CellSPU:

- Fix bug 3185, with misc other cleanups.
- Needed to implement SPUInstrInfo::InsertBranch(). CAUTION: Not sure what
  gets or needs to get passed to InsertBranch() to insert a conditional
  branch. This will abort for now until a good test case shows up.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@60811 91177308-0d34-0410-b5e6-96231b3b80d8

9 files changed, 286 insertions, 84 deletions

diff --git a/lib/Target/CellSPU/AsmPrinter/SPUAsmPrinter.cpp b/lib/Target/CellSPU/AsmPrinter/SPUAsmPrinter.cpp
index 94d8cd5..589a260 100644
--- a/lib/Target/CellSPU/AsmPrinter/SPUAsmPrinter.cpp
+++ b/lib/Target/CellSPU/AsmPrinter/SPUAsmPrinter.cpp
@@ -220,6 +220,18 @@ namespace {
     }
 
     void printPCRelativeOperand(const MachineInstr *MI, unsigned OpNo) {
+      // Used to generate a ".-<target>", but it turns out that the assembler
+      // really wants the target.
+      //
+      // N.B.: This operand is used for call targets. Branch hints are another
+      // animal entirely.
+      printOp(MI->getOperand(OpNo));
+    }
+
+    void printHBROperand(const MachineInstr *MI, unsigned OpNo) {
+      // HBR operands are generated in front of branches, hence, the
+      // program counter plus the target.
+      O << ".+";
       printOp(MI->getOperand(OpNo));
     }
 
diff --git a/lib/Target/CellSPU/SPUISelDAGToDAG.cpp b/lib/Target/CellSPU/SPUISelDAGToDAG.cpp
index 14f3edd..9ac0e2e 100644
--- a/lib/Target/CellSPU/SPUISelDAGToDAG.cpp
+++ b/lib/Target/CellSPU/SPUISelDAGToDAG.cpp
@@ -557,10 +557,7 @@ SPUDAGToDAGISel::DFormAddressPredicate(SDValue Op, SDValue N, SDValue &Base,
       else
         Addr = N;                       // Register
 
-      if (OpOpc == ISD::STORE)
-        Offs = Op.getOperand(3);
-      else
-        Offs = Op.getOperand(2);        // LOAD
+      Offs = ((OpOpc == ISD::STORE) ? Op.getOperand(3) : Op.getOperand(2));
 
       if (Offs.getOpcode() == ISD::Constant || Offs.getOpcode() == ISD::UNDEF) {
         if (Offs.getOpcode() == ISD::UNDEF)
@@ -570,6 +567,16 @@ SPUDAGToDAGISel::DFormAddressPredicate(SDValue Op, SDValue N, SDValue &Base,
         Index = Addr;
         return true;
       }
+    } else {
+      /* If otherwise unadorned, default to D-form address with 0 offset: */
+      if (Opc == ISD::CopyFromReg) {
+	Index = N.getOperand(1);
+      } else {
+	Index = N;
+      }
+
+      Base = CurDAG->getTargetConstant(0, Index.getValueType());
+      return true;
     }
   }
 
diff --git a/lib/Target/CellSPU/SPUISelLowering.cpp b/lib/Target/CellSPU/SPUISelLowering.cpp
index 031d068..c3c31e0 100644
--- a/lib/Target/CellSPU/SPUISelLowering.cpp
+++ b/lib/Target/CellSPU/SPUISelLowering.cpp
@@ -28,7 +28,6 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Target/TargetOptions.h"
-#include "llvm/CodeGen/SchedulerRegistry.h"
 
 #include <map>
 
@@ -131,9 +130,6 @@ SPUTargetLowering::SPUTargetLowering(SPUTargetMachine &TM)
   addRegisterClass(MVT::f64,  SPU::R64FPRegisterClass);
   addRegisterClass(MVT::i128, SPU::GPRCRegisterClass);
 
-  // Initialize libcalls:
-  setLibcallName(RTLIB::MUL_I64, "__muldi3");
-
   // SPU has no sign or zero extended loads for i1, i8, i16:
   setLoadExtAction(ISD::EXTLOAD,  MVT::i1, Promote);
   setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
@@ -237,10 +233,12 @@ SPUTargetLowering::SPUTargetLowering(SPUTargetMachine &TM)
   setOperationAction(ISD::MUL,  MVT::i64,    Expand);   // libcall
 
   // SMUL_LOHI, UMUL_LOHI
-  setOperationAction(ISD::SMUL_LOHI, MVT::i32, Custom);
-  setOperationAction(ISD::UMUL_LOHI, MVT::i32, Custom);
-  setOperationAction(ISD::SMUL_LOHI, MVT::i64, Custom);
-  setOperationAction(ISD::UMUL_LOHI, MVT::i64, Custom);
+#if 0
+  setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
+  setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
+  setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);
+  setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
+#endif
 
   // Need to custom handle (some) common i8, i64 math ops
   setOperationAction(ISD::ADD,  MVT::i64,    Custom);
diff --git a/lib/Target/CellSPU/SPUInstrInfo.cpp b/lib/Target/CellSPU/SPUInstrInfo.cpp
index bb89306..442d491 100644
--- a/lib/Target/CellSPU/SPUInstrInfo.cpp
+++ b/lib/Target/CellSPU/SPUInstrInfo.cpp
@@ -21,6 +21,26 @@
 
 using namespace llvm;
 
+namespace {
+  //! Predicate for an unconditional branch instruction
+  inline bool isUncondBranch(const MachineInstr *I) {
+    unsigned opc = I->getOpcode();
+
+    return (opc == SPU::BR
+	    || opc == SPU::BRA
+	    || opc == SPU::BI);
+  }
+
+  inline bool isCondBranch(const MachineInstr *I) {
+    unsigned opc = I->getOpcode();
+
+    return (opc == SPU::BRNZ
+	    || opc == SPU::BRZ
+	    || opc == SPU::BRHNZ
+	    || opc == SPU::BRHZ);
+  }
+}
+
 SPUInstrInfo::SPUInstrInfo(SPUTargetMachine &tm)
   : TargetInstrInfoImpl(SPUInsts, sizeof(SPUInsts)/sizeof(SPUInsts[0])),
     TM(tm),
@@ -131,14 +151,28 @@ SPUInstrInfo::isLoadFromStackSlot(const MachineInstr *MI,
   case SPU::LQDr128:
   case SPU::LQDr64:
   case SPU::LQDr32:
-  case SPU::LQDr16:
+  case SPU::LQDr16: {
+    const MachineOperand MOp1 = MI->getOperand(1);
+    const MachineOperand MOp2 = MI->getOperand(2);
+    if (MOp1.isImm()
+	&& (MOp2.isFI()
+	    || (MOp2.isReg() && MOp2.getReg() == SPU::R1))) {
+      if (MOp2.isFI())
+	FrameIndex = MOp2.getIndex();
+      else
+	FrameIndex = MOp1.getImm() / SPUFrameInfo::stackSlotSize();
+      return MI->getOperand(0).getReg();
+    }
+    break;
+  }
   case SPU::LQXv4i32:
   case SPU::LQXr128:
   case SPU::LQXr64:
   case SPU::LQXr32:
   case SPU::LQXr16:
-    if (MI->getOperand(1).isImm() && !MI->getOperand(1).getImm() &&
-        MI->getOperand(2).isFI()) {
+    if (MI->getOperand(1).isReg() && MI->getOperand(2).isReg()
+	&& (MI->getOperand(2).getReg() == SPU::R1
+	    || MI->getOperand(1).getReg() == SPU::R1)) {
       FrameIndex = MI->getOperand(2).getIndex();
       return MI->getOperand(0).getReg();
     }
@@ -161,7 +195,20 @@ SPUInstrInfo::isStoreToStackSlot(const MachineInstr *MI,
   case SPU::STQDr64:
   case SPU::STQDr32:
   case SPU::STQDr16:
-  case SPU::STQDr8:
+  case SPU::STQDr8: {
+    const MachineOperand MOp1 = MI->getOperand(1);
+    const MachineOperand MOp2 = MI->getOperand(2);
+    if (MOp1.isImm()
+	&& (MOp2.isFI()
+	    || (MOp2.isReg() && MOp2.getReg() == SPU::R1))) {
+      if (MOp2.isFI())
+	FrameIndex = MOp2.getIndex();
+      else
+	FrameIndex = MOp1.getImm() / SPUFrameInfo::stackSlotSize();
+      return MI->getOperand(0).getReg();
+    }
+    break;
+  }
   case SPU::STQXv16i8:
   case SPU::STQXv8i16:
   case SPU::STQXv4i32:
@@ -172,8 +219,9 @@ SPUInstrInfo::isStoreToStackSlot(const MachineInstr *MI,
   case SPU::STQXr32:
   case SPU::STQXr16:
   case SPU::STQXr8:
-    if (MI->getOperand(1).isImm() && !MI->getOperand(1).getImm() &&
-        MI->getOperand(2).isFI()) {
+    if (MI->getOperand(1).isReg() && MI->getOperand(2).isReg()
+	&& (MI->getOperand(2).getReg() == SPU::R1
+	    || MI->getOperand(1).getReg() == SPU::R1)) {
       FrameIndex = MI->getOperand(2).getIndex();
       return MI->getOperand(0).getReg();
     }
@@ -193,11 +241,6 @@ bool SPUInstrInfo::copyRegToReg(MachineBasicBlock &MBB,
   // we instruction select bitconvert i64 -> f64 as a noop for example, so our
   // types have no specific meaning.
   
-  //if (DestRC != SrcRC) {
-  //  cerr << "SPUInstrInfo::copyRegToReg(): DestRC != SrcRC not supported!\n";
-  //  abort();
-  //}
-
   if (DestRC == SPU::R8CRegisterClass) {
     BuildMI(MBB, MI, get(SPU::ORBIr8), DestReg).addReg(SrcReg).addImm(0);
   } else if (DestRC == SPU::R16CRegisterClass) {
@@ -234,30 +277,21 @@ SPUInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
                                      const TargetRegisterClass *RC) const
 {
   unsigned opc;
+  bool isValidFrameIdx = (FrameIdx < SPUFrameInfo::maxFrameOffset());
   if (RC == SPU::GPRCRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::STQDr128
-      : SPU::STQXr128;
+    opc = (isValidFrameIdx ? SPU::STQDr128 : SPU::STQXr128);
   } else if (RC == SPU::R64CRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::STQDr64
-      : SPU::STQXr64;
+    opc = (isValidFrameIdx ? SPU::STQDr64 : SPU::STQXr64);
   } else if (RC == SPU::R64FPRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::STQDr64
-      : SPU::STQXr64;
+    opc = (isValidFrameIdx ? SPU::STQDr64 : SPU::STQXr64);
   } else if (RC == SPU::R32CRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::STQDr32
-      : SPU::STQXr32;
+    opc = (isValidFrameIdx ? SPU::STQDr32 : SPU::STQXr32);
   } else if (RC == SPU::R32FPRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::STQDr32
-      : SPU::STQXr32;
+    opc = (isValidFrameIdx ? SPU::STQDr32 : SPU::STQXr32);
   } else if (RC == SPU::R16CRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset()) ?
-      SPU::STQDr16
-      : SPU::STQXr16;
+    opc = (isValidFrameIdx ? SPU::STQDr16 : SPU::STQXr16);
+  } else if (RC == SPU::R8CRegisterClass) {
+    opc = (isValidFrameIdx ? SPU::STQDr8 : SPU::STQXr8);
   } else {
     assert(0 && "Unknown regclass!");
     abort();
@@ -317,30 +351,21 @@ SPUInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
                                         const TargetRegisterClass *RC) const
 {
   unsigned opc;
+  bool isValidFrameIdx = (FrameIdx < SPUFrameInfo::maxFrameOffset());
   if (RC == SPU::GPRCRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::LQDr128
-      : SPU::LQXr128;
+    opc = (isValidFrameIdx ? SPU::LQDr128 : SPU::LQXr128);
   } else if (RC == SPU::R64CRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::LQDr64
-      : SPU::LQXr64;
+    opc = (isValidFrameIdx ? SPU::LQDr64 : SPU::LQXr64);
   } else if (RC == SPU::R64FPRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::LQDr64
-      : SPU::LQXr64;
+    opc = (isValidFrameIdx ? SPU::LQDr64 : SPU::LQXr64);
   } else if (RC == SPU::R32CRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::LQDr32
-      : SPU::LQXr32;
+    opc = (isValidFrameIdx ? SPU::LQDr32 : SPU::LQXr32);
   } else if (RC == SPU::R32FPRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::LQDr32
-      : SPU::LQXr32;
+    opc = (isValidFrameIdx ? SPU::LQDr32 : SPU::LQXr32);
   } else if (RC == SPU::R16CRegisterClass) {
-    opc = (FrameIdx < SPUFrameInfo::maxFrameOffset())
-      ? SPU::LQDr16
-      : SPU::LQXr16;
+    opc = (isValidFrameIdx ? SPU::LQDr16 : SPU::LQXr16);
+  } else if (RC == SPU::R8CRegisterClass) {
+    opc = (isValidFrameIdx ? SPU::LQDr8 : SPU::LQXr8);
   } else {
     assert(0 && "Unknown regclass in loadRegFromStackSlot!");
     abort();
@@ -353,9 +378,9 @@ SPUInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
   \note We are really pessimistic here about what kind of a load we're doing.
  */
 void SPUInstrInfo::loadRegFromAddr(MachineFunction &MF, unsigned DestReg,
-                                      SmallVectorImpl<MachineOperand> &Addr,
-                                      const TargetRegisterClass *RC,
-                                      SmallVectorImpl<MachineInstr*> &NewMIs)
+                                   SmallVectorImpl<MachineOperand> &Addr,
+                                   const TargetRegisterClass *RC,
+                                   SmallVectorImpl<MachineInstr*> &NewMIs)
     const {
   cerr << "loadRegToAddr() invoked!\n";
   abort();
@@ -438,3 +463,126 @@ SPUInstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
 #endif
 }
 
+//! Branch analysis
+/*
+  \note This code was kiped from PPC. There may be more branch analysis for
+  CellSPU than what's currently done here.
+ */
+bool
+SPUInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
+			    MachineBasicBlock *&FBB,
+			    SmallVectorImpl<MachineOperand> &Cond) const {
+  // If the block has no terminators, it just falls into the block after it.
+  MachineBasicBlock::iterator I = MBB.end();
+  if (I == MBB.begin() || !isUnpredicatedTerminator(--I))
+    return false;
+
+  // Get the last instruction in the block.
+  MachineInstr *LastInst = I;
+  
+  // If there is only one terminator instruction, process it.
+  if (I == MBB.begin() || !isUnpredicatedTerminator(--I)) {
+    if (isUncondBranch(LastInst)) {
+      TBB = LastInst->getOperand(0).getMBB();
+      return false;
+    } else if (isCondBranch(LastInst)) {
+      // Block ends with fall-through condbranch.
+      TBB = LastInst->getOperand(1).getMBB();
+      Cond.push_back(LastInst->getOperand(0));
+      Cond.push_back(LastInst->getOperand(1));
+      return false;
+    }
+    // Otherwise, don't know what this is.
+    return true;
+  }
+  
+  // Get the instruction before it if it's a terminator.
+  MachineInstr *SecondLastInst = I;
+
+  // If there are three terminators, we don't know what sort of block this is.
+  if (SecondLastInst && I != MBB.begin() &&
+      isUnpredicatedTerminator(--I))
+    return true;
+  
+  // If the block ends with a conditional and unconditional branch, handle it.
+  if (isCondBranch(SecondLastInst) && isUncondBranch(LastInst)) {
+    TBB =  SecondLastInst->getOperand(1).getMBB();
+    Cond.push_back(SecondLastInst->getOperand(0));
+    Cond.push_back(SecondLastInst->getOperand(1));
+    FBB = LastInst->getOperand(0).getMBB();
+    return false;
+  }
+  
+  // If the block ends with two unconditional branches, handle it.  The second
+  // one is not executed, so remove it.
+  if (isUncondBranch(SecondLastInst) && isUncondBranch(LastInst)) {
+    TBB = SecondLastInst->getOperand(0).getMBB();
+    I = LastInst;
+    I->eraseFromParent();
+    return false;
+  }
+
+  // Otherwise, can't handle this.
+  return true;
+}
+    
+unsigned
+SPUInstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
+  MachineBasicBlock::iterator I = MBB.end();
+  if (I == MBB.begin())
+    return 0;
+  --I;
+  if (!isCondBranch(I) && !isUncondBranch(I))
+    return 0;
+
+  // Remove the first branch.
+  I->eraseFromParent();
+  I = MBB.end();
+  if (I == MBB.begin())
+    return 1;
+
+  --I;
+  if (isCondBranch(I))
+    return 1;
+
+  // Remove the second branch.
+  I->eraseFromParent();
+  return 2;
+}
+    
+unsigned
+SPUInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
+			   MachineBasicBlock *FBB,
+			   const SmallVectorImpl<MachineOperand> &Cond) const {
+  // Shouldn't be a fall through.
+  assert(TBB && "InsertBranch must not be told to insert a fallthrough");
+  assert((Cond.size() == 2 || Cond.size() == 0) && 
+         "SPU branch conditions have two components!");
+  
+  // One-way branch.
+  if (FBB == 0) {
+    if (Cond.empty())   // Unconditional branch
+      BuildMI(&MBB, get(SPU::BR)).addMBB(TBB);
+    else {              // Conditional branch
+      /* BuildMI(&MBB, get(SPU::BRNZ))
+        .addImm(Cond[0].getImm()).addReg(Cond[1].getReg()).addMBB(TBB); */
+      cerr << "SPUInstrInfo::InsertBranch conditional branch logic needed\n";
+      abort();
+    }
+    return 1;
+  }
+  
+  // Two-way Conditional Branch.
+#if 0
+  BuildMI(&MBB, get(SPU::BRNZ))
+    .addImm(Cond[0].getImm()).addReg(Cond[1].getReg()).addMBB(TBB);
+  BuildMI(&MBB, get(SPU::BR)).addMBB(FBB);
+#else
+  cerr << "SPUInstrInfo::InsertBranch conditional branch logic needed\n";
+  abort();
+#endif
+
+  return 2;
+}
+
+
diff --git a/lib/Target/CellSPU/SPUInstrInfo.h b/lib/Target/CellSPU/SPUInstrInfo.h
index 5c59b68..27766ba 100644
--- a/lib/Target/CellSPU/SPUInstrInfo.h
+++ b/lib/Target/CellSPU/SPUInstrInfo.h
@@ -91,7 +91,17 @@ namespace llvm {
                                                 MachineInstr* LoadMI) const {
       return 0;
     }
-  };
+
+    virtual bool AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
+			       MachineBasicBlock *&FBB,
+			       SmallVectorImpl<MachineOperand> &Cond) const;
+    
+    virtual unsigned RemoveBranch(MachineBasicBlock &MBB) const;
+    
+    virtual unsigned InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
+			      MachineBasicBlock *FBB,
+			      const SmallVectorImpl<MachineOperand> &Cond) const;
+   };
 }
 
 #endif
diff --git a/lib/Target/CellSPU/SPUInstrInfo.td b/lib/Target/CellSPU/SPUInstrInfo.td
index 678f8e9..2338a03 100644
--- a/lib/Target/CellSPU/SPUInstrInfo.td
+++ b/lib/Target/CellSPU/SPUInstrInfo.td
@@ -1124,7 +1124,7 @@ defm ANDC : AndComplement;
 
 class ANDBIInst<dag OOL, dag IOL, list<dag> pattern>:
     RI10Form<0b01101000, OOL, IOL, "andbi\t$rT, $rA, $val",
-             IntegerOp, pattern>;
+             ByteOp, pattern>;
 
 multiclass AndByteImm
 {
@@ -1141,7 +1141,7 @@ defm ANDBI : AndByteImm;
 
 class ANDHIInst<dag OOL, dag IOL, list<dag> pattern> :
     RI10Form<0b10101000, OOL, IOL, "andhi\t$rT, $rA, $val",
-             IntegerOp, pattern>;
+             ByteOp, pattern>;
 
 multiclass AndHalfwordImm
 {
@@ -3394,25 +3394,39 @@ let isTerminator = 1, isBarrier = 1 in {
 // Single precision floating point instructions
 //===----------------------------------------------------------------------===//
 
-def FAv4f32:
-    RRForm<0b00100011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "fa\t$rT, $rA, $rB", SPrecFP,
-      [(set (v4f32 VECREG:$rT), (fadd (v4f32 VECREG:$rA), (v4f32 VECREG:$rB)))]>;
+class FAInst<dag OOL, dag IOL, list<dag> pattern>:
+    RRForm<0b01011000100, OOL, IOL, "fa\t$rT, $rA, $rB",
+	   SPrecFP, pattern>;
+class FAVecInst<ValueType vectype>:
+    FAInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
+             [(set (vectype VECREG:$rT),
+		   (fadd (vectype VECREG:$rA), (vectype VECREG:$rB)))]>;
+multiclass SFPAdd
+{
+  def v4f32: FAVecInst<v4f32>;
+  def r32:   FAInst<(outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB),
+		    [(set R32FP:$rT, (fadd R32FP:$rA, R32FP:$rB))]>;
+}
 
-def FAf32 :
-    RRForm<0b00100011010, (outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB),
-      "fa\t$rT, $rA, $rB", SPrecFP,
-      [(set R32FP:$rT, (fadd R32FP:$rA, R32FP:$rB))]>;
+defm FA : SFPAdd;
 
-def FSv4f32:
-    RRForm<0b00100011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "fs\t$rT, $rA, $rB", SPrecFP,
-      [(set (v4f32 VECREG:$rT), (fsub (v4f32 VECREG:$rA), (v4f32 VECREG:$rB)))]>;
+class FSInst<dag OOL, dag IOL, list<dag> pattern>:
+    RRForm<0b01011000100, OOL, IOL, "fs\t$rT, $rA, $rB",
+	   SPrecFP, pattern>;
 
-def FSf32 :
-    RRForm<0b10100011010, (outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB),
-      "fs\t$rT, $rA, $rB", SPrecFP,
-      [(set R32FP:$rT, (fsub R32FP:$rA, R32FP:$rB))]>;
+class FSVecInst<ValueType vectype>:
+    FSInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
+    	   [(set (vectype VECREG:$rT),
+	         (fsub (vectype VECREG:$rA), (vectype VECREG:$rB)))]>;
+
+multiclass SFPSub
+{
+  def v4f32: FSVecInst<v4f32>;
+  def r32:   FSInst<(outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB),
+		    [(set R32FP:$rT, (fsub R32FP:$rA, R32FP:$rB))]>;
+}
+
+defm FS : SFPSub;
 
 // Floating point reciprocal estimate
 def FREv4f32 :
@@ -3842,6 +3856,12 @@ def : Pat<(fabs (v2f64 VECREG:$rA)),
                       (v2f64 (ANDBIv16i8 (FSMBIv16i8 0xffff), 0x7f)))>;
 
 //===----------------------------------------------------------------------===//
+// Hint for branch instructions:
+//===----------------------------------------------------------------------===//
+
+/* def HBR : SPUInstr<(outs), (ins), "hbr\t" */
+
+//===----------------------------------------------------------------------===//
 // Execution, Load NOP (execute NOPs belong in even pipeline, load NOPs belong
 // in the odd pipeline)
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/CellSPU/SPUOperands.td b/lib/Target/CellSPU/SPUOperands.td
index 4398227..d788f83 100644
--- a/lib/Target/CellSPU/SPUOperands.td
+++ b/lib/Target/CellSPU/SPUOperands.td
@@ -575,7 +575,7 @@ def calltarget : Operand<iPTR> {
   let MIOperandInfo = (ops u18imm:$calldest);
 }
 
-// Relative call target
+// PC relative call target
 def relcalltarget : Operand<iPTR> {
   let PrintMethod = "printPCRelativeOperand";
   let MIOperandInfo = (ops s16imm:$calldest);
@@ -586,6 +586,11 @@ def brtarget : Operand<OtherVT> {
   let PrintMethod = "printPCRelativeOperand";
 }
 
+// Hint for branch target
+def hbrtarget : Operand<OtherVT> {
+  let PrintMethod = "printHBROperand";
+}
+
 // Indirect call target
 def indcalltarget : Operand<iPTR> {
   let PrintMethod = "printCallOperand";
diff --git a/lib/Target/CellSPU/SPUTargetMachine.cpp b/lib/Target/CellSPU/SPUTargetMachine.cpp
index 48b8583..217da2c 100644
--- a/lib/Target/CellSPU/SPUTargetMachine.cpp
+++ b/lib/Target/CellSPU/SPUTargetMachine.cpp
@@ -18,6 +18,8 @@
 #include "llvm/Module.h"
 #include "llvm/PassManager.h"
 #include "llvm/Target/TargetMachineRegistry.h"
+#include "llvm/CodeGen/RegAllocRegistry.h"
+#include "llvm/CodeGen/SchedulerRegistry.h"
 
 using namespace llvm;
 
diff --git a/lib/Target/CellSPU/SPUTargetMachine.h b/lib/Target/CellSPU/SPUTargetMachine.h
index fa72a0a..1396ff2 100644
--- a/lib/Target/CellSPU/SPUTargetMachine.h
+++ b/lib/Target/CellSPU/SPUTargetMachine.h
@@ -83,8 +83,8 @@ public:
   }
   
   // Pass Pipeline Configuration
-  virtual bool addInstSelector(PassManagerBase &PM, bool Fast);
-  virtual bool addAssemblyEmitter(PassManagerBase &PM, bool Fast, 
+  virtual bool addInstSelector(PassManagerBase &PM, bool /*Fast*/);
+  virtual bool addAssemblyEmitter(PassManagerBase &PM, bool /*Fast*/, 
                                   raw_ostream &Out);
 };