diff --git a/lib/Target/Alpha/AlphaAsmPrinter.cpp b/lib/Target/Alpha/AlphaAsmPrinter.cpp
index 348453b..acfeb91 100644
--- a/lib/Target/Alpha/AlphaAsmPrinter.cpp
+++ b/lib/Target/Alpha/AlphaAsmPrinter.cpp
@@ -119,7 +119,7 @@
     
   case MachineOperand::MO_MachineBasicBlock: {
     MachineBasicBlock *MBBOp = MO.getMachineBasicBlock();
-    O << "$LBB" << Mang->getValueName(MBBOp->getParent()->getFunction())
+    O << "LBB" << Mang->getValueName(MBBOp->getParent()->getFunction())
       << "_" << MBBOp->getNumber() << "\t" << CommentString << " "
       << MBBOp->getBasicBlock()->getName();
     return;
@@ -169,7 +169,7 @@
 
   // Print out labels for the function.
   O << "\t.text\n";
-  emitAlignment(4);
+  emitAlignment(3);
   O << "\t.globl\t" << CurrentFnName << "\n";
   O << "\t.ent\t" << CurrentFnName << "\n";
 
@@ -179,7 +179,7 @@
   for (MachineFunction::const_iterator I = MF.begin(), E = MF.end();
        I != E; ++I) {
     // Print a label for the basic block.
-    O << "$LBB" << CurrentFnName << "_" << I->getNumber() << ":\t"
+    O << "LBB" << CurrentFnName << "_" << I->getNumber() << ":\t"
       << CommentString << " " << I->getBasicBlock()->getName() << "\n";
     for (MachineBasicBlock::const_iterator II = I->begin(), E = I->end();
          II != E; ++II) {
@@ -190,7 +190,7 @@
   }
   ++LabelNumber;
 
-  O << "\t.end\t" << CurrentFnName << "\n";
+  O << "\t.end " << CurrentFnName << "\n";
 
   // We didn't modify anything.
   return false;
diff --git a/lib/Target/Alpha/AlphaISelPattern.cpp b/lib/Target/Alpha/AlphaISelPattern.cpp
index 11299bd..41028da 100644
--- a/lib/Target/Alpha/AlphaISelPattern.cpp
+++ b/lib/Target/Alpha/AlphaISelPattern.cpp
@@ -447,19 +447,38 @@
       //a = b: c = 0
       //a < b: c < 0
       //a > b: c > 0
-      unsigned Tmp1 = SelectExpr(SetCC->getOperand(0));
-      unsigned Tmp2 = SelectExpr(SetCC->getOperand(1));
-      unsigned Tmp3 = MakeReg(MVT::f64);
-      BuildMI(BB, Alpha::SUBT, 2, Tmp3).addReg(Tmp1).addReg(Tmp2);
+
+      bool invTest = false;
+      unsigned Tmp3;
+
+      ConstantFPSDNode *CN;
+      if ((CN = dyn_cast<ConstantFPSDNode>(SetCC->getOperand(1)))
+          && (CN->isExactlyValue(+0.0) || CN->isExactlyValue(-0.0)))
+        Tmp3 = SelectExpr(SetCC->getOperand(0));
+      else if ((CN = dyn_cast<ConstantFPSDNode>(SetCC->getOperand(0)))
+          && (CN->isExactlyValue(+0.0) || CN->isExactlyValue(-0.0)))
+      {
+        Tmp3 = SelectExpr(SetCC->getOperand(1));
+        invTest = true;
+      }
+      else
+      {
+        unsigned Tmp1 = SelectExpr(SetCC->getOperand(0));
+        unsigned Tmp2 = SelectExpr(SetCC->getOperand(1));
+        bool isD = SetCC->getOperand(0).getValueType() == MVT::f64;
+        Tmp3 = MakeReg(isD ? MVT::f64 : MVT::f32);
+        BuildMI(BB, isD ? Alpha::SUBT : Alpha::SUBS, 2, Tmp3)
+          .addReg(Tmp1).addReg(Tmp2);
+      }
 
       switch (SetCC->getCondition()) {
       default: CC.Val->dump(); assert(0 && "Unknown FP comparison!");
-      case ISD::SETEQ: Opc = Alpha::FBEQ; break;
-      case ISD::SETLT: Opc = Alpha::FBLT; break;
-      case ISD::SETLE: Opc = Alpha::FBLE; break;
-      case ISD::SETGT: Opc = Alpha::FBGT; break;
-      case ISD::SETGE: Opc = Alpha::FBGE; break;
-      case ISD::SETNE: Opc = Alpha::FBNE; break;
+      case ISD::SETEQ: Opc = invTest ? Alpha::FBNE : Alpha::FBEQ; break;
+      case ISD::SETLT: Opc = invTest ? Alpha::FBGT : Alpha::FBLT; break;
+      case ISD::SETLE: Opc = invTest ? Alpha::FBGE : Alpha::FBLE; break;
+      case ISD::SETGT: Opc = invTest ? Alpha::FBLT : Alpha::FBGT; break;
+      case ISD::SETGE: Opc = invTest ? Alpha::FBLE : Alpha::FBGE; break;
+      case ISD::SETNE: Opc = invTest ? Alpha::FBEQ : Alpha::FBNE; break;
       }
       BuildMI(BB, Opc, 2).addReg(Tmp3).addMBB(Dest);
       return;
@@ -598,9 +617,19 @@
     case ISD::SUB: Opc = DestType == MVT::f64 ? Alpha::SUBT : Alpha::SUBS; break;
     case ISD::SDIV: Opc = DestType == MVT::f64 ? Alpha::DIVT : Alpha::DIVS; break;
     };
-    Tmp1 = SelectExpr(N.getOperand(0));
-    Tmp2 = SelectExpr(N.getOperand(1));
-    BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2);
+
+    ConstantFPSDNode *CN;
+    if (opcode == ISD::SUB 
+        && (CN = dyn_cast<ConstantFPSDNode>(N.getOperand(0)))
+        && (CN->isExactlyValue(+0.0) || CN->isExactlyValue(-0.0)))
+    {
+      Tmp2 = SelectExpr(N.getOperand(1));
+      BuildMI(BB, Alpha::CPYSN, 2, Result).addReg(Tmp2).addReg(Tmp2);
+    } else {
+      Tmp1 = SelectExpr(N.getOperand(0));
+      Tmp2 = SelectExpr(N.getOperand(1));
+      BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2);
+    }
     return Result;
 
   case ISD::EXTLOAD:
@@ -1301,13 +1330,14 @@
     // special calling conventions
     //Restore GP because it is a call after all...
     switch(opcode) {
-    case ISD::UREM: AlphaLowering.restoreGP(BB); Opc = Alpha::REMQU; break;
-    case ISD::SREM: AlphaLowering.restoreGP(BB); Opc = Alpha::REMQ; break;
-    case ISD::UDIV: AlphaLowering.restoreGP(BB); Opc = Alpha::DIVQU; break;
-    case ISD::SDIV: AlphaLowering.restoreGP(BB); Opc = Alpha::DIVQ; break;
+    case ISD::UREM: Opc = Alpha::REMQU; break;
+    case ISD::SREM: Opc = Alpha::REMQ; break;
+    case ISD::UDIV: Opc = Alpha::DIVQU; break;
+    case ISD::SDIV: Opc = Alpha::DIVQ; break;
     }
     Tmp1 = SelectExpr(N.getOperand(0));
     Tmp2 = SelectExpr(N.getOperand(1));
+    AlphaLowering.restoreGP(BB);
     BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2);
     return Result;
 
diff --git a/lib/Target/Alpha/AlphaInstrInfo.td b/lib/Target/Alpha/AlphaInstrInfo.td
index 27ad8c0..fdf6a9a 100644
--- a/lib/Target/Alpha/AlphaInstrInfo.td
+++ b/lib/Target/Alpha/AlphaInstrInfo.td
@@ -25,7 +25,7 @@
 def s64imm  : Operand<i64>;
 
 def PHI : PseudoInstAlpha<(ops ), "#phi">;
-def IDEF : PseudoInstAlpha<(ops ), "#idef">;
+def IDEF : PseudoInstAlpha<(ops GPRC:$RA), "#idef $RA">;
 def WTF : PseudoInstAlpha<(ops ), "#wtf">;
 def ADJUSTSTACKUP : PseudoInstAlpha<(ops ), "ADJUP">;
 def ADJUSTSTACKDOWN : PseudoInstAlpha<(ops ), "ADJDOWN">;
@@ -337,18 +337,19 @@
 def CPYSN : FPForm<0x17, 0x021, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "cpysn $RA,$RB,$RC">; //Copy sign negate
 
 //Basic Floating point ops
-def ADDS  : FPForm<0x16, 0x080, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "adds $RA,$RB,$RC">;  //Add S_floating
-def ADDT  : FPForm<0x16, 0x0A0, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "addt $RA,$RB,$RC">;  //Add T_floating
-def SUBS  : FPForm<0x16, 0x081, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "subs $RA,$RB,$RC">;  //Subtract S_floating
-def SUBT  : FPForm<0x16, 0x0A1, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "subt $RA,$RB,$RC">;  //Subtract T_floating
-def DIVS  : FPForm<0x16, 0x083, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "divs $RA,$RB,$RC">;  //Divide S_floating
-def DIVT  : FPForm<0x16, 0x0A3, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "divt $RA,$RB,$RC">;  //Divide T_floating
-def MULS  : FPForm<0x16, 0x082, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "muls $RA,$RB,$RC">;  //Multiply S_floating
-def MULT  : FPForm<0x16, 0x0A2, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "mult $RA,$RB,$RC">;  //Multiply T_floating
+def ADDS  : FPForm<0x16, 0x080, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "adds/sui $RA,$RB,$RC">;  //Add S_floating
+def ADDT  : FPForm<0x16, 0x0A0, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "addt/sui $RA,$RB,$RC">;  //Add T_floating
+def SUBS  : FPForm<0x16, 0x081, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "subs/sui $RA,$RB,$RC">;  //Subtract S_floating
+def SUBT  : FPForm<0x16, 0x0A1, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "subt/sui $RA,$RB,$RC">;  //Subtract T_floating
+def DIVS  : FPForm<0x16, 0x083, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "divs/sui $RA,$RB,$RC">;  //Divide S_floating
+def DIVT  : FPForm<0x16, 0x0A3, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "divt/sui $RA,$RB,$RC">;  //Divide T_floating
+def MULS  : FPForm<0x16, 0x082, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "muls/sui $RA,$RB,$RC">;  //Multiply S_floating
+def MULT  : FPForm<0x16, 0x0A2, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "mult/sui $RA,$RB,$RC">;  //Multiply T_floating
 def SQRTS : FPForm<0x14, 0x08B, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "sqrts $RA,$RB,$RC">;  //Square root S_floating
 def SQRTT : FPForm<0x14, 0x0AB, (ops FPRC:$RC, FPRC:$RA, FPRC:$RB), "sqrtt $RA,$RB,$RC">;  //Square root T_floating
 
 //INT reg to FP reg and back again
+//not supported on 21164
 def FTOIS : FPForm<0x1C, 0x078, (ops FPRC:$RC, GPRC:$RA), "ftois $RA,$RC">; //Floating to integer move, S_floating
 def FTOIT : FPForm<0x1C, 0x070, (ops FPRC:$RC, GPRC:$RA), "ftoit $RA,$RC">; //Floating to integer move, T_floating
 def ITOFS : FPForm<0x14, 0x004, (ops FPRC:$RC, GPRC:$RA), "itofs $RA,$RC">; //Integer to floating move, S_floating