Vendor import of llvm release_32 branch r168974 (effectively, 3.2 RC2):

http://llvm.org/svn/llvm-project/llvm/branches/release_32@168974

23 files changed, 1228 insertions, 248 deletions

diff --git a/lib/Target/Hexagon/CMakeLists.txt b/lib/Target/Hexagon/CMakeLists.txt
index 1f2d8ac..306084b 100644
--- a/lib/Target/Hexagon/CMakeLists.txt
+++ b/lib/Target/Hexagon/CMakeLists.txt
@@ -16,6 +16,7 @@ add_llvm_target(HexagonCodeGen
   HexagonExpandPredSpillCode.cpp
   HexagonFrameLowering.cpp
   HexagonHardwareLoops.cpp
+  HexagonMachineScheduler.cpp
   HexagonMCInstLower.cpp
   HexagonInstrInfo.cpp
   HexagonISelDAGToDAG.cpp
diff --git a/lib/Target/Hexagon/HexagonAsmPrinter.cpp b/lib/Target/Hexagon/HexagonAsmPrinter.cpp
index 5fa4740..c15bce6 100644
--- a/lib/Target/Hexagon/HexagonAsmPrinter.cpp
+++ b/lib/Target/Hexagon/HexagonAsmPrinter.cpp
@@ -46,7 +46,7 @@
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Target/Mangler.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLoweringObjectFile.h"
 #include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Target/TargetInstrInfo.h"
diff --git a/lib/Target/Hexagon/HexagonCallingConvLower.cpp b/lib/Target/Hexagon/HexagonCallingConvLower.cpp
index ba8e679..73f9d9a 100644
--- a/lib/Target/Hexagon/HexagonCallingConvLower.cpp
+++ b/lib/Target/Hexagon/HexagonCallingConvLower.cpp
@@ -16,7 +16,7 @@
 #include "HexagonCallingConvLower.h"
 #include "Hexagon.h"
 #include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
diff --git a/lib/Target/Hexagon/HexagonISelLowering.cpp b/lib/Target/Hexagon/HexagonISelLowering.cpp
index 703a128..1c891f1 100644
--- a/lib/Target/Hexagon/HexagonISelLowering.cpp
+++ b/lib/Target/Hexagon/HexagonISelLowering.cpp
@@ -1350,6 +1350,8 @@ HexagonTargetLowering::HexagonTargetLowering(HexagonTargetMachine
     } else {
       setOperationAction(ISD::BR_JT, MVT::Other, Expand);
     }
+    // Increase jump tables cutover to 5, was 4.
+    setMinimumJumpTableEntries(5);
 
     setOperationAction(ISD::BR_CC, MVT::i32, Expand);
 
diff --git a/lib/Target/Hexagon/HexagonInstrFormats.td b/lib/Target/Hexagon/HexagonInstrFormats.td
index e472d49..a64c7a1 100644
--- a/lib/Target/Hexagon/HexagonInstrFormats.td
+++ b/lib/Target/Hexagon/HexagonInstrFormats.td
@@ -56,6 +56,16 @@ class InstHexagon<dag outs, dag ins, string asmstr, list<dag> pattern,
   bits<1> isPredicated = 0;
   let TSFlags{6} = isPredicated;
 
+  // Dot new value store instructions.
+  bits<1> isNVStore = 0;
+  let TSFlags{8} = isNVStore;
+
+  // Fields used for relation models.
+  string BaseOpcode = "";
+  string CextOpcode = "";
+  string PredSense = "";
+  string PNewValue = "";
+  string InputType = "";    // Input is "imm" or "reg" type.
   // *** The code above must match HexagonBaseInfo.h ***
 }
 
diff --git a/lib/Target/Hexagon/HexagonInstrInfo.cpp b/lib/Target/Hexagon/HexagonInstrInfo.cpp
index c8f933d..8435440 100644
--- a/lib/Target/Hexagon/HexagonInstrInfo.cpp
+++ b/lib/Target/Hexagon/HexagonInstrInfo.cpp
@@ -25,6 +25,7 @@
 #include "llvm/CodeGen/PseudoSourceValue.h"
 #include "llvm/Support/MathExtras.h"
 #define GET_INSTRINFO_CTOR
+#define GET_INSTRMAP_INFO
 #include "HexagonGenInstrInfo.inc"
 #include "HexagonGenDFAPacketizer.inc"
 
@@ -1915,6 +1916,15 @@ unsigned HexagonInstrInfo::getInvertedPredicatedOpcode(const int Opc) const {
 
 int HexagonInstrInfo::
 getMatchingCondBranchOpcode(int Opc, bool invertPredicate) const {
+  enum Hexagon::PredSense inPredSense;
+  inPredSense = invertPredicate ? Hexagon::PredSense_false :
+                                  Hexagon::PredSense_true;
+  int CondOpcode = Hexagon::getPredOpcode(Opc, inPredSense);
+  if (CondOpcode >= 0) // Valid Conditional opcode/instruction
+    return CondOpcode;
+
+  // This switch case will be removed once all the instructions have been
+  // modified to use relation maps.
   switch(Opc) {
   case Hexagon::TFR:
     return !invertPredicate ? Hexagon::TFR_cPt :
@@ -1934,24 +1944,6 @@ getMatchingCondBranchOpcode(int Opc, bool invertPredicate) const {
   case Hexagon::JMP_EQriPt_nv_V4:
     return !invertPredicate ? Hexagon::JMP_EQriPt_nv_V4 :
                               Hexagon::JMP_EQriNotPt_nv_V4;
-  case Hexagon::ADD_ri:
-    return !invertPredicate ? Hexagon::ADD_ri_cPt :
-                              Hexagon::ADD_ri_cNotPt;
-  case Hexagon::ADD_rr:
-    return !invertPredicate ? Hexagon::ADD_rr_cPt :
-                              Hexagon::ADD_rr_cNotPt;
-  case Hexagon::XOR_rr:
-    return !invertPredicate ? Hexagon::XOR_rr_cPt :
-                              Hexagon::XOR_rr_cNotPt;
-  case Hexagon::AND_rr:
-    return !invertPredicate ? Hexagon::AND_rr_cPt :
-                              Hexagon::AND_rr_cNotPt;
-  case Hexagon::OR_rr:
-    return !invertPredicate ? Hexagon::OR_rr_cPt :
-                              Hexagon::OR_rr_cNotPt;
-  case Hexagon::SUB_rr:
-    return !invertPredicate ? Hexagon::SUB_rr_cPt :
-                              Hexagon::SUB_rr_cNotPt;
   case Hexagon::COMBINE_rr:
     return !invertPredicate ? Hexagon::COMBINE_rr_cPt :
                               Hexagon::COMBINE_rr_cNotPt;
diff --git a/lib/Target/Hexagon/HexagonInstrInfo.td b/lib/Target/Hexagon/HexagonInstrInfo.td
index c0c0df6..1d4a706 100644
--- a/lib/Target/Hexagon/HexagonInstrInfo.td
+++ b/lib/Target/Hexagon/HexagonInstrInfo.td
@@ -15,6 +15,18 @@ include "HexagonInstrFormats.td"
 include "HexagonImmediates.td"
 
 //===----------------------------------------------------------------------===//
+// Classes used for relation maps.
+//===----------------------------------------------------------------------===//
+// PredRel - Filter class used to relate non-predicated instructions with their
+// predicated forms.
+class PredRel;
+// PredNewRel - Filter class used to relate predicated instructions with their
+// predicate-new forms.
+class PredNewRel: PredRel;
+// ImmRegRel - Filter class used to relate instructions having reg-reg form
+// with their reg-imm counterparts.
+class ImmRegRel;
+//===----------------------------------------------------------------------===//
 // Hexagon Instruction Predicate Definitions.
 //===----------------------------------------------------------------------===//
 def HasV2T                      : Predicate<"Subtarget.hasV2TOps()">;
@@ -148,37 +160,91 @@ multiclass CMP32_ri_s8<string OpcStr, PatFrag OpNode> {
 }
 
 //===----------------------------------------------------------------------===//
-// ALU32/ALU +
+// ALU32/ALU (Instructions with register-register form)
 //===----------------------------------------------------------------------===//
-// Add.
-let isCommutable = 1, isPredicable = 1 in
-def ADD_rr : ALU32_rr<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2),
-            "$dst = add($src1, $src2)",
-            [(set (i32 IntRegs:$dst), (add (i32 IntRegs:$src1),
-                                           (i32 IntRegs:$src2)))]>;
+multiclass ALU32_Pbase<string mnemonic, bit isNot,
+                       bit isPredNew> {
 
-let isPredicable = 1 in
-def ADD_ri : ALU32_ri<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, s16Imm:$src2),
-            "$dst = add($src1, #$src2)",
-            [(set (i32 IntRegs:$dst), (add (i32 IntRegs:$src1),
-                                           s16ImmPred:$src2))]>;
+  let PNewValue = #!if(isPredNew, "new", "") in
+  def #NAME# : ALU32_rr<(outs IntRegs:$dst),
+            (ins PredRegs:$src1, IntRegs:$src2, IntRegs: $src3),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew,".new) $dst = ",
+            ") $dst = ")#mnemonic#"($src2, $src3)",
+            []>;
+}
 
-// Logical operations.
-let isPredicable = 1 in
-def XOR_rr : ALU32_rr<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2),
-            "$dst = xor($src1, $src2)",
-            [(set (i32 IntRegs:$dst), (xor (i32 IntRegs:$src1),
-                                           (i32 IntRegs:$src2)))]>;
+multiclass ALU32_Pred<string mnemonic, bit PredNot> {
+  let PredSense = #!if(PredNot, "false", "true") in {
+    defm _c#NAME# : ALU32_Pbase<mnemonic, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME# : ALU32_Pbase<mnemonic, PredNot, 1>;
+  }
+}
 
-let isCommutable = 1, isPredicable = 1 in
-def AND_rr : ALU32_rr<(outs IntRegs:$dst),
+let InputType = "reg" in
+multiclass ALU32_base<string mnemonic, string CextOp, SDNode OpNode> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_rr in {
+    let isPredicable = 1 in
+    def #NAME# : ALU32_rr<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2),
-            "$dst = and($src1, $src2)",
-            [(set (i32 IntRegs:$dst), (and (i32 IntRegs:$src1),
-                                           (i32 IntRegs:$src2)))]>;
+            "$dst = "#mnemonic#"($src1, $src2)",
+            [(set (i32 IntRegs:$dst), (OpNode (i32 IntRegs:$src1),
+                                              (i32 IntRegs:$src2)))]>;
+
+    let neverHasSideEffects = 1, isPredicated = 1 in {
+      defm Pt : ALU32_Pred<mnemonic, 0>;
+      defm NotPt : ALU32_Pred<mnemonic, 1>;
+    }
+  }
+}
+
+let isCommutable = 1 in {
+  defm ADD_rr : ALU32_base<"add", "ADD", add>, ImmRegRel, PredNewRel;
+  defm AND_rr : ALU32_base<"and", "AND", and>, ImmRegRel, PredNewRel;
+  defm XOR_rr : ALU32_base<"xor", "XOR", xor>, ImmRegRel, PredNewRel;
+  defm OR_rr  : ALU32_base<"or", "OR", or>, ImmRegRel, PredNewRel;
+}
+
+defm SUB_rr : ALU32_base<"sub", "SUB", sub>, ImmRegRel, PredNewRel;
+
+//===----------------------------------------------------------------------===//
+// ALU32/ALU (ADD with register-immediate form)
+//===----------------------------------------------------------------------===//
+multiclass ALU32ri_Pbase<string mnemonic, bit isNot, bit isPredNew> {
+  let PNewValue = #!if(isPredNew, "new", "") in
+  def #NAME# : ALU32_ri<(outs IntRegs:$dst),
+            (ins PredRegs:$src1, IntRegs:$src2, s8Imm: $src3),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew,".new) $dst = ",
+            ") $dst = ")#mnemonic#"($src2, #$src3)",
+            []>;
+}
+
+multiclass ALU32ri_Pred<string mnemonic, bit PredNot> {
+  let PredSense = #!if(PredNot, "false", "true") in {
+    defm _c#NAME# : ALU32ri_Pbase<mnemonic, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME# : ALU32ri_Pbase<mnemonic, PredNot, 1>;
+  }
+}
+
+let InputType = "imm" in
+multiclass ALU32ri_base<string mnemonic, string CextOp, SDNode OpNode> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_ri in {
+    let isPredicable = 1 in
+    def #NAME# : ALU32_ri<(outs IntRegs:$dst),
+            (ins IntRegs:$src1, s16Imm:$src2),
+            "$dst = "#mnemonic#"($src1, #$src2)",
+            [(set (i32 IntRegs:$dst), (OpNode (i32 IntRegs:$src1),
+                                              (s16ImmPred:$src2)))]>;
+
+    let neverHasSideEffects = 1, isPredicated = 1 in {
+      defm Pt : ALU32ri_Pred<mnemonic, 0>;
+      defm NotPt : ALU32ri_Pred<mnemonic, 1>;
+    }
+  }
+}
+
+defm ADD_ri : ALU32ri_base<"add", "ADD", add>, ImmRegRel, PredNewRel;
 
 def OR_ri : ALU32_ri<(outs IntRegs:$dst),
             (ins IntRegs:$src1, s10Imm:$src2),
@@ -197,13 +263,6 @@ def AND_ri : ALU32_ri<(outs IntRegs:$dst),
             [(set (i32 IntRegs:$dst), (and (i32 IntRegs:$src1),
                                            s10ImmPred:$src2))]>;
 
-let isCommutable = 1, isPredicable = 1 in
-def OR_rr : ALU32_rr<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2),
-            "$dst = or($src1, $src2)",
-            [(set (i32 IntRegs:$dst), (or (i32 IntRegs:$src1),
-                                          (i32 IntRegs:$src2)))]>;
-
 // Negate.
 def NEG : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
           "$dst = neg($src1)",
@@ -214,14 +273,6 @@ def NOP : ALU32_rr<(outs), (ins),
           "nop",
           []>;
 
-// Subtract.
-let isPredicable = 1 in
-def SUB_rr : ALU32_rr<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2),
-            "$dst = sub($src1, $src2)",
-            [(set (i32 IntRegs:$dst), (sub (i32 IntRegs:$src1),
-                                           (i32 IntRegs:$src2)))]>;
-
 // Rd32=sub(#s10,Rs32)
 def SUB_ri : ALU32_ri<(outs IntRegs:$dst),
             (ins s10Imm:$src1, IntRegs:$src2),
@@ -348,56 +399,6 @@ def ZXTH : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
 // ALU32/PRED +
 //===----------------------------------------------------------------------===//
 
-// Conditional add.
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ADD_ri_cPt : ALU32_ri<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, s8Imm:$src3),
-            "if ($src1) $dst = add($src2, #$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ADD_ri_cNotPt : ALU32_ri<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, s8Imm:$src3),
-            "if (!$src1) $dst = add($src2, #$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ADD_ri_cdnPt : ALU32_ri<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, s8Imm:$src3),
-            "if ($src1.new) $dst = add($src2, #$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ADD_ri_cdnNotPt : ALU32_ri<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, s8Imm:$src3),
-            "if (!$src1.new) $dst = add($src2, #$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ADD_rr_cPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if ($src1) $dst = add($src2, $src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ADD_rr_cNotPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if (!$src1) $dst = add($src2, $src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ADD_rr_cdnPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if ($src1.new) $dst = add($src2, $src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ADD_rr_cdnNotPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if (!$src1.new) $dst = add($src2, $src3)",
-            []>;
-
-
 // Conditional combine.
 
 let neverHasSideEffects = 1, isPredicated = 1 in
@@ -424,108 +425,6 @@ def COMBINE_rr_cdnNotPt : ALU32_rr<(outs DoubleRegs:$dst),
             "if (!$src1.new) $dst = combine($src2, $src3)",
             []>;
 
-// Conditional logical operations.
-
-let isPredicated = 1 in
-def XOR_rr_cPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if ($src1) $dst = xor($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def XOR_rr_cNotPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if (!$src1) $dst = xor($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def XOR_rr_cdnPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if ($src1.new) $dst = xor($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def XOR_rr_cdnNotPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if (!$src1.new) $dst = xor($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def AND_rr_cPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if ($src1) $dst = and($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def AND_rr_cNotPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if (!$src1) $dst = and($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def AND_rr_cdnPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if ($src1.new) $dst = and($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def AND_rr_cdnNotPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if (!$src1.new) $dst = and($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def OR_rr_cPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if ($src1) $dst = or($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def OR_rr_cNotPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if (!$src1) $dst = or($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def OR_rr_cdnPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if ($src1.new) $dst = or($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def OR_rr_cdnNotPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if (!$src1.new) $dst = or($src2, $src3)",
-            []>;
-
-
-// Conditional subtract.
-
-let isPredicated = 1 in
-def SUB_rr_cPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if ($src1) $dst = sub($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def SUB_rr_cNotPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if (!$src1) $dst = sub($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def SUB_rr_cdnPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if ($src1.new) $dst = sub($src2, $src3)",
-            []>;
-
-let isPredicated = 1 in
-def SUB_rr_cdnNotPt : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "if (!$src1.new) $dst = sub($src2, $src3)",
-            []>;
-
-
 // Conditional transfer.
 let neverHasSideEffects = 1, isPredicated = 1 in
 def TFR_cPt : ALU32_rr<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2),
@@ -3546,4 +3445,31 @@ include "HexagonInstrInfoV5.td"
 // V5 Instructions -
 //===----------------------------------------------------------------------===//
 
+//===----------------------------------------------------------------------===//
+// Generate mapping table to relate non-predicate instructions with their
+// predicated formats - true and false.
+//
+
+def getPredOpcode : InstrMapping {
+  let FilterClass = "PredRel";
+  // Instructions with the same BaseOpcode and isNVStore values form a row.
+  let RowFields = ["BaseOpcode", "isNVStore", "PNewValue"];
+  // Instructions with the same predicate sense form a column.
+  let ColFields = ["PredSense"];
+  // The key column is the unpredicated instructions.
+  let KeyCol = [""];
+  // Value columns are PredSense=true and PredSense=false
+  let ValueCols = [["true"], ["false"]];
+}
 
+//===----------------------------------------------------------------------===//
+// Generate mapping table to relate predicated instructions with their .new
+// format.
+//
+def getPredNewOpcode : InstrMapping {
+  let FilterClass = "PredNewRel";
+  let RowFields = ["BaseOpcode", "PredSense", "isNVStore"];
+  let ColFields = ["PNewValue"];
+  let KeyCol = [""];
+  let ValueCols = [["new"]];
+}
diff --git a/lib/Target/Hexagon/HexagonMachineScheduler.cpp b/lib/Target/Hexagon/HexagonMachineScheduler.cpp
new file mode 100644
index 0000000..0e9ef48
--- /dev/null
+++ b/lib/Target/Hexagon/HexagonMachineScheduler.cpp
@@ -0,0 +1,681 @@
+//===- HexagonMachineScheduler.cpp - MI Scheduler for Hexagon -------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// MachineScheduler schedules machine instructions after phi elimination. It
+// preserves LiveIntervals so it can be invoked before register allocation.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "misched"
+
+#include "HexagonMachineScheduler.h"
+
+#include <queue>
+
+using namespace llvm;
+
+/// Platform specific modifications to DAG.
+void VLIWMachineScheduler::postprocessDAG() {
+  SUnit* LastSequentialCall = NULL;
+  // Currently we only catch the situation when compare gets scheduled
+  // before preceding call.
+  for (unsigned su = 0, e = SUnits.size(); su != e; ++su) {
+    // Remember the call.
+    if (SUnits[su].getInstr()->isCall())
+      LastSequentialCall = &(SUnits[su]);
+    // Look for a compare that defines a predicate.
+    else if (SUnits[su].getInstr()->isCompare() && LastSequentialCall)
+      SUnits[su].addPred(SDep(LastSequentialCall, SDep::Barrier));
+  }
+}
+
+/// Check if scheduling of this SU is possible
+/// in the current packet.
+/// It is _not_ precise (statefull), it is more like
+/// another heuristic. Many corner cases are figured
+/// empirically.
+bool VLIWResourceModel::isResourceAvailable(SUnit *SU) {
+  if (!SU || !SU->getInstr())
+    return false;
+
+  // First see if the pipeline could receive this instruction
+  // in the current cycle.
+  switch (SU->getInstr()->getOpcode()) {
+  default:
+    if (!ResourcesModel->canReserveResources(SU->getInstr()))
+      return false;
+  case TargetOpcode::EXTRACT_SUBREG:
+  case TargetOpcode::INSERT_SUBREG:
+  case TargetOpcode::SUBREG_TO_REG:
+  case TargetOpcode::REG_SEQUENCE:
+  case TargetOpcode::IMPLICIT_DEF:
+  case TargetOpcode::COPY:
+  case TargetOpcode::INLINEASM:
+    break;
+  }
+
+  // Now see if there are no other dependencies to instructions already
+  // in the packet.
+  for (unsigned i = 0, e = Packet.size(); i != e; ++i) {
+    if (Packet[i]->Succs.size() == 0)
+      continue;
+    for (SUnit::const_succ_iterator I = Packet[i]->Succs.begin(),
+         E = Packet[i]->Succs.end(); I != E; ++I) {
+      // Since we do not add pseudos to packets, might as well
+      // ignore order dependencies.
+      if (I->isCtrl())
+        continue;
+
+      if (I->getSUnit() == SU)
+        return false;
+    }
+  }
+  return true;
+}
+
+/// Keep track of available resources.
+bool VLIWResourceModel::reserveResources(SUnit *SU) {
+  bool startNewCycle = false;
+  // Artificially reset state.
+  if (!SU) {
+    ResourcesModel->clearResources();
+    Packet.clear();
+    TotalPackets++;
+    return false;
+  }
+  // If this SU does not fit in the packet
+  // start a new one.
+  if (!isResourceAvailable(SU)) {
+    ResourcesModel->clearResources();
+    Packet.clear();
+    TotalPackets++;
+    startNewCycle = true;
+  }
+
+  switch (SU->getInstr()->getOpcode()) {
+  default:
+    ResourcesModel->reserveResources(SU->getInstr());
+    break;
+  case TargetOpcode::EXTRACT_SUBREG:
+  case TargetOpcode::INSERT_SUBREG:
+  case TargetOpcode::SUBREG_TO_REG:
+  case TargetOpcode::REG_SEQUENCE:
+  case TargetOpcode::IMPLICIT_DEF:
+  case TargetOpcode::KILL:
+  case TargetOpcode::PROLOG_LABEL:
+  case TargetOpcode::EH_LABEL:
+  case TargetOpcode::COPY:
+  case TargetOpcode::INLINEASM:
+    break;
+  }
+  Packet.push_back(SU);
+
+#ifndef NDEBUG
+  DEBUG(dbgs() << "Packet[" << TotalPackets << "]:\n");
+  for (unsigned i = 0, e = Packet.size(); i != e; ++i) {
+    DEBUG(dbgs() << "\t[" << i << "] SU(");
+    DEBUG(dbgs() << Packet[i]->NodeNum << ")\t");
+    DEBUG(Packet[i]->getInstr()->dump());
+  }
+#endif
+
+  // If packet is now full, reset the state so in the next cycle
+  // we start fresh.
+  if (Packet.size() >= SchedModel->getIssueWidth()) {
+    ResourcesModel->clearResources();
+    Packet.clear();
+    TotalPackets++;
+    startNewCycle = true;
+  }
+
+  return startNewCycle;
+}
+
+/// schedule - Called back from MachineScheduler::runOnMachineFunction
+/// after setting up the current scheduling region. [RegionBegin, RegionEnd)
+/// only includes instructions that have DAG nodes, not scheduling boundaries.
+void VLIWMachineScheduler::schedule() {
+  DEBUG(dbgs()
+        << "********** MI Converging Scheduling VLIW BB#" << BB->getNumber()
+        << " " << BB->getName()
+        << " in_func " << BB->getParent()->getFunction()->getName()
+        << " at loop depth "  << MLI.getLoopDepth(BB)
+        << " \n");
+
+  buildDAGWithRegPressure();
+
+  // Postprocess the DAG to add platform specific artificial dependencies.
+  postprocessDAG();
+
+  // To view Height/Depth correctly, they should be accessed at least once.
+  DEBUG(unsigned maxH = 0;
+        for (unsigned su = 0, e = SUnits.size(); su != e; ++su)
+          if (SUnits[su].getHeight() > maxH)
+            maxH = SUnits[su].getHeight();
+        dbgs() << "Max Height " << maxH << "\n";);
+  DEBUG(unsigned maxD = 0;
+        for (unsigned su = 0, e = SUnits.size(); su != e; ++su)
+          if (SUnits[su].getDepth() > maxD)
+            maxD = SUnits[su].getDepth();
+        dbgs() << "Max Depth " << maxD << "\n";);
+  DEBUG(for (unsigned su = 0, e = SUnits.size(); su != e; ++su)
+          SUnits[su].dumpAll(this));
+
+  initQueues();
+
+  bool IsTopNode = false;
+  while (SUnit *SU = SchedImpl->pickNode(IsTopNode)) {
+    if (!checkSchedLimit())
+      break;
+
+    scheduleMI(SU, IsTopNode);
+
+    updateQueues(SU, IsTopNode);
+  }
+  assert(CurrentTop == CurrentBottom && "Nonempty unscheduled zone.");
+
+  placeDebugValues();
+}
+
+void ConvergingVLIWScheduler::initialize(ScheduleDAGMI *dag) {
+  DAG = static_cast<VLIWMachineScheduler*>(dag);
+  SchedModel = DAG->getSchedModel();
+  TRI = DAG->TRI;
+  Top.init(DAG, SchedModel);
+  Bot.init(DAG, SchedModel);
+
+  // Initialize the HazardRecognizers. If itineraries don't exist, are empty, or
+  // are disabled, then these HazardRecs will be disabled.
+  const InstrItineraryData *Itin = DAG->getSchedModel()->getInstrItineraries();
+  const TargetMachine &TM = DAG->MF.getTarget();
+  Top.HazardRec = TM.getInstrInfo()->CreateTargetMIHazardRecognizer(Itin, DAG);
+  Bot.HazardRec = TM.getInstrInfo()->CreateTargetMIHazardRecognizer(Itin, DAG);
+
+  Top.ResourceModel = new VLIWResourceModel(TM, DAG->getSchedModel());
+  Bot.ResourceModel = new VLIWResourceModel(TM, DAG->getSchedModel());
+
+  assert((!llvm::ForceTopDown || !llvm::ForceBottomUp) &&
+         "-misched-topdown incompatible with -misched-bottomup");
+}
+
+void ConvergingVLIWScheduler::releaseTopNode(SUnit *SU) {
+  if (SU->isScheduled)
+    return;
+
+  for (SUnit::succ_iterator I = SU->Preds.begin(), E = SU->Preds.end();
+       I != E; ++I) {
+    unsigned PredReadyCycle = I->getSUnit()->TopReadyCycle;
+    unsigned MinLatency = I->getMinLatency();
+#ifndef NDEBUG
+    Top.MaxMinLatency = std::max(MinLatency, Top.MaxMinLatency);
+#endif
+    if (SU->TopReadyCycle < PredReadyCycle + MinLatency)
+      SU->TopReadyCycle = PredReadyCycle + MinLatency;
+  }
+  Top.releaseNode(SU, SU->TopReadyCycle);
+}
+
+void ConvergingVLIWScheduler::releaseBottomNode(SUnit *SU) {
+  if (SU->isScheduled)
+    return;
+
+  assert(SU->getInstr() && "Scheduled SUnit must have instr");
+
+  for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
+       I != E; ++I) {
+    unsigned SuccReadyCycle = I->getSUnit()->BotReadyCycle;
+    unsigned MinLatency = I->getMinLatency();
+#ifndef NDEBUG
+    Bot.MaxMinLatency = std::max(MinLatency, Bot.MaxMinLatency);
+#endif
+    if (SU->BotReadyCycle < SuccReadyCycle + MinLatency)
+      SU->BotReadyCycle = SuccReadyCycle + MinLatency;
+  }
+  Bot.releaseNode(SU, SU->BotReadyCycle);
+}
+
+/// Does this SU have a hazard within the current instruction group.
+///
+/// The scheduler supports two modes of hazard recognition. The first is the
+/// ScheduleHazardRecognizer API. It is a fully general hazard recognizer that
+/// supports highly complicated in-order reservation tables
+/// (ScoreboardHazardRecognizer) and arbitrary target-specific logic.
+///
+/// The second is a streamlined mechanism that checks for hazards based on
+/// simple counters that the scheduler itself maintains. It explicitly checks
+/// for instruction dispatch limitations, including the number of micro-ops that
+/// can dispatch per cycle.
+///
+/// TODO: Also check whether the SU must start a new group.
+bool ConvergingVLIWScheduler::SchedBoundary::checkHazard(SUnit *SU) {
+  if (HazardRec->isEnabled())
+    return HazardRec->getHazardType(SU) != ScheduleHazardRecognizer::NoHazard;
+
+  unsigned uops = SchedModel->getNumMicroOps(SU->getInstr());
+  if (IssueCount + uops > SchedModel->getIssueWidth())
+    return true;
+
+  return false;
+}
+
+void ConvergingVLIWScheduler::SchedBoundary::releaseNode(SUnit *SU,
+                                                     unsigned ReadyCycle) {
+  if (ReadyCycle < MinReadyCycle)
+    MinReadyCycle = ReadyCycle;
+
+  // Check for interlocks first. For the purpose of other heuristics, an
+  // instruction that cannot issue appears as if it's not in the ReadyQueue.
+  if (ReadyCycle > CurrCycle || checkHazard(SU))
+
+    Pending.push(SU);
+  else
+    Available.push(SU);
+}
+
+/// Move the boundary of scheduled code by one cycle.
+void ConvergingVLIWScheduler::SchedBoundary::bumpCycle() {
+  unsigned Width = SchedModel->getIssueWidth();
+  IssueCount = (IssueCount <= Width) ? 0 : IssueCount - Width;
+
+  assert(MinReadyCycle < UINT_MAX && "MinReadyCycle uninitialized");
+  unsigned NextCycle = std::max(CurrCycle + 1, MinReadyCycle);
+
+  if (!HazardRec->isEnabled()) {
+    // Bypass HazardRec virtual calls.
+    CurrCycle = NextCycle;
+  } else {
+    // Bypass getHazardType calls in case of long latency.
+    for (; CurrCycle != NextCycle; ++CurrCycle) {
+      if (isTop())
+        HazardRec->AdvanceCycle();
+      else
+        HazardRec->RecedeCycle();
+    }
+  }
+  CheckPending = true;
+
+  DEBUG(dbgs() << "*** " << Available.getName() << " cycle "
+        << CurrCycle << '\n');
+}
+
+/// Move the boundary of scheduled code by one SUnit.
+void ConvergingVLIWScheduler::SchedBoundary::bumpNode(SUnit *SU) {
+  bool startNewCycle = false;
+
+  // Update the reservation table.
+  if (HazardRec->isEnabled()) {
+    if (!isTop() && SU->isCall) {
+      // Calls are scheduled with their preceding instructions. For bottom-up
+      // scheduling, clear the pipeline state before emitting.
+      HazardRec->Reset();
+    }
+    HazardRec->EmitInstruction(SU);
+  }
+
+  // Update DFA model.
+  startNewCycle = ResourceModel->reserveResources(SU);
+
+  // Check the instruction group dispatch limit.
+  // TODO: Check if this SU must end a dispatch group.
+  IssueCount += SchedModel->getNumMicroOps(SU->getInstr());
+  if (startNewCycle) {
+    DEBUG(dbgs() << "*** Max instrs at cycle " << CurrCycle << '\n');
+    bumpCycle();
+  }
+  else
+    DEBUG(dbgs() << "*** IssueCount " << IssueCount
+          << " at cycle " << CurrCycle << '\n');
+}
+
+/// Release pending ready nodes in to the available queue. This makes them
+/// visible to heuristics.
+void ConvergingVLIWScheduler::SchedBoundary::releasePending() {
+  // If the available queue is empty, it is safe to reset MinReadyCycle.
+  if (Available.empty())
+    MinReadyCycle = UINT_MAX;
+
+  // Check to see if any of the pending instructions are ready to issue.  If
+  // so, add them to the available queue.
+  for (unsigned i = 0, e = Pending.size(); i != e; ++i) {
+    SUnit *SU = *(Pending.begin()+i);
+    unsigned ReadyCycle = isTop() ? SU->TopReadyCycle : SU->BotReadyCycle;
+
+    if (ReadyCycle < MinReadyCycle)
+      MinReadyCycle = ReadyCycle;
+
+    if (ReadyCycle > CurrCycle)
+      continue;
+
+    if (checkHazard(SU))
+      continue;
+
+    Available.push(SU);
+    Pending.remove(Pending.begin()+i);
+    --i; --e;
+  }
+  CheckPending = false;
+}
+
+/// Remove SU from the ready set for this boundary.
+void ConvergingVLIWScheduler::SchedBoundary::removeReady(SUnit *SU) {
+  if (Available.isInQueue(SU))
+    Available.remove(Available.find(SU));
+  else {
+    assert(Pending.isInQueue(SU) && "bad ready count");
+    Pending.remove(Pending.find(SU));
+  }
+}
+
+/// If this queue only has one ready candidate, return it. As a side effect,
+/// advance the cycle until at least one node is ready. If multiple instructions
+/// are ready, return NULL.
+SUnit *ConvergingVLIWScheduler::SchedBoundary::pickOnlyChoice() {
+  if (CheckPending)
+    releasePending();
+
+  for (unsigned i = 0; Available.empty(); ++i) {
+    assert(i <= (HazardRec->getMaxLookAhead() + MaxMinLatency) &&
+           "permanent hazard"); (void)i;
+    ResourceModel->reserveResources(0);
+    bumpCycle();
+    releasePending();
+  }
+  if (Available.size() == 1)
+    return *Available.begin();
+  return NULL;
+}
+
+#ifndef NDEBUG
+void ConvergingVLIWScheduler::traceCandidate(const char *Label,
+                                             const ReadyQueue &Q,
+                                             SUnit *SU, PressureElement P) {
+  dbgs() << Label << " " << Q.getName() << " ";
+  if (P.isValid())
+    dbgs() << TRI->getRegPressureSetName(P.PSetID) << ":" << P.UnitIncrease
+           << " ";
+  else
+    dbgs() << "     ";
+  SU->dump(DAG);
+}
+#endif
+
+/// getSingleUnscheduledPred - If there is exactly one unscheduled predecessor
+/// of SU, return it, otherwise return null.
+static SUnit *getSingleUnscheduledPred(SUnit *SU) {
+  SUnit *OnlyAvailablePred = 0;
+  for (SUnit::const_pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
+       I != E; ++I) {
+    SUnit &Pred = *I->getSUnit();
+    if (!Pred.isScheduled) {
+      // We found an available, but not scheduled, predecessor.  If it's the
+      // only one we have found, keep track of it... otherwise give up.
+      if (OnlyAvailablePred && OnlyAvailablePred != &Pred)
+        return 0;
+      OnlyAvailablePred = &Pred;
+    }
+  }
+  return OnlyAvailablePred;
+}
+
+/// getSingleUnscheduledSucc - If there is exactly one unscheduled successor
+/// of SU, return it, otherwise return null.
+static SUnit *getSingleUnscheduledSucc(SUnit *SU) {
+  SUnit *OnlyAvailableSucc = 0;
+  for (SUnit::const_succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
+       I != E; ++I) {
+    SUnit &Succ = *I->getSUnit();
+    if (!Succ.isScheduled) {
+      // We found an available, but not scheduled, successor.  If it's the
+      // only one we have found, keep track of it... otherwise give up.
+      if (OnlyAvailableSucc && OnlyAvailableSucc != &Succ)
+        return 0;
+      OnlyAvailableSucc = &Succ;
+    }
+  }
+  return OnlyAvailableSucc;
+}
+
+// Constants used to denote relative importance of
+// heuristic components for cost computation.
+static const unsigned PriorityOne = 200;
+static const unsigned PriorityTwo = 100;
+static const unsigned PriorityThree = 50;
+static const unsigned PriorityFour = 20;
+static const unsigned ScaleTwo = 10;
+static const unsigned FactorOne = 2;
+
+/// Single point to compute overall scheduling cost.
+/// TODO: More heuristics will be used soon.
+int ConvergingVLIWScheduler::SchedulingCost(ReadyQueue &Q, SUnit *SU,
+                                            SchedCandidate &Candidate,
+                                            RegPressureDelta &Delta,
+                                            bool verbose) {
+  // Initial trivial priority.
+  int ResCount = 1;
+
+  // Do not waste time on a node that is already scheduled.
+  if (!SU || SU->isScheduled)
+    return ResCount;
+
+  // Forced priority is high.
+  if (SU->isScheduleHigh)
+    ResCount += PriorityOne;
+
+  // Critical path first.
+  if (Q.getID() == TopQID) {
+    ResCount += (SU->getHeight() * ScaleTwo);
+
+    // If resources are available for it, multiply the
+    // chance of scheduling.
+    if (Top.ResourceModel->isResourceAvailable(SU))
+      ResCount <<= FactorOne;
+  } else {
+    ResCount += (SU->getDepth() * ScaleTwo);
+
+    // If resources are available for it, multiply the
+    // chance of scheduling.
+    if (Bot.ResourceModel->isResourceAvailable(SU))
+      ResCount <<= FactorOne;
+  }
+
+  unsigned NumNodesBlocking = 0;
+  if (Q.getID() == TopQID) {
+    // How many SUs does it block from scheduling?
+    // Look at all of the successors of this node.
+    // Count the number of nodes that
+    // this node is the sole unscheduled node for.
+    for (SUnit::const_succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
+         I != E; ++I)
+      if (getSingleUnscheduledPred(I->getSUnit()) == SU)
+        ++NumNodesBlocking;
+  } else {
+    // How many unscheduled predecessors block this node?
+    for (SUnit::const_pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
+         I != E; ++I)
+      if (getSingleUnscheduledSucc(I->getSUnit()) == SU)
+        ++NumNodesBlocking;
+  }
+  ResCount += (NumNodesBlocking * ScaleTwo);
+
+  // Factor in reg pressure as a heuristic.
+  ResCount -= (Delta.Excess.UnitIncrease*PriorityThree);
+  ResCount -= (Delta.CriticalMax.UnitIncrease*PriorityThree);
+
+  DEBUG(if (verbose) dbgs() << " Total(" << ResCount << ")");
+
+  return ResCount;
+}
+
+/// Pick the best candidate from the top queue.
+///
+/// TODO: getMaxPressureDelta results can be mostly cached for each SUnit during
+/// DAG building. To adjust for the current scheduling location we need to
+/// maintain the number of vreg uses remaining to be top-scheduled.
+ConvergingVLIWScheduler::CandResult ConvergingVLIWScheduler::
+pickNodeFromQueue(ReadyQueue &Q, const RegPressureTracker &RPTracker,
+                  SchedCandidate &Candidate) {
+  DEBUG(Q.dump());
+
+  // getMaxPressureDelta temporarily modifies the tracker.
+  RegPressureTracker &TempTracker = const_cast<RegPressureTracker&>(RPTracker);
+
+  // BestSU remains NULL if no top candidates beat the best existing candidate.
+  CandResult FoundCandidate = NoCand;
+  for (ReadyQueue::iterator I = Q.begin(), E = Q.end(); I != E; ++I) {
+    RegPressureDelta RPDelta;
+    TempTracker.getMaxPressureDelta((*I)->getInstr(), RPDelta,
+                                    DAG->getRegionCriticalPSets(),
+                                    DAG->getRegPressure().MaxSetPressure);
+
+    int CurrentCost = SchedulingCost(Q, *I, Candidate, RPDelta, false);
+
+    // Initialize the candidate if needed.
+    if (!Candidate.SU) {
+      Candidate.SU = *I;
+      Candidate.RPDelta = RPDelta;
+      Candidate.SCost = CurrentCost;
+      FoundCandidate = NodeOrder;
+      continue;
+    }
+
+    // Best cost.
+    if (CurrentCost > Candidate.SCost) {
+      DEBUG(traceCandidate("CCAND", Q, *I));
+      Candidate.SU = *I;
+      Candidate.RPDelta = RPDelta;
+      Candidate.SCost = CurrentCost;
+      FoundCandidate = BestCost;
+      continue;
+    }
+
+    // Fall through to original instruction order.
+    // Only consider node order if Candidate was chosen from this Q.
+    if (FoundCandidate == NoCand)
+      continue;
+  }
+  return FoundCandidate;
+}
+
+/// Pick the best candidate node from either the top or bottom queue.
+SUnit *ConvergingVLIWScheduler::pickNodeBidrectional(bool &IsTopNode) {
+  // Schedule as far as possible in the direction of no choice. This is most
+  // efficient, but also provides the best heuristics for CriticalPSets.
+  if (SUnit *SU = Bot.pickOnlyChoice()) {
+    IsTopNode = false;
+    return SU;
+  }
+  if (SUnit *SU = Top.pickOnlyChoice()) {
+    IsTopNode = true;
+    return SU;
+  }
+  SchedCandidate BotCand;
+  // Prefer bottom scheduling when heuristics are silent.
+  CandResult BotResult = pickNodeFromQueue(Bot.Available,
+                                           DAG->getBotRPTracker(), BotCand);
+  assert(BotResult != NoCand && "failed to find the first candidate");
+
+  // If either Q has a single candidate that provides the least increase in
+  // Excess pressure, we can immediately schedule from that Q.
+  //
+  // RegionCriticalPSets summarizes the pressure within the scheduled region and
+  // affects picking from either Q. If scheduling in one direction must
+  // increase pressure for one of the excess PSets, then schedule in that
+  // direction first to provide more freedom in the other direction.
+  if (BotResult == SingleExcess || BotResult == SingleCritical) {
+    IsTopNode = false;
+    return BotCand.SU;
+  }
+  // Check if the top Q has a better candidate.
+  SchedCandidate TopCand;
+  CandResult TopResult = pickNodeFromQueue(Top.Available,
+                                           DAG->getTopRPTracker(), TopCand);
+  assert(TopResult != NoCand && "failed to find the first candidate");
+
+  if (TopResult == SingleExcess || TopResult == SingleCritical) {
+    IsTopNode = true;
+    return TopCand.SU;
+  }
+  // If either Q has a single candidate that minimizes pressure above the
+  // original region's pressure pick it.
+  if (BotResult == SingleMax) {
+    IsTopNode = false;
+    return BotCand.SU;
+  }
+  if (TopResult == SingleMax) {
+    IsTopNode = true;
+    return TopCand.SU;
+  }
+  if (TopCand.SCost > BotCand.SCost) {
+    IsTopNode = true;
+    return TopCand.SU;
+  }
+  // Otherwise prefer the bottom candidate in node order.
+  IsTopNode = false;
+  return BotCand.SU;
+}
+
+/// Pick the best node to balance the schedule. Implements MachineSchedStrategy.
+SUnit *ConvergingVLIWScheduler::pickNode(bool &IsTopNode) {
+  if (DAG->top() == DAG->bottom()) {
+    assert(Top.Available.empty() && Top.Pending.empty() &&
+           Bot.Available.empty() && Bot.Pending.empty() && "ReadyQ garbage");
+    return NULL;
+  }
+  SUnit *SU;
+  if (llvm::ForceTopDown) {
+    SU = Top.pickOnlyChoice();
+    if (!SU) {
+      SchedCandidate TopCand;
+      CandResult TopResult =
+        pickNodeFromQueue(Top.Available, DAG->getTopRPTracker(), TopCand);
+      assert(TopResult != NoCand && "failed to find the first candidate");
+      (void)TopResult;
+      SU = TopCand.SU;
+    }
+    IsTopNode = true;
+  } else if (llvm::ForceBottomUp) {
+    SU = Bot.pickOnlyChoice();
+    if (!SU) {
+      SchedCandidate BotCand;
+      CandResult BotResult =
+        pickNodeFromQueue(Bot.Available, DAG->getBotRPTracker(), BotCand);
+      assert(BotResult != NoCand && "failed to find the first candidate");
+      (void)BotResult;
+      SU = BotCand.SU;
+    }
+    IsTopNode = false;
+  } else {
+    SU = pickNodeBidrectional(IsTopNode);
+  }
+  if (SU->isTopReady())
+    Top.removeReady(SU);
+  if (SU->isBottomReady())
+    Bot.removeReady(SU);
+
+  DEBUG(dbgs() << "*** " << (IsTopNode ? "Top" : "Bottom")
+        << " Scheduling Instruction in cycle "
+        << (IsTopNode ? Top.CurrCycle : Bot.CurrCycle) << '\n';
+        SU->dump(DAG));
+  return SU;
+}
+
+/// Update the scheduler's state after scheduling a node. This is the same node
+/// that was just returned by pickNode(). However, VLIWMachineScheduler needs
+/// to update it's state based on the current cycle before MachineSchedStrategy
+/// does.
+void ConvergingVLIWScheduler::schedNode(SUnit *SU, bool IsTopNode) {
+  if (IsTopNode) {
+    SU->TopReadyCycle = Top.CurrCycle;
+    Top.bumpNode(SU);
+  } else {
+    SU->BotReadyCycle = Bot.CurrCycle;
+    Bot.bumpNode(SU);
+  }
+}
+
diff --git a/lib/Target/Hexagon/HexagonMachineScheduler.h b/lib/Target/Hexagon/HexagonMachineScheduler.h
new file mode 100644
index 0000000..fe0242a
--- /dev/null
+++ b/lib/Target/Hexagon/HexagonMachineScheduler.h
@@ -0,0 +1,244 @@
+//===-- HexagonMachineScheduler.h - Custom Hexagon MI scheduler.      ----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Custom Hexagon MI scheduler.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef HEXAGONASMPRINTER_H
+#define HEXAGONASMPRINTER_H
+
+#include "llvm/CodeGen/LiveIntervalAnalysis.h"
+#include "llvm/CodeGen/MachineScheduler.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/CodeGen/RegisterClassInfo.h"
+#include "llvm/CodeGen/RegisterPressure.h"
+#include "llvm/CodeGen/ResourcePriorityQueue.h"
+#include "llvm/CodeGen/ScheduleDAGInstrs.h"
+#include "llvm/CodeGen/ScheduleHazardRecognizer.h"
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/OwningPtr.h"
+#include "llvm/ADT/PriorityQueue.h"
+
+using namespace llvm;
+
+namespace llvm {
+//===----------------------------------------------------------------------===//
+// ConvergingVLIWScheduler - Implementation of the standard
+// MachineSchedStrategy.
+//===----------------------------------------------------------------------===//
+
+class VLIWResourceModel {
+  /// ResourcesModel - Represents VLIW state.
+  /// Not limited to VLIW targets per say, but assumes
+  /// definition of DFA by a target.
+  DFAPacketizer *ResourcesModel;
+
+  const TargetSchedModel *SchedModel;
+
+  /// Local packet/bundle model. Purely
+  /// internal to the MI schedulre at the time.
+  std::vector<SUnit*> Packet;
+
+  /// Total packets created.
+  unsigned TotalPackets;
+
+public:
+VLIWResourceModel(const TargetMachine &TM, const TargetSchedModel *SM) :
+    SchedModel(SM), TotalPackets(0) {
+    ResourcesModel = TM.getInstrInfo()->CreateTargetScheduleState(&TM,NULL);
+
+    // This hard requirement could be relaxed,
+    // but for now do not let it proceed.
+    assert(ResourcesModel && "Unimplemented CreateTargetScheduleState.");
+
+    Packet.resize(SchedModel->getIssueWidth());
+    Packet.clear();
+    ResourcesModel->clearResources();
+  }
+
+  ~VLIWResourceModel() {
+    delete ResourcesModel;
+  }
+
+  void resetPacketState() {
+    Packet.clear();
+  }
+
+  void resetDFA() {
+    ResourcesModel->clearResources();
+  }
+
+  void reset() {
+    Packet.clear();
+    ResourcesModel->clearResources();
+  }
+
+  bool isResourceAvailable(SUnit *SU);
+  bool reserveResources(SUnit *SU);
+  unsigned getTotalPackets() const { return TotalPackets; }
+};
+
+/// Extend the standard ScheduleDAGMI to provide more context and override the
+/// top-level schedule() driver.
+class VLIWMachineScheduler : public ScheduleDAGMI {
+public:
+  VLIWMachineScheduler(MachineSchedContext *C, MachineSchedStrategy *S):
+    ScheduleDAGMI(C, S) {}
+
+  /// Schedule - This is called back from ScheduleDAGInstrs::Run() when it's
+  /// time to do some work.
+  virtual void schedule();
+  /// Perform platform specific DAG postprocessing.
+  void postprocessDAG();
+};
+
+/// ConvergingVLIWScheduler shrinks the unscheduled zone using heuristics
+/// to balance the schedule.
+class ConvergingVLIWScheduler : public MachineSchedStrategy {
+
+  /// Store the state used by ConvergingVLIWScheduler heuristics, required
+  ///  for the lifetime of one invocation of pickNode().
+  struct SchedCandidate {
+    // The best SUnit candidate.
+    SUnit *SU;
+
+    // Register pressure values for the best candidate.
+    RegPressureDelta RPDelta;
+
+    // Best scheduling cost.
+    int SCost;
+
+    SchedCandidate(): SU(NULL), SCost(0) {}
+  };
+  /// Represent the type of SchedCandidate found within a single queue.
+  enum CandResult {
+    NoCand, NodeOrder, SingleExcess, SingleCritical, SingleMax, MultiPressure,
+    BestCost};
+
+  /// Each Scheduling boundary is associated with ready queues. It tracks the
+  /// current cycle in whichever direction at has moved, and maintains the state
+  /// of "hazards" and other interlocks at the current cycle.
+  struct SchedBoundary {
+    VLIWMachineScheduler *DAG;
+    const TargetSchedModel *SchedModel;
+
+    ReadyQueue Available;
+    ReadyQueue Pending;
+    bool CheckPending;
+
+    ScheduleHazardRecognizer *HazardRec;
+    VLIWResourceModel *ResourceModel;
+
+    unsigned CurrCycle;
+    unsigned IssueCount;
+
+    /// MinReadyCycle - Cycle of the soonest available instruction.
+    unsigned MinReadyCycle;
+
+    // Remember the greatest min operand latency.
+    unsigned MaxMinLatency;
+
+    /// Pending queues extend the ready queues with the same ID and the
+    /// PendingFlag set.
+    SchedBoundary(unsigned ID, const Twine &Name):
+      DAG(0), SchedModel(0), Available(ID, Name+".A"),
+      Pending(ID << ConvergingVLIWScheduler::LogMaxQID, Name+".P"),
+      CheckPending(false), HazardRec(0), ResourceModel(0),
+      CurrCycle(0), IssueCount(0),
+      MinReadyCycle(UINT_MAX), MaxMinLatency(0) {}
+
+    ~SchedBoundary() {
+      delete ResourceModel;
+      delete HazardRec;
+    }
+
+    void init(VLIWMachineScheduler *dag, const TargetSchedModel *smodel) {
+      DAG = dag;
+      SchedModel = smodel;
+    }
+
+    bool isTop() const {
+      return Available.getID() == ConvergingVLIWScheduler::TopQID;
+    }
+
+    bool checkHazard(SUnit *SU);
+
+    void releaseNode(SUnit *SU, unsigned ReadyCycle);
+
+    void bumpCycle();
+
+    void bumpNode(SUnit *SU);
+
+    void releasePending();
+
+    void removeReady(SUnit *SU);
+
+    SUnit *pickOnlyChoice();
+  };
+
+  VLIWMachineScheduler *DAG;
+  const TargetSchedModel *SchedModel;
+  const TargetRegisterInfo *TRI;
+
+  // State of the top and bottom scheduled instruction boundaries.
+  SchedBoundary Top;
+  SchedBoundary Bot;
+
+public:
+  /// SUnit::NodeQueueId: 0 (none), 1 (top), 2 (bot), 3 (both)
+  enum {
+    TopQID = 1,
+    BotQID = 2,
+    LogMaxQID = 2
+  };
+
+  ConvergingVLIWScheduler():
+    DAG(0), SchedModel(0), TRI(0), Top(TopQID, "TopQ"), Bot(BotQID, "BotQ") {}
+
+  virtual void initialize(ScheduleDAGMI *dag);
+
+  virtual SUnit *pickNode(bool &IsTopNode);
+
+  virtual void schedNode(SUnit *SU, bool IsTopNode);
+
+  virtual void releaseTopNode(SUnit *SU);
+
+  virtual void releaseBottomNode(SUnit *SU);
+
+  unsigned ReportPackets() {
+    return Top.ResourceModel->getTotalPackets() +
+           Bot.ResourceModel->getTotalPackets();
+  }
+
+protected:
+  SUnit *pickNodeBidrectional(bool &IsTopNode);
+
+  int SchedulingCost(ReadyQueue &Q,
+                     SUnit *SU, SchedCandidate &Candidate,
+                     RegPressureDelta &Delta, bool verbose);
+
+  CandResult pickNodeFromQueue(ReadyQueue &Q,
+                               const RegPressureTracker &RPTracker,
+                               SchedCandidate &Candidate);
+#ifndef NDEBUG
+  void traceCandidate(const char *Label, const ReadyQueue &Q, SUnit *SU,
+                      PressureElement P = PressureElement());
+#endif
+};
+
+} // namespace
+
+
+#endif
diff --git a/lib/Target/Hexagon/HexagonNewValueJump.cpp b/lib/Target/Hexagon/HexagonNewValueJump.cpp
index 7ece408..1e91c39 100644
--- a/lib/Target/Hexagon/HexagonNewValueJump.cpp
+++ b/lib/Target/Hexagon/HexagonNewValueJump.cpp
@@ -337,7 +337,7 @@ bool HexagonNewValueJump::runOnMachineFunction(MachineFunction &MF) {
 
   DEBUG(dbgs() << "********** Hexagon New Value Jump **********\n"
                << "********** Function: "
-               << MF.getFunction()->getName() << "\n");
+               << MF.getName() << "\n");
 
 #if 0
   // for now disable this, if we move NewValueJump before register
diff --git a/lib/Target/Hexagon/HexagonPeephole.cpp b/lib/Target/Hexagon/HexagonPeephole.cpp
index 55cbc09..a295015 100644
--- a/lib/Target/Hexagon/HexagonPeephole.cpp
+++ b/lib/Target/Hexagon/HexagonPeephole.cpp
@@ -109,6 +109,7 @@ bool HexagonPeephole::runOnMachineFunction(MachineFunction &MF) {
   MRI = &MF.getRegInfo();
 
   DenseMap<unsigned, unsigned> PeepholeMap;
+  DenseMap<unsigned, std::pair<unsigned, unsigned> > PeepholeDoubleRegsMap;
 
   if (DisableHexagonPeephole) return false;
 
@@ -117,6 +118,7 @@ bool HexagonPeephole::runOnMachineFunction(MachineFunction &MF) {
        MBBb != MBBe; ++MBBb) {
     MachineBasicBlock* MBB = MBBb;
     PeepholeMap.clear();
+    PeepholeDoubleRegsMap.clear();
 
     // Traverse the basic block.
     for (MachineBasicBlock::iterator MII = MBB->begin(); MII != MBB->end();
@@ -140,6 +142,24 @@ bool HexagonPeephole::runOnMachineFunction(MachineFunction &MF) {
         }
       }
 
+      // Look for this sequence below
+      // %vregDoubleReg1 = LSRd_ri %vregDoubleReg0, 32
+      // %vregIntReg = COPY %vregDoubleReg1:subreg_loreg.
+      // and convert into
+      // %vregIntReg = COPY %vregDoubleReg0:subreg_hireg.
+      if (MI->getOpcode() == Hexagon::LSRd_ri) {
+        assert(MI->getNumOperands() == 3);
+        MachineOperand &Dst = MI->getOperand(0);
+        MachineOperand &Src1 = MI->getOperand(1);
+        MachineOperand &Src2 = MI->getOperand(2);
+        if (Src2.getImm() != 32)
+          continue;
+        unsigned DstReg = Dst.getReg();
+        unsigned SrcReg = Src1.getReg();
+        PeepholeDoubleRegsMap[DstReg] =
+          std::make_pair(*&SrcReg, 1/*Hexagon::subreg_hireg*/);
+      }
+
       // Look for P=NOT(P).
       if (!DisablePNotP &&
           (MI->getOpcode() == Hexagon::NOT_p)) {
@@ -178,6 +198,21 @@ bool HexagonPeephole::runOnMachineFunction(MachineFunction &MF) {
             // Change the 1st operand.
             MI->RemoveOperand(1);
             MI->addOperand(MachineOperand::CreateReg(PeepholeSrc, false));
+          } else  {
+            DenseMap<unsigned, std::pair<unsigned, unsigned> >::iterator DI =
+              PeepholeDoubleRegsMap.find(SrcReg);
+            if (DI != PeepholeDoubleRegsMap.end()) {
+              std::pair<unsigned,unsigned> PeepholeSrc = DI->second;
+              MI->RemoveOperand(1);
+              MI->addOperand(MachineOperand::CreateReg(PeepholeSrc.first,
+                                                       false /*isDef*/,
+                                                       false /*isImp*/,
+                                                       false /*isKill*/,
+                                                       false /*isDead*/,
+                                                       false /*isUndef*/,
+                                                       false /*isEarlyClobber*/,
+                                                       PeepholeSrc.second));
+            }
           }
         }
       }
diff --git a/lib/Target/Hexagon/HexagonRegisterInfo.cpp b/lib/Target/Hexagon/HexagonRegisterInfo.cpp
index 2c23674..3742486 100644
--- a/lib/Target/Hexagon/HexagonRegisterInfo.cpp
+++ b/lib/Target/Hexagon/HexagonRegisterInfo.cpp
@@ -310,6 +310,58 @@ void HexagonRegisterInfo::getInitialFrameState(std::vector<MachineMove>
   Moves.push_back(MachineMove(0, Dst, Src));
 }
 
+// Get the weight in units of pressure for this register class.
+const RegClassWeight &
+HexagonRegisterInfo::getRegClassWeight(const TargetRegisterClass *RC) const {
+  // Each TargetRegisterClass has a per register weight, and weight
+  // limit which must be less than the limits of its pressure sets.
+  static const RegClassWeight RCWeightTable[] = {
+    {1, 32}, // IntRegs
+    {1, 8},  // CRRegs
+    {1, 4},  // PredRegs
+    {2, 16}, // DoubleRegs
+    {0, 0} };
+  return RCWeightTable[RC->getID()];
+}
+
+/// Get the number of dimensions of register pressure.
+unsigned HexagonRegisterInfo::getNumRegPressureSets() const {
+  return 4;
+}
+
+/// Get the name of this register unit pressure set.
+const char *HexagonRegisterInfo::getRegPressureSetName(unsigned Idx) const {
+  static const char *const RegPressureSetName[] = {
+    "IntRegsRegSet",
+    "CRRegsRegSet",
+    "PredRegsRegSet",
+    "DoubleRegsRegSet"
+  };
+  assert((Idx < 4) && "Index out of bounds");
+  return RegPressureSetName[Idx];
+}
+
+/// Get the register unit pressure limit for this dimension.
+/// This limit must be adjusted dynamically for reserved registers.
+unsigned HexagonRegisterInfo::getRegPressureSetLimit(unsigned Idx) const {
+  static const int RegPressureLimit [] = { 16, 4, 2, 8 };
+  assert((Idx < 4) && "Index out of bounds");
+  return RegPressureLimit[Idx];
+}
+
+const int*
+HexagonRegisterInfo::getRegClassPressureSets(const TargetRegisterClass *RC)
+  const {
+  static const int RCSetsTable[] = {
+    0,  -1,  // IntRegs
+    1,  -1,  // CRRegs
+    2,  -1,  // PredRegs
+    0,  -1,  // DoubleRegs
+    -1 };
+  static const unsigned RCSetStartTable[] = { 0, 2, 4, 6, 0 };
+  unsigned SetListStart = RCSetStartTable[RC->getID()];
+  return &RCSetsTable[SetListStart];
+}
 unsigned HexagonRegisterInfo::getEHExceptionRegister() const {
   llvm_unreachable("What is the exception register");
 }
diff --git a/lib/Target/Hexagon/HexagonRegisterInfo.h b/lib/Target/Hexagon/HexagonRegisterInfo.h
index 85355ae..8820d13 100644
--- a/lib/Target/Hexagon/HexagonRegisterInfo.h
+++ b/lib/Target/Hexagon/HexagonRegisterInfo.h
@@ -87,6 +87,11 @@ struct HexagonRegisterInfo : public HexagonGenRegisterInfo {
   // Exception handling queries.
   unsigned getEHExceptionRegister() const;
   unsigned getEHHandlerRegister() const;
+  const RegClassWeight &getRegClassWeight(const TargetRegisterClass *RC) const;
+  unsigned getNumRegPressureSets() const;
+  const char *getRegPressureSetName(unsigned Idx) const;
+  unsigned getRegPressureSetLimit(unsigned Idx) const;
+  const int* getRegClassPressureSets(const TargetRegisterClass *RC) const;
 };
 
 } // end namespace llvm
diff --git a/lib/Target/Hexagon/HexagonRemoveSZExtArgs.cpp b/lib/Target/Hexagon/HexagonRemoveSZExtArgs.cpp
index 2468f0b..4d93dd1 100644
--- a/lib/Target/Hexagon/HexagonRemoveSZExtArgs.cpp
+++ b/lib/Target/Hexagon/HexagonRemoveSZExtArgs.cpp
@@ -50,7 +50,7 @@ bool HexagonRemoveExtendArgs::runOnFunction(Function &F) {
   unsigned Idx = 1;
   for (Function::arg_iterator AI = F.arg_begin(), AE = F.arg_end(); AI != AE;
        ++AI, ++Idx) {
-    if (F.paramHasAttr(Idx, Attribute::SExt)) {
+    if (F.getParamAttributes(Idx).hasAttribute(Attributes::SExt)) {
       Argument* Arg = AI;
       if (!isa<PointerType>(Arg->getType())) {
         for (Instruction::use_iterator UI = Arg->use_begin();
diff --git a/lib/Target/Hexagon/HexagonSchedule.td b/lib/Target/Hexagon/HexagonSchedule.td
index d1076b8..b5ff69a 100644
--- a/lib/Target/Hexagon/HexagonSchedule.td
+++ b/lib/Target/Hexagon/HexagonSchedule.td
@@ -47,6 +47,7 @@ def HexagonModel : SchedMachineModel {
   // Max issue per cycle == bundle width.
   let IssueWidth = 4;
   let Itineraries = HexagonItineraries;
+  let LoadLatency = 1;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/Hexagon/HexagonScheduleV4.td b/lib/Target/Hexagon/HexagonScheduleV4.td
index 9b41126..5668ae8 100644
--- a/lib/Target/Hexagon/HexagonScheduleV4.td
+++ b/lib/Target/Hexagon/HexagonScheduleV4.td
@@ -58,6 +58,7 @@ def HexagonModelV4 : SchedMachineModel {
   // Max issue per cycle == bundle width.
   let IssueWidth = 4;
   let Itineraries = HexagonItinerariesV4;
+  let LoadLatency = 1;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/Hexagon/HexagonSubtarget.cpp b/lib/Target/Hexagon/HexagonSubtarget.cpp
index 5d087db..4bacb8f 100644
--- a/lib/Target/Hexagon/HexagonSubtarget.cpp
+++ b/lib/Target/Hexagon/HexagonSubtarget.cpp
@@ -40,28 +40,27 @@ EnableIEEERndNear(
 
 HexagonSubtarget::HexagonSubtarget(StringRef TT, StringRef CPU, StringRef FS):
   HexagonGenSubtargetInfo(TT, CPU, FS),
-  HexagonArchVersion(V2),
   CPUString(CPU.str()) {
-  ParseSubtargetFeatures(CPU, FS);
 
-  switch(HexagonArchVersion) {
-  case HexagonSubtarget::V2:
-    break;
-  case HexagonSubtarget::V3:
-    EnableV3 = true;
-    break;
-  case HexagonSubtarget::V4:
-    break;
-  case HexagonSubtarget::V5:
-    break;
-  default:
-    // If the programmer has not specified a Hexagon version, default
-    // to -mv4.
+  // If the programmer has not specified a Hexagon version, default to -mv4.
+  if (CPUString.empty())
     CPUString = "hexagonv4";
-    HexagonArchVersion = HexagonSubtarget::V4;
-    break;
+
+  if (CPUString == "hexagonv2") {
+    HexagonArchVersion = V2;
+  } else if (CPUString == "hexagonv3") {
+    EnableV3 = true;
+    HexagonArchVersion = V3;
+  } else if (CPUString == "hexagonv4") {
+    HexagonArchVersion = V4;
+  } else if (CPUString == "hexagonv5") {
+    HexagonArchVersion = V5;
+  } else {
+    llvm_unreachable("Unrecognized Hexagon processor version");
   }
 
+  ParseSubtargetFeatures(CPUString, FS);
+
   // Initialize scheduling itinerary for the specified CPU.
   InstrItins = getInstrItineraryForCPU(CPUString);
 
diff --git a/lib/Target/Hexagon/HexagonTargetMachine.cpp b/lib/Target/Hexagon/HexagonTargetMachine.cpp
index a7b291f..30866e9 100644
--- a/lib/Target/Hexagon/HexagonTargetMachine.cpp
+++ b/lib/Target/Hexagon/HexagonTargetMachine.cpp
@@ -14,6 +14,7 @@
 #include "HexagonTargetMachine.h"
 #include "Hexagon.h"
 #include "HexagonISelLowering.h"
+#include "HexagonMachineScheduler.h"
 #include "llvm/Module.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/PassManager.h"
@@ -29,6 +30,11 @@ opt<bool> DisableHardwareLoops(
                         "disable-hexagon-hwloops", cl::Hidden,
                         cl::desc("Disable Hardware Loops for Hexagon target"));
 
+static cl::
+opt<bool> DisableHexagonMISched("disable-hexagon-misched",
+                                cl::Hidden, cl::ZeroOrMore, cl::init(false),
+                                cl::desc("Disable Hexagon MI Scheduling"));
+
 /// HexagonTargetMachineModule - Note that this is used on hosts that
 /// cannot link in a library unless there are references into the
 /// library.  In particular, it seems that it is not possible to get
@@ -42,6 +48,13 @@ extern "C" void LLVMInitializeHexagonTarget() {
   RegisterTargetMachine<HexagonTargetMachine> X(TheHexagonTarget);
 }
 
+static ScheduleDAGInstrs *createVLIWMachineSched(MachineSchedContext *C) {
+  return new VLIWMachineScheduler(C, new ConvergingVLIWScheduler());
+}
+
+static MachineSchedRegistry
+SchedCustomRegistry("hexagon", "Run Hexagon's custom scheduler",
+                    createVLIWMachineSched);
 
 /// HexagonTargetMachine ctor - Create an ILP32 architecture model.
 ///
@@ -55,13 +68,14 @@ HexagonTargetMachine::HexagonTargetMachine(const Target &T, StringRef TT,
                                            CodeModel::Model CM,
                                            CodeGenOpt::Level OL)
   : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
-    DataLayout("e-p:32:32:32-"
+    DL("e-p:32:32:32-"
                 "i64:64:64-i32:32:32-i16:16:16-i1:32:32-"
                 "f64:64:64-f32:32:32-a0:0-n32") ,
     Subtarget(TT, CPU, FS), InstrInfo(Subtarget), TLInfo(*this),
     TSInfo(*this),
     FrameLowering(Subtarget),
-    InstrItins(&Subtarget.getInstrItineraryData()) {
+    InstrItins(&Subtarget.getInstrItineraryData()),
+    STTI(&TLInfo), VTTI(&TLInfo) {
   setMCUseCFI(false);
 }
 
@@ -74,7 +88,7 @@ bool HexagonTargetMachine::addPassesForOptimizations(PassManagerBase &PM) {
   PM.add(createDeadCodeEliminationPass());
   PM.add(createConstantPropagationPass());
   PM.add(createLoopUnrollPass());
-  PM.add(createLoopStrengthReducePass(getTargetLowering()));
+  PM.add(createLoopStrengthReducePass());
   return true;
 }
 
@@ -83,7 +97,13 @@ namespace {
 class HexagonPassConfig : public TargetPassConfig {
 public:
   HexagonPassConfig(HexagonTargetMachine *TM, PassManagerBase &PM)
-    : TargetPassConfig(TM, PM) {}
+    : TargetPassConfig(TM, PM) {
+    // Enable MI scheduler.
+    if (!DisableHexagonMISched) {
+      enablePass(&MachineSchedulerID);
+      MachineSchedRegistry::setDefault(createVLIWMachineSched);
+    }
+  }
 
   HexagonTargetMachine &getHexagonTargetMachine() const {
     return getTM<HexagonTargetMachine>();
diff --git a/lib/Target/Hexagon/HexagonTargetMachine.h b/lib/Target/Hexagon/HexagonTargetMachine.h
index 0336965..7a4215c 100644
--- a/lib/Target/Hexagon/HexagonTargetMachine.h
+++ b/lib/Target/Hexagon/HexagonTargetMachine.h
@@ -20,20 +20,23 @@
 #include "HexagonSelectionDAGInfo.h"
 #include "HexagonFrameLowering.h"
 #include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
+#include "llvm/Target/TargetTransformImpl.h"
 
 namespace llvm {
 
 class Module;
 
 class HexagonTargetMachine : public LLVMTargetMachine {
-  const TargetData DataLayout;       // Calculates type size & alignment.
+  const DataLayout DL;       // Calculates type size & alignment.
   HexagonSubtarget Subtarget;
   HexagonInstrInfo InstrInfo;
   HexagonTargetLowering TLInfo;
   HexagonSelectionDAGInfo TSInfo;
   HexagonFrameLowering FrameLowering;
   const InstrItineraryData* InstrItins;
+  ScalarTargetTransformImpl STTI;
+  VectorTargetTransformImpl VTTI;
 
 public:
   HexagonTargetMachine(const Target &T, StringRef TT,StringRef CPU,
@@ -68,7 +71,15 @@ public:
     return &TSInfo;
   }
 
-  virtual const TargetData       *getTargetData() const { return &DataLayout; }
+  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
+    return &STTI;
+  }
+
+  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
+    return &VTTI;
+  }
+
+  virtual const DataLayout       *getDataLayout() const { return &DL; }
   static unsigned getModuleMatchQuality(const Module &M);
 
   // Pass Pipeline Configuration.
diff --git a/lib/Target/Hexagon/HexagonTargetObjectFile.cpp b/lib/Target/Hexagon/HexagonTargetObjectFile.cpp
index 32cc709..f4d7761 100644
--- a/lib/Target/Hexagon/HexagonTargetObjectFile.cpp
+++ b/lib/Target/Hexagon/HexagonTargetObjectFile.cpp
@@ -16,7 +16,7 @@
 #include "HexagonTargetMachine.h"
 #include "llvm/Function.h"
 #include "llvm/GlobalVariable.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/DerivedTypes.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/Support/ELF.h"
@@ -73,7 +73,7 @@ IsGlobalInSmallSection(const GlobalValue *GV, const TargetMachine &TM,
 
   if (Kind.isBSS() || Kind.isDataNoRel() || Kind.isCommon()) {
     Type *Ty = GV->getType()->getElementType();
-    return IsInSmallSection(TM.getTargetData()->getTypeAllocSize(Ty));
+    return IsInSmallSection(TM.getDataLayout()->getTypeAllocSize(Ty));
   }
 
   return false;
diff --git a/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp b/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
index a03ed03..3d5f685 100644
--- a/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
+++ b/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
@@ -3474,8 +3474,8 @@ bool HexagonPacketizerList::isLegalToPacketizeTogether(SUnit *SUI, SUnit *SUJ) {
       // 1. Two loads unless they are volatile.
       // 2. Two stores in V4 unless they are volatile.
       else if ((DepType == SDep::Order) &&
-               !I->hasVolatileMemoryRef() &&
-               !J->hasVolatileMemoryRef()) {
+               !I->hasOrderedMemoryRef() &&
+               !J->hasOrderedMemoryRef()) {
         if (QRI->Subtarget.hasV4TOps() &&
             // hexagonv4 allows dual store.
             MCIDI.mayStore() && MCIDJ.mayStore()) {
diff --git a/lib/Target/Hexagon/HexagonVarargsCallingConvention.h b/lib/Target/Hexagon/HexagonVarargsCallingConvention.h
index 9305c27..c607b5d 100644
--- a/lib/Target/Hexagon/HexagonVarargsCallingConvention.h
+++ b/lib/Target/Hexagon/HexagonVarargsCallingConvention.h
@@ -75,9 +75,9 @@ static bool CC_Hexagon32_VarArgs(unsigned ValNo, EVT ValVT,
 
   const Type* ArgTy = LocVT.getTypeForEVT(State.getContext());
   unsigned Alignment =
-    State.getTarget().getTargetData()->getABITypeAlignment(ArgTy);
+    State.getTarget().getDataLayout()->getABITypeAlignment(ArgTy);
   unsigned Size =
-    State.getTarget().getTargetData()->getTypeSizeInBits(ArgTy) / 8;
+    State.getTarget().getDataLayout()->getTypeSizeInBits(ArgTy) / 8;
 
   // If it's passed by value, then we need the size of the aggregate not of
   // the pointer.
@@ -130,9 +130,9 @@ static bool RetCC_Hexagon32_VarArgs(unsigned ValNo, EVT ValVT,
 
   const Type* ArgTy = LocVT.getTypeForEVT(State.getContext());
   unsigned Alignment =
-    State.getTarget().getTargetData()->getABITypeAlignment(ArgTy);
+    State.getTarget().getDataLayout()->getABITypeAlignment(ArgTy);
   unsigned Size =
-    State.getTarget().getTargetData()->getTypeSizeInBits(ArgTy) / 8;
+    State.getTarget().getDataLayout()->getTypeSizeInBits(ArgTy) / 8;
 
   unsigned Offset3 = State.AllocateStack(Size, Alignment);
   State.addLoc(CCValAssign::getMem(ValNo, ValVT.getSimpleVT(), Offset3,
diff --git a/lib/Target/Hexagon/MCTargetDesc/HexagonMCAsmInfo.cpp b/lib/Target/Hexagon/MCTargetDesc/HexagonMCAsmInfo.cpp
index d6e6c36..86f75d1 100644
--- a/lib/Target/Hexagon/MCTargetDesc/HexagonMCAsmInfo.cpp
+++ b/lib/Target/Hexagon/MCTargetDesc/HexagonMCAsmInfo.cpp
@@ -24,7 +24,7 @@ HexagonMCAsmInfo::HexagonMCAsmInfo(const Target &T, StringRef TT) {
   HasLEB128 = true;
 
   PrivateGlobalPrefix = ".L";
-  LCOMMDirectiveType = LCOMM::ByteAlignment;
+  LCOMMDirectiveAlignmentType = LCOMM::ByteAlignment;
   InlineAsmStart = "# InlineAsm Start";
   InlineAsmEnd = "# InlineAsm End";
   ZeroDirective = "\t.space\t";