+liveins:         
+  - { reg: '%rbp' }
+  - { reg: '%rax' }
+frameInfo:       
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    0
+  adjustsStack:    false
+  hasCalls:        false
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: %rax, %rbp, %rbx
+    ; CHECK: %ebx = LEA64_32r killed %rax, 4, killed %rax, 5, _
+    ; CHECK: %ebp = LEA64_32r killed %rbx, 4, killed %rbx, 0, _
+    ; CHECK: %ebp = ADD32ri8 %ebp, 5
+   
+    CMP64rr   %rax, killed %rbx, implicit-def %eflags
+    %ebx = LEA64_32r killed %rax, 4, killed %rax, 5, _
+    JE_1 %bb.1, implicit %eflags
+    RETQ %ebx
+  bb.1:
+    liveins: %rax, %rbp, %rbx
+    %ebp = LEA64_32r killed %rbx, 4, killed %rbx, 5, _
+    RETQ %ebp
+
+...
+
+
+