Merge llvm 3.5.0 release from ^/vendor/llvm/dist, resolve conflicts, and

preserve our customizations, where necessary.

47 files changed, 12664 insertions, 1206 deletions

diff --git a/contrib/llvm/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.cpp b/contrib/llvm/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.cpp
index d5be0e42ab0c..80b2f621fb94 100644
--- a/contrib/llvm/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.cpp
@@ -11,20 +11,21 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "asm-printer"
 #include "InstPrinter/NVPTXInstPrinter.h"
-#include "NVPTX.h"
 #include "MCTargetDesc/NVPTXBaseInfo.h"
+#include "NVPTX.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
-#include "llvm/MC/MCSymbol.h"
 #include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/MC/MCSymbol.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormattedStream.h"
 #include <cctype>
 using namespace llvm;
 
+#define DEBUG_TYPE "asm-printer"
+
 #include "NVPTXGenAsmWriter.inc"
 
 
@@ -56,13 +57,13 @@ void NVPTXInstPrinter::printRegName(raw_ostream &OS, unsigned RegNo) const {
     OS << "%r";
     break;
   case 4:
-    OS << "%rl";
+    OS << "%rd";
     break;
   case 5:
     OS << "%f";
     break;
   case 6:
-    OS << "%fl";
+    OS << "%fd";
     break;
   }
 
diff --git a/contrib/llvm/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.h b/contrib/llvm/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.h
index 93029ae8d39e..1fb3c57390c2 100644
--- a/contrib/llvm/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.h
+++ b/contrib/llvm/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.h
@@ -27,8 +27,8 @@ public:
   NVPTXInstPrinter(const MCAsmInfo &MAI, const MCInstrInfo &MII,
                    const MCRegisterInfo &MRI, const MCSubtargetInfo &STI);
 
-  virtual void printRegName(raw_ostream &OS, unsigned RegNo) const;
-  virtual void printInst(const MCInst *MI, raw_ostream &OS, StringRef Annot);
+  void printRegName(raw_ostream &OS, unsigned RegNo) const override;
+  void printInst(const MCInst *MI, raw_ostream &OS, StringRef Annot) override;
 
   // Autogenerated by tblgen.
   void printInstruction(const MCInst *MI, raw_ostream &O);
@@ -37,15 +37,15 @@ public:
 
   void printOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
   void printCvtMode(const MCInst *MI, int OpNum, raw_ostream &O,
-                    const char *Modifier = 0);
+                    const char *Modifier = nullptr);
   void printCmpMode(const MCInst *MI, int OpNum, raw_ostream &O,
-                    const char *Modifier = 0);
+                    const char *Modifier = nullptr);
   void printLdStCode(const MCInst *MI, int OpNum,
-                     raw_ostream &O, const char *Modifier = 0);
+                     raw_ostream &O, const char *Modifier = nullptr);
   void printMemOperand(const MCInst *MI, int OpNum,
-                       raw_ostream &O, const char *Modifier = 0);
+                       raw_ostream &O, const char *Modifier = nullptr);
   void printProtoIdent(const MCInst *MI, int OpNum,
-                       raw_ostream &O, const char *Modifier = 0);
+                       raw_ostream &O, const char *Modifier = nullptr);
 };
 
 }
diff --git a/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXBaseInfo.h b/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXBaseInfo.h
index edf4a80b11e6..16ec19c25f16 100644
--- a/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXBaseInfo.h
+++ b/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXBaseInfo.h
@@ -43,14 +43,16 @@ enum PropertyAnnotation {
   PROPERTY_ISSAMPLER,
   PROPERTY_ISREADONLY_IMAGE_PARAM,
   PROPERTY_ISWRITEONLY_IMAGE_PARAM,
+  PROPERTY_ISREADWRITE_IMAGE_PARAM,
   PROPERTY_ISKERNEL_FUNCTION,
   PROPERTY_ALIGN,
+  PROPERTY_MANAGED,
 
   // last property
   PROPERTY_LAST
 };
 
-const unsigned AnnotationNameLen = 8; // length of each annotation name
+const unsigned AnnotationNameLen = 9; // length of each annotation name
 const char PropertyAnnotationNames[PROPERTY_LAST + 1][AnnotationNameLen + 1] = {
   "maxntidx",                         // PROPERTY_MAXNTID_X
   "maxntidy",                         // PROPERTY_MAXNTID_Y
@@ -64,8 +66,10 @@ const char PropertyAnnotationNames[PROPERTY_LAST + 1][AnnotationNameLen + 1] = {
   "sampler",                          // PROPERTY_ISSAMPLER
   "rdoimage",                         // PROPERTY_ISREADONLY_IMAGE_PARAM
   "wroimage",                         // PROPERTY_ISWRITEONLY_IMAGE_PARAM
+  "rdwrimage",                        // PROPERTY_ISREADWRITE_IMAGE_PARAM
   "kernel",                           // PROPERTY_ISKERNEL_FUNCTION
   "align",                            // PROPERTY_ALIGN
+  "managed",                          // PROPERTY_MANAGED
 
               // last property
   "proplast", // PROPERTY_LAST
@@ -80,6 +84,17 @@ __attribute__((unused))
 #endif
     static const char *NamedMDForAnnotations = "nvvm.annotations";
 
+namespace NVPTXII {
+enum {
+  // These must be kept in sync with TSFlags in NVPTXInstrFormats.td
+  IsTexFlag = 0x80,
+  IsSuldMask = 0x300,
+  IsSuldShift = 8,
+  IsSustFlag = 0x400,
+  IsSurfTexQueryFlag = 0x800,
+  IsTexModeUnifiedFlag = 0x1000
+};
+}
 }
 
 #endif
diff --git a/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXMCAsmInfo.cpp b/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXMCAsmInfo.cpp
index f2784b836b70..366341afe1b8 100644
--- a/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXMCAsmInfo.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXMCAsmInfo.cpp
@@ -33,8 +33,6 @@ NVPTXMCAsmInfo::NVPTXMCAsmInfo(const StringRef &TT) {
 
   CommentString = "//";
 
-  PrivateGlobalPrefix = "$L__";
-
   HasSetDirective = false;
 
   HasSingleParameterDotFile = false;
@@ -49,7 +47,6 @@ NVPTXMCAsmInfo::NVPTXMCAsmInfo(const StringRef &TT) {
   Data16bitsDirective = " .b16 ";
   Data32bitsDirective = " .b32 ";
   Data64bitsDirective = " .b64 ";
-  PrivateGlobalPrefix = "";
   ZeroDirective = " .b8";
   AsciiDirective = " .b8";
   AscizDirective = " .b8";
diff --git a/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXMCTargetDesc.cpp b/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXMCTargetDesc.cpp
index 871bac94f84c..158ca901e586 100644
--- a/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXMCTargetDesc.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXMCTargetDesc.cpp
@@ -12,14 +12,16 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTXMCTargetDesc.h"
-#include "NVPTXMCAsmInfo.h"
 #include "InstPrinter/NVPTXInstPrinter.h"
+#include "NVPTXMCAsmInfo.h"
 #include "llvm/MC/MCCodeGenInfo.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
 #include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/Support/TargetRegistry.h"
 
+using namespace llvm;
+
 #define GET_INSTRINFO_MC_DESC
 #include "NVPTXGenInstrInfo.inc"
 
@@ -29,8 +31,6 @@
 #define GET_REGINFO_MC_DESC
 #include "NVPTXGenRegisterInfo.inc"
 
-using namespace llvm;
-
 static MCInstrInfo *createNVPTXMCInstrInfo() {
   MCInstrInfo *X = new MCInstrInfo();
   InitNVPTXMCInstrInfo(X);
@@ -66,7 +66,7 @@ static MCInstPrinter *createNVPTXMCInstPrinter(const Target &T,
                                                const MCSubtargetInfo &STI) {
   if (SyntaxVariant == 0)
     return new NVPTXInstPrinter(MAI, MII, MRI, STI);
-  return 0;
+  return nullptr;
 }
 
 // Force static initialization.
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTX.h b/contrib/llvm/lib/Target/NVPTX/NVPTX.h
index 490b49d9ead6..e74c808f8554 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTX.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTX.h
@@ -61,10 +61,14 @@ inline static const char *NVPTXCondCodeToString(NVPTXCC::CondCodes CC) {
 
 FunctionPass *
 createNVPTXISelDag(NVPTXTargetMachine &TM, llvm::CodeGenOpt::Level OptLevel);
+ModulePass *createNVPTXAssignValidGlobalNamesPass();
 ModulePass *createGenericToNVVMPass();
+FunctionPass *createNVPTXFavorNonGenericAddrSpacesPass();
 ModulePass *createNVVMReflectPass();
 ModulePass *createNVVMReflectPass(const StringMap<int>& Mapping);
 MachineFunctionPass *createNVPTXPrologEpilogPass();
+MachineFunctionPass *createNVPTXReplaceImageHandlesPass();
+FunctionPass *createNVPTXImageOptimizerPass();
 
 bool isImageOrSamplerVal(const Value *, const Module *);
 
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTX.td b/contrib/llvm/lib/Target/NVPTX/NVPTX.td
index 6183a755c320..93fabf615369 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTX.td
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTX.td
@@ -34,12 +34,18 @@ def SM30 : SubtargetFeature<"sm_30", "SmVersion", "30",
                             "Target SM 3.0">;
 def SM35 : SubtargetFeature<"sm_35", "SmVersion", "35",
                             "Target SM 3.5">;
+def SM50 : SubtargetFeature<"sm_50", "SmVersion", "50",
+                            "Target SM 5.0">;
 
 // PTX Versions
 def PTX30 : SubtargetFeature<"ptx30", "PTXVersion", "30",
                              "Use PTX version 3.0">;
 def PTX31 : SubtargetFeature<"ptx31", "PTXVersion", "31",
                              "Use PTX version 3.1">;
+def PTX32 : SubtargetFeature<"ptx32", "PTXVersion", "32",
+                             "Use PTX version 3.2">;
+def PTX40 : SubtargetFeature<"ptx40", "PTXVersion", "40",
+                             "Use PTX version 4.0">;
 
 //===----------------------------------------------------------------------===//
 // NVPTX supported processors.
@@ -52,17 +58,12 @@ def : Proc<"sm_20", [SM20]>;
 def : Proc<"sm_21", [SM21]>;
 def : Proc<"sm_30", [SM30]>;
 def : Proc<"sm_35", [SM35]>;
+def : Proc<"sm_50", [SM50]>;
 
 
 def NVPTXInstrInfo : InstrInfo {
 }
 
-def NVPTXAsmWriter : AsmWriter {
-  bit isMCAsmWriter = 1;
-  string AsmWriterClassName  = "InstPrinter";
-}
-
 def NVPTX : Target {
   let InstructionSet = NVPTXInstrInfo;
-  let AssemblyWriters = [NVPTXAsmWriter];
 }
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXAllocaHoisting.h b/contrib/llvm/lib/Target/NVPTX/NVPTXAllocaHoisting.h
index 19d73c5783cb..5b610687e391 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXAllocaHoisting.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXAllocaHoisting.h
@@ -30,16 +30,17 @@ public:
   static char ID; // Pass ID
   NVPTXAllocaHoisting() : FunctionPass(ID) {}
 
-  void getAnalysisUsage(AnalysisUsage &AU) const {
-    AU.addRequired<DataLayout>();
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.addRequired<DataLayoutPass>();
+    AU.addPreserved("stack-protector");
     AU.addPreserved<MachineFunctionAnalysis>();
   }
 
-  virtual const char *getPassName() const {
+  const char *getPassName() const override {
     return "NVPTX specific alloca hoisting";
   }
 
-  virtual bool runOnFunction(Function &function);
+  bool runOnFunction(Function &function) override;
 };
 
 extern FunctionPass *createAllocaHoisting();
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
index 7552fe704115..187b88c1d54a 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
@@ -13,26 +13,27 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTXAsmPrinter.h"
+#include "InstPrinter/NVPTXInstPrinter.h"
 #include "MCTargetDesc/NVPTXMCAsmInfo.h"
 #include "NVPTX.h"
 #include "NVPTXInstrInfo.h"
+#include "NVPTXMachineFunctionInfo.h"
 #include "NVPTXMCExpr.h"
 #include "NVPTXRegisterInfo.h"
 #include "NVPTXTargetMachine.h"
 #include "NVPTXUtilities.h"
-#include "InstPrinter/NVPTXInstPrinter.h"
 #include "cl_common_defines.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/Analysis/ConstantFolding.h"
-#include "llvm/Assembly/Writer.h"
 #include "llvm/CodeGen/Analysis.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/DebugInfo.h"
+#include "llvm/IR/DebugInfo.h"
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Mangler.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/Operator.h"
 #include "llvm/MC/MCStreamer.h"
@@ -43,7 +44,6 @@
 #include "llvm/Support/Path.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/TimeValue.h"
-#include "llvm/Target/Mangler.h"
 #include "llvm/Target/TargetLoweringObjectFile.h"
 #include <sstream>
 using namespace llvm;
@@ -132,7 +132,7 @@ const MCExpr *nvptx::LowerConstant(const Constant *CV, AsmPrinter &AP) {
     return MCSymbolRefExpr::Create(AP.GetBlockAddressSymbol(BA), Ctx);
 
   const ConstantExpr *CE = dyn_cast<ConstantExpr>(CV);
-  if (CE == 0)
+  if (!CE)
     llvm_unreachable("Unknown constant value to lower!");
 
   switch (CE->getOpcode()) {
@@ -149,10 +149,25 @@ const MCExpr *nvptx::LowerConstant(const Constant *CV, AsmPrinter &AP) {
       std::string S;
       raw_string_ostream OS(S);
       OS << "Unsupported expression in static initializer: ";
-      WriteAsOperand(OS, CE, /*PrintType=*/ false,
-                     !AP.MF ? 0 : AP.MF->getFunction()->getParent());
+      CE->printAsOperand(OS, /*PrintType=*/ false,
+                         !AP.MF ? nullptr : AP.MF->getFunction()->getParent());
       report_fatal_error(OS.str());
     }
+  case Instruction::AddrSpaceCast: {
+    // Strip any addrspace(1)->addrspace(0) addrspace casts. These will be
+    // handled by the generic() logic in the MCExpr printer
+    PointerType *DstTy            = cast<PointerType>(CE->getType());
+    PointerType *SrcTy            = cast<PointerType>(CE->getOperand(0)->getType());
+    if (SrcTy->getAddressSpace() == 1 && DstTy->getAddressSpace() == 0) {
+      return LowerConstant(cast<const Constant>(CE->getOperand(0)), AP);
+    }
+    std::string S;
+    raw_string_ostream OS(S);
+    OS << "Unsupported expression in static initializer: ";
+    CE->printAsOperand(OS, /*PrintType=*/ false,
+                       !AP.MF ? nullptr : AP.MF->getFunction()->getParent());
+    report_fatal_error(OS.str());
+  }
   case Instruction::GetElementPtr: {
     const DataLayout &TD = *AP.TM.getDataLayout();
     // Generate a symbolic expression for the byte address
@@ -308,16 +323,80 @@ void NVPTXAsmPrinter::EmitInstruction(const MachineInstr *MI) {
 
   MCInst Inst;
   lowerToMCInst(MI, Inst);
-  OutStreamer.EmitInstruction(Inst);
+  EmitToStreamer(OutStreamer, Inst);
+}
+
+// Handle symbol backtracking for targets that do not support image handles
+bool NVPTXAsmPrinter::lowerImageHandleOperand(const MachineInstr *MI,
+                                           unsigned OpNo, MCOperand &MCOp) {
+  const MachineOperand &MO = MI->getOperand(OpNo);
+  const MCInstrDesc &MCID = MI->getDesc();
+
+  if (MCID.TSFlags & NVPTXII::IsTexFlag) {
+    // This is a texture fetch, so operand 4 is a texref and operand 5 is
+    // a samplerref
+    if (OpNo == 4 && MO.isImm()) {
+      lowerImageHandleSymbol(MO.getImm(), MCOp);
+      return true;
+    }
+    if (OpNo == 5 && MO.isImm() && !(MCID.TSFlags & NVPTXII::IsTexModeUnifiedFlag)) {
+      lowerImageHandleSymbol(MO.getImm(), MCOp);
+      return true;
+    }
+
+    return false;
+  } else if (MCID.TSFlags & NVPTXII::IsSuldMask) {
+    unsigned VecSize =
+      1 << (((MCID.TSFlags & NVPTXII::IsSuldMask) >> NVPTXII::IsSuldShift) - 1);
+
+    // For a surface load of vector size N, the Nth operand will be the surfref
+    if (OpNo == VecSize && MO.isImm()) {
+      lowerImageHandleSymbol(MO.getImm(), MCOp);
+      return true;
+    }
+
+    return false;
+  } else if (MCID.TSFlags & NVPTXII::IsSustFlag) {
+    // This is a surface store, so operand 0 is a surfref
+    if (OpNo == 0 && MO.isImm()) {
+      lowerImageHandleSymbol(MO.getImm(), MCOp);
+      return true;
+    }
+
+    return false;
+  } else if (MCID.TSFlags & NVPTXII::IsSurfTexQueryFlag) {
+    // This is a query, so operand 1 is a surfref/texref
+    if (OpNo == 1 && MO.isImm()) {
+      lowerImageHandleSymbol(MO.getImm(), MCOp);
+      return true;
+    }
+
+    return false;
+  }
+
+  return false;
+}
+
+void NVPTXAsmPrinter::lowerImageHandleSymbol(unsigned Index, MCOperand &MCOp) {
+  // Ewwww
+  TargetMachine &TM = const_cast<TargetMachine&>(MF->getTarget());
+  NVPTXTargetMachine &nvTM = static_cast<NVPTXTargetMachine&>(TM);
+  const NVPTXMachineFunctionInfo *MFI = MF->getInfo<NVPTXMachineFunctionInfo>();
+  const char *Sym = MFI->getImageHandleSymbol(Index);
+  std::string *SymNamePtr =
+    nvTM.getManagedStrPool()->getManagedString(Sym);
+  MCOp = GetSymbolRef(OutContext.GetOrCreateSymbol(
+    StringRef(SymNamePtr->c_str())));
 }
 
 void NVPTXAsmPrinter::lowerToMCInst(const MachineInstr *MI, MCInst &OutMI) {
   OutMI.setOpcode(MI->getOpcode());
+  const NVPTXSubtarget &ST = TM.getSubtarget<NVPTXSubtarget>();
 
   // Special: Do not mangle symbol operand of CALL_PROTOTYPE
   if (MI->getOpcode() == NVPTX::CALL_PROTOTYPE) {
     const MachineOperand &MO = MI->getOperand(0);
-    OutMI.addOperand(GetSymbolRef(MO,
+    OutMI.addOperand(GetSymbolRef(
       OutContext.GetOrCreateSymbol(Twine(MO.getSymbolName()))));
     return;
   }
@@ -326,6 +405,13 @@ void NVPTXAsmPrinter::lowerToMCInst(const MachineInstr *MI, MCInst &OutMI) {
     const MachineOperand &MO = MI->getOperand(i);
 
     MCOperand MCOp;
+    if (!ST.hasImageHandles()) {
+      if (lowerImageHandleOperand(MI, i, MCOp)) {
+        OutMI.addOperand(MCOp);
+        continue;
+      }
+    }
+
     if (lowerOperand(MO, MCOp))
       OutMI.addOperand(MCOp);
   }
@@ -346,10 +432,10 @@ bool NVPTXAsmPrinter::lowerOperand(const MachineOperand &MO,
         MO.getMBB()->getSymbol(), OutContext));
     break;
   case MachineOperand::MO_ExternalSymbol:
-    MCOp = GetSymbolRef(MO, GetExternalSymbolSymbol(MO.getSymbolName()));
+    MCOp = GetSymbolRef(GetExternalSymbolSymbol(MO.getSymbolName()));
     break;
   case MachineOperand::MO_GlobalAddress:
-    MCOp = GetSymbolRef(MO, getSymbol(MO.getGlobal()));
+    MCOp = GetSymbolRef(getSymbol(MO.getGlobal()));
     break;
   case MachineOperand::MO_FPImmediate: {
     const ConstantFP *Cnt = MO.getFPImm();
@@ -408,8 +494,7 @@ unsigned NVPTXAsmPrinter::encodeVirtualRegister(unsigned Reg) {
   }
 }
 
-MCOperand NVPTXAsmPrinter::GetSymbolRef(const MachineOperand &MO,
-                                        const MCSymbol *Symbol) {
+MCOperand NVPTXAsmPrinter::GetSymbolRef(const MCSymbol *Symbol) {
   const MCExpr *Expr;
   Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_None,
                                  OutContext);
@@ -430,7 +515,7 @@ void NVPTXAsmPrinter::printReturnValStr(const Function *F, raw_ostream &O) {
   O << " (";
 
   if (isABI) {
-    if (Ty->isPrimitiveType() || Ty->isIntegerTy()) {
+    if (Ty->isFloatingPointTy() || Ty->isIntegerTy()) {
       unsigned size = 0;
       if (const IntegerType *ITy = dyn_cast<IntegerType>(Ty)) {
         size = ITy->getBitWidth();
@@ -447,23 +532,7 @@ void NVPTXAsmPrinter::printReturnValStr(const Function *F, raw_ostream &O) {
         << " func_retval0";
     } else {
       if ((Ty->getTypeID() == Type::StructTyID) || isa<VectorType>(Ty)) {
-        SmallVector<EVT, 16> vtparts;
-        ComputeValueVTs(*TLI, Ty, vtparts);
-        unsigned totalsz = 0;
-        for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
-          unsigned elems = 1;
-          EVT elemtype = vtparts[i];
-          if (vtparts[i].isVector()) {
-            elems = vtparts[i].getVectorNumElements();
-            elemtype = vtparts[i].getVectorElementType();
-          }
-          for (unsigned j = 0, je = elems; j != je; ++j) {
-            unsigned sz = elemtype.getSizeInBits();
-            if (elemtype.isInteger() && (sz < 8))
-              sz = 8;
-            totalsz += sz / 8;
-          }
-        }
+        unsigned totalsz = TD->getTypeAllocSize(Ty);
         unsigned retAlignment = 0;
         if (!llvm::getAlign(*F, 0, retAlignment))
           retAlignment = TD->getABITypeAlignment(Ty);
@@ -700,12 +769,11 @@ static bool usedInGlobalVarDef(const Constant *C) {
     return true;
   }
 
-  for (Value::const_use_iterator ui = C->use_begin(), ue = C->use_end();
-       ui != ue; ++ui) {
-    const Constant *C = dyn_cast<Constant>(*ui);
-    if (usedInGlobalVarDef(C))
-      return true;
-  }
+  for (const User *U : C->users())
+    if (const Constant *C = dyn_cast<Constant>(U))
+      if (usedInGlobalVarDef(C))
+        return true;
+
   return false;
 }
 
@@ -731,11 +799,10 @@ static bool usedInOneFunc(const User *U, Function const *&oneFunc) {
                           (md->getName().str() == "llvm.dbg.sp")))
       return true;
 
-  for (User::const_use_iterator ui = U->use_begin(), ue = U->use_end();
-       ui != ue; ++ui) {
-    if (usedInOneFunc(*ui, oneFunc) == false)
+  for (const User *UU : U->users())
+    if (usedInOneFunc(UU, oneFunc) == false)
       return false;
-  }
+
   return true;
 }
 
@@ -753,7 +820,7 @@ static bool canDemoteGlobalVar(const GlobalVariable *gv, Function const *&f) {
   if (Pty->getAddressSpace() != llvm::ADDRESS_SPACE_SHARED)
     return false;
 
-  const Function *oneFunc = 0;
+  const Function *oneFunc = nullptr;
 
   bool flag = usedInOneFunc(gv, oneFunc);
   if (flag == false)
@@ -766,12 +833,11 @@ static bool canDemoteGlobalVar(const GlobalVariable *gv, Function const *&f) {
 
 static bool useFuncSeen(const Constant *C,
                         llvm::DenseMap<const Function *, bool> &seenMap) {
-  for (Value::const_use_iterator ui = C->use_begin(), ue = C->use_end();
-       ui != ue; ++ui) {
-    if (const Constant *cu = dyn_cast<Constant>(*ui)) {
+  for (const User *U : C->users()) {
+    if (const Constant *cu = dyn_cast<Constant>(U)) {
       if (useFuncSeen(cu, seenMap))
         return true;
-    } else if (const Instruction *I = dyn_cast<Instruction>(*ui)) {
+    } else if (const Instruction *I = dyn_cast<Instruction>(U)) {
       const BasicBlock *bb = I->getParent();
       if (!bb)
         continue;
@@ -798,10 +864,8 @@ void NVPTXAsmPrinter::emitDeclarations(const Module &M, raw_ostream &O) {
       emitDeclaration(F, O);
       continue;
     }
-    for (Value::const_use_iterator iter = F->use_begin(),
-                                   iterEnd = F->use_end();
-         iter != iterEnd; ++iter) {
-      if (const Constant *C = dyn_cast<Constant>(*iter)) {
+    for (const User *U : F->users()) {
+      if (const Constant *C = dyn_cast<Constant>(U)) {
         if (usedInGlobalVarDef(C)) {
           // The use is in the initialization of a global variable
           // that is a function pointer, so print a declaration
@@ -817,9 +881,9 @@ void NVPTXAsmPrinter::emitDeclarations(const Module &M, raw_ostream &O) {
         }
       }
 
-      if (!isa<Instruction>(*iter))
+      if (!isa<Instruction>(U))
         continue;
-      const Instruction *instr = cast<Instruction>(*iter);
+      const Instruction *instr = cast<Instruction>(U);
       const BasicBlock *bb = instr->getParent();
       if (!bb)
         continue;
@@ -844,10 +908,7 @@ void NVPTXAsmPrinter::recordAndEmitFilenames(Module &M) {
   DbgFinder.processModule(M);
 
   unsigned i = 1;
-  for (DebugInfoFinder::iterator I = DbgFinder.compile_unit_begin(),
-                                 E = DbgFinder.compile_unit_end();
-       I != E; ++I) {
-    DICompileUnit DIUnit(*I);
+  for (DICompileUnit DIUnit : DbgFinder.compile_units()) {
     StringRef Filename(DIUnit.getFilename());
     StringRef Dirname(DIUnit.getDirectory());
     SmallString<128> FullPathName = Dirname;
@@ -862,10 +923,7 @@ void NVPTXAsmPrinter::recordAndEmitFilenames(Module &M) {
     ++i;
   }
 
-  for (DebugInfoFinder::iterator I = DbgFinder.subprogram_begin(),
-                                 E = DbgFinder.subprogram_end();
-       I != E; ++I) {
-    DISubprogram SP(*I);
+  for (DISubprogram SP : DbgFinder.subprograms()) {
     StringRef Filename(SP.getFilename());
     StringRef Dirname(SP.getDirectory());
     SmallString<128> FullPathName = Dirname;
@@ -895,7 +953,7 @@ bool NVPTXAsmPrinter::doInitialization(Module &M) {
   const_cast<TargetLoweringObjectFile &>(getObjFileLowering())
       .Initialize(OutContext, TM);
 
-  Mang = new Mangler(&TM);
+  Mang = new Mangler(TM.getDataLayout());
 
   // Emit header before any dwarf directives are emitted below.
   emitHeader(M, OS1);
@@ -1022,6 +1080,8 @@ bool NVPTXAsmPrinter::doFinalization(Module &M) {
   for (i = 0; i < n; i++)
     global_list.insert(global_list.end(), gv_array[i]);
 
+  clearAnnotationCache(&M);
+
   delete[] gv_array;
   return ret;
 
@@ -1043,6 +1103,10 @@ bool NVPTXAsmPrinter::doFinalization(Module &M) {
 // external global variable with init     -> .visible
 // external without init                  -> .extern
 // appending                              -> not allowed, assert.
+// for any linkage other than
+// internal, private, linker_private,
+// linker_private_weak, linker_private_weak_def_auto,
+// we emit                                -> .weak.
 
 void NVPTXAsmPrinter::emitLinkageDirective(const GlobalValue *V,
                                            raw_ostream &O) {
@@ -1068,6 +1132,9 @@ void NVPTXAsmPrinter::emitLinkageDirective(const GlobalValue *V,
         msg.append(V->getName().str());
       msg.append("has unsupported appending linkage type");
       llvm_unreachable(msg.c_str());
+    } else if (!V->hasInternalLinkage() &&
+               !V->hasPrivateLinkage()) {
+      O << ".weak ";
     }
   }
 }
@@ -1078,10 +1145,15 @@ void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
 
   // Skip meta data
   if (GVar->hasSection()) {
-    if (GVar->getSection() == "llvm.metadata")
+    if (GVar->getSection() == StringRef("llvm.metadata"))
       return;
   }
 
+  // Skip LLVM intrinsic global variables
+  if (GVar->getName().startswith("llvm.") ||
+      GVar->getName().startswith("nvvm."))
+    return;
+
   const DataLayout *TD = TM.getDataLayout();
 
   // GlobalVariables are always constant pointers themselves.
@@ -1093,6 +1165,10 @@ void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
       O << ".visible ";
     else
       O << ".extern ";
+  } else if (GVar->hasLinkOnceLinkage() || GVar->hasWeakLinkage() ||
+             GVar->hasAvailableExternallyLinkage() ||
+             GVar->hasCommonLinkage()) {
+    O << ".weak ";
   }
 
   if (llvm::isTexture(*GVar)) {
@@ -1117,10 +1193,10 @@ void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
   if (llvm::isSampler(*GVar)) {
     O << ".global .samplerref " << llvm::getSamplerName(*GVar);
 
-    const Constant *Initializer = NULL;
+    const Constant *Initializer = nullptr;
     if (GVar->hasInitializer())
       Initializer = GVar->getInitializer();
-    const ConstantInt *CI = NULL;
+    const ConstantInt *CI = nullptr;
     if (Initializer)
       CI = dyn_cast<ConstantInt>(Initializer);
     if (CI) {
@@ -1160,7 +1236,7 @@ void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
         O << "linear";
         break;
       case 2:
-        assert(0 && "Anisotropic filtering is not supported");
+        llvm_unreachable("Anisotropic filtering is not supported");
       default:
         O << "nearest";
         break;
@@ -1187,7 +1263,7 @@ void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
       return;
   }
 
-  const Function *demotedFunc = 0;
+  const Function *demotedFunc = nullptr;
   if (!processDemoted && canDemoteGlobalVar(GVar, demotedFunc)) {
     O << "// " << GVar->getName().str() << " has been demoted\n";
     if (localDecls.find(demotedFunc) != localDecls.end())
@@ -1202,12 +1278,17 @@ void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
 
   O << ".";
   emitPTXAddressSpace(PTy->getAddressSpace(), O);
+
+  if (isManaged(*GVar)) {
+    O << " .attribute(.managed)";
+  }
+
   if (GVar->getAlignment() == 0)
     O << " .align " << (int) TD->getPrefTypeAlignment(ETy);
   else
     O << " .align " << GVar->getAlignment();
 
-  if (ETy->isPrimitiveType() || ETy->isIntegerTy() || isa<PointerType>(ETy)) {
+  if (ETy->isSingleValueType()) {
     O << " .";
     // Special case: ABI requires that we use .u8 for predicates
     if (ETy->isIntegerTy(1))
@@ -1219,13 +1300,24 @@ void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
 
     // Ptx allows variable initilization only for constant and global state
     // spaces.
-    if (((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
-         (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) &&
-        GVar->hasInitializer()) {
-      const Constant *Initializer = GVar->getInitializer();
-      if (!Initializer->isNullValue()) {
-        O << " = ";
-        printScalarConstant(Initializer, O);
+    if (GVar->hasInitializer()) {
+      if ((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
+          (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) {
+        const Constant *Initializer = GVar->getInitializer();
+        // 'undef' is treated as there is no value spefied.
+        if (!Initializer->isNullValue() && !isa<UndefValue>(Initializer)) {
+          O << " = ";
+          printScalarConstant(Initializer, O);
+        }
+      } else {
+        // The frontend adds zero-initializer to variables that don't have an
+        // initial value, so skip warning for this case.
+        if (!GVar->getInitializer()->isNullValue()) {
+          std::string warnMsg = "initial value of '" + GVar->getName().str() +
+              "' is not allowed in addrspace(" +
+              llvm::utostr_32(PTy->getAddressSpace()) + ")";
+          report_fatal_error(warnMsg.c_str());
+        }
       }
     }
   } else {
@@ -1284,7 +1376,7 @@ void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
       }
       break;
     default:
-      assert(0 && "type not supported yet");
+      llvm_unreachable("type not supported yet");
     }
 
   }
@@ -1359,7 +1451,7 @@ NVPTXAsmPrinter::getPTXFundamentalTypeStr(const Type *Ty, bool useB4PTR) const {
       return "u32";
   }
   llvm_unreachable("unexpected type");
-  return NULL;
+  return nullptr;
 }
 
 void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable *GVar,
@@ -1378,7 +1470,7 @@ void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable *GVar,
   else
     O << " .align " << GVar->getAlignment();
 
-  if (ETy->isPrimitiveType() || ETy->isIntegerTy() || isa<PointerType>(ETy)) {
+  if (ETy->isSingleValueType()) {
     O << " .";
     O << getPTXFundamentalTypeStr(ETy);
     O << " ";
@@ -1404,13 +1496,13 @@ void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable *GVar,
     O << "]";
     break;
   default:
-    assert(0 && "type not supported yet");
+    llvm_unreachable("type not supported yet");
   }
   return;
 }
 
 static unsigned int getOpenCLAlignment(const DataLayout *TD, Type *Ty) {
-  if (Ty->isPrimitiveType() || Ty->isIntegerTy() || isa<PointerType>(Ty))
+  if (Ty->isSingleValueType())
     return TD->getPrefTypeAlignment(Ty);
 
   const ArrayType *ATy = dyn_cast<ArrayType>(Ty);
@@ -1507,24 +1599,38 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
     first = false;
 
     // Handle image/sampler parameters
-    if (llvm::isSampler(*I) || llvm::isImage(*I)) {
-      if (llvm::isImage(*I)) {
-        std::string sname = I->getName();
-        if (llvm::isImageWriteOnly(*I))
-          O << "\t.param .surfref " << *getSymbol(F) << "_param_"
-            << paramIndex;
-        else // Default image is read_only
-          O << "\t.param .texref " << *getSymbol(F) << "_param_"
-            << paramIndex;
-      } else // Should be llvm::isSampler(*I)
-        O << "\t.param .samplerref " << *getSymbol(F) << "_param_"
-          << paramIndex;
-      continue;
+    if (isKernelFunction(*F)) {
+      if (isSampler(*I) || isImage(*I)) {
+        if (isImage(*I)) {
+          std::string sname = I->getName();
+          if (isImageWriteOnly(*I) || isImageReadWrite(*I)) {
+            if (nvptxSubtarget.hasImageHandles())
+              O << "\t.param .u64 .ptr .surfref ";
+            else
+              O << "\t.param .surfref ";
+            O << *CurrentFnSym << "_param_" << paramIndex;
+          }
+          else { // Default image is read_only
+            if (nvptxSubtarget.hasImageHandles())
+              O << "\t.param .u64 .ptr .texref ";
+            else
+              O << "\t.param .texref ";
+            O << *CurrentFnSym << "_param_" << paramIndex;
+          }
+        } else {
+          if (nvptxSubtarget.hasImageHandles())
+            O << "\t.param .u64 .ptr .samplerref ";
+          else
+            O << "\t.param .samplerref ";
+          O << *CurrentFnSym << "_param_" << paramIndex;
+        }
+        continue;
+      }
     }
 
     if (PAL.hasAttribute(paramIndex + 1, Attribute::ByVal) == false) {
-      if (Ty->isVectorTy()) {
-        // Just print .param .b8 .align <a> .param[size];
+      if (Ty->isAggregateType() || Ty->isVectorTy()) {
+        // Just print .param .align <a> .b8 .param[size];
         // <a> = PAL.getparamalignment
         // size = typeallocsize of element type
         unsigned align = PAL.getParamAlignment(paramIndex + 1);
@@ -1580,7 +1686,7 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
         continue;
       }
       // Non-kernel function, just print .param .b<size> for ABI
-      // and .reg .b<size> for non ABY
+      // and .reg .b<size> for non-ABI
       unsigned sz = 0;
       if (isa<IntegerType>(Ty)) {
         sz = cast<IntegerType>(Ty)->getBitWidth();
@@ -1604,7 +1710,7 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
     Type *ETy = PTy->getElementType();
 
     if (isABI || isKernelFunc) {
-      // Just print .param .b8 .align <a> .param[size];
+      // Just print .param .align <a> .b8 .param[size];
       // <a> = PAL.getparamalignment
       // size = typeallocsize of element type
       unsigned align = PAL.getParamAlignment(paramIndex + 1);
@@ -1702,9 +1808,9 @@ void NVPTXAsmPrinter::setAndEmitFunctionVirtualRegisters(
   // O << "\t.reg .s16 %rc<" << NVPTXNumRegisters << ">;\n";
   // O << "\t.reg .s16 %rs<" << NVPTXNumRegisters << ">;\n";
   // O << "\t.reg .s32 %r<" << NVPTXNumRegisters << ">;\n";
-  // O << "\t.reg .s64 %rl<" << NVPTXNumRegisters << ">;\n";
+  // O << "\t.reg .s64 %rd<" << NVPTXNumRegisters << ">;\n";
   // O << "\t.reg .f32 %f<" << NVPTXNumRegisters << ">;\n";
-  // O << "\t.reg .f64 %fl<" << NVPTXNumRegisters << ">;\n";
+  // O << "\t.reg .f64 %fd<" << NVPTXNumRegisters << ">;\n";
 
   // Emit declaration of the virtual registers or 'physical' registers for
   // each register class
@@ -1764,13 +1870,35 @@ void NVPTXAsmPrinter::printScalarConstant(const Constant *CPV, raw_ostream &O) {
     return;
   }
   if (const GlobalValue *GVar = dyn_cast<GlobalValue>(CPV)) {
-    O << *getSymbol(GVar);
+    PointerType *PTy = dyn_cast<PointerType>(GVar->getType());
+    bool IsNonGenericPointer = false;
+    if (PTy && PTy->getAddressSpace() != 0) {
+      IsNonGenericPointer = true;
+    }
+    if (EmitGeneric && !isa<Function>(CPV) && !IsNonGenericPointer) {
+      O << "generic(";
+      O << *getSymbol(GVar);
+      O << ")";
+    } else {
+      O << *getSymbol(GVar);
+    }
     return;
   }
   if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
     const Value *v = Cexpr->stripPointerCasts();
+    PointerType *PTy = dyn_cast<PointerType>(Cexpr->getType());
+    bool IsNonGenericPointer = false;
+    if (PTy && PTy->getAddressSpace() != 0) {
+      IsNonGenericPointer = true;
+    }
     if (const GlobalValue *GVar = dyn_cast<GlobalValue>(v)) {
-      O << *getSymbol(GVar);
+      if (EmitGeneric && !isa<Function>(v) && !IsNonGenericPointer) {
+        O << "generic(";
+        O << *getSymbol(GVar);
+        O << ")";
+      } else {
+        O << *getSymbol(GVar);
+      }
       return;
     } else {
       O << *LowerConstant(CPV, *this);
@@ -2087,21 +2215,6 @@ void NVPTXAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
     O << *getSymbol(MO.getGlobal());
     break;
 
-  case MachineOperand::MO_ExternalSymbol: {
-    const char *symbname = MO.getSymbolName();
-    if (strstr(symbname, ".PARAM") == symbname) {
-      unsigned index;
-      sscanf(symbname + 6, "%u[];", &index);
-      printParamName(index, O);
-    } else if (strstr(symbname, ".HLPPARAM") == symbname) {
-      unsigned index;
-      sscanf(symbname + 9, "%u[];", &index);
-      O << *CurrentFnSym << "_param_" << index << "_offset";
-    } else
-      O << symbname;
-    break;
-  }
-
   case MachineOperand::MO_MachineBasicBlock:
     O << *MO.getMBB()->getSymbol();
     return;
@@ -2148,7 +2261,7 @@ void NVPTXAsmPrinter::emitSrcInText(StringRef filename, unsigned line) {
 }
 
 LineReader *NVPTXAsmPrinter::getReader(std::string filename) {
-  if (reader == NULL) {
+  if (!reader) {
     reader = new LineReader(filename);
   }
 
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.h b/contrib/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.h
index 3abe5d166826..a9f9bdd6d3d8 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.h
@@ -27,7 +27,6 @@
 #include "llvm/MC/MCSymbol.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/FormattedStream.h"
-#include "llvm/Target/Mangler.h"
 #include "llvm/Target/TargetMachine.h"
 #include <fstream>
 
@@ -97,6 +96,7 @@ class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
     unsigned curpos;
     raw_ostream &O;
     NVPTXAsmPrinter &AP;
+    bool EmitGeneric;
 
   public:
     AggBuffer(unsigned _size, raw_ostream &_O, NVPTXAsmPrinter &_AP)
@@ -105,6 +105,7 @@ class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
       size = _size;
       curpos = 0;
       numSymbols = 0;
+      EmitGeneric = AP.EmitGeneric;
     }
     ~AggBuffer() { delete[] buffer; }
     unsigned addBytes(unsigned char *Ptr, int Num, int Bytes) {
@@ -156,7 +157,18 @@ class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
             const Value *v = Symbols[nSym];
             if (const GlobalValue *GVar = dyn_cast<GlobalValue>(v)) {
               MCSymbol *Name = AP.getSymbol(GVar);
-              O << *Name;
+              PointerType *PTy = dyn_cast<PointerType>(GVar->getType());
+              bool IsNonGenericPointer = false;
+              if (PTy && PTy->getAddressSpace() != 0) {
+                IsNonGenericPointer = true;
+              }
+              if (EmitGeneric && !isa<Function>(v) && !IsNonGenericPointer) {
+                O << "generic(";
+                O << *Name;
+                O << ")";
+              } else {
+                O << *Name;
+              }
             } else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(v)) {
               O << *nvptx::LowerConstant(Cexpr, AP);
             } else
@@ -177,35 +189,32 @@ class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
 
   friend class AggBuffer;
 
-  virtual void emitSrcInText(StringRef filename, unsigned line);
+  void emitSrcInText(StringRef filename, unsigned line);
 
 private:
-  virtual const char *getPassName() const { return "NVPTX Assembly Printer"; }
+  const char *getPassName() const override { return "NVPTX Assembly Printer"; }
 
   const Function *F;
   std::string CurrentFnName;
 
-  void EmitFunctionEntryLabel();
-  void EmitFunctionBodyStart();
-  void EmitFunctionBodyEnd();
-  void emitImplicitDef(const MachineInstr *MI) const;
+  void EmitFunctionEntryLabel() override;
+  void EmitFunctionBodyStart() override;
+  void EmitFunctionBodyEnd() override;
+  void emitImplicitDef(const MachineInstr *MI) const override;
 
-  void EmitInstruction(const MachineInstr *);
+  void EmitInstruction(const MachineInstr *) override;
   void lowerToMCInst(const MachineInstr *MI, MCInst &OutMI);
   bool lowerOperand(const MachineOperand &MO, MCOperand &MCOp);
-  MCOperand GetSymbolRef(const MachineOperand &MO, const MCSymbol *Symbol);
+  MCOperand GetSymbolRef(const MCSymbol *Symbol);
   unsigned encodeVirtualRegister(unsigned Reg);
 
-  void EmitAlignment(unsigned NumBits, const GlobalValue *GV = 0) const {}
+  void EmitAlignment(unsigned NumBits, const GlobalValue *GV = nullptr) const {}
 
-  void printGlobalVariable(const GlobalVariable *GVar);
   void printVecModifiedImmediate(const MachineOperand &MO, const char *Modifier,
                                  raw_ostream &O);
   void printMemOperand(const MachineInstr *MI, int opNum, raw_ostream &O,
-                       const char *Modifier = 0);
+                       const char *Modifier = nullptr);
   void printImplicitDef(const MachineInstr *MI, raw_ostream &O) const;
-  // definition autogenerated.
-  void printInstruction(const MachineInstr *MI, raw_ostream &O);
   void printModuleLevelGV(const GlobalVariable *GVar, raw_ostream &O,
                           bool = false);
   void printParamName(int paramIndex, raw_ostream &O);
@@ -225,15 +234,15 @@ private:
   void printReturnValStr(const MachineFunction &MF, raw_ostream &O);
   bool PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
                        unsigned AsmVariant, const char *ExtraCode,
-                       raw_ostream &);
+                       raw_ostream &) override;
   void printOperand(const MachineInstr *MI, int opNum, raw_ostream &O,
-                    const char *Modifier = 0);
+                    const char *Modifier = nullptr);
   bool PrintAsmMemoryOperand(const MachineInstr *MI, unsigned OpNo,
                              unsigned AsmVariant, const char *ExtraCode,
-                             raw_ostream &);
+                             raw_ostream &) override;
 protected:
-  bool doInitialization(Module &M);
-  bool doFinalization(Module &M);
+  bool doInitialization(Module &M) override;
+  bool doFinalization(Module &M) override;
 
 private:
   std::string CurrentBankselLabelInBasicBlock;
@@ -278,14 +287,33 @@ private:
   static const char *getRegisterName(unsigned RegNo);
   void emitDemotedVars(const Function *, raw_ostream &);
 
+  bool lowerImageHandleOperand(const MachineInstr *MI, unsigned OpNo,
+                               MCOperand &MCOp);
+  void lowerImageHandleSymbol(unsigned Index, MCOperand &MCOp);
+
   LineReader *reader;
   LineReader *getReader(std::string);
+
+  // Used to control the need to emit .generic() in the initializer of
+  // module scope variables.
+  // Although ptx supports the hybrid mode like the following,
+  //    .global .u32 a;
+  //    .global .u32 b;
+  //    .global .u32 addr[] = {a, generic(b)}
+  // we have difficulty representing the difference in the NVVM IR.
+  //
+  // Since the address value should always be generic in CUDA C and always
+  // be specific in OpenCL, we use this simple control here.
+  //
+  bool EmitGeneric;
+
 public:
   NVPTXAsmPrinter(TargetMachine &TM, MCStreamer &Streamer)
       : AsmPrinter(TM, Streamer),
         nvptxSubtarget(TM.getSubtarget<NVPTXSubtarget>()) {
     CurrentBankselLabelInBasicBlock = "";
-    reader = NULL;
+    reader = nullptr;
+    EmitGeneric = (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA);
   }
 
   ~NVPTXAsmPrinter() {
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXAssignValidGlobalNames.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXAssignValidGlobalNames.cpp
new file mode 100644
index 000000000000..962b12312026
--- /dev/null
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXAssignValidGlobalNames.cpp
@@ -0,0 +1,84 @@
+//===-- NVPTXAssignValidGlobalNames.cpp - Assign valid names to globals ---===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Clean up the names of global variables in the module to not contain symbols
+// that are invalid in PTX.
+//
+// Currently NVPTX, like other backends, relies on generic symbol name
+// sanitizing done by MC. However, the ptxas assembler is more stringent and
+// disallows some additional characters in symbol names. This pass makes sure
+// such names do not reach MC at all.
+//
+//===----------------------------------------------------------------------===//
+
+#include "NVPTX.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Module.h"
+#include "llvm/PassManager.h"
+#include "llvm/Support/raw_ostream.h"
+#include <string>
+
+using namespace llvm;
+
+namespace {
+/// \brief NVPTXAssignValidGlobalNames
+class NVPTXAssignValidGlobalNames : public ModulePass {
+public:
+  static char ID;
+  NVPTXAssignValidGlobalNames() : ModulePass(ID) {}
+
+  bool runOnModule(Module &M) override;
+
+  /// \brief Clean up the name to remove symbols invalid in PTX.
+  std::string cleanUpName(StringRef Name);
+};
+}
+
+char NVPTXAssignValidGlobalNames::ID = 0;
+
+namespace llvm {
+void initializeNVPTXAssignValidGlobalNamesPass(PassRegistry &);
+}
+
+INITIALIZE_PASS(NVPTXAssignValidGlobalNames, "nvptx-assign-valid-global-names",
+                "Assign valid PTX names to globals", false, false)
+
+bool NVPTXAssignValidGlobalNames::runOnModule(Module &M) {
+  for (GlobalVariable &GV : M.globals()) {
+    // We are only allowed to rename local symbols.
+    if (GV.hasLocalLinkage()) {
+      // setName doesn't do extra work if the name does not change.
+      // Note: this does not create collisions - if setName is asked to set the
+      // name to something that already exists, it adds a proper postfix to
+      // avoid collisions.
+      GV.setName(cleanUpName(GV.getName()));
+    }
+  }
+
+  return true;
+}
+
+std::string NVPTXAssignValidGlobalNames::cleanUpName(StringRef Name) {
+  std::string ValidName;
+  raw_string_ostream ValidNameStream(ValidName);
+  for (unsigned I = 0, E = Name.size(); I != E; ++I) {
+    char C = Name[I];
+    if (C == '.' || C == '@') {
+      ValidNameStream << "_$_";
+    } else {
+      ValidNameStream << C;
+    }
+  }
+
+  return ValidNameStream.str();
+}
+
+ModulePass *llvm::createNVPTXAssignValidGlobalNamesPass() {
+  return new NVPTXAssignValidGlobalNames();
+}
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXFavorNonGenericAddrSpaces.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXFavorNonGenericAddrSpaces.cpp
new file mode 100644
index 000000000000..f3a095d829e4
--- /dev/null
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXFavorNonGenericAddrSpaces.cpp
@@ -0,0 +1,195 @@
+//===-- NVPTXFavorNonGenericAddrSpace.cpp - ---------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// When a load/store accesses the generic address space, checks whether the
+// address is casted from a non-generic address space. If so, remove this
+// addrspacecast because accessing non-generic address spaces is typically
+// faster. Besides seeking addrspacecasts, this optimization also traces into
+// the base pointer of a GEP.
+//
+// For instance, the code below loads a float from an array allocated in
+// addrspace(3).
+//
+// %0 = addrspacecast [10 x float] addrspace(3)* @a to [10 x float]*
+// %1 = gep [10 x float]* %0, i64 0, i64 %i
+// %2 = load float* %1 ; emits ld.f32
+//
+// First, function hoistAddrSpaceCastFromGEP reorders the addrspacecast
+// and the GEP to expose more optimization opportunities to function
+// optimizeMemoryInst. The intermediate code looks like:
+//
+// %0 = gep [10 x float] addrspace(3)* @a, i64 0, i64 %i
+// %1 = addrspacecast float addrspace(3)* %0 to float*
+// %2 = load float* %1 ; still emits ld.f32, but will be optimized shortly
+//
+// Then, function optimizeMemoryInstruction detects a load from addrspacecast'ed
+// generic pointers, and folds the load and the addrspacecast into a load from
+// the original address space. The final code looks like:
+//
+// %0 = gep [10 x float] addrspace(3)* @a, i64 0, i64 %i
+// %2 = load float addrspace(3)* %0 ; emits ld.shared.f32
+//
+// This pass may remove an addrspacecast in a different BB. Therefore, we
+// implement it as a FunctionPass.
+//
+//===----------------------------------------------------------------------===//
+
+#include "NVPTX.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Operator.h"
+#include "llvm/Support/CommandLine.h"
+
+using namespace llvm;
+
+// An option to disable this optimization. Enable it by default.
+static cl::opt<bool> DisableFavorNonGeneric(
+  "disable-nvptx-favor-non-generic",
+  cl::init(false),
+  cl::desc("Do not convert generic address space usage "
+           "to non-generic address space usage"),
+  cl::Hidden);
+
+namespace {
+/// \brief NVPTXFavorNonGenericAddrSpaces
+class NVPTXFavorNonGenericAddrSpaces : public FunctionPass {
+public:
+  static char ID;
+  NVPTXFavorNonGenericAddrSpaces() : FunctionPass(ID) {}
+
+  bool runOnFunction(Function &F) override;
+
+  /// Optimizes load/store instructions. Idx is the index of the pointer operand
+  /// (0 for load, and 1 for store). Returns true if it changes anything.
+  bool optimizeMemoryInstruction(Instruction *I, unsigned Idx);
+  /// Transforms "gep (addrspacecast X), indices" into "addrspacecast (gep X,
+  /// indices)".  This reordering exposes to optimizeMemoryInstruction more
+  /// optimization opportunities on loads and stores. Returns true if it changes
+  /// the program.
+  bool hoistAddrSpaceCastFromGEP(GEPOperator *GEP);
+};
+}
+
+char NVPTXFavorNonGenericAddrSpaces::ID = 0;
+
+namespace llvm {
+void initializeNVPTXFavorNonGenericAddrSpacesPass(PassRegistry &);
+}
+INITIALIZE_PASS(NVPTXFavorNonGenericAddrSpaces, "nvptx-favor-non-generic",
+                "Remove unnecessary non-generic-to-generic addrspacecasts",
+                false, false)
+
+// Decides whether removing Cast is valid and beneficial. Cast can be an
+// instruction or a constant expression.
+static bool IsEliminableAddrSpaceCast(Operator *Cast) {
+  // Returns false if not even an addrspacecast.
+  if (Cast->getOpcode() != Instruction::AddrSpaceCast)
+    return false;
+
+  Value *Src = Cast->getOperand(0);
+  PointerType *SrcTy = cast<PointerType>(Src->getType());
+  PointerType *DestTy = cast<PointerType>(Cast->getType());
+  // TODO: For now, we only handle the case where the addrspacecast only changes
+  // the address space but not the type. If the type also changes, we could
+  // still get rid of the addrspacecast by adding an extra bitcast, but we
+  // rarely see such scenarios.
+  if (SrcTy->getElementType() != DestTy->getElementType())
+    return false;
+
+  // Checks whether the addrspacecast is from a non-generic address space to the
+  // generic address space.
+  return (SrcTy->getAddressSpace() != AddressSpace::ADDRESS_SPACE_GENERIC &&
+          DestTy->getAddressSpace() == AddressSpace::ADDRESS_SPACE_GENERIC);
+}
+
+bool NVPTXFavorNonGenericAddrSpaces::hoistAddrSpaceCastFromGEP(
+    GEPOperator *GEP) {
+  Operator *Cast = dyn_cast<Operator>(GEP->getPointerOperand());
+  if (!Cast)
+    return false;
+
+  if (!IsEliminableAddrSpaceCast(Cast))
+    return false;
+
+  SmallVector<Value *, 8> Indices(GEP->idx_begin(), GEP->idx_end());
+  if (Instruction *GEPI = dyn_cast<Instruction>(GEP)) {
+    // %1 = gep (addrspacecast X), indices
+    // =>
+    // %0 = gep X, indices
+    // %1 = addrspacecast %0
+    GetElementPtrInst *NewGEPI = GetElementPtrInst::Create(Cast->getOperand(0),
+                                                           Indices,
+                                                           GEP->getName(),
+                                                           GEPI);
+    NewGEPI->setIsInBounds(GEP->isInBounds());
+    GEP->replaceAllUsesWith(
+        new AddrSpaceCastInst(NewGEPI, GEP->getType(), "", GEPI));
+  } else {
+    // GEP is a constant expression.
+    Constant *NewGEPCE = ConstantExpr::getGetElementPtr(
+        cast<Constant>(Cast->getOperand(0)),
+        Indices,
+        GEP->isInBounds());
+    GEP->replaceAllUsesWith(
+        ConstantExpr::getAddrSpaceCast(NewGEPCE, GEP->getType()));
+  }
+
+  return true;
+}
+
+bool NVPTXFavorNonGenericAddrSpaces::optimizeMemoryInstruction(Instruction *MI,
+                                                               unsigned Idx) {
+  // If the pointer operand is a GEP, hoist the addrspacecast if any from the
+  // GEP to expose more optimization opportunites.
+  if (GEPOperator *GEP = dyn_cast<GEPOperator>(MI->getOperand(Idx))) {
+    hoistAddrSpaceCastFromGEP(GEP);
+  }
+
+  // load/store (addrspacecast X) => load/store X if shortcutting the
+  // addrspacecast is valid and can improve performance.
+  //
+  // e.g.,
+  // %1 = addrspacecast float addrspace(3)* %0 to float*
+  // %2 = load float* %1
+  // ->
+  // %2 = load float addrspace(3)* %0
+  //
+  // Note: the addrspacecast can also be a constant expression.
+  if (Operator *Cast = dyn_cast<Operator>(MI->getOperand(Idx))) {
+    if (IsEliminableAddrSpaceCast(Cast)) {
+      MI->setOperand(Idx, Cast->getOperand(0));
+      return true;
+    }
+  }
+
+  return false;
+}
+
+bool NVPTXFavorNonGenericAddrSpaces::runOnFunction(Function &F) {
+  if (DisableFavorNonGeneric)
+    return false;
+
+  bool Changed = false;
+  for (Function::iterator B = F.begin(), BE = F.end(); B != BE; ++B) {
+    for (BasicBlock::iterator I = B->begin(), IE = B->end(); I != IE; ++I) {
+      if (isa<LoadInst>(I)) {
+        // V = load P
+        Changed |= optimizeMemoryInstruction(I, 0);
+      } else if (isa<StoreInst>(I)) {
+        // store V, P
+        Changed |= optimizeMemoryInstruction(I, 1);
+      }
+    }
+  }
+  return Changed;
+}
+
+FunctionPass *llvm::createNVPTXFavorNonGenericAddrSpacesPass() {
+  return new NVPTXFavorNonGenericAddrSpaces();
+}
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXFrameLowering.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXFrameLowering.cpp
index 9030584f06fc..8b088412dbba 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXFrameLowering.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXFrameLowering.cpp
@@ -26,6 +26,10 @@
 
 using namespace llvm;
 
+NVPTXFrameLowering::NVPTXFrameLowering(NVPTXSubtarget &STI)
+    : TargetFrameLowering(TargetFrameLowering::StackGrowsUp, 8, 0),
+      is64bit(STI.is64Bit()) {}
+
 bool NVPTXFrameLowering::hasFP(const MachineFunction &MF) const { return true; }
 
 void NVPTXFrameLowering::emitPrologue(MachineFunction &MF) const {
@@ -43,17 +47,21 @@ void NVPTXFrameLowering::emitPrologue(MachineFunction &MF) const {
     // cvta.local %SP, %SPL;
     if (is64bit) {
       unsigned LocalReg = MRI.createVirtualRegister(&NVPTX::Int64RegsRegClass);
-      MachineInstr *MI = BuildMI(
-          MBB, MBBI, dl, tm.getInstrInfo()->get(NVPTX::cvta_local_yes_64),
-          NVPTX::VRFrame).addReg(LocalReg);
-      BuildMI(MBB, MI, dl, tm.getInstrInfo()->get(NVPTX::MOV_DEPOT_ADDR_64),
+      MachineInstr *MI =
+          BuildMI(MBB, MBBI, dl,
+                  MF.getTarget().getInstrInfo()->get(NVPTX::cvta_local_yes_64),
+                  NVPTX::VRFrame).addReg(LocalReg);
+      BuildMI(MBB, MI, dl,
+              MF.getTarget().getInstrInfo()->get(NVPTX::MOV_DEPOT_ADDR_64),
               LocalReg).addImm(MF.getFunctionNumber());
     } else {
       unsigned LocalReg = MRI.createVirtualRegister(&NVPTX::Int32RegsRegClass);
-      MachineInstr *MI = BuildMI(
-          MBB, MBBI, dl, tm.getInstrInfo()->get(NVPTX::cvta_local_yes),
-          NVPTX::VRFrame).addReg(LocalReg);
-      BuildMI(MBB, MI, dl, tm.getInstrInfo()->get(NVPTX::MOV_DEPOT_ADDR),
+      MachineInstr *MI =
+          BuildMI(MBB, MBBI, dl,
+                  MF.getTarget().getInstrInfo()->get(NVPTX::cvta_local_yes),
+                  NVPTX::VRFrame).addReg(LocalReg);
+      BuildMI(MBB, MI, dl,
+              MF.getTarget().getInstrInfo()->get(NVPTX::MOV_DEPOT_ADDR),
               LocalReg).addImm(MF.getFunctionNumber());
     }
   }
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXFrameLowering.h b/contrib/llvm/lib/Target/NVPTX/NVPTXFrameLowering.h
index 819f1dd3f4be..56fb673de0eb 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXFrameLowering.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXFrameLowering.h
@@ -17,24 +17,20 @@
 #include "llvm/Target/TargetFrameLowering.h"
 
 namespace llvm {
-class NVPTXTargetMachine;
-
+class NVPTXSubtarget;
 class NVPTXFrameLowering : public TargetFrameLowering {
-  NVPTXTargetMachine &tm;
   bool is64bit;
 
 public:
-  explicit NVPTXFrameLowering(NVPTXTargetMachine &_tm, bool _is64bit)
-      : TargetFrameLowering(TargetFrameLowering::StackGrowsUp, 8, 0), tm(_tm),
-        is64bit(_is64bit) {}
+  explicit NVPTXFrameLowering(NVPTXSubtarget &STI);
 
-  virtual bool hasFP(const MachineFunction &MF) const;
-  virtual void emitPrologue(MachineFunction &MF) const;
-  virtual void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
+  bool hasFP(const MachineFunction &MF) const override;
+  void emitPrologue(MachineFunction &MF) const override;
+  void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const override;
 
   void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
+                                  MachineBasicBlock &MBB,
+                                  MachineBasicBlock::iterator I) const override;
 };
 
 } // End llvm namespace
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXGenericToNVVM.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXGenericToNVVM.cpp
index 9fb0dd88b758..faa9fdb424b6 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXGenericToNVVM.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXGenericToNVVM.cpp
@@ -13,20 +13,19 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTX.h"
-#include "NVPTXUtilities.h"
 #include "MCTargetDesc/NVPTXBaseInfo.h"
-
-#include "llvm/PassManager.h"
+#include "NVPTXUtilities.h"
+#include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/CodeGen/ValueTypes.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/Operator.h"
-#include "llvm/ADT/ValueMap.h"
-#include "llvm/CodeGen/MachineFunctionAnalysis.h"
-#include "llvm/CodeGen/ValueTypes.h"
-#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/ValueMap.h"
+#include "llvm/PassManager.h"
 
 using namespace llvm;
 
@@ -41,10 +40,9 @@ public:
 
   GenericToNVVM() : ModulePass(ID) {}
 
-  virtual bool runOnModule(Module &M);
+  bool runOnModule(Module &M) override;
 
-  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
-  }
+  void getAnalysisUsage(AnalysisUsage &AU) const override {}
 
 private:
   Value *getOrInsertCVTA(Module *M, Function *F, GlobalVariable *GV,
@@ -64,7 +62,7 @@ private:
   GVMapTy GVMap;
   ConstantToValueMapTy ConstantToValueMap;
 };
-}
+} // end namespace
 
 char GenericToNVVM::ID = 0;
 
@@ -86,10 +84,11 @@ bool GenericToNVVM::runOnModule(Module &M) {
     GlobalVariable *GV = I++;
     if (GV->getType()->getAddressSpace() == llvm::ADDRESS_SPACE_GENERIC &&
         !llvm::isTexture(*GV) && !llvm::isSurface(*GV) &&
-        !GV->getName().startswith("llvm.")) {
+        !llvm::isSampler(*GV) && !GV->getName().startswith("llvm.")) {
       GlobalVariable *NewGV = new GlobalVariable(
           M, GV->getType()->getElementType(), GV->isConstant(),
-          GV->getLinkage(), GV->hasInitializer() ? GV->getInitializer() : NULL,
+          GV->getLinkage(),
+          GV->hasInitializer() ? GV->getInitializer() : nullptr,
           "", GV, GV->getThreadLocalMode(), llvm::ADDRESS_SPACE_GLOBAL);
       NewGV->copyAttributesFrom(GV);
       GVMap[GV] = NewGV;
@@ -147,10 +146,8 @@ bool GenericToNVVM::runOnModule(Module &M) {
     // variable initializers, as other uses have been already been removed
     // while walking through the instructions in function definitions.
     for (Value::use_iterator UI = GV->use_begin(), UE = GV->use_end();
-         UI != UE;) {
-      Use &U = (UI++).getUse();
-      U.set(BitCastNewGV);
-    }
+         UI != UE;)
+      (UI++)->set(BitCastNewGV);
     std::string Name = GV->getName();
     GV->removeDeadConstantUsers();
     GV->eraseFromParent();
@@ -165,7 +162,7 @@ Value *GenericToNVVM::getOrInsertCVTA(Module *M, Function *F,
                                       GlobalVariable *GV,
                                       IRBuilder<> &Builder) {
   PointerType *GVType = GV->getType();
-  Value *CVTA = NULL;
+  Value *CVTA = nullptr;
 
   // See if the address space conversion requires the operand to be bitcast
   // to i8 addrspace(n)* first.
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp
index 4b8b306a705a..05205fba1aff 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp
@@ -20,16 +20,9 @@
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetIntrinsicInfo.h"
 
-#undef DEBUG_TYPE
-#define DEBUG_TYPE "nvptx-isel"
-
 using namespace llvm;
 
-static cl::opt<int>
-FMAContractLevel("nvptx-fma-level", cl::ZeroOrMore, cl::Hidden,
-                 cl::desc("NVPTX Specific: FMA contraction (0: don't do it"
-                          " 1: do it  2: do it aggressively"),
-                 cl::init(2));
+#define DEBUG_TYPE "nvptx-isel"
 
 static cl::opt<int> UsePrecDivF32(
     "nvptx-prec-divf32", cl::ZeroOrMore, cl::Hidden,
@@ -59,16 +52,6 @@ NVPTXDAGToDAGISel::NVPTXDAGToDAGISel(NVPTXTargetMachine &tm,
                                      CodeGenOpt::Level OptLevel)
     : SelectionDAGISel(tm, OptLevel),
       Subtarget(tm.getSubtarget<NVPTXSubtarget>()) {
-
-  doFMAF32 = (OptLevel > 0) && Subtarget.hasFMAF32() && (FMAContractLevel >= 1);
-  doFMAF64 = (OptLevel > 0) && Subtarget.hasFMAF64() && (FMAContractLevel >= 1);
-  doFMAF32AGG =
-      (OptLevel > 0) && Subtarget.hasFMAF32() && (FMAContractLevel == 2);
-  doFMAF64AGG =
-      (OptLevel > 0) && Subtarget.hasFMAF64() && (FMAContractLevel == 2);
-
-  allowFMA = (FMAContractLevel >= 1);
-
   doMulWide = (OptLevel > 0);
 }
 
@@ -114,16 +97,21 @@ bool NVPTXDAGToDAGISel::useF32FTZ() const {
   }
 }
 
+bool NVPTXDAGToDAGISel::allowFMA() const {
+  const NVPTXTargetLowering *TL = Subtarget.getTargetLowering();
+  return TL->allowFMA(*MF, OptLevel);
+}
+
 /// Select - Select instructions not customized! Used for
 /// expanded, promoted and normal instructions.
 SDNode *NVPTXDAGToDAGISel::Select(SDNode *N) {
 
   if (N->isMachineOpcode()) {
     N->setNodeId(-1);
-    return NULL; // Already selected.
+    return nullptr; // Already selected.
   }
 
-  SDNode *ResNode = NULL;
+  SDNode *ResNode = nullptr;
   switch (N->getOpcode()) {
   case ISD::LOAD:
     ResNode = SelectLoad(N);
@@ -139,7 +127,7 @@ SDNode *NVPTXDAGToDAGISel::Select(SDNode *N) {
   case NVPTXISD::LDGV4:
   case NVPTXISD::LDUV2:
   case NVPTXISD::LDUV4:
-    ResNode = SelectLDGLDUVector(N);
+    ResNode = SelectLDGLDU(N);
     break;
   case NVPTXISD::StoreV2:
   case NVPTXISD::StoreV4:
@@ -162,6 +150,358 @@ SDNode *NVPTXDAGToDAGISel::Select(SDNode *N) {
   case NVPTXISD::StoreParamU32:
     ResNode = SelectStoreParam(N);
     break;
+  case ISD::INTRINSIC_WO_CHAIN:
+    ResNode = SelectIntrinsicNoChain(N);
+    break;
+  case ISD::INTRINSIC_W_CHAIN:
+    ResNode = SelectIntrinsicChain(N);
+    break;
+  case NVPTXISD::Tex1DFloatS32:
+  case NVPTXISD::Tex1DFloatFloat:
+  case NVPTXISD::Tex1DFloatFloatLevel:
+  case NVPTXISD::Tex1DFloatFloatGrad:
+  case NVPTXISD::Tex1DS32S32:
+  case NVPTXISD::Tex1DS32Float:
+  case NVPTXISD::Tex1DS32FloatLevel:
+  case NVPTXISD::Tex1DS32FloatGrad:
+  case NVPTXISD::Tex1DU32S32:
+  case NVPTXISD::Tex1DU32Float:
+  case NVPTXISD::Tex1DU32FloatLevel:
+  case NVPTXISD::Tex1DU32FloatGrad:
+  case NVPTXISD::Tex1DArrayFloatS32:
+  case NVPTXISD::Tex1DArrayFloatFloat:
+  case NVPTXISD::Tex1DArrayFloatFloatLevel:
+  case NVPTXISD::Tex1DArrayFloatFloatGrad:
+  case NVPTXISD::Tex1DArrayS32S32:
+  case NVPTXISD::Tex1DArrayS32Float:
+  case NVPTXISD::Tex1DArrayS32FloatLevel:
+  case NVPTXISD::Tex1DArrayS32FloatGrad:
+  case NVPTXISD::Tex1DArrayU32S32:
+  case NVPTXISD::Tex1DArrayU32Float:
+  case NVPTXISD::Tex1DArrayU32FloatLevel:
+  case NVPTXISD::Tex1DArrayU32FloatGrad:
+  case NVPTXISD::Tex2DFloatS32:
+  case NVPTXISD::Tex2DFloatFloat:
+  case NVPTXISD::Tex2DFloatFloatLevel:
+  case NVPTXISD::Tex2DFloatFloatGrad:
+  case NVPTXISD::Tex2DS32S32:
+  case NVPTXISD::Tex2DS32Float:
+  case NVPTXISD::Tex2DS32FloatLevel:
+  case NVPTXISD::Tex2DS32FloatGrad:
+  case NVPTXISD::Tex2DU32S32:
+  case NVPTXISD::Tex2DU32Float:
+  case NVPTXISD::Tex2DU32FloatLevel:
+  case NVPTXISD::Tex2DU32FloatGrad:
+  case NVPTXISD::Tex2DArrayFloatS32:
+  case NVPTXISD::Tex2DArrayFloatFloat:
+  case NVPTXISD::Tex2DArrayFloatFloatLevel:
+  case NVPTXISD::Tex2DArrayFloatFloatGrad:
+  case NVPTXISD::Tex2DArrayS32S32:
+  case NVPTXISD::Tex2DArrayS32Float:
+  case NVPTXISD::Tex2DArrayS32FloatLevel:
+  case NVPTXISD::Tex2DArrayS32FloatGrad:
+  case NVPTXISD::Tex2DArrayU32S32:
+  case NVPTXISD::Tex2DArrayU32Float:
+  case NVPTXISD::Tex2DArrayU32FloatLevel:
+  case NVPTXISD::Tex2DArrayU32FloatGrad:
+  case NVPTXISD::Tex3DFloatS32:
+  case NVPTXISD::Tex3DFloatFloat:
+  case NVPTXISD::Tex3DFloatFloatLevel:
+  case NVPTXISD::Tex3DFloatFloatGrad:
+  case NVPTXISD::Tex3DS32S32:
+  case NVPTXISD::Tex3DS32Float:
+  case NVPTXISD::Tex3DS32FloatLevel:
+  case NVPTXISD::Tex3DS32FloatGrad:
+  case NVPTXISD::Tex3DU32S32:
+  case NVPTXISD::Tex3DU32Float:
+  case NVPTXISD::Tex3DU32FloatLevel:
+  case NVPTXISD::Tex3DU32FloatGrad:
+  case NVPTXISD::TexCubeFloatFloat:
+  case NVPTXISD::TexCubeFloatFloatLevel:
+  case NVPTXISD::TexCubeS32Float:
+  case NVPTXISD::TexCubeS32FloatLevel:
+  case NVPTXISD::TexCubeU32Float:
+  case NVPTXISD::TexCubeU32FloatLevel:
+  case NVPTXISD::TexCubeArrayFloatFloat:
+  case NVPTXISD::TexCubeArrayFloatFloatLevel:
+  case NVPTXISD::TexCubeArrayS32Float:
+  case NVPTXISD::TexCubeArrayS32FloatLevel:
+  case NVPTXISD::TexCubeArrayU32Float:
+  case NVPTXISD::TexCubeArrayU32FloatLevel:
+  case NVPTXISD::Tld4R2DFloatFloat:
+  case NVPTXISD::Tld4G2DFloatFloat:
+  case NVPTXISD::Tld4B2DFloatFloat:
+  case NVPTXISD::Tld4A2DFloatFloat:
+  case NVPTXISD::Tld4R2DS64Float:
+  case NVPTXISD::Tld4G2DS64Float:
+  case NVPTXISD::Tld4B2DS64Float:
+  case NVPTXISD::Tld4A2DS64Float:
+  case NVPTXISD::Tld4R2DU64Float:
+  case NVPTXISD::Tld4G2DU64Float:
+  case NVPTXISD::Tld4B2DU64Float:
+  case NVPTXISD::Tld4A2DU64Float:
+  case NVPTXISD::TexUnified1DFloatS32:
+  case NVPTXISD::TexUnified1DFloatFloat:
+  case NVPTXISD::TexUnified1DFloatFloatLevel:
+  case NVPTXISD::TexUnified1DFloatFloatGrad:
+  case NVPTXISD::TexUnified1DS32S32:
+  case NVPTXISD::TexUnified1DS32Float:
+  case NVPTXISD::TexUnified1DS32FloatLevel:
+  case NVPTXISD::TexUnified1DS32FloatGrad:
+  case NVPTXISD::TexUnified1DU32S32:
+  case NVPTXISD::TexUnified1DU32Float:
+  case NVPTXISD::TexUnified1DU32FloatLevel:
+  case NVPTXISD::TexUnified1DU32FloatGrad:
+  case NVPTXISD::TexUnified1DArrayFloatS32:
+  case NVPTXISD::TexUnified1DArrayFloatFloat:
+  case NVPTXISD::TexUnified1DArrayFloatFloatLevel:
+  case NVPTXISD::TexUnified1DArrayFloatFloatGrad:
+  case NVPTXISD::TexUnified1DArrayS32S32:
+  case NVPTXISD::TexUnified1DArrayS32Float:
+  case NVPTXISD::TexUnified1DArrayS32FloatLevel:
+  case NVPTXISD::TexUnified1DArrayS32FloatGrad:
+  case NVPTXISD::TexUnified1DArrayU32S32:
+  case NVPTXISD::TexUnified1DArrayU32Float:
+  case NVPTXISD::TexUnified1DArrayU32FloatLevel:
+  case NVPTXISD::TexUnified1DArrayU32FloatGrad:
+  case NVPTXISD::TexUnified2DFloatS32:
+  case NVPTXISD::TexUnified2DFloatFloat:
+  case NVPTXISD::TexUnified2DFloatFloatLevel:
+  case NVPTXISD::TexUnified2DFloatFloatGrad:
+  case NVPTXISD::TexUnified2DS32S32:
+  case NVPTXISD::TexUnified2DS32Float:
+  case NVPTXISD::TexUnified2DS32FloatLevel:
+  case NVPTXISD::TexUnified2DS32FloatGrad:
+  case NVPTXISD::TexUnified2DU32S32:
+  case NVPTXISD::TexUnified2DU32Float:
+  case NVPTXISD::TexUnified2DU32FloatLevel:
+  case NVPTXISD::TexUnified2DU32FloatGrad:
+  case NVPTXISD::TexUnified2DArrayFloatS32:
+  case NVPTXISD::TexUnified2DArrayFloatFloat:
+  case NVPTXISD::TexUnified2DArrayFloatFloatLevel:
+  case NVPTXISD::TexUnified2DArrayFloatFloatGrad:
+  case NVPTXISD::TexUnified2DArrayS32S32:
+  case NVPTXISD::TexUnified2DArrayS32Float:
+  case NVPTXISD::TexUnified2DArrayS32FloatLevel:
+  case NVPTXISD::TexUnified2DArrayS32FloatGrad:
+  case NVPTXISD::TexUnified2DArrayU32S32:
+  case NVPTXISD::TexUnified2DArrayU32Float:
+  case NVPTXISD::TexUnified2DArrayU32FloatLevel:
+  case NVPTXISD::TexUnified2DArrayU32FloatGrad:
+  case NVPTXISD::TexUnified3DFloatS32:
+  case NVPTXISD::TexUnified3DFloatFloat:
+  case NVPTXISD::TexUnified3DFloatFloatLevel:
+  case NVPTXISD::TexUnified3DFloatFloatGrad:
+  case NVPTXISD::TexUnified3DS32S32:
+  case NVPTXISD::TexUnified3DS32Float:
+  case NVPTXISD::TexUnified3DS32FloatLevel:
+  case NVPTXISD::TexUnified3DS32FloatGrad:
+  case NVPTXISD::TexUnified3DU32S32:
+  case NVPTXISD::TexUnified3DU32Float:
+  case NVPTXISD::TexUnified3DU32FloatLevel:
+  case NVPTXISD::TexUnified3DU32FloatGrad:
+  case NVPTXISD::TexUnifiedCubeFloatFloat:
+  case NVPTXISD::TexUnifiedCubeFloatFloatLevel:
+  case NVPTXISD::TexUnifiedCubeS32Float:
+  case NVPTXISD::TexUnifiedCubeS32FloatLevel:
+  case NVPTXISD::TexUnifiedCubeU32Float:
+  case NVPTXISD::TexUnifiedCubeU32FloatLevel:
+  case NVPTXISD::TexUnifiedCubeArrayFloatFloat:
+  case NVPTXISD::TexUnifiedCubeArrayFloatFloatLevel:
+  case NVPTXISD::TexUnifiedCubeArrayS32Float:
+  case NVPTXISD::TexUnifiedCubeArrayS32FloatLevel:
+  case NVPTXISD::TexUnifiedCubeArrayU32Float:
+  case NVPTXISD::TexUnifiedCubeArrayU32FloatLevel:
+  case NVPTXISD::Tld4UnifiedR2DFloatFloat:
+  case NVPTXISD::Tld4UnifiedG2DFloatFloat:
+  case NVPTXISD::Tld4UnifiedB2DFloatFloat:
+  case NVPTXISD::Tld4UnifiedA2DFloatFloat:
+  case NVPTXISD::Tld4UnifiedR2DS64Float:
+  case NVPTXISD::Tld4UnifiedG2DS64Float:
+  case NVPTXISD::Tld4UnifiedB2DS64Float:
+  case NVPTXISD::Tld4UnifiedA2DS64Float:
+  case NVPTXISD::Tld4UnifiedR2DU64Float:
+  case NVPTXISD::Tld4UnifiedG2DU64Float:
+  case NVPTXISD::Tld4UnifiedB2DU64Float:
+  case NVPTXISD::Tld4UnifiedA2DU64Float:
+    ResNode = SelectTextureIntrinsic(N);
+    break;
+  case NVPTXISD::Suld1DI8Clamp:
+  case NVPTXISD::Suld1DI16Clamp:
+  case NVPTXISD::Suld1DI32Clamp:
+  case NVPTXISD::Suld1DI64Clamp:
+  case NVPTXISD::Suld1DV2I8Clamp:
+  case NVPTXISD::Suld1DV2I16Clamp:
+  case NVPTXISD::Suld1DV2I32Clamp:
+  case NVPTXISD::Suld1DV2I64Clamp:
+  case NVPTXISD::Suld1DV4I8Clamp:
+  case NVPTXISD::Suld1DV4I16Clamp:
+  case NVPTXISD::Suld1DV4I32Clamp:
+  case NVPTXISD::Suld1DArrayI8Clamp:
+  case NVPTXISD::Suld1DArrayI16Clamp:
+  case NVPTXISD::Suld1DArrayI32Clamp:
+  case NVPTXISD::Suld1DArrayI64Clamp:
+  case NVPTXISD::Suld1DArrayV2I8Clamp:
+  case NVPTXISD::Suld1DArrayV2I16Clamp:
+  case NVPTXISD::Suld1DArrayV2I32Clamp:
+  case NVPTXISD::Suld1DArrayV2I64Clamp:
+  case NVPTXISD::Suld1DArrayV4I8Clamp:
+  case NVPTXISD::Suld1DArrayV4I16Clamp:
+  case NVPTXISD::Suld1DArrayV4I32Clamp:
+  case NVPTXISD::Suld2DI8Clamp:
+  case NVPTXISD::Suld2DI16Clamp:
+  case NVPTXISD::Suld2DI32Clamp:
+  case NVPTXISD::Suld2DI64Clamp:
+  case NVPTXISD::Suld2DV2I8Clamp:
+  case NVPTXISD::Suld2DV2I16Clamp:
+  case NVPTXISD::Suld2DV2I32Clamp:
+  case NVPTXISD::Suld2DV2I64Clamp:
+  case NVPTXISD::Suld2DV4I8Clamp:
+  case NVPTXISD::Suld2DV4I16Clamp:
+  case NVPTXISD::Suld2DV4I32Clamp:
+  case NVPTXISD::Suld2DArrayI8Clamp:
+  case NVPTXISD::Suld2DArrayI16Clamp:
+  case NVPTXISD::Suld2DArrayI32Clamp:
+  case NVPTXISD::Suld2DArrayI64Clamp:
+  case NVPTXISD::Suld2DArrayV2I8Clamp:
+  case NVPTXISD::Suld2DArrayV2I16Clamp:
+  case NVPTXISD::Suld2DArrayV2I32Clamp:
+  case NVPTXISD::Suld2DArrayV2I64Clamp:
+  case NVPTXISD::Suld2DArrayV4I8Clamp:
+  case NVPTXISD::Suld2DArrayV4I16Clamp:
+  case NVPTXISD::Suld2DArrayV4I32Clamp:
+  case NVPTXISD::Suld3DI8Clamp:
+  case NVPTXISD::Suld3DI16Clamp:
+  case NVPTXISD::Suld3DI32Clamp:
+  case NVPTXISD::Suld3DI64Clamp:
+  case NVPTXISD::Suld3DV2I8Clamp:
+  case NVPTXISD::Suld3DV2I16Clamp:
+  case NVPTXISD::Suld3DV2I32Clamp:
+  case NVPTXISD::Suld3DV2I64Clamp:
+  case NVPTXISD::Suld3DV4I8Clamp:
+  case NVPTXISD::Suld3DV4I16Clamp:
+  case NVPTXISD::Suld3DV4I32Clamp:
+  case NVPTXISD::Suld1DI8Trap:
+  case NVPTXISD::Suld1DI16Trap:
+  case NVPTXISD::Suld1DI32Trap:
+  case NVPTXISD::Suld1DI64Trap:
+  case NVPTXISD::Suld1DV2I8Trap:
+  case NVPTXISD::Suld1DV2I16Trap:
+  case NVPTXISD::Suld1DV2I32Trap:
+  case NVPTXISD::Suld1DV2I64Trap:
+  case NVPTXISD::Suld1DV4I8Trap:
+  case NVPTXISD::Suld1DV4I16Trap:
+  case NVPTXISD::Suld1DV4I32Trap:
+  case NVPTXISD::Suld1DArrayI8Trap:
+  case NVPTXISD::Suld1DArrayI16Trap:
+  case NVPTXISD::Suld1DArrayI32Trap:
+  case NVPTXISD::Suld1DArrayI64Trap:
+  case NVPTXISD::Suld1DArrayV2I8Trap:
+  case NVPTXISD::Suld1DArrayV2I16Trap:
+  case NVPTXISD::Suld1DArrayV2I32Trap:
+  case NVPTXISD::Suld1DArrayV2I64Trap:
+  case NVPTXISD::Suld1DArrayV4I8Trap:
+  case NVPTXISD::Suld1DArrayV4I16Trap:
+  case NVPTXISD::Suld1DArrayV4I32Trap:
+  case NVPTXISD::Suld2DI8Trap:
+  case NVPTXISD::Suld2DI16Trap:
+  case NVPTXISD::Suld2DI32Trap:
+  case NVPTXISD::Suld2DI64Trap:
+  case NVPTXISD::Suld2DV2I8Trap:
+  case NVPTXISD::Suld2DV2I16Trap:
+  case NVPTXISD::Suld2DV2I32Trap:
+  case NVPTXISD::Suld2DV2I64Trap:
+  case NVPTXISD::Suld2DV4I8Trap:
+  case NVPTXISD::Suld2DV4I16Trap:
+  case NVPTXISD::Suld2DV4I32Trap:
+  case NVPTXISD::Suld2DArrayI8Trap:
+  case NVPTXISD::Suld2DArrayI16Trap:
+  case NVPTXISD::Suld2DArrayI32Trap:
+  case NVPTXISD::Suld2DArrayI64Trap:
+  case NVPTXISD::Suld2DArrayV2I8Trap:
+  case NVPTXISD::Suld2DArrayV2I16Trap:
+  case NVPTXISD::Suld2DArrayV2I32Trap:
+  case NVPTXISD::Suld2DArrayV2I64Trap:
+  case NVPTXISD::Suld2DArrayV4I8Trap:
+  case NVPTXISD::Suld2DArrayV4I16Trap:
+  case NVPTXISD::Suld2DArrayV4I32Trap:
+  case NVPTXISD::Suld3DI8Trap:
+  case NVPTXISD::Suld3DI16Trap:
+  case NVPTXISD::Suld3DI32Trap:
+  case NVPTXISD::Suld3DI64Trap:
+  case NVPTXISD::Suld3DV2I8Trap:
+  case NVPTXISD::Suld3DV2I16Trap:
+  case NVPTXISD::Suld3DV2I32Trap:
+  case NVPTXISD::Suld3DV2I64Trap:
+  case NVPTXISD::Suld3DV4I8Trap:
+  case NVPTXISD::Suld3DV4I16Trap:
+  case NVPTXISD::Suld3DV4I32Trap:
+  case NVPTXISD::Suld1DI8Zero:
+  case NVPTXISD::Suld1DI16Zero:
+  case NVPTXISD::Suld1DI32Zero:
+  case NVPTXISD::Suld1DI64Zero:
+  case NVPTXISD::Suld1DV2I8Zero:
+  case NVPTXISD::Suld1DV2I16Zero:
+  case NVPTXISD::Suld1DV2I32Zero:
+  case NVPTXISD::Suld1DV2I64Zero:
+  case NVPTXISD::Suld1DV4I8Zero:
+  case NVPTXISD::Suld1DV4I16Zero:
+  case NVPTXISD::Suld1DV4I32Zero:
+  case NVPTXISD::Suld1DArrayI8Zero:
+  case NVPTXISD::Suld1DArrayI16Zero:
+  case NVPTXISD::Suld1DArrayI32Zero:
+  case NVPTXISD::Suld1DArrayI64Zero:
+  case NVPTXISD::Suld1DArrayV2I8Zero:
+  case NVPTXISD::Suld1DArrayV2I16Zero:
+  case NVPTXISD::Suld1DArrayV2I32Zero:
+  case NVPTXISD::Suld1DArrayV2I64Zero:
+  case NVPTXISD::Suld1DArrayV4I8Zero:
+  case NVPTXISD::Suld1DArrayV4I16Zero:
+  case NVPTXISD::Suld1DArrayV4I32Zero:
+  case NVPTXISD::Suld2DI8Zero:
+  case NVPTXISD::Suld2DI16Zero:
+  case NVPTXISD::Suld2DI32Zero:
+  case NVPTXISD::Suld2DI64Zero:
+  case NVPTXISD::Suld2DV2I8Zero:
+  case NVPTXISD::Suld2DV2I16Zero:
+  case NVPTXISD::Suld2DV2I32Zero:
+  case NVPTXISD::Suld2DV2I64Zero:
+  case NVPTXISD::Suld2DV4I8Zero:
+  case NVPTXISD::Suld2DV4I16Zero:
+  case NVPTXISD::Suld2DV4I32Zero:
+  case NVPTXISD::Suld2DArrayI8Zero:
+  case NVPTXISD::Suld2DArrayI16Zero:
+  case NVPTXISD::Suld2DArrayI32Zero:
+  case NVPTXISD::Suld2DArrayI64Zero:
+  case NVPTXISD::Suld2DArrayV2I8Zero:
+  case NVPTXISD::Suld2DArrayV2I16Zero:
+  case NVPTXISD::Suld2DArrayV2I32Zero:
+  case NVPTXISD::Suld2DArrayV2I64Zero:
+  case NVPTXISD::Suld2DArrayV4I8Zero:
+  case NVPTXISD::Suld2DArrayV4I16Zero:
+  case NVPTXISD::Suld2DArrayV4I32Zero:
+  case NVPTXISD::Suld3DI8Zero:
+  case NVPTXISD::Suld3DI16Zero:
+  case NVPTXISD::Suld3DI32Zero:
+  case NVPTXISD::Suld3DI64Zero:
+  case NVPTXISD::Suld3DV2I8Zero:
+  case NVPTXISD::Suld3DV2I16Zero:
+  case NVPTXISD::Suld3DV2I32Zero:
+  case NVPTXISD::Suld3DV2I64Zero:
+  case NVPTXISD::Suld3DV4I8Zero:
+  case NVPTXISD::Suld3DV4I16Zero:
+  case NVPTXISD::Suld3DV4I32Zero:
+    ResNode = SelectSurfaceIntrinsic(N);
+    break;
+  case ISD::AND:
+  case ISD::SRA:
+  case ISD::SRL:
+    // Try to select BFE
+    ResNode = SelectBFE(N);
+    break;
+  case ISD::ADDRSPACECAST:
+    ResNode = SelectAddrSpaceCast(N);
+    break;
   default:
     break;
   }
@@ -170,9 +510,24 @@ SDNode *NVPTXDAGToDAGISel::Select(SDNode *N) {
   return SelectCode(N);
 }
 
+SDNode *NVPTXDAGToDAGISel::SelectIntrinsicChain(SDNode *N) {
+  unsigned IID = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
+  switch (IID) {
+  default:
+    return NULL;
+  case Intrinsic::nvvm_ldg_global_f:
+  case Intrinsic::nvvm_ldg_global_i:
+  case Intrinsic::nvvm_ldg_global_p:
+  case Intrinsic::nvvm_ldu_global_f:
+  case Intrinsic::nvvm_ldu_global_i:
+  case Intrinsic::nvvm_ldu_global_p:
+    return SelectLDGLDU(N);
+  }
+}
+
 static unsigned int getCodeAddrSpace(MemSDNode *N,
                                      const NVPTXSubtarget &Subtarget) {
-  const Value *Src = N->getSrcValue();
+  const Value *Src = N->getMemOperand()->getValue();
 
   if (!Src)
     return NVPTX::PTXLdStInstCode::GENERIC;
@@ -191,18 +546,96 @@ static unsigned int getCodeAddrSpace(MemSDNode *N,
   return NVPTX::PTXLdStInstCode::GENERIC;
 }
 
+SDNode *NVPTXDAGToDAGISel::SelectIntrinsicNoChain(SDNode *N) {
+  unsigned IID = cast<ConstantSDNode>(N->getOperand(0))->getZExtValue();
+  switch (IID) {
+  default:
+    return nullptr;
+  case Intrinsic::nvvm_texsurf_handle_internal:
+    return SelectTexSurfHandle(N);
+  }
+}
+
+SDNode *NVPTXDAGToDAGISel::SelectTexSurfHandle(SDNode *N) {
+  // Op 0 is the intrinsic ID
+  SDValue Wrapper = N->getOperand(1);
+  SDValue GlobalVal = Wrapper.getOperand(0);
+  return CurDAG->getMachineNode(NVPTX::texsurf_handles, SDLoc(N), MVT::i64,
+                                GlobalVal);
+}
+
+SDNode *NVPTXDAGToDAGISel::SelectAddrSpaceCast(SDNode *N) {
+  SDValue Src = N->getOperand(0);
+  AddrSpaceCastSDNode *CastN = cast<AddrSpaceCastSDNode>(N);
+  unsigned SrcAddrSpace = CastN->getSrcAddressSpace();
+  unsigned DstAddrSpace = CastN->getDestAddressSpace();
+
+  assert(SrcAddrSpace != DstAddrSpace &&
+         "addrspacecast must be between different address spaces");
+
+  if (DstAddrSpace == ADDRESS_SPACE_GENERIC) {
+    // Specific to generic
+    unsigned Opc;
+    switch (SrcAddrSpace) {
+    default: report_fatal_error("Bad address space in addrspacecast");
+    case ADDRESS_SPACE_GLOBAL:
+      Opc = Subtarget.is64Bit() ? NVPTX::cvta_global_yes_64
+                                : NVPTX::cvta_global_yes;
+      break;
+    case ADDRESS_SPACE_SHARED:
+      Opc = Subtarget.is64Bit() ? NVPTX::cvta_shared_yes_64
+                                : NVPTX::cvta_shared_yes;
+      break;
+    case ADDRESS_SPACE_CONST:
+      Opc = Subtarget.is64Bit() ? NVPTX::cvta_const_yes_64
+                                : NVPTX::cvta_const_yes;
+      break;
+    case ADDRESS_SPACE_LOCAL:
+      Opc = Subtarget.is64Bit() ? NVPTX::cvta_local_yes_64
+                                : NVPTX::cvta_local_yes;
+      break;
+    }
+    return CurDAG->getMachineNode(Opc, SDLoc(N), N->getValueType(0), Src);
+  } else {
+    // Generic to specific
+    if (SrcAddrSpace != 0)
+      report_fatal_error("Cannot cast between two non-generic address spaces");
+    unsigned Opc;
+    switch (DstAddrSpace) {
+    default: report_fatal_error("Bad address space in addrspacecast");
+    case ADDRESS_SPACE_GLOBAL:
+      Opc = Subtarget.is64Bit() ? NVPTX::cvta_to_global_yes_64
+                                : NVPTX::cvta_to_global_yes;
+      break;
+    case ADDRESS_SPACE_SHARED:
+      Opc = Subtarget.is64Bit() ? NVPTX::cvta_to_shared_yes_64
+                                : NVPTX::cvta_to_shared_yes;
+      break;
+    case ADDRESS_SPACE_CONST:
+      Opc = Subtarget.is64Bit() ? NVPTX::cvta_to_const_yes_64
+                                : NVPTX::cvta_to_const_yes;
+      break;
+    case ADDRESS_SPACE_LOCAL:
+      Opc = Subtarget.is64Bit() ? NVPTX::cvta_to_local_yes_64
+                                : NVPTX::cvta_to_local_yes;
+      break;
+    }
+    return CurDAG->getMachineNode(Opc, SDLoc(N), N->getValueType(0), Src);
+  }
+}
+
 SDNode *NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
   SDLoc dl(N);
   LoadSDNode *LD = cast<LoadSDNode>(N);
   EVT LoadedVT = LD->getMemoryVT();
-  SDNode *NVPTXLD = NULL;
+  SDNode *NVPTXLD = nullptr;
 
   // do not support pre/post inc/dec
   if (LD->isIndexed())
-    return NULL;
+    return nullptr;
 
   if (!LoadedVT.isSimple())
-    return NULL;
+    return nullptr;
 
   // Address Space Setting
   unsigned int codeAddrSpace = getCodeAddrSpace(LD, Subtarget);
@@ -225,7 +658,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
     else if (num == 4)
       vecType = NVPTX::PTXLdStInstCode::V4;
     else
-      return NULL;
+      return nullptr;
   }
 
   // Type Setting: fromType + fromTypeWidth
@@ -274,7 +707,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
       Opcode = NVPTX::LD_f64_avar;
       break;
     default:
-      return NULL;
+      return nullptr;
     }
     SDValue Ops[] = { getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
                       getI32Imm(vecType), getI32Imm(fromType),
@@ -303,7 +736,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
       Opcode = NVPTX::LD_f64_asi;
       break;
     default:
-      return NULL;
+      return nullptr;
     }
     SDValue Ops[] = { getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
                       getI32Imm(vecType), getI32Imm(fromType),
@@ -333,7 +766,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
         Opcode = NVPTX::LD_f64_ari_64;
         break;
       default:
-        return NULL;
+        return nullptr;
       }
     } else {
       switch (TargetVT) {
@@ -356,7 +789,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
         Opcode = NVPTX::LD_f64_ari;
         break;
       default:
-        return NULL;
+        return nullptr;
       }
     }
     SDValue Ops[] = { getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
@@ -385,7 +818,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
         Opcode = NVPTX::LD_f64_areg_64;
         break;
       default:
-        return NULL;
+        return nullptr;
       }
     } else {
       switch (TargetVT) {
@@ -408,7 +841,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
         Opcode = NVPTX::LD_f64_areg;
         break;
       default:
-        return NULL;
+        return nullptr;
       }
     }
     SDValue Ops[] = { getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
@@ -417,7 +850,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
     NVPTXLD = CurDAG->getMachineNode(Opcode, dl, TargetVT, MVT::Other, Ops);
   }
 
-  if (NVPTXLD != NULL) {
+  if (NVPTXLD) {
     MachineSDNode::mmo_iterator MemRefs0 = MF->allocateMemRefsArray(1);
     MemRefs0[0] = cast<MemSDNode>(N)->getMemOperand();
     cast<MachineSDNode>(NVPTXLD)->setMemRefs(MemRefs0, MemRefs0 + 1);
@@ -438,7 +871,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
   EVT LoadedVT = MemSD->getMemoryVT();
 
   if (!LoadedVT.isSimple())
-    return NULL;
+    return nullptr;
 
   // Address Space Setting
   unsigned int CodeAddrSpace = getCodeAddrSpace(MemSD, Subtarget);
@@ -484,7 +917,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
     VecType = NVPTX::PTXLdStInstCode::V4;
     break;
   default:
-    return NULL;
+    return nullptr;
   }
 
   EVT EltVT = N->getValueType(0);
@@ -492,11 +925,11 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
   if (SelectDirectAddr(Op1, Addr)) {
     switch (N->getOpcode()) {
     default:
-      return NULL;
+      return nullptr;
     case NVPTXISD::LoadV2:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::LDV_i8_v2_avar;
         break;
@@ -520,7 +953,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
     case NVPTXISD::LoadV4:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::LDV_i8_v4_avar;
         break;
@@ -546,11 +979,11 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
                  : SelectADDRsi(Op1.getNode(), Op1, Base, Offset)) {
     switch (N->getOpcode()) {
     default:
-      return NULL;
+      return nullptr;
     case NVPTXISD::LoadV2:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::LDV_i8_v2_asi;
         break;
@@ -574,7 +1007,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
     case NVPTXISD::LoadV4:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::LDV_i8_v4_asi;
         break;
@@ -601,11 +1034,11 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
     if (Subtarget.is64Bit()) {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
       case NVPTXISD::LoadV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::LDV_i8_v2_ari_64;
           break;
@@ -629,7 +1062,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
       case NVPTXISD::LoadV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::LDV_i8_v4_ari_64;
           break;
@@ -648,11 +1081,11 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
     } else {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
       case NVPTXISD::LoadV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::LDV_i8_v2_ari;
           break;
@@ -676,7 +1109,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
       case NVPTXISD::LoadV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::LDV_i8_v4_ari;
           break;
@@ -703,11 +1136,11 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
     if (Subtarget.is64Bit()) {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
       case NVPTXISD::LoadV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::LDV_i8_v2_areg_64;
           break;
@@ -731,7 +1164,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
       case NVPTXISD::LoadV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::LDV_i8_v4_areg_64;
           break;
@@ -750,11 +1183,11 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
     } else {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
       case NVPTXISD::LoadV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::LDV_i8_v2_areg;
           break;
@@ -778,7 +1211,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
       case NVPTXISD::LoadV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::LDV_i8_v4_areg;
           break;
@@ -809,26 +1242,105 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
   return LD;
 }
 
-SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
+SDNode *NVPTXDAGToDAGISel::SelectLDGLDU(SDNode *N) {
 
   SDValue Chain = N->getOperand(0);
-  SDValue Op1 = N->getOperand(1);
+  SDValue Op1;
+  MemSDNode *Mem;
+  bool IsLDG = true;
+
+  // If this is an LDG intrinsic, the address is the third operand. Its its an
+  // LDG/LDU SD node (from custom vector handling), then its the second operand
+  if (N->getOpcode() == ISD::INTRINSIC_W_CHAIN) {
+    Op1 = N->getOperand(2);
+    Mem = cast<MemIntrinsicSDNode>(N);
+    unsigned IID = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
+    switch (IID) {
+    default:
+      return NULL;
+    case Intrinsic::nvvm_ldg_global_f:
+    case Intrinsic::nvvm_ldg_global_i:
+    case Intrinsic::nvvm_ldg_global_p:
+      IsLDG = true;
+      break;
+    case Intrinsic::nvvm_ldu_global_f:
+    case Intrinsic::nvvm_ldu_global_i:
+    case Intrinsic::nvvm_ldu_global_p:
+      IsLDG = false;
+      break;
+    }
+  } else {
+    Op1 = N->getOperand(1);
+    Mem = cast<MemSDNode>(N);
+  }
+
   unsigned Opcode;
   SDLoc DL(N);
   SDNode *LD;
-  MemSDNode *Mem = cast<MemSDNode>(N);
   SDValue Base, Offset, Addr;
 
-  EVT EltVT = Mem->getMemoryVT().getVectorElementType();
+  EVT EltVT = Mem->getMemoryVT();
+  if (EltVT.isVector()) {
+    EltVT = EltVT.getVectorElementType();
+  }
 
   if (SelectDirectAddr(Op1, Addr)) {
     switch (N->getOpcode()) {
     default:
-      return NULL;
+      return nullptr;
+    case ISD::INTRINSIC_W_CHAIN:
+      if (IsLDG) {
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return nullptr;
+        case MVT::i8:
+          Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i8avar;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i16avar;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i32avar;
+          break;
+        case MVT::i64:
+          Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i64avar;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::INT_PTX_LDG_GLOBAL_f32avar;
+          break;
+        case MVT::f64:
+          Opcode = NVPTX::INT_PTX_LDG_GLOBAL_f64avar;
+          break;
+        }
+      } else {
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return nullptr;
+        case MVT::i8:
+          Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i8avar;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i16avar;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i32avar;
+          break;
+        case MVT::i64:
+          Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i64avar;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::INT_PTX_LDU_GLOBAL_f32avar;
+          break;
+        case MVT::f64:
+          Opcode = NVPTX::INT_PTX_LDU_GLOBAL_f64avar;
+          break;
+        }
+      }
+      break;
     case NVPTXISD::LDGV2:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::INT_PTX_LDG_G_v2i8_ELE_avar;
         break;
@@ -852,7 +1364,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
     case NVPTXISD::LDUV2:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::INT_PTX_LDU_G_v2i8_ELE_avar;
         break;
@@ -876,7 +1388,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
     case NVPTXISD::LDGV4:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::INT_PTX_LDG_G_v4i8_ELE_avar;
         break;
@@ -894,7 +1406,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
     case NVPTXISD::LDUV4:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::INT_PTX_LDU_G_v4i8_ELE_avar;
         break;
@@ -912,19 +1424,67 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
     }
 
     SDValue Ops[] = { Addr, Chain };
-    LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(),
-                                ArrayRef<SDValue>(Ops, 2));
+    LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(), Ops);
   } else if (Subtarget.is64Bit()
                  ? SelectADDRri64(Op1.getNode(), Op1, Base, Offset)
                  : SelectADDRri(Op1.getNode(), Op1, Base, Offset)) {
     if (Subtarget.is64Bit()) {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
+      case ISD::INTRINSIC_W_CHAIN:
+        if (IsLDG) {
+          switch (EltVT.getSimpleVT().SimpleTy) {
+          default:
+            return nullptr;
+          case MVT::i8:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i8ari64;
+            break;
+          case MVT::i16:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i16ari64;
+            break;
+          case MVT::i32:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i32ari64;
+            break;
+          case MVT::i64:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i64ari64;
+            break;
+          case MVT::f32:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_f32ari64;
+            break;
+          case MVT::f64:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_f64ari64;
+            break;
+          }
+        } else {
+          switch (EltVT.getSimpleVT().SimpleTy) {
+          default:
+            return nullptr;
+          case MVT::i8:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i8ari64;
+            break;
+          case MVT::i16:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i16ari64;
+            break;
+          case MVT::i32:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i32ari64;
+            break;
+          case MVT::i64:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i64ari64;
+            break;
+          case MVT::f32:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_f32ari64;
+            break;
+          case MVT::f64:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_f64ari64;
+            break;
+          }
+        }
+        break;
       case NVPTXISD::LDGV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDG_G_v2i8_ELE_ari64;
           break;
@@ -948,7 +1508,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDUV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDU_G_v2i8_ELE_ari64;
           break;
@@ -972,7 +1532,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDGV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDG_G_v4i8_ELE_ari64;
           break;
@@ -990,7 +1550,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDUV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDU_G_v4i8_ELE_ari64;
           break;
@@ -1009,11 +1569,60 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
     } else {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
+      case ISD::INTRINSIC_W_CHAIN:
+        if (IsLDG) {
+          switch (EltVT.getSimpleVT().SimpleTy) {
+          default:
+            return nullptr;
+          case MVT::i8:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i8ari;
+            break;
+          case MVT::i16:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i16ari;
+            break;
+          case MVT::i32:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i32ari;
+            break;
+          case MVT::i64:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i64ari;
+            break;
+          case MVT::f32:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_f32ari;
+            break;
+          case MVT::f64:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_f64ari;
+            break;
+          }
+        } else {
+          switch (EltVT.getSimpleVT().SimpleTy) {
+          default:
+            return nullptr;
+          case MVT::i8:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i8ari;
+            break;
+          case MVT::i16:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i16ari;
+            break;
+          case MVT::i32:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i32ari;
+            break;
+          case MVT::i64:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i64ari;
+            break;
+          case MVT::f32:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_f32ari;
+            break;
+          case MVT::f64:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_f64ari;
+            break;
+          }
+        }
+        break;
       case NVPTXISD::LDGV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDG_G_v2i8_ELE_ari32;
           break;
@@ -1037,7 +1646,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDUV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDU_G_v2i8_ELE_ari32;
           break;
@@ -1061,7 +1670,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDGV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDG_G_v4i8_ELE_ari32;
           break;
@@ -1079,7 +1688,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDUV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDU_G_v4i8_ELE_ari32;
           break;
@@ -1099,17 +1708,65 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
 
     SDValue Ops[] = { Base, Offset, Chain };
 
-    LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(),
-                                ArrayRef<SDValue>(Ops, 3));
+    LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(), Ops);
   } else {
     if (Subtarget.is64Bit()) {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
+      case ISD::INTRINSIC_W_CHAIN:
+        if (IsLDG) {
+          switch (EltVT.getSimpleVT().SimpleTy) {
+          default:
+            return nullptr;
+          case MVT::i8:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i8areg64;
+            break;
+          case MVT::i16:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i16areg64;
+            break;
+          case MVT::i32:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i32areg64;
+            break;
+          case MVT::i64:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i64areg64;
+            break;
+          case MVT::f32:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_f32areg64;
+            break;
+          case MVT::f64:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_f64areg64;
+            break;
+          }
+        } else {
+          switch (EltVT.getSimpleVT().SimpleTy) {
+          default:
+            return nullptr;
+          case MVT::i8:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i8areg64;
+            break;
+          case MVT::i16:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i16areg64;
+            break;
+          case MVT::i32:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i32areg64;
+            break;
+          case MVT::i64:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i64areg64;
+            break;
+          case MVT::f32:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_f32areg64;
+            break;
+          case MVT::f64:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_f64areg64;
+            break;
+          }
+        }
+        break;
       case NVPTXISD::LDGV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDG_G_v2i8_ELE_areg64;
           break;
@@ -1133,7 +1790,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDUV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDU_G_v2i8_ELE_areg64;
           break;
@@ -1157,7 +1814,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDGV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDG_G_v4i8_ELE_areg64;
           break;
@@ -1175,7 +1832,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDUV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDU_G_v4i8_ELE_areg64;
           break;
@@ -1194,11 +1851,60 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
     } else {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
+      case ISD::INTRINSIC_W_CHAIN:
+        if (IsLDG) {
+          switch (EltVT.getSimpleVT().SimpleTy) {
+          default:
+            return nullptr;
+          case MVT::i8:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i8areg;
+            break;
+          case MVT::i16:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i16areg;
+            break;
+          case MVT::i32:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i32areg;
+            break;
+          case MVT::i64:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_i64areg;
+            break;
+          case MVT::f32:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_f32areg;
+            break;
+          case MVT::f64:
+            Opcode = NVPTX::INT_PTX_LDG_GLOBAL_f64areg;
+            break;
+          }
+        } else {
+          switch (EltVT.getSimpleVT().SimpleTy) {
+          default:
+            return nullptr;
+          case MVT::i8:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i8areg;
+            break;
+          case MVT::i16:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i16areg;
+            break;
+          case MVT::i32:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i32areg;
+            break;
+          case MVT::i64:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_i64areg;
+            break;
+          case MVT::f32:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_f32areg;
+            break;
+          case MVT::f64:
+            Opcode = NVPTX::INT_PTX_LDU_GLOBAL_f64areg;
+            break;
+          }
+        }
+        break;
       case NVPTXISD::LDGV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDG_G_v2i8_ELE_areg32;
           break;
@@ -1222,7 +1928,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDUV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDU_G_v2i8_ELE_areg32;
           break;
@@ -1246,7 +1952,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDGV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDG_G_v4i8_ELE_areg32;
           break;
@@ -1264,7 +1970,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
       case NVPTXISD::LDUV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::INT_PTX_LDU_G_v4i8_ELE_areg32;
           break;
@@ -1283,12 +1989,11 @@ SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
     }
 
     SDValue Ops[] = { Op1, Chain };
-    LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(),
-                                ArrayRef<SDValue>(Ops, 2));
+    LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(), Ops);
   }
 
   MachineSDNode::mmo_iterator MemRefs0 = MF->allocateMemRefsArray(1);
-  MemRefs0[0] = cast<MemSDNode>(N)->getMemOperand();
+  MemRefs0[0] = Mem->getMemOperand();
   cast<MachineSDNode>(LD)->setMemRefs(MemRefs0, MemRefs0 + 1);
 
   return LD;
@@ -1298,14 +2003,14 @@ SDNode *NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
   SDLoc dl(N);
   StoreSDNode *ST = cast<StoreSDNode>(N);
   EVT StoreVT = ST->getMemoryVT();
-  SDNode *NVPTXST = NULL;
+  SDNode *NVPTXST = nullptr;
 
   // do not support pre/post inc/dec
   if (ST->isIndexed())
-    return NULL;
+    return nullptr;
 
   if (!StoreVT.isSimple())
-    return NULL;
+    return nullptr;
 
   // Address Space Setting
   unsigned int codeAddrSpace = getCodeAddrSpace(ST, Subtarget);
@@ -1328,7 +2033,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
     else if (num == 4)
       vecType = NVPTX::PTXLdStInstCode::V4;
     else
-      return NULL;
+      return nullptr;
   }
 
   // Type Setting: toType + toTypeWidth
@@ -1372,7 +2077,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
       Opcode = NVPTX::ST_f64_avar;
       break;
     default:
-      return NULL;
+      return nullptr;
     }
     SDValue Ops[] = { N1, getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
                       getI32Imm(vecType), getI32Imm(toType),
@@ -1401,7 +2106,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
       Opcode = NVPTX::ST_f64_asi;
       break;
     default:
-      return NULL;
+      return nullptr;
     }
     SDValue Ops[] = { N1, getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
                       getI32Imm(vecType), getI32Imm(toType),
@@ -1431,7 +2136,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
         Opcode = NVPTX::ST_f64_ari_64;
         break;
       default:
-        return NULL;
+        return nullptr;
       }
     } else {
       switch (SourceVT) {
@@ -1454,7 +2159,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
         Opcode = NVPTX::ST_f64_ari;
         break;
       default:
-        return NULL;
+        return nullptr;
       }
     }
     SDValue Ops[] = { N1, getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
@@ -1483,7 +2188,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
         Opcode = NVPTX::ST_f64_areg_64;
         break;
       default:
-        return NULL;
+        return nullptr;
       }
     } else {
       switch (SourceVT) {
@@ -1506,7 +2211,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
         Opcode = NVPTX::ST_f64_areg;
         break;
       default:
-        return NULL;
+        return nullptr;
       }
     }
     SDValue Ops[] = { N1, getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
@@ -1515,7 +2220,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
     NVPTXST = CurDAG->getMachineNode(Opcode, dl, MVT::Other, Ops);
   }
 
-  if (NVPTXST != NULL) {
+  if (NVPTXST) {
     MachineSDNode::mmo_iterator MemRefs0 = MF->allocateMemRefsArray(1);
     MemRefs0[0] = cast<MemSDNode>(N)->getMemOperand();
     cast<MachineSDNode>(NVPTXST)->setMemRefs(MemRefs0, MemRefs0 + 1);
@@ -1582,7 +2287,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
     N2 = N->getOperand(5);
     break;
   default:
-    return NULL;
+    return nullptr;
   }
 
   StOps.push_back(getI32Imm(IsVolatile));
@@ -1594,11 +2299,11 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
   if (SelectDirectAddr(N2, Addr)) {
     switch (N->getOpcode()) {
     default:
-      return NULL;
+      return nullptr;
     case NVPTXISD::StoreV2:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::STV_i8_v2_avar;
         break;
@@ -1622,7 +2327,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
     case NVPTXISD::StoreV4:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::STV_i8_v4_avar;
         break;
@@ -1644,11 +2349,11 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
                  : SelectADDRsi(N2.getNode(), N2, Base, Offset)) {
     switch (N->getOpcode()) {
     default:
-      return NULL;
+      return nullptr;
     case NVPTXISD::StoreV2:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::STV_i8_v2_asi;
         break;
@@ -1672,7 +2377,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
     case NVPTXISD::StoreV4:
       switch (EltVT.getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i8:
         Opcode = NVPTX::STV_i8_v4_asi;
         break;
@@ -1696,11 +2401,11 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
     if (Subtarget.is64Bit()) {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
       case NVPTXISD::StoreV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::STV_i8_v2_ari_64;
           break;
@@ -1724,7 +2429,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
       case NVPTXISD::StoreV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::STV_i8_v4_ari_64;
           break;
@@ -1743,11 +2448,11 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
     } else {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
       case NVPTXISD::StoreV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::STV_i8_v2_ari;
           break;
@@ -1771,7 +2476,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
       case NVPTXISD::StoreV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::STV_i8_v4_ari;
           break;
@@ -1794,11 +2499,11 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
     if (Subtarget.is64Bit()) {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
       case NVPTXISD::StoreV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::STV_i8_v2_areg_64;
           break;
@@ -1822,7 +2527,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
       case NVPTXISD::StoreV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::STV_i8_v4_areg_64;
           break;
@@ -1841,11 +2546,11 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
     } else {
       switch (N->getOpcode()) {
       default:
-        return NULL;
+        return nullptr;
       case NVPTXISD::StoreV2:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::STV_i8_v2_areg;
           break;
@@ -1869,7 +2574,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
       case NVPTXISD::StoreV4:
         switch (EltVT.getSimpleVT().SimpleTy) {
         default:
-          return NULL;
+          return nullptr;
         case MVT::i8:
           Opcode = NVPTX::STV_i8_v4_areg;
           break;
@@ -1910,7 +2615,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadParam(SDNode *Node) {
   unsigned VecSize;
   switch (Node->getOpcode()) {
   default:
-    return NULL;
+    return nullptr;
   case NVPTXISD::LoadParam:
     VecSize = 1;
     break;
@@ -1929,11 +2634,11 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadParam(SDNode *Node) {
 
   switch (VecSize) {
   default:
-    return NULL;
+    return nullptr;
   case 1:
     switch (MemVT.getSimpleVT().SimpleTy) {
     default:
-      return NULL;
+      return nullptr;
     case MVT::i1:
       Opc = NVPTX::LoadParamMemI8;
       break;
@@ -1960,7 +2665,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadParam(SDNode *Node) {
   case 2:
     switch (MemVT.getSimpleVT().SimpleTy) {
     default:
-      return NULL;
+      return nullptr;
     case MVT::i1:
       Opc = NVPTX::LoadParamMemV2I8;
       break;
@@ -1987,7 +2692,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadParam(SDNode *Node) {
   case 4:
     switch (MemVT.getSimpleVT().SimpleTy) {
     default:
-      return NULL;
+      return nullptr;
     case MVT::i1:
       Opc = NVPTX::LoadParamMemV4I8;
       break;
@@ -2014,7 +2719,7 @@ SDNode *NVPTXDAGToDAGISel::SelectLoadParam(SDNode *Node) {
     VTs = CurDAG->getVTList(EltVT, EltVT, MVT::Other, MVT::Glue);
   } else {
     EVT EVTs[] = { EltVT, EltVT, EltVT, EltVT, MVT::Other, MVT::Glue };
-    VTs = CurDAG->getVTList(&EVTs[0], 5);
+    VTs = CurDAG->getVTList(EVTs);
   }
 
   unsigned OffsetVal = cast<ConstantSDNode>(Offset)->getZExtValue();
@@ -2040,7 +2745,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreRetval(SDNode *N) {
   unsigned NumElts = 1;
   switch (N->getOpcode()) {
   default:
-    return NULL;
+    return nullptr;
   case NVPTXISD::StoreRetval:
     NumElts = 1;
     break;
@@ -2065,11 +2770,11 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreRetval(SDNode *N) {
   unsigned Opcode = 0;
   switch (NumElts) {
   default:
-    return NULL;
+    return nullptr;
   case 1:
     switch (Mem->getMemoryVT().getSimpleVT().SimpleTy) {
     default:
-      return NULL;
+      return nullptr;
     case MVT::i1:
       Opcode = NVPTX::StoreRetvalI8;
       break;
@@ -2096,7 +2801,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreRetval(SDNode *N) {
   case 2:
     switch (Mem->getMemoryVT().getSimpleVT().SimpleTy) {
     default:
-      return NULL;
+      return nullptr;
     case MVT::i1:
       Opcode = NVPTX::StoreRetvalV2I8;
       break;
@@ -2123,7 +2828,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreRetval(SDNode *N) {
   case 4:
     switch (Mem->getMemoryVT().getSimpleVT().SimpleTy) {
     default:
-      return NULL;
+      return nullptr;
     case MVT::i1:
       Opcode = NVPTX::StoreRetvalV4I8;
       break;
@@ -2166,7 +2871,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreParam(SDNode *N) {
   unsigned NumElts = 1;
   switch (N->getOpcode()) {
   default:
-    return NULL;
+    return nullptr;
   case NVPTXISD::StoreParamU32:
   case NVPTXISD::StoreParamS32:
   case NVPTXISD::StoreParam:
@@ -2197,11 +2902,11 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreParam(SDNode *N) {
   default:
     switch (NumElts) {
     default:
-      return NULL;
+      return nullptr;
     case 1:
       switch (Mem->getMemoryVT().getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i1:
         Opcode = NVPTX::StoreParamI8;
         break;
@@ -2228,7 +2933,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreParam(SDNode *N) {
     case 2:
       switch (Mem->getMemoryVT().getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i1:
         Opcode = NVPTX::StoreParamV2I8;
         break;
@@ -2255,7 +2960,7 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreParam(SDNode *N) {
     case 4:
       switch (Mem->getMemoryVT().getSimpleVT().SimpleTy) {
       default:
-        return NULL;
+        return nullptr;
       case MVT::i1:
         Opcode = NVPTX::StoreParamV4I8;
         break;
@@ -2308,6 +3013,1940 @@ SDNode *NVPTXDAGToDAGISel::SelectStoreParam(SDNode *N) {
   return Ret;
 }
 
+SDNode *NVPTXDAGToDAGISel::SelectTextureIntrinsic(SDNode *N) {
+  SDValue Chain = N->getOperand(0);
+  SDNode *Ret = nullptr;
+  unsigned Opc = 0;
+  SmallVector<SDValue, 8> Ops;
+
+  switch (N->getOpcode()) {
+  default: return nullptr;
+  case NVPTXISD::Tex1DFloatS32:
+    Opc = NVPTX::TEX_1D_F32_S32;
+    break;
+  case NVPTXISD::Tex1DFloatFloat:
+    Opc = NVPTX::TEX_1D_F32_F32;
+    break;
+  case NVPTXISD::Tex1DFloatFloatLevel:
+    Opc = NVPTX::TEX_1D_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex1DFloatFloatGrad:
+    Opc = NVPTX::TEX_1D_F32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex1DS32S32:
+    Opc = NVPTX::TEX_1D_S32_S32;
+    break;
+  case NVPTXISD::Tex1DS32Float:
+    Opc = NVPTX::TEX_1D_S32_F32;
+    break;
+  case NVPTXISD::Tex1DS32FloatLevel:
+    Opc = NVPTX::TEX_1D_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex1DS32FloatGrad:
+    Opc = NVPTX::TEX_1D_S32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex1DU32S32:
+    Opc = NVPTX::TEX_1D_U32_S32;
+    break;
+  case NVPTXISD::Tex1DU32Float:
+    Opc = NVPTX::TEX_1D_U32_F32;
+    break;
+  case NVPTXISD::Tex1DU32FloatLevel:
+    Opc = NVPTX::TEX_1D_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex1DU32FloatGrad:
+    Opc = NVPTX::TEX_1D_U32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex1DArrayFloatS32:
+    Opc = NVPTX::TEX_1D_ARRAY_F32_S32;
+    break;
+  case NVPTXISD::Tex1DArrayFloatFloat:
+    Opc = NVPTX::TEX_1D_ARRAY_F32_F32;
+    break;
+  case NVPTXISD::Tex1DArrayFloatFloatLevel:
+    Opc = NVPTX::TEX_1D_ARRAY_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex1DArrayFloatFloatGrad:
+    Opc = NVPTX::TEX_1D_ARRAY_F32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex1DArrayS32S32:
+    Opc = NVPTX::TEX_1D_ARRAY_S32_S32;
+    break;
+  case NVPTXISD::Tex1DArrayS32Float:
+    Opc = NVPTX::TEX_1D_ARRAY_S32_F32;
+    break;
+  case NVPTXISD::Tex1DArrayS32FloatLevel:
+    Opc = NVPTX::TEX_1D_ARRAY_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex1DArrayS32FloatGrad:
+    Opc = NVPTX::TEX_1D_ARRAY_S32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex1DArrayU32S32:
+    Opc = NVPTX::TEX_1D_ARRAY_U32_S32;
+    break;
+  case NVPTXISD::Tex1DArrayU32Float:
+    Opc = NVPTX::TEX_1D_ARRAY_U32_F32;
+    break;
+  case NVPTXISD::Tex1DArrayU32FloatLevel:
+    Opc = NVPTX::TEX_1D_ARRAY_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex1DArrayU32FloatGrad:
+    Opc = NVPTX::TEX_1D_ARRAY_U32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex2DFloatS32:
+    Opc = NVPTX::TEX_2D_F32_S32;
+    break;
+  case NVPTXISD::Tex2DFloatFloat:
+    Opc = NVPTX::TEX_2D_F32_F32;
+    break;
+  case NVPTXISD::Tex2DFloatFloatLevel:
+    Opc = NVPTX::TEX_2D_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex2DFloatFloatGrad:
+    Opc = NVPTX::TEX_2D_F32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex2DS32S32:
+    Opc = NVPTX::TEX_2D_S32_S32;
+    break;
+  case NVPTXISD::Tex2DS32Float:
+    Opc = NVPTX::TEX_2D_S32_F32;
+    break;
+  case NVPTXISD::Tex2DS32FloatLevel:
+    Opc = NVPTX::TEX_2D_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex2DS32FloatGrad:
+    Opc = NVPTX::TEX_2D_S32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex2DU32S32:
+    Opc = NVPTX::TEX_2D_U32_S32;
+    break;
+  case NVPTXISD::Tex2DU32Float:
+    Opc = NVPTX::TEX_2D_U32_F32;
+    break;
+  case NVPTXISD::Tex2DU32FloatLevel:
+    Opc = NVPTX::TEX_2D_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex2DU32FloatGrad:
+    Opc = NVPTX::TEX_2D_U32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex2DArrayFloatS32:
+    Opc = NVPTX::TEX_2D_ARRAY_F32_S32;
+    break;
+  case NVPTXISD::Tex2DArrayFloatFloat:
+    Opc = NVPTX::TEX_2D_ARRAY_F32_F32;
+    break;
+  case NVPTXISD::Tex2DArrayFloatFloatLevel:
+    Opc = NVPTX::TEX_2D_ARRAY_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex2DArrayFloatFloatGrad:
+    Opc = NVPTX::TEX_2D_ARRAY_F32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex2DArrayS32S32:
+    Opc = NVPTX::TEX_2D_ARRAY_S32_S32;
+    break;
+  case NVPTXISD::Tex2DArrayS32Float:
+    Opc = NVPTX::TEX_2D_ARRAY_S32_F32;
+    break;
+  case NVPTXISD::Tex2DArrayS32FloatLevel:
+    Opc = NVPTX::TEX_2D_ARRAY_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex2DArrayS32FloatGrad:
+    Opc = NVPTX::TEX_2D_ARRAY_S32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex2DArrayU32S32:
+    Opc = NVPTX::TEX_2D_ARRAY_U32_S32;
+    break;
+  case NVPTXISD::Tex2DArrayU32Float:
+    Opc = NVPTX::TEX_2D_ARRAY_U32_F32;
+    break;
+  case NVPTXISD::Tex2DArrayU32FloatLevel:
+    Opc = NVPTX::TEX_2D_ARRAY_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex2DArrayU32FloatGrad:
+    Opc = NVPTX::TEX_2D_ARRAY_U32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex3DFloatS32:
+    Opc = NVPTX::TEX_3D_F32_S32;
+    break;
+  case NVPTXISD::Tex3DFloatFloat:
+    Opc = NVPTX::TEX_3D_F32_F32;
+    break;
+  case NVPTXISD::Tex3DFloatFloatLevel:
+    Opc = NVPTX::TEX_3D_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex3DFloatFloatGrad:
+    Opc = NVPTX::TEX_3D_F32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex3DS32S32:
+    Opc = NVPTX::TEX_3D_S32_S32;
+    break;
+  case NVPTXISD::Tex3DS32Float:
+    Opc = NVPTX::TEX_3D_S32_F32;
+    break;
+  case NVPTXISD::Tex3DS32FloatLevel:
+    Opc = NVPTX::TEX_3D_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex3DS32FloatGrad:
+    Opc = NVPTX::TEX_3D_S32_F32_GRAD;
+    break;
+  case NVPTXISD::Tex3DU32S32:
+    Opc = NVPTX::TEX_3D_U32_S32;
+    break;
+  case NVPTXISD::Tex3DU32Float:
+    Opc = NVPTX::TEX_3D_U32_F32;
+    break;
+  case NVPTXISD::Tex3DU32FloatLevel:
+    Opc = NVPTX::TEX_3D_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tex3DU32FloatGrad:
+    Opc = NVPTX::TEX_3D_U32_F32_GRAD;
+    break;
+  case NVPTXISD::TexCubeFloatFloat:
+    Opc = NVPTX::TEX_CUBE_F32_F32;
+    break;
+  case NVPTXISD::TexCubeFloatFloatLevel:
+    Opc = NVPTX::TEX_CUBE_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexCubeS32Float:
+    Opc = NVPTX::TEX_CUBE_S32_F32;
+    break;
+  case NVPTXISD::TexCubeS32FloatLevel:
+    Opc = NVPTX::TEX_CUBE_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexCubeU32Float:
+    Opc = NVPTX::TEX_CUBE_U32_F32;
+    break;
+  case NVPTXISD::TexCubeU32FloatLevel:
+    Opc = NVPTX::TEX_CUBE_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexCubeArrayFloatFloat:
+    Opc = NVPTX::TEX_CUBE_ARRAY_F32_F32;
+    break;
+  case NVPTXISD::TexCubeArrayFloatFloatLevel:
+    Opc = NVPTX::TEX_CUBE_ARRAY_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexCubeArrayS32Float:
+    Opc = NVPTX::TEX_CUBE_ARRAY_S32_F32;
+    break;
+  case NVPTXISD::TexCubeArrayS32FloatLevel:
+    Opc = NVPTX::TEX_CUBE_ARRAY_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexCubeArrayU32Float:
+    Opc = NVPTX::TEX_CUBE_ARRAY_U32_F32;
+    break;
+  case NVPTXISD::TexCubeArrayU32FloatLevel:
+    Opc = NVPTX::TEX_CUBE_ARRAY_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tld4R2DFloatFloat:
+    Opc = NVPTX::TLD4_R_2D_F32_F32;
+    break;
+  case NVPTXISD::Tld4G2DFloatFloat:
+    Opc = NVPTX::TLD4_G_2D_F32_F32;
+    break;
+  case NVPTXISD::Tld4B2DFloatFloat:
+    Opc = NVPTX::TLD4_B_2D_F32_F32;
+    break;
+  case NVPTXISD::Tld4A2DFloatFloat:
+    Opc = NVPTX::TLD4_A_2D_F32_F32;
+    break;
+  case NVPTXISD::Tld4R2DS64Float:
+    Opc = NVPTX::TLD4_R_2D_S32_F32;
+    break;
+  case NVPTXISD::Tld4G2DS64Float:
+    Opc = NVPTX::TLD4_G_2D_S32_F32;
+    break;
+  case NVPTXISD::Tld4B2DS64Float:
+    Opc = NVPTX::TLD4_B_2D_S32_F32;
+    break;
+  case NVPTXISD::Tld4A2DS64Float:
+    Opc = NVPTX::TLD4_A_2D_S32_F32;
+    break;
+  case NVPTXISD::Tld4R2DU64Float:
+    Opc = NVPTX::TLD4_R_2D_U32_F32;
+    break;
+  case NVPTXISD::Tld4G2DU64Float:
+    Opc = NVPTX::TLD4_G_2D_U32_F32;
+    break;
+  case NVPTXISD::Tld4B2DU64Float:
+    Opc = NVPTX::TLD4_B_2D_U32_F32;
+    break;
+  case NVPTXISD::Tld4A2DU64Float:
+    Opc = NVPTX::TLD4_A_2D_U32_F32;
+    break;
+  case NVPTXISD::TexUnified1DFloatS32:
+    Opc = NVPTX::TEX_UNIFIED_1D_F32_S32;
+    break;
+  case NVPTXISD::TexUnified1DFloatFloat:
+    Opc = NVPTX::TEX_UNIFIED_1D_F32_F32;
+    break;
+  case NVPTXISD::TexUnified1DFloatFloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_1D_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified1DFloatFloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_1D_F32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified1DS32S32:
+    Opc = NVPTX::TEX_UNIFIED_1D_S32_S32;
+    break;
+  case NVPTXISD::TexUnified1DS32Float:
+    Opc = NVPTX::TEX_UNIFIED_1D_S32_F32;
+    break;
+  case NVPTXISD::TexUnified1DS32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_1D_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified1DS32FloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_1D_S32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified1DU32S32:
+    Opc = NVPTX::TEX_UNIFIED_1D_U32_S32;
+    break;
+  case NVPTXISD::TexUnified1DU32Float:
+    Opc = NVPTX::TEX_UNIFIED_1D_U32_F32;
+    break;
+  case NVPTXISD::TexUnified1DU32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_1D_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified1DU32FloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_1D_U32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified1DArrayFloatS32:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_F32_S32;
+    break;
+  case NVPTXISD::TexUnified1DArrayFloatFloat:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_F32_F32;
+    break;
+  case NVPTXISD::TexUnified1DArrayFloatFloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified1DArrayFloatFloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_F32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified1DArrayS32S32:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_S32_S32;
+    break;
+  case NVPTXISD::TexUnified1DArrayS32Float:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_S32_F32;
+    break;
+  case NVPTXISD::TexUnified1DArrayS32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified1DArrayS32FloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_S32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified1DArrayU32S32:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_U32_S32;
+    break;
+  case NVPTXISD::TexUnified1DArrayU32Float:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_U32_F32;
+    break;
+  case NVPTXISD::TexUnified1DArrayU32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified1DArrayU32FloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_1D_ARRAY_U32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified2DFloatS32:
+    Opc = NVPTX::TEX_UNIFIED_2D_F32_S32;
+    break;
+  case NVPTXISD::TexUnified2DFloatFloat:
+    Opc = NVPTX::TEX_UNIFIED_2D_F32_F32;
+    break;
+  case NVPTXISD::TexUnified2DFloatFloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_2D_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified2DFloatFloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_2D_F32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified2DS32S32:
+    Opc = NVPTX::TEX_UNIFIED_2D_S32_S32;
+    break;
+  case NVPTXISD::TexUnified2DS32Float:
+    Opc = NVPTX::TEX_UNIFIED_2D_S32_F32;
+    break;
+  case NVPTXISD::TexUnified2DS32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_2D_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified2DS32FloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_2D_S32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified2DU32S32:
+    Opc = NVPTX::TEX_UNIFIED_2D_U32_S32;
+    break;
+  case NVPTXISD::TexUnified2DU32Float:
+    Opc = NVPTX::TEX_UNIFIED_2D_U32_F32;
+    break;
+  case NVPTXISD::TexUnified2DU32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_2D_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified2DU32FloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_2D_U32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified2DArrayFloatS32:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_F32_S32;
+    break;
+  case NVPTXISD::TexUnified2DArrayFloatFloat:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_F32_F32;
+    break;
+  case NVPTXISD::TexUnified2DArrayFloatFloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified2DArrayFloatFloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_F32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified2DArrayS32S32:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_S32_S32;
+    break;
+  case NVPTXISD::TexUnified2DArrayS32Float:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_S32_F32;
+    break;
+  case NVPTXISD::TexUnified2DArrayS32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified2DArrayS32FloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_S32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified2DArrayU32S32:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_U32_S32;
+    break;
+  case NVPTXISD::TexUnified2DArrayU32Float:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_U32_F32;
+    break;
+  case NVPTXISD::TexUnified2DArrayU32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified2DArrayU32FloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_2D_ARRAY_U32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified3DFloatS32:
+    Opc = NVPTX::TEX_UNIFIED_3D_F32_S32;
+    break;
+  case NVPTXISD::TexUnified3DFloatFloat:
+    Opc = NVPTX::TEX_UNIFIED_3D_F32_F32;
+    break;
+  case NVPTXISD::TexUnified3DFloatFloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_3D_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified3DFloatFloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_3D_F32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified3DS32S32:
+    Opc = NVPTX::TEX_UNIFIED_3D_S32_S32;
+    break;
+  case NVPTXISD::TexUnified3DS32Float:
+    Opc = NVPTX::TEX_UNIFIED_3D_S32_F32;
+    break;
+  case NVPTXISD::TexUnified3DS32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_3D_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified3DS32FloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_3D_S32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnified3DU32S32:
+    Opc = NVPTX::TEX_UNIFIED_3D_U32_S32;
+    break;
+  case NVPTXISD::TexUnified3DU32Float:
+    Opc = NVPTX::TEX_UNIFIED_3D_U32_F32;
+    break;
+  case NVPTXISD::TexUnified3DU32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_3D_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnified3DU32FloatGrad:
+    Opc = NVPTX::TEX_UNIFIED_3D_U32_F32_GRAD;
+    break;
+  case NVPTXISD::TexUnifiedCubeFloatFloat:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_F32_F32;
+    break;
+  case NVPTXISD::TexUnifiedCubeFloatFloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnifiedCubeS32Float:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_S32_F32;
+    break;
+  case NVPTXISD::TexUnifiedCubeS32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnifiedCubeU32Float:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_U32_F32;
+    break;
+  case NVPTXISD::TexUnifiedCubeU32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnifiedCubeArrayFloatFloat:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_ARRAY_F32_F32;
+    break;
+  case NVPTXISD::TexUnifiedCubeArrayFloatFloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_ARRAY_F32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnifiedCubeArrayS32Float:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_ARRAY_S32_F32;
+    break;
+  case NVPTXISD::TexUnifiedCubeArrayS32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_ARRAY_S32_F32_LEVEL;
+    break;
+  case NVPTXISD::TexUnifiedCubeArrayU32Float:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_ARRAY_U32_F32;
+    break;
+  case NVPTXISD::TexUnifiedCubeArrayU32FloatLevel:
+    Opc = NVPTX::TEX_UNIFIED_CUBE_ARRAY_U32_F32_LEVEL;
+    break;
+  case NVPTXISD::Tld4UnifiedR2DFloatFloat:
+    Opc = NVPTX::TLD4_UNIFIED_R_2D_F32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedG2DFloatFloat:
+    Opc = NVPTX::TLD4_UNIFIED_G_2D_F32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedB2DFloatFloat:
+    Opc = NVPTX::TLD4_UNIFIED_B_2D_F32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedA2DFloatFloat:
+    Opc = NVPTX::TLD4_UNIFIED_A_2D_F32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedR2DS64Float:
+    Opc = NVPTX::TLD4_UNIFIED_R_2D_S32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedG2DS64Float:
+    Opc = NVPTX::TLD4_UNIFIED_G_2D_S32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedB2DS64Float:
+    Opc = NVPTX::TLD4_UNIFIED_B_2D_S32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedA2DS64Float:
+    Opc = NVPTX::TLD4_UNIFIED_A_2D_S32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedR2DU64Float:
+    Opc = NVPTX::TLD4_UNIFIED_R_2D_U32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedG2DU64Float:
+    Opc = NVPTX::TLD4_UNIFIED_G_2D_U32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedB2DU64Float:
+    Opc = NVPTX::TLD4_UNIFIED_B_2D_U32_F32;
+    break;
+  case NVPTXISD::Tld4UnifiedA2DU64Float:
+    Opc = NVPTX::TLD4_UNIFIED_A_2D_U32_F32;
+    break;
+  }
+
+  // Copy over operands
+  for (unsigned i = 1; i < N->getNumOperands(); ++i) {
+    Ops.push_back(N->getOperand(i));
+  }
+
+  Ops.push_back(Chain);
+  Ret = CurDAG->getMachineNode(Opc, SDLoc(N), N->getVTList(), Ops);
+  return Ret;
+}
+
+SDNode *NVPTXDAGToDAGISel::SelectSurfaceIntrinsic(SDNode *N) {
+  SDValue Chain = N->getOperand(0);
+  SDValue TexHandle = N->getOperand(1);
+  SDNode *Ret = nullptr;
+  unsigned Opc = 0;
+  SmallVector<SDValue, 8> Ops;
+  switch (N->getOpcode()) {
+  default: return nullptr;
+  case NVPTXISD::Suld1DI8Clamp:
+    Opc = NVPTX::SULD_1D_I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI16Clamp:
+    Opc = NVPTX::SULD_1D_I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI32Clamp:
+    Opc = NVPTX::SULD_1D_I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI64Clamp:
+    Opc = NVPTX::SULD_1D_I64_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I8Clamp:
+    Opc = NVPTX::SULD_1D_V2I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I16Clamp:
+    Opc = NVPTX::SULD_1D_V2I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I32Clamp:
+    Opc = NVPTX::SULD_1D_V2I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I64Clamp:
+    Opc = NVPTX::SULD_1D_V2I64_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV4I8Clamp:
+    Opc = NVPTX::SULD_1D_V4I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV4I16Clamp:
+    Opc = NVPTX::SULD_1D_V4I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV4I32Clamp:
+    Opc = NVPTX::SULD_1D_V4I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI8Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI16Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI32Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI64Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_I64_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I8Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I16Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I32Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I64Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I64_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV4I8Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_V4I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV4I16Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_V4I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV4I32Clamp:
+    Opc = NVPTX::SULD_1D_ARRAY_V4I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI8Clamp:
+    Opc = NVPTX::SULD_2D_I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI16Clamp:
+    Opc = NVPTX::SULD_2D_I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI32Clamp:
+    Opc = NVPTX::SULD_2D_I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI64Clamp:
+    Opc = NVPTX::SULD_2D_I64_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I8Clamp:
+    Opc = NVPTX::SULD_2D_V2I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I16Clamp:
+    Opc = NVPTX::SULD_2D_V2I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I32Clamp:
+    Opc = NVPTX::SULD_2D_V2I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I64Clamp:
+    Opc = NVPTX::SULD_2D_V2I64_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV4I8Clamp:
+    Opc = NVPTX::SULD_2D_V4I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV4I16Clamp:
+    Opc = NVPTX::SULD_2D_V4I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV4I32Clamp:
+    Opc = NVPTX::SULD_2D_V4I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI8Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI16Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI32Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI64Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_I64_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I8Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I16Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I32Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I64Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I64_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV4I8Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_V4I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV4I16Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_V4I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV4I32Clamp:
+    Opc = NVPTX::SULD_2D_ARRAY_V4I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI8Clamp:
+    Opc = NVPTX::SULD_3D_I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI16Clamp:
+    Opc = NVPTX::SULD_3D_I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI32Clamp:
+    Opc = NVPTX::SULD_3D_I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI64Clamp:
+    Opc = NVPTX::SULD_3D_I64_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I8Clamp:
+    Opc = NVPTX::SULD_3D_V2I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I16Clamp:
+    Opc = NVPTX::SULD_3D_V2I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I32Clamp:
+    Opc = NVPTX::SULD_3D_V2I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I64Clamp:
+    Opc = NVPTX::SULD_3D_V2I64_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV4I8Clamp:
+    Opc = NVPTX::SULD_3D_V4I8_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV4I16Clamp:
+    Opc = NVPTX::SULD_3D_V4I16_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV4I32Clamp:
+    Opc = NVPTX::SULD_3D_V4I32_CLAMP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI8Trap:
+    Opc = NVPTX::SULD_1D_I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI16Trap:
+    Opc = NVPTX::SULD_1D_I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI32Trap:
+    Opc = NVPTX::SULD_1D_I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI64Trap:
+    Opc = NVPTX::SULD_1D_I64_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I8Trap:
+    Opc = NVPTX::SULD_1D_V2I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I16Trap:
+    Opc = NVPTX::SULD_1D_V2I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I32Trap:
+    Opc = NVPTX::SULD_1D_V2I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I64Trap:
+    Opc = NVPTX::SULD_1D_V2I64_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV4I8Trap:
+    Opc = NVPTX::SULD_1D_V4I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV4I16Trap:
+    Opc = NVPTX::SULD_1D_V4I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV4I32Trap:
+    Opc = NVPTX::SULD_1D_V4I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI8Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI16Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI32Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI64Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_I64_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I8Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I16Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I32Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I64Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I64_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV4I8Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_V4I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV4I16Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_V4I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV4I32Trap:
+    Opc = NVPTX::SULD_1D_ARRAY_V4I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI8Trap:
+    Opc = NVPTX::SULD_2D_I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI16Trap:
+    Opc = NVPTX::SULD_2D_I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI32Trap:
+    Opc = NVPTX::SULD_2D_I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI64Trap:
+    Opc = NVPTX::SULD_2D_I64_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I8Trap:
+    Opc = NVPTX::SULD_2D_V2I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I16Trap:
+    Opc = NVPTX::SULD_2D_V2I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I32Trap:
+    Opc = NVPTX::SULD_2D_V2I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I64Trap:
+    Opc = NVPTX::SULD_2D_V2I64_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV4I8Trap:
+    Opc = NVPTX::SULD_2D_V4I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV4I16Trap:
+    Opc = NVPTX::SULD_2D_V4I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV4I32Trap:
+    Opc = NVPTX::SULD_2D_V4I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI8Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI16Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI32Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI64Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_I64_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I8Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I16Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I32Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I64Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I64_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV4I8Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_V4I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV4I16Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_V4I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV4I32Trap:
+    Opc = NVPTX::SULD_2D_ARRAY_V4I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI8Trap:
+    Opc = NVPTX::SULD_3D_I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI16Trap:
+    Opc = NVPTX::SULD_3D_I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI32Trap:
+    Opc = NVPTX::SULD_3D_I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI64Trap:
+    Opc = NVPTX::SULD_3D_I64_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I8Trap:
+    Opc = NVPTX::SULD_3D_V2I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I16Trap:
+    Opc = NVPTX::SULD_3D_V2I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I32Trap:
+    Opc = NVPTX::SULD_3D_V2I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I64Trap:
+    Opc = NVPTX::SULD_3D_V2I64_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV4I8Trap:
+    Opc = NVPTX::SULD_3D_V4I8_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV4I16Trap:
+    Opc = NVPTX::SULD_3D_V4I16_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV4I32Trap:
+    Opc = NVPTX::SULD_3D_V4I32_TRAP;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI8Zero:
+    Opc = NVPTX::SULD_1D_I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI16Zero:
+    Opc = NVPTX::SULD_1D_I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI32Zero:
+    Opc = NVPTX::SULD_1D_I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DI64Zero:
+    Opc = NVPTX::SULD_1D_I64_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I8Zero:
+    Opc = NVPTX::SULD_1D_V2I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I16Zero:
+    Opc = NVPTX::SULD_1D_V2I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I32Zero:
+    Opc = NVPTX::SULD_1D_V2I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV2I64Zero:
+    Opc = NVPTX::SULD_1D_V2I64_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV4I8Zero:
+    Opc = NVPTX::SULD_1D_V4I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV4I16Zero:
+    Opc = NVPTX::SULD_1D_V4I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DV4I32Zero:
+    Opc = NVPTX::SULD_1D_V4I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI8Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI16Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI32Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayI64Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_I64_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I8Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I16Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I32Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV2I64Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_V2I64_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV4I8Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_V4I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV4I16Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_V4I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld1DArrayV4I32Zero:
+    Opc = NVPTX::SULD_1D_ARRAY_V4I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI8Zero:
+    Opc = NVPTX::SULD_2D_I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI16Zero:
+    Opc = NVPTX::SULD_2D_I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI32Zero:
+    Opc = NVPTX::SULD_2D_I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DI64Zero:
+    Opc = NVPTX::SULD_2D_I64_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I8Zero:
+    Opc = NVPTX::SULD_2D_V2I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I16Zero:
+    Opc = NVPTX::SULD_2D_V2I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I32Zero:
+    Opc = NVPTX::SULD_2D_V2I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV2I64Zero:
+    Opc = NVPTX::SULD_2D_V2I64_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV4I8Zero:
+    Opc = NVPTX::SULD_2D_V4I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV4I16Zero:
+    Opc = NVPTX::SULD_2D_V4I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DV4I32Zero:
+    Opc = NVPTX::SULD_2D_V4I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI8Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI16Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI32Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayI64Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_I64_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I8Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I16Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I32Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV2I64Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_V2I64_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV4I8Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_V4I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV4I16Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_V4I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld2DArrayV4I32Zero:
+    Opc = NVPTX::SULD_2D_ARRAY_V4I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI8Zero:
+    Opc = NVPTX::SULD_3D_I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI16Zero:
+    Opc = NVPTX::SULD_3D_I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI32Zero:
+    Opc = NVPTX::SULD_3D_I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DI64Zero:
+    Opc = NVPTX::SULD_3D_I64_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I8Zero:
+    Opc = NVPTX::SULD_3D_V2I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I16Zero:
+    Opc = NVPTX::SULD_3D_V2I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I32Zero:
+    Opc = NVPTX::SULD_3D_V2I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV2I64Zero:
+    Opc = NVPTX::SULD_3D_V2I64_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV4I8Zero:
+    Opc = NVPTX::SULD_3D_V4I8_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV4I16Zero:
+    Opc = NVPTX::SULD_3D_V4I16_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  case NVPTXISD::Suld3DV4I32Zero:
+    Opc = NVPTX::SULD_3D_V4I32_ZERO;
+    Ops.push_back(TexHandle);
+    Ops.push_back(N->getOperand(2));
+    Ops.push_back(N->getOperand(3));
+    Ops.push_back(N->getOperand(4));
+    Ops.push_back(Chain);
+    break;
+  }
+  Ret = CurDAG->getMachineNode(Opc, SDLoc(N), N->getVTList(), Ops);
+  return Ret;
+}
+
+
+/// SelectBFE - Look for instruction sequences that can be made more efficient
+/// by using the 'bfe' (bit-field extract) PTX instruction
+SDNode *NVPTXDAGToDAGISel::SelectBFE(SDNode *N) {
+  SDValue LHS = N->getOperand(0);
+  SDValue RHS = N->getOperand(1);
+  SDValue Len;
+  SDValue Start;
+  SDValue Val;
+  bool IsSigned = false;
+
+  if (N->getOpcode() == ISD::AND) {
+    // Canonicalize the operands
+    // We want 'and %val, %mask'
+    if (isa<ConstantSDNode>(LHS) && !isa<ConstantSDNode>(RHS)) {
+      std::swap(LHS, RHS);
+    }
+
+    ConstantSDNode *Mask = dyn_cast<ConstantSDNode>(RHS);
+    if (!Mask) {
+      // We need a constant mask on the RHS of the AND
+      return NULL;
+    }
+
+    // Extract the mask bits
+    uint64_t MaskVal = Mask->getZExtValue();
+    if (!isMask_64(MaskVal)) {
+      // We *could* handle shifted masks here, but doing so would require an
+      // 'and' operation to fix up the low-order bits so we would trade
+      // shr+and for bfe+and, which has the same throughput
+      return NULL;
+    }
+
+    // How many bits are in our mask?
+    uint64_t NumBits = CountTrailingOnes_64(MaskVal);
+    Len = CurDAG->getTargetConstant(NumBits, MVT::i32);
+
+    if (LHS.getOpcode() == ISD::SRL || LHS.getOpcode() == ISD::SRA) {
+      // We have a 'srl/and' pair, extract the effective start bit and length
+      Val = LHS.getNode()->getOperand(0);
+      Start = LHS.getNode()->getOperand(1);
+      ConstantSDNode *StartConst = dyn_cast<ConstantSDNode>(Start);
+      if (StartConst) {
+        uint64_t StartVal = StartConst->getZExtValue();
+        // How many "good" bits do we have left?  "good" is defined here as bits
+        // that exist in the original value, not shifted in.
+        uint64_t GoodBits = Start.getValueType().getSizeInBits() - StartVal;
+        if (NumBits > GoodBits) {
+          // Do not handle the case where bits have been shifted in. In theory
+          // we could handle this, but the cost is likely higher than just
+          // emitting the srl/and pair.
+          return NULL;
+        }
+        Start = CurDAG->getTargetConstant(StartVal, MVT::i32);
+      } else {
+        // Do not handle the case where the shift amount (can be zero if no srl
+        // was found) is not constant. We could handle this case, but it would
+        // require run-time logic that would be more expensive than just
+        // emitting the srl/and pair.
+        return NULL;
+      }
+    } else {
+      // Do not handle the case where the LHS of the and is not a shift. While
+      // it would be trivial to handle this case, it would just transform
+      // 'and' -> 'bfe', but 'and' has higher-throughput.
+      return NULL;
+    }
+  } else if (N->getOpcode() == ISD::SRL || N->getOpcode() == ISD::SRA) {
+    if (LHS->getOpcode() == ISD::AND) {
+      ConstantSDNode *ShiftCnst = dyn_cast<ConstantSDNode>(RHS);
+      if (!ShiftCnst) {
+        // Shift amount must be constant
+        return NULL;
+      }
+
+      uint64_t ShiftAmt = ShiftCnst->getZExtValue();
+
+      SDValue AndLHS = LHS->getOperand(0);
+      SDValue AndRHS = LHS->getOperand(1);
+
+      // Canonicalize the AND to have the mask on the RHS
+      if (isa<ConstantSDNode>(AndLHS)) {
+        std::swap(AndLHS, AndRHS);
+      }
+
+      ConstantSDNode *MaskCnst = dyn_cast<ConstantSDNode>(AndRHS);
+      if (!MaskCnst) {
+        // Mask must be constant
+        return NULL;
+      }
+
+      uint64_t MaskVal = MaskCnst->getZExtValue();
+      uint64_t NumZeros;
+      uint64_t NumBits;
+      if (isMask_64(MaskVal)) {
+        NumZeros = 0;
+        // The number of bits in the result bitfield will be the number of
+        // trailing ones (the AND) minus the number of bits we shift off
+        NumBits = CountTrailingOnes_64(MaskVal) - ShiftAmt;
+      } else if (isShiftedMask_64(MaskVal)) {
+        NumZeros = countTrailingZeros(MaskVal);
+        unsigned NumOnes = CountTrailingOnes_64(MaskVal >> NumZeros);
+        // The number of bits in the result bitfield will be the number of
+        // trailing zeros plus the number of set bits in the mask minus the
+        // number of bits we shift off
+        NumBits = NumZeros + NumOnes - ShiftAmt;
+      } else {
+        // This is not a mask we can handle
+        return NULL;
+      }
+
+      if (ShiftAmt < NumZeros) {
+        // Handling this case would require extra logic that would make this
+        // transformation non-profitable
+        return NULL;
+      }
+
+      Val = AndLHS;
+      Start = CurDAG->getTargetConstant(ShiftAmt, MVT::i32);
+      Len = CurDAG->getTargetConstant(NumBits, MVT::i32);
+    } else if (LHS->getOpcode() == ISD::SHL) {
+      // Here, we have a pattern like:
+      //
+      // (sra (shl val, NN), MM)
+      // or
+      // (srl (shl val, NN), MM)
+      //
+      // If MM >= NN, we can efficiently optimize this with bfe
+      Val = LHS->getOperand(0);
+
+      SDValue ShlRHS = LHS->getOperand(1);
+      ConstantSDNode *ShlCnst = dyn_cast<ConstantSDNode>(ShlRHS);
+      if (!ShlCnst) {
+        // Shift amount must be constant
+        return NULL;
+      }
+      uint64_t InnerShiftAmt = ShlCnst->getZExtValue();
+
+      SDValue ShrRHS = RHS;
+      ConstantSDNode *ShrCnst = dyn_cast<ConstantSDNode>(ShrRHS);
+      if (!ShrCnst) {
+        // Shift amount must be constant
+        return NULL;
+      }
+      uint64_t OuterShiftAmt = ShrCnst->getZExtValue();
+
+      // To avoid extra codegen and be profitable, we need Outer >= Inner
+      if (OuterShiftAmt < InnerShiftAmt) {
+        return NULL;
+      }
+
+      // If the outer shift is more than the type size, we have no bitfield to
+      // extract (since we also check that the inner shift is <= the outer shift
+      // then this also implies that the inner shift is < the type size)
+      if (OuterShiftAmt >= Val.getValueType().getSizeInBits()) {
+        return NULL;
+      }
+
+      Start =
+        CurDAG->getTargetConstant(OuterShiftAmt - InnerShiftAmt, MVT::i32);
+      Len =
+        CurDAG->getTargetConstant(Val.getValueType().getSizeInBits() -
+                                  OuterShiftAmt, MVT::i32);
+
+      if (N->getOpcode() == ISD::SRA) {
+        // If we have a arithmetic right shift, we need to use the signed bfe
+        // variant
+        IsSigned = true;
+      }
+    } else {
+      // No can do...
+      return NULL;
+    }
+  } else {
+    // No can do...
+    return NULL;
+  }
+
+
+  unsigned Opc;
+  // For the BFE operations we form here from "and" and "srl", always use the
+  // unsigned variants.
+  if (Val.getValueType() == MVT::i32) {
+    if (IsSigned) {
+      Opc = NVPTX::BFE_S32rii;
+    } else {
+      Opc = NVPTX::BFE_U32rii;
+    }
+  } else if (Val.getValueType() == MVT::i64) {
+    if (IsSigned) {
+      Opc = NVPTX::BFE_S64rii;
+    } else {
+      Opc = NVPTX::BFE_U64rii;
+    }
+  } else {
+    // We cannot handle this type
+    return NULL;
+  }
+
+  SDValue Ops[] = {
+    Val, Start, Len
+  };
+
+  SDNode *Ret =
+    CurDAG->getMachineNode(Opc, SDLoc(N), N->getVTList(), Ops);
+
+  return Ret;
+}
+
 // SelectDirectAddr - Match a direct address for DAG.
 // A direct address could be a globaladdress or externalsymbol.
 bool NVPTXDAGToDAGISel::SelectDirectAddr(SDValue N, SDValue &Address) {
@@ -2401,14 +5040,18 @@ bool NVPTXDAGToDAGISel::SelectADDRri64(SDNode *OpNode, SDValue Addr,
 
 bool NVPTXDAGToDAGISel::ChkMemSDNodeAddressSpace(SDNode *N,
                                                  unsigned int spN) const {
-  const Value *Src = NULL;
+  const Value *Src = nullptr;
   // Even though MemIntrinsicSDNode is a subclas of MemSDNode,
   // the classof() for MemSDNode does not include MemIntrinsicSDNode
   // (See SelectionDAGNodes.h). So we need to check for both.
   if (MemSDNode *mN = dyn_cast<MemSDNode>(N)) {
-    Src = mN->getSrcValue();
+    if (spN == 0 && mN->getMemOperand()->getPseudoValue())
+      return true;
+    Src = mN->getMemOperand()->getValue();
   } else if (MemSDNode *mN = dyn_cast<MemIntrinsicSDNode>(N)) {
-    Src = mN->getSrcValue();
+    if (spN == 0 && mN->getMemOperand()->getPseudoValue())
+      return true;
+    Src = mN->getMemOperand()->getValue();
   }
   if (!Src)
     return false;
@@ -2440,24 +5083,3 @@ bool NVPTXDAGToDAGISel::SelectInlineAsmMemoryOperand(
   }
   return true;
 }
-
-// Return true if N is a undef or a constant.
-// If N was undef, return a (i8imm 0) in Retval
-// If N was imm, convert it to i8imm and return in Retval
-// Note: The convert to i8imm is required, otherwise the
-// pattern matcher inserts a bunch of IMOVi8rr to convert
-// the imm to i8imm, and this causes instruction selection
-// to fail.
-bool NVPTXDAGToDAGISel::UndefOrImm(SDValue Op, SDValue N, SDValue &Retval) {
-  if (!(N.getOpcode() == ISD::UNDEF) && !(N.getOpcode() == ISD::Constant))
-    return false;
-
-  if (N.getOpcode() == ISD::UNDEF)
-    Retval = CurDAG->getTargetConstant(0, MVT::i8);
-  else {
-    ConstantSDNode *cn = cast<ConstantSDNode>(N.getNode());
-    unsigned retval = cn->getZExtValue();
-    Retval = CurDAG->getTargetConstant(retval, MVT::i8);
-  }
-  return true;
-}
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.h b/contrib/llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.h
index d961e5014532..c62fc253c33d 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.h
@@ -11,8 +11,6 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "nvptx-isel"
-
 #include "NVPTX.h"
 #include "NVPTXISelLowering.h"
 #include "NVPTXRegisterInfo.h"
@@ -26,47 +24,48 @@ namespace {
 
 class LLVM_LIBRARY_VISIBILITY NVPTXDAGToDAGISel : public SelectionDAGISel {
 
-  // If true, generate corresponding FPCONTRACT. This is
-  // language dependent (i.e. CUDA and OpenCL works differently).
-  bool doFMAF64;
-  bool doFMAF32;
-  bool doFMAF64AGG;
-  bool doFMAF32AGG;
-  bool allowFMA;
-
   // If true, generate mul.wide from sext and mul
   bool doMulWide;
 
   int getDivF32Level() const;
   bool usePrecSqrtF32() const;
   bool useF32FTZ() const;
+  bool allowFMA() const;
 
 public:
   explicit NVPTXDAGToDAGISel(NVPTXTargetMachine &tm,
                              CodeGenOpt::Level   OptLevel);
 
   // Pass Name
-  virtual const char *getPassName() const {
+  const char *getPassName() const override {
     return "NVPTX DAG->DAG Pattern Instruction Selection";
   }
 
   const NVPTXSubtarget &Subtarget;
 
-  virtual bool SelectInlineAsmMemoryOperand(
-      const SDValue &Op, char ConstraintCode, std::vector<SDValue> &OutOps);
+  bool SelectInlineAsmMemoryOperand(const SDValue &Op,
+                                    char ConstraintCode,
+                                    std::vector<SDValue> &OutOps) override;
 private:
 // Include the pieces autogenerated from the target description.
 #include "NVPTXGenDAGISel.inc"
 
-  SDNode *Select(SDNode *N);
+  SDNode *Select(SDNode *N) override;
+  SDNode *SelectIntrinsicNoChain(SDNode *N);
+  SDNode *SelectIntrinsicChain(SDNode *N);
+  SDNode *SelectTexSurfHandle(SDNode *N);
   SDNode *SelectLoad(SDNode *N);
   SDNode *SelectLoadVector(SDNode *N);
-  SDNode *SelectLDGLDUVector(SDNode *N);
+  SDNode *SelectLDGLDU(SDNode *N);
   SDNode *SelectStore(SDNode *N);
   SDNode *SelectStoreVector(SDNode *N);
   SDNode *SelectLoadParam(SDNode *N);
   SDNode *SelectStoreRetval(SDNode *N);
   SDNode *SelectStoreParam(SDNode *N);
+  SDNode *SelectAddrSpaceCast(SDNode *N);
+  SDNode *SelectTextureIntrinsic(SDNode *N);
+  SDNode *SelectSurfaceIntrinsic(SDNode *N);
+  SDNode *SelectBFE(SDNode *N);
         
   inline SDValue getI32Imm(unsigned Imm) {
     return CurDAG->getTargetConstant(Imm, MVT::i32);
@@ -91,7 +90,5 @@ private:
 
   bool ChkMemSDNodeAddressSpace(SDNode *N, unsigned int spN) const;
 
-  bool UndefOrImm(SDValue Op, SDValue N, SDValue &Retval);
-
 };
 }
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
index 6a8be753c87c..d76b20a29eb7 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
@@ -22,6 +22,7 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/IR/CallSite.h"
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalValue.h"
@@ -29,10 +30,10 @@
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/Module.h"
 #include "llvm/MC/MCSectionELF.h"
-#include "llvm/Support/CallSite.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 #include <sstream>
 
@@ -47,6 +48,12 @@ static cl::opt<bool> sched4reg(
     "nvptx-sched4reg",
     cl::desc("NVPTX Specific: schedule for register pressue"), cl::init(false));
 
+static cl::opt<unsigned>
+FMAContractLevelOpt("nvptx-fma-level", cl::ZeroOrMore, cl::Hidden,
+                    cl::desc("NVPTX Specific: FMA contraction (0: don't do it"
+                             " 1: do it  2: do it aggressively"),
+                    cl::init(2));
+
 static bool IsPTXVectorType(MVT VT) {
   switch (VT.SimpleTy) {
   default:
@@ -75,7 +82,7 @@ static bool IsPTXVectorType(MVT VT) {
 /// LowerCall, and LowerReturn.
 static void ComputePTXValueVTs(const TargetLowering &TLI, Type *Ty,
                                SmallVectorImpl<EVT> &ValueVTs,
-                               SmallVectorImpl<uint64_t> *Offsets = 0,
+                               SmallVectorImpl<uint64_t> *Offsets = nullptr,
                                uint64_t StartingOffset = 0) {
   SmallVector<EVT, 16> TempVTs;
   SmallVector<uint64_t, 16> TempOffsets;
@@ -111,6 +118,7 @@ NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
   MaxStoresPerMemmove = (unsigned) 0xFFFFFFFF;
 
   setBooleanContents(ZeroOrNegativeOneBooleanContent);
+  setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
 
   // Jump is Expensive. Don't create extra control flow for 'and', 'or'
   // condition branches.
@@ -130,7 +138,13 @@ NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
   addRegisterClass(MVT::f64, &NVPTX::Float64RegsRegClass);
 
   // Operations not directly supported by NVPTX.
-  setOperationAction(ISD::SELECT_CC, MVT::Other, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::i1, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::i8, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::i16, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::i32, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::i64, Expand);
   setOperationAction(ISD::BR_CC, MVT::f32, Expand);
   setOperationAction(ISD::BR_CC, MVT::f64, Expand);
   setOperationAction(ISD::BR_CC, MVT::i1, Expand);
@@ -146,6 +160,13 @@ NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
   setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8 , Legal);
   setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
 
+  setOperationAction(ISD::SHL_PARTS, MVT::i32  , Custom);
+  setOperationAction(ISD::SRA_PARTS, MVT::i32  , Custom);
+  setOperationAction(ISD::SRL_PARTS, MVT::i32  , Custom);
+  setOperationAction(ISD::SHL_PARTS, MVT::i64  , Custom);
+  setOperationAction(ISD::SRA_PARTS, MVT::i64  , Custom);
+  setOperationAction(ISD::SRL_PARTS, MVT::i64  , Custom);
+
   if (nvptxSubtarget.hasROT64()) {
     setOperationAction(ISD::ROTL, MVT::i64, Legal);
     setOperationAction(ISD::ROTR, MVT::i64, Legal);
@@ -182,8 +203,11 @@ NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
   setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
 
   // Turn FP extload into load/fextend
+  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
   setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
   // Turn FP truncstore into trunc + store.
+  setTruncStoreAction(MVT::f32, MVT::f16, Expand);
+  setTruncStoreAction(MVT::f64, MVT::f16, Expand);
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
 
   // PTX does not support load / store predicate registers
@@ -237,6 +261,13 @@ NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
   setOperationAction(ISD::CTPOP, MVT::i32, Legal);
   setOperationAction(ISD::CTPOP, MVT::i64, Legal);
 
+  // We have some custom DAG combine patterns for these nodes
+  setTargetDAGCombine(ISD::ADD);
+  setTargetDAGCombine(ISD::AND);
+  setTargetDAGCombine(ISD::FADD);
+  setTargetDAGCombine(ISD::MUL);
+  setTargetDAGCombine(ISD::SHL);
+
   // Now deduce the information based on the above mentioned
   // actions
   computeRegisterProperties();
@@ -245,7 +276,7 @@ NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
 const char *NVPTXTargetLowering::getTargetNodeName(unsigned Opcode) const {
   switch (Opcode) {
   default:
-    return 0;
+    return nullptr;
   case NVPTXISD::CALL:
     return "NVPTXISD::CALL";
   case NVPTXISD::RET_FLAG:
@@ -328,11 +359,509 @@ const char *NVPTXTargetLowering::getTargetNodeName(unsigned Opcode) const {
     return "NVPTXISD::StoreV2";
   case NVPTXISD::StoreV4:
     return "NVPTXISD::StoreV4";
+  case NVPTXISD::FUN_SHFL_CLAMP:
+    return "NVPTXISD::FUN_SHFL_CLAMP";
+  case NVPTXISD::FUN_SHFR_CLAMP:
+    return "NVPTXISD::FUN_SHFR_CLAMP";
+  case NVPTXISD::IMAD:
+    return "NVPTXISD::IMAD";
+  case NVPTXISD::MUL_WIDE_SIGNED:
+    return "NVPTXISD::MUL_WIDE_SIGNED";
+  case NVPTXISD::MUL_WIDE_UNSIGNED:
+    return "NVPTXISD::MUL_WIDE_UNSIGNED";
+  case NVPTXISD::Tex1DFloatS32:        return "NVPTXISD::Tex1DFloatS32";
+  case NVPTXISD::Tex1DFloatFloat:      return "NVPTXISD::Tex1DFloatFloat";
+  case NVPTXISD::Tex1DFloatFloatLevel:
+    return "NVPTXISD::Tex1DFloatFloatLevel";
+  case NVPTXISD::Tex1DFloatFloatGrad:
+    return "NVPTXISD::Tex1DFloatFloatGrad";
+  case NVPTXISD::Tex1DS32S32:          return "NVPTXISD::Tex1DS32S32";
+  case NVPTXISD::Tex1DS32Float:        return "NVPTXISD::Tex1DS32Float";
+  case NVPTXISD::Tex1DS32FloatLevel:
+    return "NVPTXISD::Tex1DS32FloatLevel";
+  case NVPTXISD::Tex1DS32FloatGrad:
+    return "NVPTXISD::Tex1DS32FloatGrad";
+  case NVPTXISD::Tex1DU32S32:          return "NVPTXISD::Tex1DU32S32";
+  case NVPTXISD::Tex1DU32Float:        return "NVPTXISD::Tex1DU32Float";
+  case NVPTXISD::Tex1DU32FloatLevel:
+    return "NVPTXISD::Tex1DU32FloatLevel";
+  case NVPTXISD::Tex1DU32FloatGrad:
+    return "NVPTXISD::Tex1DU32FloatGrad";
+  case NVPTXISD::Tex1DArrayFloatS32:   return "NVPTXISD::Tex1DArrayFloatS32";
+  case NVPTXISD::Tex1DArrayFloatFloat: return "NVPTXISD::Tex1DArrayFloatFloat";
+  case NVPTXISD::Tex1DArrayFloatFloatLevel:
+    return "NVPTXISD::Tex1DArrayFloatFloatLevel";
+  case NVPTXISD::Tex1DArrayFloatFloatGrad:
+    return "NVPTXISD::Tex1DArrayFloatFloatGrad";
+  case NVPTXISD::Tex1DArrayS32S32:     return "NVPTXISD::Tex1DArrayS32S32";
+  case NVPTXISD::Tex1DArrayS32Float:   return "NVPTXISD::Tex1DArrayS32Float";
+  case NVPTXISD::Tex1DArrayS32FloatLevel:
+    return "NVPTXISD::Tex1DArrayS32FloatLevel";
+  case NVPTXISD::Tex1DArrayS32FloatGrad:
+    return "NVPTXISD::Tex1DArrayS32FloatGrad";
+  case NVPTXISD::Tex1DArrayU32S32:     return "NVPTXISD::Tex1DArrayU32S32";
+  case NVPTXISD::Tex1DArrayU32Float:   return "NVPTXISD::Tex1DArrayU32Float";
+  case NVPTXISD::Tex1DArrayU32FloatLevel:
+    return "NVPTXISD::Tex1DArrayU32FloatLevel";
+  case NVPTXISD::Tex1DArrayU32FloatGrad:
+    return "NVPTXISD::Tex1DArrayU32FloatGrad";
+  case NVPTXISD::Tex2DFloatS32:        return "NVPTXISD::Tex2DFloatS32";
+  case NVPTXISD::Tex2DFloatFloat:      return "NVPTXISD::Tex2DFloatFloat";
+  case NVPTXISD::Tex2DFloatFloatLevel:
+    return "NVPTXISD::Tex2DFloatFloatLevel";
+  case NVPTXISD::Tex2DFloatFloatGrad:
+    return "NVPTXISD::Tex2DFloatFloatGrad";
+  case NVPTXISD::Tex2DS32S32:          return "NVPTXISD::Tex2DS32S32";
+  case NVPTXISD::Tex2DS32Float:        return "NVPTXISD::Tex2DS32Float";
+  case NVPTXISD::Tex2DS32FloatLevel:
+    return "NVPTXISD::Tex2DS32FloatLevel";
+  case NVPTXISD::Tex2DS32FloatGrad:
+    return "NVPTXISD::Tex2DS32FloatGrad";
+  case NVPTXISD::Tex2DU32S32:          return "NVPTXISD::Tex2DU32S32";
+  case NVPTXISD::Tex2DU32Float:        return "NVPTXISD::Tex2DU32Float";
+  case NVPTXISD::Tex2DU32FloatLevel:
+    return "NVPTXISD::Tex2DU32FloatLevel";
+  case NVPTXISD::Tex2DU32FloatGrad:
+    return "NVPTXISD::Tex2DU32FloatGrad";
+  case NVPTXISD::Tex2DArrayFloatS32:   return "NVPTXISD::Tex2DArrayFloatS32";
+  case NVPTXISD::Tex2DArrayFloatFloat: return "NVPTXISD::Tex2DArrayFloatFloat";
+  case NVPTXISD::Tex2DArrayFloatFloatLevel:
+    return "NVPTXISD::Tex2DArrayFloatFloatLevel";
+  case NVPTXISD::Tex2DArrayFloatFloatGrad:
+    return "NVPTXISD::Tex2DArrayFloatFloatGrad";
+  case NVPTXISD::Tex2DArrayS32S32:     return "NVPTXISD::Tex2DArrayS32S32";
+  case NVPTXISD::Tex2DArrayS32Float:   return "NVPTXISD::Tex2DArrayS32Float";
+  case NVPTXISD::Tex2DArrayS32FloatLevel:
+    return "NVPTXISD::Tex2DArrayS32FloatLevel";
+  case NVPTXISD::Tex2DArrayS32FloatGrad:
+    return "NVPTXISD::Tex2DArrayS32FloatGrad";
+  case NVPTXISD::Tex2DArrayU32S32:     return "NVPTXISD::Tex2DArrayU32S32";
+  case NVPTXISD::Tex2DArrayU32Float:   return "NVPTXISD::Tex2DArrayU32Float";
+  case NVPTXISD::Tex2DArrayU32FloatLevel:
+    return "NVPTXISD::Tex2DArrayU32FloatLevel";
+  case NVPTXISD::Tex2DArrayU32FloatGrad:
+    return "NVPTXISD::Tex2DArrayU32FloatGrad";
+  case NVPTXISD::Tex3DFloatS32:        return "NVPTXISD::Tex3DFloatS32";
+  case NVPTXISD::Tex3DFloatFloat:      return "NVPTXISD::Tex3DFloatFloat";
+  case NVPTXISD::Tex3DFloatFloatLevel:
+    return "NVPTXISD::Tex3DFloatFloatLevel";
+  case NVPTXISD::Tex3DFloatFloatGrad:
+    return "NVPTXISD::Tex3DFloatFloatGrad";
+  case NVPTXISD::Tex3DS32S32:          return "NVPTXISD::Tex3DS32S32";
+  case NVPTXISD::Tex3DS32Float:        return "NVPTXISD::Tex3DS32Float";
+  case NVPTXISD::Tex3DS32FloatLevel:
+    return "NVPTXISD::Tex3DS32FloatLevel";
+  case NVPTXISD::Tex3DS32FloatGrad:
+    return "NVPTXISD::Tex3DS32FloatGrad";
+  case NVPTXISD::Tex3DU32S32:          return "NVPTXISD::Tex3DU32S32";
+  case NVPTXISD::Tex3DU32Float:        return "NVPTXISD::Tex3DU32Float";
+  case NVPTXISD::Tex3DU32FloatLevel:
+    return "NVPTXISD::Tex3DU32FloatLevel";
+  case NVPTXISD::Tex3DU32FloatGrad:
+    return "NVPTXISD::Tex3DU32FloatGrad";
+  case NVPTXISD::TexCubeFloatFloat:      return "NVPTXISD::TexCubeFloatFloat";
+  case NVPTXISD::TexCubeFloatFloatLevel:
+    return "NVPTXISD::TexCubeFloatFloatLevel";
+  case NVPTXISD::TexCubeS32Float:        return "NVPTXISD::TexCubeS32Float";
+  case NVPTXISD::TexCubeS32FloatLevel:
+    return "NVPTXISD::TexCubeS32FloatLevel";
+  case NVPTXISD::TexCubeU32Float:        return "NVPTXISD::TexCubeU32Float";
+  case NVPTXISD::TexCubeU32FloatLevel:
+    return "NVPTXISD::TexCubeU32FloatLevel";
+  case NVPTXISD::TexCubeArrayFloatFloat:
+    return "NVPTXISD::TexCubeArrayFloatFloat";
+  case NVPTXISD::TexCubeArrayFloatFloatLevel:
+    return "NVPTXISD::TexCubeArrayFloatFloatLevel";
+  case NVPTXISD::TexCubeArrayS32Float:
+    return "NVPTXISD::TexCubeArrayS32Float";
+  case NVPTXISD::TexCubeArrayS32FloatLevel:
+    return "NVPTXISD::TexCubeArrayS32FloatLevel";
+  case NVPTXISD::TexCubeArrayU32Float:
+    return "NVPTXISD::TexCubeArrayU32Float";
+  case NVPTXISD::TexCubeArrayU32FloatLevel:
+    return "NVPTXISD::TexCubeArrayU32FloatLevel";
+  case NVPTXISD::Tld4R2DFloatFloat:
+    return "NVPTXISD::Tld4R2DFloatFloat";
+  case NVPTXISD::Tld4G2DFloatFloat:
+    return "NVPTXISD::Tld4G2DFloatFloat";
+  case NVPTXISD::Tld4B2DFloatFloat:
+    return "NVPTXISD::Tld4B2DFloatFloat";
+  case NVPTXISD::Tld4A2DFloatFloat:
+    return "NVPTXISD::Tld4A2DFloatFloat";
+  case NVPTXISD::Tld4R2DS64Float:
+    return "NVPTXISD::Tld4R2DS64Float";
+  case NVPTXISD::Tld4G2DS64Float:
+    return "NVPTXISD::Tld4G2DS64Float";
+  case NVPTXISD::Tld4B2DS64Float:
+    return "NVPTXISD::Tld4B2DS64Float";
+  case NVPTXISD::Tld4A2DS64Float:
+    return "NVPTXISD::Tld4A2DS64Float";
+  case NVPTXISD::Tld4R2DU64Float:
+    return "NVPTXISD::Tld4R2DU64Float";
+  case NVPTXISD::Tld4G2DU64Float:
+    return "NVPTXISD::Tld4G2DU64Float";
+  case NVPTXISD::Tld4B2DU64Float:
+    return "NVPTXISD::Tld4B2DU64Float";
+  case NVPTXISD::Tld4A2DU64Float:
+    return "NVPTXISD::Tld4A2DU64Float";
+
+  case NVPTXISD::TexUnified1DFloatS32:
+    return "NVPTXISD::TexUnified1DFloatS32";
+  case NVPTXISD::TexUnified1DFloatFloat:
+    return "NVPTXISD::TexUnified1DFloatFloat";
+  case NVPTXISD::TexUnified1DFloatFloatLevel:
+    return "NVPTXISD::TexUnified1DFloatFloatLevel";
+  case NVPTXISD::TexUnified1DFloatFloatGrad:
+    return "NVPTXISD::TexUnified1DFloatFloatGrad";
+  case NVPTXISD::TexUnified1DS32S32:
+    return "NVPTXISD::TexUnified1DS32S32";
+  case NVPTXISD::TexUnified1DS32Float:
+    return "NVPTXISD::TexUnified1DS32Float";
+  case NVPTXISD::TexUnified1DS32FloatLevel:
+    return "NVPTXISD::TexUnified1DS32FloatLevel";
+  case NVPTXISD::TexUnified1DS32FloatGrad:
+    return "NVPTXISD::TexUnified1DS32FloatGrad";
+  case NVPTXISD::TexUnified1DU32S32:
+    return "NVPTXISD::TexUnified1DU32S32";
+  case NVPTXISD::TexUnified1DU32Float:
+    return "NVPTXISD::TexUnified1DU32Float";
+  case NVPTXISD::TexUnified1DU32FloatLevel:
+    return "NVPTXISD::TexUnified1DU32FloatLevel";
+  case NVPTXISD::TexUnified1DU32FloatGrad:
+    return "NVPTXISD::TexUnified1DU32FloatGrad";
+  case NVPTXISD::TexUnified1DArrayFloatS32:
+    return "NVPTXISD::TexUnified1DArrayFloatS32";
+  case NVPTXISD::TexUnified1DArrayFloatFloat:
+    return "NVPTXISD::TexUnified1DArrayFloatFloat";
+  case NVPTXISD::TexUnified1DArrayFloatFloatLevel:
+    return "NVPTXISD::TexUnified1DArrayFloatFloatLevel";
+  case NVPTXISD::TexUnified1DArrayFloatFloatGrad:
+    return "NVPTXISD::TexUnified1DArrayFloatFloatGrad";
+  case NVPTXISD::TexUnified1DArrayS32S32:
+    return "NVPTXISD::TexUnified1DArrayS32S32";
+  case NVPTXISD::TexUnified1DArrayS32Float:
+    return "NVPTXISD::TexUnified1DArrayS32Float";
+  case NVPTXISD::TexUnified1DArrayS32FloatLevel:
+    return "NVPTXISD::TexUnified1DArrayS32FloatLevel";
+  case NVPTXISD::TexUnified1DArrayS32FloatGrad:
+    return "NVPTXISD::TexUnified1DArrayS32FloatGrad";
+  case NVPTXISD::TexUnified1DArrayU32S32:
+    return "NVPTXISD::TexUnified1DArrayU32S32";
+  case NVPTXISD::TexUnified1DArrayU32Float:
+    return "NVPTXISD::TexUnified1DArrayU32Float";
+  case NVPTXISD::TexUnified1DArrayU32FloatLevel:
+    return "NVPTXISD::TexUnified1DArrayU32FloatLevel";
+  case NVPTXISD::TexUnified1DArrayU32FloatGrad:
+    return "NVPTXISD::TexUnified1DArrayU32FloatGrad";
+  case NVPTXISD::TexUnified2DFloatS32:
+    return "NVPTXISD::TexUnified2DFloatS32";
+  case NVPTXISD::TexUnified2DFloatFloat:
+    return "NVPTXISD::TexUnified2DFloatFloat";
+  case NVPTXISD::TexUnified2DFloatFloatLevel:
+    return "NVPTXISD::TexUnified2DFloatFloatLevel";
+  case NVPTXISD::TexUnified2DFloatFloatGrad:
+    return "NVPTXISD::TexUnified2DFloatFloatGrad";
+  case NVPTXISD::TexUnified2DS32S32:
+    return "NVPTXISD::TexUnified2DS32S32";
+  case NVPTXISD::TexUnified2DS32Float:
+    return "NVPTXISD::TexUnified2DS32Float";
+  case NVPTXISD::TexUnified2DS32FloatLevel:
+    return "NVPTXISD::TexUnified2DS32FloatLevel";
+  case NVPTXISD::TexUnified2DS32FloatGrad:
+    return "NVPTXISD::TexUnified2DS32FloatGrad";
+  case NVPTXISD::TexUnified2DU32S32:
+    return "NVPTXISD::TexUnified2DU32S32";
+  case NVPTXISD::TexUnified2DU32Float:
+    return "NVPTXISD::TexUnified2DU32Float";
+  case NVPTXISD::TexUnified2DU32FloatLevel:
+    return "NVPTXISD::TexUnified2DU32FloatLevel";
+  case NVPTXISD::TexUnified2DU32FloatGrad:
+    return "NVPTXISD::TexUnified2DU32FloatGrad";
+  case NVPTXISD::TexUnified2DArrayFloatS32:
+    return "NVPTXISD::TexUnified2DArrayFloatS32";
+  case NVPTXISD::TexUnified2DArrayFloatFloat:
+    return "NVPTXISD::TexUnified2DArrayFloatFloat";
+  case NVPTXISD::TexUnified2DArrayFloatFloatLevel:
+    return "NVPTXISD::TexUnified2DArrayFloatFloatLevel";
+  case NVPTXISD::TexUnified2DArrayFloatFloatGrad:
+    return "NVPTXISD::TexUnified2DArrayFloatFloatGrad";
+  case NVPTXISD::TexUnified2DArrayS32S32:
+    return "NVPTXISD::TexUnified2DArrayS32S32";
+  case NVPTXISD::TexUnified2DArrayS32Float:
+    return "NVPTXISD::TexUnified2DArrayS32Float";
+  case NVPTXISD::TexUnified2DArrayS32FloatLevel:
+    return "NVPTXISD::TexUnified2DArrayS32FloatLevel";
+  case NVPTXISD::TexUnified2DArrayS32FloatGrad:
+    return "NVPTXISD::TexUnified2DArrayS32FloatGrad";
+  case NVPTXISD::TexUnified2DArrayU32S32:
+    return "NVPTXISD::TexUnified2DArrayU32S32";
+  case NVPTXISD::TexUnified2DArrayU32Float:
+    return "NVPTXISD::TexUnified2DArrayU32Float";
+  case NVPTXISD::TexUnified2DArrayU32FloatLevel:
+    return "NVPTXISD::TexUnified2DArrayU32FloatLevel";
+  case NVPTXISD::TexUnified2DArrayU32FloatGrad:
+    return "NVPTXISD::TexUnified2DArrayU32FloatGrad";
+  case NVPTXISD::TexUnified3DFloatS32:
+    return "NVPTXISD::TexUnified3DFloatS32";
+  case NVPTXISD::TexUnified3DFloatFloat:
+    return "NVPTXISD::TexUnified3DFloatFloat";
+  case NVPTXISD::TexUnified3DFloatFloatLevel:
+    return "NVPTXISD::TexUnified3DFloatFloatLevel";
+  case NVPTXISD::TexUnified3DFloatFloatGrad:
+    return "NVPTXISD::TexUnified3DFloatFloatGrad";
+  case NVPTXISD::TexUnified3DS32S32:
+    return "NVPTXISD::TexUnified3DS32S32";
+  case NVPTXISD::TexUnified3DS32Float:
+    return "NVPTXISD::TexUnified3DS32Float";
+  case NVPTXISD::TexUnified3DS32FloatLevel:
+    return "NVPTXISD::TexUnified3DS32FloatLevel";
+  case NVPTXISD::TexUnified3DS32FloatGrad:
+    return "NVPTXISD::TexUnified3DS32FloatGrad";
+  case NVPTXISD::TexUnified3DU32S32:
+    return "NVPTXISD::TexUnified3DU32S32";
+  case NVPTXISD::TexUnified3DU32Float:
+    return "NVPTXISD::TexUnified3DU32Float";
+  case NVPTXISD::TexUnified3DU32FloatLevel:
+    return "NVPTXISD::TexUnified3DU32FloatLevel";
+  case NVPTXISD::TexUnified3DU32FloatGrad:
+    return "NVPTXISD::TexUnified3DU32FloatGrad";
+  case NVPTXISD::TexUnifiedCubeFloatFloat:
+    return "NVPTXISD::TexUnifiedCubeFloatFloat";
+  case NVPTXISD::TexUnifiedCubeFloatFloatLevel:
+    return "NVPTXISD::TexUnifiedCubeFloatFloatLevel";
+  case NVPTXISD::TexUnifiedCubeS32Float:
+    return "NVPTXISD::TexUnifiedCubeS32Float";
+  case NVPTXISD::TexUnifiedCubeS32FloatLevel:
+    return "NVPTXISD::TexUnifiedCubeS32FloatLevel";
+  case NVPTXISD::TexUnifiedCubeU32Float:
+    return "NVPTXISD::TexUnifiedCubeU32Float";
+  case NVPTXISD::TexUnifiedCubeU32FloatLevel:
+    return "NVPTXISD::TexUnifiedCubeU32FloatLevel";
+  case NVPTXISD::TexUnifiedCubeArrayFloatFloat:
+    return "NVPTXISD::TexUnifiedCubeArrayFloatFloat";
+  case NVPTXISD::TexUnifiedCubeArrayFloatFloatLevel:
+    return "NVPTXISD::TexUnifiedCubeArrayFloatFloatLevel";
+  case NVPTXISD::TexUnifiedCubeArrayS32Float:
+    return "NVPTXISD::TexUnifiedCubeArrayS32Float";
+  case NVPTXISD::TexUnifiedCubeArrayS32FloatLevel:
+    return "NVPTXISD::TexUnifiedCubeArrayS32FloatLevel";
+  case NVPTXISD::TexUnifiedCubeArrayU32Float:
+    return "NVPTXISD::TexUnifiedCubeArrayU32Float";
+  case NVPTXISD::TexUnifiedCubeArrayU32FloatLevel:
+    return "NVPTXISD::TexUnifiedCubeArrayU32FloatLevel";
+  case NVPTXISD::Tld4UnifiedR2DFloatFloat:
+    return "NVPTXISD::Tld4UnifiedR2DFloatFloat";
+  case NVPTXISD::Tld4UnifiedG2DFloatFloat:
+    return "NVPTXISD::Tld4UnifiedG2DFloatFloat";
+  case NVPTXISD::Tld4UnifiedB2DFloatFloat:
+    return "NVPTXISD::Tld4UnifiedB2DFloatFloat";
+  case NVPTXISD::Tld4UnifiedA2DFloatFloat:
+    return "NVPTXISD::Tld4UnifiedA2DFloatFloat";
+  case NVPTXISD::Tld4UnifiedR2DS64Float:
+    return "NVPTXISD::Tld4UnifiedR2DS64Float";
+  case NVPTXISD::Tld4UnifiedG2DS64Float:
+    return "NVPTXISD::Tld4UnifiedG2DS64Float";
+  case NVPTXISD::Tld4UnifiedB2DS64Float:
+    return "NVPTXISD::Tld4UnifiedB2DS64Float";
+  case NVPTXISD::Tld4UnifiedA2DS64Float:
+    return "NVPTXISD::Tld4UnifiedA2DS64Float";
+  case NVPTXISD::Tld4UnifiedR2DU64Float:
+    return "NVPTXISD::Tld4UnifiedR2DU64Float";
+  case NVPTXISD::Tld4UnifiedG2DU64Float:
+    return "NVPTXISD::Tld4UnifiedG2DU64Float";
+  case NVPTXISD::Tld4UnifiedB2DU64Float:
+    return "NVPTXISD::Tld4UnifiedB2DU64Float";
+  case NVPTXISD::Tld4UnifiedA2DU64Float:
+    return "NVPTXISD::Tld4UnifiedA2DU64Float";
+
+  case NVPTXISD::Suld1DI8Clamp:          return "NVPTXISD::Suld1DI8Clamp";
+  case NVPTXISD::Suld1DI16Clamp:         return "NVPTXISD::Suld1DI16Clamp";
+  case NVPTXISD::Suld1DI32Clamp:         return "NVPTXISD::Suld1DI32Clamp";
+  case NVPTXISD::Suld1DI64Clamp:         return "NVPTXISD::Suld1DI64Clamp";
+  case NVPTXISD::Suld1DV2I8Clamp:        return "NVPTXISD::Suld1DV2I8Clamp";
+  case NVPTXISD::Suld1DV2I16Clamp:       return "NVPTXISD::Suld1DV2I16Clamp";
+  case NVPTXISD::Suld1DV2I32Clamp:       return "NVPTXISD::Suld1DV2I32Clamp";
+  case NVPTXISD::Suld1DV2I64Clamp:       return "NVPTXISD::Suld1DV2I64Clamp";
+  case NVPTXISD::Suld1DV4I8Clamp:        return "NVPTXISD::Suld1DV4I8Clamp";
+  case NVPTXISD::Suld1DV4I16Clamp:       return "NVPTXISD::Suld1DV4I16Clamp";
+  case NVPTXISD::Suld1DV4I32Clamp:       return "NVPTXISD::Suld1DV4I32Clamp";
+
+  case NVPTXISD::Suld1DArrayI8Clamp:   return "NVPTXISD::Suld1DArrayI8Clamp";
+  case NVPTXISD::Suld1DArrayI16Clamp:  return "NVPTXISD::Suld1DArrayI16Clamp";
+  case NVPTXISD::Suld1DArrayI32Clamp:  return "NVPTXISD::Suld1DArrayI32Clamp";
+  case NVPTXISD::Suld1DArrayI64Clamp:  return "NVPTXISD::Suld1DArrayI64Clamp";
+  case NVPTXISD::Suld1DArrayV2I8Clamp: return "NVPTXISD::Suld1DArrayV2I8Clamp";
+  case NVPTXISD::Suld1DArrayV2I16Clamp:return "NVPTXISD::Suld1DArrayV2I16Clamp";
+  case NVPTXISD::Suld1DArrayV2I32Clamp:return "NVPTXISD::Suld1DArrayV2I32Clamp";
+  case NVPTXISD::Suld1DArrayV2I64Clamp:return "NVPTXISD::Suld1DArrayV2I64Clamp";
+  case NVPTXISD::Suld1DArrayV4I8Clamp: return "NVPTXISD::Suld1DArrayV4I8Clamp";
+  case NVPTXISD::Suld1DArrayV4I16Clamp:return "NVPTXISD::Suld1DArrayV4I16Clamp";
+  case NVPTXISD::Suld1DArrayV4I32Clamp:return "NVPTXISD::Suld1DArrayV4I32Clamp";
+
+  case NVPTXISD::Suld2DI8Clamp:          return "NVPTXISD::Suld2DI8Clamp";
+  case NVPTXISD::Suld2DI16Clamp:         return "NVPTXISD::Suld2DI16Clamp";
+  case NVPTXISD::Suld2DI32Clamp:         return "NVPTXISD::Suld2DI32Clamp";
+  case NVPTXISD::Suld2DI64Clamp:         return "NVPTXISD::Suld2DI64Clamp";
+  case NVPTXISD::Suld2DV2I8Clamp:        return "NVPTXISD::Suld2DV2I8Clamp";
+  case NVPTXISD::Suld2DV2I16Clamp:       return "NVPTXISD::Suld2DV2I16Clamp";
+  case NVPTXISD::Suld2DV2I32Clamp:       return "NVPTXISD::Suld2DV2I32Clamp";
+  case NVPTXISD::Suld2DV2I64Clamp:       return "NVPTXISD::Suld2DV2I64Clamp";
+  case NVPTXISD::Suld2DV4I8Clamp:        return "NVPTXISD::Suld2DV4I8Clamp";
+  case NVPTXISD::Suld2DV4I16Clamp:       return "NVPTXISD::Suld2DV4I16Clamp";
+  case NVPTXISD::Suld2DV4I32Clamp:       return "NVPTXISD::Suld2DV4I32Clamp";
+
+  case NVPTXISD::Suld2DArrayI8Clamp:   return "NVPTXISD::Suld2DArrayI8Clamp";
+  case NVPTXISD::Suld2DArrayI16Clamp:  return "NVPTXISD::Suld2DArrayI16Clamp";
+  case NVPTXISD::Suld2DArrayI32Clamp:  return "NVPTXISD::Suld2DArrayI32Clamp";
+  case NVPTXISD::Suld2DArrayI64Clamp:  return "NVPTXISD::Suld2DArrayI64Clamp";
+  case NVPTXISD::Suld2DArrayV2I8Clamp: return "NVPTXISD::Suld2DArrayV2I8Clamp";
+  case NVPTXISD::Suld2DArrayV2I16Clamp:return "NVPTXISD::Suld2DArrayV2I16Clamp";
+  case NVPTXISD::Suld2DArrayV2I32Clamp:return "NVPTXISD::Suld2DArrayV2I32Clamp";
+  case NVPTXISD::Suld2DArrayV2I64Clamp:return "NVPTXISD::Suld2DArrayV2I64Clamp";
+  case NVPTXISD::Suld2DArrayV4I8Clamp: return "NVPTXISD::Suld2DArrayV4I8Clamp";
+  case NVPTXISD::Suld2DArrayV4I16Clamp:return "NVPTXISD::Suld2DArrayV4I16Clamp";
+  case NVPTXISD::Suld2DArrayV4I32Clamp:return "NVPTXISD::Suld2DArrayV4I32Clamp";
+
+  case NVPTXISD::Suld3DI8Clamp:          return "NVPTXISD::Suld3DI8Clamp";
+  case NVPTXISD::Suld3DI16Clamp:         return "NVPTXISD::Suld3DI16Clamp";
+  case NVPTXISD::Suld3DI32Clamp:         return "NVPTXISD::Suld3DI32Clamp";
+  case NVPTXISD::Suld3DI64Clamp:         return "NVPTXISD::Suld3DI64Clamp";
+  case NVPTXISD::Suld3DV2I8Clamp:        return "NVPTXISD::Suld3DV2I8Clamp";
+  case NVPTXISD::Suld3DV2I16Clamp:       return "NVPTXISD::Suld3DV2I16Clamp";
+  case NVPTXISD::Suld3DV2I32Clamp:       return "NVPTXISD::Suld3DV2I32Clamp";
+  case NVPTXISD::Suld3DV2I64Clamp:       return "NVPTXISD::Suld3DV2I64Clamp";
+  case NVPTXISD::Suld3DV4I8Clamp:        return "NVPTXISD::Suld3DV4I8Clamp";
+  case NVPTXISD::Suld3DV4I16Clamp:       return "NVPTXISD::Suld3DV4I16Clamp";
+  case NVPTXISD::Suld3DV4I32Clamp:       return "NVPTXISD::Suld3DV4I32Clamp";
+
+  case NVPTXISD::Suld1DI8Trap:          return "NVPTXISD::Suld1DI8Trap";
+  case NVPTXISD::Suld1DI16Trap:         return "NVPTXISD::Suld1DI16Trap";
+  case NVPTXISD::Suld1DI32Trap:         return "NVPTXISD::Suld1DI32Trap";
+  case NVPTXISD::Suld1DI64Trap:         return "NVPTXISD::Suld1DI64Trap";
+  case NVPTXISD::Suld1DV2I8Trap:        return "NVPTXISD::Suld1DV2I8Trap";
+  case NVPTXISD::Suld1DV2I16Trap:       return "NVPTXISD::Suld1DV2I16Trap";
+  case NVPTXISD::Suld1DV2I32Trap:       return "NVPTXISD::Suld1DV2I32Trap";
+  case NVPTXISD::Suld1DV2I64Trap:       return "NVPTXISD::Suld1DV2I64Trap";
+  case NVPTXISD::Suld1DV4I8Trap:        return "NVPTXISD::Suld1DV4I8Trap";
+  case NVPTXISD::Suld1DV4I16Trap:       return "NVPTXISD::Suld1DV4I16Trap";
+  case NVPTXISD::Suld1DV4I32Trap:       return "NVPTXISD::Suld1DV4I32Trap";
+
+  case NVPTXISD::Suld1DArrayI8Trap:     return "NVPTXISD::Suld1DArrayI8Trap";
+  case NVPTXISD::Suld1DArrayI16Trap:    return "NVPTXISD::Suld1DArrayI16Trap";
+  case NVPTXISD::Suld1DArrayI32Trap:    return "NVPTXISD::Suld1DArrayI32Trap";
+  case NVPTXISD::Suld1DArrayI64Trap:    return "NVPTXISD::Suld1DArrayI64Trap";
+  case NVPTXISD::Suld1DArrayV2I8Trap:   return "NVPTXISD::Suld1DArrayV2I8Trap";
+  case NVPTXISD::Suld1DArrayV2I16Trap:  return "NVPTXISD::Suld1DArrayV2I16Trap";
+  case NVPTXISD::Suld1DArrayV2I32Trap:  return "NVPTXISD::Suld1DArrayV2I32Trap";
+  case NVPTXISD::Suld1DArrayV2I64Trap:  return "NVPTXISD::Suld1DArrayV2I64Trap";
+  case NVPTXISD::Suld1DArrayV4I8Trap:   return "NVPTXISD::Suld1DArrayV4I8Trap";
+  case NVPTXISD::Suld1DArrayV4I16Trap:  return "NVPTXISD::Suld1DArrayV4I16Trap";
+  case NVPTXISD::Suld1DArrayV4I32Trap:  return "NVPTXISD::Suld1DArrayV4I32Trap";
+
+  case NVPTXISD::Suld2DI8Trap:          return "NVPTXISD::Suld2DI8Trap";
+  case NVPTXISD::Suld2DI16Trap:         return "NVPTXISD::Suld2DI16Trap";
+  case NVPTXISD::Suld2DI32Trap:         return "NVPTXISD::Suld2DI32Trap";
+  case NVPTXISD::Suld2DI64Trap:         return "NVPTXISD::Suld2DI64Trap";
+  case NVPTXISD::Suld2DV2I8Trap:        return "NVPTXISD::Suld2DV2I8Trap";
+  case NVPTXISD::Suld2DV2I16Trap:       return "NVPTXISD::Suld2DV2I16Trap";
+  case NVPTXISD::Suld2DV2I32Trap:       return "NVPTXISD::Suld2DV2I32Trap";
+  case NVPTXISD::Suld2DV2I64Trap:       return "NVPTXISD::Suld2DV2I64Trap";
+  case NVPTXISD::Suld2DV4I8Trap:        return "NVPTXISD::Suld2DV4I8Trap";
+  case NVPTXISD::Suld2DV4I16Trap:       return "NVPTXISD::Suld2DV4I16Trap";
+  case NVPTXISD::Suld2DV4I32Trap:       return "NVPTXISD::Suld2DV4I32Trap";
+
+  case NVPTXISD::Suld2DArrayI8Trap:     return "NVPTXISD::Suld2DArrayI8Trap";
+  case NVPTXISD::Suld2DArrayI16Trap:    return "NVPTXISD::Suld2DArrayI16Trap";
+  case NVPTXISD::Suld2DArrayI32Trap:    return "NVPTXISD::Suld2DArrayI32Trap";
+  case NVPTXISD::Suld2DArrayI64Trap:    return "NVPTXISD::Suld2DArrayI64Trap";
+  case NVPTXISD::Suld2DArrayV2I8Trap:   return "NVPTXISD::Suld2DArrayV2I8Trap";
+  case NVPTXISD::Suld2DArrayV2I16Trap:  return "NVPTXISD::Suld2DArrayV2I16Trap";
+  case NVPTXISD::Suld2DArrayV2I32Trap:  return "NVPTXISD::Suld2DArrayV2I32Trap";
+  case NVPTXISD::Suld2DArrayV2I64Trap:  return "NVPTXISD::Suld2DArrayV2I64Trap";
+  case NVPTXISD::Suld2DArrayV4I8Trap:   return "NVPTXISD::Suld2DArrayV4I8Trap";
+  case NVPTXISD::Suld2DArrayV4I16Trap:  return "NVPTXISD::Suld2DArrayV4I16Trap";
+  case NVPTXISD::Suld2DArrayV4I32Trap:  return "NVPTXISD::Suld2DArrayV4I32Trap";
+
+  case NVPTXISD::Suld3DI8Trap:          return "NVPTXISD::Suld3DI8Trap";
+  case NVPTXISD::Suld3DI16Trap:         return "NVPTXISD::Suld3DI16Trap";
+  case NVPTXISD::Suld3DI32Trap:         return "NVPTXISD::Suld3DI32Trap";
+  case NVPTXISD::Suld3DI64Trap:         return "NVPTXISD::Suld3DI64Trap";
+  case NVPTXISD::Suld3DV2I8Trap:        return "NVPTXISD::Suld3DV2I8Trap";
+  case NVPTXISD::Suld3DV2I16Trap:       return "NVPTXISD::Suld3DV2I16Trap";
+  case NVPTXISD::Suld3DV2I32Trap:       return "NVPTXISD::Suld3DV2I32Trap";
+  case NVPTXISD::Suld3DV2I64Trap:       return "NVPTXISD::Suld3DV2I64Trap";
+  case NVPTXISD::Suld3DV4I8Trap:        return "NVPTXISD::Suld3DV4I8Trap";
+  case NVPTXISD::Suld3DV4I16Trap:       return "NVPTXISD::Suld3DV4I16Trap";
+  case NVPTXISD::Suld3DV4I32Trap:       return "NVPTXISD::Suld3DV4I32Trap";
+
+  case NVPTXISD::Suld1DI8Zero:          return "NVPTXISD::Suld1DI8Zero";
+  case NVPTXISD::Suld1DI16Zero:         return "NVPTXISD::Suld1DI16Zero";
+  case NVPTXISD::Suld1DI32Zero:         return "NVPTXISD::Suld1DI32Zero";
+  case NVPTXISD::Suld1DI64Zero:         return "NVPTXISD::Suld1DI64Zero";
+  case NVPTXISD::Suld1DV2I8Zero:        return "NVPTXISD::Suld1DV2I8Zero";
+  case NVPTXISD::Suld1DV2I16Zero:       return "NVPTXISD::Suld1DV2I16Zero";
+  case NVPTXISD::Suld1DV2I32Zero:       return "NVPTXISD::Suld1DV2I32Zero";
+  case NVPTXISD::Suld1DV2I64Zero:       return "NVPTXISD::Suld1DV2I64Zero";
+  case NVPTXISD::Suld1DV4I8Zero:        return "NVPTXISD::Suld1DV4I8Zero";
+  case NVPTXISD::Suld1DV4I16Zero:       return "NVPTXISD::Suld1DV4I16Zero";
+  case NVPTXISD::Suld1DV4I32Zero:       return "NVPTXISD::Suld1DV4I32Zero";
+
+  case NVPTXISD::Suld1DArrayI8Zero:     return "NVPTXISD::Suld1DArrayI8Zero";
+  case NVPTXISD::Suld1DArrayI16Zero:    return "NVPTXISD::Suld1DArrayI16Zero";
+  case NVPTXISD::Suld1DArrayI32Zero:    return "NVPTXISD::Suld1DArrayI32Zero";
+  case NVPTXISD::Suld1DArrayI64Zero:    return "NVPTXISD::Suld1DArrayI64Zero";
+  case NVPTXISD::Suld1DArrayV2I8Zero:   return "NVPTXISD::Suld1DArrayV2I8Zero";
+  case NVPTXISD::Suld1DArrayV2I16Zero:  return "NVPTXISD::Suld1DArrayV2I16Zero";
+  case NVPTXISD::Suld1DArrayV2I32Zero:  return "NVPTXISD::Suld1DArrayV2I32Zero";
+  case NVPTXISD::Suld1DArrayV2I64Zero:  return "NVPTXISD::Suld1DArrayV2I64Zero";
+  case NVPTXISD::Suld1DArrayV4I8Zero:   return "NVPTXISD::Suld1DArrayV4I8Zero";
+  case NVPTXISD::Suld1DArrayV4I16Zero:  return "NVPTXISD::Suld1DArrayV4I16Zero";
+  case NVPTXISD::Suld1DArrayV4I32Zero:  return "NVPTXISD::Suld1DArrayV4I32Zero";
+
+  case NVPTXISD::Suld2DI8Zero:          return "NVPTXISD::Suld2DI8Zero";
+  case NVPTXISD::Suld2DI16Zero:         return "NVPTXISD::Suld2DI16Zero";
+  case NVPTXISD::Suld2DI32Zero:         return "NVPTXISD::Suld2DI32Zero";
+  case NVPTXISD::Suld2DI64Zero:         return "NVPTXISD::Suld2DI64Zero";
+  case NVPTXISD::Suld2DV2I8Zero:        return "NVPTXISD::Suld2DV2I8Zero";
+  case NVPTXISD::Suld2DV2I16Zero:       return "NVPTXISD::Suld2DV2I16Zero";
+  case NVPTXISD::Suld2DV2I32Zero:       return "NVPTXISD::Suld2DV2I32Zero";
+  case NVPTXISD::Suld2DV2I64Zero:       return "NVPTXISD::Suld2DV2I64Zero";
+  case NVPTXISD::Suld2DV4I8Zero:        return "NVPTXISD::Suld2DV4I8Zero";
+  case NVPTXISD::Suld2DV4I16Zero:       return "NVPTXISD::Suld2DV4I16Zero";
+  case NVPTXISD::Suld2DV4I32Zero:       return "NVPTXISD::Suld2DV4I32Zero";
+
+  case NVPTXISD::Suld2DArrayI8Zero:     return "NVPTXISD::Suld2DArrayI8Zero";
+  case NVPTXISD::Suld2DArrayI16Zero:    return "NVPTXISD::Suld2DArrayI16Zero";
+  case NVPTXISD::Suld2DArrayI32Zero:    return "NVPTXISD::Suld2DArrayI32Zero";
+  case NVPTXISD::Suld2DArrayI64Zero:    return "NVPTXISD::Suld2DArrayI64Zero";
+  case NVPTXISD::Suld2DArrayV2I8Zero:   return "NVPTXISD::Suld2DArrayV2I8Zero";
+  case NVPTXISD::Suld2DArrayV2I16Zero:  return "NVPTXISD::Suld2DArrayV2I16Zero";
+  case NVPTXISD::Suld2DArrayV2I32Zero:  return "NVPTXISD::Suld2DArrayV2I32Zero";
+  case NVPTXISD::Suld2DArrayV2I64Zero:  return "NVPTXISD::Suld2DArrayV2I64Zero";
+  case NVPTXISD::Suld2DArrayV4I8Zero:   return "NVPTXISD::Suld2DArrayV4I8Zero";
+  case NVPTXISD::Suld2DArrayV4I16Zero:  return "NVPTXISD::Suld2DArrayV4I16Zero";
+  case NVPTXISD::Suld2DArrayV4I32Zero:  return "NVPTXISD::Suld2DArrayV4I32Zero";
+
+  case NVPTXISD::Suld3DI8Zero:          return "NVPTXISD::Suld3DI8Zero";
+  case NVPTXISD::Suld3DI16Zero:         return "NVPTXISD::Suld3DI16Zero";
+  case NVPTXISD::Suld3DI32Zero:         return "NVPTXISD::Suld3DI32Zero";
+  case NVPTXISD::Suld3DI64Zero:         return "NVPTXISD::Suld3DI64Zero";
+  case NVPTXISD::Suld3DV2I8Zero:        return "NVPTXISD::Suld3DV2I8Zero";
+  case NVPTXISD::Suld3DV2I16Zero:       return "NVPTXISD::Suld3DV2I16Zero";
+  case NVPTXISD::Suld3DV2I32Zero:       return "NVPTXISD::Suld3DV2I32Zero";
+  case NVPTXISD::Suld3DV2I64Zero:       return "NVPTXISD::Suld3DV2I64Zero";
+  case NVPTXISD::Suld3DV4I8Zero:        return "NVPTXISD::Suld3DV4I8Zero";
+  case NVPTXISD::Suld3DV4I16Zero:       return "NVPTXISD::Suld3DV4I16Zero";
+  case NVPTXISD::Suld3DV4I32Zero:       return "NVPTXISD::Suld3DV4I32Zero";
   }
 }
 
-bool NVPTXTargetLowering::shouldSplitVectorElementType(EVT VT) const {
-  return VT == MVT::i1;
+TargetLoweringBase::LegalizeTypeAction
+NVPTXTargetLowering::getPreferredVectorAction(EVT VT) const {
+  if (VT.getVectorNumElements() != 1 && VT.getScalarType() == MVT::i1)
+    return TypeSplitVector;
+
+  return TargetLoweringBase::getPreferredVectorAction(VT);
 }
 
 SDValue
@@ -361,7 +890,7 @@ NVPTXTargetLowering::getPrototype(Type *retTy, const ArgListTy &Args,
     O << "()";
   } else {
     O << "(";
-    if (retTy->isPrimitiveType() || retTy->isIntegerTy()) {
+    if (retTy->isFloatingPointTy() || retTy->isIntegerTy()) {
       unsigned size = 0;
       if (const IntegerType *ITy = dyn_cast<IntegerType>(retTy)) {
         size = ITy->getBitWidth();
@@ -377,26 +906,12 @@ NVPTXTargetLowering::getPrototype(Type *retTy, const ArgListTy &Args,
     } else if (isa<PointerType>(retTy)) {
       O << ".param .b" << getPointerTy().getSizeInBits() << " _";
     } else {
-      if ((retTy->getTypeID() == Type::StructTyID) || isa<VectorType>(retTy)) {
-        SmallVector<EVT, 16> vtparts;
-        ComputeValueVTs(*this, retTy, vtparts);
-        unsigned totalsz = 0;
-        for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
-          unsigned elems = 1;
-          EVT elemtype = vtparts[i];
-          if (vtparts[i].isVector()) {
-            elems = vtparts[i].getVectorNumElements();
-            elemtype = vtparts[i].getVectorElementType();
-          }
-          // TODO: no need to loop
-          for (unsigned j = 0, je = elems; j != je; ++j) {
-            unsigned sz = elemtype.getSizeInBits();
-            if (elemtype.isInteger() && (sz < 8))
-              sz = 8;
-            totalsz += sz / 8;
-          }
-        }
-        O << ".param .align " << retAlignment << " .b8 _[" << totalsz << "]";
+      if((retTy->getTypeID() == Type::StructTyID) ||
+         isa<VectorType>(retTy)) {
+        O << ".param .align "
+          << retAlignment
+          << " .b8 _["
+          << getDataLayout()->getTypeAllocSize(retTy) << "]";
       } else {
         assert(false && "Unknown return type");
       }
@@ -526,7 +1041,7 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   SDValue Chain = CLI.Chain;
   SDValue Callee = CLI.Callee;
   bool &isTailCall = CLI.IsTailCall;
-  ArgListTy &Args = CLI.Args;
+  ArgListTy &Args = CLI.getArgs();
   Type *retTy = CLI.RetTy;
   ImmutableCallSite *CS = CLI.CS;
 
@@ -565,7 +1080,8 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       if (Ty->isAggregateType()) {
         // aggregate
         SmallVector<EVT, 16> vtparts;
-        ComputeValueVTs(*this, Ty, vtparts);
+        SmallVector<uint64_t, 16> Offsets;
+        ComputePTXValueVTs(*this, Ty, vtparts, &Offsets, 0);
 
         unsigned align = getArgumentAlignment(Callee, CS, Ty, paramCount + 1);
         // declare .param .align <align> .b8 .param<n>[<size>];
@@ -575,36 +1091,28 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                                       DAG.getConstant(paramCount, MVT::i32),
                                       DAG.getConstant(sz, MVT::i32), InFlag };
         Chain = DAG.getNode(NVPTXISD::DeclareParam, dl, DeclareParamVTs,
-                            DeclareParamOps, 5);
+                            DeclareParamOps);
         InFlag = Chain.getValue(1);
-        unsigned curOffset = 0;
         for (unsigned j = 0, je = vtparts.size(); j != je; ++j) {
-          unsigned elems = 1;
           EVT elemtype = vtparts[j];
-          if (vtparts[j].isVector()) {
-            elems = vtparts[j].getVectorNumElements();
-            elemtype = vtparts[j].getVectorElementType();
-          }
-          for (unsigned k = 0, ke = elems; k != ke; ++k) {
-            unsigned sz = elemtype.getSizeInBits();
-            if (elemtype.isInteger() && (sz < 8))
-              sz = 8;
-            SDValue StVal = OutVals[OIdx];
-            if (elemtype.getSizeInBits() < 16) {
-              StVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i16, StVal);
-            }
-            SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
-            SDValue CopyParamOps[] = { Chain,
-                                       DAG.getConstant(paramCount, MVT::i32),
-                                       DAG.getConstant(curOffset, MVT::i32),
-                                       StVal, InFlag };
-            Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreParam, dl,
-                                            CopyParamVTs, &CopyParamOps[0], 5,
-                                            elemtype, MachinePointerInfo());
-            InFlag = Chain.getValue(1);
-            curOffset += sz / 8;
-            ++OIdx;
+          unsigned ArgAlign = GreatestCommonDivisor64(align, Offsets[j]);
+          if (elemtype.isInteger() && (sz < 8))
+            sz = 8;
+          SDValue StVal = OutVals[OIdx];
+          if (elemtype.getSizeInBits() < 16) {
+            StVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i16, StVal);
           }
+          SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
+          SDValue CopyParamOps[] = { Chain,
+                                     DAG.getConstant(paramCount, MVT::i32),
+                                     DAG.getConstant(Offsets[j], MVT::i32),
+                                     StVal, InFlag };
+          Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreParam, dl,
+                                          CopyParamVTs, CopyParamOps,
+                                          elemtype, MachinePointerInfo(),
+                                          ArgAlign);
+          InFlag = Chain.getValue(1);
+          ++OIdx;
         }
         if (vtparts.size() > 0)
           --OIdx;
@@ -621,7 +1129,7 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                                       DAG.getConstant(paramCount, MVT::i32),
                                       DAG.getConstant(sz, MVT::i32), InFlag };
         Chain = DAG.getNode(NVPTXISD::DeclareParam, dl, DeclareParamVTs,
-                            DeclareParamOps, 5);
+                            DeclareParamOps);
         InFlag = Chain.getValue(1);
         unsigned NumElts = ObjectVT.getVectorNumElements();
         EVT EltVT = ObjectVT.getVectorElementType();
@@ -644,7 +1152,7 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                                      DAG.getConstant(0, MVT::i32), Elt,
                                      InFlag };
           Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreParam, dl,
-                                          CopyParamVTs, &CopyParamOps[0], 5,
+                                          CopyParamVTs, CopyParamOps,
                                           MemVT, MachinePointerInfo());
           InFlag = Chain.getValue(1);
         } else if (NumElts == 2) {
@@ -661,7 +1169,7 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                                      DAG.getConstant(0, MVT::i32), Elt0, Elt1,
                                      InFlag };
           Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreParamV2, dl,
-                                          CopyParamVTs, &CopyParamOps[0], 6,
+                                          CopyParamVTs, CopyParamOps,
                                           MemVT, MachinePointerInfo());
           InFlag = Chain.getValue(1);
         } else {
@@ -735,9 +1243,8 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
             Ops.push_back(InFlag);
 
             SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
-            Chain = DAG.getMemIntrinsicNode(Opc, dl, CopyParamVTs, &Ops[0],
-                                            Ops.size(), MemVT,
-                                            MachinePointerInfo());
+            Chain = DAG.getMemIntrinsicNode(Opc, dl, CopyParamVTs, Ops,
+                                            MemVT, MachinePointerInfo());
             InFlag = Chain.getValue(1);
             curOffset += PerStoreOffset;
           }
@@ -762,7 +1269,7 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                                     DAG.getConstant(sz, MVT::i32),
                                     DAG.getConstant(0, MVT::i32), InFlag };
       Chain = DAG.getNode(NVPTXISD::DeclareScalarParam, dl, DeclareParamVTs,
-                          DeclareParamOps, 5);
+                          DeclareParamOps);
       InFlag = Chain.getValue(1);
       SDValue OutV = OutVals[OIdx];
       if (needExtend) {
@@ -781,7 +1288,7 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
         opcode = NVPTXISD::StoreParamU32;
       else if (Outs[OIdx].Flags.isSExt())
         opcode = NVPTXISD::StoreParamS32;
-      Chain = DAG.getMemIntrinsicNode(opcode, dl, CopyParamVTs, CopyParamOps, 5,
+      Chain = DAG.getMemIntrinsicNode(opcode, dl, CopyParamVTs, CopyParamOps,
                                       VT, MachinePointerInfo());
 
       InFlag = Chain.getValue(1);
@@ -790,13 +1297,15 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     }
     // struct or vector
     SmallVector<EVT, 16> vtparts;
+    SmallVector<uint64_t, 16> Offsets;
     const PointerType *PTy = dyn_cast<PointerType>(Args[i].Ty);
     assert(PTy && "Type of a byval parameter should be pointer");
-    ComputeValueVTs(*this, PTy->getElementType(), vtparts);
+    ComputePTXValueVTs(*this, PTy->getElementType(), vtparts, &Offsets, 0);
 
     // declare .param .align <align> .b8 .param<n>[<size>];
     unsigned sz = Outs[OIdx].Flags.getByValSize();
     SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
+    unsigned ArgAlign = Outs[OIdx].Flags.getByValAlign();
     // The ByValAlign in the Outs[OIdx].Flags is alway set at this point,
     // so we don't need to worry about natural alignment or not.
     // See TargetLowering::LowerCallTo().
@@ -806,40 +1315,30 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       InFlag
     };
     Chain = DAG.getNode(NVPTXISD::DeclareParam, dl, DeclareParamVTs,
-                        DeclareParamOps, 5);
+                        DeclareParamOps);
     InFlag = Chain.getValue(1);
-    unsigned curOffset = 0;
     for (unsigned j = 0, je = vtparts.size(); j != je; ++j) {
-      unsigned elems = 1;
       EVT elemtype = vtparts[j];
-      if (vtparts[j].isVector()) {
-        elems = vtparts[j].getVectorNumElements();
-        elemtype = vtparts[j].getVectorElementType();
+      int curOffset = Offsets[j];
+      unsigned PartAlign = GreatestCommonDivisor64(ArgAlign, curOffset);
+      SDValue srcAddr =
+          DAG.getNode(ISD::ADD, dl, getPointerTy(), OutVals[OIdx],
+                      DAG.getConstant(curOffset, getPointerTy()));
+      SDValue theVal = DAG.getLoad(elemtype, dl, tempChain, srcAddr,
+                                   MachinePointerInfo(), false, false, false,
+                                   PartAlign);
+      if (elemtype.getSizeInBits() < 16) {
+        theVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i16, theVal);
       }
-      for (unsigned k = 0, ke = elems; k != ke; ++k) {
-        unsigned sz = elemtype.getSizeInBits();
-        if (elemtype.isInteger() && (sz < 8))
-          sz = 8;
-        SDValue srcAddr =
-            DAG.getNode(ISD::ADD, dl, getPointerTy(), OutVals[OIdx],
-                        DAG.getConstant(curOffset, getPointerTy()));
-        SDValue theVal = DAG.getLoad(elemtype, dl, tempChain, srcAddr,
-                                     MachinePointerInfo(), false, false, false,
-                                     0);
-        if (elemtype.getSizeInBits() < 16) {
-          theVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i16, theVal);
-        }
-        SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
-        SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
-                                   DAG.getConstant(curOffset, MVT::i32), theVal,
-                                   InFlag };
-        Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreParam, dl, CopyParamVTs,
-                                        CopyParamOps, 5, elemtype,
-                                        MachinePointerInfo());
+      SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
+      SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
+                                 DAG.getConstant(curOffset, MVT::i32), theVal,
+                                 InFlag };
+      Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreParam, dl, CopyParamVTs,
+                                      CopyParamOps, elemtype,
+                                      MachinePointerInfo());
 
-        InFlag = Chain.getValue(1);
-        curOffset += sz / 8;
-      }
+      InFlag = Chain.getValue(1);
     }
     ++paramCount;
   }
@@ -856,8 +1355,7 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     //  .param .align 16 .b8 retval0[<size-in-bytes>], or
     //  .param .b<size-in-bits> retval0
     unsigned resultsz = TD->getTypeAllocSizeInBits(retTy);
-    if (retTy->isPrimitiveType() || retTy->isIntegerTy() ||
-        retTy->isPointerTy()) {
+    if (retTy->isSingleValueType()) {
       // Scalar needs to be at least 32bit wide
       if (resultsz < 32)
         resultsz = 32;
@@ -866,7 +1364,7 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                                   DAG.getConstant(resultsz, MVT::i32),
                                   DAG.getConstant(0, MVT::i32), InFlag };
       Chain = DAG.getNode(NVPTXISD::DeclareRet, dl, DeclareRetVTs,
-                          DeclareRetOps, 5);
+                          DeclareRetOps);
       InFlag = Chain.getValue(1);
     } else {
       retAlignment = getArgumentAlignment(Callee, CS, retTy, 0);
@@ -876,7 +1374,7 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                                   DAG.getConstant(resultsz / 8, MVT::i32),
                                   DAG.getConstant(0, MVT::i32), InFlag };
       Chain = DAG.getNode(NVPTXISD::DeclareRetParam, dl, DeclareRetVTs,
-                          DeclareRetOps, 5);
+                          DeclareRetOps);
       InFlag = Chain.getValue(1);
     }
   }
@@ -896,7 +1394,7 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     SDValue ProtoOps[] = {
       Chain, DAG.getTargetExternalSymbol(ProtoStr, MVT::i32), InFlag,
     };
-    Chain = DAG.getNode(NVPTXISD::CallPrototype, dl, ProtoVTs, &ProtoOps[0], 3);
+    Chain = DAG.getNode(NVPTXISD::CallPrototype, dl, ProtoVTs, ProtoOps);
     InFlag = Chain.getValue(1);
   }
   // Op to just print "call"
@@ -905,20 +1403,20 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     Chain, DAG.getConstant((Ins.size() == 0) ? 0 : 1, MVT::i32), InFlag
   };
   Chain = DAG.getNode(Func ? (NVPTXISD::PrintCallUni) : (NVPTXISD::PrintCall),
-                      dl, PrintCallVTs, PrintCallOps, 3);
+                      dl, PrintCallVTs, PrintCallOps);
   InFlag = Chain.getValue(1);
 
   // Ops to print out the function name
   SDVTList CallVoidVTs = DAG.getVTList(MVT::Other, MVT::Glue);
   SDValue CallVoidOps[] = { Chain, Callee, InFlag };
-  Chain = DAG.getNode(NVPTXISD::CallVoid, dl, CallVoidVTs, CallVoidOps, 3);
+  Chain = DAG.getNode(NVPTXISD::CallVoid, dl, CallVoidVTs, CallVoidOps);
   InFlag = Chain.getValue(1);
 
   // Ops to print out the param list
   SDVTList CallArgBeginVTs = DAG.getVTList(MVT::Other, MVT::Glue);
   SDValue CallArgBeginOps[] = { Chain, InFlag };
   Chain = DAG.getNode(NVPTXISD::CallArgBegin, dl, CallArgBeginVTs,
-                      CallArgBeginOps, 2);
+                      CallArgBeginOps);
   InFlag = Chain.getValue(1);
 
   for (unsigned i = 0, e = paramCount; i != e; ++i) {
@@ -930,27 +1428,25 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     SDVTList CallArgVTs = DAG.getVTList(MVT::Other, MVT::Glue);
     SDValue CallArgOps[] = { Chain, DAG.getConstant(1, MVT::i32),
                              DAG.getConstant(i, MVT::i32), InFlag };
-    Chain = DAG.getNode(opcode, dl, CallArgVTs, CallArgOps, 4);
+    Chain = DAG.getNode(opcode, dl, CallArgVTs, CallArgOps);
     InFlag = Chain.getValue(1);
   }
   SDVTList CallArgEndVTs = DAG.getVTList(MVT::Other, MVT::Glue);
   SDValue CallArgEndOps[] = { Chain, DAG.getConstant(Func ? 1 : 0, MVT::i32),
                               InFlag };
-  Chain =
-      DAG.getNode(NVPTXISD::CallArgEnd, dl, CallArgEndVTs, CallArgEndOps, 3);
+  Chain = DAG.getNode(NVPTXISD::CallArgEnd, dl, CallArgEndVTs, CallArgEndOps);
   InFlag = Chain.getValue(1);
 
   if (!Func) {
     SDVTList PrototypeVTs = DAG.getVTList(MVT::Other, MVT::Glue);
     SDValue PrototypeOps[] = { Chain, DAG.getConstant(uniqueCallSite, MVT::i32),
                                InFlag };
-    Chain = DAG.getNode(NVPTXISD::Prototype, dl, PrototypeVTs, PrototypeOps, 3);
+    Chain = DAG.getNode(NVPTXISD::Prototype, dl, PrototypeVTs, PrototypeOps);
     InFlag = Chain.getValue(1);
   }
 
   // Generate loads from param memory/moves from registers for result
   if (Ins.size() > 0) {
-    unsigned resoffset = 0;
     if (retTy && retTy->isVectorTy()) {
       EVT ObjectVT = getValueType(retTy);
       unsigned NumElts = ObjectVT.getVectorNumElements();
@@ -959,29 +1455,29 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                                                         ObjectVT) == NumElts &&
              "Vector was not scalarized");
       unsigned sz = EltVT.getSizeInBits();
-      bool needTruncate = sz < 16 ? true : false;
+      bool needTruncate = sz < 8 ? true : false;
 
       if (NumElts == 1) {
         // Just a simple load
-        std::vector<EVT> LoadRetVTs;
-        if (needTruncate) {
-          // If loading i1 result, generate
-          //   load i16
+        SmallVector<EVT, 4> LoadRetVTs;
+        if (EltVT == MVT::i1 || EltVT == MVT::i8) {
+          // If loading i1/i8 result, generate
+          //   load.b8 i16
+          //   if i1
           //   trunc i16 to i1
           LoadRetVTs.push_back(MVT::i16);
         } else
           LoadRetVTs.push_back(EltVT);
         LoadRetVTs.push_back(MVT::Other);
         LoadRetVTs.push_back(MVT::Glue);
-        std::vector<SDValue> LoadRetOps;
+        SmallVector<SDValue, 4> LoadRetOps;
         LoadRetOps.push_back(Chain);
         LoadRetOps.push_back(DAG.getConstant(1, MVT::i32));
         LoadRetOps.push_back(DAG.getConstant(0, MVT::i32));
         LoadRetOps.push_back(InFlag);
         SDValue retval = DAG.getMemIntrinsicNode(
             NVPTXISD::LoadParam, dl,
-            DAG.getVTList(&LoadRetVTs[0], LoadRetVTs.size()), &LoadRetOps[0],
-            LoadRetOps.size(), EltVT, MachinePointerInfo());
+            DAG.getVTList(LoadRetVTs), LoadRetOps, EltVT, MachinePointerInfo());
         Chain = retval.getValue(1);
         InFlag = retval.getValue(2);
         SDValue Ret0 = retval;
@@ -990,10 +1486,11 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
         InVals.push_back(Ret0);
       } else if (NumElts == 2) {
         // LoadV2
-        std::vector<EVT> LoadRetVTs;
-        if (needTruncate) {
-          // If loading i1 result, generate
-          //   load i16
+        SmallVector<EVT, 4> LoadRetVTs;
+        if (EltVT == MVT::i1 || EltVT == MVT::i8) {
+          // If loading i1/i8 result, generate
+          //   load.b8 i16
+          //   if i1
           //   trunc i16 to i1
           LoadRetVTs.push_back(MVT::i16);
           LoadRetVTs.push_back(MVT::i16);
@@ -1003,15 +1500,14 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
         }
         LoadRetVTs.push_back(MVT::Other);
         LoadRetVTs.push_back(MVT::Glue);
-        std::vector<SDValue> LoadRetOps;
+        SmallVector<SDValue, 4> LoadRetOps;
         LoadRetOps.push_back(Chain);
         LoadRetOps.push_back(DAG.getConstant(1, MVT::i32));
         LoadRetOps.push_back(DAG.getConstant(0, MVT::i32));
         LoadRetOps.push_back(InFlag);
         SDValue retval = DAG.getMemIntrinsicNode(
             NVPTXISD::LoadParamV2, dl,
-            DAG.getVTList(&LoadRetVTs[0], LoadRetVTs.size()), &LoadRetOps[0],
-            LoadRetOps.size(), EltVT, MachinePointerInfo());
+            DAG.getVTList(LoadRetVTs), LoadRetOps, EltVT, MachinePointerInfo());
         Chain = retval.getValue(2);
         InFlag = retval.getValue(3);
         SDValue Ret0 = retval.getValue(0);
@@ -1037,9 +1533,10 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
         EVT VecVT = EVT::getVectorVT(F->getContext(), EltVT, VecSize);
         for (unsigned i = 0; i < NumElts; i += VecSize) {
           SmallVector<EVT, 8> LoadRetVTs;
-          if (needTruncate) {
-            // If loading i1 result, generate
-            //   load i16
+          if (EltVT == MVT::i1 || EltVT == MVT::i8) {
+            // If loading i1/i8 result, generate
+            //   load.b8 i16
+            //   if i1
             //   trunc i16 to i1
             for (unsigned j = 0; j < VecSize; ++j)
               LoadRetVTs.push_back(MVT::i16);
@@ -1055,8 +1552,8 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
           LoadRetOps.push_back(DAG.getConstant(Ofst, MVT::i32));
           LoadRetOps.push_back(InFlag);
           SDValue retval = DAG.getMemIntrinsicNode(
-              Opc, dl, DAG.getVTList(&LoadRetVTs[0], LoadRetVTs.size()),
-              &LoadRetOps[0], LoadRetOps.size(), EltVT, MachinePointerInfo());
+              Opc, dl, DAG.getVTList(LoadRetVTs),
+              LoadRetOps, EltVT, MachinePointerInfo());
           if (VecSize == 2) {
             Chain = retval.getValue(2);
             InFlag = retval.getValue(3);
@@ -1078,10 +1575,13 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       }
     } else {
       SmallVector<EVT, 16> VTs;
-      ComputePTXValueVTs(*this, retTy, VTs);
+      SmallVector<uint64_t, 16> Offsets;
+      ComputePTXValueVTs(*this, retTy, VTs, &Offsets, 0);
       assert(VTs.size() == Ins.size() && "Bad value decomposition");
+      unsigned RetAlign = getArgumentAlignment(Callee, CS, retTy, 0);
       for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
         unsigned sz = VTs[i].getSizeInBits();
+        unsigned AlignI = GreatestCommonDivisor64(RetAlign, Offsets[i]);
         bool needTruncate = sz < 8 ? true : false;
         if (VTs[i].isInteger() && (sz < 8))
           sz = 8;
@@ -1107,19 +1607,18 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
         SmallVector<SDValue, 4> LoadRetOps;
         LoadRetOps.push_back(Chain);
         LoadRetOps.push_back(DAG.getConstant(1, MVT::i32));
-        LoadRetOps.push_back(DAG.getConstant(resoffset, MVT::i32));
+        LoadRetOps.push_back(DAG.getConstant(Offsets[i], MVT::i32));
         LoadRetOps.push_back(InFlag);
         SDValue retval = DAG.getMemIntrinsicNode(
             NVPTXISD::LoadParam, dl,
-            DAG.getVTList(&LoadRetVTs[0], LoadRetVTs.size()), &LoadRetOps[0],
-            LoadRetOps.size(), TheLoadType, MachinePointerInfo());
+            DAG.getVTList(LoadRetVTs), LoadRetOps,
+            TheLoadType, MachinePointerInfo(), AlignI);
         Chain = retval.getValue(1);
         InFlag = retval.getValue(2);
         SDValue Ret0 = retval.getValue(0);
         if (needTruncate)
           Ret0 = DAG.getNode(ISD::TRUNCATE, dl, Ins[i].VT, Ret0);
         InVals.push_back(Ret0);
-        resoffset += sz / 8;
       }
     }
   }
@@ -1154,8 +1653,128 @@ NVPTXTargetLowering::LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const {
                                 DAG.getIntPtrConstant(j)));
     }
   }
-  return DAG.getNode(ISD::BUILD_VECTOR, dl, Node->getValueType(0), &Ops[0],
-                     Ops.size());
+  return DAG.getNode(ISD::BUILD_VECTOR, dl, Node->getValueType(0), Ops);
+}
+
+/// LowerShiftRightParts - Lower SRL_PARTS, SRA_PARTS, which
+/// 1) returns two i32 values and take a 2 x i32 value to shift plus a shift
+///    amount, or
+/// 2) returns two i64 values and take a 2 x i64 value to shift plus a shift
+///    amount.
+SDValue NVPTXTargetLowering::LowerShiftRightParts(SDValue Op,
+                                                  SelectionDAG &DAG) const {
+  assert(Op.getNumOperands() == 3 && "Not a double-shift!");
+  assert(Op.getOpcode() == ISD::SRA_PARTS || Op.getOpcode() == ISD::SRL_PARTS);
+
+  EVT VT = Op.getValueType();
+  unsigned VTBits = VT.getSizeInBits();
+  SDLoc dl(Op);
+  SDValue ShOpLo = Op.getOperand(0);
+  SDValue ShOpHi = Op.getOperand(1);
+  SDValue ShAmt  = Op.getOperand(2);
+  unsigned Opc = (Op.getOpcode() == ISD::SRA_PARTS) ? ISD::SRA : ISD::SRL;
+
+  if (VTBits == 32 && nvptxSubtarget.getSmVersion() >= 35) {
+
+    // For 32bit and sm35, we can use the funnel shift 'shf' instruction.
+    // {dHi, dLo} = {aHi, aLo} >> Amt
+    //   dHi = aHi >> Amt
+    //   dLo = shf.r.clamp aLo, aHi, Amt
+
+    SDValue Hi = DAG.getNode(Opc, dl, VT, ShOpHi, ShAmt);
+    SDValue Lo = DAG.getNode(NVPTXISD::FUN_SHFR_CLAMP, dl, VT, ShOpLo, ShOpHi,
+                             ShAmt);
+
+    SDValue Ops[2] = { Lo, Hi };
+    return DAG.getMergeValues(Ops, dl);
+  }
+  else {
+
+    // {dHi, dLo} = {aHi, aLo} >> Amt
+    // - if (Amt>=size) then
+    //      dLo = aHi >> (Amt-size)
+    //      dHi = aHi >> Amt (this is either all 0 or all 1)
+    //   else
+    //      dLo = (aLo >>logic Amt) | (aHi << (size-Amt))
+    //      dHi = aHi >> Amt
+
+    SDValue RevShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32,
+                                   DAG.getConstant(VTBits, MVT::i32), ShAmt);
+    SDValue Tmp1 = DAG.getNode(ISD::SRL, dl, VT, ShOpLo, ShAmt);
+    SDValue ExtraShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32, ShAmt,
+                                     DAG.getConstant(VTBits, MVT::i32));
+    SDValue Tmp2 = DAG.getNode(ISD::SHL, dl, VT, ShOpHi, RevShAmt);
+    SDValue FalseVal = DAG.getNode(ISD::OR, dl, VT, Tmp1, Tmp2);
+    SDValue TrueVal = DAG.getNode(Opc, dl, VT, ShOpHi, ExtraShAmt);
+
+    SDValue Cmp = DAG.getSetCC(dl, MVT::i1, ShAmt,
+                               DAG.getConstant(VTBits, MVT::i32), ISD::SETGE);
+    SDValue Hi = DAG.getNode(Opc, dl, VT, ShOpHi, ShAmt);
+    SDValue Lo = DAG.getNode(ISD::SELECT, dl, VT, Cmp, TrueVal, FalseVal);
+
+    SDValue Ops[2] = { Lo, Hi };
+    return DAG.getMergeValues(Ops, dl);
+  }
+}
+
+/// LowerShiftLeftParts - Lower SHL_PARTS, which
+/// 1) returns two i32 values and take a 2 x i32 value to shift plus a shift
+///    amount, or
+/// 2) returns two i64 values and take a 2 x i64 value to shift plus a shift
+///    amount.
+SDValue NVPTXTargetLowering::LowerShiftLeftParts(SDValue Op,
+                                                 SelectionDAG &DAG) const {
+  assert(Op.getNumOperands() == 3 && "Not a double-shift!");
+  assert(Op.getOpcode() == ISD::SHL_PARTS);
+
+  EVT VT = Op.getValueType();
+  unsigned VTBits = VT.getSizeInBits();
+  SDLoc dl(Op);
+  SDValue ShOpLo = Op.getOperand(0);
+  SDValue ShOpHi = Op.getOperand(1);
+  SDValue ShAmt  = Op.getOperand(2);
+
+  if (VTBits == 32 && nvptxSubtarget.getSmVersion() >= 35) {
+
+    // For 32bit and sm35, we can use the funnel shift 'shf' instruction.
+    // {dHi, dLo} = {aHi, aLo} << Amt
+    //   dHi = shf.l.clamp aLo, aHi, Amt
+    //   dLo = aLo << Amt
+
+    SDValue Hi = DAG.getNode(NVPTXISD::FUN_SHFL_CLAMP, dl, VT, ShOpLo, ShOpHi,
+                             ShAmt);
+    SDValue Lo = DAG.getNode(ISD::SHL, dl, VT, ShOpLo, ShAmt);
+
+    SDValue Ops[2] = { Lo, Hi };
+    return DAG.getMergeValues(Ops, dl);
+  }
+  else {
+
+    // {dHi, dLo} = {aHi, aLo} << Amt
+    // - if (Amt>=size) then
+    //      dLo = aLo << Amt (all 0)
+    //      dLo = aLo << (Amt-size)
+    //   else
+    //      dLo = aLo << Amt
+    //      dHi = (aHi << Amt) | (aLo >> (size-Amt))
+
+    SDValue RevShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32,
+                                   DAG.getConstant(VTBits, MVT::i32), ShAmt);
+    SDValue Tmp1 = DAG.getNode(ISD::SHL, dl, VT, ShOpHi, ShAmt);
+    SDValue ExtraShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32, ShAmt,
+                                     DAG.getConstant(VTBits, MVT::i32));
+    SDValue Tmp2 = DAG.getNode(ISD::SRL, dl, VT, ShOpLo, RevShAmt);
+    SDValue FalseVal = DAG.getNode(ISD::OR, dl, VT, Tmp1, Tmp2);
+    SDValue TrueVal = DAG.getNode(ISD::SHL, dl, VT, ShOpLo, ExtraShAmt);
+
+    SDValue Cmp = DAG.getSetCC(dl, MVT::i1, ShAmt,
+                               DAG.getConstant(VTBits, MVT::i32), ISD::SETGE);
+    SDValue Lo = DAG.getNode(ISD::SHL, dl, VT, ShOpLo, ShAmt);
+    SDValue Hi = DAG.getNode(ISD::SELECT, dl, VT, Cmp, TrueVal, FalseVal);
+
+    SDValue Ops[2] = { Lo, Hi };
+    return DAG.getMergeValues(Ops, dl);
+  }
 }
 
 SDValue
@@ -1178,6 +1797,11 @@ NVPTXTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
     return LowerSTORE(Op, DAG);
   case ISD::LOAD:
     return LowerLOAD(Op, DAG);
+  case ISD::SHL_PARTS:
+    return LowerShiftLeftParts(Op, DAG);
+  case ISD::SRA_PARTS:
+  case ISD::SRL_PARTS:
+    return LowerShiftRightParts(Op, DAG);
   default:
     llvm_unreachable("Custom lowering not defined for operation");
   }
@@ -1210,7 +1834,7 @@ SDValue NVPTXTargetLowering::LowerLOADi1(SDValue Op, SelectionDAG &DAG) const {
   // load, so we build a MergeValues node for it. See ExpandUnalignedLoad()
   // in LegalizeDAG.cpp which also uses MergeValues.
   SDValue Ops[] = { result, LD->getChain() };
-  return DAG.getMergeValues(Ops, 2, dl);
+  return DAG.getMergeValues(Ops, dl);
 }
 
 SDValue NVPTXTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
@@ -1253,13 +1877,28 @@ NVPTXTargetLowering::LowerSTOREVector(SDValue Op, SelectionDAG &DAG) const {
       break;
     }
 
+    MemSDNode *MemSD = cast<MemSDNode>(N);
+    const DataLayout *TD = getDataLayout();
+
+    unsigned Align = MemSD->getAlignment();
+    unsigned PrefAlign =
+      TD->getPrefTypeAlignment(ValVT.getTypeForEVT(*DAG.getContext()));
+    if (Align < PrefAlign) {
+      // This store is not sufficiently aligned, so bail out and let this vector
+      // store be scalarized.  Note that we may still be able to emit smaller
+      // vector stores.  For example, if we are storing a <4 x float> with an
+      // alignment of 8, this check will fail but the legalizer will try again
+      // with 2 x <2 x float>, which will succeed with an alignment of 8.
+      return SDValue();
+    }
+
     unsigned Opcode = 0;
     EVT EltVT = ValVT.getVectorElementType();
     unsigned NumElts = ValVT.getVectorNumElements();
 
     // Since StoreV2 is a target node, we cannot rely on DAG type legalization.
     // Therefore, we must ensure the type is legal.  For i1 and i8, we set the
-    // stored type to i16 and propogate the "real" type as the memory type.
+    // stored type to i16 and propagate the "real" type as the memory type.
     bool NeedExt = false;
     if (EltVT.getSizeInBits() < 16)
       NeedExt = true;
@@ -1295,10 +1934,8 @@ NVPTXTargetLowering::LowerSTOREVector(SDValue Op, SelectionDAG &DAG) const {
       Ops.push_back(N->getOperand(i));
     }
 
-    MemSDNode *MemSD = cast<MemSDNode>(N);
-
     SDValue NewSt = DAG.getMemIntrinsicNode(
-        Opcode, DL, DAG.getVTList(MVT::Other), &Ops[0], Ops.size(),
+        Opcode, DL, DAG.getVTList(MVT::Other), Ops,
         MemSD->getMemoryVT(), MemSD->getMemOperand());
 
     //return DCI.CombineTo(N, NewSt, true);
@@ -1391,7 +2028,7 @@ SDValue NVPTXTargetLowering::LowerFormalArguments(
 
   const Function *F = MF.getFunction();
   const AttributeSet &PAL = F->getAttributes();
-  const TargetLowering *TLI = nvTM->getTargetLowering();
+  const TargetLowering *TLI = DAG.getTarget().getTargetLowering();
 
   SDValue Root = DAG.getRoot();
   std::vector<SDValue> OutChains;
@@ -1430,7 +2067,7 @@ SDValue NVPTXTargetLowering::LowerFormalArguments(
     if (isImageOrSamplerVal(
             theArgs[i],
             (theArgs[i]->getParent() ? theArgs[i]->getParent()->getParent()
-                                     : 0))) {
+                                     : nullptr))) {
       assert(isKernel && "Only kernels can have image/sampler params");
       InVals.push_back(DAG.getConstant(i + 1, MVT::i32));
       continue;
@@ -1445,8 +2082,7 @@ SDValue NVPTXTargetLowering::LowerFormalArguments(
         assert(vtparts.size() > 0 && "empty aggregate type not expected");
         for (unsigned parti = 0, parte = vtparts.size(); parti != parte;
              ++parti) {
-          EVT partVT = vtparts[parti];
-          InVals.push_back(DAG.getNode(ISD::UNDEF, dl, partVT));
+          InVals.push_back(DAG.getNode(ISD::UNDEF, dl, Ins[InsIdx].VT));
           ++InsIdx;
         }
         if (vtparts.size() > 0)
@@ -1684,8 +2320,7 @@ SDValue NVPTXTargetLowering::LowerFormalArguments(
   //}
 
   if (!OutChains.empty())
-    DAG.setRoot(DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &OutChains[0],
-                            OutChains.size()));
+    DAG.setRoot(DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains));
 
   return Chain;
 }
@@ -1727,7 +2362,7 @@ NVPTXTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
         StoreVal = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i16, StoreVal);
       SDValue Ops[] = { Chain, DAG.getConstant(0, MVT::i32), StoreVal };
       Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreRetval, dl,
-                                      DAG.getVTList(MVT::Other), &Ops[0], 3,
+                                      DAG.getVTList(MVT::Other), Ops,
                                       EltVT, MachinePointerInfo());
 
     } else if (NumElts == 2) {
@@ -1743,7 +2378,7 @@ NVPTXTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
       SDValue Ops[] = { Chain, DAG.getConstant(0, MVT::i32), StoreVal0,
                         StoreVal1 };
       Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreRetvalV2, dl,
-                                      DAG.getVTList(MVT::Other), &Ops[0], 4,
+                                      DAG.getVTList(MVT::Other), Ops,
                                       EltVT, MachinePointerInfo());
     } else {
       // V4 stores
@@ -1763,7 +2398,7 @@ NVPTXTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
       unsigned Offset = 0;
 
       EVT VecVT =
-          EVT::getVectorVT(F->getContext(), OutVals[0].getValueType(), VecSize);
+          EVT::getVectorVT(F->getContext(), EltVT, VecSize);
       unsigned PerStoreOffset =
           TD->getTypeAllocSize(VecVT.getTypeForEVT(F->getContext()));
 
@@ -1815,19 +2450,17 @@ NVPTXTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
 
         // Chain = DAG.getNode(Opc, dl, MVT::Other, &Ops[0], Ops.size());
         Chain =
-            DAG.getMemIntrinsicNode(Opc, dl, DAG.getVTList(MVT::Other), &Ops[0],
-                                    Ops.size(), EltVT, MachinePointerInfo());
+            DAG.getMemIntrinsicNode(Opc, dl, DAG.getVTList(MVT::Other), Ops,
+                                    EltVT, MachinePointerInfo());
         Offset += PerStoreOffset;
       }
     }
   } else {
     SmallVector<EVT, 16> ValVTs;
-    // const_cast is necessary since we are still using an LLVM version from
-    // before the type system re-write.
-    ComputePTXValueVTs(*this, RetTy, ValVTs);
+    SmallVector<uint64_t, 16> Offsets;
+    ComputePTXValueVTs(*this, RetTy, ValVTs, &Offsets, 0);
     assert(ValVTs.size() == OutVals.size() && "Bad return value decomposition");
 
-    unsigned SizeSoFar = 0;
     for (unsigned i = 0, e = Outs.size(); i != e; ++i) {
       SDValue theVal = OutVals[i];
       EVT TheValType = theVal.getValueType();
@@ -1851,16 +2484,14 @@ NVPTXTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
         else if (TmpVal.getValueType().getSizeInBits() < 16)
           TmpVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i16, TmpVal);
 
-        SDValue Ops[] = { Chain, DAG.getConstant(SizeSoFar, MVT::i32), TmpVal };
+        SDValue Ops[] = {
+          Chain,
+          DAG.getConstant(Offsets[i], MVT::i32),
+          TmpVal };
         Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreRetval, dl,
-                                        DAG.getVTList(MVT::Other), &Ops[0],
-                                        3, TheStoreType,
+                                        DAG.getVTList(MVT::Other), Ops,
+                                        TheStoreType,
                                         MachinePointerInfo());
-        if(TheValType.isVector())
-          SizeSoFar += 
-            TheStoreType.getVectorElementType().getStoreSizeInBits() / 8;
-        else
-          SizeSoFar += TheStoreType.getStoreSizeInBits()/8;
       }
     }
   }
@@ -1892,6 +2523,702 @@ bool NVPTXTargetLowering::isTypeSupportedInIntrinsic(MVT VT) const {
   return false;
 }
 
+static unsigned getOpcForTextureInstr(unsigned Intrinsic) {
+  switch (Intrinsic) {
+  default:
+    return 0;
+
+  case Intrinsic::nvvm_tex_1d_v4f32_s32:
+    return NVPTXISD::Tex1DFloatS32;
+  case Intrinsic::nvvm_tex_1d_v4f32_f32:
+    return NVPTXISD::Tex1DFloatFloat;
+  case Intrinsic::nvvm_tex_1d_level_v4f32_f32:
+    return NVPTXISD::Tex1DFloatFloatLevel;
+  case Intrinsic::nvvm_tex_1d_grad_v4f32_f32:
+    return NVPTXISD::Tex1DFloatFloatGrad;
+  case Intrinsic::nvvm_tex_1d_v4s32_s32:
+    return NVPTXISD::Tex1DS32S32;
+  case Intrinsic::nvvm_tex_1d_v4s32_f32:
+    return NVPTXISD::Tex1DS32Float;
+  case Intrinsic::nvvm_tex_1d_level_v4s32_f32:
+    return NVPTXISD::Tex1DS32FloatLevel;
+  case Intrinsic::nvvm_tex_1d_grad_v4s32_f32:
+    return NVPTXISD::Tex1DS32FloatGrad;
+  case Intrinsic::nvvm_tex_1d_v4u32_s32:
+    return NVPTXISD::Tex1DU32S32;
+  case Intrinsic::nvvm_tex_1d_v4u32_f32:
+    return NVPTXISD::Tex1DU32Float;
+  case Intrinsic::nvvm_tex_1d_level_v4u32_f32:
+    return NVPTXISD::Tex1DU32FloatLevel;
+  case Intrinsic::nvvm_tex_1d_grad_v4u32_f32:
+    return NVPTXISD::Tex1DU32FloatGrad;
+
+  case Intrinsic::nvvm_tex_1d_array_v4f32_s32:
+    return NVPTXISD::Tex1DArrayFloatS32;
+  case Intrinsic::nvvm_tex_1d_array_v4f32_f32:
+    return NVPTXISD::Tex1DArrayFloatFloat;
+  case Intrinsic::nvvm_tex_1d_array_level_v4f32_f32:
+    return NVPTXISD::Tex1DArrayFloatFloatLevel;
+  case Intrinsic::nvvm_tex_1d_array_grad_v4f32_f32:
+    return NVPTXISD::Tex1DArrayFloatFloatGrad;
+  case Intrinsic::nvvm_tex_1d_array_v4s32_s32:
+    return NVPTXISD::Tex1DArrayS32S32;
+  case Intrinsic::nvvm_tex_1d_array_v4s32_f32:
+    return NVPTXISD::Tex1DArrayS32Float;
+  case Intrinsic::nvvm_tex_1d_array_level_v4s32_f32:
+    return NVPTXISD::Tex1DArrayS32FloatLevel;
+  case Intrinsic::nvvm_tex_1d_array_grad_v4s32_f32:
+    return NVPTXISD::Tex1DArrayS32FloatGrad;
+  case Intrinsic::nvvm_tex_1d_array_v4u32_s32:
+    return NVPTXISD::Tex1DArrayU32S32;
+  case Intrinsic::nvvm_tex_1d_array_v4u32_f32:
+    return NVPTXISD::Tex1DArrayU32Float;
+  case Intrinsic::nvvm_tex_1d_array_level_v4u32_f32:
+    return NVPTXISD::Tex1DArrayU32FloatLevel;
+  case Intrinsic::nvvm_tex_1d_array_grad_v4u32_f32:
+    return NVPTXISD::Tex1DArrayU32FloatGrad;
+
+  case Intrinsic::nvvm_tex_2d_v4f32_s32:
+    return NVPTXISD::Tex2DFloatS32;
+  case Intrinsic::nvvm_tex_2d_v4f32_f32:
+    return NVPTXISD::Tex2DFloatFloat;
+  case Intrinsic::nvvm_tex_2d_level_v4f32_f32:
+    return NVPTXISD::Tex2DFloatFloatLevel;
+  case Intrinsic::nvvm_tex_2d_grad_v4f32_f32:
+    return NVPTXISD::Tex2DFloatFloatGrad;
+  case Intrinsic::nvvm_tex_2d_v4s32_s32:
+    return NVPTXISD::Tex2DS32S32;
+  case Intrinsic::nvvm_tex_2d_v4s32_f32:
+    return NVPTXISD::Tex2DS32Float;
+  case Intrinsic::nvvm_tex_2d_level_v4s32_f32:
+    return NVPTXISD::Tex2DS32FloatLevel;
+  case Intrinsic::nvvm_tex_2d_grad_v4s32_f32:
+    return NVPTXISD::Tex2DS32FloatGrad;
+  case Intrinsic::nvvm_tex_2d_v4u32_s32:
+    return NVPTXISD::Tex2DU32S32;
+  case Intrinsic::nvvm_tex_2d_v4u32_f32:
+    return NVPTXISD::Tex2DU32Float;
+  case Intrinsic::nvvm_tex_2d_level_v4u32_f32:
+    return NVPTXISD::Tex2DU32FloatLevel;
+  case Intrinsic::nvvm_tex_2d_grad_v4u32_f32:
+    return NVPTXISD::Tex2DU32FloatGrad;
+
+  case Intrinsic::nvvm_tex_2d_array_v4f32_s32:
+    return NVPTXISD::Tex2DArrayFloatS32;
+  case Intrinsic::nvvm_tex_2d_array_v4f32_f32:
+    return NVPTXISD::Tex2DArrayFloatFloat;
+  case Intrinsic::nvvm_tex_2d_array_level_v4f32_f32:
+    return NVPTXISD::Tex2DArrayFloatFloatLevel;
+  case Intrinsic::nvvm_tex_2d_array_grad_v4f32_f32:
+    return NVPTXISD::Tex2DArrayFloatFloatGrad;
+  case Intrinsic::nvvm_tex_2d_array_v4s32_s32:
+    return NVPTXISD::Tex2DArrayS32S32;
+  case Intrinsic::nvvm_tex_2d_array_v4s32_f32:
+    return NVPTXISD::Tex2DArrayS32Float;
+  case Intrinsic::nvvm_tex_2d_array_level_v4s32_f32:
+    return NVPTXISD::Tex2DArrayS32FloatLevel;
+  case Intrinsic::nvvm_tex_2d_array_grad_v4s32_f32:
+    return NVPTXISD::Tex2DArrayS32FloatGrad;
+  case Intrinsic::nvvm_tex_2d_array_v4u32_s32:
+    return NVPTXISD::Tex2DArrayU32S32;
+  case Intrinsic::nvvm_tex_2d_array_v4u32_f32:
+    return NVPTXISD::Tex2DArrayU32Float;
+  case Intrinsic::nvvm_tex_2d_array_level_v4u32_f32:
+    return NVPTXISD::Tex2DArrayU32FloatLevel;
+  case Intrinsic::nvvm_tex_2d_array_grad_v4u32_f32:
+    return NVPTXISD::Tex2DArrayU32FloatGrad;
+
+  case Intrinsic::nvvm_tex_3d_v4f32_s32:
+    return NVPTXISD::Tex3DFloatS32;
+  case Intrinsic::nvvm_tex_3d_v4f32_f32:
+    return NVPTXISD::Tex3DFloatFloat;
+  case Intrinsic::nvvm_tex_3d_level_v4f32_f32:
+    return NVPTXISD::Tex3DFloatFloatLevel;
+  case Intrinsic::nvvm_tex_3d_grad_v4f32_f32:
+    return NVPTXISD::Tex3DFloatFloatGrad;
+  case Intrinsic::nvvm_tex_3d_v4s32_s32:
+    return NVPTXISD::Tex3DS32S32;
+  case Intrinsic::nvvm_tex_3d_v4s32_f32:
+    return NVPTXISD::Tex3DS32Float;
+  case Intrinsic::nvvm_tex_3d_level_v4s32_f32:
+    return NVPTXISD::Tex3DS32FloatLevel;
+  case Intrinsic::nvvm_tex_3d_grad_v4s32_f32:
+    return NVPTXISD::Tex3DS32FloatGrad;
+  case Intrinsic::nvvm_tex_3d_v4u32_s32:
+    return NVPTXISD::Tex3DU32S32;
+  case Intrinsic::nvvm_tex_3d_v4u32_f32:
+    return NVPTXISD::Tex3DU32Float;
+  case Intrinsic::nvvm_tex_3d_level_v4u32_f32:
+    return NVPTXISD::Tex3DU32FloatLevel;
+  case Intrinsic::nvvm_tex_3d_grad_v4u32_f32:
+    return NVPTXISD::Tex3DU32FloatGrad;
+
+  case Intrinsic::nvvm_tex_cube_v4f32_f32:
+    return NVPTXISD::TexCubeFloatFloat;
+  case Intrinsic::nvvm_tex_cube_level_v4f32_f32:
+    return NVPTXISD::TexCubeFloatFloatLevel;
+  case Intrinsic::nvvm_tex_cube_v4s32_f32:
+    return NVPTXISD::TexCubeS32Float;
+  case Intrinsic::nvvm_tex_cube_level_v4s32_f32:
+    return NVPTXISD::TexCubeS32FloatLevel;
+  case Intrinsic::nvvm_tex_cube_v4u32_f32:
+    return NVPTXISD::TexCubeU32Float;
+  case Intrinsic::nvvm_tex_cube_level_v4u32_f32:
+    return NVPTXISD::TexCubeU32FloatLevel;
+
+  case Intrinsic::nvvm_tex_cube_array_v4f32_f32:
+    return NVPTXISD::TexCubeArrayFloatFloat;
+  case Intrinsic::nvvm_tex_cube_array_level_v4f32_f32:
+    return NVPTXISD::TexCubeArrayFloatFloatLevel;
+  case Intrinsic::nvvm_tex_cube_array_v4s32_f32:
+    return NVPTXISD::TexCubeArrayS32Float;
+  case Intrinsic::nvvm_tex_cube_array_level_v4s32_f32:
+    return NVPTXISD::TexCubeArrayS32FloatLevel;
+  case Intrinsic::nvvm_tex_cube_array_v4u32_f32:
+    return NVPTXISD::TexCubeArrayU32Float;
+  case Intrinsic::nvvm_tex_cube_array_level_v4u32_f32:
+    return NVPTXISD::TexCubeArrayU32FloatLevel;
+
+  case Intrinsic::nvvm_tld4_r_2d_v4f32_f32:
+    return NVPTXISD::Tld4R2DFloatFloat;
+  case Intrinsic::nvvm_tld4_g_2d_v4f32_f32:
+    return NVPTXISD::Tld4G2DFloatFloat;
+  case Intrinsic::nvvm_tld4_b_2d_v4f32_f32:
+    return NVPTXISD::Tld4B2DFloatFloat;
+  case Intrinsic::nvvm_tld4_a_2d_v4f32_f32:
+    return NVPTXISD::Tld4A2DFloatFloat;
+  case Intrinsic::nvvm_tld4_r_2d_v4s32_f32:
+    return NVPTXISD::Tld4R2DS64Float;
+  case Intrinsic::nvvm_tld4_g_2d_v4s32_f32:
+    return NVPTXISD::Tld4G2DS64Float;
+  case Intrinsic::nvvm_tld4_b_2d_v4s32_f32:
+    return NVPTXISD::Tld4B2DS64Float;
+  case Intrinsic::nvvm_tld4_a_2d_v4s32_f32:
+    return NVPTXISD::Tld4A2DS64Float;
+  case Intrinsic::nvvm_tld4_r_2d_v4u32_f32:
+    return NVPTXISD::Tld4R2DU64Float;
+  case Intrinsic::nvvm_tld4_g_2d_v4u32_f32:
+    return NVPTXISD::Tld4G2DU64Float;
+  case Intrinsic::nvvm_tld4_b_2d_v4u32_f32:
+    return NVPTXISD::Tld4B2DU64Float;
+  case Intrinsic::nvvm_tld4_a_2d_v4u32_f32:
+    return NVPTXISD::Tld4A2DU64Float;
+
+  case Intrinsic::nvvm_tex_unified_1d_v4f32_s32:
+    return NVPTXISD::TexUnified1DFloatS32;
+  case Intrinsic::nvvm_tex_unified_1d_v4f32_f32:
+    return NVPTXISD::TexUnified1DFloatFloat;
+  case Intrinsic::nvvm_tex_unified_1d_level_v4f32_f32:
+    return NVPTXISD::TexUnified1DFloatFloatLevel;
+  case Intrinsic::nvvm_tex_unified_1d_grad_v4f32_f32:
+    return NVPTXISD::TexUnified1DFloatFloatGrad;
+  case Intrinsic::nvvm_tex_unified_1d_v4s32_s32:
+    return NVPTXISD::TexUnified1DS32S32;
+  case Intrinsic::nvvm_tex_unified_1d_v4s32_f32:
+    return NVPTXISD::TexUnified1DS32Float;
+  case Intrinsic::nvvm_tex_unified_1d_level_v4s32_f32:
+    return NVPTXISD::TexUnified1DS32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_1d_grad_v4s32_f32:
+    return NVPTXISD::TexUnified1DS32FloatGrad;
+  case Intrinsic::nvvm_tex_unified_1d_v4u32_s32:
+    return NVPTXISD::TexUnified1DU32S32;
+  case Intrinsic::nvvm_tex_unified_1d_v4u32_f32:
+    return NVPTXISD::TexUnified1DU32Float;
+  case Intrinsic::nvvm_tex_unified_1d_level_v4u32_f32:
+    return NVPTXISD::TexUnified1DU32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_1d_grad_v4u32_f32:
+    return NVPTXISD::TexUnified1DU32FloatGrad;
+
+  case Intrinsic::nvvm_tex_unified_1d_array_v4f32_s32:
+    return NVPTXISD::TexUnified1DArrayFloatS32;
+  case Intrinsic::nvvm_tex_unified_1d_array_v4f32_f32:
+    return NVPTXISD::TexUnified1DArrayFloatFloat;
+  case Intrinsic::nvvm_tex_unified_1d_array_level_v4f32_f32:
+    return NVPTXISD::TexUnified1DArrayFloatFloatLevel;
+  case Intrinsic::nvvm_tex_unified_1d_array_grad_v4f32_f32:
+    return NVPTXISD::TexUnified1DArrayFloatFloatGrad;
+  case Intrinsic::nvvm_tex_unified_1d_array_v4s32_s32:
+    return NVPTXISD::TexUnified1DArrayS32S32;
+  case Intrinsic::nvvm_tex_unified_1d_array_v4s32_f32:
+    return NVPTXISD::TexUnified1DArrayS32Float;
+  case Intrinsic::nvvm_tex_unified_1d_array_level_v4s32_f32:
+    return NVPTXISD::TexUnified1DArrayS32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_1d_array_grad_v4s32_f32:
+    return NVPTXISD::TexUnified1DArrayS32FloatGrad;
+  case Intrinsic::nvvm_tex_unified_1d_array_v4u32_s32:
+    return NVPTXISD::TexUnified1DArrayU32S32;
+  case Intrinsic::nvvm_tex_unified_1d_array_v4u32_f32:
+    return NVPTXISD::TexUnified1DArrayU32Float;
+  case Intrinsic::nvvm_tex_unified_1d_array_level_v4u32_f32:
+    return NVPTXISD::TexUnified1DArrayU32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_1d_array_grad_v4u32_f32:
+    return NVPTXISD::TexUnified1DArrayU32FloatGrad;
+
+  case Intrinsic::nvvm_tex_unified_2d_v4f32_s32:
+    return NVPTXISD::TexUnified2DFloatS32;
+  case Intrinsic::nvvm_tex_unified_2d_v4f32_f32:
+    return NVPTXISD::TexUnified2DFloatFloat;
+  case Intrinsic::nvvm_tex_unified_2d_level_v4f32_f32:
+    return NVPTXISD::TexUnified2DFloatFloatLevel;
+  case Intrinsic::nvvm_tex_unified_2d_grad_v4f32_f32:
+    return NVPTXISD::TexUnified2DFloatFloatGrad;
+  case Intrinsic::nvvm_tex_unified_2d_v4s32_s32:
+    return NVPTXISD::TexUnified2DS32S32;
+  case Intrinsic::nvvm_tex_unified_2d_v4s32_f32:
+    return NVPTXISD::TexUnified2DS32Float;
+  case Intrinsic::nvvm_tex_unified_2d_level_v4s32_f32:
+    return NVPTXISD::TexUnified2DS32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_2d_grad_v4s32_f32:
+    return NVPTXISD::TexUnified2DS32FloatGrad;
+  case Intrinsic::nvvm_tex_unified_2d_v4u32_s32:
+    return NVPTXISD::TexUnified2DU32S32;
+  case Intrinsic::nvvm_tex_unified_2d_v4u32_f32:
+    return NVPTXISD::TexUnified2DU32Float;
+  case Intrinsic::nvvm_tex_unified_2d_level_v4u32_f32:
+    return NVPTXISD::TexUnified2DU32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_2d_grad_v4u32_f32:
+    return NVPTXISD::TexUnified2DU32FloatGrad;
+
+  case Intrinsic::nvvm_tex_unified_2d_array_v4f32_s32:
+    return NVPTXISD::TexUnified2DArrayFloatS32;
+  case Intrinsic::nvvm_tex_unified_2d_array_v4f32_f32:
+    return NVPTXISD::TexUnified2DArrayFloatFloat;
+  case Intrinsic::nvvm_tex_unified_2d_array_level_v4f32_f32:
+    return NVPTXISD::TexUnified2DArrayFloatFloatLevel;
+  case Intrinsic::nvvm_tex_unified_2d_array_grad_v4f32_f32:
+    return NVPTXISD::TexUnified2DArrayFloatFloatGrad;
+  case Intrinsic::nvvm_tex_unified_2d_array_v4s32_s32:
+    return NVPTXISD::TexUnified2DArrayS32S32;
+  case Intrinsic::nvvm_tex_unified_2d_array_v4s32_f32:
+    return NVPTXISD::TexUnified2DArrayS32Float;
+  case Intrinsic::nvvm_tex_unified_2d_array_level_v4s32_f32:
+    return NVPTXISD::TexUnified2DArrayS32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_2d_array_grad_v4s32_f32:
+    return NVPTXISD::TexUnified2DArrayS32FloatGrad;
+  case Intrinsic::nvvm_tex_unified_2d_array_v4u32_s32:
+    return NVPTXISD::TexUnified2DArrayU32S32;
+  case Intrinsic::nvvm_tex_unified_2d_array_v4u32_f32:
+    return NVPTXISD::TexUnified2DArrayU32Float;
+  case Intrinsic::nvvm_tex_unified_2d_array_level_v4u32_f32:
+    return NVPTXISD::TexUnified2DArrayU32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_2d_array_grad_v4u32_f32:
+    return NVPTXISD::TexUnified2DArrayU32FloatGrad;
+
+  case Intrinsic::nvvm_tex_unified_3d_v4f32_s32:
+    return NVPTXISD::TexUnified3DFloatS32;
+  case Intrinsic::nvvm_tex_unified_3d_v4f32_f32:
+    return NVPTXISD::TexUnified3DFloatFloat;
+  case Intrinsic::nvvm_tex_unified_3d_level_v4f32_f32:
+    return NVPTXISD::TexUnified3DFloatFloatLevel;
+  case Intrinsic::nvvm_tex_unified_3d_grad_v4f32_f32:
+    return NVPTXISD::TexUnified3DFloatFloatGrad;
+  case Intrinsic::nvvm_tex_unified_3d_v4s32_s32:
+    return NVPTXISD::TexUnified3DS32S32;
+  case Intrinsic::nvvm_tex_unified_3d_v4s32_f32:
+    return NVPTXISD::TexUnified3DS32Float;
+  case Intrinsic::nvvm_tex_unified_3d_level_v4s32_f32:
+    return NVPTXISD::TexUnified3DS32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_3d_grad_v4s32_f32:
+    return NVPTXISD::TexUnified3DS32FloatGrad;
+  case Intrinsic::nvvm_tex_unified_3d_v4u32_s32:
+    return NVPTXISD::TexUnified3DU32S32;
+  case Intrinsic::nvvm_tex_unified_3d_v4u32_f32:
+    return NVPTXISD::TexUnified3DU32Float;
+  case Intrinsic::nvvm_tex_unified_3d_level_v4u32_f32:
+    return NVPTXISD::TexUnified3DU32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_3d_grad_v4u32_f32:
+    return NVPTXISD::TexUnified3DU32FloatGrad;
+
+  case Intrinsic::nvvm_tex_unified_cube_v4f32_f32:
+    return NVPTXISD::TexUnifiedCubeFloatFloat;
+  case Intrinsic::nvvm_tex_unified_cube_level_v4f32_f32:
+    return NVPTXISD::TexUnifiedCubeFloatFloatLevel;
+  case Intrinsic::nvvm_tex_unified_cube_v4s32_f32:
+    return NVPTXISD::TexUnifiedCubeS32Float;
+  case Intrinsic::nvvm_tex_unified_cube_level_v4s32_f32:
+    return NVPTXISD::TexUnifiedCubeS32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_cube_v4u32_f32:
+    return NVPTXISD::TexUnifiedCubeU32Float;
+  case Intrinsic::nvvm_tex_unified_cube_level_v4u32_f32:
+    return NVPTXISD::TexUnifiedCubeU32FloatLevel;
+
+  case Intrinsic::nvvm_tex_unified_cube_array_v4f32_f32:
+    return NVPTXISD::TexUnifiedCubeArrayFloatFloat;
+  case Intrinsic::nvvm_tex_unified_cube_array_level_v4f32_f32:
+    return NVPTXISD::TexUnifiedCubeArrayFloatFloatLevel;
+  case Intrinsic::nvvm_tex_unified_cube_array_v4s32_f32:
+    return NVPTXISD::TexUnifiedCubeArrayS32Float;
+  case Intrinsic::nvvm_tex_unified_cube_array_level_v4s32_f32:
+    return NVPTXISD::TexUnifiedCubeArrayS32FloatLevel;
+  case Intrinsic::nvvm_tex_unified_cube_array_v4u32_f32:
+    return NVPTXISD::TexUnifiedCubeArrayU32Float;
+  case Intrinsic::nvvm_tex_unified_cube_array_level_v4u32_f32:
+    return NVPTXISD::TexUnifiedCubeArrayU32FloatLevel;
+
+  case Intrinsic::nvvm_tld4_unified_r_2d_v4f32_f32:
+    return NVPTXISD::Tld4UnifiedR2DFloatFloat;
+  case Intrinsic::nvvm_tld4_unified_g_2d_v4f32_f32:
+    return NVPTXISD::Tld4UnifiedG2DFloatFloat;
+  case Intrinsic::nvvm_tld4_unified_b_2d_v4f32_f32:
+    return NVPTXISD::Tld4UnifiedB2DFloatFloat;
+  case Intrinsic::nvvm_tld4_unified_a_2d_v4f32_f32:
+    return NVPTXISD::Tld4UnifiedA2DFloatFloat;
+  case Intrinsic::nvvm_tld4_unified_r_2d_v4s32_f32:
+    return NVPTXISD::Tld4UnifiedR2DS64Float;
+  case Intrinsic::nvvm_tld4_unified_g_2d_v4s32_f32:
+    return NVPTXISD::Tld4UnifiedG2DS64Float;
+  case Intrinsic::nvvm_tld4_unified_b_2d_v4s32_f32:
+    return NVPTXISD::Tld4UnifiedB2DS64Float;
+  case Intrinsic::nvvm_tld4_unified_a_2d_v4s32_f32:
+    return NVPTXISD::Tld4UnifiedA2DS64Float;
+  case Intrinsic::nvvm_tld4_unified_r_2d_v4u32_f32:
+    return NVPTXISD::Tld4UnifiedR2DU64Float;
+  case Intrinsic::nvvm_tld4_unified_g_2d_v4u32_f32:
+    return NVPTXISD::Tld4UnifiedG2DU64Float;
+  case Intrinsic::nvvm_tld4_unified_b_2d_v4u32_f32:
+    return NVPTXISD::Tld4UnifiedB2DU64Float;
+  case Intrinsic::nvvm_tld4_unified_a_2d_v4u32_f32:
+    return NVPTXISD::Tld4UnifiedA2DU64Float;
+  }
+}
+
+static unsigned getOpcForSurfaceInstr(unsigned Intrinsic) {
+  switch (Intrinsic) {
+  default:
+    return 0;
+  case Intrinsic::nvvm_suld_1d_i8_clamp:
+    return NVPTXISD::Suld1DI8Clamp;
+  case Intrinsic::nvvm_suld_1d_i16_clamp:
+    return NVPTXISD::Suld1DI16Clamp;
+  case Intrinsic::nvvm_suld_1d_i32_clamp:
+    return NVPTXISD::Suld1DI32Clamp;
+  case Intrinsic::nvvm_suld_1d_i64_clamp:
+    return NVPTXISD::Suld1DI64Clamp;
+  case Intrinsic::nvvm_suld_1d_v2i8_clamp:
+    return NVPTXISD::Suld1DV2I8Clamp;
+  case Intrinsic::nvvm_suld_1d_v2i16_clamp:
+    return NVPTXISD::Suld1DV2I16Clamp;
+  case Intrinsic::nvvm_suld_1d_v2i32_clamp:
+    return NVPTXISD::Suld1DV2I32Clamp;
+  case Intrinsic::nvvm_suld_1d_v2i64_clamp:
+    return NVPTXISD::Suld1DV2I64Clamp;
+  case Intrinsic::nvvm_suld_1d_v4i8_clamp:
+    return NVPTXISD::Suld1DV4I8Clamp;
+  case Intrinsic::nvvm_suld_1d_v4i16_clamp:
+    return NVPTXISD::Suld1DV4I16Clamp;
+  case Intrinsic::nvvm_suld_1d_v4i32_clamp:
+    return NVPTXISD::Suld1DV4I32Clamp;
+  case Intrinsic::nvvm_suld_1d_array_i8_clamp:
+    return NVPTXISD::Suld1DArrayI8Clamp;
+  case Intrinsic::nvvm_suld_1d_array_i16_clamp:
+    return NVPTXISD::Suld1DArrayI16Clamp;
+  case Intrinsic::nvvm_suld_1d_array_i32_clamp:
+    return NVPTXISD::Suld1DArrayI32Clamp;
+  case Intrinsic::nvvm_suld_1d_array_i64_clamp:
+    return NVPTXISD::Suld1DArrayI64Clamp;
+  case Intrinsic::nvvm_suld_1d_array_v2i8_clamp:
+    return NVPTXISD::Suld1DArrayV2I8Clamp;
+  case Intrinsic::nvvm_suld_1d_array_v2i16_clamp:
+    return NVPTXISD::Suld1DArrayV2I16Clamp;
+  case Intrinsic::nvvm_suld_1d_array_v2i32_clamp:
+    return NVPTXISD::Suld1DArrayV2I32Clamp;
+  case Intrinsic::nvvm_suld_1d_array_v2i64_clamp:
+    return NVPTXISD::Suld1DArrayV2I64Clamp;
+  case Intrinsic::nvvm_suld_1d_array_v4i8_clamp:
+    return NVPTXISD::Suld1DArrayV4I8Clamp;
+  case Intrinsic::nvvm_suld_1d_array_v4i16_clamp:
+    return NVPTXISD::Suld1DArrayV4I16Clamp;
+  case Intrinsic::nvvm_suld_1d_array_v4i32_clamp:
+    return NVPTXISD::Suld1DArrayV4I32Clamp;
+  case Intrinsic::nvvm_suld_2d_i8_clamp:
+    return NVPTXISD::Suld2DI8Clamp;
+  case Intrinsic::nvvm_suld_2d_i16_clamp:
+    return NVPTXISD::Suld2DI16Clamp;
+  case Intrinsic::nvvm_suld_2d_i32_clamp:
+    return NVPTXISD::Suld2DI32Clamp;
+  case Intrinsic::nvvm_suld_2d_i64_clamp:
+    return NVPTXISD::Suld2DI64Clamp;
+  case Intrinsic::nvvm_suld_2d_v2i8_clamp:
+    return NVPTXISD::Suld2DV2I8Clamp;
+  case Intrinsic::nvvm_suld_2d_v2i16_clamp:
+    return NVPTXISD::Suld2DV2I16Clamp;
+  case Intrinsic::nvvm_suld_2d_v2i32_clamp:
+    return NVPTXISD::Suld2DV2I32Clamp;
+  case Intrinsic::nvvm_suld_2d_v2i64_clamp:
+    return NVPTXISD::Suld2DV2I64Clamp;
+  case Intrinsic::nvvm_suld_2d_v4i8_clamp:
+    return NVPTXISD::Suld2DV4I8Clamp;
+  case Intrinsic::nvvm_suld_2d_v4i16_clamp:
+    return NVPTXISD::Suld2DV4I16Clamp;
+  case Intrinsic::nvvm_suld_2d_v4i32_clamp:
+    return NVPTXISD::Suld2DV4I32Clamp;
+  case Intrinsic::nvvm_suld_2d_array_i8_clamp:
+    return NVPTXISD::Suld2DArrayI8Clamp;
+  case Intrinsic::nvvm_suld_2d_array_i16_clamp:
+    return NVPTXISD::Suld2DArrayI16Clamp;
+  case Intrinsic::nvvm_suld_2d_array_i32_clamp:
+    return NVPTXISD::Suld2DArrayI32Clamp;
+  case Intrinsic::nvvm_suld_2d_array_i64_clamp:
+    return NVPTXISD::Suld2DArrayI64Clamp;
+  case Intrinsic::nvvm_suld_2d_array_v2i8_clamp:
+    return NVPTXISD::Suld2DArrayV2I8Clamp;
+  case Intrinsic::nvvm_suld_2d_array_v2i16_clamp:
+    return NVPTXISD::Suld2DArrayV2I16Clamp;
+  case Intrinsic::nvvm_suld_2d_array_v2i32_clamp:
+    return NVPTXISD::Suld2DArrayV2I32Clamp;
+  case Intrinsic::nvvm_suld_2d_array_v2i64_clamp:
+    return NVPTXISD::Suld2DArrayV2I64Clamp;
+  case Intrinsic::nvvm_suld_2d_array_v4i8_clamp:
+    return NVPTXISD::Suld2DArrayV4I8Clamp;
+  case Intrinsic::nvvm_suld_2d_array_v4i16_clamp:
+    return NVPTXISD::Suld2DArrayV4I16Clamp;
+  case Intrinsic::nvvm_suld_2d_array_v4i32_clamp:
+    return NVPTXISD::Suld2DArrayV4I32Clamp;
+  case Intrinsic::nvvm_suld_3d_i8_clamp:
+    return NVPTXISD::Suld3DI8Clamp;
+  case Intrinsic::nvvm_suld_3d_i16_clamp:
+    return NVPTXISD::Suld3DI16Clamp;
+  case Intrinsic::nvvm_suld_3d_i32_clamp:
+    return NVPTXISD::Suld3DI32Clamp;
+  case Intrinsic::nvvm_suld_3d_i64_clamp:
+    return NVPTXISD::Suld3DI64Clamp;
+  case Intrinsic::nvvm_suld_3d_v2i8_clamp:
+    return NVPTXISD::Suld3DV2I8Clamp;
+  case Intrinsic::nvvm_suld_3d_v2i16_clamp:
+    return NVPTXISD::Suld3DV2I16Clamp;
+  case Intrinsic::nvvm_suld_3d_v2i32_clamp:
+    return NVPTXISD::Suld3DV2I32Clamp;
+  case Intrinsic::nvvm_suld_3d_v2i64_clamp:
+    return NVPTXISD::Suld3DV2I64Clamp;
+  case Intrinsic::nvvm_suld_3d_v4i8_clamp:
+    return NVPTXISD::Suld3DV4I8Clamp;
+  case Intrinsic::nvvm_suld_3d_v4i16_clamp:
+    return NVPTXISD::Suld3DV4I16Clamp;
+  case Intrinsic::nvvm_suld_3d_v4i32_clamp:
+    return NVPTXISD::Suld3DV4I32Clamp;
+  case Intrinsic::nvvm_suld_1d_i8_trap:
+    return NVPTXISD::Suld1DI8Trap;
+  case Intrinsic::nvvm_suld_1d_i16_trap:
+    return NVPTXISD::Suld1DI16Trap;
+  case Intrinsic::nvvm_suld_1d_i32_trap:
+    return NVPTXISD::Suld1DI32Trap;
+  case Intrinsic::nvvm_suld_1d_i64_trap:
+    return NVPTXISD::Suld1DI64Trap;
+  case Intrinsic::nvvm_suld_1d_v2i8_trap:
+    return NVPTXISD::Suld1DV2I8Trap;
+  case Intrinsic::nvvm_suld_1d_v2i16_trap:
+    return NVPTXISD::Suld1DV2I16Trap;
+  case Intrinsic::nvvm_suld_1d_v2i32_trap:
+    return NVPTXISD::Suld1DV2I32Trap;
+  case Intrinsic::nvvm_suld_1d_v2i64_trap:
+    return NVPTXISD::Suld1DV2I64Trap;
+  case Intrinsic::nvvm_suld_1d_v4i8_trap:
+    return NVPTXISD::Suld1DV4I8Trap;
+  case Intrinsic::nvvm_suld_1d_v4i16_trap:
+    return NVPTXISD::Suld1DV4I16Trap;
+  case Intrinsic::nvvm_suld_1d_v4i32_trap:
+    return NVPTXISD::Suld1DV4I32Trap;
+  case Intrinsic::nvvm_suld_1d_array_i8_trap:
+    return NVPTXISD::Suld1DArrayI8Trap;
+  case Intrinsic::nvvm_suld_1d_array_i16_trap:
+    return NVPTXISD::Suld1DArrayI16Trap;
+  case Intrinsic::nvvm_suld_1d_array_i32_trap:
+    return NVPTXISD::Suld1DArrayI32Trap;
+  case Intrinsic::nvvm_suld_1d_array_i64_trap:
+    return NVPTXISD::Suld1DArrayI64Trap;
+  case Intrinsic::nvvm_suld_1d_array_v2i8_trap:
+    return NVPTXISD::Suld1DArrayV2I8Trap;
+  case Intrinsic::nvvm_suld_1d_array_v2i16_trap:
+    return NVPTXISD::Suld1DArrayV2I16Trap;
+  case Intrinsic::nvvm_suld_1d_array_v2i32_trap:
+    return NVPTXISD::Suld1DArrayV2I32Trap;
+  case Intrinsic::nvvm_suld_1d_array_v2i64_trap:
+    return NVPTXISD::Suld1DArrayV2I64Trap;
+  case Intrinsic::nvvm_suld_1d_array_v4i8_trap:
+    return NVPTXISD::Suld1DArrayV4I8Trap;
+  case Intrinsic::nvvm_suld_1d_array_v4i16_trap:
+    return NVPTXISD::Suld1DArrayV4I16Trap;
+  case Intrinsic::nvvm_suld_1d_array_v4i32_trap:
+    return NVPTXISD::Suld1DArrayV4I32Trap;
+  case Intrinsic::nvvm_suld_2d_i8_trap:
+    return NVPTXISD::Suld2DI8Trap;
+  case Intrinsic::nvvm_suld_2d_i16_trap:
+    return NVPTXISD::Suld2DI16Trap;
+  case Intrinsic::nvvm_suld_2d_i32_trap:
+    return NVPTXISD::Suld2DI32Trap;
+  case Intrinsic::nvvm_suld_2d_i64_trap:
+    return NVPTXISD::Suld2DI64Trap;
+  case Intrinsic::nvvm_suld_2d_v2i8_trap:
+    return NVPTXISD::Suld2DV2I8Trap;
+  case Intrinsic::nvvm_suld_2d_v2i16_trap:
+    return NVPTXISD::Suld2DV2I16Trap;
+  case Intrinsic::nvvm_suld_2d_v2i32_trap:
+    return NVPTXISD::Suld2DV2I32Trap;
+  case Intrinsic::nvvm_suld_2d_v2i64_trap:
+    return NVPTXISD::Suld2DV2I64Trap;
+  case Intrinsic::nvvm_suld_2d_v4i8_trap:
+    return NVPTXISD::Suld2DV4I8Trap;
+  case Intrinsic::nvvm_suld_2d_v4i16_trap:
+    return NVPTXISD::Suld2DV4I16Trap;
+  case Intrinsic::nvvm_suld_2d_v4i32_trap:
+    return NVPTXISD::Suld2DV4I32Trap;
+  case Intrinsic::nvvm_suld_2d_array_i8_trap:
+    return NVPTXISD::Suld2DArrayI8Trap;
+  case Intrinsic::nvvm_suld_2d_array_i16_trap:
+    return NVPTXISD::Suld2DArrayI16Trap;
+  case Intrinsic::nvvm_suld_2d_array_i32_trap:
+    return NVPTXISD::Suld2DArrayI32Trap;
+  case Intrinsic::nvvm_suld_2d_array_i64_trap:
+    return NVPTXISD::Suld2DArrayI64Trap;
+  case Intrinsic::nvvm_suld_2d_array_v2i8_trap:
+    return NVPTXISD::Suld2DArrayV2I8Trap;
+  case Intrinsic::nvvm_suld_2d_array_v2i16_trap:
+    return NVPTXISD::Suld2DArrayV2I16Trap;
+  case Intrinsic::nvvm_suld_2d_array_v2i32_trap:
+    return NVPTXISD::Suld2DArrayV2I32Trap;
+  case Intrinsic::nvvm_suld_2d_array_v2i64_trap:
+    return NVPTXISD::Suld2DArrayV2I64Trap;
+  case Intrinsic::nvvm_suld_2d_array_v4i8_trap:
+    return NVPTXISD::Suld2DArrayV4I8Trap;
+  case Intrinsic::nvvm_suld_2d_array_v4i16_trap:
+    return NVPTXISD::Suld2DArrayV4I16Trap;
+  case Intrinsic::nvvm_suld_2d_array_v4i32_trap:
+    return NVPTXISD::Suld2DArrayV4I32Trap;
+  case Intrinsic::nvvm_suld_3d_i8_trap:
+    return NVPTXISD::Suld3DI8Trap;
+  case Intrinsic::nvvm_suld_3d_i16_trap:
+    return NVPTXISD::Suld3DI16Trap;
+  case Intrinsic::nvvm_suld_3d_i32_trap:
+    return NVPTXISD::Suld3DI32Trap;
+  case Intrinsic::nvvm_suld_3d_i64_trap:
+    return NVPTXISD::Suld3DI64Trap;
+  case Intrinsic::nvvm_suld_3d_v2i8_trap:
+    return NVPTXISD::Suld3DV2I8Trap;
+  case Intrinsic::nvvm_suld_3d_v2i16_trap:
+    return NVPTXISD::Suld3DV2I16Trap;
+  case Intrinsic::nvvm_suld_3d_v2i32_trap:
+    return NVPTXISD::Suld3DV2I32Trap;
+  case Intrinsic::nvvm_suld_3d_v2i64_trap:
+    return NVPTXISD::Suld3DV2I64Trap;
+  case Intrinsic::nvvm_suld_3d_v4i8_trap:
+    return NVPTXISD::Suld3DV4I8Trap;
+  case Intrinsic::nvvm_suld_3d_v4i16_trap:
+    return NVPTXISD::Suld3DV4I16Trap;
+  case Intrinsic::nvvm_suld_3d_v4i32_trap:
+    return NVPTXISD::Suld3DV4I32Trap;
+  case Intrinsic::nvvm_suld_1d_i8_zero:
+    return NVPTXISD::Suld1DI8Zero;
+  case Intrinsic::nvvm_suld_1d_i16_zero:
+    return NVPTXISD::Suld1DI16Zero;
+  case Intrinsic::nvvm_suld_1d_i32_zero:
+    return NVPTXISD::Suld1DI32Zero;
+  case Intrinsic::nvvm_suld_1d_i64_zero:
+    return NVPTXISD::Suld1DI64Zero;
+  case Intrinsic::nvvm_suld_1d_v2i8_zero:
+    return NVPTXISD::Suld1DV2I8Zero;
+  case Intrinsic::nvvm_suld_1d_v2i16_zero:
+    return NVPTXISD::Suld1DV2I16Zero;
+  case Intrinsic::nvvm_suld_1d_v2i32_zero:
+    return NVPTXISD::Suld1DV2I32Zero;
+  case Intrinsic::nvvm_suld_1d_v2i64_zero:
+    return NVPTXISD::Suld1DV2I64Zero;
+  case Intrinsic::nvvm_suld_1d_v4i8_zero:
+    return NVPTXISD::Suld1DV4I8Zero;
+  case Intrinsic::nvvm_suld_1d_v4i16_zero:
+    return NVPTXISD::Suld1DV4I16Zero;
+  case Intrinsic::nvvm_suld_1d_v4i32_zero:
+    return NVPTXISD::Suld1DV4I32Zero;
+  case Intrinsic::nvvm_suld_1d_array_i8_zero:
+    return NVPTXISD::Suld1DArrayI8Zero;
+  case Intrinsic::nvvm_suld_1d_array_i16_zero:
+    return NVPTXISD::Suld1DArrayI16Zero;
+  case Intrinsic::nvvm_suld_1d_array_i32_zero:
+    return NVPTXISD::Suld1DArrayI32Zero;
+  case Intrinsic::nvvm_suld_1d_array_i64_zero:
+    return NVPTXISD::Suld1DArrayI64Zero;
+  case Intrinsic::nvvm_suld_1d_array_v2i8_zero:
+    return NVPTXISD::Suld1DArrayV2I8Zero;
+  case Intrinsic::nvvm_suld_1d_array_v2i16_zero:
+    return NVPTXISD::Suld1DArrayV2I16Zero;
+  case Intrinsic::nvvm_suld_1d_array_v2i32_zero:
+    return NVPTXISD::Suld1DArrayV2I32Zero;
+  case Intrinsic::nvvm_suld_1d_array_v2i64_zero:
+    return NVPTXISD::Suld1DArrayV2I64Zero;
+  case Intrinsic::nvvm_suld_1d_array_v4i8_zero:
+    return NVPTXISD::Suld1DArrayV4I8Zero;
+  case Intrinsic::nvvm_suld_1d_array_v4i16_zero:
+    return NVPTXISD::Suld1DArrayV4I16Zero;
+  case Intrinsic::nvvm_suld_1d_array_v4i32_zero:
+    return NVPTXISD::Suld1DArrayV4I32Zero;
+  case Intrinsic::nvvm_suld_2d_i8_zero:
+    return NVPTXISD::Suld2DI8Zero;
+  case Intrinsic::nvvm_suld_2d_i16_zero:
+    return NVPTXISD::Suld2DI16Zero;
+  case Intrinsic::nvvm_suld_2d_i32_zero:
+    return NVPTXISD::Suld2DI32Zero;
+  case Intrinsic::nvvm_suld_2d_i64_zero:
+    return NVPTXISD::Suld2DI64Zero;
+  case Intrinsic::nvvm_suld_2d_v2i8_zero:
+    return NVPTXISD::Suld2DV2I8Zero;
+  case Intrinsic::nvvm_suld_2d_v2i16_zero:
+    return NVPTXISD::Suld2DV2I16Zero;
+  case Intrinsic::nvvm_suld_2d_v2i32_zero:
+    return NVPTXISD::Suld2DV2I32Zero;
+  case Intrinsic::nvvm_suld_2d_v2i64_zero:
+    return NVPTXISD::Suld2DV2I64Zero;
+  case Intrinsic::nvvm_suld_2d_v4i8_zero:
+    return NVPTXISD::Suld2DV4I8Zero;
+  case Intrinsic::nvvm_suld_2d_v4i16_zero:
+    return NVPTXISD::Suld2DV4I16Zero;
+  case Intrinsic::nvvm_suld_2d_v4i32_zero:
+    return NVPTXISD::Suld2DV4I32Zero;
+  case Intrinsic::nvvm_suld_2d_array_i8_zero:
+    return NVPTXISD::Suld2DArrayI8Zero;
+  case Intrinsic::nvvm_suld_2d_array_i16_zero:
+    return NVPTXISD::Suld2DArrayI16Zero;
+  case Intrinsic::nvvm_suld_2d_array_i32_zero:
+    return NVPTXISD::Suld2DArrayI32Zero;
+  case Intrinsic::nvvm_suld_2d_array_i64_zero:
+    return NVPTXISD::Suld2DArrayI64Zero;
+  case Intrinsic::nvvm_suld_2d_array_v2i8_zero:
+    return NVPTXISD::Suld2DArrayV2I8Zero;
+  case Intrinsic::nvvm_suld_2d_array_v2i16_zero:
+    return NVPTXISD::Suld2DArrayV2I16Zero;
+  case Intrinsic::nvvm_suld_2d_array_v2i32_zero:
+    return NVPTXISD::Suld2DArrayV2I32Zero;
+  case Intrinsic::nvvm_suld_2d_array_v2i64_zero:
+    return NVPTXISD::Suld2DArrayV2I64Zero;
+  case Intrinsic::nvvm_suld_2d_array_v4i8_zero:
+    return NVPTXISD::Suld2DArrayV4I8Zero;
+  case Intrinsic::nvvm_suld_2d_array_v4i16_zero:
+    return NVPTXISD::Suld2DArrayV4I16Zero;
+  case Intrinsic::nvvm_suld_2d_array_v4i32_zero:
+    return NVPTXISD::Suld2DArrayV4I32Zero;
+  case Intrinsic::nvvm_suld_3d_i8_zero:
+    return NVPTXISD::Suld3DI8Zero;
+  case Intrinsic::nvvm_suld_3d_i16_zero:
+    return NVPTXISD::Suld3DI16Zero;
+  case Intrinsic::nvvm_suld_3d_i32_zero:
+    return NVPTXISD::Suld3DI32Zero;
+  case Intrinsic::nvvm_suld_3d_i64_zero:
+    return NVPTXISD::Suld3DI64Zero;
+  case Intrinsic::nvvm_suld_3d_v2i8_zero:
+    return NVPTXISD::Suld3DV2I8Zero;
+  case Intrinsic::nvvm_suld_3d_v2i16_zero:
+    return NVPTXISD::Suld3DV2I16Zero;
+  case Intrinsic::nvvm_suld_3d_v2i32_zero:
+    return NVPTXISD::Suld3DV2I32Zero;
+  case Intrinsic::nvvm_suld_3d_v2i64_zero:
+    return NVPTXISD::Suld3DV2I64Zero;
+  case Intrinsic::nvvm_suld_3d_v4i8_zero:
+    return NVPTXISD::Suld3DV4I8Zero;
+  case Intrinsic::nvvm_suld_3d_v4i16_zero:
+    return NVPTXISD::Suld3DV4I16Zero;
+  case Intrinsic::nvvm_suld_3d_v4i32_zero:
+    return NVPTXISD::Suld3DV4I32Zero;
+  }
+}
+
 // llvm.ptx.memcpy.const and llvm.ptx.memmove.const need to be modeled as
 // TgtMemIntrinsic
 // because we need the information that is only available in the "Value" type
@@ -1928,23 +3255,456 @@ bool NVPTXTargetLowering::getTgtMemIntrinsic(
 
   case Intrinsic::nvvm_ldu_global_i:
   case Intrinsic::nvvm_ldu_global_f:
-  case Intrinsic::nvvm_ldu_global_p:
+  case Intrinsic::nvvm_ldu_global_p: {
 
     Info.opc = ISD::INTRINSIC_W_CHAIN;
     if (Intrinsic == Intrinsic::nvvm_ldu_global_i)
       Info.memVT = getValueType(I.getType());
-    else if (Intrinsic == Intrinsic::nvvm_ldu_global_p)
+    else if(Intrinsic == Intrinsic::nvvm_ldu_global_p)
+      Info.memVT = getPointerTy();
+    else
       Info.memVT = getValueType(I.getType());
+    Info.ptrVal = I.getArgOperand(0);
+    Info.offset = 0;
+    Info.vol = 0;
+    Info.readMem = true;
+    Info.writeMem = false;
+
+    // alignment is available as metadata.
+    // Grab it and set the alignment.
+    assert(I.hasMetadataOtherThanDebugLoc() && "Must have alignment metadata");
+    MDNode *AlignMD = I.getMetadata("align");
+    assert(AlignMD && "Must have a non-null MDNode");
+    assert(AlignMD->getNumOperands() == 1 && "Must have a single operand");
+    Value *Align = AlignMD->getOperand(0);
+    int64_t Alignment = cast<ConstantInt>(Align)->getZExtValue();
+    Info.align = Alignment;
+
+    return true;
+  }
+  case Intrinsic::nvvm_ldg_global_i:
+  case Intrinsic::nvvm_ldg_global_f:
+  case Intrinsic::nvvm_ldg_global_p: {
+
+    Info.opc = ISD::INTRINSIC_W_CHAIN;
+    if (Intrinsic == Intrinsic::nvvm_ldg_global_i)
+      Info.memVT = getValueType(I.getType());
+    else if(Intrinsic == Intrinsic::nvvm_ldg_global_p)
+      Info.memVT = getPointerTy();
     else
-      Info.memVT = MVT::f32;
+      Info.memVT = getValueType(I.getType());
     Info.ptrVal = I.getArgOperand(0);
     Info.offset = 0;
     Info.vol = 0;
     Info.readMem = true;
     Info.writeMem = false;
-    Info.align = 0;
+
+    // alignment is available as metadata.
+    // Grab it and set the alignment.
+    assert(I.hasMetadataOtherThanDebugLoc() && "Must have alignment metadata");
+    MDNode *AlignMD = I.getMetadata("align");
+    assert(AlignMD && "Must have a non-null MDNode");
+    assert(AlignMD->getNumOperands() == 1 && "Must have a single operand");
+    Value *Align = AlignMD->getOperand(0);
+    int64_t Alignment = cast<ConstantInt>(Align)->getZExtValue();
+    Info.align = Alignment;
+
     return true;
+  }
 
+  case Intrinsic::nvvm_tex_1d_v4f32_s32:
+  case Intrinsic::nvvm_tex_1d_v4f32_f32:
+  case Intrinsic::nvvm_tex_1d_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_1d_grad_v4f32_f32:
+  case Intrinsic::nvvm_tex_1d_array_v4f32_s32:
+  case Intrinsic::nvvm_tex_1d_array_v4f32_f32:
+  case Intrinsic::nvvm_tex_1d_array_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_1d_array_grad_v4f32_f32:
+  case Intrinsic::nvvm_tex_2d_v4f32_s32:
+  case Intrinsic::nvvm_tex_2d_v4f32_f32:
+  case Intrinsic::nvvm_tex_2d_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_2d_grad_v4f32_f32:
+  case Intrinsic::nvvm_tex_2d_array_v4f32_s32:
+  case Intrinsic::nvvm_tex_2d_array_v4f32_f32:
+  case Intrinsic::nvvm_tex_2d_array_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_2d_array_grad_v4f32_f32:
+  case Intrinsic::nvvm_tex_3d_v4f32_s32:
+  case Intrinsic::nvvm_tex_3d_v4f32_f32:
+  case Intrinsic::nvvm_tex_3d_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_3d_grad_v4f32_f32:
+  case Intrinsic::nvvm_tex_cube_v4f32_f32:
+  case Intrinsic::nvvm_tex_cube_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_cube_array_v4f32_f32:
+  case Intrinsic::nvvm_tex_cube_array_level_v4f32_f32:
+  case Intrinsic::nvvm_tld4_r_2d_v4f32_f32:
+  case Intrinsic::nvvm_tld4_g_2d_v4f32_f32:
+  case Intrinsic::nvvm_tld4_b_2d_v4f32_f32:
+  case Intrinsic::nvvm_tld4_a_2d_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_v4f32_s32:
+  case Intrinsic::nvvm_tex_unified_1d_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_grad_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_array_v4f32_s32:
+  case Intrinsic::nvvm_tex_unified_1d_array_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_array_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_array_grad_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_v4f32_s32:
+  case Intrinsic::nvvm_tex_unified_2d_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_grad_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_array_v4f32_s32:
+  case Intrinsic::nvvm_tex_unified_2d_array_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_array_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_array_grad_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_3d_v4f32_s32:
+  case Intrinsic::nvvm_tex_unified_3d_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_3d_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_3d_grad_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_level_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_array_v4f32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_array_level_v4f32_f32:
+  case Intrinsic::nvvm_tld4_unified_r_2d_v4f32_f32:
+  case Intrinsic::nvvm_tld4_unified_g_2d_v4f32_f32:
+  case Intrinsic::nvvm_tld4_unified_b_2d_v4f32_f32:
+  case Intrinsic::nvvm_tld4_unified_a_2d_v4f32_f32: {
+    Info.opc = getOpcForTextureInstr(Intrinsic);
+    Info.memVT = MVT::v4f32;
+    Info.ptrVal = nullptr;
+    Info.offset = 0;
+    Info.vol = 0;
+    Info.readMem = true;
+    Info.writeMem = false;
+    Info.align = 16;
+    return true;
+  }
+  case Intrinsic::nvvm_tex_1d_v4s32_s32:
+  case Intrinsic::nvvm_tex_1d_v4s32_f32:
+  case Intrinsic::nvvm_tex_1d_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_1d_grad_v4s32_f32:
+  case Intrinsic::nvvm_tex_1d_array_v4s32_s32:
+  case Intrinsic::nvvm_tex_1d_array_v4s32_f32:
+  case Intrinsic::nvvm_tex_1d_array_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_1d_array_grad_v4s32_f32:
+  case Intrinsic::nvvm_tex_2d_v4s32_s32:
+  case Intrinsic::nvvm_tex_2d_v4s32_f32:
+  case Intrinsic::nvvm_tex_2d_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_2d_grad_v4s32_f32:
+  case Intrinsic::nvvm_tex_2d_array_v4s32_s32:
+  case Intrinsic::nvvm_tex_2d_array_v4s32_f32:
+  case Intrinsic::nvvm_tex_2d_array_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_2d_array_grad_v4s32_f32:
+  case Intrinsic::nvvm_tex_3d_v4s32_s32:
+  case Intrinsic::nvvm_tex_3d_v4s32_f32:
+  case Intrinsic::nvvm_tex_3d_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_3d_grad_v4s32_f32:
+  case Intrinsic::nvvm_tex_cube_v4s32_f32:
+  case Intrinsic::nvvm_tex_cube_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_cube_array_v4s32_f32:
+  case Intrinsic::nvvm_tex_cube_array_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_cube_v4u32_f32:
+  case Intrinsic::nvvm_tex_cube_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_cube_array_v4u32_f32:
+  case Intrinsic::nvvm_tex_cube_array_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_1d_v4u32_s32:
+  case Intrinsic::nvvm_tex_1d_v4u32_f32:
+  case Intrinsic::nvvm_tex_1d_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_1d_grad_v4u32_f32:
+  case Intrinsic::nvvm_tex_1d_array_v4u32_s32:
+  case Intrinsic::nvvm_tex_1d_array_v4u32_f32:
+  case Intrinsic::nvvm_tex_1d_array_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_1d_array_grad_v4u32_f32:
+  case Intrinsic::nvvm_tex_2d_v4u32_s32:
+  case Intrinsic::nvvm_tex_2d_v4u32_f32:
+  case Intrinsic::nvvm_tex_2d_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_2d_grad_v4u32_f32:
+  case Intrinsic::nvvm_tex_2d_array_v4u32_s32:
+  case Intrinsic::nvvm_tex_2d_array_v4u32_f32:
+  case Intrinsic::nvvm_tex_2d_array_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_2d_array_grad_v4u32_f32:
+  case Intrinsic::nvvm_tex_3d_v4u32_s32:
+  case Intrinsic::nvvm_tex_3d_v4u32_f32:
+  case Intrinsic::nvvm_tex_3d_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_3d_grad_v4u32_f32:
+  case Intrinsic::nvvm_tld4_r_2d_v4s32_f32:
+  case Intrinsic::nvvm_tld4_g_2d_v4s32_f32:
+  case Intrinsic::nvvm_tld4_b_2d_v4s32_f32:
+  case Intrinsic::nvvm_tld4_a_2d_v4s32_f32:
+  case Intrinsic::nvvm_tld4_r_2d_v4u32_f32:
+  case Intrinsic::nvvm_tld4_g_2d_v4u32_f32:
+  case Intrinsic::nvvm_tld4_b_2d_v4u32_f32:
+  case Intrinsic::nvvm_tld4_a_2d_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_v4s32_s32:
+  case Intrinsic::nvvm_tex_unified_1d_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_grad_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_array_v4s32_s32:
+  case Intrinsic::nvvm_tex_unified_1d_array_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_array_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_array_grad_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_v4s32_s32:
+  case Intrinsic::nvvm_tex_unified_2d_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_grad_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_array_v4s32_s32:
+  case Intrinsic::nvvm_tex_unified_2d_array_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_array_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_array_grad_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_3d_v4s32_s32:
+  case Intrinsic::nvvm_tex_unified_3d_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_3d_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_3d_grad_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_v4u32_s32:
+  case Intrinsic::nvvm_tex_unified_1d_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_grad_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_array_v4u32_s32:
+  case Intrinsic::nvvm_tex_unified_1d_array_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_array_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_1d_array_grad_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_v4u32_s32:
+  case Intrinsic::nvvm_tex_unified_2d_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_grad_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_array_v4u32_s32:
+  case Intrinsic::nvvm_tex_unified_2d_array_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_array_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_2d_array_grad_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_3d_v4u32_s32:
+  case Intrinsic::nvvm_tex_unified_3d_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_3d_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_3d_grad_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_array_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_array_level_v4s32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_level_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_array_v4u32_f32:
+  case Intrinsic::nvvm_tex_unified_cube_array_level_v4u32_f32:
+  case Intrinsic::nvvm_tld4_unified_r_2d_v4s32_f32:
+  case Intrinsic::nvvm_tld4_unified_g_2d_v4s32_f32:
+  case Intrinsic::nvvm_tld4_unified_b_2d_v4s32_f32:
+  case Intrinsic::nvvm_tld4_unified_a_2d_v4s32_f32:
+  case Intrinsic::nvvm_tld4_unified_r_2d_v4u32_f32:
+  case Intrinsic::nvvm_tld4_unified_g_2d_v4u32_f32:
+  case Intrinsic::nvvm_tld4_unified_b_2d_v4u32_f32:
+  case Intrinsic::nvvm_tld4_unified_a_2d_v4u32_f32: {
+    Info.opc = getOpcForTextureInstr(Intrinsic);
+    Info.memVT = MVT::v4i32;
+    Info.ptrVal = nullptr;
+    Info.offset = 0;
+    Info.vol = 0;
+    Info.readMem = true;
+    Info.writeMem = false;
+    Info.align = 16;
+    return true;
+  }
+  case Intrinsic::nvvm_suld_1d_i8_clamp:
+  case Intrinsic::nvvm_suld_1d_v2i8_clamp:
+  case Intrinsic::nvvm_suld_1d_v4i8_clamp:
+  case Intrinsic::nvvm_suld_1d_array_i8_clamp:
+  case Intrinsic::nvvm_suld_1d_array_v2i8_clamp:
+  case Intrinsic::nvvm_suld_1d_array_v4i8_clamp:
+  case Intrinsic::nvvm_suld_2d_i8_clamp:
+  case Intrinsic::nvvm_suld_2d_v2i8_clamp:
+  case Intrinsic::nvvm_suld_2d_v4i8_clamp:
+  case Intrinsic::nvvm_suld_2d_array_i8_clamp:
+  case Intrinsic::nvvm_suld_2d_array_v2i8_clamp:
+  case Intrinsic::nvvm_suld_2d_array_v4i8_clamp:
+  case Intrinsic::nvvm_suld_3d_i8_clamp:
+  case Intrinsic::nvvm_suld_3d_v2i8_clamp:
+  case Intrinsic::nvvm_suld_3d_v4i8_clamp:
+  case Intrinsic::nvvm_suld_1d_i8_trap:
+  case Intrinsic::nvvm_suld_1d_v2i8_trap:
+  case Intrinsic::nvvm_suld_1d_v4i8_trap:
+  case Intrinsic::nvvm_suld_1d_array_i8_trap:
+  case Intrinsic::nvvm_suld_1d_array_v2i8_trap:
+  case Intrinsic::nvvm_suld_1d_array_v4i8_trap:
+  case Intrinsic::nvvm_suld_2d_i8_trap:
+  case Intrinsic::nvvm_suld_2d_v2i8_trap:
+  case Intrinsic::nvvm_suld_2d_v4i8_trap:
+  case Intrinsic::nvvm_suld_2d_array_i8_trap:
+  case Intrinsic::nvvm_suld_2d_array_v2i8_trap:
+  case Intrinsic::nvvm_suld_2d_array_v4i8_trap:
+  case Intrinsic::nvvm_suld_3d_i8_trap:
+  case Intrinsic::nvvm_suld_3d_v2i8_trap:
+  case Intrinsic::nvvm_suld_3d_v4i8_trap:
+  case Intrinsic::nvvm_suld_1d_i8_zero:
+  case Intrinsic::nvvm_suld_1d_v2i8_zero:
+  case Intrinsic::nvvm_suld_1d_v4i8_zero:
+  case Intrinsic::nvvm_suld_1d_array_i8_zero:
+  case Intrinsic::nvvm_suld_1d_array_v2i8_zero:
+  case Intrinsic::nvvm_suld_1d_array_v4i8_zero:
+  case Intrinsic::nvvm_suld_2d_i8_zero:
+  case Intrinsic::nvvm_suld_2d_v2i8_zero:
+  case Intrinsic::nvvm_suld_2d_v4i8_zero:
+  case Intrinsic::nvvm_suld_2d_array_i8_zero:
+  case Intrinsic::nvvm_suld_2d_array_v2i8_zero:
+  case Intrinsic::nvvm_suld_2d_array_v4i8_zero:
+  case Intrinsic::nvvm_suld_3d_i8_zero:
+  case Intrinsic::nvvm_suld_3d_v2i8_zero:
+  case Intrinsic::nvvm_suld_3d_v4i8_zero: {
+    Info.opc = getOpcForSurfaceInstr(Intrinsic);
+    Info.memVT = MVT::i8;
+    Info.ptrVal = nullptr;
+    Info.offset = 0;
+    Info.vol = 0;
+    Info.readMem = true;
+    Info.writeMem = false;
+    Info.align = 16;
+    return true;
+  }
+  case Intrinsic::nvvm_suld_1d_i16_clamp:
+  case Intrinsic::nvvm_suld_1d_v2i16_clamp:
+  case Intrinsic::nvvm_suld_1d_v4i16_clamp:
+  case Intrinsic::nvvm_suld_1d_array_i16_clamp:
+  case Intrinsic::nvvm_suld_1d_array_v2i16_clamp:
+  case Intrinsic::nvvm_suld_1d_array_v4i16_clamp:
+  case Intrinsic::nvvm_suld_2d_i16_clamp:
+  case Intrinsic::nvvm_suld_2d_v2i16_clamp:
+  case Intrinsic::nvvm_suld_2d_v4i16_clamp:
+  case Intrinsic::nvvm_suld_2d_array_i16_clamp:
+  case Intrinsic::nvvm_suld_2d_array_v2i16_clamp:
+  case Intrinsic::nvvm_suld_2d_array_v4i16_clamp:
+  case Intrinsic::nvvm_suld_3d_i16_clamp:
+  case Intrinsic::nvvm_suld_3d_v2i16_clamp:
+  case Intrinsic::nvvm_suld_3d_v4i16_clamp:
+  case Intrinsic::nvvm_suld_1d_i16_trap:
+  case Intrinsic::nvvm_suld_1d_v2i16_trap:
+  case Intrinsic::nvvm_suld_1d_v4i16_trap:
+  case Intrinsic::nvvm_suld_1d_array_i16_trap:
+  case Intrinsic::nvvm_suld_1d_array_v2i16_trap:
+  case Intrinsic::nvvm_suld_1d_array_v4i16_trap:
+  case Intrinsic::nvvm_suld_2d_i16_trap:
+  case Intrinsic::nvvm_suld_2d_v2i16_trap:
+  case Intrinsic::nvvm_suld_2d_v4i16_trap:
+  case Intrinsic::nvvm_suld_2d_array_i16_trap:
+  case Intrinsic::nvvm_suld_2d_array_v2i16_trap:
+  case Intrinsic::nvvm_suld_2d_array_v4i16_trap:
+  case Intrinsic::nvvm_suld_3d_i16_trap:
+  case Intrinsic::nvvm_suld_3d_v2i16_trap:
+  case Intrinsic::nvvm_suld_3d_v4i16_trap:
+  case Intrinsic::nvvm_suld_1d_i16_zero:
+  case Intrinsic::nvvm_suld_1d_v2i16_zero:
+  case Intrinsic::nvvm_suld_1d_v4i16_zero:
+  case Intrinsic::nvvm_suld_1d_array_i16_zero:
+  case Intrinsic::nvvm_suld_1d_array_v2i16_zero:
+  case Intrinsic::nvvm_suld_1d_array_v4i16_zero:
+  case Intrinsic::nvvm_suld_2d_i16_zero:
+  case Intrinsic::nvvm_suld_2d_v2i16_zero:
+  case Intrinsic::nvvm_suld_2d_v4i16_zero:
+  case Intrinsic::nvvm_suld_2d_array_i16_zero:
+  case Intrinsic::nvvm_suld_2d_array_v2i16_zero:
+  case Intrinsic::nvvm_suld_2d_array_v4i16_zero:
+  case Intrinsic::nvvm_suld_3d_i16_zero:
+  case Intrinsic::nvvm_suld_3d_v2i16_zero:
+  case Intrinsic::nvvm_suld_3d_v4i16_zero: {
+    Info.opc = getOpcForSurfaceInstr(Intrinsic);
+    Info.memVT = MVT::i16;
+    Info.ptrVal = nullptr;
+    Info.offset = 0;
+    Info.vol = 0;
+    Info.readMem = true;
+    Info.writeMem = false;
+    Info.align = 16;
+    return true;
+  }
+  case Intrinsic::nvvm_suld_1d_i32_clamp:
+  case Intrinsic::nvvm_suld_1d_v2i32_clamp:
+  case Intrinsic::nvvm_suld_1d_v4i32_clamp:
+  case Intrinsic::nvvm_suld_1d_array_i32_clamp:
+  case Intrinsic::nvvm_suld_1d_array_v2i32_clamp:
+  case Intrinsic::nvvm_suld_1d_array_v4i32_clamp:
+  case Intrinsic::nvvm_suld_2d_i32_clamp:
+  case Intrinsic::nvvm_suld_2d_v2i32_clamp:
+  case Intrinsic::nvvm_suld_2d_v4i32_clamp:
+  case Intrinsic::nvvm_suld_2d_array_i32_clamp:
+  case Intrinsic::nvvm_suld_2d_array_v2i32_clamp:
+  case Intrinsic::nvvm_suld_2d_array_v4i32_clamp:
+  case Intrinsic::nvvm_suld_3d_i32_clamp:
+  case Intrinsic::nvvm_suld_3d_v2i32_clamp:
+  case Intrinsic::nvvm_suld_3d_v4i32_clamp:
+  case Intrinsic::nvvm_suld_1d_i32_trap:
+  case Intrinsic::nvvm_suld_1d_v2i32_trap:
+  case Intrinsic::nvvm_suld_1d_v4i32_trap:
+  case Intrinsic::nvvm_suld_1d_array_i32_trap:
+  case Intrinsic::nvvm_suld_1d_array_v2i32_trap:
+  case Intrinsic::nvvm_suld_1d_array_v4i32_trap:
+  case Intrinsic::nvvm_suld_2d_i32_trap:
+  case Intrinsic::nvvm_suld_2d_v2i32_trap:
+  case Intrinsic::nvvm_suld_2d_v4i32_trap:
+  case Intrinsic::nvvm_suld_2d_array_i32_trap:
+  case Intrinsic::nvvm_suld_2d_array_v2i32_trap:
+  case Intrinsic::nvvm_suld_2d_array_v4i32_trap:
+  case Intrinsic::nvvm_suld_3d_i32_trap:
+  case Intrinsic::nvvm_suld_3d_v2i32_trap:
+  case Intrinsic::nvvm_suld_3d_v4i32_trap:
+  case Intrinsic::nvvm_suld_1d_i32_zero:
+  case Intrinsic::nvvm_suld_1d_v2i32_zero:
+  case Intrinsic::nvvm_suld_1d_v4i32_zero:
+  case Intrinsic::nvvm_suld_1d_array_i32_zero:
+  case Intrinsic::nvvm_suld_1d_array_v2i32_zero:
+  case Intrinsic::nvvm_suld_1d_array_v4i32_zero:
+  case Intrinsic::nvvm_suld_2d_i32_zero:
+  case Intrinsic::nvvm_suld_2d_v2i32_zero:
+  case Intrinsic::nvvm_suld_2d_v4i32_zero:
+  case Intrinsic::nvvm_suld_2d_array_i32_zero:
+  case Intrinsic::nvvm_suld_2d_array_v2i32_zero:
+  case Intrinsic::nvvm_suld_2d_array_v4i32_zero:
+  case Intrinsic::nvvm_suld_3d_i32_zero:
+  case Intrinsic::nvvm_suld_3d_v2i32_zero:
+  case Intrinsic::nvvm_suld_3d_v4i32_zero: {
+    Info.opc = getOpcForSurfaceInstr(Intrinsic);
+    Info.memVT = MVT::i32;
+    Info.ptrVal = nullptr;
+    Info.offset = 0;
+    Info.vol = 0;
+    Info.readMem = true;
+    Info.writeMem = false;
+    Info.align = 16;
+    return true;
+  }
+  case Intrinsic::nvvm_suld_1d_i64_clamp:
+  case Intrinsic::nvvm_suld_1d_v2i64_clamp:
+  case Intrinsic::nvvm_suld_1d_array_i64_clamp:
+  case Intrinsic::nvvm_suld_1d_array_v2i64_clamp:
+  case Intrinsic::nvvm_suld_2d_i64_clamp:
+  case Intrinsic::nvvm_suld_2d_v2i64_clamp:
+  case Intrinsic::nvvm_suld_2d_array_i64_clamp:
+  case Intrinsic::nvvm_suld_2d_array_v2i64_clamp:
+  case Intrinsic::nvvm_suld_3d_i64_clamp:
+  case Intrinsic::nvvm_suld_3d_v2i64_clamp:
+  case Intrinsic::nvvm_suld_1d_i64_trap:
+  case Intrinsic::nvvm_suld_1d_v2i64_trap:
+  case Intrinsic::nvvm_suld_1d_array_i64_trap:
+  case Intrinsic::nvvm_suld_1d_array_v2i64_trap:
+  case Intrinsic::nvvm_suld_2d_i64_trap:
+  case Intrinsic::nvvm_suld_2d_v2i64_trap:
+  case Intrinsic::nvvm_suld_2d_array_i64_trap:
+  case Intrinsic::nvvm_suld_2d_array_v2i64_trap:
+  case Intrinsic::nvvm_suld_3d_i64_trap:
+  case Intrinsic::nvvm_suld_3d_v2i64_trap:
+  case Intrinsic::nvvm_suld_1d_i64_zero:
+  case Intrinsic::nvvm_suld_1d_v2i64_zero:
+  case Intrinsic::nvvm_suld_1d_array_i64_zero:
+  case Intrinsic::nvvm_suld_1d_array_v2i64_zero:
+  case Intrinsic::nvvm_suld_2d_i64_zero:
+  case Intrinsic::nvvm_suld_2d_v2i64_zero:
+  case Intrinsic::nvvm_suld_2d_array_i64_zero:
+  case Intrinsic::nvvm_suld_2d_array_v2i64_zero:
+  case Intrinsic::nvvm_suld_3d_i64_zero:
+  case Intrinsic::nvvm_suld_3d_v2i64_zero: {
+    Info.opc = getOpcForSurfaceInstr(Intrinsic);
+    Info.memVT = MVT::i64;
+    Info.ptrVal = nullptr;
+    Info.offset = 0;
+    Info.vol = 0;
+    Info.readMem = true;
+    Info.writeMem = false;
+    Info.align = 16;
+    return true;
+  }
   }
   return false;
 }
@@ -1999,6 +3759,7 @@ NVPTXTargetLowering::getConstraintType(const std::string &Constraint) const {
     switch (Constraint[0]) {
     default:
       break;
+    case 'b':
     case 'r':
     case 'h':
     case 'c':
@@ -2018,6 +3779,8 @@ NVPTXTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
                                                   MVT VT) const {
   if (Constraint.size() == 1) {
     switch (Constraint[0]) {
+    case 'b':
+      return std::make_pair(0U, &NVPTX::Int1RegsRegClass);
     case 'c':
       return std::make_pair(0U, &NVPTX::Int16RegsRegClass);
     case 'h':
@@ -2041,8 +3804,434 @@ unsigned NVPTXTargetLowering::getFunctionAlignment(const Function *) const {
   return 4;
 }
 
+//===----------------------------------------------------------------------===//
+//                         NVPTX DAG Combining
+//===----------------------------------------------------------------------===//
+
+bool NVPTXTargetLowering::allowFMA(MachineFunction &MF,
+                                   CodeGenOpt::Level OptLevel) const {
+  const Function *F = MF.getFunction();
+  const TargetOptions &TO = MF.getTarget().Options;
+
+  // Always honor command-line argument
+  if (FMAContractLevelOpt.getNumOccurrences() > 0) {
+    return FMAContractLevelOpt > 0;
+  } else if (OptLevel == 0) {
+    // Do not contract if we're not optimizing the code
+    return false;
+  } else if (TO.AllowFPOpFusion == FPOpFusion::Fast || TO.UnsafeFPMath) {
+    // Honor TargetOptions flags that explicitly say fusion is okay
+    return true;
+  } else if (F->hasFnAttribute("unsafe-fp-math")) {
+    // Check for unsafe-fp-math=true coming from Clang
+    Attribute Attr = F->getFnAttribute("unsafe-fp-math");
+    StringRef Val = Attr.getValueAsString();
+    if (Val == "true")
+      return true;
+  }
+
+  // We did not have a clear indication that fusion is allowed, so assume not
+  return false;
+}
+
+/// PerformADDCombineWithOperands - Try DAG combinations for an ADD with
+/// operands N0 and N1.  This is a helper for PerformADDCombine that is
+/// called with the default operands, and if that fails, with commuted
+/// operands.
+static SDValue PerformADDCombineWithOperands(SDNode *N, SDValue N0, SDValue N1,
+                                           TargetLowering::DAGCombinerInfo &DCI,
+                                             const NVPTXSubtarget &Subtarget,
+                                             CodeGenOpt::Level OptLevel) {
+  SelectionDAG  &DAG = DCI.DAG;
+  // Skip non-integer, non-scalar case
+  EVT VT=N0.getValueType();
+  if (VT.isVector())
+    return SDValue();
+
+  // fold (add (mul a, b), c) -> (mad a, b, c)
+  //
+  if (N0.getOpcode() == ISD::MUL) {
+    assert (VT.isInteger());
+    // For integer:
+    // Since integer multiply-add costs the same as integer multiply
+    // but is more costly than integer add, do the fusion only when
+    // the mul is only used in the add.
+    if (OptLevel==CodeGenOpt::None || VT != MVT::i32 ||
+        !N0.getNode()->hasOneUse())
+      return SDValue();
+
+    // Do the folding
+    return DAG.getNode(NVPTXISD::IMAD, SDLoc(N), VT,
+                       N0.getOperand(0), N0.getOperand(1), N1);
+  }
+  else if (N0.getOpcode() == ISD::FMUL) {
+    if (VT == MVT::f32 || VT == MVT::f64) {
+      NVPTXTargetLowering *TLI =
+        (NVPTXTargetLowering *)&DAG.getTargetLoweringInfo();
+      if (!TLI->allowFMA(DAG.getMachineFunction(), OptLevel))
+        return SDValue();
+
+      // For floating point:
+      // Do the fusion only when the mul has less than 5 uses and all
+      // are add.
+      // The heuristic is that if a use is not an add, then that use
+      // cannot be fused into fma, therefore mul is still needed anyway.
+      // If there are more than 4 uses, even if they are all add, fusing
+      // them will increase register pressue.
+      //
+      int numUses = 0;
+      int nonAddCount = 0;
+      for (SDNode::use_iterator UI = N0.getNode()->use_begin(),
+           UE = N0.getNode()->use_end();
+           UI != UE; ++UI) {
+        numUses++;
+        SDNode *User = *UI;
+        if (User->getOpcode() != ISD::FADD)
+          ++nonAddCount;
+      }
+      if (numUses >= 5)
+        return SDValue();
+      if (nonAddCount) {
+        int orderNo = N->getIROrder();
+        int orderNo2 = N0.getNode()->getIROrder();
+        // simple heuristics here for considering potential register
+        // pressure, the logics here is that the differnce are used
+        // to measure the distance between def and use, the longer distance
+        // more likely cause register pressure.
+        if (orderNo - orderNo2 < 500)
+          return SDValue();
+
+        // Now, check if at least one of the FMUL's operands is live beyond the node N,
+        // which guarantees that the FMA will not increase register pressure at node N.
+        bool opIsLive = false;
+        const SDNode *left = N0.getOperand(0).getNode();
+        const SDNode *right = N0.getOperand(1).getNode();
+
+        if (dyn_cast<ConstantSDNode>(left) || dyn_cast<ConstantSDNode>(right))
+          opIsLive = true;
+
+        if (!opIsLive)
+          for (SDNode::use_iterator UI = left->use_begin(), UE = left->use_end(); UI != UE; ++UI) {
+            SDNode *User = *UI;
+            int orderNo3 = User->getIROrder();
+            if (orderNo3 > orderNo) {
+              opIsLive = true;
+              break;
+            }
+          }
+
+        if (!opIsLive)
+          for (SDNode::use_iterator UI = right->use_begin(), UE = right->use_end(); UI != UE; ++UI) {
+            SDNode *User = *UI;
+            int orderNo3 = User->getIROrder();
+            if (orderNo3 > orderNo) {
+              opIsLive = true;
+              break;
+            }
+          }
+
+        if (!opIsLive)
+          return SDValue();
+      }
+
+      return DAG.getNode(ISD::FMA, SDLoc(N), VT,
+                         N0.getOperand(0), N0.getOperand(1), N1);
+    }
+  }
+
+  return SDValue();
+}
+
+/// PerformADDCombine - Target-specific dag combine xforms for ISD::ADD.
+///
+static SDValue PerformADDCombine(SDNode *N,
+                                 TargetLowering::DAGCombinerInfo &DCI,
+                                 const NVPTXSubtarget &Subtarget,
+                                 CodeGenOpt::Level OptLevel) {
+  SDValue N0 = N->getOperand(0);
+  SDValue N1 = N->getOperand(1);
+
+  // First try with the default operand order.
+  SDValue Result = PerformADDCombineWithOperands(N, N0, N1, DCI, Subtarget,
+                                                 OptLevel);
+  if (Result.getNode())
+    return Result;
+
+  // If that didn't work, try again with the operands commuted.
+  return PerformADDCombineWithOperands(N, N1, N0, DCI, Subtarget, OptLevel);
+}
+
+static SDValue PerformANDCombine(SDNode *N,
+                                 TargetLowering::DAGCombinerInfo &DCI) {
+  // The type legalizer turns a vector load of i8 values into a zextload to i16
+  // registers, optionally ANY_EXTENDs it (if target type is integer),
+  // and ANDs off the high 8 bits. Since we turn this load into a
+  // target-specific DAG node, the DAG combiner fails to eliminate these AND
+  // nodes. Do that here.
+  SDValue Val = N->getOperand(0);
+  SDValue Mask = N->getOperand(1);
+
+  if (isa<ConstantSDNode>(Val)) {
+    std::swap(Val, Mask);
+  }
+
+  SDValue AExt;
+  // Generally, we will see zextload -> IMOV16rr -> ANY_EXTEND -> and
+  if (Val.getOpcode() == ISD::ANY_EXTEND) {
+    AExt = Val;
+    Val = Val->getOperand(0);
+  }
+
+  if (Val->isMachineOpcode() && Val->getMachineOpcode() == NVPTX::IMOV16rr) {
+    Val = Val->getOperand(0);
+  }
+
+  if (Val->getOpcode() == NVPTXISD::LoadV2 ||
+      Val->getOpcode() == NVPTXISD::LoadV4) {
+    ConstantSDNode *MaskCnst = dyn_cast<ConstantSDNode>(Mask);
+    if (!MaskCnst) {
+      // Not an AND with a constant
+      return SDValue();
+    }
+
+    uint64_t MaskVal = MaskCnst->getZExtValue();
+    if (MaskVal != 0xff) {
+      // Not an AND that chops off top 8 bits
+      return SDValue();
+    }
+
+    MemSDNode *Mem = dyn_cast<MemSDNode>(Val);
+    if (!Mem) {
+      // Not a MemSDNode?!?
+      return SDValue();
+    }
+
+    EVT MemVT = Mem->getMemoryVT();
+    if (MemVT != MVT::v2i8 && MemVT != MVT::v4i8) {
+      // We only handle the i8 case
+      return SDValue();
+    }
+
+    unsigned ExtType =
+      cast<ConstantSDNode>(Val->getOperand(Val->getNumOperands()-1))->
+        getZExtValue();
+    if (ExtType == ISD::SEXTLOAD) {
+      // If for some reason the load is a sextload, the and is needed to zero
+      // out the high 8 bits
+      return SDValue();
+    }
+
+    bool AddTo = false;
+    if (AExt.getNode() != 0) {
+      // Re-insert the ext as a zext.
+      Val = DCI.DAG.getNode(ISD::ZERO_EXTEND, SDLoc(N),
+                            AExt.getValueType(), Val);
+      AddTo = true;
+    }
+
+    // If we get here, the AND is unnecessary.  Just replace it with the load
+    DCI.CombineTo(N, Val, AddTo);
+  }
+
+  return SDValue();
+}
+
+enum OperandSignedness {
+  Signed = 0,
+  Unsigned,
+  Unknown
+};
+
+/// IsMulWideOperandDemotable - Checks if the provided DAG node is an operand
+/// that can be demoted to \p OptSize bits without loss of information. The
+/// signedness of the operand, if determinable, is placed in \p S.
+static bool IsMulWideOperandDemotable(SDValue Op,
+                                      unsigned OptSize,
+                                      OperandSignedness &S) {
+  S = Unknown;
+
+  if (Op.getOpcode() == ISD::SIGN_EXTEND ||
+      Op.getOpcode() == ISD::SIGN_EXTEND_INREG) {
+    EVT OrigVT = Op.getOperand(0).getValueType();
+    if (OrigVT.getSizeInBits() == OptSize) {
+      S = Signed;
+      return true;
+    }
+  } else if (Op.getOpcode() == ISD::ZERO_EXTEND) {
+    EVT OrigVT = Op.getOperand(0).getValueType();
+    if (OrigVT.getSizeInBits() == OptSize) {
+      S = Unsigned;
+      return true;
+    }
+  }
+
+  return false;
+}
+
+/// AreMulWideOperandsDemotable - Checks if the given LHS and RHS operands can
+/// be demoted to \p OptSize bits without loss of information. If the operands
+/// contain a constant, it should appear as the RHS operand. The signedness of
+/// the operands is placed in \p IsSigned.
+static bool AreMulWideOperandsDemotable(SDValue LHS, SDValue RHS,
+                                        unsigned OptSize,
+                                        bool &IsSigned) {
+
+  OperandSignedness LHSSign;
+
+  // The LHS operand must be a demotable op
+  if (!IsMulWideOperandDemotable(LHS, OptSize, LHSSign))
+    return false;
+
+  // We should have been able to determine the signedness from the LHS
+  if (LHSSign == Unknown)
+    return false;
+
+  IsSigned = (LHSSign == Signed);
+
+  // The RHS can be a demotable op or a constant
+  if (ConstantSDNode *CI = dyn_cast<ConstantSDNode>(RHS)) {
+    APInt Val = CI->getAPIntValue();
+    if (LHSSign == Unsigned) {
+      if (Val.isIntN(OptSize)) {
+        return true;
+      }
+      return false;
+    } else {
+      if (Val.isSignedIntN(OptSize)) {
+        return true;
+      }
+      return false;
+    }
+  } else {
+    OperandSignedness RHSSign;
+    if (!IsMulWideOperandDemotable(RHS, OptSize, RHSSign))
+      return false;
+
+    if (LHSSign != RHSSign)
+      return false;
+
+    return true;
+  }
+}
+
+/// TryMULWIDECombine - Attempt to replace a multiply of M bits with a multiply
+/// of M/2 bits that produces an M-bit result (i.e. mul.wide). This transform
+/// works on both multiply DAG nodes and SHL DAG nodes with a constant shift
+/// amount.
+static SDValue TryMULWIDECombine(SDNode *N,
+                                 TargetLowering::DAGCombinerInfo &DCI) {
+  EVT MulType = N->getValueType(0);
+  if (MulType != MVT::i32 && MulType != MVT::i64) {
+    return SDValue();
+  }
+
+  unsigned OptSize = MulType.getSizeInBits() >> 1;
+  SDValue LHS = N->getOperand(0);
+  SDValue RHS = N->getOperand(1);
+
+  // Canonicalize the multiply so the constant (if any) is on the right
+  if (N->getOpcode() == ISD::MUL) {
+    if (isa<ConstantSDNode>(LHS)) {
+      std::swap(LHS, RHS);
+    }
+  }
+
+  // If we have a SHL, determine the actual multiply amount
+  if (N->getOpcode() == ISD::SHL) {
+    ConstantSDNode *ShlRHS = dyn_cast<ConstantSDNode>(RHS);
+    if (!ShlRHS) {
+      return SDValue();
+    }
+
+    APInt ShiftAmt = ShlRHS->getAPIntValue();
+    unsigned BitWidth = MulType.getSizeInBits();
+    if (ShiftAmt.sge(0) && ShiftAmt.slt(BitWidth)) {
+      APInt MulVal = APInt(BitWidth, 1) << ShiftAmt;
+      RHS = DCI.DAG.getConstant(MulVal, MulType);
+    } else {
+      return SDValue();
+    }
+  }
+
+  bool Signed;
+  // Verify that our operands are demotable
+  if (!AreMulWideOperandsDemotable(LHS, RHS, OptSize, Signed)) {
+    return SDValue();
+  }
+
+  EVT DemotedVT;
+  if (MulType == MVT::i32) {
+    DemotedVT = MVT::i16;
+  } else {
+    DemotedVT = MVT::i32;
+  }
+
+  // Truncate the operands to the correct size. Note that these are just for
+  // type consistency and will (likely) be eliminated in later phases.
+  SDValue TruncLHS =
+    DCI.DAG.getNode(ISD::TRUNCATE, SDLoc(N), DemotedVT, LHS);
+  SDValue TruncRHS =
+    DCI.DAG.getNode(ISD::TRUNCATE, SDLoc(N), DemotedVT, RHS);
+
+  unsigned Opc;
+  if (Signed) {
+    Opc = NVPTXISD::MUL_WIDE_SIGNED;
+  } else {
+    Opc = NVPTXISD::MUL_WIDE_UNSIGNED;
+  }
+
+  return DCI.DAG.getNode(Opc, SDLoc(N), MulType, TruncLHS, TruncRHS);
+}
+
+/// PerformMULCombine - Runs PTX-specific DAG combine patterns on MUL nodes.
+static SDValue PerformMULCombine(SDNode *N,
+                                 TargetLowering::DAGCombinerInfo &DCI,
+                                 CodeGenOpt::Level OptLevel) {
+  if (OptLevel > 0) {
+    // Try mul.wide combining at OptLevel > 0
+    SDValue Ret = TryMULWIDECombine(N, DCI);
+    if (Ret.getNode())
+      return Ret;
+  }
+
+  return SDValue();
+}
+
+/// PerformSHLCombine - Runs PTX-specific DAG combine patterns on SHL nodes.
+static SDValue PerformSHLCombine(SDNode *N,
+                                 TargetLowering::DAGCombinerInfo &DCI,
+                                 CodeGenOpt::Level OptLevel) {
+  if (OptLevel > 0) {
+    // Try mul.wide combining at OptLevel > 0
+    SDValue Ret = TryMULWIDECombine(N, DCI);
+    if (Ret.getNode())
+      return Ret;
+  }
+
+  return SDValue();
+}
+
+SDValue NVPTXTargetLowering::PerformDAGCombine(SDNode *N,
+                                               DAGCombinerInfo &DCI) const {
+  CodeGenOpt::Level OptLevel = getTargetMachine().getOptLevel();
+  switch (N->getOpcode()) {
+    default: break;
+    case ISD::ADD:
+    case ISD::FADD:
+      return PerformADDCombine(N, DCI, nvptxSubtarget, OptLevel);
+    case ISD::MUL:
+      return PerformMULCombine(N, DCI, OptLevel);
+    case ISD::SHL:
+      return PerformSHLCombine(N, DCI, OptLevel);
+    case ISD::AND:
+      return PerformANDCombine(N, DCI);
+  }
+  return SDValue();
+}
+
 /// ReplaceVectorLoad - Convert vector loads into multi-output scalar loads.
 static void ReplaceLoadVector(SDNode *N, SelectionDAG &DAG,
+                              const DataLayout *TD,
                               SmallVectorImpl<SDValue> &Results) {
   EVT ResVT = N->getValueType(0);
   SDLoc DL(N);
@@ -2070,12 +4259,26 @@ static void ReplaceLoadVector(SDNode *N, SelectionDAG &DAG,
     break;
   }
 
+  LoadSDNode *LD = cast<LoadSDNode>(N);
+
+  unsigned Align = LD->getAlignment();
+  unsigned PrefAlign =
+    TD->getPrefTypeAlignment(ResVT.getTypeForEVT(*DAG.getContext()));
+  if (Align < PrefAlign) {
+    // This load is not sufficiently aligned, so bail out and let this vector
+    // load be scalarized.  Note that we may still be able to emit smaller
+    // vector loads.  For example, if we are loading a <4 x float> with an
+    // alignment of 8, this check will fail but the legalizer will try again
+    // with 2 x <2 x float>, which will succeed with an alignment of 8.
+    return;
+  }
+
   EVT EltVT = ResVT.getVectorElementType();
   unsigned NumElts = ResVT.getVectorNumElements();
 
   // Since LoadV2 is a target node, we cannot rely on DAG type legalization.
   // Therefore, we must ensure the type is legal.  For i1 and i8, we set the
-  // loaded type to i16 and propogate the "real" type as the memory type.
+  // loaded type to i16 and propagate the "real" type as the memory type.
   bool NeedTrunc = false;
   if (EltVT.getSizeInBits() < 16) {
     EltVT = MVT::i16;
@@ -2095,7 +4298,7 @@ static void ReplaceLoadVector(SDNode *N, SelectionDAG &DAG,
   case 4: {
     Opcode = NVPTXISD::LoadV4;
     EVT ListVTs[] = { EltVT, EltVT, EltVT, EltVT, MVT::Other };
-    LdResVTs = DAG.getVTList(ListVTs, 5);
+    LdResVTs = DAG.getVTList(ListVTs);
     break;
   }
   }
@@ -2106,14 +4309,12 @@ static void ReplaceLoadVector(SDNode *N, SelectionDAG &DAG,
   for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
     OtherOps.push_back(N->getOperand(i));
 
-  LoadSDNode *LD = cast<LoadSDNode>(N);
-
   // The select routine does not have access to the LoadSDNode instance, so
   // pass along the extension information
   OtherOps.push_back(DAG.getIntPtrConstant(LD->getExtensionType()));
 
-  SDValue NewLD = DAG.getMemIntrinsicNode(Opcode, DL, LdResVTs, &OtherOps[0],
-                                          OtherOps.size(), LD->getMemoryVT(),
+  SDValue NewLD = DAG.getMemIntrinsicNode(Opcode, DL, LdResVTs, OtherOps,
+                                          LD->getMemoryVT(),
                                           LD->getMemOperand());
 
   SmallVector<SDValue, 4> ScalarRes;
@@ -2127,8 +4328,7 @@ static void ReplaceLoadVector(SDNode *N, SelectionDAG &DAG,
 
   SDValue LoadChain = NewLD.getValue(NumElts);
 
-  SDValue BuildVec =
-      DAG.getNode(ISD::BUILD_VECTOR, DL, ResVT, &ScalarRes[0], NumElts);
+  SDValue BuildVec = DAG.getNode(ISD::BUILD_VECTOR, DL, ResVT, ScalarRes);
 
   Results.push_back(BuildVec);
   Results.push_back(LoadChain);
@@ -2162,7 +4362,7 @@ static void ReplaceINTRINSIC_W_CHAIN(SDNode *N, SelectionDAG &DAG,
       // Since LDU/LDG are target nodes, we cannot rely on DAG type
       // legalization.
       // Therefore, we must ensure the type is legal.  For i1 and i8, we set the
-      // loaded type to i16 and propogate the "real" type as the memory type.
+      // loaded type to i16 and propagate the "real" type as the memory type.
       bool NeedTrunc = false;
       if (EltVT.getSizeInBits() < 16) {
         EltVT = MVT::i16;
@@ -2208,7 +4408,7 @@ static void ReplaceINTRINSIC_W_CHAIN(SDNode *N, SelectionDAG &DAG,
           break;
         }
         EVT ListVTs[] = { EltVT, EltVT, EltVT, EltVT, MVT::Other };
-        LdResVTs = DAG.getVTList(ListVTs, 5);
+        LdResVTs = DAG.getVTList(ListVTs);
         break;
       }
       }
@@ -2225,9 +4425,9 @@ static void ReplaceINTRINSIC_W_CHAIN(SDNode *N, SelectionDAG &DAG,
 
       MemIntrinsicSDNode *MemSD = cast<MemIntrinsicSDNode>(N);
 
-      SDValue NewLD = DAG.getMemIntrinsicNode(
-          Opcode, DL, LdResVTs, &OtherOps[0], OtherOps.size(),
-          MemSD->getMemoryVT(), MemSD->getMemOperand());
+      SDValue NewLD = DAG.getMemIntrinsicNode(Opcode, DL, LdResVTs, OtherOps,
+                                              MemSD->getMemoryVT(),
+                                              MemSD->getMemOperand());
 
       SmallVector<SDValue, 4> ScalarRes;
 
@@ -2242,7 +4442,7 @@ static void ReplaceINTRINSIC_W_CHAIN(SDNode *N, SelectionDAG &DAG,
       SDValue LoadChain = NewLD.getValue(NumElts);
 
       SDValue BuildVec =
-          DAG.getNode(ISD::BUILD_VECTOR, DL, ResVT, &ScalarRes[0], NumElts);
+          DAG.getNode(ISD::BUILD_VECTOR, DL, ResVT, ScalarRes);
 
       Results.push_back(BuildVec);
       Results.push_back(LoadChain);
@@ -2264,8 +4464,8 @@ static void ReplaceINTRINSIC_W_CHAIN(SDNode *N, SelectionDAG &DAG,
       // We make sure the memory type is i8, which will be used during isel
       // to select the proper instruction.
       SDValue NewLD =
-          DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, LdResVTs, &Ops[0],
-                                  Ops.size(), MVT::i8, MemSD->getMemOperand());
+          DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, LdResVTs, Ops,
+                                  MVT::i8, MemSD->getMemOperand());
 
       Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i8,
                                     NewLD.getValue(0)));
@@ -2281,7 +4481,7 @@ void NVPTXTargetLowering::ReplaceNodeResults(
   default:
     report_fatal_error("Unhandled custom legalization");
   case ISD::LOAD:
-    ReplaceLoadVector(N, DAG, Results);
+    ReplaceLoadVector(N, DAG, getDataLayout(), Results);
     return;
   case ISD::INTRINSIC_W_CHAIN:
     ReplaceINTRINSIC_W_CHAIN(N, DAG, Results);
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXISelLowering.h b/contrib/llvm/lib/Target/NVPTX/NVPTXISelLowering.h
index 66e708fcea07..bef6ed9faad6 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXISelLowering.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXISelLowering.h
@@ -16,7 +16,6 @@
 #define NVPTXISELLOWERING_H
 
 #include "NVPTX.h"
-#include "NVPTXSubtarget.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/Target/TargetLowering.h"
 
@@ -50,6 +49,11 @@ enum NodeType {
   CallSeqBegin,
   CallSeqEnd,
   CallPrototype,
+  FUN_SHFL_CLAMP,
+  FUN_SHFR_CLAMP,
+  MUL_WIDE_SIGNED,
+  MUL_WIDE_UNSIGNED,
+  IMAD,
   Dummy,
 
   LoadV2 = ISD::FIRST_TARGET_MEMORY_OPCODE,
@@ -70,78 +74,440 @@ enum NodeType {
   StoreParamU32, // to zext and store a <32bit value, not used currently 
   StoreRetval,
   StoreRetvalV2,
-  StoreRetvalV4
+  StoreRetvalV4,
+
+  // Texture intrinsics
+  Tex1DFloatS32,
+  Tex1DFloatFloat,
+  Tex1DFloatFloatLevel,
+  Tex1DFloatFloatGrad,
+  Tex1DS32S32,
+  Tex1DS32Float,
+  Tex1DS32FloatLevel,
+  Tex1DS32FloatGrad,
+  Tex1DU32S32,
+  Tex1DU32Float,
+  Tex1DU32FloatLevel,
+  Tex1DU32FloatGrad,
+  Tex1DArrayFloatS32,
+  Tex1DArrayFloatFloat,
+  Tex1DArrayFloatFloatLevel,
+  Tex1DArrayFloatFloatGrad,
+  Tex1DArrayS32S32,
+  Tex1DArrayS32Float,
+  Tex1DArrayS32FloatLevel,
+  Tex1DArrayS32FloatGrad,
+  Tex1DArrayU32S32,
+  Tex1DArrayU32Float,
+  Tex1DArrayU32FloatLevel,
+  Tex1DArrayU32FloatGrad,
+  Tex2DFloatS32,
+  Tex2DFloatFloat,
+  Tex2DFloatFloatLevel,
+  Tex2DFloatFloatGrad,
+  Tex2DS32S32,
+  Tex2DS32Float,
+  Tex2DS32FloatLevel,
+  Tex2DS32FloatGrad,
+  Tex2DU32S32,
+  Tex2DU32Float,
+  Tex2DU32FloatLevel,
+  Tex2DU32FloatGrad,
+  Tex2DArrayFloatS32,
+  Tex2DArrayFloatFloat,
+  Tex2DArrayFloatFloatLevel,
+  Tex2DArrayFloatFloatGrad,
+  Tex2DArrayS32S32,
+  Tex2DArrayS32Float,
+  Tex2DArrayS32FloatLevel,
+  Tex2DArrayS32FloatGrad,
+  Tex2DArrayU32S32,
+  Tex2DArrayU32Float,
+  Tex2DArrayU32FloatLevel,
+  Tex2DArrayU32FloatGrad,
+  Tex3DFloatS32,
+  Tex3DFloatFloat,
+  Tex3DFloatFloatLevel,
+  Tex3DFloatFloatGrad,
+  Tex3DS32S32,
+  Tex3DS32Float,
+  Tex3DS32FloatLevel,
+  Tex3DS32FloatGrad,
+  Tex3DU32S32,
+  Tex3DU32Float,
+  Tex3DU32FloatLevel,
+  Tex3DU32FloatGrad,
+  TexCubeFloatFloat,
+  TexCubeFloatFloatLevel,
+  TexCubeS32Float,
+  TexCubeS32FloatLevel,
+  TexCubeU32Float,
+  TexCubeU32FloatLevel,
+  TexCubeArrayFloatFloat,
+  TexCubeArrayFloatFloatLevel,
+  TexCubeArrayS32Float,
+  TexCubeArrayS32FloatLevel,
+  TexCubeArrayU32Float,
+  TexCubeArrayU32FloatLevel,
+  Tld4R2DFloatFloat,
+  Tld4G2DFloatFloat,
+  Tld4B2DFloatFloat,
+  Tld4A2DFloatFloat,
+  Tld4R2DS64Float,
+  Tld4G2DS64Float,
+  Tld4B2DS64Float,
+  Tld4A2DS64Float,
+  Tld4R2DU64Float,
+  Tld4G2DU64Float,
+  Tld4B2DU64Float,
+  Tld4A2DU64Float,
+  TexUnified1DFloatS32,
+  TexUnified1DFloatFloat,
+  TexUnified1DFloatFloatLevel,
+  TexUnified1DFloatFloatGrad,
+  TexUnified1DS32S32,
+  TexUnified1DS32Float,
+  TexUnified1DS32FloatLevel,
+  TexUnified1DS32FloatGrad,
+  TexUnified1DU32S32,
+  TexUnified1DU32Float,
+  TexUnified1DU32FloatLevel,
+  TexUnified1DU32FloatGrad,
+  TexUnified1DArrayFloatS32,
+  TexUnified1DArrayFloatFloat,
+  TexUnified1DArrayFloatFloatLevel,
+  TexUnified1DArrayFloatFloatGrad,
+  TexUnified1DArrayS32S32,
+  TexUnified1DArrayS32Float,
+  TexUnified1DArrayS32FloatLevel,
+  TexUnified1DArrayS32FloatGrad,
+  TexUnified1DArrayU32S32,
+  TexUnified1DArrayU32Float,
+  TexUnified1DArrayU32FloatLevel,
+  TexUnified1DArrayU32FloatGrad,
+  TexUnified2DFloatS32,
+  TexUnified2DFloatFloat,
+  TexUnified2DFloatFloatLevel,
+  TexUnified2DFloatFloatGrad,
+  TexUnified2DS32S32,
+  TexUnified2DS32Float,
+  TexUnified2DS32FloatLevel,
+  TexUnified2DS32FloatGrad,
+  TexUnified2DU32S32,
+  TexUnified2DU32Float,
+  TexUnified2DU32FloatLevel,
+  TexUnified2DU32FloatGrad,
+  TexUnified2DArrayFloatS32,
+  TexUnified2DArrayFloatFloat,
+  TexUnified2DArrayFloatFloatLevel,
+  TexUnified2DArrayFloatFloatGrad,
+  TexUnified2DArrayS32S32,
+  TexUnified2DArrayS32Float,
+  TexUnified2DArrayS32FloatLevel,
+  TexUnified2DArrayS32FloatGrad,
+  TexUnified2DArrayU32S32,
+  TexUnified2DArrayU32Float,
+  TexUnified2DArrayU32FloatLevel,
+  TexUnified2DArrayU32FloatGrad,
+  TexUnified3DFloatS32,
+  TexUnified3DFloatFloat,
+  TexUnified3DFloatFloatLevel,
+  TexUnified3DFloatFloatGrad,
+  TexUnified3DS32S32,
+  TexUnified3DS32Float,
+  TexUnified3DS32FloatLevel,
+  TexUnified3DS32FloatGrad,
+  TexUnified3DU32S32,
+  TexUnified3DU32Float,
+  TexUnified3DU32FloatLevel,
+  TexUnified3DU32FloatGrad,
+  TexUnifiedCubeFloatFloat,
+  TexUnifiedCubeFloatFloatLevel,
+  TexUnifiedCubeS32Float,
+  TexUnifiedCubeS32FloatLevel,
+  TexUnifiedCubeU32Float,
+  TexUnifiedCubeU32FloatLevel,
+  TexUnifiedCubeArrayFloatFloat,
+  TexUnifiedCubeArrayFloatFloatLevel,
+  TexUnifiedCubeArrayS32Float,
+  TexUnifiedCubeArrayS32FloatLevel,
+  TexUnifiedCubeArrayU32Float,
+  TexUnifiedCubeArrayU32FloatLevel,
+  Tld4UnifiedR2DFloatFloat,
+  Tld4UnifiedG2DFloatFloat,
+  Tld4UnifiedB2DFloatFloat,
+  Tld4UnifiedA2DFloatFloat,
+  Tld4UnifiedR2DS64Float,
+  Tld4UnifiedG2DS64Float,
+  Tld4UnifiedB2DS64Float,
+  Tld4UnifiedA2DS64Float,
+  Tld4UnifiedR2DU64Float,
+  Tld4UnifiedG2DU64Float,
+  Tld4UnifiedB2DU64Float,
+  Tld4UnifiedA2DU64Float,
+
+  // Surface intrinsics
+  Suld1DI8Clamp,
+  Suld1DI16Clamp,
+  Suld1DI32Clamp,
+  Suld1DI64Clamp,
+  Suld1DV2I8Clamp,
+  Suld1DV2I16Clamp,
+  Suld1DV2I32Clamp,
+  Suld1DV2I64Clamp,
+  Suld1DV4I8Clamp,
+  Suld1DV4I16Clamp,
+  Suld1DV4I32Clamp,
+
+  Suld1DArrayI8Clamp,
+  Suld1DArrayI16Clamp,
+  Suld1DArrayI32Clamp,
+  Suld1DArrayI64Clamp,
+  Suld1DArrayV2I8Clamp,
+  Suld1DArrayV2I16Clamp,
+  Suld1DArrayV2I32Clamp,
+  Suld1DArrayV2I64Clamp,
+  Suld1DArrayV4I8Clamp,
+  Suld1DArrayV4I16Clamp,
+  Suld1DArrayV4I32Clamp,
+
+  Suld2DI8Clamp,
+  Suld2DI16Clamp,
+  Suld2DI32Clamp,
+  Suld2DI64Clamp,
+  Suld2DV2I8Clamp,
+  Suld2DV2I16Clamp,
+  Suld2DV2I32Clamp,
+  Suld2DV2I64Clamp,
+  Suld2DV4I8Clamp,
+  Suld2DV4I16Clamp,
+  Suld2DV4I32Clamp,
+
+  Suld2DArrayI8Clamp,
+  Suld2DArrayI16Clamp,
+  Suld2DArrayI32Clamp,
+  Suld2DArrayI64Clamp,
+  Suld2DArrayV2I8Clamp,
+  Suld2DArrayV2I16Clamp,
+  Suld2DArrayV2I32Clamp,
+  Suld2DArrayV2I64Clamp,
+  Suld2DArrayV4I8Clamp,
+  Suld2DArrayV4I16Clamp,
+  Suld2DArrayV4I32Clamp,
+
+  Suld3DI8Clamp,
+  Suld3DI16Clamp,
+  Suld3DI32Clamp,
+  Suld3DI64Clamp,
+  Suld3DV2I8Clamp,
+  Suld3DV2I16Clamp,
+  Suld3DV2I32Clamp,
+  Suld3DV2I64Clamp,
+  Suld3DV4I8Clamp,
+  Suld3DV4I16Clamp,
+  Suld3DV4I32Clamp,
+
+  Suld1DI8Trap,
+  Suld1DI16Trap,
+  Suld1DI32Trap,
+  Suld1DI64Trap,
+  Suld1DV2I8Trap,
+  Suld1DV2I16Trap,
+  Suld1DV2I32Trap,
+  Suld1DV2I64Trap,
+  Suld1DV4I8Trap,
+  Suld1DV4I16Trap,
+  Suld1DV4I32Trap,
+
+  Suld1DArrayI8Trap,
+  Suld1DArrayI16Trap,
+  Suld1DArrayI32Trap,
+  Suld1DArrayI64Trap,
+  Suld1DArrayV2I8Trap,
+  Suld1DArrayV2I16Trap,
+  Suld1DArrayV2I32Trap,
+  Suld1DArrayV2I64Trap,
+  Suld1DArrayV4I8Trap,
+  Suld1DArrayV4I16Trap,
+  Suld1DArrayV4I32Trap,
+
+  Suld2DI8Trap,
+  Suld2DI16Trap,
+  Suld2DI32Trap,
+  Suld2DI64Trap,
+  Suld2DV2I8Trap,
+  Suld2DV2I16Trap,
+  Suld2DV2I32Trap,
+  Suld2DV2I64Trap,
+  Suld2DV4I8Trap,
+  Suld2DV4I16Trap,
+  Suld2DV4I32Trap,
+
+  Suld2DArrayI8Trap,
+  Suld2DArrayI16Trap,
+  Suld2DArrayI32Trap,
+  Suld2DArrayI64Trap,
+  Suld2DArrayV2I8Trap,
+  Suld2DArrayV2I16Trap,
+  Suld2DArrayV2I32Trap,
+  Suld2DArrayV2I64Trap,
+  Suld2DArrayV4I8Trap,
+  Suld2DArrayV4I16Trap,
+  Suld2DArrayV4I32Trap,
+
+  Suld3DI8Trap,
+  Suld3DI16Trap,
+  Suld3DI32Trap,
+  Suld3DI64Trap,
+  Suld3DV2I8Trap,
+  Suld3DV2I16Trap,
+  Suld3DV2I32Trap,
+  Suld3DV2I64Trap,
+  Suld3DV4I8Trap,
+  Suld3DV4I16Trap,
+  Suld3DV4I32Trap,
+
+  Suld1DI8Zero,
+  Suld1DI16Zero,
+  Suld1DI32Zero,
+  Suld1DI64Zero,
+  Suld1DV2I8Zero,
+  Suld1DV2I16Zero,
+  Suld1DV2I32Zero,
+  Suld1DV2I64Zero,
+  Suld1DV4I8Zero,
+  Suld1DV4I16Zero,
+  Suld1DV4I32Zero,
+
+  Suld1DArrayI8Zero,
+  Suld1DArrayI16Zero,
+  Suld1DArrayI32Zero,
+  Suld1DArrayI64Zero,
+  Suld1DArrayV2I8Zero,
+  Suld1DArrayV2I16Zero,
+  Suld1DArrayV2I32Zero,
+  Suld1DArrayV2I64Zero,
+  Suld1DArrayV4I8Zero,
+  Suld1DArrayV4I16Zero,
+  Suld1DArrayV4I32Zero,
+
+  Suld2DI8Zero,
+  Suld2DI16Zero,
+  Suld2DI32Zero,
+  Suld2DI64Zero,
+  Suld2DV2I8Zero,
+  Suld2DV2I16Zero,
+  Suld2DV2I32Zero,
+  Suld2DV2I64Zero,
+  Suld2DV4I8Zero,
+  Suld2DV4I16Zero,
+  Suld2DV4I32Zero,
+
+  Suld2DArrayI8Zero,
+  Suld2DArrayI16Zero,
+  Suld2DArrayI32Zero,
+  Suld2DArrayI64Zero,
+  Suld2DArrayV2I8Zero,
+  Suld2DArrayV2I16Zero,
+  Suld2DArrayV2I32Zero,
+  Suld2DArrayV2I64Zero,
+  Suld2DArrayV4I8Zero,
+  Suld2DArrayV4I16Zero,
+  Suld2DArrayV4I32Zero,
+
+  Suld3DI8Zero,
+  Suld3DI16Zero,
+  Suld3DI32Zero,
+  Suld3DI64Zero,
+  Suld3DV2I8Zero,
+  Suld3DV2I16Zero,
+  Suld3DV2I32Zero,
+  Suld3DV2I64Zero,
+  Suld3DV4I8Zero,
+  Suld3DV4I16Zero,
+  Suld3DV4I32Zero
 };
 }
 
+class NVPTXSubtarget;
+
 //===--------------------------------------------------------------------===//
 // TargetLowering Implementation
 //===--------------------------------------------------------------------===//
 class NVPTXTargetLowering : public TargetLowering {
 public:
   explicit NVPTXTargetLowering(NVPTXTargetMachine &TM);
-  virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
 
   SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerGlobalAddress(const GlobalValue *GV, int64_t Offset,
                              SelectionDAG &DAG) const;
 
-  virtual const char *getTargetNodeName(unsigned Opcode) const;
+  const char *getTargetNodeName(unsigned Opcode) const override;
 
   bool isTypeSupportedInIntrinsic(MVT VT) const;
 
   bool getTgtMemIntrinsic(IntrinsicInfo &Info, const CallInst &I,
-                          unsigned Intrinsic) const;
+                          unsigned Intrinsic) const override;
 
   /// isLegalAddressingMode - Return true if the addressing mode represented
   /// by AM is legal for this target, for a load/store of the specified type
   /// Used to guide target specific optimizations, like loop strength
   /// reduction (LoopStrengthReduce.cpp) and memory optimization for
   /// address mode (CodeGenPrepare.cpp)
-  virtual bool isLegalAddressingMode(const AddrMode &AM, Type *Ty) const;
+  bool isLegalAddressingMode(const AddrMode &AM, Type *Ty) const override;
 
   /// getFunctionAlignment - Return the Log2 alignment of this function.
-  virtual unsigned getFunctionAlignment(const Function *F) const;
+  unsigned getFunctionAlignment(const Function *F) const;
 
-  virtual EVT getSetCCResultType(LLVMContext &, EVT VT) const {
+  EVT getSetCCResultType(LLVMContext &Ctx, EVT VT) const override {
     if (VT.isVector())
-      return MVT::getVectorVT(MVT::i1, VT.getVectorNumElements());
+      return EVT::getVectorVT(Ctx, MVT::i1, VT.getVectorNumElements());
     return MVT::i1;
   }
 
-  ConstraintType getConstraintType(const std::string &Constraint) const;
+  ConstraintType
+  getConstraintType(const std::string &Constraint) const override;
   std::pair<unsigned, const TargetRegisterClass *>
-  getRegForInlineAsmConstraint(const std::string &Constraint, MVT VT) const;
+  getRegForInlineAsmConstraint(const std::string &Constraint,
+                               MVT VT) const override;
 
-  virtual SDValue LowerFormalArguments(
+  SDValue LowerFormalArguments(
       SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
       const SmallVectorImpl<ISD::InputArg> &Ins, SDLoc dl, SelectionDAG &DAG,
-      SmallVectorImpl<SDValue> &InVals) const;
+      SmallVectorImpl<SDValue> &InVals) const override;
 
-  virtual SDValue
-  LowerCall(CallLoweringInfo &CLI, SmallVectorImpl<SDValue> &InVals) const;
+  SDValue LowerCall(CallLoweringInfo &CLI,
+                    SmallVectorImpl<SDValue> &InVals) const override;
 
   std::string getPrototype(Type *, const ArgListTy &,
                            const SmallVectorImpl<ISD::OutputArg> &,
                            unsigned retAlignment,
                            const ImmutableCallSite *CS) const;
 
-  virtual SDValue
+  SDValue
   LowerReturn(SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
               const SmallVectorImpl<ISD::OutputArg> &Outs,
               const SmallVectorImpl<SDValue> &OutVals, SDLoc dl,
-              SelectionDAG &DAG) const;
+              SelectionDAG &DAG) const override;
 
-  virtual void LowerAsmOperandForConstraint(SDValue Op, std::string &Constraint,
-                                            std::vector<SDValue> &Ops,
-                                            SelectionDAG &DAG) const;
+  void LowerAsmOperandForConstraint(SDValue Op, std::string &Constraint,
+                                    std::vector<SDValue> &Ops,
+                                    SelectionDAG &DAG) const override;
 
   NVPTXTargetMachine *nvTM;
 
   // PTX always uses 32-bit shift amounts
-  virtual MVT getScalarShiftAmountTy(EVT LHSTy) const { return MVT::i32; }
+  MVT getScalarShiftAmountTy(EVT LHSTy) const override { return MVT::i32; }
+
+  TargetLoweringBase::LegalizeTypeAction
+  getPreferredVectorAction(EVT VT) const override;
 
-  virtual bool shouldSplitVectorElementType(EVT VT) const;
+  bool allowFMA(MachineFunction &MF, CodeGenOpt::Level OptLevel) const;
+
+  virtual bool isFMAFasterThanFMulAndFAdd(EVT) const {
+    return true;
+  }
 
 private:
   const NVPTXSubtarget &nvptxSubtarget; // cache the subtarget here
@@ -160,8 +526,12 @@ private:
   SDValue LowerSTOREi1(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerSTOREVector(SDValue Op, SelectionDAG &DAG) const;
 
-  virtual void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue> &Results,
-                                  SelectionDAG &DAG) const;
+  SDValue LowerShiftRightParts(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerShiftLeftParts(SDValue Op, SelectionDAG &DAG) const;
+
+  void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue> &Results,
+                          SelectionDAG &DAG) const override;
+  SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
 
   unsigned getArgumentAlignment(SDValue Callee, const ImmutableCallSite *CS,
                                 Type *Ty, unsigned Idx) const;
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXImageOptimizer.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXImageOptimizer.cpp
new file mode 100644
index 000000000000..a98fb37f6e25
--- /dev/null
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXImageOptimizer.cpp
@@ -0,0 +1,178 @@
+//===-- NVPTXImageOptimizer.cpp - Image optimization pass -----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source 
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass implements IR-level optimizations of image access code,
+// including:
+//
+// 1. Eliminate istypep intrinsics when image access qualifier is known
+//
+//===----------------------------------------------------------------------===//
+
+#include "NVPTX.h"
+#include "NVPTXUtilities.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Module.h"
+#include "llvm/Pass.h"
+#include "llvm/Analysis/ConstantFolding.h"
+
+using namespace llvm;
+
+namespace {
+class NVPTXImageOptimizer : public FunctionPass {
+private:
+  static char ID;
+  SmallVector<Instruction*, 4> InstrToDelete;
+
+public:
+  NVPTXImageOptimizer();
+
+  bool runOnFunction(Function &F) override;
+
+private:
+  bool replaceIsTypePSampler(Instruction &I);
+  bool replaceIsTypePSurface(Instruction &I);
+  bool replaceIsTypePTexture(Instruction &I);
+  Value *cleanupValue(Value *V);
+  void replaceWith(Instruction *From, ConstantInt *To);
+};
+}
+
+char NVPTXImageOptimizer::ID = 0;
+
+NVPTXImageOptimizer::NVPTXImageOptimizer()
+  : FunctionPass(ID) {}
+
+bool NVPTXImageOptimizer::runOnFunction(Function &F) {
+  bool Changed = false;
+  InstrToDelete.clear();
+
+  // Look for call instructions in the function
+  for (Function::iterator BI = F.begin(), BE = F.end(); BI != BE;
+       ++BI) {
+    for (BasicBlock::iterator I = (*BI).begin(), E = (*BI).end();
+         I != E; ++I) {
+      Instruction &Instr = *I;
+      if (CallInst *CI = dyn_cast<CallInst>(I)) {
+        Function *CalledF = CI->getCalledFunction();
+        if (CalledF && CalledF->isIntrinsic()) {
+          // This is an intrinsic function call, check if its an istypep
+          switch (CalledF->getIntrinsicID()) {
+          default: break;
+          case Intrinsic::nvvm_istypep_sampler:
+            Changed |= replaceIsTypePSampler(Instr);
+            break;
+          case Intrinsic::nvvm_istypep_surface:
+            Changed |= replaceIsTypePSurface(Instr);
+            break;
+          case Intrinsic::nvvm_istypep_texture:
+            Changed |= replaceIsTypePTexture(Instr);
+            break;
+          }
+        }
+      }
+    }
+  }
+
+  // Delete any istypep instances we replaced in the IR
+  for (unsigned i = 0, e = InstrToDelete.size(); i != e; ++i)
+    InstrToDelete[i]->eraseFromParent();
+
+  return Changed;
+}
+
+bool NVPTXImageOptimizer::replaceIsTypePSampler(Instruction &I) {
+  Value *TexHandle = cleanupValue(I.getOperand(0));
+  if (isSampler(*TexHandle)) {
+    // This is an OpenCL sampler, so it must be a samplerref
+    replaceWith(&I, ConstantInt::getTrue(I.getContext()));
+    return true;
+  } else if (isImageWriteOnly(*TexHandle) ||
+             isImageReadWrite(*TexHandle) ||
+             isImageReadOnly(*TexHandle)) {
+    // This is an OpenCL image, so it cannot be a samplerref
+    replaceWith(&I, ConstantInt::getFalse(I.getContext()));
+    return true;
+  } else {
+    // The image type is unknown, so we cannot eliminate the intrinsic
+    return false;
+  }
+}
+
+bool NVPTXImageOptimizer::replaceIsTypePSurface(Instruction &I) {
+  Value *TexHandle = cleanupValue(I.getOperand(0));
+  if (isImageReadWrite(*TexHandle) ||
+      isImageWriteOnly(*TexHandle)) {
+    // This is an OpenCL read-only/read-write image, so it must be a surfref
+    replaceWith(&I, ConstantInt::getTrue(I.getContext()));
+    return true;
+  } else if (isImageReadOnly(*TexHandle) ||
+             isSampler(*TexHandle)) {
+    // This is an OpenCL read-only/ imageor sampler, so it cannot be
+    // a surfref
+    replaceWith(&I, ConstantInt::getFalse(I.getContext()));
+    return true;
+  } else {
+    // The image type is unknown, so we cannot eliminate the intrinsic
+    return false;
+  }
+}
+
+bool NVPTXImageOptimizer::replaceIsTypePTexture(Instruction &I) {
+  Value *TexHandle = cleanupValue(I.getOperand(0));
+  if (isImageReadOnly(*TexHandle)) {
+    // This is an OpenCL read-only image, so it must be a texref
+    replaceWith(&I, ConstantInt::getTrue(I.getContext()));
+    return true;
+  } else if (isImageWriteOnly(*TexHandle) ||
+             isImageReadWrite(*TexHandle) ||
+             isSampler(*TexHandle)) {
+    // This is an OpenCL read-write/write-only image or a sampler, so it
+    // cannot be a texref
+    replaceWith(&I, ConstantInt::getFalse(I.getContext()));
+    return true;
+  } else {
+    // The image type is unknown, so we cannot eliminate the intrinsic
+    return false;
+  }
+}
+
+void NVPTXImageOptimizer::replaceWith(Instruction *From, ConstantInt *To) {
+  // We implement "poor man's DCE" here to make sure any code that is no longer
+  // live is actually unreachable and can be trivially eliminated by the
+  // unreachable block elimination pass.
+  for (CallInst::use_iterator UI = From->use_begin(), UE = From->use_end();
+       UI != UE; ++UI) {
+    if (BranchInst *BI = dyn_cast<BranchInst>(*UI)) {
+      if (BI->isUnconditional()) continue;
+      BasicBlock *Dest;
+      if (To->isZero())
+        // Get false block
+        Dest = BI->getSuccessor(1);
+      else
+        // Get true block
+        Dest = BI->getSuccessor(0);
+      BranchInst::Create(Dest, BI);
+      InstrToDelete.push_back(BI);
+    }
+  }
+  From->replaceAllUsesWith(To);
+  InstrToDelete.push_back(From);
+}
+
+Value *NVPTXImageOptimizer::cleanupValue(Value *V) {
+  if (ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(V)) {
+    return cleanupValue(EVI->getAggregateOperand());
+  }
+  return V;
+}
+
+FunctionPass *llvm::createNVPTXImageOptimizerPass() {
+  return new NVPTXImageOptimizer();
+}
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXInstrFormats.td b/contrib/llvm/lib/Target/NVPTX/NVPTXInstrFormats.td
index f11f1b8f96fc..ffcb5d5273a2 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXInstrFormats.td
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXInstrFormats.td
@@ -36,8 +36,24 @@ class NVPTXInst<dag outs, dag ins, string asmstr, list<dag> pattern>
   bit IsLoad = 0;
   bit IsStore = 0;
 
-  let TSFlags{3-0} = VecInstType;
-  let TSFlags{4-4} = IsSimpleMove;
-  let TSFlags{5-5} = IsLoad;
-  let TSFlags{6-6} = IsStore;
+  bit IsTex = 0;
+  bit IsSust = 0;
+  bit IsSurfTexQuery = 0;
+  bit IsTexModeUnified = 0;
+
+  // The following field is encoded as log2 of the vector size minus one,
+  // with 0 meaning the operation is not a surface instruction.  For example,
+  // if IsSuld == 2, then the instruction is a suld instruction with vector size
+  // 2**(2-1) = 2.
+  bits<2> IsSuld = 0;
+
+  let TSFlags{3-0}   = VecInstType;
+  let TSFlags{4-4}   = IsSimpleMove;
+  let TSFlags{5-5}   = IsLoad;
+  let TSFlags{6-6}   = IsStore;
+  let TSFlags{7}     = IsTex;
+  let TSFlags{9-8}   = IsSuld;
+  let TSFlags{10}    = IsSust;
+  let TSFlags{11}    = IsSurfTexQuery;
+  let TSFlags{12}    = IsTexModeUnified;
 }
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.cpp
index 86ddd3817709..b5b4fbed0799 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.cpp
@@ -14,8 +14,6 @@
 #include "NVPTX.h"
 #include "NVPTXInstrInfo.h"
 #include "NVPTXTargetMachine.h"
-#define GET_INSTRINFO_CTOR_DTOR
-#include "NVPTXGenInstrInfo.inc"
 #include "llvm/IR/Function.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -24,12 +22,15 @@
 
 using namespace llvm;
 
+#define GET_INSTRINFO_CTOR_DTOR
+#include "NVPTXGenInstrInfo.inc"
+
 // Pin the vtable to this file.
 void NVPTXInstrInfo::anchor() {}
 
 // FIXME: Add the subtarget support on this constructor.
-NVPTXInstrInfo::NVPTXInstrInfo(NVPTXTargetMachine &tm)
-    : NVPTXGenInstrInfo(), TM(tm), RegInfo(*TM.getSubtargetImpl()) {}
+NVPTXInstrInfo::NVPTXInstrInfo(NVPTXSubtarget &STI)
+    : NVPTXGenInstrInfo(), RegInfo(STI) {}
 
 void NVPTXInstrInfo::copyPhysReg(
     MachineBasicBlock &MBB, MachineBasicBlock::iterator I, DebugLoc DL,
@@ -256,7 +257,7 @@ unsigned NVPTXInstrInfo::InsertBranch(
          "NVPTX branch conditions have two components!");
 
   // One-way branch.
-  if (FBB == 0) {
+  if (!FBB) {
     if (Cond.empty()) // Unconditional branch
       BuildMI(&MBB, DL, get(NVPTX::GOTO)).addMBB(TBB);
     else // Conditional branch
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.h b/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.h
index 600fc5c60a1b..2ac29748676a 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.h
@@ -24,13 +24,12 @@
 namespace llvm {
 
 class NVPTXInstrInfo : public NVPTXGenInstrInfo {
-  NVPTXTargetMachine &TM;
   const NVPTXRegisterInfo RegInfo;
   virtual void anchor();
 public:
-  explicit NVPTXInstrInfo(NVPTXTargetMachine &TM);
+  explicit NVPTXInstrInfo(NVPTXSubtarget &STI);
 
-  virtual const NVPTXRegisterInfo &getRegisterInfo() const { return RegInfo; }
+  const NVPTXRegisterInfo &getRegisterInfo() const { return RegInfo; }
 
   /* The following virtual functions are used in register allocation.
    * They are not implemented because the existing interface and the logic
@@ -50,9 +49,9 @@ public:
    *                               const TargetRegisterClass *RC) const;
    */
 
-  virtual void copyPhysReg(
+  void copyPhysReg(
       MachineBasicBlock &MBB, MachineBasicBlock::iterator I, DebugLoc DL,
-      unsigned DestReg, unsigned SrcReg, bool KillSrc) const;
+      unsigned DestReg, unsigned SrcReg, bool KillSrc) const override;
   virtual bool isMoveInstr(const MachineInstr &MI, unsigned &SrcReg,
                            unsigned &DestReg) const;
   bool isLoadInstr(const MachineInstr &MI, unsigned &AddrSpace) const;
@@ -61,13 +60,13 @@ public:
 
   virtual bool CanTailMerge(const MachineInstr *MI) const;
   // Branch analysis.
-  virtual bool AnalyzeBranch(
+  bool AnalyzeBranch(
       MachineBasicBlock &MBB, MachineBasicBlock *&TBB, MachineBasicBlock *&FBB,
-      SmallVectorImpl<MachineOperand> &Cond, bool AllowModify) const;
-  virtual unsigned RemoveBranch(MachineBasicBlock &MBB) const;
-  virtual unsigned InsertBranch(
+      SmallVectorImpl<MachineOperand> &Cond, bool AllowModify) const override;
+  unsigned RemoveBranch(MachineBasicBlock &MBB) const override;
+  unsigned InsertBranch(
       MachineBasicBlock &MBB, MachineBasicBlock *TBB, MachineBasicBlock *FBB,
-      const SmallVectorImpl<MachineOperand> &Cond, DebugLoc DL) const;
+      const SmallVectorImpl<MachineOperand> &Cond, DebugLoc DL) const override;
   unsigned getLdStCodeAddrSpace(const MachineInstr &MI) const {
     return MI.getOperand(2).getImm();
   }
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.td b/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.td
index b23f1e4b458d..9900b8c8433f 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.td
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXInstrInfo.td
@@ -139,17 +139,10 @@ def hasGenericLdSt : Predicate<"Subtarget.hasGenericLdSt()">;
 def doF32FTZ : Predicate<"useF32FTZ()">;
 def doNoF32FTZ : Predicate<"!useF32FTZ()">;
 
-def doFMAF32      : Predicate<"doFMAF32">;
-def doFMAF32_ftz  : Predicate<"(doFMAF32 && useF32FTZ())">;
-def doFMAF32AGG      : Predicate<"doFMAF32AGG">;
-def doFMAF32AGG_ftz  : Predicate<"(doFMAF32AGG && useF32FTZ())">;
-def doFMAF64      : Predicate<"doFMAF64">;
-def doFMAF64AGG      : Predicate<"doFMAF64AGG">;
-
 def doMulWide      : Predicate<"doMulWide">;
 
-def allowFMA : Predicate<"allowFMA">;
-def allowFMA_ftz : Predicate<"(allowFMA && useF32FTZ())">;
+def allowFMA : Predicate<"allowFMA()">;
+def noFMA : Predicate<"!allowFMA()">;
 
 def do_DIVF32_APPROX : Predicate<"getDivF32Level()==0">;
 def do_DIVF32_FULL : Predicate<"getDivF32Level()==1">;
@@ -158,9 +151,12 @@ def do_SQRTF32_APPROX : Predicate<"!usePrecSqrtF32()">;
 def do_SQRTF32_RN : Predicate<"usePrecSqrtF32()">;
 
 def hasHWROT32 : Predicate<"Subtarget.hasHWROT32()">;
+def noHWROT32 : Predicate<"!Subtarget.hasHWROT32()">;
 
 def true : Predicate<"1">;
 
+def hasPTX31 : Predicate<"Subtarget.getPTXVersion() >= 31">;
+
 
 //===----------------------------------------------------------------------===//
 // Some Common Instruction Class Templates
@@ -219,13 +215,13 @@ multiclass F3<string OpcStr, SDNode OpNode> {
                       !strconcat(OpcStr, ".ftz.f32 \t$dst, $a, $b;"),
                       [(set Float32Regs:$dst,
                         (OpNode Float32Regs:$a, Float32Regs:$b))]>,
-                      Requires<[allowFMA_ftz]>;
+                      Requires<[allowFMA, doF32FTZ]>;
    def f32ri_ftz : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, f32imm:$b),
                       !strconcat(OpcStr, ".ftz.f32 \t$dst, $a, $b;"),
                       [(set Float32Regs:$dst,
                         (OpNode Float32Regs:$a, fpimm:$b))]>,
-                      Requires<[allowFMA_ftz]>;
+                      Requires<[allowFMA, doF32FTZ]>;
    def f32rr : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, Float32Regs:$b),
                       !strconcat(OpcStr, ".f32 \t$dst, $a, $b;"),
@@ -245,34 +241,38 @@ multiclass F3_rn<string OpcStr, SDNode OpNode> {
                       (ins Float64Regs:$a, Float64Regs:$b),
                       !strconcat(OpcStr, ".rn.f64 \t$dst, $a, $b;"),
                       [(set Float64Regs:$dst,
-                        (OpNode Float64Regs:$a, Float64Regs:$b))]>;
+                        (OpNode Float64Regs:$a, Float64Regs:$b))]>,
+                      Requires<[noFMA]>;
    def f64ri : NVPTXInst<(outs Float64Regs:$dst),
                       (ins Float64Regs:$a, f64imm:$b),
                       !strconcat(OpcStr, ".rn.f64 \t$dst, $a, $b;"),
                       [(set Float64Regs:$dst,
-                        (OpNode Float64Regs:$a, fpimm:$b))]>;
+                        (OpNode Float64Regs:$a, fpimm:$b))]>,
+                      Requires<[noFMA]>;
    def f32rr_ftz : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, Float32Regs:$b),
                       !strconcat(OpcStr, ".rn.ftz.f32 \t$dst, $a, $b;"),
                       [(set Float32Regs:$dst,
                         (OpNode Float32Regs:$a, Float32Regs:$b))]>,
-                      Requires<[doF32FTZ]>;
+                      Requires<[noFMA, doF32FTZ]>;
    def f32ri_ftz : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, f32imm:$b),
                       !strconcat(OpcStr, ".rn.ftz.f32 \t$dst, $a, $b;"),
                       [(set Float32Regs:$dst,
                         (OpNode Float32Regs:$a, fpimm:$b))]>,
-                      Requires<[doF32FTZ]>;
+                      Requires<[noFMA, doF32FTZ]>;
    def f32rr : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, Float32Regs:$b),
                       !strconcat(OpcStr, ".rn.f32 \t$dst, $a, $b;"),
                       [(set Float32Regs:$dst,
-                        (OpNode Float32Regs:$a, Float32Regs:$b))]>;
+                        (OpNode Float32Regs:$a, Float32Regs:$b))]>,
+                      Requires<[noFMA]>;
    def f32ri : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, f32imm:$b),
                       !strconcat(OpcStr, ".rn.f32 \t$dst, $a, $b;"),
                       [(set Float32Regs:$dst,
-                        (OpNode Float32Regs:$a, fpimm:$b))]>;
+                        (OpNode Float32Regs:$a, fpimm:$b))]>,
+                      Requires<[noFMA]>;
 }
 
 multiclass F2<string OpcStr, SDNode OpNode> {
@@ -461,33 +461,45 @@ def SHL2MUL16 : SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(temp.shl(v), MVT::i16);
 }]>;
 
-def MULWIDES64 : NVPTXInst<(outs Int64Regs:$dst),
-                           (ins Int32Regs:$a, Int32Regs:$b),
+def MULWIDES64
+  : NVPTXInst<(outs Int64Regs:$dst), (ins Int32Regs:$a, Int32Regs:$b),
+              "mul.wide.s32 \t$dst, $a, $b;", []>;
+def MULWIDES64Imm
+  : NVPTXInst<(outs Int64Regs:$dst), (ins Int32Regs:$a, i32imm:$b),
                            "mul.wide.s32 \t$dst, $a, $b;", []>;
-def MULWIDES64Imm : NVPTXInst<(outs Int64Regs:$dst),
-                            (ins Int32Regs:$a, i64imm:$b),
+def MULWIDES64Imm64
+  : NVPTXInst<(outs Int64Regs:$dst), (ins Int32Regs:$a, i64imm:$b),
                            "mul.wide.s32 \t$dst, $a, $b;", []>;
 
-def MULWIDEU64 : NVPTXInst<(outs Int64Regs:$dst),
-                           (ins Int32Regs:$a, Int32Regs:$b),
+def MULWIDEU64
+  : NVPTXInst<(outs Int64Regs:$dst), (ins Int32Regs:$a, Int32Regs:$b),
+              "mul.wide.u32 \t$dst, $a, $b;", []>;
+def MULWIDEU64Imm
+  : NVPTXInst<(outs Int64Regs:$dst), (ins Int32Regs:$a, i32imm:$b),
                            "mul.wide.u32 \t$dst, $a, $b;", []>;
-def MULWIDEU64Imm : NVPTXInst<(outs Int64Regs:$dst),
-                            (ins Int32Regs:$a, i64imm:$b),
+def MULWIDEU64Imm64
+  : NVPTXInst<(outs Int64Regs:$dst), (ins Int32Regs:$a, i64imm:$b),
                            "mul.wide.u32 \t$dst, $a, $b;", []>;
 
-def MULWIDES32 : NVPTXInst<(outs Int32Regs:$dst),
-                            (ins Int16Regs:$a, Int16Regs:$b),
+def MULWIDES32
+  : NVPTXInst<(outs Int32Regs:$dst), (ins Int16Regs:$a, Int16Regs:$b),
                            "mul.wide.s16 \t$dst, $a, $b;", []>;
-def MULWIDES32Imm : NVPTXInst<(outs Int32Regs:$dst),
-                            (ins Int16Regs:$a, i32imm:$b),
+def MULWIDES32Imm
+  : NVPTXInst<(outs Int32Regs:$dst), (ins Int16Regs:$a, i16imm:$b),
+              "mul.wide.s16 \t$dst, $a, $b;", []>;
+def MULWIDES32Imm32
+  : NVPTXInst<(outs Int32Regs:$dst), (ins Int16Regs:$a, i32imm:$b),
                            "mul.wide.s16 \t$dst, $a, $b;", []>;
 
-def MULWIDEU32 : NVPTXInst<(outs Int32Regs:$dst),
-                            (ins Int16Regs:$a, Int16Regs:$b),
-                           "mul.wide.u16 \t$dst, $a, $b;", []>;
-def MULWIDEU32Imm : NVPTXInst<(outs Int32Regs:$dst),
-                            (ins Int16Regs:$a, i32imm:$b),
+def MULWIDEU32
+  : NVPTXInst<(outs Int32Regs:$dst), (ins Int16Regs:$a, Int16Regs:$b),
+              "mul.wide.u16 \t$dst, $a, $b;", []>;
+def MULWIDEU32Imm
+  : NVPTXInst<(outs Int32Regs:$dst), (ins Int16Regs:$a, i16imm:$b),
                            "mul.wide.u16 \t$dst, $a, $b;", []>;
+def MULWIDEU32Imm32
+  : NVPTXInst<(outs Int32Regs:$dst), (ins Int16Regs:$a, i32imm:$b),
+                            "mul.wide.u16 \t$dst, $a, $b;", []>;
 
 def : Pat<(shl (sext Int32Regs:$a), (i32 Int5Const:$b)),
           (MULWIDES64Imm Int32Regs:$a, (SHL2MUL32 node:$b))>,
@@ -507,25 +519,63 @@ def : Pat<(mul (sext Int32Regs:$a), (sext Int32Regs:$b)),
           (MULWIDES64 Int32Regs:$a, Int32Regs:$b)>,
           Requires<[doMulWide]>;
 def : Pat<(mul (sext Int32Regs:$a), (i64 SInt32Const:$b)),
-          (MULWIDES64Imm Int32Regs:$a, (i64 SInt32Const:$b))>,
+          (MULWIDES64Imm64 Int32Regs:$a, (i64 SInt32Const:$b))>,
           Requires<[doMulWide]>;
 
 def : Pat<(mul (zext Int32Regs:$a), (zext Int32Regs:$b)),
-          (MULWIDEU64 Int32Regs:$a, Int32Regs:$b)>, Requires<[doMulWide]>;
+          (MULWIDEU64 Int32Regs:$a, Int32Regs:$b)>,
+      Requires<[doMulWide]>;
 def : Pat<(mul (zext Int32Regs:$a), (i64 UInt32Const:$b)),
-          (MULWIDEU64Imm Int32Regs:$a, (i64 UInt32Const:$b))>,
+          (MULWIDEU64Imm64 Int32Regs:$a, (i64 UInt32Const:$b))>,
           Requires<[doMulWide]>;
 
 def : Pat<(mul (sext Int16Regs:$a), (sext Int16Regs:$b)),
-          (MULWIDES32 Int16Regs:$a, Int16Regs:$b)>, Requires<[doMulWide]>;
+          (MULWIDES32 Int16Regs:$a, Int16Regs:$b)>,
+      Requires<[doMulWide]>;
 def : Pat<(mul (sext Int16Regs:$a), (i32 SInt16Const:$b)),
-          (MULWIDES32Imm Int16Regs:$a, (i32 SInt16Const:$b))>,
+          (MULWIDES32Imm32 Int16Regs:$a, (i32 SInt16Const:$b))>,
           Requires<[doMulWide]>;
 
 def : Pat<(mul (zext Int16Regs:$a), (zext Int16Regs:$b)),
-          (MULWIDEU32 Int16Regs:$a, Int16Regs:$b)>, Requires<[doMulWide]>;
+          (MULWIDEU32 Int16Regs:$a, Int16Regs:$b)>,
+      Requires<[doMulWide]>;
 def : Pat<(mul (zext Int16Regs:$a), (i32 UInt16Const:$b)),
-          (MULWIDEU32Imm Int16Regs:$a, (i32 UInt16Const:$b))>,
+          (MULWIDEU32Imm32 Int16Regs:$a, (i32 UInt16Const:$b))>,
+          Requires<[doMulWide]>;
+
+
+def SDTMulWide
+  : SDTypeProfile<1, 2, [SDTCisSameAs<1, 2>]>;
+def mul_wide_signed
+  : SDNode<"NVPTXISD::MUL_WIDE_SIGNED", SDTMulWide>;
+def mul_wide_unsigned
+  : SDNode<"NVPTXISD::MUL_WIDE_UNSIGNED", SDTMulWide>;
+
+def : Pat<(i32 (mul_wide_signed Int16Regs:$a, Int16Regs:$b)),
+          (MULWIDES32 Int16Regs:$a, Int16Regs:$b)>,
+      Requires<[doMulWide]>;
+def : Pat<(i32 (mul_wide_signed Int16Regs:$a, imm:$b)),
+          (MULWIDES32Imm Int16Regs:$a, imm:$b)>,
+          Requires<[doMulWide]>;
+def : Pat<(i32 (mul_wide_unsigned Int16Regs:$a, Int16Regs:$b)),
+          (MULWIDEU32 Int16Regs:$a, Int16Regs:$b)>,
+          Requires<[doMulWide]>;
+def : Pat<(i32 (mul_wide_unsigned Int16Regs:$a, imm:$b)),
+          (MULWIDEU32Imm Int16Regs:$a, imm:$b)>,
+          Requires<[doMulWide]>;
+
+
+def : Pat<(i64 (mul_wide_signed Int32Regs:$a, Int32Regs:$b)),
+          (MULWIDES64 Int32Regs:$a, Int32Regs:$b)>,
+          Requires<[doMulWide]>;
+def : Pat<(i64 (mul_wide_signed Int32Regs:$a, imm:$b)),
+          (MULWIDES64Imm Int32Regs:$a, imm:$b)>,
+          Requires<[doMulWide]>;
+def : Pat<(i64 (mul_wide_unsigned Int32Regs:$a, Int32Regs:$b)),
+          (MULWIDEU64 Int32Regs:$a, Int32Regs:$b)>,
+          Requires<[doMulWide]>;
+def : Pat<(i64 (mul_wide_unsigned Int32Regs:$a, imm:$b)),
+          (MULWIDEU64Imm Int32Regs:$a, imm:$b)>,
           Requires<[doMulWide]>;
 
 defm MULT : I3<"mul.lo.s", mul>;
@@ -541,69 +591,75 @@ defm SREM : I3<"rem.s", srem>;
 defm UREM : I3<"rem.u", urem>;
 // The ri version will not be selected as DAGCombiner::visitUREM will lower it.
 
+def SDTIMAD
+  : SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>, SDTCisInt<0>,
+                         SDTCisInt<2>, SDTCisSameAs<0, 2>,
+                         SDTCisSameAs<0, 3>]>;
+def imad
+  : SDNode<"NVPTXISD::IMAD", SDTIMAD>;
+
 def MAD16rrr : NVPTXInst<(outs Int16Regs:$dst),
                       (ins Int16Regs:$a, Int16Regs:$b, Int16Regs:$c),
                       "mad.lo.s16 \t$dst, $a, $b, $c;",
-                      [(set Int16Regs:$dst, (add
-                        (mul Int16Regs:$a, Int16Regs:$b), Int16Regs:$c))]>;
+                      [(set Int16Regs:$dst,
+                         (imad Int16Regs:$a, Int16Regs:$b, Int16Regs:$c))]>;
 def MAD16rri : NVPTXInst<(outs Int16Regs:$dst),
                       (ins Int16Regs:$a, Int16Regs:$b, i16imm:$c),
                       "mad.lo.s16 \t$dst, $a, $b, $c;",
-                      [(set Int16Regs:$dst, (add
-                        (mul Int16Regs:$a, Int16Regs:$b), imm:$c))]>;
+                      [(set Int16Regs:$dst,
+                         (imad Int16Regs:$a, Int16Regs:$b, imm:$c))]>;
 def MAD16rir : NVPTXInst<(outs Int16Regs:$dst),
                       (ins Int16Regs:$a, i16imm:$b, Int16Regs:$c),
                       "mad.lo.s16 \t$dst, $a, $b, $c;",
-                      [(set Int16Regs:$dst, (add
-                        (mul Int16Regs:$a, imm:$b), Int16Regs:$c))]>;
+                      [(set Int16Regs:$dst,
+                        (imad Int16Regs:$a, imm:$b, Int16Regs:$c))]>;
 def MAD16rii : NVPTXInst<(outs Int16Regs:$dst),
     (ins Int16Regs:$a, i16imm:$b, i16imm:$c),
                       "mad.lo.s16 \t$dst, $a, $b, $c;",
-                      [(set Int16Regs:$dst, (add (mul Int16Regs:$a, imm:$b),
-                        imm:$c))]>;
+                      [(set Int16Regs:$dst,
+                        (imad Int16Regs:$a, imm:$b, imm:$c))]>;
 
 def MAD32rrr : NVPTXInst<(outs Int32Regs:$dst),
                       (ins Int32Regs:$a, Int32Regs:$b, Int32Regs:$c),
                       "mad.lo.s32 \t$dst, $a, $b, $c;",
-                      [(set Int32Regs:$dst, (add
-                        (mul Int32Regs:$a, Int32Regs:$b), Int32Regs:$c))]>;
+                      [(set Int32Regs:$dst,
+                        (imad Int32Regs:$a, Int32Regs:$b, Int32Regs:$c))]>;
 def MAD32rri : NVPTXInst<(outs Int32Regs:$dst),
                       (ins Int32Regs:$a, Int32Regs:$b, i32imm:$c),
                       "mad.lo.s32 \t$dst, $a, $b, $c;",
-                      [(set Int32Regs:$dst, (add
-                        (mul Int32Regs:$a, Int32Regs:$b), imm:$c))]>;
+                      [(set Int32Regs:$dst,
+                        (imad Int32Regs:$a, Int32Regs:$b, imm:$c))]>;
 def MAD32rir : NVPTXInst<(outs Int32Regs:$dst),
                       (ins Int32Regs:$a, i32imm:$b, Int32Regs:$c),
                       "mad.lo.s32 \t$dst, $a, $b, $c;",
-                      [(set Int32Regs:$dst, (add
-                        (mul Int32Regs:$a, imm:$b), Int32Regs:$c))]>;
+                      [(set Int32Regs:$dst,
+                        (imad Int32Regs:$a, imm:$b, Int32Regs:$c))]>;
 def MAD32rii : NVPTXInst<(outs Int32Regs:$dst),
                       (ins Int32Regs:$a, i32imm:$b, i32imm:$c),
                       "mad.lo.s32 \t$dst, $a, $b, $c;",
-                      [(set Int32Regs:$dst, (add
-                        (mul Int32Regs:$a, imm:$b), imm:$c))]>;
+                      [(set Int32Regs:$dst,
+                        (imad Int32Regs:$a, imm:$b, imm:$c))]>;
 
 def MAD64rrr : NVPTXInst<(outs Int64Regs:$dst),
                       (ins Int64Regs:$a, Int64Regs:$b, Int64Regs:$c),
                       "mad.lo.s64 \t$dst, $a, $b, $c;",
-                      [(set Int64Regs:$dst, (add
-                        (mul Int64Regs:$a, Int64Regs:$b), Int64Regs:$c))]>;
+                      [(set Int64Regs:$dst,
+                        (imad Int64Regs:$a, Int64Regs:$b, Int64Regs:$c))]>;
 def MAD64rri : NVPTXInst<(outs Int64Regs:$dst),
                       (ins Int64Regs:$a, Int64Regs:$b, i64imm:$c),
                       "mad.lo.s64 \t$dst, $a, $b, $c;",
-                      [(set Int64Regs:$dst, (add
-                        (mul Int64Regs:$a, Int64Regs:$b), imm:$c))]>;
+                      [(set Int64Regs:$dst,
+                        (imad Int64Regs:$a, Int64Regs:$b, imm:$c))]>;
 def MAD64rir : NVPTXInst<(outs Int64Regs:$dst),
                       (ins Int64Regs:$a, i64imm:$b, Int64Regs:$c),
                       "mad.lo.s64 \t$dst, $a, $b, $c;",
-                      [(set Int64Regs:$dst, (add
-                        (mul Int64Regs:$a, imm:$b), Int64Regs:$c))]>;
+                      [(set Int64Regs:$dst,
+                        (imad Int64Regs:$a, imm:$b, Int64Regs:$c))]>;
 def MAD64rii : NVPTXInst<(outs Int64Regs:$dst),
                       (ins Int64Regs:$a, i64imm:$b, i64imm:$c),
                       "mad.lo.s64 \t$dst, $a, $b, $c;",
-                      [(set Int64Regs:$dst, (add
-                        (mul Int64Regs:$a, imm:$b), imm:$c))]>;
-
+                      [(set Int64Regs:$dst,
+                        (imad Int64Regs:$a, imm:$b, imm:$c))]>;
 
 def INEG16 : NVPTXInst<(outs Int16Regs:$dst), (ins Int16Regs:$src),
                      "neg.s16 \t$dst, $src;",
@@ -689,12 +745,24 @@ def FDIV32approxrr_ftz : NVPTXInst<(outs Float32Regs:$dst),
                       [(set Float32Regs:$dst,
                         (fdiv Float32Regs:$a, Float32Regs:$b))]>,
                       Requires<[do_DIVF32_APPROX, doF32FTZ]>;
+def FDIV32approxri_ftz : NVPTXInst<(outs Float32Regs:$dst),
+                      (ins Float32Regs:$a, f32imm:$b),
+                      "div.approx.ftz.f32 \t$dst, $a, $b;",
+                      [(set Float32Regs:$dst,
+                        (fdiv Float32Regs:$a, fpimm:$b))]>,
+                      Requires<[do_DIVF32_APPROX, doF32FTZ]>;
 def FDIV32approxrr     : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, Float32Regs:$b),
                       "div.approx.f32 \t$dst, $a, $b;",
                       [(set Float32Regs:$dst,
                         (fdiv Float32Regs:$a, Float32Regs:$b))]>,
                       Requires<[do_DIVF32_APPROX]>;
+def FDIV32approxri : NVPTXInst<(outs Float32Regs:$dst),
+                      (ins Float32Regs:$a, f32imm:$b),
+                      "div.approx.f32 \t$dst, $a, $b;",
+                      [(set Float32Regs:$dst,
+                        (fdiv Float32Regs:$a, fpimm:$b))]>,
+                      Requires<[do_DIVF32_APPROX]>;
 //
 // F32 Semi-accurate reciprocal
 //
@@ -797,36 +865,26 @@ multiclass FPCONTRACT32<string OpcStr, Predicate Pred> {
    def rrr : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, Float32Regs:$b, Float32Regs:$c),
                       !strconcat(OpcStr, " \t$dst, $a, $b, $c;"),
-                      [(set Float32Regs:$dst, (fadd
-                        (fmul Float32Regs:$a, Float32Regs:$b),
-                        Float32Regs:$c))]>, Requires<[Pred]>;
-   // This is to WAR a weird bug in Tablegen that does not automatically
-   // generate the following permutated rule rrr2 from the above rrr.
-   // So we explicitly add it here. This happens to FMA32 only.
-   // See the comments at FMAD32 and FMA32 for more information.
-   def rrr2 : NVPTXInst<(outs Float32Regs:$dst),
-                        (ins Float32Regs:$a, Float32Regs:$b, Float32Regs:$c),
-                      !strconcat(OpcStr, " \t$dst, $a, $b, $c;"),
-                      [(set Float32Regs:$dst, (fadd Float32Regs:$c,
-                        (fmul Float32Regs:$a, Float32Regs:$b)))]>,
+                      [(set Float32Regs:$dst,
+                        (fma Float32Regs:$a, Float32Regs:$b, Float32Regs:$c))]>,
                       Requires<[Pred]>;
    def rri : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, Float32Regs:$b, f32imm:$c),
                       !strconcat(OpcStr, " \t$dst, $a, $b, $c;"),
-                      [(set Float32Regs:$dst, (fadd
-                        (fmul Float32Regs:$a, Float32Regs:$b), fpimm:$c))]>,
+                      [(set Float32Regs:$dst,
+                        (fma Float32Regs:$a, Float32Regs:$b, fpimm:$c))]>,
                       Requires<[Pred]>;
    def rir : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, f32imm:$b, Float32Regs:$c),
                       !strconcat(OpcStr, " \t$dst, $a, $b, $c;"),
-                      [(set Float32Regs:$dst, (fadd
-                        (fmul Float32Regs:$a, fpimm:$b), Float32Regs:$c))]>,
+                      [(set Float32Regs:$dst,
+                        (fma Float32Regs:$a, fpimm:$b, Float32Regs:$c))]>,
                       Requires<[Pred]>;
    def rii : NVPTXInst<(outs Float32Regs:$dst),
                       (ins Float32Regs:$a, f32imm:$b, f32imm:$c),
                       !strconcat(OpcStr, " \t$dst, $a, $b, $c;"),
-                      [(set Float32Regs:$dst, (fadd
-                        (fmul Float32Regs:$a, fpimm:$b), fpimm:$c))]>,
+                      [(set Float32Regs:$dst,
+                        (fma Float32Regs:$a, fpimm:$b, fpimm:$c))]>,
                       Requires<[Pred]>;
 }
 
@@ -834,73 +892,32 @@ multiclass FPCONTRACT64<string OpcStr, Predicate Pred> {
    def rrr : NVPTXInst<(outs Float64Regs:$dst),
                       (ins Float64Regs:$a, Float64Regs:$b, Float64Regs:$c),
                       !strconcat(OpcStr, " \t$dst, $a, $b, $c;"),
-                      [(set Float64Regs:$dst, (fadd
-                        (fmul Float64Regs:$a, Float64Regs:$b),
-                        Float64Regs:$c))]>, Requires<[Pred]>;
+                      [(set Float64Regs:$dst,
+                        (fma Float64Regs:$a, Float64Regs:$b, Float64Regs:$c))]>,
+                      Requires<[Pred]>;
    def rri : NVPTXInst<(outs Float64Regs:$dst),
                       (ins Float64Regs:$a, Float64Regs:$b, f64imm:$c),
                       !strconcat(OpcStr, " \t$dst, $a, $b, $c;"),
-                      [(set Float64Regs:$dst, (fadd (fmul Float64Regs:$a,
-                        Float64Regs:$b), fpimm:$c))]>, Requires<[Pred]>;
+                      [(set Float64Regs:$dst,
+                        (fma Float64Regs:$a, Float64Regs:$b, fpimm:$c))]>,
+                      Requires<[Pred]>;
    def rir : NVPTXInst<(outs Float64Regs:$dst),
                       (ins Float64Regs:$a, f64imm:$b, Float64Regs:$c),
                       !strconcat(OpcStr, " \t$dst, $a, $b, $c;"),
-                      [(set Float64Regs:$dst, (fadd
-                        (fmul Float64Regs:$a, fpimm:$b), Float64Regs:$c))]>,
+                      [(set Float64Regs:$dst,
+                        (fma Float64Regs:$a, fpimm:$b, Float64Regs:$c))]>,
                       Requires<[Pred]>;
    def rii : NVPTXInst<(outs Float64Regs:$dst),
                       (ins Float64Regs:$a, f64imm:$b, f64imm:$c),
                       !strconcat(OpcStr, " \t$dst, $a, $b, $c;"),
-                      [(set Float64Regs:$dst, (fadd
-                        (fmul Float64Regs:$a, fpimm:$b), fpimm:$c))]>,
+                      [(set Float64Regs:$dst,
+                        (fma Float64Regs:$a, fpimm:$b, fpimm:$c))]>,
                       Requires<[Pred]>;
 }
 
-// Due to a unknown reason (most likely a bug in tablegen), tablegen does not
-// automatically generate the rrr2 rule from
-// the rrr rule (see FPCONTRACT32) for FMA32, though it does for FMAD32.
-// If we reverse the order of the following two lines, then rrr2 rule will be
-// generated for FMA32, but not for rrr.
-// Therefore, we manually write the rrr2 rule in FPCONTRACT32.
-defm FMA32_ftz  : FPCONTRACT32<"fma.rn.ftz.f32", doFMAF32_ftz>;
-defm FMA32  : FPCONTRACT32<"fma.rn.f32", doFMAF32>;
-defm FMA64  : FPCONTRACT64<"fma.rn.f64", doFMAF64>;
-
-// b*c-a => fmad(b, c, -a)
-multiclass FPCONTRACT32_SUB_PAT_MAD<NVPTXInst Inst, Predicate Pred> {
-  def : Pat<(fsub (fmul Float32Regs:$b, Float32Regs:$c), Float32Regs:$a),
-          (Inst Float32Regs:$b, Float32Regs:$c, (FNEGf32 Float32Regs:$a))>,
-          Requires<[Pred]>;
-}
-
-// a-b*c => fmad(-b,c, a)
-// - legal because a-b*c <=> a+(-b*c) <=> a+(-b)*c
-// b*c-a => fmad(b, c, -a)
-// - legal because b*c-a <=> b*c+(-a)
-multiclass FPCONTRACT32_SUB_PAT<NVPTXInst Inst, Predicate Pred> {
-  def : Pat<(fsub Float32Regs:$a, (fmul Float32Regs:$b, Float32Regs:$c)),
-          (Inst (FNEGf32 Float32Regs:$b), Float32Regs:$c, Float32Regs:$a)>,
-          Requires<[Pred]>;
-  def : Pat<(fsub (fmul Float32Regs:$b, Float32Regs:$c), Float32Regs:$a),
-          (Inst Float32Regs:$b, Float32Regs:$c, (FNEGf32 Float32Regs:$a))>,
-          Requires<[Pred]>;
-}
-
-// a-b*c => fmad(-b,c, a)
-// b*c-a => fmad(b, c, -a)
-multiclass FPCONTRACT64_SUB_PAT<NVPTXInst Inst, Predicate Pred> {
-  def : Pat<(fsub Float64Regs:$a, (fmul Float64Regs:$b, Float64Regs:$c)),
-          (Inst (FNEGf64 Float64Regs:$b), Float64Regs:$c, Float64Regs:$a)>,
-          Requires<[Pred]>;
-
-  def : Pat<(fsub (fmul Float64Regs:$b, Float64Regs:$c), Float64Regs:$a),
-          (Inst Float64Regs:$b, Float64Regs:$c, (FNEGf64 Float64Regs:$a))>,
-          Requires<[Pred]>;
-}
-
-defm FMAF32ext_ftz  : FPCONTRACT32_SUB_PAT<FMA32_ftzrrr, doFMAF32AGG_ftz>;
-defm FMAF32ext  : FPCONTRACT32_SUB_PAT<FMA32rrr, doFMAF32AGG>;
-defm FMAF64ext  : FPCONTRACT64_SUB_PAT<FMA64rrr, doFMAF64AGG>;
+defm FMA32_ftz  : FPCONTRACT32<"fma.rn.ftz.f32", doF32FTZ>;
+defm FMA32  : FPCONTRACT32<"fma.rn.f32", true>;
+defm FMA64  : FPCONTRACT64<"fma.rn.f64", true>;
 
 def SINF:  NVPTXInst<(outs Float32Regs:$dst), (ins Float32Regs:$src),
                       "sin.approx.f32 \t$dst, $src;",
@@ -1071,6 +1088,43 @@ multiclass RSHIFT_FORMAT<string OpcStr, SDNode OpNode> {
 defm SRA : RSHIFT_FORMAT<"shr.s", sra>;
 defm SRL : RSHIFT_FORMAT<"shr.u", srl>;
 
+//
+// Rotate: use ptx shf instruction if available.
+//
+
+// 32 bit r2 = rotl r1, n
+//    =>
+//        r2 = shf.l r1, r1, n
+def ROTL32imm_hw : NVPTXInst<(outs Int32Regs:$dst),
+                             (ins Int32Regs:$src, i32imm:$amt),
+              "shf.l.wrap.b32 \t$dst, $src, $src, $amt;",
+    [(set Int32Regs:$dst, (rotl Int32Regs:$src, (i32 imm:$amt)))]>,
+    Requires<[hasHWROT32]> ;
+
+def ROTL32reg_hw : NVPTXInst<(outs Int32Regs:$dst),
+                             (ins Int32Regs:$src, Int32Regs:$amt),
+              "shf.l.wrap.b32 \t$dst, $src, $src, $amt;",
+    [(set Int32Regs:$dst, (rotl Int32Regs:$src, Int32Regs:$amt))]>,
+    Requires<[hasHWROT32]>;
+
+// 32 bit r2 = rotr r1, n
+//    =>
+//        r2 = shf.r r1, r1, n
+def ROTR32imm_hw : NVPTXInst<(outs Int32Regs:$dst),
+                             (ins Int32Regs:$src, i32imm:$amt),
+              "shf.r.wrap.b32 \t$dst, $src, $src, $amt;",
+    [(set Int32Regs:$dst, (rotr Int32Regs:$src, (i32 imm:$amt)))]>,
+    Requires<[hasHWROT32]>;
+
+def ROTR32reg_hw : NVPTXInst<(outs Int32Regs:$dst),
+                             (ins Int32Regs:$src, Int32Regs:$amt),
+              "shf.r.wrap.b32 \t$dst, $src, $src, $amt;",
+    [(set Int32Regs:$dst, (rotr Int32Regs:$src, Int32Regs:$amt))]>,
+    Requires<[hasHWROT32]>;
+
+//
+// Rotate: if ptx shf instruction is not available, then use shift+add
+//
 // 32bit
 def ROT32imm_sw : NVPTXInst<(outs Int32Regs:$dst),
   (ins Int32Regs:$src, i32imm:$amt1, i32imm:$amt2),
@@ -1088,9 +1142,11 @@ def SUB_FRM_32 : SDNodeXForm<imm, [{
 }]>;
 
 def : Pat<(rotl Int32Regs:$src, (i32 imm:$amt)),
-          (ROT32imm_sw Int32Regs:$src, imm:$amt, (SUB_FRM_32 node:$amt))>;
+          (ROT32imm_sw Int32Regs:$src, imm:$amt, (SUB_FRM_32 node:$amt))>,
+      Requires<[noHWROT32]>;
 def : Pat<(rotr Int32Regs:$src, (i32 imm:$amt)),
-          (ROT32imm_sw Int32Regs:$src, (SUB_FRM_32 node:$amt), imm:$amt)>;
+          (ROT32imm_sw Int32Regs:$src, (SUB_FRM_32 node:$amt), imm:$amt)>,
+      Requires<[noHWROT32]>;
 
 def ROTL32reg_sw : NVPTXInst<(outs Int32Regs:$dst), (ins Int32Regs:$src,
     Int32Regs:$amt),
@@ -1103,7 +1159,8 @@ def ROTL32reg_sw : NVPTXInst<(outs Int32Regs:$dst), (ins Int32Regs:$src,
     !strconcat("shr.b32 \t%rhs, $src, %amt2;\n\t",
     !strconcat("add.u32 \t$dst, %lhs, %rhs;\n\t",
     !strconcat("}}", ""))))))))),
-    [(set Int32Regs:$dst, (rotl Int32Regs:$src, Int32Regs:$amt))]>;
+    [(set Int32Regs:$dst, (rotl Int32Regs:$src, Int32Regs:$amt))]>,
+    Requires<[noHWROT32]>;
 
 def ROTR32reg_sw : NVPTXInst<(outs Int32Regs:$dst), (ins Int32Regs:$src,
     Int32Regs:$amt),
@@ -1116,7 +1173,8 @@ def ROTR32reg_sw : NVPTXInst<(outs Int32Regs:$dst), (ins Int32Regs:$src,
     !strconcat("shl.b32 \t%rhs, $src, %amt2;\n\t",
     !strconcat("add.u32 \t$dst, %lhs, %rhs;\n\t",
     !strconcat("}}", ""))))))))),
-    [(set Int32Regs:$dst, (rotr Int32Regs:$src, Int32Regs:$amt))]>;
+    [(set Int32Regs:$dst, (rotr Int32Regs:$src, Int32Regs:$amt))]>,
+    Requires<[noHWROT32]>;
 
 // 64bit
 def ROT64imm_sw : NVPTXInst<(outs Int64Regs:$dst), (ins Int64Regs:$src,
@@ -1165,6 +1223,29 @@ def ROTR64reg_sw : NVPTXInst<(outs Int64Regs:$dst), (ins Int64Regs:$src,
     !strconcat("}}", ""))))))))),
     [(set Int64Regs:$dst, (rotr Int64Regs:$src, Int32Regs:$amt))]>;
 
+// BFE - bit-field extract
+
+multiclass BFE<string TyStr, RegisterClass RC> {
+  // BFE supports both 32-bit and 64-bit values, but the start and length
+  // operands are always 32-bit
+  def rrr
+    : NVPTXInst<(outs RC:$d),
+                (ins RC:$a, Int32Regs:$b, Int32Regs:$c),
+                !strconcat("bfe.", TyStr, " \t$d, $a, $b, $c;"), []>;
+  def rri
+    : NVPTXInst<(outs RC:$d),
+                (ins RC:$a, Int32Regs:$b, i32imm:$c),
+                !strconcat("bfe.", TyStr, " \t$d, $a, $b, $c;"), []>;
+  def rii
+    : NVPTXInst<(outs RC:$d),
+                (ins RC:$a, i32imm:$b, i32imm:$c),
+                !strconcat("bfe.", TyStr, " \t$d, $a, $b, $c;"), []>;
+}
+
+defm BFE_S32 : BFE<"s32", Int32Regs>;
+defm BFE_U32 : BFE<"u32", Int32Regs>;
+defm BFE_S64 : BFE<"s64", Int64Regs>;
+defm BFE_U64 : BFE<"u64", Int64Regs>;
 
 //-----------------------------------
 // General Comparison
@@ -1280,6 +1361,32 @@ def : Pat<(i1 (select Int1Regs:$p, Int1Regs:$a, Int1Regs:$b)),
               (ORb1rr (ANDb1rr Int1Regs:$p, Int1Regs:$a),
               (ANDb1rr (NOT1 Int1Regs:$p), Int1Regs:$b))>;
 
+//
+// Funnnel shift in clamp mode
+//
+// - SDNodes are created so they can be used in the DAG code,
+//   e.g. NVPTXISelLowering (LowerShiftLeftParts and LowerShiftRightParts)
+//
+def SDTIntShiftDOp: SDTypeProfile<1, 3,
+                                  [SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>,
+                                   SDTCisInt<0>, SDTCisInt<3>]>;
+def FUN_SHFL_CLAMP : SDNode<"NVPTXISD::FUN_SHFL_CLAMP", SDTIntShiftDOp, []>;
+def FUN_SHFR_CLAMP : SDNode<"NVPTXISD::FUN_SHFR_CLAMP", SDTIntShiftDOp, []>;
+
+def FUNSHFLCLAMP : NVPTXInst<(outs Int32Regs:$dst),
+                             (ins Int32Regs:$lo, Int32Regs:$hi, Int32Regs:$amt),
+                  "shf.l.clamp.b32 \t$dst, $lo, $hi, $amt;",
+                  [(set Int32Regs:$dst,
+                     (FUN_SHFL_CLAMP Int32Regs:$lo,
+                        Int32Regs:$hi, Int32Regs:$amt))]>;
+
+def FUNSHFRCLAMP : NVPTXInst<(outs Int32Regs:$dst),
+                             (ins Int32Regs:$lo, Int32Regs:$hi, Int32Regs:$amt),
+                  "shf.r.clamp.b32 \t$dst, $lo, $hi, $amt;",
+                  [(set Int32Regs:$dst,
+                     (FUN_SHFR_CLAMP Int32Regs:$lo,
+                        Int32Regs:$hi, Int32Regs:$amt))]>;
+
 //-----------------------------------
 // Data Movement (Load / Store, Move)
 //-----------------------------------
@@ -1807,7 +1914,7 @@ def StoreParamV2I8   : StoreParamV2Inst<Int16Regs, ".b8">;
 def StoreParamV4I32    : NVPTXInst<(outs), (ins Int32Regs:$val, Int32Regs:$val2,
                                                Int32Regs:$val3, Int32Regs:$val4,
                                                 i32imm:$a, i32imm:$b),
-                   "st.param.b32\t[param$a+$b], {{$val, $val2, $val3, $val4}};",
+                "st.param.v4.b32\t[param$a+$b], {{$val, $val2, $val3, $val4}};",
                          []>;
 
 def StoreParamV4I16    : NVPTXInst<(outs), (ins Int16Regs:$val, Int16Regs:$val2,
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td b/contrib/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
index 14049b1cf8eb..14e51aa309ea 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
@@ -792,13 +792,18 @@ def INT_NVVM_H2F : F_MATH_1<!strconcat("{{\n\t",
             "}}")))),
           Float32Regs, Int16Regs, int_nvvm_h2f>;
 
-def : Pat<(f32 (f16_to_f32 Int16Regs:$a)),
+def : Pat<(f32 (f16_to_fp Int16Regs:$a)),
           (CVT_f32_f16 Int16Regs:$a, CvtNONE)>;
-def : Pat<(i16 (f32_to_f16 Float32Regs:$a)),
+def : Pat<(i16 (fp_to_f16 Float32Regs:$a)),
           (CVT_f16_f32 Float32Regs:$a, CvtRN_FTZ)>, Requires<[doF32FTZ]>;
-def : Pat<(i16 (f32_to_f16 Float32Regs:$a)),
+def : Pat<(i16 (fp_to_f16 Float32Regs:$a)),
           (CVT_f16_f32 Float32Regs:$a, CvtRN)>;
 
+def : Pat<(f64 (f16_to_fp Int16Regs:$a)),
+          (CVT_f64_f16 Int16Regs:$a, CvtNONE)>;
+def : Pat<(i16 (fp_to_f16 Float64Regs:$a)),
+          (CVT_f16_f64 Float64Regs:$a, CvtRN)>;
+
 //
 // Bitcast
 //
@@ -1057,12 +1062,24 @@ def atomic_load_max_32_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
   (atomic_load_max_32 node:$a, node:$b)>;
 def atomic_load_max_32_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
   (atomic_load_max_32 node:$a, node:$b)>;
+def atomic_load_max_64_g: ATOMIC_GLOBAL_CHK<(ops node:$a, node:$b)
+  , (atomic_load_max_64 node:$a, node:$b)>;
+def atomic_load_max_64_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
+  (atomic_load_max_64 node:$a, node:$b)>;
+def atomic_load_max_64_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
+  (atomic_load_max_64 node:$a, node:$b)>;
 def atomic_load_umax_32_g: ATOMIC_GLOBAL_CHK<(ops node:$a, node:$b),
   (atomic_load_umax_32 node:$a, node:$b)>;
 def atomic_load_umax_32_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
   (atomic_load_umax_32 node:$a, node:$b)>;
 def atomic_load_umax_32_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
   (atomic_load_umax_32 node:$a, node:$b)>;
+def atomic_load_umax_64_g: ATOMIC_GLOBAL_CHK<(ops node:$a, node:$b),
+  (atomic_load_umax_64 node:$a, node:$b)>;
+def atomic_load_umax_64_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
+  (atomic_load_umax_64 node:$a, node:$b)>;
+def atomic_load_umax_64_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
+  (atomic_load_umax_64 node:$a, node:$b)>;
 
 defm INT_PTX_ATOM_LOAD_MAX_G_32 : F_ATOMIC_2<Int32Regs, ".global", ".s32",
   ".max", atomic_load_max_32_g, i32imm, imm, hasAtomRedG32>;
@@ -1072,6 +1089,14 @@ defm INT_PTX_ATOM_LOAD_MAX_GEN_32 : F_ATOMIC_2<Int32Regs, "", ".s32", ".max",
   atomic_load_max_32_gen, i32imm, imm, hasAtomRedGen32>;
 defm INT_PTX_ATOM_LOAD_MAX_GEN_32_USE_G : F_ATOMIC_2<Int32Regs, ".global",
   ".s32", ".max", atomic_load_max_32_gen, i32imm, imm, useAtomRedG32forGen32>;
+defm INT_PTX_ATOM_LOAD_MAX_G_64 : F_ATOMIC_2<Int64Regs, ".global", ".s64",
+  ".max", atomic_load_max_64_g, i64imm, imm, hasAtomRedG64>;
+defm INT_PTX_ATOM_LOAD_MAX_S_64 : F_ATOMIC_2<Int64Regs, ".shared", ".s64",
+  ".max", atomic_load_max_64_s, i64imm, imm, hasAtomRedS64>;
+defm INT_PTX_ATOM_LOAD_MAX_GEN_64 : F_ATOMIC_2<Int64Regs, "", ".s64", ".max",
+  atomic_load_max_64_gen, i64imm, imm, hasAtomRedGen64>;
+defm INT_PTX_ATOM_LOAD_MAX_GEN_64_USE_G : F_ATOMIC_2<Int64Regs, ".global",
+  ".s64", ".max", atomic_load_max_64_gen, i64imm, imm, useAtomRedG64forGen64>;
 defm INT_PTX_ATOM_LOAD_UMAX_G_32 : F_ATOMIC_2<Int32Regs, ".global", ".u32",
   ".max", atomic_load_umax_32_g, i32imm, imm, hasAtomRedG32>;
 defm INT_PTX_ATOM_LOAD_UMAX_S_32 : F_ATOMIC_2<Int32Regs, ".shared", ".u32",
@@ -1080,6 +1105,14 @@ defm INT_PTX_ATOM_LOAD_UMAX_GEN_32 : F_ATOMIC_2<Int32Regs, "", ".u32", ".max",
   atomic_load_umax_32_gen, i32imm, imm, hasAtomRedGen32>;
 defm INT_PTX_ATOM_LOAD_UMAX_GEN_32_USE_G : F_ATOMIC_2<Int32Regs, ".global",
   ".u32", ".max", atomic_load_umax_32_gen, i32imm, imm, useAtomRedG32forGen32>;
+defm INT_PTX_ATOM_LOAD_UMAX_G_64 : F_ATOMIC_2<Int64Regs, ".global", ".u64",
+  ".max", atomic_load_umax_64_g, i64imm, imm, hasAtomRedG64>;
+defm INT_PTX_ATOM_LOAD_UMAX_S_64 : F_ATOMIC_2<Int64Regs, ".shared", ".u64",
+  ".max", atomic_load_umax_64_s, i64imm, imm, hasAtomRedS64>;
+defm INT_PTX_ATOM_LOAD_UMAX_GEN_64 : F_ATOMIC_2<Int64Regs, "", ".u64", ".max",
+  atomic_load_umax_64_gen, i64imm, imm, hasAtomRedGen64>;
+defm INT_PTX_ATOM_LOAD_UMAX_GEN_64_USE_G : F_ATOMIC_2<Int64Regs, ".global",
+  ".u64", ".max", atomic_load_umax_64_gen, i64imm, imm, useAtomRedG64forGen64>;
 
 // atom_min
 
@@ -1089,12 +1122,24 @@ def atomic_load_min_32_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
   (atomic_load_min_32 node:$a, node:$b)>;
 def atomic_load_min_32_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
   (atomic_load_min_32 node:$a, node:$b)>;
+def atomic_load_min_64_g: ATOMIC_GLOBAL_CHK<(ops node:$a, node:$b),
+  (atomic_load_min_64 node:$a, node:$b)>;
+def atomic_load_min_64_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
+  (atomic_load_min_64 node:$a, node:$b)>;
+def atomic_load_min_64_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
+  (atomic_load_min_64 node:$a, node:$b)>;
 def atomic_load_umin_32_g: ATOMIC_GLOBAL_CHK<(ops node:$a, node:$b),
   (atomic_load_umin_32 node:$a, node:$b)>;
 def atomic_load_umin_32_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
   (atomic_load_umin_32 node:$a, node:$b)>;
 def atomic_load_umin_32_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
   (atomic_load_umin_32 node:$a, node:$b)>;
+def atomic_load_umin_64_g: ATOMIC_GLOBAL_CHK<(ops node:$a, node:$b),
+  (atomic_load_umin_64 node:$a, node:$b)>;
+def atomic_load_umin_64_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
+  (atomic_load_umin_64 node:$a, node:$b)>;
+def atomic_load_umin_64_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
+  (atomic_load_umin_64 node:$a, node:$b)>;
 
 defm INT_PTX_ATOM_LOAD_MIN_G_32 : F_ATOMIC_2<Int32Regs, ".global", ".s32",
   ".min", atomic_load_min_32_g, i32imm, imm, hasAtomRedG32>;
@@ -1104,6 +1149,14 @@ defm INT_PTX_ATOM_LOAD_MIN_GEN_32 : F_ATOMIC_2<Int32Regs, "", ".s32", ".min",
   atomic_load_min_32_gen, i32imm, imm, hasAtomRedGen32>;
 defm INT_PTX_ATOM_LOAD_MIN_GEN_32_USE_G : F_ATOMIC_2<Int32Regs, ".global",
   ".s32", ".min", atomic_load_min_32_gen, i32imm, imm, useAtomRedG32forGen32>;
+defm INT_PTX_ATOM_LOAD_MIN_G_64 : F_ATOMIC_2<Int64Regs, ".global", ".s64",
+  ".min", atomic_load_min_64_g, i64imm, imm, hasAtomRedG64>;
+defm INT_PTX_ATOM_LOAD_MIN_S_64 : F_ATOMIC_2<Int64Regs, ".shared", ".s64",
+  ".min", atomic_load_min_64_s, i64imm, imm, hasAtomRedS64>;
+defm INT_PTX_ATOM_LOAD_MIN_GEN_64 : F_ATOMIC_2<Int64Regs, "", ".s64", ".min",
+  atomic_load_min_64_gen, i64imm, imm, hasAtomRedGen64>;
+defm INT_PTX_ATOM_LOAD_MIN_GEN_64_USE_G : F_ATOMIC_2<Int64Regs, ".global",
+  ".s64", ".min", atomic_load_min_64_gen, i64imm, imm, useAtomRedG64forGen64>;
 defm INT_PTX_ATOM_LOAD_UMIN_G_32 : F_ATOMIC_2<Int32Regs, ".global", ".u32",
   ".min", atomic_load_umin_32_g, i32imm, imm, hasAtomRedG32>;
 defm INT_PTX_ATOM_LOAD_UMIN_S_32 : F_ATOMIC_2<Int32Regs, ".shared", ".u32",
@@ -1112,6 +1165,14 @@ defm INT_PTX_ATOM_LOAD_UMIN_GEN_32 : F_ATOMIC_2<Int32Regs, "", ".u32", ".min",
   atomic_load_umin_32_gen, i32imm, imm, hasAtomRedGen32>;
 defm INT_PTX_ATOM_LOAD_UMIN_GEN_32_USE_G : F_ATOMIC_2<Int32Regs, ".global",
   ".u32", ".min", atomic_load_umin_32_gen, i32imm, imm, useAtomRedG32forGen32>;
+defm INT_PTX_ATOM_LOAD_UMIN_G_64 : F_ATOMIC_2<Int64Regs, ".global", ".u64",
+  ".min", atomic_load_umin_64_g, i64imm, imm, hasAtomRedG64>;
+defm INT_PTX_ATOM_LOAD_UMIN_S_64 : F_ATOMIC_2<Int64Regs, ".shared", ".u64",
+  ".min", atomic_load_umin_64_s, i64imm, imm, hasAtomRedS64>;
+defm INT_PTX_ATOM_LOAD_UMIN_GEN_64 : F_ATOMIC_2<Int64Regs, "", ".u64", ".min",
+  atomic_load_umin_64_gen, i64imm, imm, hasAtomRedGen64>;
+defm INT_PTX_ATOM_LOAD_UMIN_GEN_64_USE_G : F_ATOMIC_2<Int64Regs, ".global",
+  ".u64", ".min", atomic_load_umin_64_gen, i64imm, imm, useAtomRedG64forGen64>;
 
 // atom_inc  atom_dec
 
@@ -1153,6 +1214,12 @@ def atomic_load_and_32_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
   (atomic_load_and_32 node:$a, node:$b)>;
 def atomic_load_and_32_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
   (atomic_load_and_32 node:$a, node:$b)>;
+def atomic_load_and_64_g: ATOMIC_GLOBAL_CHK<(ops node:$a, node:$b),
+  (atomic_load_and_64 node:$a, node:$b)>;
+def atomic_load_and_64_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
+  (atomic_load_and_64 node:$a, node:$b)>;
+def atomic_load_and_64_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
+  (atomic_load_and_64 node:$a, node:$b)>;
 
 defm INT_PTX_ATOM_AND_G_32 : F_ATOMIC_2<Int32Regs, ".global", ".b32", ".and",
   atomic_load_and_32_g, i32imm, imm, hasAtomRedG32>;
@@ -1162,6 +1229,14 @@ defm INT_PTX_ATOM_AND_GEN_32 : F_ATOMIC_2<Int32Regs, "", ".b32", ".and",
   atomic_load_and_32_gen, i32imm, imm, hasAtomRedGen32>;
 defm INT_PTX_ATOM_AND_GEN_32_USE_G : F_ATOMIC_2<Int32Regs, ".global", ".b32",
   ".and", atomic_load_and_32_gen, i32imm, imm, useAtomRedG32forGen32>;
+defm INT_PTX_ATOM_AND_G_64 : F_ATOMIC_2<Int64Regs, ".global", ".b64", ".and",
+  atomic_load_and_64_g, i64imm, imm, hasAtomRedG64>;
+defm INT_PTX_ATOM_AND_S_64 : F_ATOMIC_2<Int64Regs, ".shared", ".b64", ".and",
+  atomic_load_and_64_s, i64imm, imm, hasAtomRedS64>;
+defm INT_PTX_ATOM_AND_GEN_64 : F_ATOMIC_2<Int64Regs, "", ".b64", ".and",
+  atomic_load_and_64_gen, i64imm, imm, hasAtomRedGen64>;
+defm INT_PTX_ATOM_AND_GEN_64_USE_G : F_ATOMIC_2<Int64Regs, ".global", ".b64",
+  ".and", atomic_load_and_64_gen, i64imm, imm, useAtomRedG64forGen64>;
 
 // atom_or
 
@@ -1171,6 +1246,12 @@ def atomic_load_or_32_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
   (atomic_load_or_32 node:$a, node:$b)>;
 def atomic_load_or_32_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
   (atomic_load_or_32 node:$a, node:$b)>;
+def atomic_load_or_64_g: ATOMIC_GLOBAL_CHK<(ops node:$a, node:$b),
+  (atomic_load_or_64 node:$a, node:$b)>;
+def atomic_load_or_64_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
+  (atomic_load_or_64 node:$a, node:$b)>;
+def atomic_load_or_64_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
+  (atomic_load_or_64 node:$a, node:$b)>;
 
 defm INT_PTX_ATOM_OR_G_32 : F_ATOMIC_2<Int32Regs, ".global", ".b32", ".or",
   atomic_load_or_32_g, i32imm, imm, hasAtomRedG32>;
@@ -1180,6 +1261,14 @@ defm INT_PTX_ATOM_OR_GEN_32_USE_G : F_ATOMIC_2<Int32Regs, ".global", ".b32",
   ".or", atomic_load_or_32_gen, i32imm, imm, useAtomRedG32forGen32>;
 defm INT_PTX_ATOM_OR_S_32 : F_ATOMIC_2<Int32Regs, ".shared", ".b32", ".or",
   atomic_load_or_32_s, i32imm, imm, hasAtomRedS32>;
+defm INT_PTX_ATOM_OR_G_64 : F_ATOMIC_2<Int64Regs, ".global", ".b64", ".or",
+  atomic_load_or_64_g, i64imm, imm, hasAtomRedG64>;
+defm INT_PTX_ATOM_OR_GEN_64 : F_ATOMIC_2<Int64Regs, "", ".b64", ".or",
+  atomic_load_or_64_gen, i64imm, imm, hasAtomRedGen64>;
+defm INT_PTX_ATOM_OR_GEN_64_USE_G : F_ATOMIC_2<Int64Regs, ".global", ".b64",
+  ".or", atomic_load_or_64_gen, i64imm, imm, useAtomRedG64forGen64>;
+defm INT_PTX_ATOM_OR_S_64 : F_ATOMIC_2<Int64Regs, ".shared", ".b64", ".or",
+  atomic_load_or_64_s, i64imm, imm, hasAtomRedS64>;
 
 // atom_xor
 
@@ -1189,6 +1278,12 @@ def atomic_load_xor_32_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
   (atomic_load_xor_32 node:$a, node:$b)>;
 def atomic_load_xor_32_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
   (atomic_load_xor_32 node:$a, node:$b)>;
+def atomic_load_xor_64_g: ATOMIC_GLOBAL_CHK<(ops node:$a, node:$b),
+  (atomic_load_xor_64 node:$a, node:$b)>;
+def atomic_load_xor_64_s: ATOMIC_SHARED_CHK<(ops node:$a, node:$b),
+  (atomic_load_xor_64 node:$a, node:$b)>;
+def atomic_load_xor_64_gen: ATOMIC_GENERIC_CHK<(ops node:$a, node:$b),
+  (atomic_load_xor_64 node:$a, node:$b)>;
 
 defm INT_PTX_ATOM_XOR_G_32 : F_ATOMIC_2<Int32Regs, ".global", ".b32", ".xor",
   atomic_load_xor_32_g, i32imm, imm, hasAtomRedG32>;
@@ -1198,6 +1293,14 @@ defm INT_PTX_ATOM_XOR_GEN_32 : F_ATOMIC_2<Int32Regs, "", ".b32", ".xor",
   atomic_load_xor_32_gen, i32imm, imm, hasAtomRedGen32>;
 defm INT_PTX_ATOM_XOR_GEN_32_USE_G : F_ATOMIC_2<Int32Regs, ".global", ".b32",
   ".xor", atomic_load_xor_32_gen, i32imm, imm, useAtomRedG32forGen32>;
+defm INT_PTX_ATOM_XOR_G_64 : F_ATOMIC_2<Int64Regs, ".global", ".b64", ".xor",
+  atomic_load_xor_64_g, i64imm, imm, hasAtomRedG64>;
+defm INT_PTX_ATOM_XOR_S_64 : F_ATOMIC_2<Int64Regs, ".shared", ".b64", ".xor",
+  atomic_load_xor_64_s, i64imm, imm, hasAtomRedS64>;
+defm INT_PTX_ATOM_XOR_GEN_64 : F_ATOMIC_2<Int64Regs, "", ".b64", ".xor",
+  atomic_load_xor_64_gen, i64imm, imm, hasAtomRedGen64>;
+defm INT_PTX_ATOM_XOR_GEN_64_USE_G : F_ATOMIC_2<Int64Regs, ".global", ".b64",
+  ".xor", atomic_load_xor_64_gen, i64imm, imm, useAtomRedG64forGen64>;
 
 // atom_cas
 
@@ -1276,67 +1379,33 @@ def INT_PTX_SREG_WARPSIZE : F_SREG<"mov.u32 \t$dst, WARP_SZ;", Int32Regs,
 // Support for ldu on sm_20 or later
 //-----------------------------------
 
-def ldu_i8 : PatFrag<(ops node:$ptr), (int_nvvm_ldu_global_i node:$ptr), [{
-  MemIntrinsicSDNode *M = cast<MemIntrinsicSDNode>(N);
-  return M->getMemoryVT() == MVT::i8;
-}]>;
-
 // Scalar
-// @TODO: Revisit this, Changed imemAny to imem
-multiclass LDU_G<string TyStr, NVPTXRegClass regclass, Intrinsic IntOp> {
+multiclass LDU_G<string TyStr, NVPTXRegClass regclass> {
   def areg: NVPTXInst<(outs regclass:$result), (ins Int32Regs:$src),
                !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp Int32Regs:$src))]>, Requires<[hasLDU]>;
+                      []>, Requires<[hasLDU]>;
   def areg64: NVPTXInst<(outs regclass:$result), (ins Int64Regs:$src),
                !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp Int64Regs:$src))]>, Requires<[hasLDU]>;
- def avar:  NVPTXInst<(outs regclass:$result), (ins imem:$src),
+                        []>, Requires<[hasLDU]>;
+ def avar:  NVPTXInst<(outs regclass:$result), (ins imemAny:$src),
                !strconcat("ldu.global.", TyStr),
-                [(set regclass:$result, (IntOp (Wrapper tglobaladdr:$src)))]>,
-                Requires<[hasLDU]>;
+                      []>, Requires<[hasLDU]>;
  def ari :  NVPTXInst<(outs regclass:$result), (ins MEMri:$src),
                !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp ADDRri:$src))]>, Requires<[hasLDU]>;
+                      []>, Requires<[hasLDU]>;
  def ari64 :  NVPTXInst<(outs regclass:$result), (ins MEMri64:$src),
                !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp ADDRri64:$src))]>, Requires<[hasLDU]>;
+                        []>, Requires<[hasLDU]>;
 }
 
-multiclass LDU_G_NOINTRIN<string TyStr, NVPTXRegClass regclass, PatFrag IntOp> {
-  def areg: NVPTXInst<(outs regclass:$result), (ins Int32Regs:$src),
-               !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp Int32Regs:$src))]>, Requires<[hasLDU]>;
-  def areg64: NVPTXInst<(outs regclass:$result), (ins Int64Regs:$src),
-               !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp Int64Regs:$src))]>, Requires<[hasLDU]>;
- def avar:  NVPTXInst<(outs regclass:$result), (ins imem:$src),
-               !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp (Wrapper tglobaladdr:$src)))]>,
-         Requires<[hasLDU]>;
- def ari :  NVPTXInst<(outs regclass:$result), (ins MEMri:$src),
-               !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp ADDRri:$src))]>, Requires<[hasLDU]>;
- def ari64 :  NVPTXInst<(outs regclass:$result), (ins MEMri64:$src),
-               !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp ADDRri64:$src))]>, Requires<[hasLDU]>;
-}
-
-defm INT_PTX_LDU_GLOBAL_i8  : LDU_G_NOINTRIN<"u8 \t$result, [$src];", Int16Regs,
-                                             ldu_i8>;
-defm INT_PTX_LDU_GLOBAL_i16 : LDU_G<"u16 \t$result, [$src];", Int16Regs,
-int_nvvm_ldu_global_i>;
-defm INT_PTX_LDU_GLOBAL_i32 : LDU_G<"u32 \t$result, [$src];", Int32Regs,
-int_nvvm_ldu_global_i>;
-defm INT_PTX_LDU_GLOBAL_i64 : LDU_G<"u64 \t$result, [$src];", Int64Regs,
-int_nvvm_ldu_global_i>;
-defm INT_PTX_LDU_GLOBAL_f32 : LDU_G<"f32 \t$result, [$src];", Float32Regs,
-int_nvvm_ldu_global_f>;
-defm INT_PTX_LDU_GLOBAL_f64 : LDU_G<"f64 \t$result, [$src];", Float64Regs,
-int_nvvm_ldu_global_f>;
-defm INT_PTX_LDU_GLOBAL_p32 : LDU_G<"u32 \t$result, [$src];", Int32Regs,
-int_nvvm_ldu_global_p>;
-defm INT_PTX_LDU_GLOBAL_p64 : LDU_G<"u64 \t$result, [$src];", Int64Regs,
-int_nvvm_ldu_global_p>;
+defm INT_PTX_LDU_GLOBAL_i8  : LDU_G<"u8 \t$result, [$src];", Int16Regs>;
+defm INT_PTX_LDU_GLOBAL_i16 : LDU_G<"u16 \t$result, [$src];", Int16Regs>;
+defm INT_PTX_LDU_GLOBAL_i32 : LDU_G<"u32 \t$result, [$src];", Int32Regs>;
+defm INT_PTX_LDU_GLOBAL_i64 : LDU_G<"u64 \t$result, [$src];", Int64Regs>;
+defm INT_PTX_LDU_GLOBAL_f32 : LDU_G<"f32 \t$result, [$src];", Float32Regs>;
+defm INT_PTX_LDU_GLOBAL_f64 : LDU_G<"f64 \t$result, [$src];", Float64Regs>;
+defm INT_PTX_LDU_GLOBAL_p32 : LDU_G<"u32 \t$result, [$src];", Int32Regs>;
+defm INT_PTX_LDU_GLOBAL_p64 : LDU_G<"u64 \t$result, [$src];", Int64Regs>;
 
 // vector
 
@@ -1406,65 +1475,40 @@ defm INT_PTX_LDU_G_v4f32_ELE
 // Support for ldg on sm_35 or later 
 //-----------------------------------
 
-def ldg_i8 : PatFrag<(ops node:$ptr), (int_nvvm_ldg_global_i node:$ptr), [{
-  MemIntrinsicSDNode *M = cast<MemIntrinsicSDNode>(N);
-  return M->getMemoryVT() == MVT::i8;
-}]>;
-
-multiclass LDG_G<string TyStr, NVPTXRegClass regclass, Intrinsic IntOp> {
+multiclass LDG_G<string TyStr, NVPTXRegClass regclass> {
   def areg: NVPTXInst<(outs regclass:$result), (ins Int32Regs:$src),
                !strconcat("ld.global.nc.", TyStr),
-         [(set regclass:$result, (IntOp Int32Regs:$src))]>, Requires<[hasLDG]>;
+                      []>, Requires<[hasLDG]>;
   def areg64: NVPTXInst<(outs regclass:$result), (ins Int64Regs:$src),
                !strconcat("ld.global.nc.", TyStr),
-         [(set regclass:$result, (IntOp Int64Regs:$src))]>, Requires<[hasLDG]>;
- def avar:  NVPTXInst<(outs regclass:$result), (ins imem:$src),
+                        []>, Requires<[hasLDG]>;
+ def avar:  NVPTXInst<(outs regclass:$result), (ins imemAny:$src),
                !strconcat("ld.global.nc.", TyStr),
-         [(set regclass:$result, (IntOp (Wrapper tglobaladdr:$src)))]>,
-         Requires<[hasLDG]>;
+                      []>, Requires<[hasLDG]>;
  def ari :  NVPTXInst<(outs regclass:$result), (ins MEMri:$src),
                !strconcat("ld.global.nc.", TyStr),
-         [(set regclass:$result, (IntOp ADDRri:$src))]>, Requires<[hasLDG]>;
+                      []>, Requires<[hasLDG]>;
  def ari64 :  NVPTXInst<(outs regclass:$result), (ins MEMri64:$src),
                !strconcat("ld.global.nc.", TyStr),
-         [(set regclass:$result, (IntOp ADDRri64:$src))]>, Requires<[hasLDG]>;
-}
-
-multiclass LDG_G_NOINTRIN<string TyStr, NVPTXRegClass regclass, PatFrag IntOp> {
-  def areg: NVPTXInst<(outs regclass:$result), (ins Int32Regs:$src),
-               !strconcat("ld.global.nc.", TyStr),
-         [(set regclass:$result, (IntOp Int32Regs:$src))]>, Requires<[hasLDG]>;
-  def areg64: NVPTXInst<(outs regclass:$result), (ins Int64Regs:$src),
-               !strconcat("ld.global.nc.", TyStr),
-         [(set regclass:$result, (IntOp Int64Regs:$src))]>, Requires<[hasLDG]>;
- def avar:  NVPTXInst<(outs regclass:$result), (ins imem:$src),
-               !strconcat("ld.global.nc.", TyStr),
-         [(set regclass:$result, (IntOp (Wrapper tglobaladdr:$src)))]>,
-        Requires<[hasLDG]>;
- def ari :  NVPTXInst<(outs regclass:$result), (ins MEMri:$src),
-               !strconcat("ld.global.nc.", TyStr),
-         [(set regclass:$result, (IntOp ADDRri:$src))]>, Requires<[hasLDG]>;
- def ari64 :  NVPTXInst<(outs regclass:$result), (ins MEMri64:$src),
-               !strconcat("ld.global.nc.", TyStr),
-         [(set regclass:$result, (IntOp ADDRri64:$src))]>, Requires<[hasLDG]>;
+                        []>, Requires<[hasLDG]>;
 }
 
 defm INT_PTX_LDG_GLOBAL_i8
-  : LDG_G_NOINTRIN<"u8 \t$result, [$src];",  Int16Regs, ldg_i8>;
+  : LDG_G<"u8 \t$result, [$src];", Int16Regs>;
 defm INT_PTX_LDG_GLOBAL_i16
-  : LDG_G<"u16 \t$result, [$src];", Int16Regs,   int_nvvm_ldg_global_i>;
+  : LDG_G<"u16 \t$result, [$src];", Int16Regs>;
 defm INT_PTX_LDG_GLOBAL_i32
-  : LDG_G<"u32 \t$result, [$src];", Int32Regs,   int_nvvm_ldg_global_i>;
+  : LDG_G<"u32 \t$result, [$src];", Int32Regs>;
 defm INT_PTX_LDG_GLOBAL_i64
-  : LDG_G<"u64 \t$result, [$src];", Int64Regs,   int_nvvm_ldg_global_i>;
+  : LDG_G<"u64 \t$result, [$src];", Int64Regs>;
 defm INT_PTX_LDG_GLOBAL_f32
-  : LDG_G<"f32 \t$result, [$src];", Float32Regs, int_nvvm_ldg_global_f>;
+  : LDG_G<"f32 \t$result, [$src];", Float32Regs>;
 defm INT_PTX_LDG_GLOBAL_f64
-  : LDG_G<"f64 \t$result, [$src];", Float64Regs, int_nvvm_ldg_global_f>;
+  : LDG_G<"f64 \t$result, [$src];", Float64Regs>;
 defm INT_PTX_LDG_GLOBAL_p32
-  : LDG_G<"u32 \t$result, [$src];", Int32Regs,   int_nvvm_ldg_global_p>;
+  : LDG_G<"u32 \t$result, [$src];", Int32Regs>;
 defm INT_PTX_LDG_GLOBAL_p64
-  : LDG_G<"u64 \t$result, [$src];", Int64Regs,   int_nvvm_ldg_global_p>;
+  : LDG_G<"u64 \t$result, [$src];", Int64Regs>;
 
 // vector
 
@@ -1666,6 +1710,9 @@ def : Pat<(i32 (int_nvvm_ptr_gen_to_local (int_nvvm_ptr_local_to_gen
                 (MoveParam texternalsym:$src)))),
                (nvvm_move_ptr32  texternalsym:$src)>;
 
+def texsurf_handles
+  : NVPTXInst<(outs Int64Regs:$result), (ins imem:$src),
+              "mov.u64 \t$result, $src;", []>;
 
 //-----------------------------------
 // Compiler Error Warn
@@ -1686,6 +1733,5224 @@ def INT_NVVM_COMPILER_ERROR_64 : NVPTXInst<(outs), (ins Int64Regs:$a),
                 [(int_nvvm_compiler_error Int64Regs:$a)]>;
 
 
+// isspacep
+
+def ISSPACEP_CONST_32
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int32Regs:$a),
+              "isspacep.const \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_isspacep_const Int32Regs:$a))]>,
+    Requires<[hasPTX31]>;
+def ISSPACEP_CONST_64
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int64Regs:$a),
+              "isspacep.const \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_isspacep_const Int64Regs:$a))]>,
+    Requires<[hasPTX31]>;
+def ISSPACEP_GLOBAL_32
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int32Regs:$a),
+              "isspacep.global \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_isspacep_global Int32Regs:$a))]>;
+def ISSPACEP_GLOBAL_64
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int64Regs:$a),
+              "isspacep.global \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_isspacep_global Int64Regs:$a))]>;
+def ISSPACEP_LOCAL_32
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int32Regs:$a),
+              "isspacep.local \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_isspacep_local Int32Regs:$a))]>;
+def ISSPACEP_LOCAL_64
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int64Regs:$a),
+              "isspacep.local \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_isspacep_local Int64Regs:$a))]>;
+def ISSPACEP_SHARED_32
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int32Regs:$a),
+              "isspacep.shared \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_isspacep_shared Int32Regs:$a))]>;
+def ISSPACEP_SHARED_64
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int64Regs:$a),
+              "isspacep.shared \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_isspacep_shared Int64Regs:$a))]>;
+
+
+// Special register reads
+def MOV_SPECIAL : NVPTXInst<(outs Int32Regs:$d),
+                            (ins SpecialRegs:$r),
+                            "mov.b32\t$d, $r;", []>;
+
+def : Pat<(int_nvvm_read_ptx_sreg_envreg0), (MOV_SPECIAL ENVREG0)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg1), (MOV_SPECIAL ENVREG1)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg2), (MOV_SPECIAL ENVREG2)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg3), (MOV_SPECIAL ENVREG3)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg4), (MOV_SPECIAL ENVREG4)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg5), (MOV_SPECIAL ENVREG5)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg6), (MOV_SPECIAL ENVREG6)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg7), (MOV_SPECIAL ENVREG7)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg8), (MOV_SPECIAL ENVREG8)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg9), (MOV_SPECIAL ENVREG9)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg10), (MOV_SPECIAL ENVREG10)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg11), (MOV_SPECIAL ENVREG11)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg12), (MOV_SPECIAL ENVREG12)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg13), (MOV_SPECIAL ENVREG13)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg14), (MOV_SPECIAL ENVREG14)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg15), (MOV_SPECIAL ENVREG15)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg16), (MOV_SPECIAL ENVREG16)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg17), (MOV_SPECIAL ENVREG17)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg18), (MOV_SPECIAL ENVREG18)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg19), (MOV_SPECIAL ENVREG19)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg20), (MOV_SPECIAL ENVREG20)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg21), (MOV_SPECIAL ENVREG21)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg22), (MOV_SPECIAL ENVREG22)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg23), (MOV_SPECIAL ENVREG23)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg24), (MOV_SPECIAL ENVREG24)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg25), (MOV_SPECIAL ENVREG25)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg26), (MOV_SPECIAL ENVREG26)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg27), (MOV_SPECIAL ENVREG27)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg28), (MOV_SPECIAL ENVREG28)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg29), (MOV_SPECIAL ENVREG29)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg30), (MOV_SPECIAL ENVREG30)>;
+def : Pat<(int_nvvm_read_ptx_sreg_envreg31), (MOV_SPECIAL ENVREG31)>;
+
+
+// rotate builtin support
+
+def ROTATE_B32_HW_IMM
+  : NVPTXInst<(outs Int32Regs:$dst),
+              (ins  Int32Regs:$src, i32imm:$amt),
+              "shf.l.wrap.b32 \t$dst, $src, $src, $amt;",
+              [(set Int32Regs:$dst,
+                 (int_nvvm_rotate_b32 Int32Regs:$src, (i32 imm:$amt)))]>,
+              Requires<[hasHWROT32]> ;
+
+def ROTATE_B32_HW_REG
+  : NVPTXInst<(outs Int32Regs:$dst),
+              (ins  Int32Regs:$src, Int32Regs:$amt),
+              "shf.l.wrap.b32 \t$dst, $src, $src, $amt;",
+              [(set Int32Regs:$dst,
+                 (int_nvvm_rotate_b32 Int32Regs:$src, Int32Regs:$amt))]>,
+              Requires<[hasHWROT32]> ;
+
+def : Pat<(int_nvvm_rotate_b32 Int32Regs:$src, (i32 imm:$amt)),
+          (ROT32imm_sw Int32Regs:$src, imm:$amt, (SUB_FRM_32 node:$amt))>,
+      Requires<[noHWROT32]> ;
+
+def : Pat<(int_nvvm_rotate_b32 Int32Regs:$src, Int32Regs:$amt),
+          (ROTL32reg_sw Int32Regs:$src, Int32Regs:$amt)>,
+      Requires<[noHWROT32]> ;
+
+def GET_LO_INT64
+  : NVPTXInst<(outs Int32Regs:$dst), (ins Int64Regs:$src),
+              !strconcat("{{\n\t",
+              !strconcat(".reg .b32 %dummy;\n\t",
+              !strconcat("mov.b64 \t{$dst,%dummy}, $src;\n\t",
+        !strconcat("}}", "")))),
+        []> ;
+
+def GET_HI_INT64
+  : NVPTXInst<(outs Int32Regs:$dst), (ins Int64Regs:$src),
+              !strconcat("{{\n\t",
+              !strconcat(".reg .b32 %dummy;\n\t",
+              !strconcat("mov.b64 \t{%dummy,$dst}, $src;\n\t",
+        !strconcat("}}", "")))),
+        []> ;
+
+def PACK_TWO_INT32
+  : NVPTXInst<(outs Int64Regs:$dst), (ins Int32Regs:$lo, Int32Regs:$hi),
+              "mov.b64 \t$dst, {{$lo, $hi}};", []> ;
+
+def : Pat<(int_nvvm_swap_lo_hi_b64 Int64Regs:$src),
+          (PACK_TWO_INT32 (GET_HI_INT64 Int64Regs:$src),
+                          (GET_LO_INT64 Int64Regs:$src))> ;
+
+// funnel shift, requires >= sm_32
+def SHF_L_WRAP_B32_IMM
+  : NVPTXInst<(outs Int32Regs:$dst),
+              (ins  Int32Regs:$lo, Int32Regs:$hi, i32imm:$amt),
+              "shf.l.wrap.b32 \t$dst, $lo, $hi, $amt;",[]>,
+    Requires<[hasHWROT32]>;
+
+def SHF_L_WRAP_B32_REG
+  : NVPTXInst<(outs Int32Regs:$dst),
+              (ins  Int32Regs:$lo, Int32Regs:$hi, Int32Regs:$amt),
+              "shf.l.wrap.b32 \t$dst, $lo, $hi, $amt;",[]>,
+    Requires<[hasHWROT32]>;
+
+def SHF_R_WRAP_B32_IMM
+  : NVPTXInst<(outs Int32Regs:$dst),
+              (ins  Int32Regs:$lo, Int32Regs:$hi, i32imm:$amt),
+              "shf.r.wrap.b32 \t$dst, $lo, $hi, $amt;",[]>,
+    Requires<[hasHWROT32]>;
+
+def SHF_R_WRAP_B32_REG
+  : NVPTXInst<(outs Int32Regs:$dst),
+              (ins  Int32Regs:$lo, Int32Regs:$hi, Int32Regs:$amt),
+              "shf.r.wrap.b32 \t$dst, $lo, $hi, $amt;",[]>,
+    Requires<[hasHWROT32]>;
+
+// HW version of rotate 64
+def : Pat<(int_nvvm_rotate_b64 Int64Regs:$src, (i32 imm:$amt)),
+          (PACK_TWO_INT32
+            (SHF_L_WRAP_B32_IMM (GET_HI_INT64 Int64Regs:$src),
+                                (GET_LO_INT64 Int64Regs:$src), imm:$amt),
+            (SHF_L_WRAP_B32_IMM (GET_LO_INT64 Int64Regs:$src),
+                                (GET_HI_INT64 Int64Regs:$src), imm:$amt))>,
+      Requires<[hasHWROT32]>;
+
+def : Pat<(int_nvvm_rotate_b64 Int64Regs:$src, Int32Regs:$amt),
+          (PACK_TWO_INT32
+            (SHF_L_WRAP_B32_REG (GET_HI_INT64 Int64Regs:$src),
+                                (GET_LO_INT64 Int64Regs:$src), Int32Regs:$amt),
+            (SHF_L_WRAP_B32_REG (GET_LO_INT64 Int64Regs:$src),
+                               (GET_HI_INT64 Int64Regs:$src), Int32Regs:$amt))>,
+      Requires<[hasHWROT32]>;
+
+
+def : Pat<(int_nvvm_rotate_right_b64 Int64Regs:$src, (i32 imm:$amt)),
+          (PACK_TWO_INT32
+            (SHF_R_WRAP_B32_IMM (GET_LO_INT64 Int64Regs:$src),
+                                (GET_HI_INT64 Int64Regs:$src), imm:$amt),
+            (SHF_R_WRAP_B32_IMM (GET_HI_INT64 Int64Regs:$src),
+                                (GET_LO_INT64 Int64Regs:$src), imm:$amt))>,
+      Requires<[hasHWROT32]>;
+
+def : Pat<(int_nvvm_rotate_right_b64 Int64Regs:$src, Int32Regs:$amt),
+          (PACK_TWO_INT32
+            (SHF_R_WRAP_B32_REG (GET_LO_INT64 Int64Regs:$src),
+                                (GET_HI_INT64 Int64Regs:$src), Int32Regs:$amt),
+            (SHF_R_WRAP_B32_REG (GET_HI_INT64 Int64Regs:$src),
+                               (GET_LO_INT64 Int64Regs:$src), Int32Regs:$amt))>,
+      Requires<[hasHWROT32]>;
+
+// SW version of rotate 64
+def : Pat<(int_nvvm_rotate_b64 Int64Regs:$src, (i32 imm:$amt)),
+          (ROT64imm_sw Int64Regs:$src, imm:$amt, (SUB_FRM_32 node:$amt))>,
+      Requires<[noHWROT32]>;
+def : Pat<(int_nvvm_rotate_b64 Int64Regs:$src, Int32Regs:$amt),
+          (ROTL64reg_sw Int64Regs:$src, Int32Regs:$amt)>,
+      Requires<[noHWROT32]>;
+def : Pat<(int_nvvm_rotate_right_b64 Int64Regs:$src, (i32 imm:$amt)),
+          (ROT64imm_sw Int64Regs:$src, (SUB_FRM_64 node:$amt), imm:$amt)>,
+      Requires<[noHWROT32]>;
+def : Pat<(int_nvvm_rotate_right_b64 Int64Regs:$src, Int32Regs:$amt),
+          (ROTR64reg_sw Int64Regs:$src, Int32Regs:$amt)>,
+      Requires<[noHWROT32]>;
+
+
+//-----------------------------------
+// Texture Intrinsics
+//-----------------------------------
+
+// NOTE: For Fermi support, any new texture/surface/sampler intrinsics must be
+// also defined in NVPTXReplaceImageHandles.cpp
+
+// texmode_independent
+let IsTex = 1, IsTexModeUnified = 0 in {
+// Texture fetch instructions using handles
+def TEX_1D_F32_S32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$x),
+              "tex.1d.v4.f32.s32\t\\{$r, $g, $b, $a\\}, [$t, $s, \\{$x\\}];",
+              []>;
+def TEX_1D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x),
+              "tex.1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, [$t, $s, \\{$x\\}];",
+              []>;
+def TEX_1D_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$lod),
+              "tex.level.1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x\\}], $lod;",
+              []>;
+def TEX_1D_F32_F32_GRAD
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+def TEX_1D_S32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$x),
+              "tex.1d.v4.s32.s32\t\\{$r, $g, $b, $a\\}, [$t, $s, \\{$x\\}];",
+              []>;
+def TEX_1D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x),
+              "tex.1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, [$t, $s, \\{$x\\}];",
+              []>;
+def TEX_1D_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x,
+                   Float32Regs:$lod),
+              "tex.level.1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x\\}], $lod;",
+              []>;
+def TEX_1D_S32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+def TEX_1D_U32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$x),
+              "tex.1d.v4.u32.s32\t\\{$r, $g, $b, $a\\}, [$t, $s, \\{$x\\}];",
+              []>;
+def TEX_1D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x),
+              "tex.1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, [$t, $s, \\{$x\\}];",
+              []>;
+def TEX_1D_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x,
+                   Float32Regs:$lod),
+              "tex.level.1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x\\}], $lod;",
+              []>;
+def TEX_1D_U32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+
+def TEX_1D_ARRAY_F32_S32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "tex.a1d.v4.f32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}];",
+              []>;
+def TEX_1D_ARRAY_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x),
+              "tex.a1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}];",
+              []>;
+def TEX_1D_ARRAY_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$lod),
+              "tex.level.a1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}], $lod;",
+              []>;
+def TEX_1D_ARRAY_F32_F32_GRAD
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.a1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+def TEX_1D_ARRAY_S32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "tex.a1d.v4.s32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}];",
+              []>;
+def TEX_1D_ARRAY_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x),
+              "tex.a1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}];",
+              []>;
+def TEX_1D_ARRAY_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$lod),
+              "tex.level.a1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}], $lod;",
+              []>;
+def TEX_1D_ARRAY_S32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.a1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+def TEX_1D_ARRAY_U32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "tex.a1d.v4.u32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}];",
+              []>;
+def TEX_1D_ARRAY_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x),
+              "tex.a1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}];",
+              []>;
+def TEX_1D_ARRAY_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$lod),
+              "tex.level.a1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}], $lod;",
+              []>;
+def TEX_1D_ARRAY_U32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.a1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+
+def TEX_2D_F32_S32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "tex.2d.v4.f32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TEX_2D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tex.2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TEX_2D_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$lod),
+              "tex.level.2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}], $lod;",
+              []>;
+def TEX_2D_F32_F32_GRAD
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+def TEX_2D_S32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "tex.2d.v4.s32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TEX_2D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tex.2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TEX_2D_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$lod),
+              "tex.level.2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}], $lod;",
+              []>;
+def TEX_2D_S32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+def TEX_2D_U32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "tex.2d.v4.u32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TEX_2D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tex.2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TEX_2D_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$lod),
+              "tex.level.2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}], $lod;",
+              []>;
+def TEX_2D_U32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+
+def TEX_2D_ARRAY_F32_S32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+                   Int32Regs:$y),
+              "tex.a2d.v4.f32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_2D_ARRAY_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y),
+              "tex.a2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_2D_ARRAY_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y, Float32Regs:$lod),
+              "tex.level.a2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}], $lod;",
+              []>;
+def TEX_2D_ARRAY_F32_F32_GRAD
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y, Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.a2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+def TEX_2D_ARRAY_S32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+                   Int32Regs:$y),
+              "tex.a2d.v4.s32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_2D_ARRAY_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y),
+              "tex.a2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_2D_ARRAY_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y, Float32Regs:$lod),
+              "tex.level.a2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}], $lod;",
+              []>;
+def TEX_2D_ARRAY_S32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.a2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+def TEX_2D_ARRAY_U32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+                   Int32Regs:$y),
+              "tex.a2d.v4.u32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_2D_ARRAY_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y),
+              "tex.a2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_2D_ARRAY_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y, Float32Regs:$lod),
+              "tex.level.a2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}], $lod;",
+              []>;
+def TEX_2D_ARRAY_U32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.a2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+
+def TEX_3D_F32_S32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$z),
+              "tex.3d.v4.f32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_3D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z),
+              "tex.3d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_3D_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z, Float32Regs:$lod),
+              "tex.level.3d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+def TEX_3D_F32_F32_GRAD
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$gradx2, Float32Regs:$grady0,
+                   Float32Regs:$grady1, Float32Regs:$grady2),
+              "tex.grad.3d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}], "
+              "\\{$gradx0, $gradx1, $gradx2, $gradx2\\}, "
+              "\\{$grady0, $grady1, $grady2, $grady2\\};",
+              []>;
+def TEX_3D_S32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$z),
+              "tex.3d.v4.s32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_3D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z),
+              "tex.3d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_3D_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z, Float32Regs:$lod),
+              "tex.level.3d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+def TEX_3D_S32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$gradx2, Float32Regs:$grady0,
+                   Float32Regs:$grady1, Float32Regs:$grady2),
+              "tex.grad.3d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}], "
+              "\\{$gradx0, $gradx1, $gradx2, $gradx2\\}, "
+              "\\{$grady0, $grady1, $grady2, $grady2\\};",
+              []>;
+def TEX_3D_U32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$z),
+              "tex.3d.v4.u32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_3D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z),
+              "tex.3d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_3D_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z, Float32Regs:$lod),
+              "tex.level.3d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+def TEX_3D_U32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$gradx2, Float32Regs:$grady0,
+                   Float32Regs:$grady1, Float32Regs:$grady2),
+              "tex.grad.3d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}], "
+              "\\{$gradx0, $gradx1, $gradx2, $gradx2\\}, "
+              "\\{$grady0, $grady1, $grady2, $grady2\\};",
+              []>;
+
+def TEX_CUBE_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s,
+               Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.cube.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_CUBE_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.cube.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+def TEX_CUBE_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.cube.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_CUBE_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.cube.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+def TEX_CUBE_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.cube.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_CUBE_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.cube.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+
+def TEX_CUBE_ARRAY_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l,
+               Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.acube.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $z\\}];",
+              []>;
+def TEX_CUBE_ARRAY_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.acube.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $z\\}], $lod;",
+              []>;
+def TEX_CUBE_ARRAY_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.acube.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $z\\}];",
+              []>;
+def TEX_CUBE_ARRAY_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.acube.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $z\\}], $lod;",
+              []>;
+def TEX_CUBE_ARRAY_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.acube.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $z\\}];",
+              []>;
+def TEX_CUBE_ARRAY_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int64Regs:$s, Int32Regs:$l,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.acube.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, $s, \\{$l, $x, $y, $z\\}], $lod;",
+              []>;
+
+def TLD4_R_2D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$v0, Float32Regs:$v1,
+                    Float32Regs:$v2, Float32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.r.2d.v4.f32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_G_2D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$v0, Float32Regs:$v1,
+                    Float32Regs:$v2, Float32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.g.2d.v4.f32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_B_2D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$v0, Float32Regs:$v1,
+                    Float32Regs:$v2, Float32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.b.2d.v4.f32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_A_2D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$v0, Float32Regs:$v1,
+                    Float32Regs:$v2, Float32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.a.2d.v4.f32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_R_2D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.r.2d.v4.s32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_G_2D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.g.2d.v4.s32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_B_2D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.b.2d.v4.s32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_A_2D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.a.2d.v4.s32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_R_2D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.r.2d.v4.u32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_G_2D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.g.2d.v4.u32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_B_2D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.b.2d.v4.u32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+def TLD4_A_2D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Int64Regs:$s, Float32Regs:$x, Float32Regs:$y),
+              "tld4.a.2d.v4.u32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, $s, \\{$x, $y\\}];",
+              []>;
+}
+
+
+// texmode_unified
+let IsTex = 1, IsTexModeUnified = 1 in {
+// Texture fetch instructions using handles
+def TEX_UNIFIED_1D_F32_S32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$x),
+              "tex.1d.v4.f32.s32\t\\{$r, $g, $b, $a\\}, [$t, \\{$x\\}];",
+              []>;
+def TEX_UNIFIED_1D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x),
+              "tex.1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, [$t, \\{$x\\}];",
+              []>;
+def TEX_UNIFIED_1D_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$lod),
+              "tex.level.1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x\\}], $lod;",
+              []>;
+def TEX_UNIFIED_1D_F32_F32_GRAD
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+def TEX_UNIFIED_1D_S32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$x),
+              "tex.1d.v4.s32.s32\t\\{$r, $g, $b, $a\\}, [$t, \\{$x\\}];",
+              []>;
+def TEX_UNIFIED_1D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x),
+              "tex.1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, [$t, \\{$x\\}];",
+              []>;
+def TEX_UNIFIED_1D_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x,
+                   Float32Regs:$lod),
+              "tex.level.1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x\\}], $lod;",
+              []>;
+def TEX_UNIFIED_1D_S32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+def TEX_UNIFIED_1D_U32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$x),
+              "tex.1d.v4.u32.s32\t\\{$r, $g, $b, $a\\}, [$t, \\{$x\\}];",
+              []>;
+def TEX_UNIFIED_1D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x),
+              "tex.1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, [$t, \\{$x\\}];",
+              []>;
+def TEX_UNIFIED_1D_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x,
+                   Float32Regs:$lod),
+              "tex.level.1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x\\}], $lod;",
+              []>;
+def TEX_UNIFIED_1D_U32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+
+def TEX_UNIFIED_1D_ARRAY_F32_S32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Int32Regs:$x),
+              "tex.a1d.v4.f32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}];",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x),
+              "tex.a1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}];",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$lod),
+              "tex.level.a1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}], $lod;",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_F32_F32_GRAD
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.a1d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_S32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Int32Regs:$x),
+              "tex.a1d.v4.s32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}];",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x),
+              "tex.a1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}];",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$lod),
+              "tex.level.a1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}], $lod;",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_S32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.a1d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_U32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Int32Regs:$x),
+              "tex.a1d.v4.u32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}];",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x),
+              "tex.a1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}];",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$lod),
+              "tex.level.a1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}], $lod;",
+              []>;
+def TEX_UNIFIED_1D_ARRAY_U32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$gradx, Float32Regs:$grady),
+              "tex.grad.a1d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x\\}], \\{$gradx\\}, \\{$grady\\};",
+              []>;
+
+def TEX_UNIFIED_2D_F32_S32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$x, Int32Regs:$y),
+              "tex.2d.v4.f32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tex.2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$lod),
+              "tex.level.2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}], $lod;",
+              []>;
+def TEX_UNIFIED_2D_F32_F32_GRAD
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+def TEX_UNIFIED_2D_S32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$x, Int32Regs:$y),
+              "tex.2d.v4.s32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tex.2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$lod),
+              "tex.level.2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}], $lod;",
+              []>;
+def TEX_UNIFIED_2D_S32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+def TEX_UNIFIED_2D_U32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$x, Int32Regs:$y),
+              "tex.2d.v4.u32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tex.2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$lod),
+              "tex.level.2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}], $lod;",
+              []>;
+def TEX_UNIFIED_2D_U32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+
+def TEX_UNIFIED_2D_ARRAY_F32_S32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Int32Regs:$x,
+                   Int32Regs:$y),
+              "tex.a2d.v4.f32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y),
+              "tex.a2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y, Float32Regs:$lod),
+              "tex.level.a2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}], $lod;",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_F32_F32_GRAD
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y, Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.a2d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_S32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Int32Regs:$x,
+                   Int32Regs:$y),
+              "tex.a2d.v4.s32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y),
+              "tex.a2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y, Float32Regs:$lod),
+              "tex.level.a2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}], $lod;",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_S32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.a2d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_U32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Int32Regs:$x,
+                   Int32Regs:$y),
+              "tex.a2d.v4.u32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y),
+              "tex.a2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}];",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y, Float32Regs:$lod),
+              "tex.level.a2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}], $lod;",
+              []>;
+def TEX_UNIFIED_2D_ARRAY_U32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l, Float32Regs:$x,
+                   Float32Regs:$y,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$grady0, Float32Regs:$grady1),
+              "tex.grad.a2d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $y\\}], \\{$gradx0, $gradx1\\}, "
+              "\\{$grady0, $grady1\\};",
+              []>;
+
+def TEX_UNIFIED_3D_F32_S32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$z),
+              "tex.3d.v4.f32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_UNIFIED_3D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z),
+              "tex.3d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_UNIFIED_3D_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z, Float32Regs:$lod),
+              "tex.level.3d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+def TEX_UNIFIED_3D_F32_F32_GRAD
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$gradx2, Float32Regs:$grady0,
+                   Float32Regs:$grady1, Float32Regs:$grady2),
+              "tex.grad.3d.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}], "
+              "\\{$gradx0, $gradx1, $gradx2, $gradx2\\}, "
+              "\\{$grady0, $grady1, $grady2, $grady2\\};",
+              []>;
+def TEX_UNIFIED_3D_S32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$z),
+              "tex.3d.v4.s32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_UNIFIED_3D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z),
+              "tex.3d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_UNIFIED_3D_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z, Float32Regs:$lod),
+              "tex.level.3d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+def TEX_UNIFIED_3D_S32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$gradx2, Float32Regs:$grady0,
+                   Float32Regs:$grady1, Float32Regs:$grady2),
+              "tex.grad.3d.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}], "
+              "\\{$gradx0, $gradx1, $gradx2, $gradx2\\}, "
+              "\\{$grady0, $grady1, $grady2, $grady2\\};",
+              []>;
+def TEX_UNIFIED_3D_U32_S32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$z),
+              "tex.3d.v4.u32.s32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_UNIFIED_3D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z),
+              "tex.3d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_UNIFIED_3D_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z, Float32Regs:$lod),
+              "tex.level.3d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+def TEX_UNIFIED_3D_U32_F32_GRAD
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y,
+                   Float32Regs:$z,
+                   Float32Regs:$gradx0, Float32Regs:$gradx1,
+                   Float32Regs:$gradx2, Float32Regs:$grady0,
+                   Float32Regs:$grady1, Float32Regs:$grady2),
+              "tex.grad.3d.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}], "
+              "\\{$gradx0, $gradx1, $gradx2, $gradx2\\}, "
+              "\\{$grady0, $grady1, $grady2, $grady2\\};",
+              []>;
+
+def TEX_UNIFIED_CUBE_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t,
+               Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.cube.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_UNIFIED_CUBE_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.cube.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+def TEX_UNIFIED_CUBE_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.cube.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_UNIFIED_CUBE_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.cube.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+def TEX_UNIFIED_CUBE_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.cube.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}];",
+              []>;
+def TEX_UNIFIED_CUBE_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.cube.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$x, $y, $z, $z\\}], $lod;",
+              []>;
+
+def TEX_UNIFIED_CUBE_ARRAY_F32_F32
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l,
+               Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.acube.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $z\\}];",
+              []>;
+def TEX_UNIFIED_CUBE_ARRAY_F32_F32_LEVEL
+  : NVPTXInst<(outs Float32Regs:$r, Float32Regs:$g,
+                    Float32Regs:$b, Float32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.acube.v4.f32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $z\\}], $lod;",
+              []>;
+def TEX_UNIFIED_CUBE_ARRAY_S32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.acube.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $z\\}];",
+              []>;
+def TEX_UNIFIED_CUBE_ARRAY_S32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.acube.v4.s32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $z\\}], $lod;",
+              []>;
+def TEX_UNIFIED_CUBE_ARRAY_U32_F32
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z),
+              "tex.acube.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $z\\}];",
+              []>;
+def TEX_UNIFIED_CUBE_ARRAY_U32_F32_LEVEL
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g,
+                    Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$t, Int32Regs:$l,
+                   Float32Regs:$x, Float32Regs:$y, Float32Regs:$z,
+                   Float32Regs:$lod),
+              "tex.level.acube.v4.u32.f32\t\\{$r, $g, $b, $a\\}, "
+              "[$t, \\{$l, $x, $y, $z\\}], $lod;",
+              []>;
+
+def TLD4_UNIFIED_R_2D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$v0, Float32Regs:$v1,
+                    Float32Regs:$v2, Float32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.r.2d.v4.f32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_G_2D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$v0, Float32Regs:$v1,
+                    Float32Regs:$v2, Float32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.g.2d.v4.f32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_B_2D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$v0, Float32Regs:$v1,
+                    Float32Regs:$v2, Float32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.b.2d.v4.f32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_A_2D_F32_F32
+  : NVPTXInst<(outs Float32Regs:$v0, Float32Regs:$v1,
+                    Float32Regs:$v2, Float32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.a.2d.v4.f32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_R_2D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.r.2d.v4.s32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_G_2D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.g.2d.v4.s32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_B_2D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.b.2d.v4.s32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_A_2D_S32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.a.2d.v4.s32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_R_2D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.r.2d.v4.u32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_G_2D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.g.2d.v4.u32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_B_2D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.b.2d.v4.u32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+def TLD4_UNIFIED_A_2D_U32_F32
+  : NVPTXInst<(outs Int32Regs:$v0, Int32Regs:$v1,
+                    Int32Regs:$v2, Int32Regs:$v3),
+              (ins Int64Regs:$t, Float32Regs:$x, Float32Regs:$y),
+              "tld4.a.2d.v4.u32.f32\t\\{$v0, $v1, $v2, $v3\\}, "
+              "[$t, \\{$x, $y\\}];",
+              []>;
+}
+
+
+
+//=== Surface load instructions
+// .clamp variant
+let IsSuld = 1 in {
+def SULD_1D_I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b8.clamp \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b16.clamp \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b32.clamp \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_I64_CLAMP
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b64.clamp \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+
+def SULD_1D_ARRAY_I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b8.clamp \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b16.clamp \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b32.clamp \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_I64_CLAMP
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b64.clamp \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+
+def SULD_2D_I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b8.clamp \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b16.clamp \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b32.clamp \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_I64_CLAMP
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b64.clamp \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+
+def SULD_2D_ARRAY_I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b8.clamp \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b16.clamp \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b32.clamp \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_I64_CLAMP
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b64.clamp \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+
+def SULD_3D_I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b8.clamp \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b16.clamp \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b32.clamp \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_I64_CLAMP
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b64.clamp \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+}
+
+let IsSuld = 2 in {
+def SULD_1D_V2I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b8.clamp \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V2I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b16.clamp \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V2I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b32.clamp \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V2I64_CLAMP
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b64.clamp \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+
+def SULD_1D_ARRAY_V2I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b8.clamp \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V2I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b16.clamp \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V2I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b32.clamp \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V2I64_CLAMP
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b64.clamp \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+
+def SULD_2D_V2I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b8.clamp \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V2I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b16.clamp \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V2I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b32.clamp \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V2I64_CLAMP
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b64.clamp \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+
+def SULD_2D_ARRAY_V2I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b8.clamp \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V2I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b16.clamp \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V2I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b32.clamp \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V2I64_CLAMP
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b64.clamp \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+
+def SULD_3D_V2I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b8.clamp \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V2I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b16.clamp \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V2I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b32.clamp \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V2I64_CLAMP
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b64.clamp \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+}
+
+let IsSuld = 3 in {
+def SULD_1D_V4I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v4.b8.clamp \\{$r, $g, $b, $a\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V4I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v4.b16.clamp \\{$r, $g, $b, $a\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V4I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v4.b32.clamp \\{$r, $g, $b, $a\\}, [$s, \\{$x\\}];",
+              []>;
+
+def SULD_1D_ARRAY_V4I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v4.b8.clamp \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V4I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v4.b16.clamp \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V4I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v4.b32.clamp \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x\\}];",
+              []>;
+
+def SULD_2D_V4I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v4.b8.clamp \\{$r, $g, $b, $a\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V4I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v4.b16.clamp \\{$r, $g, $b, $a\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V4I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v4.b32.clamp \\{$r, $g, $b, $a\\}, [$s, \\{$x, $y\\}];",
+              []>;
+
+def SULD_2D_ARRAY_V4I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v4.b8.clamp \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V4I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v4.b16.clamp \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V4I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v4.b32.clamp \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+
+
+def SULD_3D_V4I8_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v4.b8.clamp \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V4I16_CLAMP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v4.b16.clamp \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V4I32_CLAMP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v4.b32.clamp \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+}
+
+
+// .trap variant
+let IsSuld = 1 in {
+def SULD_1D_I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b8.trap \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b16.trap \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b32.trap \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_I64_TRAP
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b64.trap \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+
+def SULD_1D_ARRAY_I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b8.trap \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b16.trap \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b32.trap \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_I64_TRAP
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b64.trap \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+
+def SULD_2D_I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b8.trap \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b16.trap \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b32.trap \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_I64_TRAP
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b64.trap \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+
+def SULD_2D_ARRAY_I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b8.trap \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b16.trap \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b32.trap \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_I64_TRAP
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b64.trap \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+
+def SULD_3D_I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b8.trap \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b16.trap \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b32.trap \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_I64_TRAP
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b64.trap \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+}
+
+let IsSuld = 2 in {
+def SULD_1D_V2I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b8.trap \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V2I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b16.trap \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V2I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b32.trap \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V2I64_TRAP
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b64.trap \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+
+def SULD_1D_ARRAY_V2I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b8.trap \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V2I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b16.trap \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V2I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b32.trap \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V2I64_TRAP
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b64.trap \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+
+def SULD_2D_V2I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b8.trap \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V2I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b16.trap \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V2I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b32.trap \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V2I64_TRAP
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b64.trap \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+
+def SULD_2D_ARRAY_V2I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b8.trap \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V2I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b16.trap \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V2I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b32.trap \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V2I64_TRAP
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b64.trap \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+
+def SULD_3D_V2I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b8.trap \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V2I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b16.trap \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V2I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b32.trap \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V2I64_TRAP
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b64.trap \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+}
+
+let IsSuld = 3 in {
+def SULD_1D_V4I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v4.b8.trap \\{$r, $g, $b, $a\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V4I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v4.b16.trap \\{$r, $g, $b, $a\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V4I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v4.b32.trap \\{$r, $g, $b, $a\\}, [$s, \\{$x\\}];",
+              []>;
+
+def SULD_1D_ARRAY_V4I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v4.b8.trap \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V4I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v4.b16.trap \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V4I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v4.b32.trap \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x\\}];",
+              []>;
+
+def SULD_2D_V4I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v4.b8.trap \\{$r, $g, $b, $a\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V4I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v4.b16.trap \\{$r, $g, $b, $a\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V4I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v4.b32.trap \\{$r, $g, $b, $a\\}, [$s, \\{$x, $y\\}];",
+              []>;
+
+def SULD_2D_ARRAY_V4I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v4.b8.trap \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V4I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v4.b16.trap \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V4I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v4.b32.trap \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+
+
+def SULD_3D_V4I8_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v4.b8.trap \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V4I16_TRAP
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v4.b16.trap \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V4I32_TRAP
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v4.b32.trap \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+}
+
+// .zero variant
+let IsSuld = 1 in {
+def SULD_1D_I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b8.zero \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b16.zero \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b32.zero \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_I64_ZERO
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.b64.zero \\{$r\\}, [$s, \\{$x\\}];",
+              []>;
+
+def SULD_1D_ARRAY_I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b8.zero \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b16.zero \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b32.zero \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_I64_ZERO
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.b64.zero \\{$r\\}, [$s, \\{$l, $x\\}];",
+              []>;
+
+def SULD_2D_I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b8.zero \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b16.zero \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b32.zero \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_I64_ZERO
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.b64.zero \\{$r\\}, [$s, \\{$x, $y\\}];",
+              []>;
+
+def SULD_2D_ARRAY_I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b8.zero \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b16.zero \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b32.zero \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_I64_ZERO
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.b64.zero \\{$r\\}, [$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+
+def SULD_3D_I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b8.zero \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b16.zero \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b32.zero \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_I64_ZERO
+  : NVPTXInst<(outs Int64Regs:$r),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.b64.zero \\{$r\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+}
+
+let IsSuld = 2 in {
+def SULD_1D_V2I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b8.zero \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V2I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b16.zero \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V2I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b32.zero \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V2I64_ZERO
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v2.b64.zero \\{$r, $g\\}, [$s, \\{$x\\}];",
+              []>;
+
+def SULD_1D_ARRAY_V2I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b8.zero \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V2I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b16.zero \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V2I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b32.zero \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V2I64_ZERO
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v2.b64.zero \\{$r, $g\\}, [$s, \\{$l, $x\\}];",
+              []>;
+
+def SULD_2D_V2I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b8.zero \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V2I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b16.zero \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V2I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b32.zero \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V2I64_ZERO
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v2.b64.zero \\{$r, $g\\}, [$s, \\{$x, $y\\}];",
+              []>;
+
+def SULD_2D_ARRAY_V2I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b8.zero \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V2I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b16.zero \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V2I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b32.zero \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V2I64_ZERO
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v2.b64.zero \\{$r, $g\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+
+def SULD_3D_V2I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b8.zero \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V2I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b16.zero \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V2I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b32.zero \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V2I64_ZERO
+  : NVPTXInst<(outs Int64Regs:$r, Int64Regs:$g),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v2.b64.zero \\{$r, $g\\}, [$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+}
+
+let IsSuld = 3 in {
+def SULD_1D_V4I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v4.b8.zero \\{$r, $g, $b, $a\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V4I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v4.b16.zero \\{$r, $g, $b, $a\\}, [$s, \\{$x\\}];",
+              []>;
+def SULD_1D_V4I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x),
+              "suld.b.1d.v4.b32.zero \\{$r, $g, $b, $a\\}, [$s, \\{$x\\}];",
+              []>;
+
+def SULD_1D_ARRAY_V4I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v4.b8.zero \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V4I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v4.b16.zero \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x\\}];",
+              []>;
+def SULD_1D_ARRAY_V4I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x),
+              "suld.b.a1d.v4.b32.zero \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x\\}];",
+              []>;
+
+def SULD_2D_V4I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v4.b8.zero \\{$r, $g, $b, $a\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V4I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v4.b16.zero \\{$r, $g, $b, $a\\}, [$s, \\{$x, $y\\}];",
+              []>;
+def SULD_2D_V4I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.2d.v4.b32.zero \\{$r, $g, $b, $a\\}, [$s, \\{$x, $y\\}];",
+              []>;
+
+def SULD_2D_ARRAY_V4I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v4.b8.zero \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V4I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v4.b16.zero \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+def SULD_2D_ARRAY_V4I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y),
+              "suld.b.a2d.v4.b32.zero \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$l, $x, $y, $y\\}];",
+              []>;
+
+
+def SULD_3D_V4I8_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v4.b8.zero \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V4I16_ZERO
+  : NVPTXInst<(outs Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v4.b16.zero \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+def SULD_3D_V4I32_ZERO
+  : NVPTXInst<(outs Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z),
+              "suld.b.3d.v4.b32.zero \\{$r, $g, $b, $a\\}, "
+              "[$s, \\{$x, $y, $z, $z\\}];",
+              []>;
+}
+
+//-----------------------------------
+// Texture Query Intrinsics
+//-----------------------------------
+
+let IsSurfTexQuery = 1 in {
+def TXQ_CHANNEL_ORDER
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "txq.channel_order.b32 \t$d, [$a];",
+              []>;
+def TXQ_CHANNEL_DATA_TYPE
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "txq.channel_data_type.b32 \t$d, [$a];",
+              []>;
+def TXQ_WIDTH
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "txq.width.b32 \t$d, [$a];",
+              []>;
+def TXQ_HEIGHT
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "txq.height.b32 \t$d, [$a];",
+              []>;
+def TXQ_DEPTH
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "txq.depth.b32 \t$d, [$a];",
+              []>;
+def TXQ_ARRAY_SIZE
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "txq.array_size.b32 \t$d, [$a];",
+              []>;
+def TXQ_NUM_SAMPLES
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "txq.num_samples.b32 \t$d, [$a];",
+              []>;
+def TXQ_NUM_MIPMAP_LEVELS
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "txq.num_mipmap_levels.b32 \t$d, [$a];",
+              []>;
+}
+
+def : Pat<(int_nvvm_txq_channel_order Int64Regs:$a),
+          (TXQ_CHANNEL_ORDER Int64Regs:$a)>;
+def : Pat<(int_nvvm_txq_channel_data_type Int64Regs:$a),
+          (TXQ_CHANNEL_DATA_TYPE Int64Regs:$a)>;
+def : Pat<(int_nvvm_txq_width Int64Regs:$a),
+          (TXQ_WIDTH Int64Regs:$a)>;
+def : Pat<(int_nvvm_txq_height Int64Regs:$a),
+          (TXQ_HEIGHT Int64Regs:$a)>;
+def : Pat<(int_nvvm_txq_depth Int64Regs:$a),
+          (TXQ_DEPTH Int64Regs:$a)>;
+def : Pat<(int_nvvm_txq_array_size Int64Regs:$a),
+          (TXQ_ARRAY_SIZE Int64Regs:$a)>;
+def : Pat<(int_nvvm_txq_num_samples Int64Regs:$a),
+          (TXQ_NUM_SAMPLES Int64Regs:$a)>;
+def : Pat<(int_nvvm_txq_num_mipmap_levels Int64Regs:$a),
+          (TXQ_NUM_MIPMAP_LEVELS Int64Regs:$a)>;
+
+
+//-----------------------------------
+// Surface Query Intrinsics
+//-----------------------------------
+
+let IsSurfTexQuery = 1 in {
+def SUQ_CHANNEL_ORDER
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "suq.channel_order.b32 \t$d, [$a];",
+              []>;
+def SUQ_CHANNEL_DATA_TYPE
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "suq.channel_data_type.b32 \t$d, [$a];",
+              []>;
+def SUQ_WIDTH
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "suq.width.b32 \t$d, [$a];",
+              []>;
+def SUQ_HEIGHT
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "suq.height.b32 \t$d, [$a];",
+              []>;
+def SUQ_DEPTH
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "suq.depth.b32 \t$d, [$a];",
+              []>;
+def SUQ_ARRAY_SIZE
+  : NVPTXInst<(outs Int32Regs:$d), (ins Int64Regs:$a),
+              "suq.array_size.b32 \t$d, [$a];",
+              []>;
+}
+
+def : Pat<(int_nvvm_suq_channel_order Int64Regs:$a),
+          (SUQ_CHANNEL_ORDER Int64Regs:$a)>;
+def : Pat<(int_nvvm_suq_channel_data_type Int64Regs:$a),
+          (SUQ_CHANNEL_DATA_TYPE Int64Regs:$a)>;
+def : Pat<(int_nvvm_suq_width Int64Regs:$a),
+          (SUQ_WIDTH Int64Regs:$a)>;
+def : Pat<(int_nvvm_suq_height Int64Regs:$a),
+          (SUQ_HEIGHT Int64Regs:$a)>;
+def : Pat<(int_nvvm_suq_depth Int64Regs:$a),
+          (SUQ_DEPTH Int64Regs:$a)>;
+def : Pat<(int_nvvm_suq_array_size Int64Regs:$a),
+          (SUQ_ARRAY_SIZE Int64Regs:$a)>;
+
+
+//===- Handle Query -------------------------------------------------------===//
+
+// TODO: These intrinsics are not yet finalized, pending PTX ISA design work
+def ISTYPEP_SAMPLER
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int64Regs:$a),
+              "istypep.samplerref \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_istypep_sampler Int64Regs:$a))]>;
+def ISTYPEP_SURFACE
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int64Regs:$a),
+              "istypep.surfref \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_istypep_surface Int64Regs:$a))]>;
+def ISTYPEP_TEXTURE
+  : NVPTXInst<(outs Int1Regs:$d), (ins Int64Regs:$a),
+              "istypep.texref \t$d, $a;",
+              [(set Int1Regs:$d, (int_nvvm_istypep_texture Int64Regs:$a))]>;
+
+//===- Surface Stores -----------------------------------------------------===//
+
+let IsSust = 1 in {
+// Unformatted
+// .clamp variant
+def SUST_B_1D_B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.1d.b8.clamp \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.1d.b16.clamp \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r),
+              "sust.b.1d.b32.clamp \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_B64_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int64Regs:$r),
+              "sust.b.1d.b64.clamp \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_V2B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+              "sust.b.1d.v2.b8.clamp \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V2B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+              "sust.b.1d.v2.b16.clamp \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V2B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+              "sust.b.1d.v2.b32.clamp \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V2B64_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int64Regs:$r, Int64Regs:$g),
+              "sust.b.1d.v2.b64.clamp \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V4B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g,
+                   Int16Regs:$b, Int16Regs:$a),
+              "sust.b.1d.v4.b8.clamp \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_V4B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g,
+                   Int16Regs:$b, Int16Regs:$a),
+              "sust.b.1d.v4.b16.clamp \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_V4B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g,
+                   Int32Regs:$b, Int32Regs:$a),
+              "sust.b.1d.v4.b32.clamp \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_1D_ARRAY_B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.a1d.b8.clamp \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.a1d.b16.clamp \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r),
+              "sust.b.a1d.b32.clamp \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_B64_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int64Regs:$r),
+              "sust.b.a1d.b64.clamp \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.a1d.v2.b8.clamp \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.a1d.v2.b16.clamp \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r,
+                   Int32Regs:$g),
+              "sust.b.a1d.v2.b32.clamp \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B64_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int64Regs:$r,
+                   Int64Regs:$g),
+              "sust.b.a1d.v2.b64.clamp \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V4B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              "sust.b.a1d.v4.b8.clamp \t[$s, \\{$idx, $x\\}], "
+              "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_ARRAY_V4B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+             "sust.b.a1d.v4.b16.clamp \t[$s, \\{$idx, $x\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_ARRAY_V4B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r,
+                   Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+             "sust.b.a1d.v4.b32.clamp \t[$s, \\{$idx, $x\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_2D_B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+              "sust.b.2d.b8.clamp \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+              "sust.b.2d.b16.clamp \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+              "sust.b.2d.b32.clamp \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_B64_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r),
+              "sust.b.2d.b64.clamp \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_V2B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.2d.v2.b8.clamp \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V2B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.2d.v2.b16.clamp \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V2B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+                   Int32Regs:$g),
+              "sust.b.2d.v2.b32.clamp \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V2B64_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r,
+                   Int64Regs:$g),
+              "sust.b.2d.v2.b64.clamp \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V4B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              "sust.b.2d.v4.b8.clamp \t[$s, \\{$x, $y\\}], "
+              "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_V4B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+             "sust.b.2d.v4.b16.clamp \t[$s, \\{$x, $y\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_V4B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+                   Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+             "sust.b.2d.v4.b32.clamp \t[$s, \\{$x, $y\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_2D_ARRAY_B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r),
+              "sust.b.a2d.b8.clamp \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r),
+              "sust.b.a2d.b16.clamp \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r),
+              "sust.b.a2d.b32.clamp \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_B64_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int64Regs:$r),
+              "sust.b.a2d.b64.clamp \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.b.a2d.v2.b8.clamp \t[$s, \\{$idx, $x, $y, $y\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g),
+             "sust.b.a2d.v2.b16.clamp \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r, Int32Regs:$g),
+             "sust.b.a2d.v2.b32.clamp \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B64_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int64Regs:$r, Int64Regs:$g),
+             "sust.b.a2d.v2.b64.clamp \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V4B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+      "sust.b.a2d.v4.b8.clamp \t[$s, \\{$idx, $x, $y, $y\\}], "
+      "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_ARRAY_V4B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+     "sust.b.a2d.v4.b16.clamp \t[$s, \\{$idx, $x, $y, $y\\}], "
+     "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_ARRAY_V4B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+     "sust.b.a2d.v4.b32.clamp \t[$s, \\{$idx, $x, $y, $y\\}], "
+     "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_3D_B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r),
+              "sust.b.3d.b8.clamp \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r),
+              "sust.b.3d.b16.clamp \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r),
+              "sust.b.3d.b32.clamp \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_B64_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int64Regs:$r),
+              "sust.b.3d.b64.clamp \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_V2B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.b.3d.v2.b8.clamp \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V2B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.b.3d.v2.b16.clamp \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V2B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r, Int32Regs:$g),
+              "sust.b.3d.v2.b32.clamp \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V2B64_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int64Regs:$r, Int64Regs:$g),
+              "sust.b.3d.v2.b64.clamp \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V4B8_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+         "sust.b.3d.v4.b8.clamp \t[$s, \\{$x, $y, $z, $z\\}], "
+         "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_3D_V4B16_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+        "sust.b.3d.v4.b16.clamp \t[$s, \\{$x, $y, $z, $z\\}], "
+        "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_3D_V4B32_CLAMP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+        "sust.b.3d.v4.b32.clamp \t[$s, \\{$x, $y, $z, $z\\}], "
+        "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+// .trap variant
+def SUST_B_1D_B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.1d.b8.trap \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.1d.b16.trap \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r),
+              "sust.b.1d.b32.trap \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_B64_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int64Regs:$r),
+              "sust.b.1d.b64.trap \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_V2B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+              "sust.b.1d.v2.b8.trap \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V2B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+              "sust.b.1d.v2.b16.trap \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V2B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+              "sust.b.1d.v2.b32.trap \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V2B64_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int64Regs:$r, Int64Regs:$g),
+              "sust.b.1d.v2.b64.trap \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V4B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g,
+                   Int16Regs:$b, Int16Regs:$a),
+              "sust.b.1d.v4.b8.trap \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_V4B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g,
+                   Int16Regs:$b, Int16Regs:$a),
+              "sust.b.1d.v4.b16.trap \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_V4B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g,
+                   Int32Regs:$b, Int32Regs:$a),
+              "sust.b.1d.v4.b32.trap \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_1D_ARRAY_B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.a1d.b8.trap \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.a1d.b16.trap \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r),
+              "sust.b.a1d.b32.trap \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_B64_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int64Regs:$r),
+              "sust.b.a1d.b64.trap \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.a1d.v2.b8.trap \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.a1d.v2.b16.trap \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r,
+                   Int32Regs:$g),
+              "sust.b.a1d.v2.b32.trap \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B64_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int64Regs:$r,
+                   Int64Regs:$g),
+              "sust.b.a1d.v2.b64.trap \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V4B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              "sust.b.a1d.v4.b8.trap \t[$s, \\{$idx, $x\\}], "
+              "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_ARRAY_V4B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+             "sust.b.a1d.v4.b16.trap \t[$s, \\{$idx, $x\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_ARRAY_V4B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r,
+                   Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+             "sust.b.a1d.v4.b32.trap \t[$s, \\{$idx, $x\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_2D_B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+              "sust.b.2d.b8.trap \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+              "sust.b.2d.b16.trap \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+              "sust.b.2d.b32.trap \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_B64_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r),
+              "sust.b.2d.b64.trap \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_V2B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.2d.v2.b8.trap \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V2B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.2d.v2.b16.trap \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V2B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+                   Int32Regs:$g),
+              "sust.b.2d.v2.b32.trap \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V2B64_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r,
+                   Int64Regs:$g),
+              "sust.b.2d.v2.b64.trap \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V4B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              "sust.b.2d.v4.b8.trap \t[$s, \\{$x, $y\\}], "
+              "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_V4B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+             "sust.b.2d.v4.b16.trap \t[$s, \\{$x, $y\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_V4B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+                   Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+             "sust.b.2d.v4.b32.trap \t[$s, \\{$x, $y\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_2D_ARRAY_B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r),
+              "sust.b.a2d.b8.trap \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r),
+              "sust.b.a2d.b16.trap \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r),
+              "sust.b.a2d.b32.trap \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_B64_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int64Regs:$r),
+              "sust.b.a2d.b64.trap \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.b.a2d.v2.b8.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g),
+             "sust.b.a2d.v2.b16.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r, Int32Regs:$g),
+             "sust.b.a2d.v2.b32.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B64_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int64Regs:$r, Int64Regs:$g),
+             "sust.b.a2d.v2.b64.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V4B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+      "sust.b.a2d.v4.b8.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+      "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_ARRAY_V4B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+     "sust.b.a2d.v4.b16.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+     "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_ARRAY_V4B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+     "sust.b.a2d.v4.b32.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+     "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_3D_B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r),
+              "sust.b.3d.b8.trap \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r),
+              "sust.b.3d.b16.trap \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r),
+              "sust.b.3d.b32.trap \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_B64_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int64Regs:$r),
+              "sust.b.3d.b64.trap \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_V2B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.b.3d.v2.b8.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V2B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.b.3d.v2.b16.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V2B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r, Int32Regs:$g),
+              "sust.b.3d.v2.b32.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V2B64_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int64Regs:$r, Int64Regs:$g),
+              "sust.b.3d.v2.b64.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V4B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+         "sust.b.3d.v4.b8.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+         "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_3D_V4B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+        "sust.b.3d.v4.b16.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+        "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_3D_V4B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+        "sust.b.3d.v4.b32.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+        "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+// .zero variant
+def SUST_B_1D_B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.1d.b8.zero \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.1d.b16.zero \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r),
+              "sust.b.1d.b32.zero \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_B64_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int64Regs:$r),
+              "sust.b.1d.b64.zero \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_V2B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+              "sust.b.1d.v2.b8.zero \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V2B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+              "sust.b.1d.v2.b16.zero \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V2B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+              "sust.b.1d.v2.b32.zero \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V2B64_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int64Regs:$r, Int64Regs:$g),
+              "sust.b.1d.v2.b64.zero \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_V4B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g,
+                   Int16Regs:$b, Int16Regs:$a),
+              "sust.b.1d.v4.b8.zero \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_V4B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g,
+                   Int16Regs:$b, Int16Regs:$a),
+              "sust.b.1d.v4.b16.zero \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_V4B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g,
+                   Int32Regs:$b, Int32Regs:$a),
+              "sust.b.1d.v4.b32.zero \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_1D_ARRAY_B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.a1d.b8.zero \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r),
+              "sust.b.a1d.b16.zero \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r),
+              "sust.b.a1d.b32.zero \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_B64_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int64Regs:$r),
+              "sust.b.a1d.b64.zero \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.a1d.v2.b8.zero \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.a1d.v2.b16.zero \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r,
+                   Int32Regs:$g),
+              "sust.b.a1d.v2.b32.zero \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V2B64_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int64Regs:$r,
+                   Int64Regs:$g),
+              "sust.b.a1d.v2.b64.zero \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_1D_ARRAY_V4B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              "sust.b.a1d.v4.b8.zero \t[$s, \\{$idx, $x\\}], "
+              "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_ARRAY_V4B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+             "sust.b.a1d.v4.b16.zero \t[$s, \\{$idx, $x\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_1D_ARRAY_V4B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r,
+                   Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+             "sust.b.a1d.v4.b32.zero \t[$s, \\{$idx, $x\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_2D_B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+              "sust.b.2d.b8.zero \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+              "sust.b.2d.b16.zero \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+              "sust.b.2d.b32.zero \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_B64_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r),
+              "sust.b.2d.b64.zero \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_V2B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.2d.v2.b8.zero \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V2B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.b.2d.v2.b16.zero \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V2B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+                   Int32Regs:$g),
+              "sust.b.2d.v2.b32.zero \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V2B64_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r,
+                   Int64Regs:$g),
+              "sust.b.2d.v2.b64.zero \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_B_2D_V4B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              "sust.b.2d.v4.b8.zero \t[$s, \\{$x, $y\\}], "
+              "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_V4B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+             "sust.b.2d.v4.b16.zero \t[$s, \\{$x, $y\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_V4B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+                   Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+             "sust.b.2d.v4.b32.zero \t[$s, \\{$x, $y\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_2D_ARRAY_B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r),
+              "sust.b.a2d.b8.zero \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r),
+              "sust.b.a2d.b16.zero \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r),
+              "sust.b.a2d.b32.zero \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_B64_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int64Regs:$r),
+              "sust.b.a2d.b64.zero \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.b.a2d.v2.b8.zero \t[$s, \\{$idx, $x, $y, $y\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g),
+             "sust.b.a2d.v2.b16.zero \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r, Int32Regs:$g),
+             "sust.b.a2d.v2.b32.zero \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V2B64_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int64Regs:$r, Int64Regs:$g),
+             "sust.b.a2d.v2.b64.zero \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_B_2D_ARRAY_V4B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+      "sust.b.a2d.v4.b8.zero \t[$s, \\{$idx, $x, $y, $y\\}], "
+      "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_ARRAY_V4B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+     "sust.b.a2d.v4.b16.zero \t[$s, \\{$idx, $x, $y, $y\\}], "
+     "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_2D_ARRAY_V4B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+     "sust.b.a2d.v4.b32.zero \t[$s, \\{$idx, $x, $y, $y\\}], "
+     "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_B_3D_B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r),
+              "sust.b.3d.b8.zero \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r),
+              "sust.b.3d.b16.zero \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r),
+              "sust.b.3d.b32.zero \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_B64_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int64Regs:$r),
+              "sust.b.3d.b64.zero \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_B_3D_V2B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.b.3d.v2.b8.zero \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V2B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.b.3d.v2.b16.zero \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V2B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r, Int32Regs:$g),
+              "sust.b.3d.v2.b32.zero \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V2B64_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int64Regs:$r, Int64Regs:$g),
+              "sust.b.3d.v2.b64.zero \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_B_3D_V4B8_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+         "sust.b.3d.v4.b8.zero \t[$s, \\{$x, $y, $z, $z\\}], "
+         "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_3D_V4B16_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+        "sust.b.3d.v4.b16.zero \t[$s, \\{$x, $y, $z, $z\\}], "
+        "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_B_3D_V4B32_ZERO
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+        "sust.b.3d.v4.b32.zero \t[$s, \\{$x, $y, $z, $z\\}], "
+        "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+
+// Formatted
+
+def SUST_P_1D_B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+              "sust.p.1d.b8.trap \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_P_1D_B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+              "sust.p.1d.b16.trap \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_P_1D_B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r),
+              "sust.p.1d.b32.trap \t[$s, \\{$x\\}], \\{$r\\};",
+              []>;
+def SUST_P_1D_V2B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+              "sust.p.1d.v2.b8.trap \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_P_1D_V2B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+              "sust.p.1d.v2.b16.trap \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_P_1D_V2B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+              "sust.p.1d.v2.b32.trap \t[$s, \\{$x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_P_1D_V4B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g,
+                   Int16Regs:$b, Int16Regs:$a),
+              "sust.p.1d.v4.b8.trap \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_P_1D_V4B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g,
+                   Int16Regs:$b, Int16Regs:$a),
+              "sust.p.1d.v4.b16.trap \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_P_1D_V4B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g,
+                   Int32Regs:$b, Int32Regs:$a),
+              "sust.p.1d.v4.b32.trap \t[$s, \\{$x\\}], \\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_P_1D_ARRAY_B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r),
+              "sust.p.a1d.b8.trap \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_P_1D_ARRAY_B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r),
+              "sust.p.a1d.b16.trap \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_P_1D_ARRAY_B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r),
+              "sust.p.a1d.b32.trap \t[$s, \\{$idx, $x\\}], \\{$r\\};",
+              []>;
+def SUST_P_1D_ARRAY_V2B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.p.a1d.v2.b8.trap \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_P_1D_ARRAY_V2B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.p.a1d.v2.b16.trap \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_P_1D_ARRAY_V2B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r,
+                   Int32Regs:$g),
+              "sust.p.a1d.v2.b32.trap \t[$s, \\{$idx, $x\\}], \\{$r, $g\\};",
+              []>;
+def SUST_P_1D_ARRAY_V4B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              "sust.p.a1d.v4.b8.trap \t[$s, \\{$idx, $x\\}], "
+              "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_P_1D_ARRAY_V4B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+             "sust.p.a1d.v4.b16.trap \t[$s, \\{$idx, $x\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_P_1D_ARRAY_V4B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$r,
+                   Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+             "sust.p.a1d.v4.b32.trap \t[$s, \\{$idx, $x\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_P_2D_B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+              "sust.p.2d.b8.trap \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_P_2D_B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+              "sust.p.2d.b16.trap \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_P_2D_B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+              "sust.p.2d.b32.trap \t[$s, \\{$x, $y\\}], \\{$r\\};",
+              []>;
+def SUST_P_2D_V2B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.p.2d.v2.b8.trap \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_P_2D_V2B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g),
+              "sust.p.2d.v2.b16.trap \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_P_2D_V2B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+                   Int32Regs:$g),
+              "sust.p.2d.v2.b32.trap \t[$s, \\{$x, $y\\}], \\{$r, $g\\};",
+              []>;
+def SUST_P_2D_V4B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+              "sust.p.2d.v4.b8.trap \t[$s, \\{$x, $y\\}], "
+              "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_P_2D_V4B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r,
+                   Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+             "sust.p.2d.v4.b16.trap \t[$s, \\{$x, $y\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_P_2D_V4B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+                   Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+             "sust.p.2d.v4.b32.trap \t[$s, \\{$x, $y\\}], "
+             "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_P_2D_ARRAY_B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r),
+              "sust.p.a2d.b8.trap \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_P_2D_ARRAY_B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r),
+              "sust.p.a2d.b16.trap \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_P_2D_ARRAY_B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r),
+              "sust.p.a2d.b32.trap \t[$s, \\{$idx, $x, $y, $y\\}], \\{$r\\};",
+              []>;
+def SUST_P_2D_ARRAY_V2B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.p.a2d.v2.b8.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_P_2D_ARRAY_V2B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g),
+             "sust.p.a2d.v2.b16.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_P_2D_ARRAY_V2B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r, Int32Regs:$g),
+             "sust.p.a2d.v2.b32.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+             "\\{$r, $g\\};",
+              []>;
+def SUST_P_2D_ARRAY_V4B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+      "sust.p.a2d.v4.b8.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+      "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_P_2D_ARRAY_V4B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+     "sust.p.a2d.v4.b16.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+     "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_P_2D_ARRAY_V4B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$idx, Int32Regs:$x, Int32Regs:$y,
+                   Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+     "sust.p.a2d.v4.b32.trap \t[$s, \\{$idx, $x, $y, $y\\}], "
+     "\\{$r, $g, $b, $a\\};",
+              []>;
+
+
+def SUST_P_3D_B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r),
+              "sust.p.3d.b8.trap \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_P_3D_B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r),
+              "sust.p.3d.b16.trap \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_P_3D_B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r),
+              "sust.p.3d.b32.trap \t[$s, \\{$x, $y, $z, $z\\}], \\{$r\\};",
+              []>;
+def SUST_P_3D_V2B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.p.3d.v2.b8.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_P_3D_V2B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g),
+              "sust.p.3d.v2.b16.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_P_3D_V2B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r, Int32Regs:$g),
+              "sust.p.3d.v2.b32.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+              "\\{$r, $g\\};",
+              []>;
+def SUST_P_3D_V4B8_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+         "sust.p.3d.v4.b8.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+         "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_P_3D_V4B16_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+        "sust.p.3d.v4.b16.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+        "\\{$r, $g, $b, $a\\};",
+              []>;
+def SUST_P_3D_V4B32_TRAP
+  : NVPTXInst<(outs),
+              (ins Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+                   Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+        "sust.p.3d.v4.b32.trap \t[$s, \\{$x, $y, $z, $z\\}], "
+        "\\{$r, $g, $b, $a\\};",
+              []>;
+}
+
+// Surface store instruction patterns
+// I'm not sure why we can't just include these in the instruction definitions,
+// but TableGen complains of type errors :(
+
+// .clamp variant
+def : Pat<(int_nvvm_sust_b_1d_i8_clamp
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_B8_CLAMP Int64Regs:$s, Int32Regs:$x, Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_i16_clamp
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_B16_CLAMP Int64Regs:$s, Int32Regs:$x, Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_i32_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$r),
+          (SUST_B_1D_B32_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_i64_clamp
+           Int64Regs:$s, Int32Regs:$x, Int64Regs:$r),
+          (SUST_B_1D_B64_CLAMP Int64Regs:$s, Int32Regs:$x, Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i8_clamp
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_V2B8_CLAMP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i16_clamp
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_V2B16_CLAMP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i32_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_1D_V2B32_CLAMP Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i64_clamp
+           Int64Regs:$s, Int32Regs:$x, Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_1D_V2B64_CLAMP Int64Regs:$s, Int32Regs:$x,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v4i8_clamp
+           Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_V4B8_CLAMP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v4i16_clamp
+           Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_V4B16_CLAMP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v4i32_clamp
+           Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_1D_V4B32_CLAMP Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_1d_array_i8_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_ARRAY_B8_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_i16_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_ARRAY_B16_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_i32_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$r),
+          (SUST_B_1D_ARRAY_B32_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_i64_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int64Regs:$r),
+          (SUST_B_1D_ARRAY_B64_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i8_clamp
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_ARRAY_V2B8_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i16_clamp
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_ARRAY_V2B16_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i32_clamp
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_1D_ARRAY_V2B32_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i64_clamp
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_1D_ARRAY_V2B64_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v4i8_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_ARRAY_V4B8_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v4i16_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_ARRAY_V4B16_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v4i32_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_1D_ARRAY_V4B32_CLAMP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_2d_i8_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_B8_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_i16_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_B16_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_i32_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+          (SUST_B_2D_B32_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_i64_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r),
+          (SUST_B_2D_B64_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i8_clamp
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_V2B8_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i16_clamp
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_V2B16_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i32_clamp
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_2D_V2B32_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i64_clamp
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_2D_V2B64_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v4i8_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_V4B8_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v4i16_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_V4B16_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v4i32_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_2D_V4B32_CLAMP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_2d_array_i8_clamp
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_ARRAY_B8_CLAMP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_i16_clamp
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_ARRAY_B16_CLAMP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_i32_clamp
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+          (SUST_B_2D_ARRAY_B32_CLAMP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_i64_clamp
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r),
+          (SUST_B_2D_ARRAY_B64_CLAMP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i8_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_ARRAY_V2B8_CLAMP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i16_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_ARRAY_V2B16_CLAMP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i32_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+           Int32Regs:$g),
+          (SUST_B_2D_ARRAY_V2B32_CLAMP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i64_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r,
+           Int64Regs:$g),
+          (SUST_B_2D_ARRAY_V2B64_CLAMP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y, Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v4i8_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_ARRAY_V4B8_CLAMP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v4i16_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_ARRAY_V4B16_CLAMP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v4i32_clamp
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_2D_ARRAY_V4B32_CLAMP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_3d_i8_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r),
+          (SUST_B_3D_B8_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_i16_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r),
+          (SUST_B_3D_B16_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_i32_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r),
+          (SUST_B_3D_B32_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_i64_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r),
+          (SUST_B_3D_B64_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i8_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_3D_V2B8_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i16_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_3D_V2B16_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i32_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_3D_V2B32_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i64_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_3D_V2B64_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v4i8_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_3D_V4B8_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v4i16_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_3D_V4B16_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v4i32_clamp
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_3D_V4B32_CLAMP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+// .trap variant
+def : Pat<(int_nvvm_sust_b_1d_i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_B8_TRAP Int64Regs:$s, Int32Regs:$x, Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_B16_TRAP Int64Regs:$s, Int32Regs:$x, Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$r),
+          (SUST_B_1D_B32_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_i64_trap
+           Int64Regs:$s, Int32Regs:$x, Int64Regs:$r),
+          (SUST_B_1D_B64_TRAP Int64Regs:$s, Int32Regs:$x, Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_V2B8_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_V2B16_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_1D_V2B32_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i64_trap
+           Int64Regs:$s, Int32Regs:$x, Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_1D_V2B64_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v4i8_trap
+           Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_V4B8_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v4i16_trap
+           Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_V4B16_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v4i32_trap
+           Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_1D_V4B32_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_1d_array_i8_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_ARRAY_B8_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_i16_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_ARRAY_B16_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_i32_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$r),
+          (SUST_B_1D_ARRAY_B32_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_i64_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int64Regs:$r),
+          (SUST_B_1D_ARRAY_B64_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i8_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_ARRAY_V2B8_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i16_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_ARRAY_V2B16_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i32_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_1D_ARRAY_V2B32_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i64_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_1D_ARRAY_V2B64_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v4i8_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_ARRAY_V4B8_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v4i16_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_ARRAY_V4B16_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v4i32_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_1D_ARRAY_V4B32_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_2d_i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_B8_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_B16_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+          (SUST_B_2D_B32_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_i64_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r),
+          (SUST_B_2D_B64_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i8_trap
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_V2B8_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i16_trap
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_V2B16_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i32_trap
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_2D_V2B32_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i64_trap
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_2D_V2B64_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v4i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_V4B8_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v4i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_V4B16_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v4i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_2D_V4B32_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_2d_array_i8_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_ARRAY_B8_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_i16_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_ARRAY_B16_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_i32_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+          (SUST_B_2D_ARRAY_B32_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_i64_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r),
+          (SUST_B_2D_ARRAY_B64_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i8_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_ARRAY_V2B8_TRAP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i16_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_ARRAY_V2B16_TRAP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i32_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+           Int32Regs:$g),
+          (SUST_B_2D_ARRAY_V2B32_TRAP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i64_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r,
+           Int64Regs:$g),
+          (SUST_B_2D_ARRAY_V2B64_TRAP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y, Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v4i8_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_ARRAY_V4B8_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v4i16_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_ARRAY_V4B16_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v4i32_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_2D_ARRAY_V4B32_TRAP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_3d_i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r),
+          (SUST_B_3D_B8_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r),
+          (SUST_B_3D_B16_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r),
+          (SUST_B_3D_B32_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_i64_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r),
+          (SUST_B_3D_B64_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_3D_V2B8_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_3D_V2B16_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_3D_V2B32_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i64_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_3D_V2B64_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v4i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_3D_V4B8_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v4i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_3D_V4B16_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v4i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_3D_V4B32_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+// .zero variant
+def : Pat<(int_nvvm_sust_b_1d_i8_zero
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_B8_ZERO Int64Regs:$s, Int32Regs:$x, Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_i16_zero
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_B16_ZERO Int64Regs:$s, Int32Regs:$x, Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_i32_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$r),
+          (SUST_B_1D_B32_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_i64_zero
+           Int64Regs:$s, Int32Regs:$x, Int64Regs:$r),
+          (SUST_B_1D_B64_ZERO Int64Regs:$s, Int32Regs:$x, Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i8_zero
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_V2B8_ZERO Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i16_zero
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_V2B16_ZERO Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i32_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_1D_V2B32_ZERO Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v2i64_zero
+           Int64Regs:$s, Int32Regs:$x, Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_1D_V2B64_ZERO Int64Regs:$s, Int32Regs:$x,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v4i8_zero
+           Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_V4B8_ZERO Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v4i16_zero
+           Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_V4B16_ZERO Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_v4i32_zero
+           Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_1D_V4B32_ZERO Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_1d_array_i8_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_ARRAY_B8_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_i16_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r),
+          (SUST_B_1D_ARRAY_B16_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_i32_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$r),
+          (SUST_B_1D_ARRAY_B32_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_i64_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int64Regs:$r),
+          (SUST_B_1D_ARRAY_B64_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i8_zero
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_ARRAY_V2B8_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i16_zero
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_1D_ARRAY_V2B16_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i32_zero
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_1D_ARRAY_V2B32_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v2i64_zero
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_1D_ARRAY_V2B64_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v4i8_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_ARRAY_V4B8_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v4i16_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_1D_ARRAY_V4B16_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_1d_array_v4i32_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_1D_ARRAY_V4B32_ZERO Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_2d_i8_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_B8_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_i16_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_B16_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_i32_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+          (SUST_B_2D_B32_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_i64_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r),
+          (SUST_B_2D_B64_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i8_zero
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_V2B8_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i16_zero
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_V2B16_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i32_zero
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_2D_V2B32_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v2i64_zero
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_2D_V2B64_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v4i8_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_V4B8_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v4i16_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_V4B16_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_v4i32_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_2D_V4B32_ZERO Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_2d_array_i8_zero
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_ARRAY_B8_ZERO Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_i16_zero
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_B_2D_ARRAY_B16_ZERO Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_i32_zero
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+          (SUST_B_2D_ARRAY_B32_ZERO Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_i64_zero
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r),
+          (SUST_B_2D_ARRAY_B64_ZERO Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i8_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_ARRAY_V2B8_ZERO Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i16_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_2D_ARRAY_V2B16_ZERO Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i32_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+           Int32Regs:$g),
+          (SUST_B_2D_ARRAY_V2B32_ZERO Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v2i64_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int64Regs:$r,
+           Int64Regs:$g),
+          (SUST_B_2D_ARRAY_V2B64_ZERO Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y, Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v4i8_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_ARRAY_V4B8_ZERO Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v4i16_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_2D_ARRAY_V4B16_ZERO Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_2d_array_v4i32_zero
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_2D_ARRAY_V4B32_ZERO Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_b_3d_i8_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r),
+          (SUST_B_3D_B8_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_i16_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r),
+          (SUST_B_3D_B16_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_i32_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r),
+          (SUST_B_3D_B32_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_i64_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r),
+          (SUST_B_3D_B64_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i8_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_3D_V2B8_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i16_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_B_3D_V2B16_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i32_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g),
+          (SUST_B_3D_V2B32_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v2i64_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r, Int64Regs:$g),
+          (SUST_B_3D_V2B64_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int64Regs:$r, Int64Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v4i8_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_3D_V4B8_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v4i16_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_B_3D_V4B16_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_b_3d_v4i32_zero
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_B_3D_V4B32_ZERO Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+
+def : Pat<(int_nvvm_sust_p_1d_i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+          (SUST_P_1D_B8_TRAP Int64Regs:$s, Int32Regs:$x, Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_1d_i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r),
+          (SUST_P_1D_B16_TRAP Int64Regs:$s, Int32Regs:$x, Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_1d_i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$r),
+          (SUST_P_1D_B32_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_1d_v2i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_P_1D_V2B8_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_1d_v2i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_P_1D_V2B16_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_1d_v2i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+          (SUST_P_1D_V2B32_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_1d_v4i8_trap
+           Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_P_1D_V4B8_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_p_1d_v4i16_trap
+           Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_P_1D_V4B16_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_p_1d_v4i32_trap
+           Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_P_1D_V4B32_TRAP Int64Regs:$s, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_p_1d_array_i8_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r),
+          (SUST_P_1D_ARRAY_B8_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_1d_array_i16_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r),
+          (SUST_P_1D_ARRAY_B16_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_1d_array_i32_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$r),
+          (SUST_P_1D_ARRAY_B32_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_1d_array_v2i8_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_P_1D_ARRAY_V2B8_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_1d_array_v2i16_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int16Regs:$r, Int16Regs:$g),
+          (SUST_P_1D_ARRAY_V2B16_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_1d_array_v2i32_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$r, Int32Regs:$g),
+          (SUST_P_1D_ARRAY_V2B32_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_1d_array_v4i8_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_P_1D_ARRAY_V4B8_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_p_1d_array_v4i16_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_P_1D_ARRAY_V4B16_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_p_1d_array_v4i32_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_P_1D_ARRAY_V4B32_TRAP Int64Regs:$s, Int32Regs:$l, Int32Regs:$x,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_p_2d_i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_P_2D_B8_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_2d_i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_P_2D_B16_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_2d_i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+          (SUST_P_2D_B32_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_2d_v2i8_trap
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r, Int16Regs:$g),
+          (SUST_P_2D_V2B8_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_2d_v2i16_trap
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r, Int16Regs:$g),
+          (SUST_P_2D_V2B16_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_2d_v2i32_trap
+          Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r, Int32Regs:$g),
+          (SUST_P_2D_V2B32_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_2d_v4i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_P_2D_V4B8_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_p_2d_v4i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_P_2D_V4B16_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_p_2d_v4i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_P_2D_V4B32_TRAP Int64Regs:$s, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_p_2d_array_i8_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_P_2D_ARRAY_B8_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_2d_array_i16_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int16Regs:$r),
+          (SUST_P_2D_ARRAY_B16_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_2d_array_i32_trap
+          Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r),
+          (SUST_P_2D_ARRAY_B32_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_2d_array_v2i8_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_P_2D_ARRAY_V2B8_TRAP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_2d_array_v2i16_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_P_2D_ARRAY_V2B16_TRAP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_2d_array_v2i32_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y, Int32Regs:$r,
+           Int32Regs:$g),
+          (SUST_P_2D_ARRAY_V2B32_TRAP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_2d_array_v4i8_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_P_2D_ARRAY_V4B8_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_p_2d_array_v4i16_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_P_2D_ARRAY_V4B16_TRAP Int64Regs:$s,
+           Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_p_2d_array_v4i32_trap
+           Int64Regs:$s, Int32Regs:$l, Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_P_2D_ARRAY_V4B32_TRAP Int64Regs:$s, Int32Regs:$l,
+           Int32Regs:$x, Int32Regs:$y,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
+
+def : Pat<(int_nvvm_sust_p_3d_i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r),
+          (SUST_P_3D_B8_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_3d_i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r),
+          (SUST_P_3D_B16_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_3d_i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r),
+          (SUST_P_3D_B32_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r)>;
+
+def : Pat<(int_nvvm_sust_p_3d_v2i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_P_3D_V2B8_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_3d_v2i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g),
+          (SUST_P_3D_V2B16_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_3d_v2i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g),
+          (SUST_P_3D_V2B32_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g)>;
+
+def : Pat<(int_nvvm_sust_p_3d_v4i8_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_P_3D_V4B8_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_p_3d_v4i16_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a),
+          (SUST_P_3D_V4B16_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int16Regs:$r, Int16Regs:$g, Int16Regs:$b, Int16Regs:$a)>;
+
+def : Pat<(int_nvvm_sust_p_3d_v4i32_trap
+           Int64Regs:$s, Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a),
+          (SUST_P_3D_V4B32_TRAP Int64Regs:$s,
+           Int32Regs:$x, Int32Regs:$y, Int32Regs:$z,
+           Int32Regs:$r, Int32Regs:$g, Int32Regs:$b, Int32Regs:$a)>;
+
+
 
 //===-- Old PTX Back-end Intrinsics ---------------------------------------===//
 
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
index 7c257b4c6a89..f0c3663551bc 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
@@ -16,12 +16,12 @@
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/InstIterator.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/LLVMContext.h"
 #include "llvm/IR/Module.h"
-#include "llvm/Support/InstIterator.h"
 
 using namespace llvm;
 
@@ -104,7 +104,7 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
   SmallVector<MemTransferInst *, 4> aggrMemcpys;
   SmallVector<MemSetInst *, 4> aggrMemsets;
 
-  DataLayout *TD = &getAnalysis<DataLayout>();
+  const DataLayout *DL = &getAnalysis<DataLayoutPass>().getDataLayout();
   LLVMContext &Context = F.getParent()->getContext();
 
   //
@@ -120,10 +120,10 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
         if (load->hasOneUse() == false)
           continue;
 
-        if (TD->getTypeStoreSize(load->getType()) < MaxAggrCopySize)
+        if (DL->getTypeStoreSize(load->getType()) < MaxAggrCopySize)
           continue;
 
-        User *use = *(load->use_begin());
+        User *use = load->user_back();
         if (StoreInst *store = dyn_cast<StoreInst>(use)) {
           if (store->getOperand(0) != load) //getValueOperand
             continue;
@@ -163,10 +163,10 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
   //
   for (unsigned i = 0, e = aggrLoads.size(); i != e; ++i) {
     LoadInst *load = aggrLoads[i];
-    StoreInst *store = dyn_cast<StoreInst>(*load->use_begin());
+    StoreInst *store = dyn_cast<StoreInst>(*load->user_begin());
     Value *srcAddr = load->getOperand(0);
     Value *dstAddr = store->getOperand(1);
-    unsigned numLoads = TD->getTypeStoreSize(load->getType());
+    unsigned numLoads = DL->getTypeStoreSize(load->getType());
     Value *len = ConstantInt::get(Type::getInt32Ty(Context), numLoads);
 
     convertTransferToLoop(store, srcAddr, dstAddr, len, load->isVolatile(),
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.h b/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.h
index 286e753fa92b..5ec1fc969687 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.h
@@ -27,16 +27,17 @@ struct NVPTXLowerAggrCopies : public FunctionPass {
 
   NVPTXLowerAggrCopies() : FunctionPass(ID) {}
 
-  void getAnalysisUsage(AnalysisUsage &AU) const {
-    AU.addRequired<DataLayout>();
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.addRequired<DataLayoutPass>();
+    AU.addPreserved("stack-protector");
     AU.addPreserved<MachineFunctionAnalysis>();
   }
 
-  virtual bool runOnFunction(Function &F);
+  bool runOnFunction(Function &F) override;
 
   static const unsigned MaxAggrCopySize = 128;
 
-  virtual const char *getPassName() const {
+  const char *getPassName() const override {
     return "Lower aggregate copies/intrinsics into loops";
   }
 };
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXMCExpr.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXMCExpr.cpp
index ca247642048a..137248b19a68 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXMCExpr.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXMCExpr.cpp
@@ -7,13 +7,14 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "nvptx-mcexpr"
 #include "NVPTXMCExpr.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/MC/MCAssembler.h"
 #include "llvm/MC/MCContext.h"
 using namespace llvm;
 
+#define DEBUG_TYPE "nvptx-mcexpr"
+
 const NVPTXFloatMCExpr*
 NVPTXFloatMCExpr::Create(VariantKind Kind, APFloat Flt, MCContext &Ctx) {
   return new (Ctx) NVPTXFloatMCExpr(Kind, Flt);
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXMCExpr.h b/contrib/llvm/lib/Target/NVPTX/NVPTXMCExpr.h
index 0efb231d7d2e..554764930a9e 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXMCExpr.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXMCExpr.h
@@ -61,18 +61,18 @@ public:
 
 /// @}
 
-  void PrintImpl(raw_ostream &OS) const;
+  void PrintImpl(raw_ostream &OS) const override;
   bool EvaluateAsRelocatableImpl(MCValue &Res,
-                                 const MCAsmLayout *Layout) const {
+                                 const MCAsmLayout *Layout) const override {
     return false;
   }
-  void AddValueSymbols(MCAssembler *) const {};
-  const MCSection *FindAssociatedSection() const {
-    return NULL;
+  void visitUsedExpr(MCStreamer &Streamer) const override {};
+  const MCSection *FindAssociatedSection() const override {
+    return nullptr;
   }
 
   // There are no TLS NVPTXMCExprs at the moment.
-  void fixELFSymbolsInTLSFixups(MCAssembler &Asm) const {}
+  void fixELFSymbolsInTLSFixups(MCAssembler &Asm) const override {}
 
   static bool classof(const MCExpr *E) {
     return E->getKind() == MCExpr::Target;
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXMachineFunctionInfo.h b/contrib/llvm/lib/Target/NVPTX/NVPTXMachineFunctionInfo.h
new file mode 100644
index 000000000000..67fb39050797
--- /dev/null
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXMachineFunctionInfo.h
@@ -0,0 +1,46 @@
+//===-- NVPTXMachineFunctionInfo.h - NVPTX-specific Function Info  --------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source 
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class is attached to a MachineFunction instance and tracks target-
+// dependent information
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/CodeGen/MachineFunction.h"
+
+namespace llvm {
+class NVPTXMachineFunctionInfo : public MachineFunctionInfo {
+private:
+  /// Stores a mapping from index to symbol name for removing image handles
+  /// on Fermi.
+  SmallVector<std::string, 8> ImageHandleList;
+
+public:
+  NVPTXMachineFunctionInfo(MachineFunction &MF) {}
+
+  /// Returns the index for the symbol \p Symbol. If the symbol was previously,
+  /// added, the same index is returned. Otherwise, the symbol is added and the
+  /// new index is returned.
+  unsigned getImageHandleSymbolIndex(const char *Symbol) {
+    // Is the symbol already present?
+    for (unsigned i = 0, e = ImageHandleList.size(); i != e; ++i)
+      if (ImageHandleList[i] == std::string(Symbol))
+        return i;
+    // Nope, insert it
+    ImageHandleList.push_back(Symbol);
+    return ImageHandleList.size()-1;
+  }
+
+  /// Returns the symbol name at the given index.
+  const char *getImageHandleSymbol(unsigned Idx) const {
+    assert(ImageHandleList.size() > Idx && "Bad index");
+    return ImageHandleList[Idx].c_str();
+  }
+};
+}
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXPrologEpilogPass.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXPrologEpilogPass.cpp
index 843ebed5e4a4..348ab0c4bf14 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXPrologEpilogPass.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXPrologEpilogPass.cpp
@@ -14,24 +14,26 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTX.h"
-#include "llvm/Pass.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetRegisterInfo.h"
+#include "llvm/Pass.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 
 using namespace llvm;
 
+#define DEBUG_TYPE "nvptx-prolog-epilog"
+
 namespace {
 class NVPTXPrologEpilogPass : public MachineFunctionPass {
 public:
   static char ID;
   NVPTXPrologEpilogPass() : MachineFunctionPass(ID) {}
 
-  virtual bool runOnMachineFunction(MachineFunction &MF);
+  bool runOnMachineFunction(MachineFunction &MF) override;
 
 private:
   void calculateFrameObjectOffsets(MachineFunction &Fn);
@@ -58,7 +60,7 @@ bool NVPTXPrologEpilogPass::runOnMachineFunction(MachineFunction &MF) {
       for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
         if (!MI->getOperand(i).isFI())
           continue;
-        TRI.eliminateFrameIndex(MI, 0, i, NULL);
+        TRI.eliminateFrameIndex(MI, 0, i, nullptr);
         Modified = true;
       }
     }
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.cpp
index 4d3a1d9b40ce..358ccce39818 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.cpp
@@ -11,8 +11,6 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "nvptx-reg-info"
-
 #include "NVPTXRegisterInfo.h"
 #include "NVPTX.h"
 #include "NVPTXSubtarget.h"
@@ -25,6 +23,8 @@
 
 using namespace llvm;
 
+#define DEBUG_TYPE "nvptx-reg-info"
+
 namespace llvm {
 std::string getNVPTXRegClassName(TargetRegisterClass const *RC) {
   if (RC == &NVPTX::Float32RegsRegClass) {
@@ -53,9 +53,9 @@ std::string getNVPTXRegClassStr(TargetRegisterClass const *RC) {
     return "%f";
   }
   if (RC == &NVPTX::Float64RegsRegClass) {
-    return "%fl";
+    return "%fd";
   } else if (RC == &NVPTX::Int64RegsRegClass) {
-    return "%rl";
+    return "%rd";
   } else if (RC == &NVPTX::Int32RegsRegClass) {
     return "%r";
   } else if (RC == &NVPTX::Int16RegsRegClass) {
@@ -78,19 +78,12 @@ NVPTXRegisterInfo::NVPTXRegisterInfo(const NVPTXSubtarget &st)
 #include "NVPTXGenRegisterInfo.inc"
 
 /// NVPTX Callee Saved Registers
-const uint16_t *
+const MCPhysReg *
 NVPTXRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
-  static const uint16_t CalleeSavedRegs[] = { 0 };
+  static const MCPhysReg CalleeSavedRegs[] = { 0 };
   return CalleeSavedRegs;
 }
 
-// NVPTX Callee Saved Reg Classes
-const TargetRegisterClass *const *
-NVPTXRegisterInfo::getCalleeSavedRegClasses(const MachineFunction *MF) const {
-  static const TargetRegisterClass *const CalleeSavedRegClasses[] = { 0 };
-  return CalleeSavedRegClasses;
-}
-
 BitVector NVPTXRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   BitVector Reserved(getNumRegs());
   return Reserved;
@@ -113,12 +106,6 @@ void NVPTXRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset);
 }
 
-int NVPTXRegisterInfo::getDwarfRegNum(unsigned RegNum, bool isEH) const {
-  return 0;
-}
-
 unsigned NVPTXRegisterInfo::getFrameRegister(const MachineFunction &MF) const {
   return NVPTX::VRFrame;
 }
-
-unsigned NVPTXRegisterInfo::getRARegister() const { return 0; }
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.h b/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.h
index 0a20f29ccb1e..a7594be121a0 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.h
@@ -16,11 +16,10 @@
 
 #include "ManagedStringPool.h"
 #include "llvm/Target/TargetRegisterInfo.h"
+#include <sstream>
 
 #define GET_REGINFO_HEADER
 #include "NVPTXGenRegisterInfo.inc"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include <sstream>
 
 namespace llvm {
 
@@ -42,22 +41,16 @@ public:
   //------------------------------------------------------
 
   // NVPTX callee saved registers
-  virtual const uint16_t *
-  getCalleeSavedRegs(const MachineFunction *MF = 0) const;
-
-  // NVPTX callee saved register classes
-  virtual const TargetRegisterClass *const *
-  getCalleeSavedRegClasses(const MachineFunction *MF) const;
+  const MCPhysReg *
+  getCalleeSavedRegs(const MachineFunction *MF = nullptr) const override;
 
-  virtual BitVector getReservedRegs(const MachineFunction &MF) const;
+  BitVector getReservedRegs(const MachineFunction &MF) const override;
 
-  virtual void eliminateFrameIndex(MachineBasicBlock::iterator MI, int SPAdj,
-                                   unsigned FIOperandNum,
-                                   RegScavenger *RS = NULL) const;
+  void eliminateFrameIndex(MachineBasicBlock::iterator MI, int SPAdj,
+                           unsigned FIOperandNum,
+                           RegScavenger *RS = nullptr) const override;
 
-  virtual int getDwarfRegNum(unsigned RegNum, bool isEH) const;
-  virtual unsigned getFrameRegister(const MachineFunction &MF) const;
-  virtual unsigned getRARegister() const;
+  unsigned getFrameRegister(const MachineFunction &MF) const override;
 
   ManagedStringPool *getStrPool() const {
     return const_cast<ManagedStringPool *>(&ManagedStrPool);
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.td b/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.td
index 7a38a66b9227..efcee6b6f2bd 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.td
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXRegisterInfo.td
@@ -35,9 +35,9 @@ foreach i = 0-4 in {
   def P#i  : NVPTXReg<"%p"#i>;  // Predicate
   def RS#i : NVPTXReg<"%rs"#i>; // 16-bit
   def R#i  : NVPTXReg<"%r"#i>;  // 32-bit
-  def RL#i : NVPTXReg<"%rl"#i>; // 64-bit
+  def RL#i : NVPTXReg<"%rd"#i>; // 64-bit
   def F#i  : NVPTXReg<"%f"#i>;  // 32-bit float
-  def FL#i : NVPTXReg<"%fl"#i>; // 64-bit float
+  def FL#i : NVPTXReg<"%fd"#i>; // 64-bit float
 
   // Arguments
   def ia#i : NVPTXReg<"%ia"#i>;
@@ -46,6 +46,10 @@ foreach i = 0-4 in {
   def da#i : NVPTXReg<"%da"#i>;
 }
 
+foreach i = 0-31 in {
+  def ENVREG#i : NVPTXReg<"%envreg"#i>;
+}
+
 //===----------------------------------------------------------------------===//
 //  Register classes
 //===----------------------------------------------------------------------===//
@@ -61,4 +65,5 @@ def Float32ArgRegs : NVPTXRegClass<[f32], 32, (add (sequence "fa%u", 0, 4))>;
 def Float64ArgRegs : NVPTXRegClass<[f64], 64, (add (sequence "da%u", 0, 4))>;
 
 // Read NVPTXRegisterInfo.cpp to see how VRFrame and VRDepot are used.
-def SpecialRegs : NVPTXRegClass<[i32], 32, (add VRFrame, VRDepot)>;
+def SpecialRegs : NVPTXRegClass<[i32], 32, (add VRFrame, VRDepot,
+                                            (sequence "ENVREG%u", 0, 31))>;
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXReplaceImageHandles.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXReplaceImageHandles.cpp
new file mode 100644
index 000000000000..20d4e272341e
--- /dev/null
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXReplaceImageHandles.cpp
@@ -0,0 +1,189 @@
+//===-- NVPTXReplaceImageHandles.cpp - Replace image handles for Fermi ----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source 
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// On Fermi, image handles are not supported. To work around this, we traverse
+// the machine code and replace image handles with concrete symbols. For this
+// to work reliably, inlining of all function call must be performed.
+//
+//===----------------------------------------------------------------------===//
+
+#include "NVPTX.h"
+#include "NVPTXMachineFunctionInfo.h"
+#include "NVPTXSubtarget.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/DenseSet.h"
+
+using namespace llvm;
+
+namespace {
+class NVPTXReplaceImageHandles : public MachineFunctionPass {
+private:
+  static char ID;
+  DenseSet<MachineInstr *> InstrsToRemove;
+
+public:
+  NVPTXReplaceImageHandles();
+
+  bool runOnMachineFunction(MachineFunction &MF);
+
+  virtual const char *getPassName() const {
+    return "NVPTX Replace Image Handles";
+  }
+private:
+  bool processInstr(MachineInstr &MI);
+  void replaceImageHandle(MachineOperand &Op, MachineFunction &MF);
+  bool findIndexForHandle(MachineOperand &Op, MachineFunction &MF,
+                          unsigned &Idx);
+};
+}
+
+char NVPTXReplaceImageHandles::ID = 0;
+
+NVPTXReplaceImageHandles::NVPTXReplaceImageHandles()
+  : MachineFunctionPass(ID) {}
+
+bool NVPTXReplaceImageHandles::runOnMachineFunction(MachineFunction &MF) {
+  bool Changed = false;
+  InstrsToRemove.clear();
+
+  for (MachineFunction::iterator BI = MF.begin(), BE = MF.end(); BI != BE;
+       ++BI) {
+    for (MachineBasicBlock::iterator I = (*BI).begin(), E = (*BI).end();
+         I != E; ++I) {
+      MachineInstr &MI = *I;
+      Changed |= processInstr(MI);
+    }
+  }
+
+  // Now clean up any handle-access instructions
+  // This is needed in debug mode when code cleanup passes are not executed,
+  // but we need the handle access to be eliminated because they are not
+  // valid instructions when image handles are disabled.
+  for (DenseSet<MachineInstr *>::iterator I = InstrsToRemove.begin(),
+       E = InstrsToRemove.end(); I != E; ++I) {
+    (*I)->eraseFromParent();
+  }
+  return Changed;
+}
+
+bool NVPTXReplaceImageHandles::processInstr(MachineInstr &MI) {
+  MachineFunction &MF = *MI.getParent()->getParent();
+  const MCInstrDesc &MCID = MI.getDesc();
+
+  if (MCID.TSFlags & NVPTXII::IsTexFlag) {
+    // This is a texture fetch, so operand 4 is a texref and operand 5 is
+    // a samplerref
+    MachineOperand &TexHandle = MI.getOperand(4);
+    replaceImageHandle(TexHandle, MF);
+
+    if (!(MCID.TSFlags & NVPTXII::IsTexModeUnifiedFlag)) {
+      MachineOperand &SampHandle = MI.getOperand(5);
+      replaceImageHandle(SampHandle, MF);
+    }
+
+    return true;
+  } else if (MCID.TSFlags & NVPTXII::IsSuldMask) {
+    unsigned VecSize =
+      1 << (((MCID.TSFlags & NVPTXII::IsSuldMask) >> NVPTXII::IsSuldShift) - 1);
+
+    // For a surface load of vector size N, the Nth operand will be the surfref
+    MachineOperand &SurfHandle = MI.getOperand(VecSize);
+
+    replaceImageHandle(SurfHandle, MF);
+
+    return true;
+  } else if (MCID.TSFlags & NVPTXII::IsSustFlag) {
+    // This is a surface store, so operand 0 is a surfref
+    MachineOperand &SurfHandle = MI.getOperand(0);
+
+    replaceImageHandle(SurfHandle, MF);
+
+    return true;
+  } else if (MCID.TSFlags & NVPTXII::IsSurfTexQueryFlag) {
+    // This is a query, so operand 1 is a surfref/texref
+    MachineOperand &Handle = MI.getOperand(1);
+
+    replaceImageHandle(Handle, MF);
+
+    return true; 
+  }
+
+  return false;
+}
+
+void NVPTXReplaceImageHandles::
+replaceImageHandle(MachineOperand &Op, MachineFunction &MF) {
+  unsigned Idx;
+  if (findIndexForHandle(Op, MF, Idx)) {
+    Op.ChangeToImmediate(Idx);
+  }
+}
+
+bool NVPTXReplaceImageHandles::
+findIndexForHandle(MachineOperand &Op, MachineFunction &MF, unsigned &Idx) {
+  const MachineRegisterInfo &MRI = MF.getRegInfo();
+  NVPTXMachineFunctionInfo *MFI = MF.getInfo<NVPTXMachineFunctionInfo>();
+
+  assert(Op.isReg() && "Handle is not in a reg?");
+
+  // Which instruction defines the handle?
+  MachineInstr &TexHandleDef = *MRI.getVRegDef(Op.getReg());
+
+  switch (TexHandleDef.getOpcode()) {
+  case NVPTX::LD_i64_avar: {
+    // The handle is a parameter value being loaded, replace with the
+    // parameter symbol
+    const NVPTXSubtarget &ST = MF.getTarget().getSubtarget<NVPTXSubtarget>();
+    if (ST.getDrvInterface() == NVPTX::CUDA) {
+      // For CUDA, we preserve the param loads coming from function arguments
+      return false;
+    }
+
+    assert(TexHandleDef.getOperand(6).isSymbol() && "Load is not a symbol!");
+    StringRef Sym = TexHandleDef.getOperand(6).getSymbolName();
+    std::string ParamBaseName = MF.getName();
+    ParamBaseName += "_param_";
+    assert(Sym.startswith(ParamBaseName) && "Invalid symbol reference");
+    unsigned Param = atoi(Sym.data()+ParamBaseName.size());
+    std::string NewSym;
+    raw_string_ostream NewSymStr(NewSym);
+    NewSymStr << MF.getFunction()->getName() << "_param_" << Param;
+
+    InstrsToRemove.insert(&TexHandleDef);
+    Idx = MFI->getImageHandleSymbolIndex(NewSymStr.str().c_str());
+    return true;
+  }
+  case NVPTX::texsurf_handles: {
+    // The handle is a global variable, replace with the global variable name
+    assert(TexHandleDef.getOperand(1).isGlobal() && "Load is not a global!");
+    const GlobalValue *GV = TexHandleDef.getOperand(1).getGlobal();
+    assert(GV->hasName() && "Global sampler must be named!");
+    InstrsToRemove.insert(&TexHandleDef);
+    Idx = MFI->getImageHandleSymbolIndex(GV->getName().data());
+    return true;
+  }
+  case NVPTX::nvvm_move_i64:
+  case TargetOpcode::COPY: {
+    bool Res = findIndexForHandle(TexHandleDef.getOperand(1), MF, Idx);
+    if (Res) {
+      InstrsToRemove.insert(&TexHandleDef);
+    }
+    return Res;
+  }
+  default:
+    llvm_unreachable("Unknown instruction operating on handle");
+  }
+}
+
+MachineFunctionPass *llvm::createNVPTXReplaceImageHandlesPass() {
+  return new NVPTXReplaceImageHandles();
+}
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXSection.h b/contrib/llvm/lib/Target/NVPTX/NVPTXSection.h
index f8a692e9b1a3..aa0436bf0da7 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXSection.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXSection.h
@@ -31,16 +31,16 @@ public:
 
   /// Override this as NVPTX has its own way of printing switching
   /// to a section.
-  virtual void PrintSwitchToSection(const MCAsmInfo &MAI,
-                                    raw_ostream &OS,
-                                    const MCExpr *Subsection) const {}
+  void PrintSwitchToSection(const MCAsmInfo &MAI,
+                            raw_ostream &OS,
+                            const MCExpr *Subsection) const override {}
 
   /// Base address of PTX sections is zero.
-  virtual bool isBaseAddressKnownZero() const { return true; }
-  virtual bool UseCodeAlign() const { return false; }
-  virtual bool isVirtualSection() const { return false; }
-  virtual std::string getLabelBeginName() const { return ""; }
-  virtual std::string getLabelEndName() const { return ""; }
+  bool isBaseAddressKnownZero() const override { return true; }
+  bool UseCodeAlign() const override { return false; }
+  bool isVirtualSection() const override { return false; }
+  std::string getLabelBeginName() const override { return ""; }
+  std::string getLabelEndName() const override { return ""; }
 };
 
 } // end namespace llvm
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXSplitBBatBar.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXSplitBBatBar.cpp
deleted file mode 100644
index b64c30880b94..000000000000
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXSplitBBatBar.cpp
+++ /dev/null
@@ -1,73 +0,0 @@
-//===- NVPTXSplitBBatBar.cpp - Split BB at Barrier  --*- C++ -*--===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-// Split basic blocks so that a basic block that contains a barrier instruction
-// only contains the barrier instruction.
-//
-//===----------------------------------------------------------------------===//
-
-#include "NVPTXSplitBBatBar.h"
-#include "NVPTXUtilities.h"
-#include "llvm/IR/Function.h"
-#include "llvm/IR/Instructions.h"
-#include "llvm/IR/IntrinsicInst.h"
-#include "llvm/IR/Intrinsics.h"
-#include "llvm/Support/InstIterator.h"
-
-using namespace llvm;
-
-namespace llvm { FunctionPass *createSplitBBatBarPass(); }
-
-char NVPTXSplitBBatBar::ID = 0;
-
-bool NVPTXSplitBBatBar::runOnFunction(Function &F) {
-
-  SmallVector<Instruction *, 4> SplitPoints;
-  bool changed = false;
-
-  // Collect all the split points in SplitPoints
-  for (Function::iterator BI = F.begin(), BE = F.end(); BI != BE; ++BI) {
-    BasicBlock::iterator IB = BI->begin();
-    BasicBlock::iterator II = IB;
-    BasicBlock::iterator IE = BI->end();
-
-    // Skit the first instruction. No splitting is needed at this
-    // point even if this is a bar.
-    while (II != IE) {
-      if (IntrinsicInst *inst = dyn_cast<IntrinsicInst>(II)) {
-        Intrinsic::ID id = inst->getIntrinsicID();
-        // If this is a barrier, split at this instruction
-        // and the next instruction.
-        if (llvm::isBarrierIntrinsic(id)) {
-          if (II != IB)
-            SplitPoints.push_back(II);
-          II++;
-          if ((II != IE) && (!II->isTerminator())) {
-            SplitPoints.push_back(II);
-            II++;
-          }
-          continue;
-        }
-      }
-      II++;
-    }
-  }
-
-  for (unsigned i = 0; i != SplitPoints.size(); i++) {
-    changed = true;
-    Instruction *inst = SplitPoints[i];
-    inst->getParent()->splitBasicBlock(inst, "bar_split");
-  }
-
-  return changed;
-}
-
-// This interface will most likely not be necessary, because this pass will
-// not be invoked by the driver, but will be used as a prerequisite to
-// another pass.
-FunctionPass *llvm::createSplitBBatBarPass() { return new NVPTXSplitBBatBar(); }
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXSplitBBatBar.h b/contrib/llvm/lib/Target/NVPTX/NVPTXSplitBBatBar.h
deleted file mode 100644
index bdafba9075a0..000000000000
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXSplitBBatBar.h
+++ /dev/null
@@ -1,41 +0,0 @@
-//===-- llvm/lib/Target/NVPTX/NVPTXSplitBBatBar.h ---------------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file contains the declaration of the NVIDIA specific declarations
-// for splitting basic blocks at barrier instructions.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef NVPTX_SPLIT_BB_AT_BAR_H
-#define NVPTX_SPLIT_BB_AT_BAR_H
-
-#include "llvm/CodeGen/MachineFunctionAnalysis.h"
-#include "llvm/Pass.h"
-
-namespace llvm {
-
-// actual analysis class, which is a functionpass
-struct NVPTXSplitBBatBar : public FunctionPass {
-  static char ID;
-
-  NVPTXSplitBBatBar() : FunctionPass(ID) {}
-  void getAnalysisUsage(AnalysisUsage &AU) const {
-    AU.addPreserved<MachineFunctionAnalysis>();
-  }
-  virtual bool runOnFunction(Function &F);
-
-  virtual const char *getPassName() const {
-    return "Split basic blocks at barrier";
-  }
-};
-
-extern FunctionPass *createSplitBBatBarPass();
-}
-
-#endif //NVPTX_SPLIT_BB_AT_BAR_H
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXSubtarget.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXSubtarget.cpp
index 9771a176d8dc..d5cded218362 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXSubtarget.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXSubtarget.cpp
@@ -12,21 +12,54 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTXSubtarget.h"
+
+using namespace llvm;
+
+#define DEBUG_TYPE "nvptx-subtarget"
+
 #define GET_SUBTARGETINFO_ENUM
 #define GET_SUBTARGETINFO_TARGET_DESC
 #define GET_SUBTARGETINFO_CTOR
 #include "NVPTXGenSubtargetInfo.inc"
 
-using namespace llvm;
-
-
 // Pin the vtable to this file.
 void NVPTXSubtarget::anchor() {}
 
+static std::string computeDataLayout(bool is64Bit) {
+  std::string Ret = "e";
+
+  if (!is64Bit)
+    Ret += "-p:32:32";
+
+  Ret += "-i64:64-v16:16-v32:32-n16:32:64";
+
+  return Ret;
+}
+
+NVPTXSubtarget &NVPTXSubtarget::initializeSubtargetDependencies(StringRef CPU,
+                                                                StringRef FS) {
+    // Provide the default CPU if we don't have one.
+  if (CPU.empty() && FS.size())
+    llvm_unreachable("we are not using FeatureStr");
+  TargetName = CPU.empty() ? "sm_20" : CPU;
+
+  ParseSubtargetFeatures(TargetName, FS);
+
+  // Set default to PTX 3.2 (CUDA 5.5)
+  if (PTXVersion == 0) {
+    PTXVersion = 32;
+  }
+
+  return *this;
+}
+
 NVPTXSubtarget::NVPTXSubtarget(const std::string &TT, const std::string &CPU,
-                               const std::string &FS, bool is64Bit)
+                               const std::string &FS, const TargetMachine &TM,
+                               bool is64Bit)
     : NVPTXGenSubtargetInfo(TT, CPU, FS), Is64Bit(is64Bit), PTXVersion(0),
-      SmVersion(20) {
+      SmVersion(20), DL(computeDataLayout(is64Bit)),
+      InstrInfo(initializeSubtargetDependencies(CPU, FS)),
+      TLInfo((NVPTXTargetMachine &)TM), TSInfo(&DL), FrameLowering(*this) {
 
   Triple T(TT);
 
@@ -34,26 +67,4 @@ NVPTXSubtarget::NVPTXSubtarget(const std::string &TT, const std::string &CPU,
     drvInterface = NVPTX::NVCL;
   else
     drvInterface = NVPTX::CUDA;
-
-  // Provide the default CPU if none
-  std::string defCPU = "sm_20";
-
-  ParseSubtargetFeatures((CPU.empty() ? defCPU : CPU), FS);
-
-  // Get the TargetName from the FS if available
-  if (FS.empty() && CPU.empty())
-    TargetName = defCPU;
-  else if (!CPU.empty())
-    TargetName = CPU;
-  else
-    llvm_unreachable("we are not using FeatureStr");
-
-  // We default to PTX 3.1, but we cannot just default to it in the initializer
-  // since the attribute parser checks if the given option is >= the default.
-  // So if we set ptx31 as the default, the ptx30 attribute would never match.
-  // Instead, we use 0 as the default and manually set 31 if the default is
-  // used.
-  if (PTXVersion == 0) {
-    PTXVersion = 31;
-  }
 }
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXSubtarget.h b/contrib/llvm/lib/Target/NVPTX/NVPTXSubtarget.h
index 004be116a96c..4c41e4e470dd 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXSubtarget.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXSubtarget.h
@@ -15,13 +15,18 @@
 #define NVPTXSUBTARGET_H
 
 #include "NVPTX.h"
+#include "NVPTXFrameLowering.h"
+#include "NVPTXISelLowering.h"
+#include "NVPTXInstrInfo.h"
+#include "NVPTXRegisterInfo.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/Target/TargetSelectionDAGInfo.h"
 #include "llvm/Target/TargetSubtargetInfo.h"
+#include <string>
 
 #define GET_SUBTARGETINFO_HEADER
 #include "NVPTXGenSubtargetInfo.inc"
 
-#include <string>
-
 namespace llvm {
 
 class NVPTXSubtarget : public NVPTXGenSubtargetInfo {
@@ -36,12 +41,30 @@ class NVPTXSubtarget : public NVPTXGenSubtargetInfo {
   // SM version x.y is represented as 10*x+y, e.g. 3.1 == 31
   unsigned int SmVersion;
 
+  const DataLayout DL; // Calculates type size & alignment
+  NVPTXInstrInfo InstrInfo;
+  NVPTXTargetLowering TLInfo;
+  TargetSelectionDAGInfo TSInfo;
+
+  // NVPTX does not have any call stack frame, but need a NVPTX specific
+  // FrameLowering class because TargetFrameLowering is abstract.
+  NVPTXFrameLowering FrameLowering;
+
 public:
   /// This constructor initializes the data members to match that
   /// of the specified module.
   ///
   NVPTXSubtarget(const std::string &TT, const std::string &CPU,
-                 const std::string &FS, bool is64Bit);
+                 const std::string &FS, const TargetMachine &TM, bool is64Bit);
+
+  const TargetFrameLowering *getFrameLowering() const { return &FrameLowering; }
+  const NVPTXInstrInfo *getInstrInfo() const { return &InstrInfo; }
+  const DataLayout *getDataLayout() const { return &DL; }
+  const NVPTXRegisterInfo *getRegisterInfo() const {
+    return &InstrInfo.getRegisterInfo();
+  }
+  const NVPTXTargetLowering *getTargetLowering() const { return &TLInfo; }
+  const TargetSelectionDAGInfo *getSelectionDAGInfo() const { return &TSInfo; }
 
   bool hasBrkPt() const { return SmVersion >= 11; }
   bool hasAtomRedG32() const { return SmVersion >= 11; }
@@ -58,13 +81,24 @@ public:
   bool hasFMAF32() const { return SmVersion >= 20; }
   bool hasFMAF64() const { return SmVersion >= 13; }
   bool hasLDG() const { return SmVersion >= 32; }
-  bool hasLDU() const { return SmVersion >= 20; }
+  bool hasLDU() const { return ((SmVersion >= 20) && (SmVersion < 30)); }
   bool hasGenericLdSt() const { return SmVersion >= 20; }
-  inline bool hasHWROT32() const { return false; }
-  inline bool hasSWROT32() const { return true; }
+  inline bool hasHWROT32() const { return SmVersion >= 32; }
+  inline bool hasSWROT32() const {
+    return ((SmVersion >= 20) && (SmVersion < 32));
+  }
   inline bool hasROT32() const { return hasHWROT32() || hasSWROT32(); }
   inline bool hasROT64() const { return SmVersion >= 20; }
 
+  bool hasImageHandles() const {
+    // Enable handles for Kepler+, where CUDA supports indirect surfaces and
+    // textures
+    if (getDrvInterface() == NVPTX::CUDA)
+      return (SmVersion >= 30);
+
+    // Disabled, otherwise
+    return false;
+  }
   bool is64Bit() const { return Is64Bit; }
 
   unsigned int getSmVersion() const { return SmVersion; }
@@ -73,22 +107,8 @@ public:
 
   unsigned getPTXVersion() const { return PTXVersion; }
 
+  NVPTXSubtarget &initializeSubtargetDependencies(StringRef CPU, StringRef FS);
   void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
-
-  std::string getDataLayout() const {
-    const char *p;
-    if (is64Bit())
-      p = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-"
-          "f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-"
-          "n16:32:64";
-    else
-      p = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-"
-          "f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-"
-          "n16:32:64";
-
-    return std::string(p);
-  }
-
 };
 
 } // End llvm namespace
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp
index 46edd6d83f65..069a1b9966f0 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp
@@ -16,16 +16,14 @@
 #include "NVPTX.h"
 #include "NVPTXAllocaHoisting.h"
 #include "NVPTXLowerAggrCopies.h"
-#include "NVPTXSplitBBatBar.h"
-#include "llvm/ADT/OwningPtr.h"
 #include "llvm/Analysis/Passes.h"
-#include "llvm/Analysis/Verifier.h"
-#include "llvm/Assembly/PrintModulePass.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/IR/DataLayout.h"
+#include "llvm/IR/IRPrintingPasses.h"
+#include "llvm/IR/Verifier.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCStreamer.h"
@@ -50,6 +48,8 @@ using namespace llvm;
 namespace llvm {
 void initializeNVVMReflectPass(PassRegistry&);
 void initializeGenericToNVVMPass(PassRegistry&);
+void initializeNVPTXAssignValidGlobalNamesPass(PassRegistry&);
+void initializeNVPTXFavorNonGenericAddrSpacesPass(PassRegistry &);
 }
 
 extern "C" void LLVMInitializeNVPTXTarget() {
@@ -61,17 +61,18 @@ extern "C" void LLVMInitializeNVPTXTarget() {
   // but it's very NVPTX-specific.
   initializeNVVMReflectPass(*PassRegistry::getPassRegistry());
   initializeGenericToNVVMPass(*PassRegistry::getPassRegistry());
+  initializeNVPTXAssignValidGlobalNamesPass(*PassRegistry::getPassRegistry());
+  initializeNVPTXFavorNonGenericAddrSpacesPass(
+    *PassRegistry::getPassRegistry());
 }
 
-NVPTXTargetMachine::NVPTXTargetMachine(
-    const Target &T, StringRef TT, StringRef CPU, StringRef FS,
-    const TargetOptions &Options, Reloc::Model RM, CodeModel::Model CM,
-    CodeGenOpt::Level OL, bool is64bit)
+NVPTXTargetMachine::NVPTXTargetMachine(const Target &T, StringRef TT,
+                                       StringRef CPU, StringRef FS,
+                                       const TargetOptions &Options,
+                                       Reloc::Model RM, CodeModel::Model CM,
+                                       CodeGenOpt::Level OL, bool is64bit)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
-      Subtarget(TT, CPU, FS, is64bit), DL(Subtarget.getDataLayout()),
-      InstrInfo(*this), TLInfo(*this), TSInfo(*this),
-      FrameLowering(
-          *this, is64bit) /*FrameInfo(TargetFrameInfo::StackGrowsUp, 8, 0)*/ {
+      Subtarget(TT, CPU, FS, *this, is64bit) {
   initAsmInfo();
 }
 
@@ -101,14 +102,15 @@ public:
     return getTM<NVPTXTargetMachine>();
   }
 
-  virtual void addIRPasses();
-  virtual bool addInstSelector();
-  virtual bool addPreRegAlloc();
-  virtual bool addPostRegAlloc();
+  void addIRPasses() override;
+  bool addInstSelector() override;
+  bool addPreRegAlloc() override;
+  bool addPostRegAlloc() override;
+  void addMachineSSAOptimization() override;
 
-  virtual FunctionPass *createTargetRegisterAllocator(bool) LLVM_OVERRIDE;
-  virtual void addFastRegAlloc(FunctionPass *RegAllocPass);
-  virtual void addOptimizedRegAlloc(FunctionPass *RegAllocPass);
+  FunctionPass *createTargetRegisterAllocator(bool) override;
+  void addFastRegAlloc(FunctionPass *RegAllocPass) override;
+  void addOptimizedRegAlloc(FunctionPass *RegAllocPass) override;
 };
 } // end anonymous namespace
 
@@ -128,15 +130,42 @@ void NVPTXPassConfig::addIRPasses() {
   disablePass(&BranchFolderPassID);
   disablePass(&TailDuplicateID);
 
+  addPass(createNVPTXImageOptimizerPass());
   TargetPassConfig::addIRPasses();
+  addPass(createNVPTXAssignValidGlobalNamesPass());
   addPass(createGenericToNVVMPass());
+  addPass(createNVPTXFavorNonGenericAddrSpacesPass());
+  addPass(createSeparateConstOffsetFromGEPPass());
+  // The SeparateConstOffsetFromGEP pass creates variadic bases that can be used
+  // by multiple GEPs. Run GVN or EarlyCSE to really reuse them. GVN generates
+  // significantly better code than EarlyCSE for some of our benchmarks.
+  if (getOptLevel() == CodeGenOpt::Aggressive)
+    addPass(createGVNPass());
+  else
+    addPass(createEarlyCSEPass());
+  // Both FavorNonGenericAddrSpaces and SeparateConstOffsetFromGEP may leave
+  // some dead code.  We could remove dead code in an ad-hoc manner, but that
+  // requires manual work and might be error-prone.
+  //
+  // The FavorNonGenericAddrSpaces pass shortcuts unnecessary addrspacecasts,
+  // and leave them unused.
+  //
+  // SeparateConstOffsetFromGEP rebuilds a new index from the old index, and the
+  // old index and some of its intermediate results may become unused.
+  addPass(createDeadCodeEliminationPass());
 }
 
 bool NVPTXPassConfig::addInstSelector() {
+  const NVPTXSubtarget &ST =
+    getTM<NVPTXTargetMachine>().getSubtarget<NVPTXSubtarget>();
+
   addPass(createLowerAggrCopies());
-  addPass(createSplitBBatBarPass());
   addPass(createAllocaHoisting());
   addPass(createNVPTXISelDag(getNVPTXTargetMachine(), getOptLevel()));
+
+  if (!ST.hasImageHandles())
+    addPass(createNVPTXReplaceImageHandlesPass());
+
   return false;
 }
 
@@ -147,7 +176,7 @@ bool NVPTXPassConfig::addPostRegAlloc() {
 }
 
 FunctionPass *NVPTXPassConfig::createTargetRegisterAllocator(bool) {
-  return 0; // No reg alloc
+  return nullptr; // No reg alloc
 }
 
 void NVPTXPassConfig::addFastRegAlloc(FunctionPass *RegAllocPass) {
@@ -179,3 +208,43 @@ void NVPTXPassConfig::addOptimizedRegAlloc(FunctionPass *RegAllocPass) {
 
   printAndVerify("After StackSlotColoring");
 }
+
+void NVPTXPassConfig::addMachineSSAOptimization() {
+  // Pre-ra tail duplication.
+  if (addPass(&EarlyTailDuplicateID))
+    printAndVerify("After Pre-RegAlloc TailDuplicate");
+
+  // Optimize PHIs before DCE: removing dead PHI cycles may make more
+  // instructions dead.
+  addPass(&OptimizePHIsID);
+
+  // This pass merges large allocas. StackSlotColoring is a different pass
+  // which merges spill slots.
+  addPass(&StackColoringID);
+
+  // If the target requests it, assign local variables to stack slots relative
+  // to one another and simplify frame index references where possible.
+  addPass(&LocalStackSlotAllocationID);
+
+  // With optimization, dead code should already be eliminated. However
+  // there is one known exception: lowered code for arguments that are only
+  // used by tail calls, where the tail calls reuse the incoming stack
+  // arguments directly (see t11 in test/CodeGen/X86/sibcall.ll).
+  addPass(&DeadMachineInstructionElimID);
+  printAndVerify("After codegen DCE pass");
+
+  // Allow targets to insert passes that improve instruction level parallelism,
+  // like if-conversion. Such passes will typically need dominator trees and
+  // loop info, just like LICM and CSE below.
+  if (addILPOpts())
+    printAndVerify("After ILP optimizations");
+
+  addPass(&MachineLICMID);
+  addPass(&MachineCSEID);
+
+  addPass(&MachineSinkingID);
+  printAndVerify("After Machine LICM, CSE and Sinking passes");
+
+  addPass(&PeepholeOptimizerID);
+  printAndVerify("After codegen peephole optimization pass");
+}
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXTargetMachine.h b/contrib/llvm/lib/Target/NVPTX/NVPTXTargetMachine.h
index 5fbcf735b48f..a7a1c8f4e171 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXTargetMachine.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXTargetMachine.h
@@ -14,13 +14,8 @@
 #ifndef NVPTX_TARGETMACHINE_H
 #define NVPTX_TARGETMACHINE_H
 
-#include "ManagedStringPool.h"
-#include "NVPTXFrameLowering.h"
-#include "NVPTXISelLowering.h"
-#include "NVPTXInstrInfo.h"
-#include "NVPTXRegisterInfo.h"
 #include "NVPTXSubtarget.h"
-#include "llvm/IR/DataLayout.h"
+#include "ManagedStringPool.h"
 #include "llvm/Target/TargetFrameLowering.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetSelectionDAGInfo.h"
@@ -31,65 +26,52 @@ namespace llvm {
 ///
 class NVPTXTargetMachine : public LLVMTargetMachine {
   NVPTXSubtarget Subtarget;
-  const DataLayout DL; // Calculates type size & alignment
-  NVPTXInstrInfo InstrInfo;
-  NVPTXTargetLowering TLInfo;
-  TargetSelectionDAGInfo TSInfo;
-
-  // NVPTX does not have any call stack frame, but need a NVPTX specific
-  // FrameLowering class because TargetFrameLowering is abstract.
-  NVPTXFrameLowering FrameLowering;
 
   // Hold Strings that can be free'd all together with NVPTXTargetMachine
   ManagedStringPool ManagedStrPool;
 
-  //bool addCommonCodeGenPasses(PassManagerBase &, CodeGenOpt::Level,
-  //                            bool DisableVerify, MCContext *&OutCtx);
-
 public:
   NVPTXTargetMachine(const Target &T, StringRef TT, StringRef CPU, StringRef FS,
                      const TargetOptions &Options, Reloc::Model RM,
                      CodeModel::Model CM, CodeGenOpt::Level OP, bool is64bit);
 
-  virtual const TargetFrameLowering *getFrameLowering() const {
-    return &FrameLowering;
+  const TargetFrameLowering *getFrameLowering() const override {
+    return getSubtargetImpl()->getFrameLowering();
   }
-  virtual const NVPTXInstrInfo *getInstrInfo() const { return &InstrInfo; }
-  virtual const DataLayout *getDataLayout() const { return &DL; }
-  virtual const NVPTXSubtarget *getSubtargetImpl() const { return &Subtarget; }
-
-  virtual const NVPTXRegisterInfo *getRegisterInfo() const {
-    return &(InstrInfo.getRegisterInfo());
+  const NVPTXInstrInfo *getInstrInfo() const override {
+    return getSubtargetImpl()->getInstrInfo();
   }
-
-  virtual NVPTXTargetLowering *getTargetLowering() const {
-    return const_cast<NVPTXTargetLowering *>(&TLInfo);
+  const DataLayout *getDataLayout() const override {
+    return getSubtargetImpl()->getDataLayout();
   }
-
-  virtual const TargetSelectionDAGInfo *getSelectionDAGInfo() const {
-    return &TSInfo;
+  const NVPTXSubtarget *getSubtargetImpl() const override { return &Subtarget; }
+  const NVPTXRegisterInfo *getRegisterInfo() const override {
+    return getSubtargetImpl()->getRegisterInfo();
   }
 
-  //virtual bool addInstSelector(PassManagerBase &PM,
-  //                             CodeGenOpt::Level OptLevel);
+  const NVPTXTargetLowering *getTargetLowering() const override {
+    return getSubtargetImpl()->getTargetLowering();
+  }
 
-  //virtual bool addPreRegAlloc(PassManagerBase &, CodeGenOpt::Level);
+  const TargetSelectionDAGInfo *getSelectionDAGInfo() const override {
+    return getSubtargetImpl()->getSelectionDAGInfo();
+  }
 
   ManagedStringPool *getManagedStrPool() const {
     return const_cast<ManagedStringPool *>(&ManagedStrPool);
   }
 
-  virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
+  TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
 
   // Emission of machine code through JITCodeEmitter is not supported.
-  virtual bool addPassesToEmitMachineCode(PassManagerBase &, JITCodeEmitter &,
-                                          bool = true) {
+  bool addPassesToEmitMachineCode(PassManagerBase &, JITCodeEmitter &,
+                                  bool = true) override {
     return true;
   }
 
   // Emission of machine code through MCJIT is not supported.
-  virtual bool addPassesToEmitMC(PassManagerBase &, MCContext *&, raw_ostream &,
-                                 bool = true) {
+  bool addPassesToEmitMC(PassManagerBase &, MCContext *&, raw_ostream &,
+                         bool = true) override {
     return true;
   }
 
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXTargetObjectFile.h b/contrib/llvm/lib/Target/NVPTX/NVPTXTargetObjectFile.h
index 2a7394b79ae6..ba8086d78880 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXTargetObjectFile.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXTargetObjectFile.h
@@ -22,31 +22,31 @@ class NVPTXTargetObjectFile : public TargetLoweringObjectFile {
 
 public:
   NVPTXTargetObjectFile() {
-    TextSection = 0;
-    DataSection = 0;
-    BSSSection = 0;
-    ReadOnlySection = 0;
+    TextSection = nullptr;
+    DataSection = nullptr;
+    BSSSection = nullptr;
+    ReadOnlySection = nullptr;
 
-    StaticCtorSection = 0;
-    StaticDtorSection = 0;
-    LSDASection = 0;
-    EHFrameSection = 0;
-    DwarfAbbrevSection = 0;
-    DwarfInfoSection = 0;
-    DwarfLineSection = 0;
-    DwarfFrameSection = 0;
-    DwarfPubTypesSection = 0;
-    DwarfDebugInlineSection = 0;
-    DwarfStrSection = 0;
-    DwarfLocSection = 0;
-    DwarfARangesSection = 0;
-    DwarfRangesSection = 0;
-    DwarfMacroInfoSection = 0;
+    StaticCtorSection = nullptr;
+    StaticDtorSection = nullptr;
+    LSDASection = nullptr;
+    EHFrameSection = nullptr;
+    DwarfAbbrevSection = nullptr;
+    DwarfInfoSection = nullptr;
+    DwarfLineSection = nullptr;
+    DwarfFrameSection = nullptr;
+    DwarfPubTypesSection = nullptr;
+    DwarfDebugInlineSection = nullptr;
+    DwarfStrSection = nullptr;
+    DwarfLocSection = nullptr;
+    DwarfARangesSection = nullptr;
+    DwarfRangesSection = nullptr;
+    DwarfMacroInfoSection = nullptr;
   }
 
   virtual ~NVPTXTargetObjectFile();
 
-  virtual void Initialize(MCContext &ctx, const TargetMachine &TM) {
+  void Initialize(MCContext &ctx, const TargetMachine &TM) override {
     TargetLoweringObjectFile::Initialize(ctx, TM);
     TextSection = new NVPTXSection(MCSection::SV_ELF, SectionKind::getText());
     DataSection =
@@ -87,13 +87,14 @@ public:
         new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
   }
 
-  virtual const MCSection *getSectionForConstant(SectionKind Kind) const {
+  const MCSection *getSectionForConstant(SectionKind Kind,
+                                         const Constant *C) const override {
     return ReadOnlySection;
   }
 
-  virtual const MCSection *
-  getExplicitSectionGlobal(const GlobalValue *GV, SectionKind Kind,
-                           Mangler *Mang, const TargetMachine &TM) const {
+  const MCSection *getExplicitSectionGlobal(const GlobalValue *GV,
+                                       SectionKind Kind, Mangler &Mang,
+                                       const TargetMachine &TM) const override {
     return DataSection;
   }
 
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXUtilities.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXUtilities.cpp
index 6786eb02240c..a9fd190b7ff0 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXUtilities.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXUtilities.cpp
@@ -22,9 +22,9 @@
 #include <map>
 #include <string>
 #include <vector>
-//#include <iostream>
 #include "llvm/Support/ManagedStatic.h"
-#include "llvm/Support/InstIterator.h"
+#include "llvm/IR/InstIterator.h"
+#include "llvm/Support/MutexGuard.h"
 
 using namespace llvm;
 
@@ -33,8 +33,15 @@ typedef std::map<const GlobalValue *, key_val_pair_t> global_val_annot_t;
 typedef std::map<const Module *, global_val_annot_t> per_module_annot_t;
 
 ManagedStatic<per_module_annot_t> annotationCache;
+static sys::Mutex Lock;
+
+void llvm::clearAnnotationCache(const llvm::Module *Mod) {
+  MutexGuard Guard(Lock);
+  annotationCache->erase(Mod);
+}
 
 static void cacheAnnotationFromMD(const MDNode *md, key_val_pair_t &retval) {
+  MutexGuard Guard(Lock);
   assert(md && "Invalid mdnode for annotation");
   assert((md->getNumOperands() % 2) == 1 && "Invalid number of operands");
   // start index = 1, to skip the global variable key
@@ -60,6 +67,7 @@ static void cacheAnnotationFromMD(const MDNode *md, key_val_pair_t &retval) {
 }
 
 static void cacheAnnotationFromMD(const Module *m, const GlobalValue *gv) {
+  MutexGuard Guard(Lock);
   NamedMDNode *NMD = m->getNamedMetadata(llvm::NamedMDForAnnotations);
   if (!NMD)
     return;
@@ -92,6 +100,7 @@ static void cacheAnnotationFromMD(const Module *m, const GlobalValue *gv) {
 
 bool llvm::findOneNVVMAnnotation(const GlobalValue *gv, std::string prop,
                                  unsigned &retval) {
+  MutexGuard Guard(Lock);
   const Module *m = gv->getParent();
   if ((*annotationCache).find(m) == (*annotationCache).end())
     cacheAnnotationFromMD(m, gv);
@@ -105,6 +114,7 @@ bool llvm::findOneNVVMAnnotation(const GlobalValue *gv, std::string prop,
 
 bool llvm::findAllNVVMAnnotation(const GlobalValue *gv, std::string prop,
                                  std::vector<unsigned> &retval) {
+  MutexGuard Guard(Lock);
   const Module *m = gv->getParent();
   if ((*annotationCache).find(m) == (*annotationCache).end())
     cacheAnnotationFromMD(m, gv);
@@ -195,8 +205,37 @@ bool llvm::isImageWriteOnly(const llvm::Value &val) {
   return false;
 }
 
+bool llvm::isImageReadWrite(const llvm::Value &val) {
+  if (const Argument *arg = dyn_cast<Argument>(&val)) {
+    const Function *func = arg->getParent();
+    std::vector<unsigned> annot;
+    if (llvm::findAllNVVMAnnotation(func,
+                                    llvm::PropertyAnnotationNames[
+                                        llvm::PROPERTY_ISREADWRITE_IMAGE_PARAM],
+                                    annot)) {
+      if (std::find(annot.begin(), annot.end(), arg->getArgNo()) != annot.end())
+        return true;
+    }
+  }
+  return false;
+}
+
 bool llvm::isImage(const llvm::Value &val) {
-  return llvm::isImageReadOnly(val) || llvm::isImageWriteOnly(val);
+  return llvm::isImageReadOnly(val) || llvm::isImageWriteOnly(val) ||
+         llvm::isImageReadWrite(val);
+}
+
+bool llvm::isManaged(const llvm::Value &val) {
+  if(const GlobalValue *gv = dyn_cast<GlobalValue>(&val)) {
+    unsigned annot;
+    if(llvm::findOneNVVMAnnotation(gv,
+                          llvm::PropertyAnnotationNames[llvm::PROPERTY_MANAGED],
+                                   annot)) {
+      assert((annot == 1) && "Unexpected annotation on a managed symbol");
+      return true;
+    }
+  }
+  return false;
 }
 
 std::string llvm::getTextureName(const llvm::Value &val) {
@@ -354,12 +393,12 @@ llvm::skipPointerTransfer(const Value *V, bool ignore_GEP_indices) {
 const Value *
 llvm::skipPointerTransfer(const Value *V, std::set<const Value *> &processed) {
   if (processed.find(V) != processed.end())
-    return NULL;
+    return nullptr;
   processed.insert(V);
 
   const Value *V2 = V->stripPointerCasts();
   if (V2 != V && processed.find(V2) != processed.end())
-    return NULL;
+    return nullptr;
   processed.insert(V2);
 
   V = V2;
@@ -375,20 +414,20 @@ llvm::skipPointerTransfer(const Value *V, std::set<const Value *> &processed) {
       continue;
     } else if (const PHINode *PN = dyn_cast<PHINode>(V)) {
       if (V != V2 && processed.find(V) != processed.end())
-        return NULL;
+        return nullptr;
       processed.insert(PN);
-      const Value *common = 0;
+      const Value *common = nullptr;
       for (unsigned i = 0; i != PN->getNumIncomingValues(); ++i) {
         const Value *pv = PN->getIncomingValue(i);
         const Value *base = skipPointerTransfer(pv, processed);
         if (base) {
-          if (common == 0)
+          if (!common)
             common = base;
           else if (common != base)
             return PN;
         }
       }
-      if (common == 0)
+      if (!common)
         return PN;
       V = common;
     }
@@ -406,7 +445,7 @@ BasicBlock *llvm::getParentBlock(Value *v) {
   if (Instruction *I = dyn_cast<Instruction>(v))
     return I->getParent();
 
-  return 0;
+  return nullptr;
 }
 
 Function *llvm::getParentFunction(Value *v) {
@@ -419,13 +458,13 @@ Function *llvm::getParentFunction(Value *v) {
   if (BasicBlock *B = dyn_cast<BasicBlock>(v))
     return B->getParent();
 
-  return 0;
+  return nullptr;
 }
 
 // Dump a block by name
 void llvm::dumpBlock(Value *v, char *blockName) {
   Function *F = getParentFunction(v);
-  if (F == 0)
+  if (!F)
     return;
 
   for (Function::iterator it = F->begin(), ie = F->end(); it != ie; ++it) {
@@ -440,8 +479,8 @@ void llvm::dumpBlock(Value *v, char *blockName) {
 // Find an instruction by name
 Instruction *llvm::getInst(Value *base, char *instName) {
   Function *F = getParentFunction(base);
-  if (F == 0)
-    return 0;
+  if (!F)
+    return nullptr;
 
   for (inst_iterator it = inst_begin(F), ie = inst_end(F); it != ie; ++it) {
     Instruction *I = &*it;
@@ -450,7 +489,7 @@ Instruction *llvm::getInst(Value *base, char *instName) {
     }
   }
 
-  return 0;
+  return nullptr;
 }
 
 // Dump an instruction by nane
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXUtilities.h b/contrib/llvm/lib/Target/NVPTX/NVPTXUtilities.h
index a208004297d0..446bfa1e112c 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXUtilities.h
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXUtilities.h
@@ -28,6 +28,8 @@ namespace llvm {
 #define NVCL_IMAGE2D_READONLY_FUNCNAME "__is_image2D_readonly"
 #define NVCL_IMAGE3D_READONLY_FUNCNAME "__is_image3D_readonly"
 
+void clearAnnotationCache(const llvm::Module *);
+
 bool findOneNVVMAnnotation(const llvm::GlobalValue *, std::string, unsigned &);
 bool findAllNVVMAnnotation(const llvm::GlobalValue *, std::string,
                            std::vector<unsigned> &);
@@ -38,6 +40,8 @@ bool isSampler(const llvm::Value &);
 bool isImage(const llvm::Value &);
 bool isImageReadOnly(const llvm::Value &);
 bool isImageWriteOnly(const llvm::Value &);
+bool isImageReadWrite(const llvm::Value &);
+bool isManaged(const llvm::Value &);
 
 std::string getTextureName(const llvm::Value &);
 std::string getSurfaceName(const llvm::Value &);
diff --git a/contrib/llvm/lib/Target/NVPTX/NVVMReflect.cpp b/contrib/llvm/lib/Target/NVPTX/NVVMReflect.cpp
index 7406207c94b5..a8d6b95ae4ec 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVVMReflect.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVVMReflect.cpp
@@ -7,7 +7,7 @@
 //
 //===----------------------------------------------------------------------===//
 //
-// This pass replaces occurences of __nvvm_reflect("string") with an
+// This pass replaces occurrences of __nvvm_reflect("string") with an
 // integer based on -nvvm-reflect-list string=<int> option given to this pass.
 // If an undefined string value is seen in a call to __nvvm_reflect("string"),
 // a default value of 0 will be used.
@@ -18,13 +18,14 @@
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringMap.h"
-#include "llvm/Pass.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Function.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/Type.h"
-#include "llvm/IR/DerivedTypes.h"
-#include "llvm/IR/Instructions.h"
-#include "llvm/IR/Constants.h"
+#include "llvm/Pass.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_os_ostream.h"
@@ -38,6 +39,8 @@
 
 using namespace llvm;
 
+#define DEBUG_TYPE "nvptx-reflect"
+
 namespace llvm { void initializeNVVMReflectPass(PassRegistry &); }
 
 namespace {
@@ -45,17 +48,16 @@ class NVVMReflect : public ModulePass {
 private:
   StringMap<int> VarMap;
   typedef DenseMap<std::string, int>::iterator VarMapIter;
-  Function *ReflectFunction;
 
 public:
   static char ID;
-  NVVMReflect() : ModulePass(ID), ReflectFunction(0) {
+  NVVMReflect() : ModulePass(ID) {
     initializeNVVMReflectPass(*PassRegistry::getPassRegistry());
     VarMap.clear();
   }
 
   NVVMReflect(const StringMap<int> &Mapping)
-  : ModulePass(ID), ReflectFunction(0) {
+  : ModulePass(ID) {
     initializeNVVMReflectPass(*PassRegistry::getPassRegistry());
     for (StringMap<int>::const_iterator I = Mapping.begin(), E = Mapping.end();
          I != E; ++I) {
@@ -63,9 +65,13 @@ public:
     }
   }
 
-  void getAnalysisUsage(AnalysisUsage &AU) const { AU.setPreservesAll(); }
-  virtual bool runOnModule(Module &);
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.setPreservesAll();
+  }
+  bool runOnModule(Module &) override;
 
+private:
+  bool handleFunction(Function *ReflectFunction);
   void setVarMap();
 };
 }
@@ -84,7 +90,7 @@ NVVMReflectEnabled("nvvm-reflect-enable", cl::init(true), cl::Hidden,
 
 char NVVMReflect::ID = 0;
 INITIALIZE_PASS(NVVMReflect, "nvvm-reflect",
-                "Replace occurences of __nvvm_reflect() calls with 0/1", false,
+                "Replace occurrences of __nvvm_reflect() calls with 0/1", false,
                 false)
 
 static cl::list<std::string>
@@ -116,19 +122,7 @@ void NVVMReflect::setVarMap() {
   }
 }
 
-bool NVVMReflect::runOnModule(Module &M) {
-  if (!NVVMReflectEnabled)
-    return false;
-
-  setVarMap();
-
-  ReflectFunction = M.getFunction(NVVM_REFLECT_FUNCTION);
-
-  // If reflect function is not used, then there will be
-  // no entry in the module.
-  if (ReflectFunction == 0)
-    return false;
-
+bool NVVMReflect::handleFunction(Function *ReflectFunction) {
   // Validate _reflect function
   assert(ReflectFunction->isDeclaration() &&
          "_reflect function should not have a body");
@@ -143,23 +137,23 @@ bool NVVMReflect::runOnModule(Module &M) {
   // ConstantArray can be found successfully, see if it can be
   // found in VarMap. If so, replace the uses of CallInst with the
   // value found in VarMap. If not, replace the use  with value 0.
-  for (Value::use_iterator I = ReflectFunction->use_begin(),
-                           E = ReflectFunction->use_end();
-       I != E; ++I) {
-    assert(isa<CallInst>(*I) && "Only a call instruction can use _reflect");
-    CallInst *Reflect = cast<CallInst>(*I);
+  for (User *U : ReflectFunction->users()) {
+    assert(isa<CallInst>(U) && "Only a call instruction can use _reflect");
+    CallInst *Reflect = cast<CallInst>(U);
 
     assert((Reflect->getNumOperands() == 2) &&
            "Only one operand expect for _reflect function");
     // In cuda, we will have an extra constant-to-generic conversion of
     // the string.
-    const Value *conv = Reflect->getArgOperand(0);
-    assert(isa<CallInst>(conv) && "Expected a const-to-gen conversion");
-    const CallInst *ConvCall = cast<CallInst>(conv);
-    const Value *str = ConvCall->getArgOperand(0);
-    assert(isa<ConstantExpr>(str) &&
+    const Value *Str = Reflect->getArgOperand(0);
+    if (isa<CallInst>(Str)) {
+      // CUDA path
+      const CallInst *ConvCall = cast<CallInst>(Str);
+      Str = ConvCall->getArgOperand(0);
+    }
+    assert(isa<ConstantExpr>(Str) &&
            "Format of _reflect function not recognized");
-    const ConstantExpr *GEP = cast<ConstantExpr>(str);
+    const ConstantExpr *GEP = cast<ConstantExpr>(Str);
 
     const Value *Sym = GEP->getOperand(0);
     assert(isa<Constant>(Sym) && "Format of _reflect function not recognized");
@@ -193,3 +187,36 @@ bool NVVMReflect::runOnModule(Module &M) {
     ToRemove[i]->eraseFromParent();
   return true;
 }
+
+bool NVVMReflect::runOnModule(Module &M) {
+  if (!NVVMReflectEnabled)
+    return false;
+
+  setVarMap();
+
+
+  bool Res = false;
+  std::string Name;
+  Type *Tys[1];
+  Type *I8Ty = Type::getInt8Ty(M.getContext());
+  Function *ReflectFunction;
+
+  // Check for standard overloaded versions of llvm.nvvm.reflect
+
+  for (unsigned i = 0; i != 5; ++i) {
+    Tys[0] = PointerType::get(I8Ty, i);
+    Name = Intrinsic::getName(Intrinsic::nvvm_reflect, Tys);
+    ReflectFunction = M.getFunction(Name);
+    if(ReflectFunction != 0) {
+      Res |= handleFunction(ReflectFunction);
+    }
+  }
+
+  ReflectFunction = M.getFunction(NVVM_REFLECT_FUNCTION);
+  // If reflect function is not used, then there will be
+  // no entry in the module.
+  if (ReflectFunction != 0)
+    Res |= handleFunction(ReflectFunction);
+
+  return Res;
+}
diff --git a/contrib/llvm/lib/Target/NVPTX/cl_common_defines.h b/contrib/llvm/lib/Target/NVPTX/cl_common_defines.h
index 45cc0b8b67f2..02c5a94c3d03 100644
--- a/contrib/llvm/lib/Target/NVPTX/cl_common_defines.h
+++ b/contrib/llvm/lib/Target/NVPTX/cl_common_defines.h
@@ -1,5 +1,5 @@
-#ifndef __CL_COMMON_DEFINES_H__
-#define __CL_COMMON_DEFINES_H__
+#ifndef CL_COMMON_DEFINES_H
+#define CL_COMMON_DEFINES_H
 // This file includes defines that are common to both kernel code and
 // the NVPTX back-end.
 
@@ -119,4 +119,4 @@ typedef enum clk_sampler_type {
 #define CLK_LOCAL_MEM_FENCE (1 << 0)
 #define CLK_GLOBAL_MEM_FENCE (1 << 1)
 
-#endif // __CL_COMMON_DEFINES_H__
+#endif // CL_COMMON_DEFINES_H