diff --git a/lib/Target/X86/InstSelectSimple.cpp b/lib/Target/X86/InstSelectSimple.cpp
index d0d80e3..7e4f689 100644
--- a/lib/Target/X86/InstSelectSimple.cpp
+++ b/lib/Target/X86/InstSelectSimple.cpp
@@ -8,6 +8,7 @@
 #include "X86InstrInfo.h"
 #include "llvm/Function.h"
 #include "llvm/iTerminators.h"
+#include "llvm/iOther.h"
 #include "llvm/Type.h"
 #include "llvm/Constants.h"
 #include "llvm/Pass.h"
@@ -55,6 +56,7 @@
     //
     void visitReturnInst(ReturnInst &RI);
     void visitAdd(BinaryOperator &B);
+    void visitShiftInst(ShiftInst &I);
 
     void visitInstruction(Instruction &I) {
       std::cerr << "Cannot instruction select: " << I;
@@ -141,6 +143,186 @@
   BuildMI(BB, X86::RET, 0);
 }
 
+/// Shift instructions: 'shl', 'sar', 'shr' - Some special cases here
+/// for constant immediate shift values, and for constant immediate
+/// shift values equal to 1. Even the general case is sort of special,
+/// because the shift amount has to be in CL, not just any old register.
+///
+void
+ISel::visitShiftInst (ShiftInst & I)
+{
+  unsigned Op0r = getReg (I.getOperand (0));
+  unsigned DestReg = getReg (I);
+  unsigned operandSize = I.getOperand (0)->getType ()->getPrimitiveSize ();
+  bool isRightShift = (I.getOpcode () == Instruction::Shr);
+  bool isOperandUnsigned = I.getType ()->isUnsigned ();
+  bool isConstantShiftAmount = (isa <ConstantUInt> (I.getOperand (1)));
+  if (ConstantUInt *CUI = dyn_cast <ConstantUInt> (I.getOperand (1)))
+    {
+      // The shift amount is constant. Get its value.
+      uint64_t shAmt = CUI->getValue ();
+      // Emit: <insn> reg, shamt  (shift-by-immediate opcode "ir" form.)
+      if (isRightShift)
+	{
+	  if (isOperandUnsigned)
+	    {
+	      // This is a shift right logical (SHR).
+	      switch (operandSize)
+		{
+		case 1:
+		  BuildMI (BB, X86::SHRir8, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 2:
+		  BuildMI (BB, X86::SHRir16, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 4:
+		  BuildMI (BB, X86::SHRir32, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 8:
+		default:
+		  visitInstruction (I);
+		  break;
+		}
+	    }
+	  else
+	    {
+	      // This is a shift right arithmetic (SAR).
+	      switch (operandSize)
+		{
+		case 1:
+		  BuildMI (BB, X86::SARir8, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 2:
+		  BuildMI (BB, X86::SARir16, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 4:
+		  BuildMI (BB, X86::SARir32, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 8:
+		default:
+		  visitInstruction (I);
+		  break;
+		}
+	    }
+	}
+      else
+	{
+	  // This is a left shift (SHL).
+	  switch (operandSize)
+	    {
+	    case 1:
+	      BuildMI (BB, X86::SHLir8, 2,
+		       DestReg).addReg (Op0r).addZImm (shAmt);
+	      break;
+	    case 2:
+	      BuildMI (BB, X86::SHLir16, 2,
+		       DestReg).addReg (Op0r).addZImm (shAmt);
+	      break;
+	    case 4:
+	      BuildMI (BB, X86::SHLir32, 2,
+		       DestReg).addReg (Op0r).addZImm (shAmt);
+	      break;
+	    case 8:
+	    default:
+	      visitInstruction (I);
+	      break;
+	    }
+	}
+    }
+  else
+    {
+      // The shift amount is non-constant.
+      //
+      // In fact, you can only shift with a variable shift amount if
+      // that amount is already in the CL register, so we have to put it
+      // there first.
+      //
+      // Get it from the register it's in.
+      unsigned Op1r = getReg (I.getOperand (1));
+      // Emit: move cl, shiftAmount (put the shift amount in CL.)
+      BuildMI (BB, X86::MOVrr8, 2, X86::CL).addReg (Op1r);
+      // Emit: <insn> reg, cl       (shift-by-CL opcode; "rr" form.)
+      if (isRightShift)
+	{
+	  if (isOperandUnsigned)
+	    {
+	      // This is a shift right logical (SHR).
+	      switch (operandSize)
+		{
+		case 1:
+		  BuildMI (BB, X86::SHRrr8, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 2:
+		  BuildMI (BB, X86::SHRrr16, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 4:
+		  BuildMI (BB, X86::SHRrr32, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 8:
+		default:
+		  visitInstruction (I);
+		  break;
+		}
+	    }
+	  else
+	    {
+	      // This is a shift right arithmetic (SAR).
+	      switch (operandSize)
+		{
+		case 1:
+		  BuildMI (BB, X86::SARrr8, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 2:
+		  BuildMI (BB, X86::SARrr16, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 4:
+		  BuildMI (BB, X86::SARrr32, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 8:
+		default:
+		  visitInstruction (I);
+		  break;
+		}
+	    }
+	}
+      else
+	{
+	  // This is a left shift (SHL).
+	  switch (operandSize)
+	    {
+	    case 1:
+	      BuildMI (BB, X86::SHLrr8, 2,
+		       DestReg).addReg (Op0r).addReg (X86::CL);
+	      break;
+	    case 2:
+	      BuildMI (BB, X86::SHLrr16, 2,
+		       DestReg).addReg (Op0r).addReg (X86::CL);
+	      break;
+	    case 4:
+	      BuildMI (BB, X86::SHLrr32, 2,
+		       DestReg).addReg (Op0r).addReg (X86::CL);
+	      break;
+	    case 8:
+	    default:
+	      visitInstruction (I);
+	      break;
+	    }
+	}
+    }
+}
+
 
 /// 'add' instruction - Simply turn this into an x86 reg,reg add instruction.
 void ISel::visitAdd(BinaryOperator &B) {
@@ -157,13 +339,18 @@
   case 4:   // UInt, Int
     BuildMI(BB, X86::ADDrr32, 2, DestReg).addReg(Op0r).addReg(Op1r);
     break;
-
   case 8:   // ULong, Long
+    // Here we have a pair of operands each occupying a pair of registers.
+    // We need to do an ADDrr32 of the least-significant pair immediately
+    // followed by an ADCrr32 (Add with Carry) of the most-significant pair.
+    // I don't know how we are representing these multi-register arguments.
   default:
     visitInstruction(B);  // abort
   }
 }
 
+
+
 /// createSimpleX86InstructionSelector - This pass converts an LLVM function
 /// into a machine code representation is a very simple peep-hole fashion.  The
 /// generated code sucks but the implementation is nice and simple.
diff --git a/lib/Target/X86/X86ISelSimple.cpp b/lib/Target/X86/X86ISelSimple.cpp
index d0d80e3..7e4f689 100644
--- a/lib/Target/X86/X86ISelSimple.cpp
+++ b/lib/Target/X86/X86ISelSimple.cpp
@@ -8,6 +8,7 @@
 #include "X86InstrInfo.h"
 #include "llvm/Function.h"
 #include "llvm/iTerminators.h"
+#include "llvm/iOther.h"
 #include "llvm/Type.h"
 #include "llvm/Constants.h"
 #include "llvm/Pass.h"
@@ -55,6 +56,7 @@
     //
     void visitReturnInst(ReturnInst &RI);
     void visitAdd(BinaryOperator &B);
+    void visitShiftInst(ShiftInst &I);
 
     void visitInstruction(Instruction &I) {
       std::cerr << "Cannot instruction select: " << I;
@@ -141,6 +143,186 @@
   BuildMI(BB, X86::RET, 0);
 }
 
+/// Shift instructions: 'shl', 'sar', 'shr' - Some special cases here
+/// for constant immediate shift values, and for constant immediate
+/// shift values equal to 1. Even the general case is sort of special,
+/// because the shift amount has to be in CL, not just any old register.
+///
+void
+ISel::visitShiftInst (ShiftInst & I)
+{
+  unsigned Op0r = getReg (I.getOperand (0));
+  unsigned DestReg = getReg (I);
+  unsigned operandSize = I.getOperand (0)->getType ()->getPrimitiveSize ();
+  bool isRightShift = (I.getOpcode () == Instruction::Shr);
+  bool isOperandUnsigned = I.getType ()->isUnsigned ();
+  bool isConstantShiftAmount = (isa <ConstantUInt> (I.getOperand (1)));
+  if (ConstantUInt *CUI = dyn_cast <ConstantUInt> (I.getOperand (1)))
+    {
+      // The shift amount is constant. Get its value.
+      uint64_t shAmt = CUI->getValue ();
+      // Emit: <insn> reg, shamt  (shift-by-immediate opcode "ir" form.)
+      if (isRightShift)
+	{
+	  if (isOperandUnsigned)
+	    {
+	      // This is a shift right logical (SHR).
+	      switch (operandSize)
+		{
+		case 1:
+		  BuildMI (BB, X86::SHRir8, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 2:
+		  BuildMI (BB, X86::SHRir16, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 4:
+		  BuildMI (BB, X86::SHRir32, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 8:
+		default:
+		  visitInstruction (I);
+		  break;
+		}
+	    }
+	  else
+	    {
+	      // This is a shift right arithmetic (SAR).
+	      switch (operandSize)
+		{
+		case 1:
+		  BuildMI (BB, X86::SARir8, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 2:
+		  BuildMI (BB, X86::SARir16, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 4:
+		  BuildMI (BB, X86::SARir32, 2,
+			   DestReg).addReg (Op0r).addZImm (shAmt);
+		  break;
+		case 8:
+		default:
+		  visitInstruction (I);
+		  break;
+		}
+	    }
+	}
+      else
+	{
+	  // This is a left shift (SHL).
+	  switch (operandSize)
+	    {
+	    case 1:
+	      BuildMI (BB, X86::SHLir8, 2,
+		       DestReg).addReg (Op0r).addZImm (shAmt);
+	      break;
+	    case 2:
+	      BuildMI (BB, X86::SHLir16, 2,
+		       DestReg).addReg (Op0r).addZImm (shAmt);
+	      break;
+	    case 4:
+	      BuildMI (BB, X86::SHLir32, 2,
+		       DestReg).addReg (Op0r).addZImm (shAmt);
+	      break;
+	    case 8:
+	    default:
+	      visitInstruction (I);
+	      break;
+	    }
+	}
+    }
+  else
+    {
+      // The shift amount is non-constant.
+      //
+      // In fact, you can only shift with a variable shift amount if
+      // that amount is already in the CL register, so we have to put it
+      // there first.
+      //
+      // Get it from the register it's in.
+      unsigned Op1r = getReg (I.getOperand (1));
+      // Emit: move cl, shiftAmount (put the shift amount in CL.)
+      BuildMI (BB, X86::MOVrr8, 2, X86::CL).addReg (Op1r);
+      // Emit: <insn> reg, cl       (shift-by-CL opcode; "rr" form.)
+      if (isRightShift)
+	{
+	  if (isOperandUnsigned)
+	    {
+	      // This is a shift right logical (SHR).
+	      switch (operandSize)
+		{
+		case 1:
+		  BuildMI (BB, X86::SHRrr8, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 2:
+		  BuildMI (BB, X86::SHRrr16, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 4:
+		  BuildMI (BB, X86::SHRrr32, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 8:
+		default:
+		  visitInstruction (I);
+		  break;
+		}
+	    }
+	  else
+	    {
+	      // This is a shift right arithmetic (SAR).
+	      switch (operandSize)
+		{
+		case 1:
+		  BuildMI (BB, X86::SARrr8, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 2:
+		  BuildMI (BB, X86::SARrr16, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 4:
+		  BuildMI (BB, X86::SARrr32, 2,
+			   DestReg).addReg (Op0r).addReg (X86::CL);
+		  break;
+		case 8:
+		default:
+		  visitInstruction (I);
+		  break;
+		}
+	    }
+	}
+      else
+	{
+	  // This is a left shift (SHL).
+	  switch (operandSize)
+	    {
+	    case 1:
+	      BuildMI (BB, X86::SHLrr8, 2,
+		       DestReg).addReg (Op0r).addReg (X86::CL);
+	      break;
+	    case 2:
+	      BuildMI (BB, X86::SHLrr16, 2,
+		       DestReg).addReg (Op0r).addReg (X86::CL);
+	      break;
+	    case 4:
+	      BuildMI (BB, X86::SHLrr32, 2,
+		       DestReg).addReg (Op0r).addReg (X86::CL);
+	      break;
+	    case 8:
+	    default:
+	      visitInstruction (I);
+	      break;
+	    }
+	}
+    }
+}
+
 
 /// 'add' instruction - Simply turn this into an x86 reg,reg add instruction.
 void ISel::visitAdd(BinaryOperator &B) {
@@ -157,13 +339,18 @@
   case 4:   // UInt, Int
     BuildMI(BB, X86::ADDrr32, 2, DestReg).addReg(Op0r).addReg(Op1r);
     break;
-
   case 8:   // ULong, Long
+    // Here we have a pair of operands each occupying a pair of registers.
+    // We need to do an ADDrr32 of the least-significant pair immediately
+    // followed by an ADCrr32 (Add with Carry) of the most-significant pair.
+    // I don't know how we are representing these multi-register arguments.
   default:
     visitInstruction(B);  // abort
   }
 }
 
+
+
 /// createSimpleX86InstructionSelector - This pass converts an LLVM function
 /// into a machine code representation is a very simple peep-hole fashion.  The
 /// generated code sucks but the implementation is nice and simple.
diff --git a/lib/Target/X86/X86InstrInfo.def b/lib/Target/X86/X86InstrInfo.def
index 2388b9b..519e768 100644
--- a/lib/Target/X86/X86InstrInfo.def
+++ b/lib/Target/X86/X86InstrInfo.def
@@ -37,15 +37,37 @@
 I(RET         , "ret",       M_RET_FLAG, X86II::Void) // ret          CB
 
 // Move instructions
-I(MOVir8      , "movb",               0, 0)           // R = imm8     B0+ rb
-I(MOVir16     , "movw",               0, 0)           // R = imm16    B8+ rw
-I(MOVir32     , "movl",               0, 0)           // R = imm32    B8+ rd
+I(MOVrr8      , "movb",               0, 0)           // R8  = R8     88/r
+I(MOVrr16     , "movw",               0, 0)           // R16 = R16    89/r
+I(MOVrr32     , "movl",               0, 0)           // R32 = R32    89/r
+I(MOVir8      , "movb",               0, 0)           // R8  = imm8   B0+ rb
+I(MOVir16     , "movw",               0, 0)           // R16 = imm16  B8+ rw
+I(MOVir32     , "movl",               0, 0)           // R32 = imm32  B8+ rd
 
 // Arithmetic instructions
 I(ADDrr8      , "addb",               0, 0)           // R8  += R8    00/r
 I(ADDrr16     , "addw",               0, 0)           // R16 += R16   01/r
 I(ADDrr32     , "addl",               0, 0)           // R32 += R32   02/r
 
+// Shift instructions
+I(SHLrr8      , "shlb",               0, 0)           // R8   <<= cl   D2/4
+I(SHLir8      , "shlb",               0, 0)           // R8   <<= imm8 C0/4 ib
+I(SHLrr16     , "shlw",               0, 0)           // R16  <<= cl   D3/4
+I(SHLir16     , "shlw",               0, 0)           // R16  <<= imm8 C1/4 ib
+I(SHLrr32     , "shll",               0, 0)           // R32  <<= cl   D3/4
+I(SHLir32     , "shll",               0, 0)           // R32  <<= imm8 C1/4 ib
+I(SHRrr8      , "shrb",               0, 0)           // R8  >>>= cl   D2/5
+I(SHRir8      , "shrb",               0, 0)           // R8  >>>= imm8 C0/5 ib
+I(SHRrr16     , "shrw",               0, 0)           // R16 >>>= cl   D3/5
+I(SHRir16     , "shrw",               0, 0)           // R16 >>>= imm8 C1/5 ib
+I(SHRrr32     , "shrl",               0, 0)           // R32 >>>= cl   D3/5
+I(SHRir32     , "shrl",               0, 0)           // R32 >>>= imm8 C1/5 ib
+I(SARrr8      , "sarb",               0, 0)           // R8   >>= cl   D2/7
+I(SARir8      , "sarb",               0, 0)           // R8   >>= imm8 C0/7 ib
+I(SARrr16     , "sarw",               0, 0)           // R16  >>= cl   D3/7
+I(SARir16     , "sarw",               0, 0)           // R16  >>= imm8 C1/7 ib
+I(SARrr32     , "sarl",               0, 0)           // R32  >>= cl   D3/7
+I(SARir32     , "sarl",               0, 0)           // R32  >>= imm8 C1/7 ib
 
 // At this point, I is dead, so undefine the macro
 #undef I