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

preserve our customizations, where necessary.

92 files changed, 11211 insertions, 4950 deletions

diff --git a/contrib/llvm/lib/Target/R600/AMDGPU.h b/contrib/llvm/lib/Target/R600/AMDGPU.h
index 025b28e..d7e94f7 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPU.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPU.h
@@ -17,6 +17,7 @@
 namespace llvm {
 
 class AMDGPUInstrPrinter;
+class AMDGPUSubtarget;
 class AMDGPUTargetMachine;
 class FunctionPass;
 class MCAsmInfo;
@@ -28,31 +29,49 @@ class TargetMachine;
 FunctionPass *createR600VectorRegMerger(TargetMachine &tm);
 FunctionPass *createR600TextureIntrinsicsReplacer();
 FunctionPass *createR600ExpandSpecialInstrsPass(TargetMachine &tm);
-FunctionPass *createR600EmitClauseMarkers(TargetMachine &tm);
+FunctionPass *createR600EmitClauseMarkers();
 FunctionPass *createR600ClauseMergePass(TargetMachine &tm);
 FunctionPass *createR600Packetizer(TargetMachine &tm);
 FunctionPass *createR600ControlFlowFinalizer(TargetMachine &tm);
-FunctionPass *createAMDGPUCFGStructurizerPass(TargetMachine &tm);
+FunctionPass *createAMDGPUCFGStructurizerPass();
 
 // SI Passes
 FunctionPass *createSITypeRewriter();
 FunctionPass *createSIAnnotateControlFlowPass();
+FunctionPass *createSILowerI1CopiesPass();
+FunctionPass *createSIShrinkInstructionsPass();
 FunctionPass *createSILowerControlFlowPass(TargetMachine &tm);
 FunctionPass *createSIFixSGPRCopiesPass(TargetMachine &tm);
+FunctionPass *createSIFixSGPRLiveRangesPass();
 FunctionPass *createSICodeEmitterPass(formatted_raw_ostream &OS);
 FunctionPass *createSIInsertWaits(TargetMachine &tm);
 
+void initializeSILowerI1CopiesPass(PassRegistry &);
+extern char &SILowerI1CopiesID;
+
 // Passes common to R600 and SI
+FunctionPass *createAMDGPUPromoteAlloca(const AMDGPUSubtarget &ST);
 Pass *createAMDGPUStructurizeCFGPass();
-FunctionPass *createAMDGPUConvertToISAPass(TargetMachine &tm);
 FunctionPass *createAMDGPUISelDag(TargetMachine &tm);
 
 /// \brief Creates an AMDGPU-specific Target Transformation Info pass.
 ImmutablePass *
 createAMDGPUTargetTransformInfoPass(const AMDGPUTargetMachine *TM);
 
+void initializeSIFixSGPRLiveRangesPass(PassRegistry&);
+extern char &SIFixSGPRLiveRangesID;
+
+
 extern Target TheAMDGPUTarget;
 
+namespace AMDGPU {
+enum TargetIndex {
+  TI_CONSTDATA_START
+};
+}
+
+#define END_OF_TEXT_LABEL_NAME "EndOfTextLabel"
+
 } // End namespace llvm
 
 namespace ShaderType {
@@ -68,7 +87,7 @@ namespace ShaderType {
 /// various memory regions on the hardware. On the CPU
 /// all of the address spaces point to the same memory,
 /// however on the GPU, each address space points to
-/// a seperate piece of memory that is unique from other
+/// a separate piece of memory that is unique from other
 /// memory locations.
 namespace AMDGPUAS {
 enum AddressSpaces {
@@ -76,8 +95,8 @@ enum AddressSpaces {
   GLOBAL_ADDRESS   = 1, ///< Address space for global memory (RAT0, VTX0).
   CONSTANT_ADDRESS = 2, ///< Address space for constant memory
   LOCAL_ADDRESS    = 3, ///< Address space for local memory.
-  REGION_ADDRESS   = 4, ///< Address space for region memory.
-  ADDRESS_NONE     = 5, ///< Address space for unknown memory.
+  FLAT_ADDRESS     = 4, ///< Address space for flat memory.
+  REGION_ADDRESS   = 5, ///< Address space for region memory.
   PARAM_D_ADDRESS  = 6, ///< Address space for direct addressible parameter memory (CONST0)
   PARAM_I_ADDRESS  = 7, ///< Address space for indirect addressible parameter memory (VTX1)
 
@@ -102,7 +121,8 @@ enum AddressSpaces {
   CONSTANT_BUFFER_13 = 21,
   CONSTANT_BUFFER_14 = 22,
   CONSTANT_BUFFER_15 = 23,
-  LAST_ADDRESS     = 24
+  ADDRESS_NONE = 24, ///< Address space for unknown memory.
+  LAST_ADDRESS = ADDRESS_NONE
 };
 
 } // namespace AMDGPUAS
diff --git a/contrib/llvm/lib/Target/R600/AMDGPU.td b/contrib/llvm/lib/Target/R600/AMDGPU.td
index 182235b..5645f1a 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPU.td
+++ b/contrib/llvm/lib/Target/R600/AMDGPU.td
@@ -7,8 +7,7 @@
 //
 //==-----------------------------------------------------------------------===//
 
-// Include AMDIL TD files
-include "AMDILBase.td"
+include "llvm/Target/Target.td"
 
 //===----------------------------------------------------------------------===//
 // Subtarget Features
@@ -26,6 +25,11 @@ def FeatureIRStructurizer : SubtargetFeature <"disable-irstructurizer",
         "false",
         "Disable IR Structurizer">;
 
+def FeaturePromoteAlloca : SubtargetFeature <"promote-alloca",
+        "EnablePromoteAlloca",
+        "true",
+        "Enable promote alloca pass">;
+
 // Target features
 
 def FeatureIfCvt : SubtargetFeature <"disable-ifcvt",
@@ -33,36 +37,50 @@ def FeatureIfCvt : SubtargetFeature <"disable-ifcvt",
         "false",
         "Disable the if conversion pass">;
 
-def FeatureFP64     : SubtargetFeature<"fp64",
+def FeatureFP64 : SubtargetFeature<"fp64",
         "FP64",
         "true",
-        "Enable 64bit double precision operations">;
+        "Enable double precision operations">;
+
+def FeatureFP64Denormals : SubtargetFeature<"fp64-denormals",
+        "FP64Denormals",
+        "true",
+        "Enable double precision denormal handling",
+        [FeatureFP64]>;
+
+// Some instructions do not support denormals despite this flag. Using
+// fp32 denormals also causes instructions to run at the double
+// precision rate for the device.
+def FeatureFP32Denormals : SubtargetFeature<"fp32-denormals",
+        "FP32Denormals",
+        "true",
+        "Enable single precision denormal handling">;
 
 def Feature64BitPtr : SubtargetFeature<"64BitPtr",
         "Is64bit",
         "true",
-        "Specify if 64bit addressing should be used.">;
-
-def Feature32on64BitPtr : SubtargetFeature<"64on32BitPtr",
-        "Is32on64bit",
-        "false",
-        "Specify if 64bit sized pointers with 32bit addressing should be used.">;
+        "Specify if 64-bit addressing should be used">;
 
 def FeatureR600ALUInst : SubtargetFeature<"R600ALUInst",
         "R600ALUInst",
         "false",
-        "Older version of ALU instructions encoding.">;
+        "Older version of ALU instructions encoding">;
 
 def FeatureVertexCache : SubtargetFeature<"HasVertexCache",
         "HasVertexCache",
         "true",
-        "Specify use of dedicated vertex cache.">;
+        "Specify use of dedicated vertex cache">;
 
 def FeatureCaymanISA : SubtargetFeature<"caymanISA",
         "CaymanISA",
         "true",
         "Use Cayman ISA">;
 
+def FeatureCFALUBug : SubtargetFeature<"cfalubug",
+        "CFALUBug",
+        "true",
+        "GPU has CF_ALU bug">;
+
 class SubtargetFeatureFetchLimit <string Value> :
                           SubtargetFeature <"fetch"#Value,
         "TexVTXClauseSize",
@@ -72,47 +90,76 @@ class SubtargetFeatureFetchLimit <string Value> :
 def FeatureFetchLimit8 : SubtargetFeatureFetchLimit <"8">;
 def FeatureFetchLimit16 : SubtargetFeatureFetchLimit <"16">;
 
+class SubtargetFeatureWavefrontSize <int Value> : SubtargetFeature<
+        "wavefrontsize"#Value,
+        "WavefrontSize",
+        !cast<string>(Value),
+        "The number of threads per wavefront">;
+
+def FeatureWavefrontSize16 : SubtargetFeatureWavefrontSize<16>;
+def FeatureWavefrontSize32 : SubtargetFeatureWavefrontSize<32>;
+def FeatureWavefrontSize64 : SubtargetFeatureWavefrontSize<64>;
+
+class SubtargetFeatureLocalMemorySize <int Value> : SubtargetFeature<
+        "localmemorysize"#Value,
+        "LocalMemorySize",
+        !cast<string>(Value),
+        "The size of local memory in bytes">;
+
 class SubtargetFeatureGeneration <string Value,
                                   list<SubtargetFeature> Implies> :
         SubtargetFeature <Value, "Gen", "AMDGPUSubtarget::"#Value,
                           Value#" GPU generation", Implies>;
 
+def FeatureLocalMemorySize0 : SubtargetFeatureLocalMemorySize<0>;
+def FeatureLocalMemorySize32768 : SubtargetFeatureLocalMemorySize<32768>;
+def FeatureLocalMemorySize65536 : SubtargetFeatureLocalMemorySize<65536>;
+
 def FeatureR600 : SubtargetFeatureGeneration<"R600",
-        [FeatureR600ALUInst, FeatureFetchLimit8]>;
+        [FeatureR600ALUInst, FeatureFetchLimit8, FeatureLocalMemorySize0]>;
 
 def FeatureR700 : SubtargetFeatureGeneration<"R700",
-        [FeatureFetchLimit16]>;
+        [FeatureFetchLimit16, FeatureLocalMemorySize0]>;
 
 def FeatureEvergreen : SubtargetFeatureGeneration<"EVERGREEN",
-        [FeatureFetchLimit16]>;
+        [FeatureFetchLimit16, FeatureLocalMemorySize32768]>;
 
 def FeatureNorthernIslands : SubtargetFeatureGeneration<"NORTHERN_ISLANDS",
-        [FeatureFetchLimit16]>;
+        [FeatureFetchLimit16, FeatureWavefrontSize64,
+         FeatureLocalMemorySize32768]
+>;
 
 def FeatureSouthernIslands : SubtargetFeatureGeneration<"SOUTHERN_ISLANDS",
-        [Feature64BitPtr, FeatureFP64]>;
+        [Feature64BitPtr, FeatureFP64, FeatureLocalMemorySize32768,
+         FeatureWavefrontSize64]>;
 
 def FeatureSeaIslands : SubtargetFeatureGeneration<"SEA_ISLANDS",
-        [Feature64BitPtr, FeatureFP64]>;
+        [Feature64BitPtr, FeatureFP64, FeatureLocalMemorySize65536,
+         FeatureWavefrontSize64]>;
 //===----------------------------------------------------------------------===//
 
 def AMDGPUInstrInfo : InstrInfo {
   let guessInstructionProperties = 1;
 }
 
-//===----------------------------------------------------------------------===//
-// Declare the target which we are implementing
-//===----------------------------------------------------------------------===//
-def AMDGPUAsmWriter : AsmWriter {
-    string AsmWriterClassName = "InstPrinter";
-    int Variant = 0;
-    bit isMCAsmWriter = 1;
-}
-
 def AMDGPU : Target {
   // Pull in Instruction Info:
   let InstructionSet = AMDGPUInstrInfo;
-  let AssemblyWriters = [AMDGPUAsmWriter];
+}
+
+// Dummy Instruction itineraries for pseudo instructions
+def ALU_NULL : FuncUnit;
+def NullALU : InstrItinClass;
+
+//===----------------------------------------------------------------------===//
+// Predicate helper class
+//===----------------------------------------------------------------------===//
+
+class PredicateControl {
+  Predicate SubtargetPredicate;
+  list<Predicate> OtherPredicates = [];
+  list<Predicate> Predicates = !listconcat([SubtargetPredicate],
+                                            OtherPredicates);
 }
 
 // Include AMDGPU TD files
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUAsmPrinter.cpp b/contrib/llvm/lib/Target/R600/AMDGPUAsmPrinter.cpp
index 67bdba2..73faaa1 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUAsmPrinter.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUAsmPrinter.cpp
@@ -16,15 +16,16 @@
 //===----------------------------------------------------------------------===//
 //
 
-
 #include "AMDGPUAsmPrinter.h"
 #include "AMDGPU.h"
+#include "AMDGPUSubtarget.h"
 #include "R600Defines.h"
 #include "R600MachineFunctionInfo.h"
 #include "R600RegisterInfo.h"
 #include "SIDefines.h"
 #include "SIMachineFunctionInfo.h"
 #include "SIRegisterInfo.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCSectionELF.h"
 #include "llvm/MC/MCStreamer.h"
@@ -35,6 +36,41 @@
 
 using namespace llvm;
 
+// TODO: This should get the default rounding mode from the kernel. We just set
+// the default here, but this could change if the OpenCL rounding mode pragmas
+// are used.
+//
+// The denormal mode here should match what is reported by the OpenCL runtime
+// for the CL_FP_DENORM bit from CL_DEVICE_{HALF|SINGLE|DOUBLE}_FP_CONFIG, but
+// can also be override to flush with the -cl-denorms-are-zero compiler flag.
+//
+// AMD OpenCL only sets flush none and reports CL_FP_DENORM for double
+// precision, and leaves single precision to flush all and does not report
+// CL_FP_DENORM for CL_DEVICE_SINGLE_FP_CONFIG. Mesa's OpenCL currently reports
+// CL_FP_DENORM for both.
+//
+// FIXME: It seems some instructions do not support single precision denormals
+// regardless of the mode (exp_*_f32, rcp_*_f32, rsq_*_f32, rsq_*f32, sqrt_f32,
+// and sin_f32, cos_f32 on most parts).
+
+// We want to use these instructions, and using fp32 denormals also causes
+// instructions to run at the double precision rate for the device so it's
+// probably best to just report no single precision denormals.
+static uint32_t getFPMode(const MachineFunction &F) {
+  const AMDGPUSubtarget& ST = F.getTarget().getSubtarget<AMDGPUSubtarget>();
+  // TODO: Is there any real use for the flush in only / flush out only modes?
+
+  uint32_t FP32Denormals =
+    ST.hasFP32Denormals() ? FP_DENORM_FLUSH_NONE : FP_DENORM_FLUSH_IN_FLUSH_OUT;
+
+  uint32_t FP64Denormals =
+    ST.hasFP64Denormals() ? FP_DENORM_FLUSH_NONE : FP_DENORM_FLUSH_IN_FLUSH_OUT;
+
+  return FP_ROUND_MODE_SP(FP_ROUND_ROUND_TO_NEAREST) |
+         FP_ROUND_MODE_DP(FP_ROUND_ROUND_TO_NEAREST) |
+         FP_DENORM_MODE_SP(FP32Denormals) |
+         FP_DENORM_MODE_DP(FP64Denormals);
+}
 
 static AsmPrinter *createAMDGPUAsmPrinterPass(TargetMachine &tm,
                                               MCStreamer &Streamer) {
@@ -46,28 +82,36 @@ extern "C" void LLVMInitializeR600AsmPrinter() {
 }
 
 AMDGPUAsmPrinter::AMDGPUAsmPrinter(TargetMachine &TM, MCStreamer &Streamer)
-    : AsmPrinter(TM, Streamer)
-{
-  DisasmEnabled = TM.getSubtarget<AMDGPUSubtarget>().dumpCode() &&
-                  ! Streamer.hasRawTextSupport();
+    : AsmPrinter(TM, Streamer) {
+  DisasmEnabled = TM.getSubtarget<AMDGPUSubtarget>().dumpCode();
+}
+
+void AMDGPUAsmPrinter::EmitEndOfAsmFile(Module &M) {
+
+  // This label is used to mark the end of the .text section.
+  const TargetLoweringObjectFile &TLOF = getObjFileLowering();
+  OutStreamer.SwitchSection(TLOF.getTextSection());
+  MCSymbol *EndOfTextLabel =
+      OutContext.GetOrCreateSymbol(StringRef(END_OF_TEXT_LABEL_NAME));
+  OutStreamer.EmitLabel(EndOfTextLabel);
 }
 
-/// We need to override this function so we can avoid
-/// the call to EmitFunctionHeader(), which the MCPureStreamer can't handle.
 bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
   SetupMachineFunction(MF);
-  if (OutStreamer.hasRawTextSupport()) {
-    OutStreamer.EmitRawText("@" + MF.getName() + ":");
-  }
+
+  OutStreamer.emitRawComment(Twine('@') + MF.getName() + Twine(':'));
 
   MCContext &Context = getObjFileLowering().getContext();
   const MCSectionELF *ConfigSection = Context.getELFSection(".AMDGPU.config",
                                               ELF::SHT_PROGBITS, 0,
                                               SectionKind::getReadOnly());
   OutStreamer.SwitchSection(ConfigSection);
+
   const AMDGPUSubtarget &STM = TM.getSubtarget<AMDGPUSubtarget>();
+  SIProgramInfo KernelInfo;
   if (STM.getGeneration() > AMDGPUSubtarget::NORTHERN_ISLANDS) {
-    EmitProgramInfoSI(MF);
+    getSIProgramInfo(KernelInfo, MF);
+    EmitProgramInfoSI(MF, KernelInfo);
   } else {
     EmitProgramInfoR600(MF);
   }
@@ -79,6 +123,34 @@ bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
   OutStreamer.SwitchSection(getObjFileLowering().getTextSection());
   EmitFunctionBody();
 
+  if (isVerbose()) {
+    const MCSectionELF *CommentSection
+      = Context.getELFSection(".AMDGPU.csdata",
+                              ELF::SHT_PROGBITS, 0,
+                              SectionKind::getReadOnly());
+    OutStreamer.SwitchSection(CommentSection);
+
+    if (STM.getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS) {
+      OutStreamer.emitRawComment(" Kernel info:", false);
+      OutStreamer.emitRawComment(" codeLenInByte = " + Twine(KernelInfo.CodeLen),
+                                 false);
+      OutStreamer.emitRawComment(" NumSgprs: " + Twine(KernelInfo.NumSGPR),
+                                 false);
+      OutStreamer.emitRawComment(" NumVgprs: " + Twine(KernelInfo.NumVGPR),
+                                 false);
+      OutStreamer.emitRawComment(" FloatMode: " + Twine(KernelInfo.FloatMode),
+                                 false);
+      OutStreamer.emitRawComment(" IeeeMode: " + Twine(KernelInfo.IEEEMode),
+                                 false);
+      OutStreamer.emitRawComment(" ScratchSize: " + Twine(KernelInfo.ScratchSize),
+                                 false);
+    } else {
+      R600MachineFunctionInfo *MFI = MF.getInfo<R600MachineFunctionInfo>();
+      OutStreamer.emitRawComment(
+        Twine("SQ_PGM_RESOURCES:STACK_SIZE = " + Twine(MFI->StackSize)));
+    }
+  }
+
   if (STM.dumpCode()) {
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
     MF.dump();
@@ -102,25 +174,21 @@ bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
   return false;
 }
 
-void AMDGPUAsmPrinter::EmitProgramInfoR600(MachineFunction &MF) {
+void AMDGPUAsmPrinter::EmitProgramInfoR600(const MachineFunction &MF) {
   unsigned MaxGPR = 0;
   bool killPixel = false;
-  const R600RegisterInfo * RI =
-                static_cast<const R600RegisterInfo*>(TM.getRegisterInfo());
-  R600MachineFunctionInfo *MFI = MF.getInfo<R600MachineFunctionInfo>();
+  const R600RegisterInfo *RI
+    = static_cast<const R600RegisterInfo*>(TM.getRegisterInfo());
+  const R600MachineFunctionInfo *MFI = MF.getInfo<R600MachineFunctionInfo>();
   const AMDGPUSubtarget &STM = TM.getSubtarget<AMDGPUSubtarget>();
 
-  for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
-                                                  BB != BB_E; ++BB) {
-    MachineBasicBlock &MBB = *BB;
-    for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
-                                                    I != E; ++I) {
-      MachineInstr &MI = *I;
+  for (const MachineBasicBlock &MBB : MF) {
+    for (const MachineInstr &MI : MBB) {
       if (MI.getOpcode() == AMDGPU::KILLGT)
         killPixel = true;
       unsigned numOperands = MI.getNumOperands();
       for (unsigned op_idx = 0; op_idx < numOperands; op_idx++) {
-        MachineOperand & MO = MI.getOperand(op_idx);
+        const MachineOperand &MO = MI.getOperand(op_idx);
         if (!MO.isReg())
           continue;
         unsigned HWReg = RI->getEncodingValue(MO.getReg()) & 0xff;
@@ -136,7 +204,7 @@ void AMDGPUAsmPrinter::EmitProgramInfoR600(MachineFunction &MF) {
   unsigned RsrcReg;
   if (STM.getGeneration() >= AMDGPUSubtarget::EVERGREEN) {
     // Evergreen / Northern Islands
-    switch (MFI->ShaderType) {
+    switch (MFI->getShaderType()) {
     default: // Fall through
     case ShaderType::COMPUTE:  RsrcReg = R_0288D4_SQ_PGM_RESOURCES_LS; break;
     case ShaderType::GEOMETRY: RsrcReg = R_028878_SQ_PGM_RESOURCES_GS; break;
@@ -145,7 +213,7 @@ void AMDGPUAsmPrinter::EmitProgramInfoR600(MachineFunction &MF) {
     }
   } else {
     // R600 / R700
-    switch (MFI->ShaderType) {
+    switch (MFI->getShaderType()) {
     default: // Fall through
     case ShaderType::GEOMETRY: // Fall through
     case ShaderType::COMPUTE:  // Fall through
@@ -160,40 +228,38 @@ void AMDGPUAsmPrinter::EmitProgramInfoR600(MachineFunction &MF) {
   OutStreamer.EmitIntValue(R_02880C_DB_SHADER_CONTROL, 4);
   OutStreamer.EmitIntValue(S_02880C_KILL_ENABLE(killPixel), 4);
 
-  if (MFI->ShaderType == ShaderType::COMPUTE) {
+  if (MFI->getShaderType() == ShaderType::COMPUTE) {
     OutStreamer.EmitIntValue(R_0288E8_SQ_LDS_ALLOC, 4);
     OutStreamer.EmitIntValue(RoundUpToAlignment(MFI->LDSSize, 4) >> 2, 4);
   }
 }
 
-void AMDGPUAsmPrinter::EmitProgramInfoSI(MachineFunction &MF) {
-  const AMDGPUSubtarget &STM = TM.getSubtarget<AMDGPUSubtarget>();
+void AMDGPUAsmPrinter::getSIProgramInfo(SIProgramInfo &ProgInfo,
+                                        const MachineFunction &MF) const {
+  uint64_t CodeSize = 0;
   unsigned MaxSGPR = 0;
   unsigned MaxVGPR = 0;
   bool VCCUsed = false;
-  const SIRegisterInfo * RI =
-                static_cast<const SIRegisterInfo*>(TM.getRegisterInfo());
+  const SIRegisterInfo *RI
+    = static_cast<const SIRegisterInfo*>(TM.getRegisterInfo());
 
-  for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
-                                                  BB != BB_E; ++BB) {
-    MachineBasicBlock &MBB = *BB;
-    for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
-                                                    I != E; ++I) {
-      MachineInstr &MI = *I;
+  for (const MachineBasicBlock &MBB : MF) {
+    for (const MachineInstr &MI : MBB) {
+      // TODO: CodeSize should account for multiple functions.
+      CodeSize += MI.getDesc().Size;
 
       unsigned numOperands = MI.getNumOperands();
       for (unsigned op_idx = 0; op_idx < numOperands; op_idx++) {
-        MachineOperand &MO = MI.getOperand(op_idx);
-        unsigned maxUsed;
+        const MachineOperand &MO = MI.getOperand(op_idx);
         unsigned width = 0;
         bool isSGPR = false;
-        unsigned reg;
-        unsigned hwReg;
+
         if (!MO.isReg()) {
           continue;
         }
-        reg = MO.getReg();
-        if (reg == AMDGPU::VCC) {
+        unsigned reg = MO.getReg();
+        if (reg == AMDGPU::VCC || reg == AMDGPU::VCC_LO ||
+	    reg == AMDGPU::VCC_HI) {
           VCCUsed = true;
           continue;
         }
@@ -240,10 +306,10 @@ void AMDGPUAsmPrinter::EmitProgramInfoSI(MachineFunction &MF) {
           isSGPR = false;
           width = 16;
         } else {
-          assert(!"Unknown register class");
+          llvm_unreachable("Unknown register class");
         }
-        hwReg = RI->getEncodingValue(reg) & 0xff;
-        maxUsed = hwReg + width - 1;
+        unsigned hwReg = RI->getEncodingValue(reg) & 0xff;
+        unsigned maxUsed = hwReg + width - 1;
         if (isSGPR) {
           MaxSGPR = maxUsed > MaxSGPR ? maxUsed : MaxSGPR;
         } else {
@@ -252,12 +318,36 @@ void AMDGPUAsmPrinter::EmitProgramInfoSI(MachineFunction &MF) {
       }
     }
   }
-  if (VCCUsed) {
+
+  if (VCCUsed)
     MaxSGPR += 2;
-  }
-  SIMachineFunctionInfo * MFI = MF.getInfo<SIMachineFunctionInfo>();
+
+  ProgInfo.NumVGPR = MaxVGPR;
+  ProgInfo.NumSGPR = MaxSGPR;
+
+  // Set the value to initialize FP_ROUND and FP_DENORM parts of the mode
+  // register.
+  ProgInfo.FloatMode = getFPMode(MF);
+
+  // XXX: Not quite sure what this does, but sc seems to unset this.
+  ProgInfo.IEEEMode = 0;
+
+  // Do not clamp NAN to 0.
+  ProgInfo.DX10Clamp = 0;
+
+  const MachineFrameInfo *FrameInfo = MF.getFrameInfo();
+  ProgInfo.ScratchSize = FrameInfo->estimateStackSize(MF);
+
+  ProgInfo.CodeLen = CodeSize;
+}
+
+void AMDGPUAsmPrinter::EmitProgramInfoSI(const MachineFunction &MF,
+                                         const SIProgramInfo &KernelInfo) {
+  const AMDGPUSubtarget &STM = TM.getSubtarget<AMDGPUSubtarget>();
+  const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
+
   unsigned RsrcReg;
-  switch (MFI->ShaderType) {
+  switch (MFI->getShaderType()) {
   default: // Fall through
   case ShaderType::COMPUTE:  RsrcReg = R_00B848_COMPUTE_PGM_RSRC1; break;
   case ShaderType::GEOMETRY: RsrcReg = R_00B228_SPI_SHADER_PGM_RSRC1_GS; break;
@@ -265,25 +355,58 @@ void AMDGPUAsmPrinter::EmitProgramInfoSI(MachineFunction &MF) {
   case ShaderType::VERTEX:   RsrcReg = R_00B128_SPI_SHADER_PGM_RSRC1_VS; break;
   }
 
-  OutStreamer.EmitIntValue(RsrcReg, 4);
-  OutStreamer.EmitIntValue(S_00B028_VGPRS(MaxVGPR / 4) | S_00B028_SGPRS(MaxSGPR / 8), 4);
-
   unsigned LDSAlignShift;
   if (STM.getGeneration() < AMDGPUSubtarget::SEA_ISLANDS) {
-    // LDS is allocated in 64 dword blocks
+    // LDS is allocated in 64 dword blocks.
     LDSAlignShift = 8;
   } else {
-    // LDS is allocated in 128 dword blocks
+    // LDS is allocated in 128 dword blocks.
     LDSAlignShift = 9;
   }
+
   unsigned LDSBlocks =
-          RoundUpToAlignment(MFI->LDSSize, 1 << LDSAlignShift) >> LDSAlignShift;
+    RoundUpToAlignment(MFI->LDSSize, 1 << LDSAlignShift) >> LDSAlignShift;
+
+  // Scratch is allocated in 256 dword blocks.
+  unsigned ScratchAlignShift = 10;
+  // We need to program the hardware with the amount of scratch memory that
+  // is used by the entire wave.  KernelInfo.ScratchSize is the amount of
+  // scratch memory used per thread.
+  unsigned ScratchBlocks =
+    RoundUpToAlignment(KernelInfo.ScratchSize * STM.getWavefrontSize(),
+                       1 << ScratchAlignShift) >> ScratchAlignShift;
+
+  if (MFI->getShaderType() == ShaderType::COMPUTE) {
+    OutStreamer.EmitIntValue(R_00B848_COMPUTE_PGM_RSRC1, 4);
+
+    const uint32_t ComputePGMRSrc1 =
+      S_00B848_VGPRS(KernelInfo.NumVGPR / 4) |
+      S_00B848_SGPRS(KernelInfo.NumSGPR / 8) |
+      S_00B848_PRIORITY(KernelInfo.Priority) |
+      S_00B848_FLOAT_MODE(KernelInfo.FloatMode) |
+      S_00B848_PRIV(KernelInfo.Priv) |
+      S_00B848_DX10_CLAMP(KernelInfo.DX10Clamp) |
+      S_00B848_IEEE_MODE(KernelInfo.DebugMode) |
+      S_00B848_IEEE_MODE(KernelInfo.IEEEMode);
+
+    OutStreamer.EmitIntValue(ComputePGMRSrc1, 4);
 
-  if (MFI->ShaderType == ShaderType::COMPUTE) {
     OutStreamer.EmitIntValue(R_00B84C_COMPUTE_PGM_RSRC2, 4);
-    OutStreamer.EmitIntValue(S_00B84C_LDS_SIZE(LDSBlocks), 4);
+    const uint32_t ComputePGMRSrc2 =
+      S_00B84C_LDS_SIZE(LDSBlocks) |
+      S_00B02C_SCRATCH_EN(ScratchBlocks > 0);
+
+    OutStreamer.EmitIntValue(ComputePGMRSrc2, 4);
+
+    OutStreamer.EmitIntValue(R_00B860_COMPUTE_TMPRING_SIZE, 4);
+    OutStreamer.EmitIntValue(S_00B860_WAVESIZE(ScratchBlocks), 4);
+  } else {
+    OutStreamer.EmitIntValue(RsrcReg, 4);
+    OutStreamer.EmitIntValue(S_00B028_VGPRS(KernelInfo.NumVGPR / 4) |
+                             S_00B028_SGPRS(KernelInfo.NumSGPR / 8), 4);
   }
-  if (MFI->ShaderType == ShaderType::PIXEL) {
+
+  if (MFI->getShaderType() == ShaderType::PIXEL) {
     OutStreamer.EmitIntValue(R_00B02C_SPI_SHADER_PGM_RSRC2_PS, 4);
     OutStreamer.EmitIntValue(S_00B02C_EXTRA_LDS_SIZE(LDSBlocks), 4);
     OutStreamer.EmitIntValue(R_0286CC_SPI_PS_INPUT_ENA, 4);
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUAsmPrinter.h b/contrib/llvm/lib/Target/R600/AMDGPUAsmPrinter.h
index 05dc9bb..19907cf 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUAsmPrinter.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPUAsmPrinter.h
@@ -16,29 +16,63 @@
 #define AMDGPU_ASMPRINTER_H
 
 #include "llvm/CodeGen/AsmPrinter.h"
-#include <string>
 #include <vector>
 
 namespace llvm {
 
 class AMDGPUAsmPrinter : public AsmPrinter {
+private:
+  struct SIProgramInfo {
+    SIProgramInfo() :
+      NumVGPR(0),
+      NumSGPR(0),
+      Priority(0),
+      FloatMode(0),
+      Priv(0),
+      DX10Clamp(0),
+      DebugMode(0),
+      IEEEMode(0),
+      ScratchSize(0),
+      CodeLen(0) {}
+
+    // Fields set in PGM_RSRC1 pm4 packet.
+    uint32_t NumVGPR;
+    uint32_t NumSGPR;
+    uint32_t Priority;
+    uint32_t FloatMode;
+    uint32_t Priv;
+    uint32_t DX10Clamp;
+    uint32_t DebugMode;
+    uint32_t IEEEMode;
+    uint32_t ScratchSize;
+
+    // Bonus information for debugging.
+    uint64_t CodeLen;
+  };
+
+  void getSIProgramInfo(SIProgramInfo &Out, const MachineFunction &MF) const;
+  void findNumUsedRegistersSI(const MachineFunction &MF,
+                              unsigned &NumSGPR,
+                              unsigned &NumVGPR) const;
+
+  /// \brief Emit register usage information so that the GPU driver
+  /// can correctly setup the GPU state.
+  void EmitProgramInfoR600(const MachineFunction &MF);
+  void EmitProgramInfoSI(const MachineFunction &MF, const SIProgramInfo &KernelInfo);
 
 public:
   explicit AMDGPUAsmPrinter(TargetMachine &TM, MCStreamer &Streamer);
 
-  virtual bool runOnMachineFunction(MachineFunction &MF);
+  bool runOnMachineFunction(MachineFunction &MF) override;
 
-  virtual const char *getPassName() const {
+  const char *getPassName() const override {
     return "AMDGPU Assembly Printer";
   }
 
-  /// \brief Emit register usage information so that the GPU driver
-  /// can correctly setup the GPU state.
-  void EmitProgramInfoR600(MachineFunction &MF);
-  void EmitProgramInfoSI(MachineFunction &MF);
-
   /// Implemented in AMDGPUMCInstLower.cpp
-  virtual void EmitInstruction(const MachineInstr *MI);
+  void EmitInstruction(const MachineInstr *MI) override;
+
+  void EmitEndOfAsmFile(Module &M) override;
 
 protected:
   bool DisasmEnabled;
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUCallingConv.td b/contrib/llvm/lib/Target/R600/AMDGPUCallingConv.td
index 65cdb24..3586c88 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUCallingConv.td
+++ b/contrib/llvm/lib/Target/R600/AMDGPUCallingConv.td
@@ -20,7 +20,7 @@ def CC_SI : CallingConv<[
   CCIfInReg<CCIfType<[f32, i32] , CCAssignToReg<[
     SGPR0, SGPR1, SGPR2, SGPR3, SGPR4, SGPR5, SGPR6, SGPR7,
     SGPR8, SGPR9, SGPR10, SGPR11, SGPR12, SGPR13, SGPR14, SGPR15,
-    SGPR16
+    SGPR16, SGPR17, SGPR18, SGPR19, SGPR20, SGPR21
   ]>>>,
 
   CCIfInReg<CCIfType<[i64] , CCAssignToRegWithShadow<
@@ -62,11 +62,11 @@ def CC_AMDGPU : CallingConv<[
   CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>().getGeneration() >= "
        "AMDGPUSubtarget::SOUTHERN_ISLANDS && "
        "State.getMachineFunction().getInfo<SIMachineFunctionInfo>()->"#
-       "ShaderType == ShaderType::COMPUTE", CCDelegateTo<CC_AMDGPU_Kernel>>,
+       "getShaderType() == ShaderType::COMPUTE", CCDelegateTo<CC_AMDGPU_Kernel>>,
   CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>().getGeneration() < "
        "AMDGPUSubtarget::SOUTHERN_ISLANDS && "
        "State.getMachineFunction().getInfo<R600MachineFunctionInfo>()->"
-       "ShaderType == ShaderType::COMPUTE", CCDelegateTo<CC_AMDGPU_Kernel>>,
+       "getShaderType() == ShaderType::COMPUTE", CCDelegateTo<CC_AMDGPU_Kernel>>,
   CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>()"#
        ".getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS", CCDelegateTo<CC_SI>>,
   CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>()"#
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUConvertToISA.cpp b/contrib/llvm/lib/Target/R600/AMDGPUConvertToISA.cpp
deleted file mode 100644
index 50297d1..0000000
--- a/contrib/llvm/lib/Target/R600/AMDGPUConvertToISA.cpp
+++ /dev/null
@@ -1,62 +0,0 @@
-//===-- AMDGPUConvertToISA.cpp - Lower AMDIL to HW ISA --------------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-/// \file
-/// \brief This pass lowers AMDIL machine instructions to the appropriate
-/// hardware instructions.
-//
-//===----------------------------------------------------------------------===//
-
-#include "AMDGPU.h"
-#include "AMDGPUInstrInfo.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
-
-using namespace llvm;
-
-namespace {
-
-class AMDGPUConvertToISAPass : public MachineFunctionPass {
-
-private:
-  static char ID;
-  TargetMachine &TM;
-
-public:
-  AMDGPUConvertToISAPass(TargetMachine &tm) :
-    MachineFunctionPass(ID), TM(tm) { }
-
-  virtual bool runOnMachineFunction(MachineFunction &MF);
-
-  virtual const char *getPassName() const {return "AMDGPU Convert to ISA";}
-
-};
-
-} // End anonymous namespace
-
-char AMDGPUConvertToISAPass::ID = 0;
-
-FunctionPass *llvm::createAMDGPUConvertToISAPass(TargetMachine &tm) {
-  return new AMDGPUConvertToISAPass(tm);
-}
-
-bool AMDGPUConvertToISAPass::runOnMachineFunction(MachineFunction &MF) {
-  const AMDGPUInstrInfo * TII =
-                      static_cast<const AMDGPUInstrInfo*>(TM.getInstrInfo());
-
-  for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
-                                                  BB != BB_E; ++BB) {
-    MachineBasicBlock &MBB = *BB;
-    for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
-                                                      I != E; ++I) {
-      MachineInstr &MI = *I;
-      TII->convertToISA(MI, MF, MBB.findDebugLoc(I));
-    }
-  }
-  return false;
-}
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUFrameLowering.cpp b/contrib/llvm/lib/Target/R600/AMDGPUFrameLowering.cpp
index 40f14d2..9e8302e 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUFrameLowering.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUFrameLowering.cpp
@@ -74,20 +74,30 @@ unsigned AMDGPUFrameLowering::getStackWidth(const MachineFunction &MF) const {
 int AMDGPUFrameLowering::getFrameIndexOffset(const MachineFunction &MF,
                                          int FI) const {
   const MachineFrameInfo *MFI = MF.getFrameInfo();
-  unsigned Offset = 0;
+  // Start the offset at 2 so we don't overwrite work group information.
+  // XXX: We should only do this when the shader actually uses this
+  // information.
+  unsigned OffsetBytes = 2 * (getStackWidth(MF) * 4);
   int UpperBound = FI == -1 ? MFI->getNumObjects() : FI;
 
   for (int i = MFI->getObjectIndexBegin(); i < UpperBound; ++i) {
-    unsigned Size = MFI->getObjectSize(i);
-    Offset += (Size / (getStackWidth(MF) * 4));
+    OffsetBytes = RoundUpToAlignment(OffsetBytes, MFI->getObjectAlignment(i));
+    OffsetBytes += MFI->getObjectSize(i);
+    // Each register holds 4 bytes, so we must always align the offset to at
+    // least 4 bytes, so that 2 frame objects won't share the same register.
+    OffsetBytes = RoundUpToAlignment(OffsetBytes, 4);
   }
-  return Offset;
+
+  if (FI != -1)
+    OffsetBytes = RoundUpToAlignment(OffsetBytes, MFI->getObjectAlignment(FI));
+
+  return OffsetBytes / (getStackWidth(MF) * 4);
 }
 
 const TargetFrameLowering::SpillSlot *
 AMDGPUFrameLowering::getCalleeSavedSpillSlots(unsigned &NumEntries) const {
   NumEntries = 0;
-  return 0;
+  return nullptr;
 }
 void
 AMDGPUFrameLowering::emitPrologue(MachineFunction &MF) const {
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUFrameLowering.h b/contrib/llvm/lib/Target/R600/AMDGPUFrameLowering.h
index cf5742e..d18ede5 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUFrameLowering.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPUFrameLowering.h
@@ -33,12 +33,13 @@ public:
 
   /// \returns The number of 32-bit sub-registers that are used when storing
   /// values to the stack.
-  virtual unsigned getStackWidth(const MachineFunction &MF) const;
-  virtual int getFrameIndexOffset(const MachineFunction &MF, int FI) const;
-  virtual const SpillSlot *getCalleeSavedSpillSlots(unsigned &NumEntries) const;
-  virtual void emitPrologue(MachineFunction &MF) const;
-  virtual void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
-  virtual bool hasFP(const MachineFunction &MF) const;
+  unsigned getStackWidth(const MachineFunction &MF) const;
+  int getFrameIndexOffset(const MachineFunction &MF, int FI) const override;
+  const SpillSlot *
+    getCalleeSavedSpillSlots(unsigned &NumEntries) const override;
+  void emitPrologue(MachineFunction &MF) const override;
+  void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const override;
+  bool hasFP(const MachineFunction &MF) const override;
 };
 } // namespace llvm
 #endif // AMDILFRAME_LOWERING_H
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUISelDAGToDAG.cpp b/contrib/llvm/lib/Target/R600/AMDGPUISelDAGToDAG.cpp
index a989135..cc17b7e 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUISelDAGToDAG.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUISelDAGToDAG.cpp
@@ -14,17 +14,18 @@
 #include "AMDGPUInstrInfo.h"
 #include "AMDGPUISelLowering.h" // For AMDGPUISD
 #include "AMDGPURegisterInfo.h"
+#include "AMDGPUSubtarget.h"
 #include "R600InstrInfo.h"
+#include "SIDefines.h"
 #include "SIISelLowering.h"
-#include "llvm/ADT/ValueMap.h"
-#include "llvm/Analysis/ValueTracking.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "SIMachineFunctionInfo.h"
+#include "llvm/CodeGen/FunctionLoweringInfo.h"
 #include "llvm/CodeGen/PseudoSourceValue.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
-#include "llvm/Support/Compiler.h"
-#include <list>
-#include <queue>
+#include "llvm/IR/Function.h"
 
 using namespace llvm;
 
@@ -43,11 +44,12 @@ public:
   AMDGPUDAGToDAGISel(TargetMachine &TM);
   virtual ~AMDGPUDAGToDAGISel();
 
-  SDNode *Select(SDNode *N);
-  virtual const char *getPassName() const;
-  virtual void PostprocessISelDAG();
+  SDNode *Select(SDNode *N) override;
+  const char *getPassName() const override;
+  void PostprocessISelDAG() override;
 
 private:
+  bool isInlineImmediate(SDNode *N) const;
   inline SDValue getSmallIPtrImm(unsigned Imm);
   bool FoldOperand(SDValue &Src, SDValue &Sel, SDValue &Neg, SDValue &Abs,
                    const R600InstrInfo *TII);
@@ -58,11 +60,9 @@ private:
   bool SelectADDRParam(SDValue Addr, SDValue& R1, SDValue& R2);
   bool SelectADDR(SDValue N, SDValue &R1, SDValue &R2);
   bool SelectADDR64(SDValue N, SDValue &R1, SDValue &R2);
-  SDValue SimplifyI24(SDValue &Op);
-  bool SelectI24(SDValue Addr, SDValue &Op);
-  bool SelectU24(SDValue Addr, SDValue &Op);
 
   static bool checkType(const Value *ptr, unsigned int addrspace);
+  static bool checkPrivateAddress(const MachineMemOperand *Op);
 
   static bool isGlobalStore(const StoreSDNode *N);
   static bool isPrivateStore(const StoreSDNode *N);
@@ -77,12 +77,28 @@ private:
   bool isLocalLoad(const LoadSDNode *N) const;
   bool isRegionLoad(const LoadSDNode *N) const;
 
+  /// \returns True if the current basic block being selected is at control
+  ///          flow depth 0.  Meaning that the current block dominates the
+  //           exit block.
+  bool isCFDepth0() const;
+
   const TargetRegisterClass *getOperandRegClass(SDNode *N, unsigned OpNo) const;
   bool SelectGlobalValueConstantOffset(SDValue Addr, SDValue& IntPtr);
-  bool SelectGlobalValueVariableOffset(SDValue Addr,
-      SDValue &BaseReg, SDValue& Offset);
+  bool SelectGlobalValueVariableOffset(SDValue Addr, SDValue &BaseReg,
+                                       SDValue& Offset);
   bool SelectADDRVTX_READ(SDValue Addr, SDValue &Base, SDValue &Offset);
   bool SelectADDRIndirect(SDValue Addr, SDValue &Base, SDValue &Offset);
+  bool SelectMUBUFAddr64(SDValue Addr, SDValue &Ptr, SDValue &Offset,
+                         SDValue &ImmOffset) const;
+  bool SelectMUBUFScratch(SDValue Addr, SDValue &RSrc, SDValue &VAddr,
+                          SDValue &SOffset, SDValue &ImmOffset) const;
+  bool SelectMUBUFAddr32(SDValue Addr, SDValue &SRsrc, SDValue &VAddr,
+                         SDValue &SOffset, SDValue &Offset, SDValue &Offen,
+                         SDValue &Idxen, SDValue &GLC, SDValue &SLC,
+                         SDValue &TFE) const;
+
+  SDNode *SelectADD_SUB_I64(SDNode *N);
+  SDNode *SelectDIV_SCALE(SDNode *N);
 
   // Include the pieces autogenerated from the target description.
 #include "AMDGPUGenDAGISel.inc"
@@ -91,8 +107,7 @@ private:
 
 /// \brief This pass converts a legalized DAG into a AMDGPU-specific
 // DAG, ready for instruction scheduling.
-FunctionPass *llvm::createAMDGPUISelDag(TargetMachine &TM
-                                       ) {
+FunctionPass *llvm::createAMDGPUISelDag(TargetMachine &TM) {
   return new AMDGPUDAGToDAGISel(TM);
 }
 
@@ -103,32 +118,39 @@ AMDGPUDAGToDAGISel::AMDGPUDAGToDAGISel(TargetMachine &TM)
 AMDGPUDAGToDAGISel::~AMDGPUDAGToDAGISel() {
 }
 
+bool AMDGPUDAGToDAGISel::isInlineImmediate(SDNode *N) const {
+  const SITargetLowering *TL
+      = static_cast<const SITargetLowering *>(getTargetLowering());
+  return TL->analyzeImmediate(N) == 0;
+}
+
 /// \brief Determine the register class for \p OpNo
 /// \returns The register class of the virtual register that will be used for
 /// the given operand number \OpNo or NULL if the register class cannot be
 /// determined.
 const TargetRegisterClass *AMDGPUDAGToDAGISel::getOperandRegClass(SDNode *N,
                                                           unsigned OpNo) const {
-  if (!N->isMachineOpcode()) {
-    return NULL;
-  }
+  if (!N->isMachineOpcode())
+    return nullptr;
+
   switch (N->getMachineOpcode()) {
   default: {
     const MCInstrDesc &Desc = TM.getInstrInfo()->get(N->getMachineOpcode());
     unsigned OpIdx = Desc.getNumDefs() + OpNo;
     if (OpIdx >= Desc.getNumOperands())
-      return NULL;
+      return nullptr;
     int RegClass = Desc.OpInfo[OpIdx].RegClass;
-    if (RegClass == -1) {
-      return NULL;
-    }
+    if (RegClass == -1)
+      return nullptr;
+
     return TM.getRegisterInfo()->getRegClass(RegClass);
   }
   case AMDGPU::REG_SEQUENCE: {
-    const TargetRegisterClass *SuperRC = TM.getRegisterInfo()->getRegClass(
-                      cast<ConstantSDNode>(N->getOperand(0))->getZExtValue());
-    unsigned SubRegIdx =
-            dyn_cast<ConstantSDNode>(N->getOperand(OpNo + 1))->getZExtValue();
+    unsigned RCID = cast<ConstantSDNode>(N->getOperand(0))->getZExtValue();
+    const TargetRegisterClass *SuperRC = TM.getRegisterInfo()->getRegClass(RCID);
+
+    SDValue SubRegOp = N->getOperand(OpNo + 1);
+    unsigned SubRegIdx = cast<ConstantSDNode>(SubRegOp)->getZExtValue();
     return TM.getRegisterInfo()->getSubClassWithSubReg(SuperRC, SubRegIdx);
   }
   }
@@ -139,7 +161,7 @@ SDValue AMDGPUDAGToDAGISel::getSmallIPtrImm(unsigned int Imm) {
 }
 
 bool AMDGPUDAGToDAGISel::SelectADDRParam(
-    SDValue Addr, SDValue& R1, SDValue& R2) {
+  SDValue Addr, SDValue& R1, SDValue& R2) {
 
   if (Addr.getOpcode() == ISD::FrameIndex) {
     if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) {
@@ -196,20 +218,35 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
   unsigned int Opc = N->getOpcode();
   if (N->isMachineOpcode()) {
     N->setNodeId(-1);
-    return NULL;   // Already selected.
+    return nullptr;   // Already selected.
   }
+
+  const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
   switch (Opc) {
   default: break;
+  // We are selecting i64 ADD here instead of custom lower it during
+  // DAG legalization, so we can fold some i64 ADDs used for address
+  // calculation into the LOAD and STORE instructions.
+  case ISD::ADD:
+  case ISD::SUB: {
+    if (N->getValueType(0) != MVT::i64 ||
+        ST.getGeneration() < AMDGPUSubtarget::SOUTHERN_ISLANDS)
+      break;
+
+    return SelectADD_SUB_I64(N);
+  }
+  case ISD::SCALAR_TO_VECTOR:
+  case AMDGPUISD::BUILD_VERTICAL_VECTOR:
   case ISD::BUILD_VECTOR: {
     unsigned RegClassID;
-    const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
     const AMDGPURegisterInfo *TRI =
                    static_cast<const AMDGPURegisterInfo*>(TM.getRegisterInfo());
     const SIRegisterInfo *SIRI =
                    static_cast<const SIRegisterInfo*>(TM.getRegisterInfo());
     EVT VT = N->getValueType(0);
     unsigned NumVectorElts = VT.getVectorNumElements();
-    assert(VT.getVectorElementType().bitsEq(MVT::i32));
+    EVT EltVT = VT.getVectorElementType();
+    assert(EltVT.bitsEq(MVT::i32));
     if (ST.getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS) {
       bool UseVReg = true;
       for (SDNode::use_iterator U = N->use_begin(), E = SDNode::use_end();
@@ -250,7 +287,12 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
       // can't be bundled by our scheduler.
       switch(NumVectorElts) {
       case 2: RegClassID = AMDGPU::R600_Reg64RegClassID; break;
-      case 4: RegClassID = AMDGPU::R600_Reg128RegClassID; break;
+      case 4:
+        if (Opc == AMDGPUISD::BUILD_VERTICAL_VECTOR)
+          RegClassID = AMDGPU::R600_Reg128VerticalRegClassID;
+        else
+          RegClassID = AMDGPU::R600_Reg128RegClassID;
+        break;
       default: llvm_unreachable("Do not know how to lower this BUILD_VECTOR");
       }
     }
@@ -258,8 +300,7 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
     SDValue RegClass = CurDAG->getTargetConstant(RegClassID, MVT::i32);
 
     if (NumVectorElts == 1) {
-      return CurDAG->SelectNodeTo(N, AMDGPU::COPY_TO_REGCLASS,
-                                  VT.getVectorElementType(),
+      return CurDAG->SelectNodeTo(N, AMDGPU::COPY_TO_REGCLASS, EltVT,
                                   N->getOperand(0), RegClass);
     }
 
@@ -268,11 +309,12 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
     // 16 = Max Num Vector Elements
     // 2 = 2 REG_SEQUENCE operands per element (value, subreg index)
     // 1 = Vector Register Class
-    SDValue RegSeqArgs[16 * 2 + 1];
+    SmallVector<SDValue, 16 * 2 + 1> RegSeqArgs(NumVectorElts * 2 + 1);
 
     RegSeqArgs[0] = CurDAG->getTargetConstant(RegClassID, MVT::i32);
     bool IsRegSeq = true;
-    for (unsigned i = 0; i < N->getNumOperands(); i++) {
+    unsigned NOps = N->getNumOperands();
+    for (unsigned i = 0; i < NOps; i++) {
       // XXX: Why is this here?
       if (dyn_cast<RegisterSDNode>(N->getOperand(i))) {
         IsRegSeq = false;
@@ -282,14 +324,27 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
       RegSeqArgs[1 + (2 * i) + 1] =
               CurDAG->getTargetConstant(TRI->getSubRegFromChannel(i), MVT::i32);
     }
+
+    if (NOps != NumVectorElts) {
+      // Fill in the missing undef elements if this was a scalar_to_vector.
+      assert(Opc == ISD::SCALAR_TO_VECTOR && NOps < NumVectorElts);
+
+      MachineSDNode *ImpDef = CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF,
+                                                     SDLoc(N), EltVT);
+      for (unsigned i = NOps; i < NumVectorElts; ++i) {
+        RegSeqArgs[1 + (2 * i)] = SDValue(ImpDef, 0);
+        RegSeqArgs[1 + (2 * i) + 1] =
+          CurDAG->getTargetConstant(TRI->getSubRegFromChannel(i), MVT::i32);
+      }
+    }
+
     if (!IsRegSeq)
       break;
     return CurDAG->SelectNodeTo(N, AMDGPU::REG_SEQUENCE, N->getVTList(),
-        RegSeqArgs, 2 * N->getNumOperands() + 1);
+                                RegSeqArgs);
   }
   case ISD::BUILD_PAIR: {
     SDValue RC, SubReg0, SubReg1;
-    const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
     if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
       break;
     }
@@ -298,7 +353,7 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
       SubReg0 = CurDAG->getTargetConstant(AMDGPU::sub0_sub1, MVT::i32);
       SubReg1 = CurDAG->getTargetConstant(AMDGPU::sub2_sub3, MVT::i32);
     } else if (N->getValueType(0) == MVT::i64) {
-      RC = CurDAG->getTargetConstant(AMDGPU::VSrc_64RegClassID, MVT::i32);
+      RC = CurDAG->getTargetConstant(AMDGPU::SReg_64RegClassID, MVT::i32);
       SubReg0 = CurDAG->getTargetConstant(AMDGPU::sub0, MVT::i32);
       SubReg1 = CurDAG->getTargetConstant(AMDGPU::sub1, MVT::i32);
     } else {
@@ -309,8 +364,37 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
     return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE,
                                   SDLoc(N), N->getValueType(0), Ops);
   }
-  case AMDGPUISD::REGISTER_LOAD: {
+
+  case ISD::Constant:
+  case ISD::ConstantFP: {
     const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
+    if (ST.getGeneration() < AMDGPUSubtarget::SOUTHERN_ISLANDS ||
+        N->getValueType(0).getSizeInBits() != 64 || isInlineImmediate(N))
+      break;
+
+    uint64_t Imm;
+    if (ConstantFPSDNode *FP = dyn_cast<ConstantFPSDNode>(N))
+      Imm = FP->getValueAPF().bitcastToAPInt().getZExtValue();
+    else {
+      ConstantSDNode *C = cast<ConstantSDNode>(N);
+      Imm = C->getZExtValue();
+    }
+
+    SDNode *Lo = CurDAG->getMachineNode(AMDGPU::S_MOV_B32, SDLoc(N), MVT::i32,
+                                CurDAG->getConstant(Imm & 0xFFFFFFFF, MVT::i32));
+    SDNode *Hi = CurDAG->getMachineNode(AMDGPU::S_MOV_B32, SDLoc(N), MVT::i32,
+                                CurDAG->getConstant(Imm >> 32, MVT::i32));
+    const SDValue Ops[] = {
+      CurDAG->getTargetConstant(AMDGPU::SReg_64RegClassID, MVT::i32),
+      SDValue(Lo, 0), CurDAG->getTargetConstant(AMDGPU::sub0, MVT::i32),
+      SDValue(Hi, 0), CurDAG->getTargetConstant(AMDGPU::sub1, MVT::i32)
+    };
+
+    return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, SDLoc(N),
+                                  N->getValueType(0), Ops);
+  }
+
+  case AMDGPUISD::REGISTER_LOAD: {
     if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS)
       break;
     SDValue Addr, Offset;
@@ -327,7 +411,6 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
                                   Ops);
   }
   case AMDGPUISD::REGISTER_STORE: {
-    const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
     if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS)
       break;
     SDValue Addr, Offset;
@@ -343,42 +426,98 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
                                         CurDAG->getVTList(MVT::Other),
                                         Ops);
   }
+
+  case AMDGPUISD::BFE_I32:
+  case AMDGPUISD::BFE_U32: {
+    if (ST.getGeneration() < AMDGPUSubtarget::SOUTHERN_ISLANDS)
+      break;
+
+    // There is a scalar version available, but unlike the vector version which
+    // has a separate operand for the offset and width, the scalar version packs
+    // the width and offset into a single operand. Try to move to the scalar
+    // version if the offsets are constant, so that we can try to keep extended
+    // loads of kernel arguments in SGPRs.
+
+    // TODO: Technically we could try to pattern match scalar bitshifts of
+    // dynamic values, but it's probably not useful.
+    ConstantSDNode *Offset = dyn_cast<ConstantSDNode>(N->getOperand(1));
+    if (!Offset)
+      break;
+
+    ConstantSDNode *Width = dyn_cast<ConstantSDNode>(N->getOperand(2));
+    if (!Width)
+      break;
+
+    bool Signed = Opc == AMDGPUISD::BFE_I32;
+
+    // Transformation function, pack the offset and width of a BFE into
+    // the format expected by the S_BFE_I32 / S_BFE_U32. In the second
+    // source, bits [5:0] contain the offset and bits [22:16] the width.
+
+    uint32_t OffsetVal = Offset->getZExtValue();
+    uint32_t WidthVal = Width->getZExtValue();
+
+    uint32_t PackedVal = OffsetVal | WidthVal << 16;
+
+    SDValue PackedOffsetWidth = CurDAG->getTargetConstant(PackedVal, MVT::i32);
+    return CurDAG->getMachineNode(Signed ? AMDGPU::S_BFE_I32 : AMDGPU::S_BFE_U32,
+                                  SDLoc(N),
+                                  MVT::i32,
+                                  N->getOperand(0),
+                                  PackedOffsetWidth);
+
+  }
+  case AMDGPUISD::DIV_SCALE: {
+    return SelectDIV_SCALE(N);
+  }
   }
   return SelectCode(N);
 }
 
 
-bool AMDGPUDAGToDAGISel::checkType(const Value *ptr, unsigned int addrspace) {
-  if (!ptr) {
+bool AMDGPUDAGToDAGISel::checkType(const Value *Ptr, unsigned AS) {
+  assert(AS != 0 && "Use checkPrivateAddress instead.");
+  if (!Ptr)
     return false;
-  }
-  Type *ptrType = ptr->getType();
-  return dyn_cast<PointerType>(ptrType)->getAddressSpace() == addrspace;
+
+  return Ptr->getType()->getPointerAddressSpace() == AS;
+}
+
+bool AMDGPUDAGToDAGISel::checkPrivateAddress(const MachineMemOperand *Op) {
+  if (Op->getPseudoValue())
+    return true;
+
+  if (PointerType *PT = dyn_cast<PointerType>(Op->getValue()->getType()))
+    return PT->getAddressSpace() == AMDGPUAS::PRIVATE_ADDRESS;
+
+  return false;
 }
 
 bool AMDGPUDAGToDAGISel::isGlobalStore(const StoreSDNode *N) {
-  return checkType(N->getSrcValue(), AMDGPUAS::GLOBAL_ADDRESS);
+  return checkType(N->getMemOperand()->getValue(), AMDGPUAS::GLOBAL_ADDRESS);
 }
 
 bool AMDGPUDAGToDAGISel::isPrivateStore(const StoreSDNode *N) {
-  return (!checkType(N->getSrcValue(), AMDGPUAS::LOCAL_ADDRESS)
-          && !checkType(N->getSrcValue(), AMDGPUAS::GLOBAL_ADDRESS)
-          && !checkType(N->getSrcValue(), AMDGPUAS::REGION_ADDRESS));
+  const Value *MemVal = N->getMemOperand()->getValue();
+  return (!checkType(MemVal, AMDGPUAS::LOCAL_ADDRESS) &&
+          !checkType(MemVal, AMDGPUAS::GLOBAL_ADDRESS) &&
+          !checkType(MemVal, AMDGPUAS::REGION_ADDRESS));
 }
 
 bool AMDGPUDAGToDAGISel::isLocalStore(const StoreSDNode *N) {
-  return checkType(N->getSrcValue(), AMDGPUAS::LOCAL_ADDRESS);
+  return checkType(N->getMemOperand()->getValue(), AMDGPUAS::LOCAL_ADDRESS);
 }
 
 bool AMDGPUDAGToDAGISel::isRegionStore(const StoreSDNode *N) {
-  return checkType(N->getSrcValue(), AMDGPUAS::REGION_ADDRESS);
+  return checkType(N->getMemOperand()->getValue(), AMDGPUAS::REGION_ADDRESS);
 }
 
 bool AMDGPUDAGToDAGISel::isConstantLoad(const LoadSDNode *N, int CbId) const {
-  if (CbId == -1) {
-    return checkType(N->getSrcValue(), AMDGPUAS::CONSTANT_ADDRESS);
-  }
-  return checkType(N->getSrcValue(), AMDGPUAS::CONSTANT_BUFFER_0 + CbId);
+  const Value *MemVal = N->getMemOperand()->getValue();
+  if (CbId == -1)
+    return checkType(MemVal, AMDGPUAS::CONSTANT_ADDRESS);
+
+  return checkType(MemVal, AMDGPUAS::CONSTANT_BUFFER_0 + CbId);
 }
 
 bool AMDGPUDAGToDAGISel::isGlobalLoad(const LoadSDNode *N) const {
@@ -389,27 +528,26 @@ bool AMDGPUDAGToDAGISel::isGlobalLoad(const LoadSDNode *N) const {
       return true;
     }
   }
-  return checkType(N->getSrcValue(), AMDGPUAS::GLOBAL_ADDRESS);
+  return checkType(N->getMemOperand()->getValue(), AMDGPUAS::GLOBAL_ADDRESS);
 }
 
 bool AMDGPUDAGToDAGISel::isParamLoad(const LoadSDNode *N) const {
-  return checkType(N->getSrcValue(), AMDGPUAS::PARAM_I_ADDRESS);
+  return checkType(N->getMemOperand()->getValue(), AMDGPUAS::PARAM_I_ADDRESS);
 }
 
 bool AMDGPUDAGToDAGISel::isLocalLoad(const  LoadSDNode *N) const {
-  return checkType(N->getSrcValue(), AMDGPUAS::LOCAL_ADDRESS);
+  return checkType(N->getMemOperand()->getValue(), AMDGPUAS::LOCAL_ADDRESS);
 }
 
 bool AMDGPUDAGToDAGISel::isRegionLoad(const  LoadSDNode *N) const {
-  return checkType(N->getSrcValue(), AMDGPUAS::REGION_ADDRESS);
+  return checkType(N->getMemOperand()->getValue(), AMDGPUAS::REGION_ADDRESS);
 }
 
 bool AMDGPUDAGToDAGISel::isCPLoad(const LoadSDNode *N) const {
   MachineMemOperand *MMO = N->getMemOperand();
-  if (checkType(N->getSrcValue(), AMDGPUAS::PRIVATE_ADDRESS)) {
+  if (checkPrivateAddress(N->getMemOperand())) {
     if (MMO) {
-      const Value *V = MMO->getValue();
-      const PseudoSourceValue *PSV = dyn_cast<PseudoSourceValue>(V);
+      const PseudoSourceValue *PSV = MMO->getPseudoValue();
       if (PSV && PSV == PseudoSourceValue::getConstantPool()) {
         return true;
       }
@@ -419,24 +557,34 @@ bool AMDGPUDAGToDAGISel::isCPLoad(const LoadSDNode *N) const {
 }
 
 bool AMDGPUDAGToDAGISel::isPrivateLoad(const LoadSDNode *N) const {
-  if (checkType(N->getSrcValue(), AMDGPUAS::PRIVATE_ADDRESS)) {
+  if (checkPrivateAddress(N->getMemOperand())) {
     // Check to make sure we are not a constant pool load or a constant load
     // that is marked as a private load
     if (isCPLoad(N) || isConstantLoad(N, -1)) {
       return false;
     }
   }
-  if (!checkType(N->getSrcValue(), AMDGPUAS::LOCAL_ADDRESS)
-      && !checkType(N->getSrcValue(), AMDGPUAS::GLOBAL_ADDRESS)
-      && !checkType(N->getSrcValue(), AMDGPUAS::REGION_ADDRESS)
-      && !checkType(N->getSrcValue(), AMDGPUAS::CONSTANT_ADDRESS)
-      && !checkType(N->getSrcValue(), AMDGPUAS::PARAM_D_ADDRESS)
-      && !checkType(N->getSrcValue(), AMDGPUAS::PARAM_I_ADDRESS)) {
+
+  const Value *MemVal = N->getMemOperand()->getValue();
+  if (!checkType(MemVal, AMDGPUAS::LOCAL_ADDRESS) &&
+      !checkType(MemVal, AMDGPUAS::GLOBAL_ADDRESS) &&
+      !checkType(MemVal, AMDGPUAS::REGION_ADDRESS) &&
+      !checkType(MemVal, AMDGPUAS::CONSTANT_ADDRESS) &&
+      !checkType(MemVal, AMDGPUAS::PARAM_D_ADDRESS) &&
+      !checkType(MemVal, AMDGPUAS::PARAM_I_ADDRESS)){
     return true;
   }
   return false;
 }
 
+bool AMDGPUDAGToDAGISel::isCFDepth0() const {
+  // FIXME: Figure out a way to use DominatorTree analysis here.
+  const BasicBlock *CurBlock = FuncInfo->MBB->getBasicBlock();
+  const Function *Fn = FuncInfo->Fn;
+  return &Fn->front() == CurBlock || &Fn->back() == CurBlock;
+}
+
+
 const char *AMDGPUDAGToDAGISel::getPassName() const {
   return "AMDGPU DAG->DAG Pattern Instruction Selection";
 }
@@ -451,7 +599,7 @@ const char *AMDGPUDAGToDAGISel::getPassName() const {
 //===----------------------------------------------------------------------===//
 
 bool AMDGPUDAGToDAGISel::SelectGlobalValueConstantOffset(SDValue Addr,
-    SDValue& IntPtr) {
+                                                         SDValue& IntPtr) {
   if (ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(Addr)) {
     IntPtr = CurDAG->getIntPtrConstant(Cst->getZExtValue() / 4, true);
     return true;
@@ -461,7 +609,7 @@ bool AMDGPUDAGToDAGISel::SelectGlobalValueConstantOffset(SDValue Addr,
 
 bool AMDGPUDAGToDAGISel::SelectGlobalValueVariableOffset(SDValue Addr,
     SDValue& BaseReg, SDValue &Offset) {
-  if (!dyn_cast<ConstantSDNode>(Addr)) {
+  if (!isa<ConstantSDNode>(Addr)) {
     BaseReg = Addr;
     Offset = CurDAG->getIntPtrConstant(0, true);
     return true;
@@ -471,7 +619,7 @@ bool AMDGPUDAGToDAGISel::SelectGlobalValueVariableOffset(SDValue Addr,
 
 bool AMDGPUDAGToDAGISel::SelectADDRVTX_READ(SDValue Addr, SDValue &Base,
                                            SDValue &Offset) {
-  ConstantSDNode * IMMOffset;
+  ConstantSDNode *IMMOffset;
 
   if (Addr.getOpcode() == ISD::ADD
       && (IMMOffset = dyn_cast<ConstantSDNode>(Addr.getOperand(1)))
@@ -515,52 +663,225 @@ bool AMDGPUDAGToDAGISel::SelectADDRIndirect(SDValue Addr, SDValue &Base,
   return true;
 }
 
-SDValue AMDGPUDAGToDAGISel::SimplifyI24(SDValue &Op) {
-  APInt Demanded = APInt(32, 0x00FFFFFF);
-  APInt KnownZero, KnownOne;
-  TargetLowering::TargetLoweringOpt TLO(*CurDAG, true, true);
-  const TargetLowering *TLI = getTargetLowering();
-  if (TLI->SimplifyDemandedBits(Op, Demanded, KnownZero, KnownOne, TLO)) {
-    CurDAG->ReplaceAllUsesWith(Op, TLO.New);
-    CurDAG->RepositionNode(Op.getNode(), TLO.New.getNode());
-    return SimplifyI24(TLO.New);
-  } else {
-    return  Op;
+SDNode *AMDGPUDAGToDAGISel::SelectADD_SUB_I64(SDNode *N) {
+  SDLoc DL(N);
+  SDValue LHS = N->getOperand(0);
+  SDValue RHS = N->getOperand(1);
+
+  bool IsAdd = (N->getOpcode() == ISD::ADD);
+
+  SDValue Sub0 = CurDAG->getTargetConstant(AMDGPU::sub0, MVT::i32);
+  SDValue Sub1 = CurDAG->getTargetConstant(AMDGPU::sub1, MVT::i32);
+
+  SDNode *Lo0 = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG,
+                                       DL, MVT::i32, LHS, Sub0);
+  SDNode *Hi0 = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG,
+                                       DL, MVT::i32, LHS, Sub1);
+
+  SDNode *Lo1 = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG,
+                                       DL, MVT::i32, RHS, Sub0);
+  SDNode *Hi1 = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG,
+                                       DL, MVT::i32, RHS, Sub1);
+
+  SDVTList VTList = CurDAG->getVTList(MVT::i32, MVT::Glue);
+  SDValue AddLoArgs[] = { SDValue(Lo0, 0), SDValue(Lo1, 0) };
+
+
+  unsigned Opc = IsAdd ? AMDGPU::S_ADD_I32 : AMDGPU::S_SUB_I32;
+  unsigned CarryOpc = IsAdd ? AMDGPU::S_ADDC_U32 : AMDGPU::S_SUBB_U32;
+
+  if (!isCFDepth0()) {
+    Opc = IsAdd ? AMDGPU::V_ADD_I32_e32 : AMDGPU::V_SUB_I32_e32;
+    CarryOpc = IsAdd ? AMDGPU::V_ADDC_U32_e32 : AMDGPU::V_SUBB_U32_e32;
   }
+
+  SDNode *AddLo = CurDAG->getMachineNode( Opc, DL, VTList, AddLoArgs);
+  SDValue Carry(AddLo, 1);
+  SDNode *AddHi
+    = CurDAG->getMachineNode(CarryOpc, DL, MVT::i32,
+                             SDValue(Hi0, 0), SDValue(Hi1, 0), Carry);
+
+  SDValue Args[5] = {
+    CurDAG->getTargetConstant(AMDGPU::SReg_64RegClassID, MVT::i32),
+    SDValue(AddLo,0),
+    Sub0,
+    SDValue(AddHi,0),
+    Sub1,
+  };
+  return CurDAG->SelectNodeTo(N, AMDGPU::REG_SEQUENCE, MVT::i64, Args);
+}
+
+SDNode *AMDGPUDAGToDAGISel::SelectDIV_SCALE(SDNode *N) {
+  SDLoc SL(N);
+  EVT VT = N->getValueType(0);
+
+  assert(VT == MVT::f32 || VT == MVT::f64);
+
+  unsigned Opc
+    = (VT == MVT::f64) ? AMDGPU::V_DIV_SCALE_F64 : AMDGPU::V_DIV_SCALE_F32;
+
+  const SDValue Zero = CurDAG->getTargetConstant(0, MVT::i32);
+
+  SDValue Ops[] = {
+    N->getOperand(0),
+    N->getOperand(1),
+    N->getOperand(2),
+    Zero,
+    Zero,
+    Zero,
+    Zero
+  };
+
+  return CurDAG->SelectNodeTo(N, Opc, VT, MVT::i1, Ops);
 }
 
-bool AMDGPUDAGToDAGISel::SelectI24(SDValue Op, SDValue &I24) {
+static SDValue wrapAddr64Rsrc(SelectionDAG *DAG, SDLoc DL, SDValue Ptr) {
+  return SDValue(DAG->getMachineNode(AMDGPU::SI_ADDR64_RSRC, DL, MVT::v4i32,
+                                     Ptr), 0);
+}
 
-  assert(Op.getValueType() == MVT::i32);
+static bool isLegalMUBUFImmOffset(const ConstantSDNode *Imm) {
+  return isUInt<12>(Imm->getZExtValue());
+}
 
-  if (CurDAG->ComputeNumSignBits(Op) == 9) {
-    I24 = SimplifyI24(Op);
+bool AMDGPUDAGToDAGISel::SelectMUBUFAddr64(SDValue Addr, SDValue &Ptr,
+                                           SDValue &Offset,
+                                           SDValue &ImmOffset) const {
+  SDLoc DL(Addr);
+
+  if (CurDAG->isBaseWithConstantOffset(Addr)) {
+    SDValue N0 = Addr.getOperand(0);
+    SDValue N1 = Addr.getOperand(1);
+    ConstantSDNode *C1 = cast<ConstantSDNode>(N1);
+
+    if (isLegalMUBUFImmOffset(C1)) {
+
+      if (N0.getOpcode() == ISD::ADD) {
+        // (add (add N2, N3), C1)
+        SDValue N2 = N0.getOperand(0);
+        SDValue N3 = N0.getOperand(1);
+        Ptr = wrapAddr64Rsrc(CurDAG, DL, N2);
+        Offset = N3;
+        ImmOffset = CurDAG->getTargetConstant(C1->getZExtValue(), MVT::i16);
+        return true;
+      }
+
+      // (add N0, C1)
+      Ptr = wrapAddr64Rsrc(CurDAG, DL, CurDAG->getTargetConstant(0, MVT::i64));;
+      Offset = N0;
+      ImmOffset = CurDAG->getTargetConstant(C1->getZExtValue(), MVT::i16);
+      return true;
+    }
+  }
+  if (Addr.getOpcode() == ISD::ADD) {
+    // (add N0, N1)
+    SDValue N0 = Addr.getOperand(0);
+    SDValue N1 = Addr.getOperand(1);
+    Ptr = wrapAddr64Rsrc(CurDAG, DL, N0);
+    Offset = N1;
+    ImmOffset = CurDAG->getTargetConstant(0, MVT::i16);
     return true;
   }
-  return false;
+
+  // default case
+  Ptr = wrapAddr64Rsrc(CurDAG, DL, CurDAG->getConstant(0, MVT::i64));
+  Offset = Addr;
+  ImmOffset = CurDAG->getTargetConstant(0, MVT::i16);
+  return true;
 }
 
-bool AMDGPUDAGToDAGISel::SelectU24(SDValue Op, SDValue &U24) {
-  APInt KnownZero;
-  APInt KnownOne;
-  CurDAG->ComputeMaskedBits(Op, KnownZero, KnownOne);
+/// \brief Return a resource descriptor with the 'Add TID' bit enabled
+///        The TID (Thread ID) is multipled by the stride value (bits [61:48]
+///        of the resource descriptor) to create an offset, which is added to the
+///        resource ponter.
+static SDValue buildScratchRSRC(SelectionDAG *DAG, SDLoc DL, SDValue Ptr) {
+
+  uint64_t Rsrc = AMDGPU::RSRC_DATA_FORMAT | AMDGPU::RSRC_TID_ENABLE |
+                  0xffffffff;
 
-  assert (Op.getValueType() == MVT::i32);
+  SDValue PtrLo = DAG->getTargetExtractSubreg(AMDGPU::sub0, DL, MVT::i32, Ptr);
+  SDValue PtrHi = DAG->getTargetExtractSubreg(AMDGPU::sub1, DL, MVT::i32, Ptr);
+  SDValue DataLo = DAG->getTargetConstant(
+      Rsrc & APInt::getAllOnesValue(32).getZExtValue(), MVT::i32);
+  SDValue DataHi = DAG->getTargetConstant(Rsrc >> 32, MVT::i32);
+
+  const SDValue Ops[] = { PtrLo, PtrHi, DataLo, DataHi };
+  return SDValue(DAG->getMachineNode(AMDGPU::SI_BUFFER_RSRC, DL,
+                                     MVT::v4i32, Ops), 0);
+}
 
-  // ANY_EXTEND and EXTLOAD operations can only be done on types smaller than
-  // i32.  These smaller types are legal to use with the i24 instructions.
-  if ((KnownZero & APInt(KnownZero.getBitWidth(), 0xFF000000)) == 0xFF000000 ||
-       Op.getOpcode() == ISD::ANY_EXTEND ||
-       ISD::isEXTLoad(Op.getNode())) {
-    U24 = SimplifyI24(Op);
+bool AMDGPUDAGToDAGISel::SelectMUBUFScratch(SDValue Addr, SDValue &Rsrc,
+                                            SDValue &VAddr, SDValue &SOffset,
+                                            SDValue &ImmOffset) const {
+
+  SDLoc DL(Addr);
+  MachineFunction &MF = CurDAG->getMachineFunction();
+  const SIRegisterInfo *TRI = static_cast<const SIRegisterInfo*>(MF.getTarget().getRegisterInfo());
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+
+
+  unsigned ScratchPtrReg =
+      TRI->getPreloadedValue(MF, SIRegisterInfo::SCRATCH_PTR);
+  unsigned ScratchOffsetReg =
+      TRI->getPreloadedValue(MF, SIRegisterInfo::SCRATCH_WAVE_OFFSET);
+
+  Rsrc = buildScratchRSRC(CurDAG, DL, CurDAG->getCopyFromReg(CurDAG->getEntryNode(), DL, MRI.getLiveInVirtReg(ScratchPtrReg), MVT::i64));
+  SOffset = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), DL,
+      MRI.getLiveInVirtReg(ScratchOffsetReg), MVT::i32);
+
+  // (add n0, c1)
+  if (CurDAG->isBaseWithConstantOffset(Addr)) {
+    SDValue N1 = Addr.getOperand(1);
+    ConstantSDNode *C1 = cast<ConstantSDNode>(N1);
+
+    if (isLegalMUBUFImmOffset(C1)) {
+      VAddr = Addr.getOperand(0);
+      ImmOffset = CurDAG->getTargetConstant(C1->getZExtValue(), MVT::i16);
+      return true;
+    }
+  }
+
+  // (add FI, n0)
+  if ((Addr.getOpcode() == ISD::ADD || Addr.getOpcode() == ISD::OR) &&
+       isa<FrameIndexSDNode>(Addr.getOperand(0))) {
+    VAddr = Addr.getOperand(1);
+    ImmOffset = Addr.getOperand(0);
     return true;
   }
-  return false;
+
+  // (FI)
+  if (isa<FrameIndexSDNode>(Addr)) {
+    VAddr = SDValue(CurDAG->getMachineNode(AMDGPU::V_MOV_B32_e32, DL, MVT::i32,
+                                          CurDAG->getConstant(0, MVT::i32)), 0);
+    ImmOffset = Addr;
+    return true;
+  }
+
+  // (node)
+  VAddr = Addr;
+  ImmOffset = CurDAG->getTargetConstant(0, MVT::i16);
+  return true;
+}
+
+bool AMDGPUDAGToDAGISel::SelectMUBUFAddr32(SDValue Addr, SDValue &SRsrc,
+                                           SDValue &VAddr, SDValue &SOffset,
+                                           SDValue &Offset, SDValue &Offen,
+                                           SDValue &Idxen, SDValue &GLC,
+                                           SDValue &SLC, SDValue &TFE) const {
+
+  GLC = CurDAG->getTargetConstant(0, MVT::i1);
+  SLC = CurDAG->getTargetConstant(0, MVT::i1);
+  TFE = CurDAG->getTargetConstant(0, MVT::i1);
+
+  Idxen = CurDAG->getTargetConstant(0, MVT::i1);
+  Offen = CurDAG->getTargetConstant(1, MVT::i1);
+
+  return SelectMUBUFScratch(Addr, SRsrc, VAddr, SOffset, Offset);
 }
 
 void AMDGPUDAGToDAGISel::PostprocessISelDAG() {
   const AMDGPUTargetLowering& Lowering =
-    (*(const AMDGPUTargetLowering*)getTargetLowering());
+    *static_cast<const AMDGPUTargetLowering*>(getTargetLowering());
   bool IsModified = false;
   do {
     IsModified = false;
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUISelLowering.cpp b/contrib/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
index 1029f30..5a46297b 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -16,9 +16,9 @@
 #include "AMDGPUISelLowering.h"
 #include "AMDGPU.h"
 #include "AMDGPUFrameLowering.h"
+#include "AMDGPUIntrinsicInfo.h"
 #include "AMDGPURegisterInfo.h"
 #include "AMDGPUSubtarget.h"
-#include "AMDILIntrinsicInfo.h"
 #include "R600MachineFunctionInfo.h"
 #include "SIMachineFunctionInfo.h"
 #include "llvm/CodeGen/CallingConvLower.h"
@@ -27,24 +27,93 @@
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
 #include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DiagnosticInfo.h"
+#include "llvm/IR/DiagnosticPrinter.h"
 
 using namespace llvm;
+
+namespace {
+
+/// Diagnostic information for unimplemented or unsupported feature reporting.
+class DiagnosticInfoUnsupported : public DiagnosticInfo {
+private:
+  const Twine &Description;
+  const Function &Fn;
+
+  static int KindID;
+
+  static int getKindID() {
+    if (KindID == 0)
+      KindID = llvm::getNextAvailablePluginDiagnosticKind();
+    return KindID;
+  }
+
+public:
+  DiagnosticInfoUnsupported(const Function &Fn, const Twine &Desc,
+                          DiagnosticSeverity Severity = DS_Error)
+    : DiagnosticInfo(getKindID(), Severity),
+      Description(Desc),
+      Fn(Fn) { }
+
+  const Function &getFunction() const { return Fn; }
+  const Twine &getDescription() const { return Description; }
+
+  void print(DiagnosticPrinter &DP) const override {
+    DP << "unsupported " << getDescription() << " in " << Fn.getName();
+  }
+
+  static bool classof(const DiagnosticInfo *DI) {
+    return DI->getKind() == getKindID();
+  }
+};
+
+int DiagnosticInfoUnsupported::KindID = 0;
+}
+
+
 static bool allocateStack(unsigned ValNo, MVT ValVT, MVT LocVT,
                       CCValAssign::LocInfo LocInfo,
                       ISD::ArgFlagsTy ArgFlags, CCState &State) {
-  unsigned Offset = State.AllocateStack(ValVT.getSizeInBits() / 8, ArgFlags.getOrigAlign());
-    State.addLoc(CCValAssign::getMem(ValNo, ValVT, Offset, LocVT, LocInfo));
+  unsigned Offset = State.AllocateStack(ValVT.getStoreSize(),
+                                        ArgFlags.getOrigAlign());
+  State.addLoc(CCValAssign::getMem(ValNo, ValVT, Offset, LocVT, LocInfo));
 
   return true;
 }
 
 #include "AMDGPUGenCallingConv.inc"
 
+// Find a larger type to do a load / store of a vector with.
+EVT AMDGPUTargetLowering::getEquivalentMemType(LLVMContext &Ctx, EVT VT) {
+  unsigned StoreSize = VT.getStoreSizeInBits();
+  if (StoreSize <= 32)
+    return EVT::getIntegerVT(Ctx, StoreSize);
+
+  assert(StoreSize % 32 == 0 && "Store size not a multiple of 32");
+  return EVT::getVectorVT(Ctx, MVT::i32, StoreSize / 32);
+}
+
+// Type for a vector that will be loaded to.
+EVT AMDGPUTargetLowering::getEquivalentLoadRegType(LLVMContext &Ctx, EVT VT) {
+  unsigned StoreSize = VT.getStoreSizeInBits();
+  if (StoreSize <= 32)
+    return EVT::getIntegerVT(Ctx, 32);
+
+  return EVT::getVectorVT(Ctx, MVT::i32, StoreSize / 32);
+}
+
 AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   TargetLowering(TM, new TargetLoweringObjectFileELF()) {
 
-  // Initialize target lowering borrowed from AMDIL
-  InitAMDILLowering();
+  Subtarget = &TM.getSubtarget<AMDGPUSubtarget>();
+
+  setOperationAction(ISD::Constant, MVT::i32, Legal);
+  setOperationAction(ISD::Constant, MVT::i64, Legal);
+  setOperationAction(ISD::ConstantFP, MVT::f32, Legal);
+  setOperationAction(ISD::ConstantFP, MVT::f64, Legal);
+
+  setOperationAction(ISD::BR_JT, MVT::Other, Expand);
+  setOperationAction(ISD::BRIND, MVT::Other, Expand);
 
   // We need to custom lower some of the intrinsics
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
@@ -59,9 +128,7 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::FFLOOR, MVT::f32, Legal);
   setOperationAction(ISD::FRINT,  MVT::f32, Legal);
   setOperationAction(ISD::FROUND, MVT::f32, Legal);
-
-  // The hardware supports ROTR, but not ROTL
-  setOperationAction(ISD::ROTL, MVT::i32, Expand);
+  setOperationAction(ISD::FTRUNC, MVT::f32, Legal);
 
   // Lower floating point store/load to integer store/load to reduce the number
   // of patterns in tablegen.
@@ -71,6 +138,9 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::STORE, MVT::v2f32, Promote);
   AddPromotedToType(ISD::STORE, MVT::v2f32, MVT::v2i32);
 
+  setOperationAction(ISD::STORE, MVT::i64, Promote);
+  AddPromotedToType(ISD::STORE, MVT::i64, MVT::v2i32);
+
   setOperationAction(ISD::STORE, MVT::v4f32, Promote);
   AddPromotedToType(ISD::STORE, MVT::v4f32, MVT::v4i32);
 
@@ -83,6 +153,9 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::STORE, MVT::f64, Promote);
   AddPromotedToType(ISD::STORE, MVT::f64, MVT::i64);
 
+  setOperationAction(ISD::STORE, MVT::v2f64, Promote);
+  AddPromotedToType(ISD::STORE, MVT::v2f64, MVT::v2i64);
+
   // Custom lowering of vector stores is required for local address space
   // stores.
   setOperationAction(ISD::STORE, MVT::v4i32, Custom);
@@ -93,16 +166,27 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setTruncStoreAction(MVT::v2i32, MVT::v2i16, Custom);
   setTruncStoreAction(MVT::v2i32, MVT::v2i8, Custom);
   setTruncStoreAction(MVT::v4i32, MVT::v4i8, Custom);
+
   // XXX: This can be change to Custom, once ExpandVectorStores can
   // handle 64-bit stores.
   setTruncStoreAction(MVT::v4i32, MVT::v4i16, Expand);
 
+  setTruncStoreAction(MVT::i64, MVT::i16, Expand);
+  setTruncStoreAction(MVT::i64, MVT::i8, Expand);
+  setTruncStoreAction(MVT::i64, MVT::i1, Expand);
+  setTruncStoreAction(MVT::v2i64, MVT::v2i1, Expand);
+  setTruncStoreAction(MVT::v4i64, MVT::v4i1, Expand);
+
+
   setOperationAction(ISD::LOAD, MVT::f32, Promote);
   AddPromotedToType(ISD::LOAD, MVT::f32, MVT::i32);
 
   setOperationAction(ISD::LOAD, MVT::v2f32, Promote);
   AddPromotedToType(ISD::LOAD, MVT::v2f32, MVT::v2i32);
 
+  setOperationAction(ISD::LOAD, MVT::i64, Promote);
+  AddPromotedToType(ISD::LOAD, MVT::i64, MVT::v2i32);
+
   setOperationAction(ISD::LOAD, MVT::v4f32, Promote);
   AddPromotedToType(ISD::LOAD, MVT::v4f32, MVT::v4i32);
 
@@ -115,10 +199,19 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::LOAD, MVT::f64, Promote);
   AddPromotedToType(ISD::LOAD, MVT::f64, MVT::i64);
 
+  setOperationAction(ISD::LOAD, MVT::v2f64, Promote);
+  AddPromotedToType(ISD::LOAD, MVT::v2f64, MVT::v2i64);
+
   setOperationAction(ISD::CONCAT_VECTORS, MVT::v4i32, Custom);
   setOperationAction(ISD::CONCAT_VECTORS, MVT::v4f32, Custom);
-  setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2i32, Custom);
+  setOperationAction(ISD::CONCAT_VECTORS, MVT::v8i32, Custom);
+  setOperationAction(ISD::CONCAT_VECTORS, MVT::v8f32, Custom);
   setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2f32, Custom);
+  setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2i32, Custom);
+  setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v4f32, Custom);
+  setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v4i32, Custom);
+  setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v8f32, Custom);
+  setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v8i32, Custom);
 
   setLoadExtAction(ISD::EXTLOAD, MVT::v2i8, Expand);
   setLoadExtAction(ISD::SEXTLOAD, MVT::v2i8, Expand);
@@ -135,27 +228,74 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
 
   setOperationAction(ISD::BR_CC, MVT::i1, Expand);
 
-  setOperationAction(ISD::FNEG, MVT::v2f32, Expand);
-  setOperationAction(ISD::FNEG, MVT::v4f32, Expand);
+  if (Subtarget->getGeneration() < AMDGPUSubtarget::SEA_ISLANDS) {
+    setOperationAction(ISD::FCEIL, MVT::f64, Custom);
+    setOperationAction(ISD::FTRUNC, MVT::f64, Custom);
+    setOperationAction(ISD::FRINT, MVT::f64, Custom);
+    setOperationAction(ISD::FFLOOR, MVT::f64, Custom);
+  }
+
+  if (!Subtarget->hasBFI()) {
+    // fcopysign can be done in a single instruction with BFI.
+    setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
+    setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
+  }
 
-  setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
+  setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
 
-  setOperationAction(ISD::MUL, MVT::i64, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
+  setTruncStoreAction(MVT::f32, MVT::f16, Expand);
+  setTruncStoreAction(MVT::f64, MVT::f16, Expand);
+
+  const MVT ScalarIntVTs[] = { MVT::i32, MVT::i64 };
+  for (MVT VT : ScalarIntVTs) {
+    setOperationAction(ISD::SREM, VT, Expand);
+    setOperationAction(ISD::SDIV, VT, Expand);
+
+    // GPU does not have divrem function for signed or unsigned.
+    setOperationAction(ISD::SDIVREM, VT, Custom);
+    setOperationAction(ISD::UDIVREM, VT, Custom);
+
+    // GPU does not have [S|U]MUL_LOHI functions as a single instruction.
+    setOperationAction(ISD::SMUL_LOHI, VT, Expand);
+    setOperationAction(ISD::UMUL_LOHI, VT, Expand);
+
+    setOperationAction(ISD::BSWAP, VT, Expand);
+    setOperationAction(ISD::CTTZ, VT, Expand);
+    setOperationAction(ISD::CTLZ, VT, Expand);
+  }
+
+  if (!Subtarget->hasBCNT(32))
+    setOperationAction(ISD::CTPOP, MVT::i32, Expand);
+
+  if (!Subtarget->hasBCNT(64))
+    setOperationAction(ISD::CTPOP, MVT::i64, Expand);
 
+  // The hardware supports 32-bit ROTR, but not ROTL.
+  setOperationAction(ISD::ROTL, MVT::i32, Expand);
+  setOperationAction(ISD::ROTL, MVT::i64, Expand);
+  setOperationAction(ISD::ROTR, MVT::i64, Expand);
+
+  setOperationAction(ISD::MUL, MVT::i64, Expand);
+  setOperationAction(ISD::MULHU, MVT::i64, Expand);
+  setOperationAction(ISD::MULHS, MVT::i64, Expand);
   setOperationAction(ISD::UDIV, MVT::i32, Expand);
-  setOperationAction(ISD::UDIVREM, MVT::i32, Custom);
   setOperationAction(ISD::UREM, MVT::i32, Expand);
-  setOperationAction(ISD::VSELECT, MVT::v2f32, Expand);
-  setOperationAction(ISD::VSELECT, MVT::v4f32, Expand);
+  setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
+  setOperationAction(ISD::SELECT_CC, MVT::i64, Expand);
 
-  static const MVT::SimpleValueType IntTypes[] = {
+  if (!Subtarget->hasFFBH())
+    setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i32, Expand);
+
+  if (!Subtarget->hasFFBL())
+    setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i32, Expand);
+
+  static const MVT::SimpleValueType VectorIntTypes[] = {
     MVT::v2i32, MVT::v4i32
   };
-  const size_t NumIntTypes = array_lengthof(IntTypes);
 
-  for (unsigned int x  = 0; x < NumIntTypes; ++x) {
-    MVT::SimpleValueType VT = IntTypes[x];
-    //Expand the following operations for the current type by default
+  for (MVT VT : VectorIntTypes) {
+    // Expand the following operations for the current type by default.
     setOperationAction(ISD::ADD,  VT, Expand);
     setOperationAction(ISD::AND,  VT, Expand);
     setOperationAction(ISD::FP_TO_SINT, VT, Expand);
@@ -163,33 +303,94 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
     setOperationAction(ISD::MUL,  VT, Expand);
     setOperationAction(ISD::OR,   VT, Expand);
     setOperationAction(ISD::SHL,  VT, Expand);
-    setOperationAction(ISD::SINT_TO_FP, VT, Expand);
-    setOperationAction(ISD::SRL,  VT, Expand);
     setOperationAction(ISD::SRA,  VT, Expand);
+    setOperationAction(ISD::SRL,  VT, Expand);
+    setOperationAction(ISD::ROTL, VT, Expand);
+    setOperationAction(ISD::ROTR, VT, Expand);
     setOperationAction(ISD::SUB,  VT, Expand);
-    setOperationAction(ISD::UDIV, VT, Expand);
+    setOperationAction(ISD::SINT_TO_FP, VT, Expand);
     setOperationAction(ISD::UINT_TO_FP, VT, Expand);
+    // TODO: Implement custom UREM / SREM routines.
+    setOperationAction(ISD::SDIV, VT, Expand);
+    setOperationAction(ISD::UDIV, VT, Expand);
+    setOperationAction(ISD::SREM, VT, Expand);
     setOperationAction(ISD::UREM, VT, Expand);
+    setOperationAction(ISD::SMUL_LOHI, VT, Expand);
+    setOperationAction(ISD::UMUL_LOHI, VT, Expand);
+    setOperationAction(ISD::SDIVREM, VT, Custom);
+    setOperationAction(ISD::UDIVREM, VT, Custom);
+    setOperationAction(ISD::ADDC, VT, Expand);
+    setOperationAction(ISD::SUBC, VT, Expand);
+    setOperationAction(ISD::ADDE, VT, Expand);
+    setOperationAction(ISD::SUBE, VT, Expand);
+    setOperationAction(ISD::SELECT, VT, Expand);
     setOperationAction(ISD::VSELECT, VT, Expand);
+    setOperationAction(ISD::SELECT_CC, VT, Expand);
     setOperationAction(ISD::XOR,  VT, Expand);
+    setOperationAction(ISD::BSWAP, VT, Expand);
+    setOperationAction(ISD::CTPOP, VT, Expand);
+    setOperationAction(ISD::CTTZ, VT, Expand);
+    setOperationAction(ISD::CTTZ_ZERO_UNDEF, VT, Expand);
+    setOperationAction(ISD::CTLZ, VT, Expand);
+    setOperationAction(ISD::CTLZ_ZERO_UNDEF, VT, Expand);
+    setOperationAction(ISD::VECTOR_SHUFFLE, VT, Expand);
   }
 
-  static const MVT::SimpleValueType FloatTypes[] = {
+  static const MVT::SimpleValueType FloatVectorTypes[] = {
     MVT::v2f32, MVT::v4f32
   };
-  const size_t NumFloatTypes = array_lengthof(FloatTypes);
 
-  for (unsigned int x = 0; x < NumFloatTypes; ++x) {
-    MVT::SimpleValueType VT = FloatTypes[x];
+  for (MVT VT : FloatVectorTypes) {
     setOperationAction(ISD::FABS, VT, Expand);
     setOperationAction(ISD::FADD, VT, Expand);
+    setOperationAction(ISD::FCEIL, VT, Expand);
+    setOperationAction(ISD::FCOS, VT, Expand);
     setOperationAction(ISD::FDIV, VT, Expand);
+    setOperationAction(ISD::FEXP2, VT, Expand);
+    setOperationAction(ISD::FLOG2, VT, Expand);
+    setOperationAction(ISD::FPOW, VT, Expand);
     setOperationAction(ISD::FFLOOR, VT, Expand);
+    setOperationAction(ISD::FTRUNC, VT, Expand);
     setOperationAction(ISD::FMUL, VT, Expand);
+    setOperationAction(ISD::FMA, VT, Expand);
     setOperationAction(ISD::FRINT, VT, Expand);
+    setOperationAction(ISD::FNEARBYINT, VT, Expand);
     setOperationAction(ISD::FSQRT, VT, Expand);
+    setOperationAction(ISD::FSIN, VT, Expand);
     setOperationAction(ISD::FSUB, VT, Expand);
+    setOperationAction(ISD::FNEG, VT, Expand);
+    setOperationAction(ISD::SELECT, VT, Expand);
+    setOperationAction(ISD::VSELECT, VT, Expand);
+    setOperationAction(ISD::SELECT_CC, VT, Expand);
+    setOperationAction(ISD::FCOPYSIGN, VT, Expand);
+    setOperationAction(ISD::VECTOR_SHUFFLE, VT, Expand);
   }
+
+  setOperationAction(ISD::FNEARBYINT, MVT::f32, Custom);
+  setOperationAction(ISD::FNEARBYINT, MVT::f64, Custom);
+
+  setTargetDAGCombine(ISD::MUL);
+  setTargetDAGCombine(ISD::SELECT_CC);
+  setTargetDAGCombine(ISD::STORE);
+
+  setSchedulingPreference(Sched::RegPressure);
+  setJumpIsExpensive(true);
+
+  setSelectIsExpensive(false);
+  PredictableSelectIsExpensive = false;
+
+  // There are no integer divide instructions, and these expand to a pretty
+  // large sequence of instructions.
+  setIntDivIsCheap(false);
+  setPow2DivIsCheap(false);
+
+  // TODO: Investigate this when 64-bit divides are implemented.
+  addBypassSlowDiv(64, 32);
+
+  // FIXME: Need to really handle these.
+  MaxStoresPerMemcpy  = 4096;
+  MaxStoresPerMemmove = 4096;
+  MaxStoresPerMemset  = 4096;
 }
 
 //===----------------------------------------------------------------------===//
@@ -200,6 +401,23 @@ MVT AMDGPUTargetLowering::getVectorIdxTy() const {
   return MVT::i32;
 }
 
+bool AMDGPUTargetLowering::isSelectSupported(SelectSupportKind SelType) const {
+  return true;
+}
+
+// The backend supports 32 and 64 bit floating point immediates.
+// FIXME: Why are we reporting vectors of FP immediates as legal?
+bool AMDGPUTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
+  EVT ScalarVT = VT.getScalarType();
+  return (ScalarVT == MVT::f32 || ScalarVT == MVT::f64);
+}
+
+// We don't want to shrink f64 / f32 constants.
+bool AMDGPUTargetLowering::ShouldShrinkFPConstant(EVT VT) const {
+  EVT ScalarVT = VT.getScalarType();
+  return (ScalarVT != MVT::f32 && ScalarVT != MVT::f64);
+}
+
 bool AMDGPUTargetLowering::isLoadBitCastBeneficial(EVT LoadTy,
                                                    EVT CastTy) const {
   if (LoadTy.getSizeInBits() != CastTy.getSizeInBits())
@@ -227,6 +445,47 @@ bool AMDGPUTargetLowering::isFNegFree(EVT VT) const {
   return VT == MVT::f32;
 }
 
+bool AMDGPUTargetLowering::isTruncateFree(EVT Source, EVT Dest) const {
+  // Truncate is just accessing a subregister.
+  return Dest.bitsLT(Source) && (Dest.getSizeInBits() % 32 == 0);
+}
+
+bool AMDGPUTargetLowering::isTruncateFree(Type *Source, Type *Dest) const {
+  // Truncate is just accessing a subregister.
+  return Dest->getPrimitiveSizeInBits() < Source->getPrimitiveSizeInBits() &&
+         (Dest->getPrimitiveSizeInBits() % 32 == 0);
+}
+
+bool AMDGPUTargetLowering::isZExtFree(Type *Src, Type *Dest) const {
+  const DataLayout *DL = getDataLayout();
+  unsigned SrcSize = DL->getTypeSizeInBits(Src->getScalarType());
+  unsigned DestSize = DL->getTypeSizeInBits(Dest->getScalarType());
+
+  return SrcSize == 32 && DestSize == 64;
+}
+
+bool AMDGPUTargetLowering::isZExtFree(EVT Src, EVT Dest) const {
+  // Any register load of a 64-bit value really requires 2 32-bit moves. For all
+  // practical purposes, the extra mov 0 to load a 64-bit is free.  As used,
+  // this will enable reducing 64-bit operations the 32-bit, which is always
+  // good.
+  return Src == MVT::i32 && Dest == MVT::i64;
+}
+
+bool AMDGPUTargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
+  return isZExtFree(Val.getValueType(), VT2);
+}
+
+bool AMDGPUTargetLowering::isNarrowingProfitable(EVT SrcVT, EVT DestVT) const {
+  // There aren't really 64-bit registers, but pairs of 32-bit ones and only a
+  // limited number of native 64-bit operations. Shrinking an operation to fit
+  // in a single 32-bit register should always be helpful. As currently used,
+  // this is much less general than the name suggests, and is only used in
+  // places trying to reduce the sizes of loads. Shrinking loads to < 32-bits is
+  // not profitable, and may actually be harmful.
+  return SrcVT.getSizeInBits() > 32 && DestVT.getSizeInBits() == 32;
+}
+
 //===---------------------------------------------------------------------===//
 // TargetLowering Callbacks
 //===---------------------------------------------------------------------===//
@@ -251,67 +510,243 @@ SDValue AMDGPUTargetLowering::LowerReturn(
 // Target specific lowering
 //===---------------------------------------------------------------------===//
 
-SDValue AMDGPUTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG)
-    const {
+SDValue AMDGPUTargetLowering::LowerCall(CallLoweringInfo &CLI,
+                                        SmallVectorImpl<SDValue> &InVals) const {
+  SDValue Callee = CLI.Callee;
+  SelectionDAG &DAG = CLI.DAG;
+
+  const Function &Fn = *DAG.getMachineFunction().getFunction();
+
+  StringRef FuncName("<unknown>");
+
+  if (const ExternalSymbolSDNode *G = dyn_cast<ExternalSymbolSDNode>(Callee))
+    FuncName = G->getSymbol();
+  else if (const GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
+    FuncName = G->getGlobal()->getName();
+
+  DiagnosticInfoUnsupported NoCalls(Fn, "call to function " + FuncName);
+  DAG.getContext()->diagnose(NoCalls);
+  return SDValue();
+}
+
+SDValue AMDGPUTargetLowering::LowerOperation(SDValue Op,
+                                             SelectionDAG &DAG) const {
   switch (Op.getOpcode()) {
   default:
     Op.getNode()->dump();
-    assert(0 && "Custom lowering code for this"
-        "instruction is not implemented yet!");
+    llvm_unreachable("Custom lowering code for this"
+                     "instruction is not implemented yet!");
     break;
-  // AMDIL DAG lowering
-  case ISD::SDIV: return LowerSDIV(Op, DAG);
-  case ISD::SREM: return LowerSREM(Op, DAG);
   case ISD::SIGN_EXTEND_INREG: return LowerSIGN_EXTEND_INREG(Op, DAG);
-  case ISD::BRCOND: return LowerBRCOND(Op, DAG);
-  // AMDGPU DAG lowering
   case ISD::CONCAT_VECTORS: return LowerCONCAT_VECTORS(Op, DAG);
   case ISD::EXTRACT_SUBVECTOR: return LowerEXTRACT_SUBVECTOR(Op, DAG);
   case ISD::FrameIndex: return LowerFrameIndex(Op, DAG);
   case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
+  case ISD::SDIV: return LowerSDIV(Op, DAG);
+  case ISD::SREM: return LowerSREM(Op, DAG);
   case ISD::UDIVREM: return LowerUDIVREM(Op, DAG);
+  case ISD::SDIVREM: return LowerSDIVREM(Op, DAG);
+  case ISD::FCEIL: return LowerFCEIL(Op, DAG);
+  case ISD::FTRUNC: return LowerFTRUNC(Op, DAG);
+  case ISD::FRINT: return LowerFRINT(Op, DAG);
+  case ISD::FNEARBYINT: return LowerFNEARBYINT(Op, DAG);
+  case ISD::FFLOOR: return LowerFFLOOR(Op, DAG);
   case ISD::UINT_TO_FP: return LowerUINT_TO_FP(Op, DAG);
   }
   return Op;
 }
 
+void AMDGPUTargetLowering::ReplaceNodeResults(SDNode *N,
+                                              SmallVectorImpl<SDValue> &Results,
+                                              SelectionDAG &DAG) const {
+  switch (N->getOpcode()) {
+  case ISD::SIGN_EXTEND_INREG:
+    // Different parts of legalization seem to interpret which type of
+    // sign_extend_inreg is the one to check for custom lowering. The extended
+    // from type is what really matters, but some places check for custom
+    // lowering of the result type. This results in trying to use
+    // ReplaceNodeResults to sext_in_reg to an illegal type, so we'll just do
+    // nothing here and let the illegal result integer be handled normally.
+    return;
+  case ISD::LOAD: {
+    SDNode *Node = LowerLOAD(SDValue(N, 0), DAG).getNode();
+    if (!Node)
+      return;
+
+    Results.push_back(SDValue(Node, 0));
+    Results.push_back(SDValue(Node, 1));
+    // XXX: LLVM seems not to replace Chain Value inside CustomWidenLowerNode
+    // function
+    DAG.ReplaceAllUsesOfValueWith(SDValue(N,1), SDValue(Node, 1));
+    return;
+  }
+  case ISD::STORE: {
+    SDValue Lowered = LowerSTORE(SDValue(N, 0), DAG);
+    if (Lowered.getNode())
+      Results.push_back(Lowered);
+    return;
+  }
+  default:
+    return;
+  }
+}
+
+// FIXME: This implements accesses to initialized globals in the constant
+// address space by copying them to private and accessing that. It does not
+// properly handle illegal types or vectors. The private vector loads are not
+// scalarized, and the illegal scalars hit an assertion. This technique will not
+// work well with large initializers, and this should eventually be
+// removed. Initialized globals should be placed into a data section that the
+// runtime will load into a buffer before the kernel is executed. Uses of the
+// global need to be replaced with a pointer loaded from an implicit kernel
+// argument into this buffer holding the copy of the data, which will remove the
+// need for any of this.
+SDValue AMDGPUTargetLowering::LowerConstantInitializer(const Constant* Init,
+                                                       const GlobalValue *GV,
+                                                       const SDValue &InitPtr,
+                                                       SDValue Chain,
+                                                       SelectionDAG &DAG) const {
+  const DataLayout *TD = getTargetMachine().getDataLayout();
+  SDLoc DL(InitPtr);
+  Type *InitTy = Init->getType();
+
+  if (const ConstantInt *CI = dyn_cast<ConstantInt>(Init)) {
+    EVT VT = EVT::getEVT(InitTy);
+    PointerType *PtrTy = PointerType::get(InitTy, AMDGPUAS::PRIVATE_ADDRESS);
+    return DAG.getStore(Chain, DL, DAG.getConstant(*CI, VT), InitPtr,
+                        MachinePointerInfo(UndefValue::get(PtrTy)), false, false,
+                        TD->getPrefTypeAlignment(InitTy));
+  }
+
+  if (const ConstantFP *CFP = dyn_cast<ConstantFP>(Init)) {
+    EVT VT = EVT::getEVT(CFP->getType());
+    PointerType *PtrTy = PointerType::get(CFP->getType(), 0);
+    return DAG.getStore(Chain, DL, DAG.getConstantFP(*CFP, VT), InitPtr,
+                 MachinePointerInfo(UndefValue::get(PtrTy)), false, false,
+                 TD->getPrefTypeAlignment(CFP->getType()));
+  }
+
+  if (StructType *ST = dyn_cast<StructType>(InitTy)) {
+    const StructLayout *SL = TD->getStructLayout(ST);
+
+    EVT PtrVT = InitPtr.getValueType();
+    SmallVector<SDValue, 8> Chains;
+
+    for (unsigned I = 0, N = ST->getNumElements(); I != N; ++I) {
+      SDValue Offset = DAG.getConstant(SL->getElementOffset(I), PtrVT);
+      SDValue Ptr = DAG.getNode(ISD::ADD, DL, PtrVT, InitPtr, Offset);
+
+      Constant *Elt = Init->getAggregateElement(I);
+      Chains.push_back(LowerConstantInitializer(Elt, GV, Ptr, Chain, DAG));
+    }
+
+    return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Chains);
+  }
+
+  if (SequentialType *SeqTy = dyn_cast<SequentialType>(InitTy)) {
+    EVT PtrVT = InitPtr.getValueType();
+
+    unsigned NumElements;
+    if (ArrayType *AT = dyn_cast<ArrayType>(SeqTy))
+      NumElements = AT->getNumElements();
+    else if (VectorType *VT = dyn_cast<VectorType>(SeqTy))
+      NumElements = VT->getNumElements();
+    else
+      llvm_unreachable("Unexpected type");
+
+    unsigned EltSize = TD->getTypeAllocSize(SeqTy->getElementType());
+    SmallVector<SDValue, 8> Chains;
+    for (unsigned i = 0; i < NumElements; ++i) {
+      SDValue Offset = DAG.getConstant(i * EltSize, PtrVT);
+      SDValue Ptr = DAG.getNode(ISD::ADD, DL, PtrVT, InitPtr, Offset);
+
+      Constant *Elt = Init->getAggregateElement(i);
+      Chains.push_back(LowerConstantInitializer(Elt, GV, Ptr, Chain, DAG));
+    }
+
+    return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Chains);
+  }
+
+  if (isa<UndefValue>(Init)) {
+    EVT VT = EVT::getEVT(InitTy);
+    PointerType *PtrTy = PointerType::get(InitTy, AMDGPUAS::PRIVATE_ADDRESS);
+    return DAG.getStore(Chain, DL, DAG.getUNDEF(VT), InitPtr,
+                        MachinePointerInfo(UndefValue::get(PtrTy)), false, false,
+                        TD->getPrefTypeAlignment(InitTy));
+  }
+
+  Init->dump();
+  llvm_unreachable("Unhandled constant initializer");
+}
+
 SDValue AMDGPUTargetLowering::LowerGlobalAddress(AMDGPUMachineFunction* MFI,
                                                  SDValue Op,
                                                  SelectionDAG &DAG) const {
 
   const DataLayout *TD = getTargetMachine().getDataLayout();
   GlobalAddressSDNode *G = cast<GlobalAddressSDNode>(Op);
+  const GlobalValue *GV = G->getGlobal();
 
-  assert(G->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS);
-  // XXX: What does the value of G->getOffset() mean?
-  assert(G->getOffset() == 0 &&
+  switch (G->getAddressSpace()) {
+  default: llvm_unreachable("Global Address lowering not implemented for this "
+                            "address space");
+  case AMDGPUAS::LOCAL_ADDRESS: {
+    // XXX: What does the value of G->getOffset() mean?
+    assert(G->getOffset() == 0 &&
          "Do not know what to do with an non-zero offset");
 
-  const GlobalValue *GV = G->getGlobal();
+    unsigned Offset;
+    if (MFI->LocalMemoryObjects.count(GV) == 0) {
+      uint64_t Size = TD->getTypeAllocSize(GV->getType()->getElementType());
+      Offset = MFI->LDSSize;
+      MFI->LocalMemoryObjects[GV] = Offset;
+      // XXX: Account for alignment?
+      MFI->LDSSize += Size;
+    } else {
+      Offset = MFI->LocalMemoryObjects[GV];
+    }
 
-  unsigned Offset;
-  if (MFI->LocalMemoryObjects.count(GV) == 0) {
-    uint64_t Size = TD->getTypeAllocSize(GV->getType()->getElementType());
-    Offset = MFI->LDSSize;
-    MFI->LocalMemoryObjects[GV] = Offset;
-    // XXX: Account for alignment?
-    MFI->LDSSize += Size;
-  } else {
-    Offset = MFI->LocalMemoryObjects[GV];
+    return DAG.getConstant(Offset, getPointerTy(G->getAddressSpace()));
   }
+  case AMDGPUAS::CONSTANT_ADDRESS: {
+    MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
+    Type *EltType = GV->getType()->getElementType();
+    unsigned Size = TD->getTypeAllocSize(EltType);
+    unsigned Alignment = TD->getPrefTypeAlignment(EltType);
+
+    MVT PrivPtrVT = getPointerTy(AMDGPUAS::PRIVATE_ADDRESS);
+    MVT ConstPtrVT = getPointerTy(AMDGPUAS::CONSTANT_ADDRESS);
+
+    int FI = FrameInfo->CreateStackObject(Size, Alignment, false);
+    SDValue InitPtr = DAG.getFrameIndex(FI, PrivPtrVT);
+
+    const GlobalVariable *Var = cast<GlobalVariable>(GV);
+    if (!Var->hasInitializer()) {
+      // This has no use, but bugpoint will hit it.
+      return DAG.getZExtOrTrunc(InitPtr, SDLoc(Op), ConstPtrVT);
+    }
 
-  return DAG.getConstant(Offset, getPointerTy(G->getAddressSpace()));
-}
+    const Constant *Init = Var->getInitializer();
+    SmallVector<SDNode*, 8> WorkList;
 
-void AMDGPUTargetLowering::ExtractVectorElements(SDValue Op, SelectionDAG &DAG,
-                                         SmallVectorImpl<SDValue> &Args,
-                                         unsigned Start,
-                                         unsigned Count) const {
-  EVT VT = Op.getValueType();
-  for (unsigned i = Start, e = Start + Count; i != e; ++i) {
-    Args.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(Op),
-                               VT.getVectorElementType(),
-                               Op, DAG.getConstant(i, MVT::i32)));
+    for (SDNode::use_iterator I = DAG.getEntryNode()->use_begin(),
+                              E = DAG.getEntryNode()->use_end(); I != E; ++I) {
+      if (I->getOpcode() != AMDGPUISD::REGISTER_LOAD && I->getOpcode() != ISD::LOAD)
+        continue;
+      WorkList.push_back(*I);
+    }
+    SDValue Chain = LowerConstantInitializer(Init, GV, InitPtr, DAG.getEntryNode(), DAG);
+    for (SmallVector<SDNode*, 8>::iterator I = WorkList.begin(),
+                                           E = WorkList.end(); I != E; ++I) {
+      SmallVector<SDValue, 8> Ops;
+      Ops.push_back(Chain);
+      for (unsigned i = 1; i < (*I)->getNumOperands(); ++i) {
+        Ops.push_back((*I)->getOperand(i));
+      }
+      DAG.UpdateNodeOperands(*I, Ops);
+    }
+    return DAG.getZExtOrTrunc(InitPtr, SDLoc(Op), ConstPtrVT);
+  }
   }
 }
 
@@ -321,26 +756,22 @@ SDValue AMDGPUTargetLowering::LowerCONCAT_VECTORS(SDValue Op,
   SDValue A = Op.getOperand(0);
   SDValue B = Op.getOperand(1);
 
-  ExtractVectorElements(A, DAG, Args, 0,
-                        A.getValueType().getVectorNumElements());
-  ExtractVectorElements(B, DAG, Args, 0,
-                        B.getValueType().getVectorNumElements());
+  DAG.ExtractVectorElements(A, Args);
+  DAG.ExtractVectorElements(B, Args);
 
-  return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(Op), Op.getValueType(),
-                     &Args[0], Args.size());
+  return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(Op), Op.getValueType(), Args);
 }
 
 SDValue AMDGPUTargetLowering::LowerEXTRACT_SUBVECTOR(SDValue Op,
                                                      SelectionDAG &DAG) const {
 
   SmallVector<SDValue, 8> Args;
-  EVT VT = Op.getValueType();
   unsigned Start = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
-  ExtractVectorElements(Op.getOperand(0), DAG, Args, Start,
-                        VT.getVectorNumElements());
+  EVT VT = Op.getValueType();
+  DAG.ExtractVectorElements(Op.getOperand(0), Args, Start,
+                            VT.getVectorNumElements());
 
-  return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(Op), Op.getValueType(),
-                     &Args[0], Args.size());
+  return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(Op), Op.getValueType(), Args);
 }
 
 SDValue AMDGPUTargetLowering::LowerFrameIndex(SDValue Op,
@@ -350,8 +781,7 @@ SDValue AMDGPUTargetLowering::LowerFrameIndex(SDValue Op,
   const AMDGPUFrameLowering *TFL =
    static_cast<const AMDGPUFrameLowering*>(getTargetMachine().getFrameLowering());
 
-  FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Op);
-  assert(FIN);
+  FrameIndexSDNode *FIN = cast<FrameIndexSDNode>(Op);
 
   unsigned FrameIndex = FIN->getIndex();
   unsigned Offset = TFL->getFrameIndexOffset(MF, FrameIndex);
@@ -367,41 +797,140 @@ SDValue AMDGPUTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
 
   switch (IntrinsicID) {
     default: return Op;
-    case AMDGPUIntrinsic::AMDIL_abs:
+    case AMDGPUIntrinsic::AMDGPU_abs:
+    case AMDGPUIntrinsic::AMDIL_abs: // Legacy name.
       return LowerIntrinsicIABS(Op, DAG);
-    case AMDGPUIntrinsic::AMDIL_exp:
-      return DAG.getNode(ISD::FEXP2, DL, VT, Op.getOperand(1));
     case AMDGPUIntrinsic::AMDGPU_lrp:
       return LowerIntrinsicLRP(Op, DAG);
-    case AMDGPUIntrinsic::AMDIL_fraction:
+    case AMDGPUIntrinsic::AMDGPU_fract:
+    case AMDGPUIntrinsic::AMDIL_fraction: // Legacy name.
       return DAG.getNode(AMDGPUISD::FRACT, DL, VT, Op.getOperand(1));
-    case AMDGPUIntrinsic::AMDIL_max:
-      return DAG.getNode(AMDGPUISD::FMAX, DL, VT, Op.getOperand(1),
-                                                  Op.getOperand(2));
+
+    case AMDGPUIntrinsic::AMDGPU_clamp:
+    case AMDGPUIntrinsic::AMDIL_clamp: // Legacy name.
+      return DAG.getNode(AMDGPUISD::CLAMP, DL, VT,
+                         Op.getOperand(1), Op.getOperand(2), Op.getOperand(3));
+
+    case Intrinsic::AMDGPU_div_scale: {
+      // 3rd parameter required to be a constant.
+      const ConstantSDNode *Param = dyn_cast<ConstantSDNode>(Op.getOperand(3));
+      if (!Param)
+        return DAG.getUNDEF(VT);
+
+      // Translate to the operands expected by the machine instruction. The
+      // first parameter must be the same as the first instruction.
+      SDValue Numerator = Op.getOperand(1);
+      SDValue Denominator = Op.getOperand(2);
+      SDValue Src0 = Param->isAllOnesValue() ? Numerator : Denominator;
+
+      return DAG.getNode(AMDGPUISD::DIV_SCALE, DL, VT,
+                         Src0, Denominator, Numerator);
+    }
+
+    case Intrinsic::AMDGPU_div_fmas:
+      return DAG.getNode(AMDGPUISD::DIV_FMAS, DL, VT,
+                         Op.getOperand(1), Op.getOperand(2), Op.getOperand(3));
+
+    case Intrinsic::AMDGPU_div_fixup:
+      return DAG.getNode(AMDGPUISD::DIV_FIXUP, DL, VT,
+                         Op.getOperand(1), Op.getOperand(2), Op.getOperand(3));
+
+    case Intrinsic::AMDGPU_trig_preop:
+      return DAG.getNode(AMDGPUISD::TRIG_PREOP, DL, VT,
+                         Op.getOperand(1), Op.getOperand(2));
+
+    case Intrinsic::AMDGPU_rcp:
+      return DAG.getNode(AMDGPUISD::RCP, DL, VT, Op.getOperand(1));
+
+    case Intrinsic::AMDGPU_rsq:
+      return DAG.getNode(AMDGPUISD::RSQ, DL, VT, Op.getOperand(1));
+
+    case AMDGPUIntrinsic::AMDGPU_legacy_rsq:
+      return DAG.getNode(AMDGPUISD::RSQ_LEGACY, DL, VT, Op.getOperand(1));
+
+    case Intrinsic::AMDGPU_rsq_clamped:
+      return DAG.getNode(AMDGPUISD::RSQ_CLAMPED, DL, VT, Op.getOperand(1));
+
     case AMDGPUIntrinsic::AMDGPU_imax:
       return DAG.getNode(AMDGPUISD::SMAX, DL, VT, Op.getOperand(1),
                                                   Op.getOperand(2));
     case AMDGPUIntrinsic::AMDGPU_umax:
       return DAG.getNode(AMDGPUISD::UMAX, DL, VT, Op.getOperand(1),
                                                   Op.getOperand(2));
-    case AMDGPUIntrinsic::AMDIL_min:
-      return DAG.getNode(AMDGPUISD::FMIN, DL, VT, Op.getOperand(1),
-                                                  Op.getOperand(2));
     case AMDGPUIntrinsic::AMDGPU_imin:
       return DAG.getNode(AMDGPUISD::SMIN, DL, VT, Op.getOperand(1),
                                                   Op.getOperand(2));
     case AMDGPUIntrinsic::AMDGPU_umin:
       return DAG.getNode(AMDGPUISD::UMIN, DL, VT, Op.getOperand(1),
                                                   Op.getOperand(2));
-    case AMDGPUIntrinsic::AMDIL_round_nearest:
+
+    case AMDGPUIntrinsic::AMDGPU_umul24:
+      return DAG.getNode(AMDGPUISD::MUL_U24, DL, VT,
+                         Op.getOperand(1), Op.getOperand(2));
+
+    case AMDGPUIntrinsic::AMDGPU_imul24:
+      return DAG.getNode(AMDGPUISD::MUL_I24, DL, VT,
+                         Op.getOperand(1), Op.getOperand(2));
+
+    case AMDGPUIntrinsic::AMDGPU_umad24:
+      return DAG.getNode(AMDGPUISD::MAD_U24, DL, VT,
+                         Op.getOperand(1), Op.getOperand(2), Op.getOperand(3));
+
+    case AMDGPUIntrinsic::AMDGPU_imad24:
+      return DAG.getNode(AMDGPUISD::MAD_I24, DL, VT,
+                         Op.getOperand(1), Op.getOperand(2), Op.getOperand(3));
+
+    case AMDGPUIntrinsic::AMDGPU_cvt_f32_ubyte0:
+      return DAG.getNode(AMDGPUISD::CVT_F32_UBYTE0, DL, VT, Op.getOperand(1));
+
+    case AMDGPUIntrinsic::AMDGPU_cvt_f32_ubyte1:
+      return DAG.getNode(AMDGPUISD::CVT_F32_UBYTE1, DL, VT, Op.getOperand(1));
+
+    case AMDGPUIntrinsic::AMDGPU_cvt_f32_ubyte2:
+      return DAG.getNode(AMDGPUISD::CVT_F32_UBYTE2, DL, VT, Op.getOperand(1));
+
+    case AMDGPUIntrinsic::AMDGPU_cvt_f32_ubyte3:
+      return DAG.getNode(AMDGPUISD::CVT_F32_UBYTE3, DL, VT, Op.getOperand(1));
+
+    case AMDGPUIntrinsic::AMDGPU_bfe_i32:
+      return DAG.getNode(AMDGPUISD::BFE_I32, DL, VT,
+                         Op.getOperand(1),
+                         Op.getOperand(2),
+                         Op.getOperand(3));
+
+    case AMDGPUIntrinsic::AMDGPU_bfe_u32:
+      return DAG.getNode(AMDGPUISD::BFE_U32, DL, VT,
+                         Op.getOperand(1),
+                         Op.getOperand(2),
+                         Op.getOperand(3));
+
+    case AMDGPUIntrinsic::AMDGPU_bfi:
+      return DAG.getNode(AMDGPUISD::BFI, DL, VT,
+                         Op.getOperand(1),
+                         Op.getOperand(2),
+                         Op.getOperand(3));
+
+    case AMDGPUIntrinsic::AMDGPU_bfm:
+      return DAG.getNode(AMDGPUISD::BFM, DL, VT,
+                         Op.getOperand(1),
+                         Op.getOperand(2));
+
+    case AMDGPUIntrinsic::AMDGPU_brev:
+      return DAG.getNode(AMDGPUISD::BREV, DL, VT, Op.getOperand(1));
+
+    case AMDGPUIntrinsic::AMDIL_exp: // Legacy name.
+      return DAG.getNode(ISD::FEXP2, DL, VT, Op.getOperand(1));
+
+    case AMDGPUIntrinsic::AMDIL_round_nearest: // Legacy name.
       return DAG.getNode(ISD::FRINT, DL, VT, Op.getOperand(1));
+    case AMDGPUIntrinsic::AMDGPU_trunc: // Legacy name.
+      return DAG.getNode(ISD::FTRUNC, DL, VT, Op.getOperand(1));
   }
 }
 
 ///IABS(a) = SMAX(sub(0, a), a)
 SDValue AMDGPUTargetLowering::LowerIntrinsicIABS(SDValue Op,
-    SelectionDAG &DAG) const {
-
+                                                 SelectionDAG &DAG) const {
   SDLoc DL(Op);
   EVT VT = Op.getValueType();
   SDValue Neg = DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, VT),
@@ -413,7 +942,7 @@ SDValue AMDGPUTargetLowering::LowerIntrinsicIABS(SDValue Op,
 /// Linear Interpolation
 /// LRP(a, b, c) = muladd(a,  b, (1 - a) * c)
 SDValue AMDGPUTargetLowering::LowerIntrinsicLRP(SDValue Op,
-    SelectionDAG &DAG) const {
+                                                SelectionDAG &DAG) const {
   SDLoc DL(Op);
   EVT VT = Op.getValueType();
   SDValue OneSubA = DAG.getNode(ISD::FSUB, DL, VT,
@@ -427,16 +956,16 @@ SDValue AMDGPUTargetLowering::LowerIntrinsicLRP(SDValue Op,
 }
 
 /// \brief Generate Min/Max node
-SDValue AMDGPUTargetLowering::LowerMinMax(SDValue Op,
-    SelectionDAG &DAG) const {
-  SDLoc DL(Op);
-  EVT VT = Op.getValueType();
+SDValue AMDGPUTargetLowering::CombineMinMax(SDNode *N,
+                                            SelectionDAG &DAG) const {
+  SDLoc DL(N);
+  EVT VT = N->getValueType(0);
 
-  SDValue LHS = Op.getOperand(0);
-  SDValue RHS = Op.getOperand(1);
-  SDValue True = Op.getOperand(2);
-  SDValue False = Op.getOperand(3);
-  SDValue CC = Op.getOperand(4);
+  SDValue LHS = N->getOperand(0);
+  SDValue RHS = N->getOperand(1);
+  SDValue True = N->getOperand(2);
+  SDValue False = N->getOperand(3);
+  SDValue CC = N->getOperand(4);
 
   if (VT != MVT::f32 ||
       !((LHS == True && RHS == False) || (LHS == False && RHS == True))) {
@@ -457,17 +986,15 @@ SDValue AMDGPUTargetLowering::LowerMinMax(SDValue Op,
   case ISD::SETTRUE2:
   case ISD::SETUO:
   case ISD::SETO:
-    assert(0 && "Operation should already be optimised !");
+    llvm_unreachable("Operation should already be optimised!");
   case ISD::SETULE:
   case ISD::SETULT:
   case ISD::SETOLE:
   case ISD::SETOLT:
   case ISD::SETLE:
   case ISD::SETLT: {
-    if (LHS == True)
-      return DAG.getNode(AMDGPUISD::FMIN, DL, VT, LHS, RHS);
-    else
-      return DAG.getNode(AMDGPUISD::FMAX, DL, VT, LHS, RHS);
+    unsigned Opc = (LHS == True) ? AMDGPUISD::FMIN : AMDGPUISD::FMAX;
+    return DAG.getNode(Opc, DL, VT, LHS, RHS);
   }
   case ISD::SETGT:
   case ISD::SETGE:
@@ -475,89 +1002,105 @@ SDValue AMDGPUTargetLowering::LowerMinMax(SDValue Op,
   case ISD::SETOGE:
   case ISD::SETUGT:
   case ISD::SETOGT: {
-    if (LHS == True)
-      return DAG.getNode(AMDGPUISD::FMAX, DL, VT, LHS, RHS);
-    else
-      return DAG.getNode(AMDGPUISD::FMIN, DL, VT, LHS, RHS);
+    unsigned Opc = (LHS == True) ? AMDGPUISD::FMAX : AMDGPUISD::FMIN;
+    return DAG.getNode(Opc, DL, VT, LHS, RHS);
   }
   case ISD::SETCC_INVALID:
-    assert(0 && "Invalid setcc condcode !");
+    llvm_unreachable("Invalid setcc condcode!");
   }
-  return Op;
+  return SDValue();
 }
 
 SDValue AMDGPUTargetLowering::SplitVectorLoad(const SDValue &Op,
                                               SelectionDAG &DAG) const {
   LoadSDNode *Load = dyn_cast<LoadSDNode>(Op);
   EVT MemEltVT = Load->getMemoryVT().getVectorElementType();
+  EVT LoadVT = Op.getValueType();
   EVT EltVT = Op.getValueType().getVectorElementType();
   EVT PtrVT = Load->getBasePtr().getValueType();
+
   unsigned NumElts = Load->getMemoryVT().getVectorNumElements();
   SmallVector<SDValue, 8> Loads;
+  SmallVector<SDValue, 8> Chains;
+
   SDLoc SL(Op);
 
   for (unsigned i = 0, e = NumElts; i != e; ++i) {
     SDValue Ptr = DAG.getNode(ISD::ADD, SL, PtrVT, Load->getBasePtr(),
                     DAG.getConstant(i * (MemEltVT.getSizeInBits() / 8), PtrVT));
-    Loads.push_back(DAG.getExtLoad(Load->getExtensionType(), SL, EltVT,
-                        Load->getChain(), Ptr,
-                        MachinePointerInfo(Load->getMemOperand()->getValue()),
-                        MemEltVT, Load->isVolatile(), Load->isNonTemporal(),
-                        Load->getAlignment()));
+
+    SDValue NewLoad
+      = DAG.getExtLoad(Load->getExtensionType(), SL, EltVT,
+                       Load->getChain(), Ptr,
+                       MachinePointerInfo(Load->getMemOperand()->getValue()),
+                       MemEltVT, Load->isVolatile(), Load->isNonTemporal(),
+                       Load->getAlignment());
+    Loads.push_back(NewLoad.getValue(0));
+    Chains.push_back(NewLoad.getValue(1));
   }
-  return DAG.getNode(ISD::BUILD_VECTOR, SL, Op.getValueType(), &Loads[0],
-                     Loads.size());
+
+  SDValue Ops[] = {
+    DAG.getNode(ISD::BUILD_VECTOR, SL, LoadVT, Loads),
+    DAG.getNode(ISD::TokenFactor, SL, MVT::Other, Chains)
+  };
+
+  return DAG.getMergeValues(Ops, SL);
 }
 
 SDValue AMDGPUTargetLowering::MergeVectorStore(const SDValue &Op,
                                                SelectionDAG &DAG) const {
-  StoreSDNode *Store = dyn_cast<StoreSDNode>(Op);
+  StoreSDNode *Store = cast<StoreSDNode>(Op);
   EVT MemVT = Store->getMemoryVT();
   unsigned MemBits = MemVT.getSizeInBits();
 
-  // Byte stores are really expensive, so if possible, try to pack
-  // 32-bit vector truncatating store into an i32 store.
-  // XXX: We could also handle optimize other vector bitwidths
+  // Byte stores are really expensive, so if possible, try to pack 32-bit vector
+  // truncating store into an i32 store.
+  // XXX: We could also handle optimize other vector bitwidths.
   if (!MemVT.isVector() || MemBits > 32) {
     return SDValue();
   }
 
   SDLoc DL(Op);
-  const SDValue &Value = Store->getValue();
+  SDValue Value = Store->getValue();
   EVT VT = Value.getValueType();
-  const SDValue &Ptr = Store->getBasePtr();
+  EVT ElemVT = VT.getVectorElementType();
+  SDValue Ptr = Store->getBasePtr();
   EVT MemEltVT = MemVT.getVectorElementType();
   unsigned MemEltBits = MemEltVT.getSizeInBits();
   unsigned MemNumElements = MemVT.getVectorNumElements();
-  EVT PackedVT = EVT::getIntegerVT(*DAG.getContext(), MemVT.getSizeInBits());
-  SDValue Mask;
-  switch(MemEltBits) {
-  case 8:
-    Mask = DAG.getConstant(0xFF, PackedVT);
-    break;
-  case 16:
-    Mask = DAG.getConstant(0xFFFF, PackedVT);
-    break;
-  default:
-    llvm_unreachable("Cannot lower this vector store");
-  }
+  unsigned PackedSize = MemVT.getStoreSizeInBits();
+  SDValue Mask = DAG.getConstant((1 << MemEltBits) - 1, MVT::i32);
+
+  assert(Value.getValueType().getScalarSizeInBits() >= 32);
+
   SDValue PackedValue;
   for (unsigned i = 0; i < MemNumElements; ++i) {
-    EVT ElemVT = VT.getVectorElementType();
     SDValue Elt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, ElemVT, Value,
                               DAG.getConstant(i, MVT::i32));
-    Elt = DAG.getZExtOrTrunc(Elt, DL, PackedVT);
-    Elt = DAG.getNode(ISD::AND, DL, PackedVT, Elt, Mask);
-    SDValue Shift = DAG.getConstant(MemEltBits * i, PackedVT);
-    Elt = DAG.getNode(ISD::SHL, DL, PackedVT, Elt, Shift);
+    Elt = DAG.getZExtOrTrunc(Elt, DL, MVT::i32);
+    Elt = DAG.getNode(ISD::AND, DL, MVT::i32, Elt, Mask); // getZeroExtendInReg
+
+    SDValue Shift = DAG.getConstant(MemEltBits * i, MVT::i32);
+    Elt = DAG.getNode(ISD::SHL, DL, MVT::i32, Elt, Shift);
+
     if (i == 0) {
       PackedValue = Elt;
     } else {
-      PackedValue = DAG.getNode(ISD::OR, DL, PackedVT, PackedValue, Elt);
+      PackedValue = DAG.getNode(ISD::OR, DL, MVT::i32, PackedValue, Elt);
     }
   }
+
+  if (PackedSize < 32) {
+    EVT PackedVT = EVT::getIntegerVT(*DAG.getContext(), PackedSize);
+    return DAG.getTruncStore(Store->getChain(), DL, PackedValue, Ptr,
+                             Store->getMemOperand()->getPointerInfo(),
+                             PackedVT,
+                             Store->isNonTemporal(), Store->isVolatile(),
+                             Store->getAlignment());
+  }
+
   return DAG.getStore(Store->getChain(), DL, PackedValue, Ptr,
-                      MachinePointerInfo(Store->getMemOperand()->getValue()),
+                      Store->getMemOperand()->getPointerInfo(),
                       Store->isVolatile(),  Store->isNonTemporal(),
                       Store->getAlignment());
 }
@@ -585,34 +1128,404 @@ SDValue AMDGPUTargetLowering::SplitVectorStore(SDValue Op,
                          MemEltVT, Store->isVolatile(), Store->isNonTemporal(),
                          Store->getAlignment()));
   }
-  return DAG.getNode(ISD::TokenFactor, SL, MVT::Other, &Chains[0], NumElts);
+  return DAG.getNode(ISD::TokenFactor, SL, MVT::Other, Chains);
+}
+
+SDValue AMDGPUTargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  LoadSDNode *Load = cast<LoadSDNode>(Op);
+  ISD::LoadExtType ExtType = Load->getExtensionType();
+  EVT VT = Op.getValueType();
+  EVT MemVT = Load->getMemoryVT();
+
+  if (ExtType != ISD::NON_EXTLOAD && !VT.isVector() && VT.getSizeInBits() > 32) {
+    // We can do the extload to 32-bits, and then need to separately extend to
+    // 64-bits.
+
+    SDValue ExtLoad32 = DAG.getExtLoad(ExtType, DL, MVT::i32,
+                                       Load->getChain(),
+                                       Load->getBasePtr(),
+                                       MemVT,
+                                       Load->getMemOperand());
+
+    SDValue Ops[] = {
+      DAG.getNode(ISD::getExtForLoadExtType(ExtType), DL, VT, ExtLoad32),
+      ExtLoad32.getValue(1)
+    };
+
+    return DAG.getMergeValues(Ops, DL);
+  }
+
+  if (ExtType == ISD::NON_EXTLOAD && VT.getSizeInBits() < 32) {
+    assert(VT == MVT::i1 && "Only i1 non-extloads expected");
+    // FIXME: Copied from PPC
+    // First, load into 32 bits, then truncate to 1 bit.
+
+    SDValue Chain = Load->getChain();
+    SDValue BasePtr = Load->getBasePtr();
+    MachineMemOperand *MMO = Load->getMemOperand();
+
+    SDValue NewLD = DAG.getExtLoad(ISD::EXTLOAD, DL, MVT::i32, Chain,
+                                   BasePtr, MVT::i8, MMO);
+
+    SDValue Ops[] = {
+      DAG.getNode(ISD::TRUNCATE, DL, VT, NewLD),
+      NewLD.getValue(1)
+    };
+
+    return DAG.getMergeValues(Ops, DL);
+  }
+
+  if (Subtarget->getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS ||
+      Load->getAddressSpace() != AMDGPUAS::PRIVATE_ADDRESS ||
+      ExtType == ISD::NON_EXTLOAD || Load->getMemoryVT().bitsGE(MVT::i32))
+    return SDValue();
+
+
+  SDValue Ptr = DAG.getNode(ISD::SRL, DL, MVT::i32, Load->getBasePtr(),
+                            DAG.getConstant(2, MVT::i32));
+  SDValue Ret = DAG.getNode(AMDGPUISD::REGISTER_LOAD, DL, Op.getValueType(),
+                            Load->getChain(), Ptr,
+                            DAG.getTargetConstant(0, MVT::i32),
+                            Op.getOperand(2));
+  SDValue ByteIdx = DAG.getNode(ISD::AND, DL, MVT::i32,
+                                Load->getBasePtr(),
+                                DAG.getConstant(0x3, MVT::i32));
+  SDValue ShiftAmt = DAG.getNode(ISD::SHL, DL, MVT::i32, ByteIdx,
+                                 DAG.getConstant(3, MVT::i32));
+
+  Ret = DAG.getNode(ISD::SRL, DL, MVT::i32, Ret, ShiftAmt);
+
+  EVT MemEltVT = MemVT.getScalarType();
+  if (ExtType == ISD::SEXTLOAD) {
+    SDValue MemEltVTNode = DAG.getValueType(MemEltVT);
+
+    SDValue Ops[] = {
+      DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i32, Ret, MemEltVTNode),
+      Load->getChain()
+    };
+
+    return DAG.getMergeValues(Ops, DL);
+  }
+
+  SDValue Ops[] = {
+    DAG.getZeroExtendInReg(Ret, DL, MemEltVT),
+    Load->getChain()
+  };
+
+  return DAG.getMergeValues(Ops, DL);
 }
 
 SDValue AMDGPUTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
   SDValue Result = AMDGPUTargetLowering::MergeVectorStore(Op, DAG);
   if (Result.getNode()) {
     return Result;
   }
 
   StoreSDNode *Store = cast<StoreSDNode>(Op);
+  SDValue Chain = Store->getChain();
   if ((Store->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS ||
        Store->getAddressSpace() == AMDGPUAS::PRIVATE_ADDRESS) &&
       Store->getValue().getValueType().isVector()) {
     return SplitVectorStore(Op, DAG);
   }
+
+  EVT MemVT = Store->getMemoryVT();
+  if (Store->getAddressSpace() == AMDGPUAS::PRIVATE_ADDRESS &&
+      MemVT.bitsLT(MVT::i32)) {
+    unsigned Mask = 0;
+    if (Store->getMemoryVT() == MVT::i8) {
+      Mask = 0xff;
+    } else if (Store->getMemoryVT() == MVT::i16) {
+      Mask = 0xffff;
+    }
+    SDValue BasePtr = Store->getBasePtr();
+    SDValue Ptr = DAG.getNode(ISD::SRL, DL, MVT::i32, BasePtr,
+                              DAG.getConstant(2, MVT::i32));
+    SDValue Dst = DAG.getNode(AMDGPUISD::REGISTER_LOAD, DL, MVT::i32,
+                              Chain, Ptr, DAG.getTargetConstant(0, MVT::i32));
+
+    SDValue ByteIdx = DAG.getNode(ISD::AND, DL, MVT::i32, BasePtr,
+                                  DAG.getConstant(0x3, MVT::i32));
+
+    SDValue ShiftAmt = DAG.getNode(ISD::SHL, DL, MVT::i32, ByteIdx,
+                                   DAG.getConstant(3, MVT::i32));
+
+    SDValue SExtValue = DAG.getNode(ISD::SIGN_EXTEND, DL, MVT::i32,
+                                    Store->getValue());
+
+    SDValue MaskedValue = DAG.getZeroExtendInReg(SExtValue, DL, MemVT);
+
+    SDValue ShiftedValue = DAG.getNode(ISD::SHL, DL, MVT::i32,
+                                       MaskedValue, ShiftAmt);
+
+    SDValue DstMask = DAG.getNode(ISD::SHL, DL, MVT::i32, DAG.getConstant(Mask, MVT::i32),
+                                  ShiftAmt);
+    DstMask = DAG.getNode(ISD::XOR, DL, MVT::i32, DstMask,
+                          DAG.getConstant(0xffffffff, MVT::i32));
+    Dst = DAG.getNode(ISD::AND, DL, MVT::i32, Dst, DstMask);
+
+    SDValue Value = DAG.getNode(ISD::OR, DL, MVT::i32, Dst, ShiftedValue);
+    return DAG.getNode(AMDGPUISD::REGISTER_STORE, DL, MVT::Other,
+                       Chain, Value, Ptr, DAG.getTargetConstant(0, MVT::i32));
+  }
   return SDValue();
 }
 
+SDValue AMDGPUTargetLowering::LowerSDIV24(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  EVT OVT = Op.getValueType();
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  MVT INTTY;
+  MVT FLTTY;
+  if (!OVT.isVector()) {
+    INTTY = MVT::i32;
+    FLTTY = MVT::f32;
+  } else if (OVT.getVectorNumElements() == 2) {
+    INTTY = MVT::v2i32;
+    FLTTY = MVT::v2f32;
+  } else if (OVT.getVectorNumElements() == 4) {
+    INTTY = MVT::v4i32;
+    FLTTY = MVT::v4f32;
+  }
+  unsigned bitsize = OVT.getScalarType().getSizeInBits();
+  // char|short jq = ia ^ ib;
+  SDValue jq = DAG.getNode(ISD::XOR, DL, OVT, LHS, RHS);
+
+  // jq = jq >> (bitsize - 2)
+  jq = DAG.getNode(ISD::SRA, DL, OVT, jq, DAG.getConstant(bitsize - 2, OVT));
+
+  // jq = jq | 0x1
+  jq = DAG.getNode(ISD::OR, DL, OVT, jq, DAG.getConstant(1, OVT));
+
+  // jq = (int)jq
+  jq = DAG.getSExtOrTrunc(jq, DL, INTTY);
+
+  // int ia = (int)LHS;
+  SDValue ia = DAG.getSExtOrTrunc(LHS, DL, INTTY);
+
+  // int ib, (int)RHS;
+  SDValue ib = DAG.getSExtOrTrunc(RHS, DL, INTTY);
+
+  // float fa = (float)ia;
+  SDValue fa = DAG.getNode(ISD::SINT_TO_FP, DL, FLTTY, ia);
+
+  // float fb = (float)ib;
+  SDValue fb = DAG.getNode(ISD::SINT_TO_FP, DL, FLTTY, ib);
+
+  // float fq = native_divide(fa, fb);
+  SDValue fq = DAG.getNode(ISD::FMUL, DL, FLTTY,
+                           fa, DAG.getNode(AMDGPUISD::RCP, DL, FLTTY, fb));
+
+  // fq = trunc(fq);
+  fq = DAG.getNode(ISD::FTRUNC, DL, FLTTY, fq);
+
+  // float fqneg = -fq;
+  SDValue fqneg = DAG.getNode(ISD::FNEG, DL, FLTTY, fq);
+
+  // float fr = mad(fqneg, fb, fa);
+  SDValue fr = DAG.getNode(ISD::FADD, DL, FLTTY,
+      DAG.getNode(ISD::MUL, DL, FLTTY, fqneg, fb), fa);
+
+  // int iq = (int)fq;
+  SDValue iq = DAG.getNode(ISD::FP_TO_SINT, DL, INTTY, fq);
+
+  // fr = fabs(fr);
+  fr = DAG.getNode(ISD::FABS, DL, FLTTY, fr);
+
+  // fb = fabs(fb);
+  fb = DAG.getNode(ISD::FABS, DL, FLTTY, fb);
+
+  // int cv = fr >= fb;
+  SDValue cv;
+  if (INTTY == MVT::i32) {
+    cv = DAG.getSetCC(DL, INTTY, fr, fb, ISD::SETOGE);
+  } else {
+    cv = DAG.getSetCC(DL, INTTY, fr, fb, ISD::SETOGE);
+  }
+  // jq = (cv ? jq : 0);
+  jq = DAG.getNode(ISD::SELECT, DL, OVT, cv, jq,
+      DAG.getConstant(0, OVT));
+  // dst = iq + jq;
+  iq = DAG.getSExtOrTrunc(iq, DL, OVT);
+  iq = DAG.getNode(ISD::ADD, DL, OVT, iq, jq);
+  return iq;
+}
+
+SDValue AMDGPUTargetLowering::LowerSDIV32(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  EVT OVT = Op.getValueType();
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  // The LowerSDIV32 function generates equivalent to the following IL.
+  // mov r0, LHS
+  // mov r1, RHS
+  // ilt r10, r0, 0
+  // ilt r11, r1, 0
+  // iadd r0, r0, r10
+  // iadd r1, r1, r11
+  // ixor r0, r0, r10
+  // ixor r1, r1, r11
+  // udiv r0, r0, r1
+  // ixor r10, r10, r11
+  // iadd r0, r0, r10
+  // ixor DST, r0, r10
+
+  // mov r0, LHS
+  SDValue r0 = LHS;
+
+  // mov r1, RHS
+  SDValue r1 = RHS;
+
+  // ilt r10, r0, 0
+  SDValue r10 = DAG.getSelectCC(DL,
+      r0, DAG.getConstant(0, OVT),
+      DAG.getConstant(-1, OVT),
+      DAG.getConstant(0, OVT),
+      ISD::SETLT);
+
+  // ilt r11, r1, 0
+  SDValue r11 = DAG.getSelectCC(DL,
+      r1, DAG.getConstant(0, OVT),
+      DAG.getConstant(-1, OVT),
+      DAG.getConstant(0, OVT),
+      ISD::SETLT);
+
+  // iadd r0, r0, r10
+  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
+
+  // iadd r1, r1, r11
+  r1 = DAG.getNode(ISD::ADD, DL, OVT, r1, r11);
+
+  // ixor r0, r0, r10
+  r0 = DAG.getNode(ISD::XOR, DL, OVT, r0, r10);
+
+  // ixor r1, r1, r11
+  r1 = DAG.getNode(ISD::XOR, DL, OVT, r1, r11);
+
+  // udiv r0, r0, r1
+  r0 = DAG.getNode(ISD::UDIV, DL, OVT, r0, r1);
+
+  // ixor r10, r10, r11
+  r10 = DAG.getNode(ISD::XOR, DL, OVT, r10, r11);
+
+  // iadd r0, r0, r10
+  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
+
+  // ixor DST, r0, r10
+  SDValue DST = DAG.getNode(ISD::XOR, DL, OVT, r0, r10);
+  return DST;
+}
+
+SDValue AMDGPUTargetLowering::LowerSDIV64(SDValue Op, SelectionDAG &DAG) const {
+  return SDValue(Op.getNode(), 0);
+}
+
+SDValue AMDGPUTargetLowering::LowerSDIV(SDValue Op, SelectionDAG &DAG) const {
+  EVT OVT = Op.getValueType().getScalarType();
+
+  if (OVT == MVT::i64)
+    return LowerSDIV64(Op, DAG);
+
+  if (OVT.getScalarType() == MVT::i32)
+    return LowerSDIV32(Op, DAG);
+
+  if (OVT == MVT::i16 || OVT == MVT::i8) {
+    // FIXME: We should be checking for the masked bits. This isn't reached
+    // because i8 and i16 are not legal types.
+    return LowerSDIV24(Op, DAG);
+  }
+
+  return SDValue(Op.getNode(), 0);
+}
+
+SDValue AMDGPUTargetLowering::LowerSREM32(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  EVT OVT = Op.getValueType();
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  // The LowerSREM32 function generates equivalent to the following IL.
+  // mov r0, LHS
+  // mov r1, RHS
+  // ilt r10, r0, 0
+  // ilt r11, r1, 0
+  // iadd r0, r0, r10
+  // iadd r1, r1, r11
+  // ixor r0, r0, r10
+  // ixor r1, r1, r11
+  // udiv r20, r0, r1
+  // umul r20, r20, r1
+  // sub r0, r0, r20
+  // iadd r0, r0, r10
+  // ixor DST, r0, r10
+
+  // mov r0, LHS
+  SDValue r0 = LHS;
+
+  // mov r1, RHS
+  SDValue r1 = RHS;
+
+  // ilt r10, r0, 0
+  SDValue r10 = DAG.getSetCC(DL, OVT, r0, DAG.getConstant(0, OVT), ISD::SETLT);
+
+  // ilt r11, r1, 0
+  SDValue r11 = DAG.getSetCC(DL, OVT, r1, DAG.getConstant(0, OVT), ISD::SETLT);
+
+  // iadd r0, r0, r10
+  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
+
+  // iadd r1, r1, r11
+  r1 = DAG.getNode(ISD::ADD, DL, OVT, r1, r11);
+
+  // ixor r0, r0, r10
+  r0 = DAG.getNode(ISD::XOR, DL, OVT, r0, r10);
+
+  // ixor r1, r1, r11
+  r1 = DAG.getNode(ISD::XOR, DL, OVT, r1, r11);
+
+  // udiv r20, r0, r1
+  SDValue r20 = DAG.getNode(ISD::UREM, DL, OVT, r0, r1);
+
+  // umul r20, r20, r1
+  r20 = DAG.getNode(AMDGPUISD::UMUL, DL, OVT, r20, r1);
+
+  // sub r0, r0, r20
+  r0 = DAG.getNode(ISD::SUB, DL, OVT, r0, r20);
+
+  // iadd r0, r0, r10
+  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
+
+  // ixor DST, r0, r10
+  SDValue DST = DAG.getNode(ISD::XOR, DL, OVT, r0, r10);
+  return DST;
+}
+
+SDValue AMDGPUTargetLowering::LowerSREM64(SDValue Op, SelectionDAG &DAG) const {
+  return SDValue(Op.getNode(), 0);
+}
+
+SDValue AMDGPUTargetLowering::LowerSREM(SDValue Op, SelectionDAG &DAG) const {
+  EVT OVT = Op.getValueType();
+
+  if (OVT.getScalarType() == MVT::i64)
+    return LowerSREM64(Op, DAG);
+
+  if (OVT.getScalarType() == MVT::i32)
+    return LowerSREM32(Op, DAG);
+
+  return SDValue(Op.getNode(), 0);
+}
+
 SDValue AMDGPUTargetLowering::LowerUDIVREM(SDValue Op,
-    SelectionDAG &DAG) const {
+                                           SelectionDAG &DAG) const {
   SDLoc DL(Op);
   EVT VT = Op.getValueType();
 
   SDValue Num = Op.getOperand(0);
   SDValue Den = Op.getOperand(1);
 
-  SmallVector<SDValue, 8> Results;
-
   // RCP =  URECIP(Den) = 2^32 / Den + e
   // e is rounding error.
   SDValue RCP = DAG.getNode(AMDGPUISD::URECIP, DL, VT, Den);
@@ -702,10 +1615,182 @@ SDValue AMDGPUTargetLowering::LowerUDIVREM(SDValue Op,
   // Rem = (Remainder_GE_Zero == 0 ? Remainder_A_Den : Rem)
   Rem = DAG.getSelectCC(DL, Remainder_GE_Zero, DAG.getConstant(0, VT),
                             Remainder_A_Den, Rem, ISD::SETEQ);
-  SDValue Ops[2];
-  Ops[0] = Div;
-  Ops[1] = Rem;
-  return DAG.getMergeValues(Ops, 2, DL);
+  SDValue Ops[2] = {
+    Div,
+    Rem
+  };
+  return DAG.getMergeValues(Ops, DL);
+}
+
+SDValue AMDGPUTargetLowering::LowerSDIVREM(SDValue Op,
+                                           SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  EVT VT = Op.getValueType();
+
+  SDValue Zero = DAG.getConstant(0, VT);
+  SDValue NegOne = DAG.getConstant(-1, VT);
+
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+
+  SDValue LHSign = DAG.getSelectCC(DL, LHS, Zero, NegOne, Zero, ISD::SETLT);
+  SDValue RHSign = DAG.getSelectCC(DL, RHS, Zero, NegOne, Zero, ISD::SETLT);
+  SDValue DSign = DAG.getNode(ISD::XOR, DL, VT, LHSign, RHSign);
+  SDValue RSign = LHSign; // Remainder sign is the same as LHS
+
+  LHS = DAG.getNode(ISD::ADD, DL, VT, LHS, LHSign);
+  RHS = DAG.getNode(ISD::ADD, DL, VT, RHS, RHSign);
+
+  LHS = DAG.getNode(ISD::XOR, DL, VT, LHS, LHSign);
+  RHS = DAG.getNode(ISD::XOR, DL, VT, RHS, RHSign);
+
+  SDValue Div = DAG.getNode(ISD::UDIVREM, DL, DAG.getVTList(VT, VT), LHS, RHS);
+  SDValue Rem = Div.getValue(1);
+
+  Div = DAG.getNode(ISD::XOR, DL, VT, Div, DSign);
+  Rem = DAG.getNode(ISD::XOR, DL, VT, Rem, RSign);
+
+  Div = DAG.getNode(ISD::SUB, DL, VT, Div, DSign);
+  Rem = DAG.getNode(ISD::SUB, DL, VT, Rem, RSign);
+
+  SDValue Res[2] = {
+    Div,
+    Rem
+  };
+  return DAG.getMergeValues(Res, DL);
+}
+
+SDValue AMDGPUTargetLowering::LowerFCEIL(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc SL(Op);
+  SDValue Src = Op.getOperand(0);
+
+  // result = trunc(src)
+  // if (src > 0.0 && src != result)
+  //   result += 1.0
+
+  SDValue Trunc = DAG.getNode(ISD::FTRUNC, SL, MVT::f64, Src);
+
+  const SDValue Zero = DAG.getConstantFP(0.0, MVT::f64);
+  const SDValue One = DAG.getConstantFP(1.0, MVT::f64);
+
+  EVT SetCCVT = getSetCCResultType(*DAG.getContext(), MVT::f64);
+
+  SDValue Lt0 = DAG.getSetCC(SL, SetCCVT, Src, Zero, ISD::SETOGT);
+  SDValue NeTrunc = DAG.getSetCC(SL, SetCCVT, Src, Trunc, ISD::SETONE);
+  SDValue And = DAG.getNode(ISD::AND, SL, SetCCVT, Lt0, NeTrunc);
+
+  SDValue Add = DAG.getNode(ISD::SELECT, SL, MVT::f64, And, One, Zero);
+  return DAG.getNode(ISD::FADD, SL, MVT::f64, Trunc, Add);
+}
+
+SDValue AMDGPUTargetLowering::LowerFTRUNC(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc SL(Op);
+  SDValue Src = Op.getOperand(0);
+
+  assert(Op.getValueType() == MVT::f64);
+
+  const SDValue Zero = DAG.getConstant(0, MVT::i32);
+  const SDValue One = DAG.getConstant(1, MVT::i32);
+
+  SDValue VecSrc = DAG.getNode(ISD::BITCAST, SL, MVT::v2i32, Src);
+
+  // Extract the upper half, since this is where we will find the sign and
+  // exponent.
+  SDValue Hi = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SL, MVT::i32, VecSrc, One);
+
+  const unsigned FractBits = 52;
+  const unsigned ExpBits = 11;
+
+  // Extract the exponent.
+  SDValue ExpPart = DAG.getNode(AMDGPUISD::BFE_I32, SL, MVT::i32,
+                                Hi,
+                                DAG.getConstant(FractBits - 32, MVT::i32),
+                                DAG.getConstant(ExpBits, MVT::i32));
+  SDValue Exp = DAG.getNode(ISD::SUB, SL, MVT::i32, ExpPart,
+                            DAG.getConstant(1023, MVT::i32));
+
+  // Extract the sign bit.
+  const SDValue SignBitMask = DAG.getConstant(UINT32_C(1) << 31, MVT::i32);
+  SDValue SignBit = DAG.getNode(ISD::AND, SL, MVT::i32, Hi, SignBitMask);
+
+  // Extend back to to 64-bits.
+  SDValue SignBit64 = DAG.getNode(ISD::BUILD_VECTOR, SL, MVT::v2i32,
+                                  Zero, SignBit);
+  SignBit64 = DAG.getNode(ISD::BITCAST, SL, MVT::i64, SignBit64);
+
+  SDValue BcInt = DAG.getNode(ISD::BITCAST, SL, MVT::i64, Src);
+  const SDValue FractMask
+    = DAG.getConstant((UINT64_C(1) << FractBits) - 1, MVT::i64);
+
+  SDValue Shr = DAG.getNode(ISD::SRA, SL, MVT::i64, FractMask, Exp);
+  SDValue Not = DAG.getNOT(SL, Shr, MVT::i64);
+  SDValue Tmp0 = DAG.getNode(ISD::AND, SL, MVT::i64, BcInt, Not);
+
+  EVT SetCCVT = getSetCCResultType(*DAG.getContext(), MVT::i32);
+
+  const SDValue FiftyOne = DAG.getConstant(FractBits - 1, MVT::i32);
+
+  SDValue ExpLt0 = DAG.getSetCC(SL, SetCCVT, Exp, Zero, ISD::SETLT);
+  SDValue ExpGt51 = DAG.getSetCC(SL, SetCCVT, Exp, FiftyOne, ISD::SETGT);
+
+  SDValue Tmp1 = DAG.getNode(ISD::SELECT, SL, MVT::i64, ExpLt0, SignBit64, Tmp0);
+  SDValue Tmp2 = DAG.getNode(ISD::SELECT, SL, MVT::i64, ExpGt51, BcInt, Tmp1);
+
+  return DAG.getNode(ISD::BITCAST, SL, MVT::f64, Tmp2);
+}
+
+SDValue AMDGPUTargetLowering::LowerFRINT(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc SL(Op);
+  SDValue Src = Op.getOperand(0);
+
+  assert(Op.getValueType() == MVT::f64);
+
+  APFloat C1Val(APFloat::IEEEdouble, "0x1.0p+52");
+  SDValue C1 = DAG.getConstantFP(C1Val, MVT::f64);
+  SDValue CopySign = DAG.getNode(ISD::FCOPYSIGN, SL, MVT::f64, C1, Src);
+
+  SDValue Tmp1 = DAG.getNode(ISD::FADD, SL, MVT::f64, Src, CopySign);
+  SDValue Tmp2 = DAG.getNode(ISD::FSUB, SL, MVT::f64, Tmp1, CopySign);
+
+  SDValue Fabs = DAG.getNode(ISD::FABS, SL, MVT::f64, Src);
+
+  APFloat C2Val(APFloat::IEEEdouble, "0x1.fffffffffffffp+51");
+  SDValue C2 = DAG.getConstantFP(C2Val, MVT::f64);
+
+  EVT SetCCVT = getSetCCResultType(*DAG.getContext(), MVT::f64);
+  SDValue Cond = DAG.getSetCC(SL, SetCCVT, Fabs, C2, ISD::SETOGT);
+
+  return DAG.getSelect(SL, MVT::f64, Cond, Src, Tmp2);
+}
+
+SDValue AMDGPUTargetLowering::LowerFNEARBYINT(SDValue Op, SelectionDAG &DAG) const {
+  // FNEARBYINT and FRINT are the same, except in their handling of FP
+  // exceptions. Those aren't really meaningful for us, and OpenCL only has
+  // rint, so just treat them as equivalent.
+  return DAG.getNode(ISD::FRINT, SDLoc(Op), Op.getValueType(), Op.getOperand(0));
+}
+
+SDValue AMDGPUTargetLowering::LowerFFLOOR(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc SL(Op);
+  SDValue Src = Op.getOperand(0);
+
+  // result = trunc(src);
+  // if (src < 0.0 && src != result)
+  //   result += -1.0.
+
+  SDValue Trunc = DAG.getNode(ISD::FTRUNC, SL, MVT::f64, Src);
+
+  const SDValue Zero = DAG.getConstantFP(0.0, MVT::f64);
+  const SDValue NegOne = DAG.getConstantFP(-1.0, MVT::f64);
+
+  EVT SetCCVT = getSetCCResultType(*DAG.getContext(), MVT::f64);
+
+  SDValue Lt0 = DAG.getSetCC(SL, SetCCVT, Src, Zero, ISD::SETOLT);
+  SDValue NeTrunc = DAG.getSetCC(SL, SetCCVT, Src, Trunc, ISD::SETONE);
+  SDValue And = DAG.getNode(ISD::AND, SL, SetCCVT, Lt0, NeTrunc);
+
+  SDValue Add = DAG.getNode(ISD::SELECT, SL, MVT::f64, And, NegOne, Zero);
+  return DAG.getNode(ISD::FADD, SL, MVT::f64, Trunc, Add);
 }
 
 SDValue AMDGPUTargetLowering::LowerUINT_TO_FP(SDValue Op,
@@ -725,7 +1810,275 @@ SDValue AMDGPUTargetLowering::LowerUINT_TO_FP(SDValue Op,
   FloatHi = DAG.getNode(ISD::FMUL, DL, MVT::f32, FloatHi,
                         DAG.getConstantFP(4294967296.0f, MVT::f32)); // 2^32
   return DAG.getNode(ISD::FADD, DL, MVT::f32, FloatLo, FloatHi);
+}
 
+SDValue AMDGPUTargetLowering::ExpandSIGN_EXTEND_INREG(SDValue Op,
+                                                      unsigned BitsDiff,
+                                                      SelectionDAG &DAG) const {
+  MVT VT = Op.getSimpleValueType();
+  SDLoc DL(Op);
+  SDValue Shift = DAG.getConstant(BitsDiff, VT);
+  // Shift left by 'Shift' bits.
+  SDValue Shl = DAG.getNode(ISD::SHL, DL, VT, Op.getOperand(0), Shift);
+  // Signed shift Right by 'Shift' bits.
+  return DAG.getNode(ISD::SRA, DL, VT, Shl, Shift);
+}
+
+SDValue AMDGPUTargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op,
+                                                     SelectionDAG &DAG) const {
+  EVT ExtraVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
+  MVT VT = Op.getSimpleValueType();
+  MVT ScalarVT = VT.getScalarType();
+
+  if (!VT.isVector())
+    return SDValue();
+
+  SDValue Src = Op.getOperand(0);
+  SDLoc DL(Op);
+
+  // TODO: Don't scalarize on Evergreen?
+  unsigned NElts = VT.getVectorNumElements();
+  SmallVector<SDValue, 8> Args;
+  DAG.ExtractVectorElements(Src, Args, 0, NElts);
+
+  SDValue VTOp = DAG.getValueType(ExtraVT.getScalarType());
+  for (unsigned I = 0; I < NElts; ++I)
+    Args[I] = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, ScalarVT, Args[I], VTOp);
+
+  return DAG.getNode(ISD::BUILD_VECTOR, DL, VT, Args);
+}
+
+//===----------------------------------------------------------------------===//
+// Custom DAG optimizations
+//===----------------------------------------------------------------------===//
+
+static bool isU24(SDValue Op, SelectionDAG &DAG) {
+  APInt KnownZero, KnownOne;
+  EVT VT = Op.getValueType();
+  DAG.computeKnownBits(Op, KnownZero, KnownOne);
+
+  return (VT.getSizeInBits() - KnownZero.countLeadingOnes()) <= 24;
+}
+
+static bool isI24(SDValue Op, SelectionDAG &DAG) {
+  EVT VT = Op.getValueType();
+
+  // In order for this to be a signed 24-bit value, bit 23, must
+  // be a sign bit.
+  return VT.getSizeInBits() >= 24 && // Types less than 24-bit should be treated
+                                     // as unsigned 24-bit values.
+         (VT.getSizeInBits() - DAG.ComputeNumSignBits(Op)) < 24;
+}
+
+static void simplifyI24(SDValue Op, TargetLowering::DAGCombinerInfo &DCI) {
+
+  SelectionDAG &DAG = DCI.DAG;
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  EVT VT = Op.getValueType();
+
+  APInt Demanded = APInt::getLowBitsSet(VT.getSizeInBits(), 24);
+  APInt KnownZero, KnownOne;
+  TargetLowering::TargetLoweringOpt TLO(DAG, true, true);
+  if (TLI.SimplifyDemandedBits(Op, Demanded, KnownZero, KnownOne, TLO))
+    DCI.CommitTargetLoweringOpt(TLO);
+}
+
+template <typename IntTy>
+static SDValue constantFoldBFE(SelectionDAG &DAG, IntTy Src0,
+                               uint32_t Offset, uint32_t Width) {
+  if (Width + Offset < 32) {
+    IntTy Result = (Src0 << (32 - Offset - Width)) >> (32 - Width);
+    return DAG.getConstant(Result, MVT::i32);
+  }
+
+  return DAG.getConstant(Src0 >> Offset, MVT::i32);
+}
+
+static bool usesAllNormalStores(SDNode *LoadVal) {
+  for (SDNode::use_iterator I = LoadVal->use_begin(); !I.atEnd(); ++I) {
+    if (!ISD::isNormalStore(*I))
+      return false;
+  }
+
+  return true;
+}
+
+// If we have a copy of an illegal type, replace it with a load / store of an
+// equivalently sized legal type. This avoids intermediate bit pack / unpack
+// instructions emitted when handling extloads and truncstores. Ideally we could
+// recognize the pack / unpack pattern to eliminate it.
+SDValue AMDGPUTargetLowering::performStoreCombine(SDNode *N,
+                                                  DAGCombinerInfo &DCI) const {
+  if (!DCI.isBeforeLegalize())
+    return SDValue();
+
+  StoreSDNode *SN = cast<StoreSDNode>(N);
+  SDValue Value = SN->getValue();
+  EVT VT = Value.getValueType();
+
+  if (isTypeLegal(VT) || SN->isVolatile() || !ISD::isNormalLoad(Value.getNode()))
+    return SDValue();
+
+  LoadSDNode *LoadVal = cast<LoadSDNode>(Value);
+  if (LoadVal->isVolatile() || !usesAllNormalStores(LoadVal))
+    return SDValue();
+
+  EVT MemVT = LoadVal->getMemoryVT();
+
+  SDLoc SL(N);
+  SelectionDAG &DAG = DCI.DAG;
+  EVT LoadVT = getEquivalentMemType(*DAG.getContext(), MemVT);
+
+  SDValue NewLoad = DAG.getLoad(ISD::UNINDEXED, ISD::NON_EXTLOAD,
+                                LoadVT, SL,
+                                LoadVal->getChain(),
+                                LoadVal->getBasePtr(),
+                                LoadVal->getOffset(),
+                                LoadVT,
+                                LoadVal->getMemOperand());
+
+  SDValue CastLoad = DAG.getNode(ISD::BITCAST, SL, VT, NewLoad.getValue(0));
+  DCI.CombineTo(LoadVal, CastLoad, NewLoad.getValue(1), false);
+
+  return DAG.getStore(SN->getChain(), SL, NewLoad,
+                      SN->getBasePtr(), SN->getMemOperand());
+}
+
+SDValue AMDGPUTargetLowering::performMulCombine(SDNode *N,
+                                                DAGCombinerInfo &DCI) const {
+  EVT VT = N->getValueType(0);
+
+  if (VT.isVector() || VT.getSizeInBits() > 32)
+    return SDValue();
+
+  SelectionDAG &DAG = DCI.DAG;
+  SDLoc DL(N);
+
+  SDValue N0 = N->getOperand(0);
+  SDValue N1 = N->getOperand(1);
+  SDValue Mul;
+
+  if (Subtarget->hasMulU24() && isU24(N0, DAG) && isU24(N1, DAG)) {
+    N0 = DAG.getZExtOrTrunc(N0, DL, MVT::i32);
+    N1 = DAG.getZExtOrTrunc(N1, DL, MVT::i32);
+    Mul = DAG.getNode(AMDGPUISD::MUL_U24, DL, MVT::i32, N0, N1);
+  } else if (Subtarget->hasMulI24() && isI24(N0, DAG) && isI24(N1, DAG)) {
+    N0 = DAG.getSExtOrTrunc(N0, DL, MVT::i32);
+    N1 = DAG.getSExtOrTrunc(N1, DL, MVT::i32);
+    Mul = DAG.getNode(AMDGPUISD::MUL_I24, DL, MVT::i32, N0, N1);
+  } else {
+    return SDValue();
+  }
+
+  // We need to use sext even for MUL_U24, because MUL_U24 is used
+  // for signed multiply of 8 and 16-bit types.
+  return DAG.getSExtOrTrunc(Mul, DL, VT);
+}
+
+SDValue AMDGPUTargetLowering::PerformDAGCombine(SDNode *N,
+                                                DAGCombinerInfo &DCI) const {
+  SelectionDAG &DAG = DCI.DAG;
+  SDLoc DL(N);
+
+  switch(N->getOpcode()) {
+    default: break;
+    case ISD::MUL:
+      return performMulCombine(N, DCI);
+    case AMDGPUISD::MUL_I24:
+    case AMDGPUISD::MUL_U24: {
+      SDValue N0 = N->getOperand(0);
+      SDValue N1 = N->getOperand(1);
+      simplifyI24(N0, DCI);
+      simplifyI24(N1, DCI);
+      return SDValue();
+    }
+    case ISD::SELECT_CC: {
+      return CombineMinMax(N, DAG);
+    }
+  case AMDGPUISD::BFE_I32:
+  case AMDGPUISD::BFE_U32: {
+    assert(!N->getValueType(0).isVector() &&
+           "Vector handling of BFE not implemented");
+    ConstantSDNode *Width = dyn_cast<ConstantSDNode>(N->getOperand(2));
+    if (!Width)
+      break;
+
+    uint32_t WidthVal = Width->getZExtValue() & 0x1f;
+    if (WidthVal == 0)
+      return DAG.getConstant(0, MVT::i32);
+
+    ConstantSDNode *Offset = dyn_cast<ConstantSDNode>(N->getOperand(1));
+    if (!Offset)
+      break;
+
+    SDValue BitsFrom = N->getOperand(0);
+    uint32_t OffsetVal = Offset->getZExtValue() & 0x1f;
+
+    bool Signed = N->getOpcode() == AMDGPUISD::BFE_I32;
+
+    if (OffsetVal == 0) {
+      // This is already sign / zero extended, so try to fold away extra BFEs.
+      unsigned SignBits =  Signed ? (32 - WidthVal + 1) : (32 - WidthVal);
+
+      unsigned OpSignBits = DAG.ComputeNumSignBits(BitsFrom);
+      if (OpSignBits >= SignBits)
+        return BitsFrom;
+
+      EVT SmallVT = EVT::getIntegerVT(*DAG.getContext(), WidthVal);
+      if (Signed) {
+        // This is a sign_extend_inreg. Replace it to take advantage of existing
+        // DAG Combines. If not eliminated, we will match back to BFE during
+        // selection.
+
+        // TODO: The sext_inreg of extended types ends, although we can could
+        // handle them in a single BFE.
+        return DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i32, BitsFrom,
+                           DAG.getValueType(SmallVT));
+      }
+
+      return DAG.getZeroExtendInReg(BitsFrom, DL, SmallVT);
+    }
+
+    if (ConstantSDNode *Val = dyn_cast<ConstantSDNode>(N->getOperand(0))) {
+      if (Signed) {
+        return constantFoldBFE<int32_t>(DAG,
+                                        Val->getSExtValue(),
+                                        OffsetVal,
+                                        WidthVal);
+      }
+
+      return constantFoldBFE<uint32_t>(DAG,
+                                       Val->getZExtValue(),
+                                       OffsetVal,
+                                       WidthVal);
+    }
+
+    APInt Demanded = APInt::getBitsSet(32,
+                                       OffsetVal,
+                                       OffsetVal + WidthVal);
+
+    if ((OffsetVal + WidthVal) >= 32) {
+      SDValue ShiftVal = DAG.getConstant(OffsetVal, MVT::i32);
+      return DAG.getNode(Signed ? ISD::SRA : ISD::SRL, DL, MVT::i32,
+                         BitsFrom, ShiftVal);
+    }
+
+    APInt KnownZero, KnownOne;
+    TargetLowering::TargetLoweringOpt TLO(DAG, !DCI.isBeforeLegalize(),
+                                          !DCI.isBeforeLegalizeOps());
+    const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+    if (TLO.ShrinkDemandedConstant(BitsFrom, Demanded) ||
+        TLI.SimplifyDemandedBits(BitsFrom, Demanded, KnownZero, KnownOne, TLO)) {
+      DCI.CommitTargetLoweringOpt(TLO);
+    }
+
+    break;
+  }
+
+  case ISD::STORE:
+    return performStoreCombine(N, DCI);
+  }
+  return SDValue();
 }
 
 //===----------------------------------------------------------------------===//
@@ -803,17 +2156,17 @@ SDValue AMDGPUTargetLowering::CreateLiveInRegister(SelectionDAG &DAG,
 
 const char* AMDGPUTargetLowering::getTargetNodeName(unsigned Opcode) const {
   switch (Opcode) {
-  default: return 0;
+  default: return nullptr;
   // AMDIL DAG nodes
   NODE_NAME_CASE(CALL);
   NODE_NAME_CASE(UMUL);
-  NODE_NAME_CASE(DIV_INF);
   NODE_NAME_CASE(RET_FLAG);
   NODE_NAME_CASE(BRANCH_COND);
 
   // AMDGPU DAG nodes
   NODE_NAME_CASE(DWORDADDR)
   NODE_NAME_CASE(FRACT)
+  NODE_NAME_CASE(CLAMP)
   NODE_NAME_CASE(FMAX)
   NODE_NAME_CASE(SMAX)
   NODE_NAME_CASE(UMAX)
@@ -821,6 +2174,24 @@ const char* AMDGPUTargetLowering::getTargetNodeName(unsigned Opcode) const {
   NODE_NAME_CASE(SMIN)
   NODE_NAME_CASE(UMIN)
   NODE_NAME_CASE(URECIP)
+  NODE_NAME_CASE(DIV_SCALE)
+  NODE_NAME_CASE(DIV_FMAS)
+  NODE_NAME_CASE(DIV_FIXUP)
+  NODE_NAME_CASE(TRIG_PREOP)
+  NODE_NAME_CASE(RCP)
+  NODE_NAME_CASE(RSQ)
+  NODE_NAME_CASE(RSQ_LEGACY)
+  NODE_NAME_CASE(RSQ_CLAMPED)
+  NODE_NAME_CASE(DOT4)
+  NODE_NAME_CASE(BFE_U32)
+  NODE_NAME_CASE(BFE_I32)
+  NODE_NAME_CASE(BFI)
+  NODE_NAME_CASE(BFM)
+  NODE_NAME_CASE(BREV)
+  NODE_NAME_CASE(MUL_U24)
+  NODE_NAME_CASE(MUL_I24)
+  NODE_NAME_CASE(MAD_U24)
+  NODE_NAME_CASE(MAD_I24)
   NODE_NAME_CASE(EXPORT)
   NODE_NAME_CASE(CONST_ADDRESS)
   NODE_NAME_CASE(REGISTER_LOAD)
@@ -831,7 +2202,124 @@ const char* AMDGPUTargetLowering::getTargetNodeName(unsigned Opcode) const {
   NODE_NAME_CASE(SAMPLEB)
   NODE_NAME_CASE(SAMPLED)
   NODE_NAME_CASE(SAMPLEL)
+  NODE_NAME_CASE(CVT_F32_UBYTE0)
+  NODE_NAME_CASE(CVT_F32_UBYTE1)
+  NODE_NAME_CASE(CVT_F32_UBYTE2)
+  NODE_NAME_CASE(CVT_F32_UBYTE3)
+  NODE_NAME_CASE(BUILD_VERTICAL_VECTOR)
+  NODE_NAME_CASE(CONST_DATA_PTR)
   NODE_NAME_CASE(STORE_MSKOR)
   NODE_NAME_CASE(TBUFFER_STORE_FORMAT)
   }
 }
+
+static void computeKnownBitsForMinMax(const SDValue Op0,
+                                      const SDValue Op1,
+                                      APInt &KnownZero,
+                                      APInt &KnownOne,
+                                      const SelectionDAG &DAG,
+                                      unsigned Depth) {
+  APInt Op0Zero, Op0One;
+  APInt Op1Zero, Op1One;
+  DAG.computeKnownBits(Op0, Op0Zero, Op0One, Depth);
+  DAG.computeKnownBits(Op1, Op1Zero, Op1One, Depth);
+
+  KnownZero = Op0Zero & Op1Zero;
+  KnownOne = Op0One & Op1One;
+}
+
+void AMDGPUTargetLowering::computeKnownBitsForTargetNode(
+  const SDValue Op,
+  APInt &KnownZero,
+  APInt &KnownOne,
+  const SelectionDAG &DAG,
+  unsigned Depth) const {
+
+  KnownZero = KnownOne = APInt(KnownOne.getBitWidth(), 0); // Don't know anything.
+
+  APInt KnownZero2;
+  APInt KnownOne2;
+  unsigned Opc = Op.getOpcode();
+
+  switch (Opc) {
+  default:
+    break;
+  case ISD::INTRINSIC_WO_CHAIN: {
+    // FIXME: The intrinsic should just use the node.
+    switch (cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue()) {
+    case AMDGPUIntrinsic::AMDGPU_imax:
+    case AMDGPUIntrinsic::AMDGPU_umax:
+    case AMDGPUIntrinsic::AMDGPU_imin:
+    case AMDGPUIntrinsic::AMDGPU_umin:
+      computeKnownBitsForMinMax(Op.getOperand(1), Op.getOperand(2),
+                                KnownZero, KnownOne, DAG, Depth);
+      break;
+    default:
+      break;
+    }
+
+    break;
+  }
+  case AMDGPUISD::SMAX:
+  case AMDGPUISD::UMAX:
+  case AMDGPUISD::SMIN:
+  case AMDGPUISD::UMIN:
+    computeKnownBitsForMinMax(Op.getOperand(0), Op.getOperand(1),
+                              KnownZero, KnownOne, DAG, Depth);
+    break;
+
+  case AMDGPUISD::BFE_I32:
+  case AMDGPUISD::BFE_U32: {
+    ConstantSDNode *CWidth = dyn_cast<ConstantSDNode>(Op.getOperand(2));
+    if (!CWidth)
+      return;
+
+    unsigned BitWidth = 32;
+    uint32_t Width = CWidth->getZExtValue() & 0x1f;
+    if (Width == 0) {
+      KnownZero = APInt::getAllOnesValue(BitWidth);
+      KnownOne = APInt::getNullValue(BitWidth);
+      return;
+    }
+
+    // FIXME: This could do a lot more. If offset is 0, should be the same as
+    // sign_extend_inreg implementation, but that involves duplicating it.
+    if (Opc == AMDGPUISD::BFE_I32)
+      KnownOne = APInt::getHighBitsSet(BitWidth, BitWidth - Width);
+    else
+      KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - Width);
+
+    break;
+  }
+  }
+}
+
+unsigned AMDGPUTargetLowering::ComputeNumSignBitsForTargetNode(
+  SDValue Op,
+  const SelectionDAG &DAG,
+  unsigned Depth) const {
+  switch (Op.getOpcode()) {
+  case AMDGPUISD::BFE_I32: {
+    ConstantSDNode *Width = dyn_cast<ConstantSDNode>(Op.getOperand(2));
+    if (!Width)
+      return 1;
+
+    unsigned SignBits = 32 - Width->getZExtValue() + 1;
+    ConstantSDNode *Offset = dyn_cast<ConstantSDNode>(Op.getOperand(1));
+    if (!Offset || !Offset->isNullValue())
+      return SignBits;
+
+    // TODO: Could probably figure something out with non-0 offsets.
+    unsigned Op0SignBits = DAG.ComputeNumSignBits(Op.getOperand(0), Depth + 1);
+    return std::max(SignBits, Op0SignBits);
+  }
+
+  case AMDGPUISD::BFE_U32: {
+    ConstantSDNode *Width = dyn_cast<ConstantSDNode>(Op.getOperand(2));
+    return Width ? 32 - (Width->getZExtValue() & 0x1f) : 1;
+  }
+
+  default:
+    return 1;
+  }
+}
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUISelLowering.h b/contrib/llvm/lib/Target/R600/AMDGPUISelLowering.h
index 2dfd3cf..624d4e0 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUISelLowering.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPUISelLowering.h
@@ -21,13 +21,18 @@
 namespace llvm {
 
 class AMDGPUMachineFunction;
+class AMDGPUSubtarget;
 class MachineRegisterInfo;
 
 class AMDGPUTargetLowering : public TargetLowering {
+protected:
+  const AMDGPUSubtarget *Subtarget;
+
 private:
-  void ExtractVectorElements(SDValue Op, SelectionDAG &DAG,
-                             SmallVectorImpl<SDValue> &Args,
-                             unsigned Start, unsigned Count) const;
+  SDValue LowerConstantInitializer(const Constant* Init, const GlobalValue *GV,
+                                   const SDValue &InitPtr,
+                                   SDValue Chain,
+                                   SelectionDAG &DAG) const;
   SDValue LowerFrameIndex(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerEXTRACT_SUBVECTOR(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const;
@@ -36,25 +41,44 @@ private:
   /// of the same bitwidth.
   SDValue MergeVectorStore(const SDValue &Op, SelectionDAG &DAG) const;
   /// \brief Split a vector store into multiple scalar stores.
-  /// \returns The resulting chain. 
+  /// \returns The resulting chain.
+
+  SDValue LowerSDIV(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSDIV24(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSDIV32(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSDIV64(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSREM(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSREM32(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSREM64(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerUDIVREM(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFCEIL(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFTRUNC(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFRINT(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFNEARBYINT(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFFLOOR(SDValue Op, SelectionDAG &DAG) const;
+
   SDValue LowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
 
+  SDValue ExpandSIGN_EXTEND_INREG(SDValue Op,
+                                  unsigned BitsDiff,
+                                  SelectionDAG &DAG) const;
+  SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const;
+
+  SDValue performStoreCombine(SDNode *N, DAGCombinerInfo &DCI) const;
+  SDValue performMulCombine(SDNode *N, DAGCombinerInfo &DCI) const;
+
 protected:
+  static EVT getEquivalentMemType(LLVMContext &Context, EVT VT);
+  static EVT getEquivalentLoadRegType(LLVMContext &Context, EVT VT);
 
-  /// \brief Helper function that adds Reg to the LiveIn list of the DAG's
-  /// MachineFunction.
-  ///
-  /// \returns a RegisterSDNode representing Reg.
-  virtual SDValue CreateLiveInRegister(SelectionDAG &DAG,
-                                       const TargetRegisterClass *RC,
-                                       unsigned Reg, EVT VT) const;
-  SDValue LowerGlobalAddress(AMDGPUMachineFunction *MFI, SDValue Op,
-                             SelectionDAG &DAG) const;
+  virtual SDValue LowerGlobalAddress(AMDGPUMachineFunction *MFI, SDValue Op,
+                                     SelectionDAG &DAG) const;
   /// \brief Split a vector load into multiple scalar loads.
   SDValue SplitVectorLoad(const SDValue &Op, SelectionDAG &DAG) const;
   SDValue SplitVectorStore(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSDIVREM(SDValue Op, SelectionDAG &DAG) const;
   bool isHWTrueValue(SDValue Op) const;
   bool isHWFalseValue(SDValue Op) const;
 
@@ -74,67 +98,69 @@ protected:
 public:
   AMDGPUTargetLowering(TargetMachine &TM);
 
-  virtual bool isFAbsFree(EVT VT) const;
-  virtual bool isFNegFree(EVT VT) const;
-  virtual MVT getVectorIdxTy() const;
-  virtual bool isLoadBitCastBeneficial(EVT, EVT) const LLVM_OVERRIDE;
-  virtual SDValue LowerReturn(SDValue Chain, CallingConv::ID CallConv,
-                              bool isVarArg,
-                              const SmallVectorImpl<ISD::OutputArg> &Outs,
-                              const SmallVectorImpl<SDValue> &OutVals,
-                              SDLoc DL, SelectionDAG &DAG) const;
-  virtual SDValue LowerCall(CallLoweringInfo &CLI,
-                            SmallVectorImpl<SDValue> &InVals) const {
-    CLI.Callee.dump();
-    llvm_unreachable("Undefined function");
-  }
+  bool isFAbsFree(EVT VT) const override;
+  bool isFNegFree(EVT VT) const override;
+  bool isTruncateFree(EVT Src, EVT Dest) const override;
+  bool isTruncateFree(Type *Src, Type *Dest) const override;
+
+  bool isZExtFree(Type *Src, Type *Dest) const override;
+  bool isZExtFree(EVT Src, EVT Dest) const override;
+  bool isZExtFree(SDValue Val, EVT VT2) const override;
+
+  bool isNarrowingProfitable(EVT VT1, EVT VT2) const override;
+
+  MVT getVectorIdxTy() const override;
+  bool isSelectSupported(SelectSupportKind) const override;
+
+  bool isFPImmLegal(const APFloat &Imm, EVT VT) const override;
+  bool ShouldShrinkFPConstant(EVT VT) const override;
+
+  bool isLoadBitCastBeneficial(EVT, EVT) const override;
+  SDValue LowerReturn(SDValue Chain, CallingConv::ID CallConv,
+                      bool isVarArg,
+                      const SmallVectorImpl<ISD::OutputArg> &Outs,
+                      const SmallVectorImpl<SDValue> &OutVals,
+                      SDLoc DL, SelectionDAG &DAG) const override;
+  SDValue LowerCall(CallLoweringInfo &CLI,
+                    SmallVectorImpl<SDValue> &InVals) const override;
+
+  SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
+  SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
+  void ReplaceNodeResults(SDNode * N,
+                          SmallVectorImpl<SDValue> &Results,
+                          SelectionDAG &DAG) const override;
 
-  virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerIntrinsicIABS(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerIntrinsicLRP(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerMinMax(SDValue Op, SelectionDAG &DAG) const;
-  virtual const char* getTargetNodeName(unsigned Opcode) const;
+  SDValue CombineMinMax(SDNode *N, SelectionDAG &DAG) const;
+  const char* getTargetNodeName(unsigned Opcode) const override;
 
-  virtual SDNode *PostISelFolding(MachineSDNode *N, SelectionDAG &DAG) const {
+  virtual SDNode *PostISelFolding(MachineSDNode *N,
+                                  SelectionDAG &DAG) const {
     return N;
   }
 
-// Functions defined in AMDILISelLowering.cpp
-public:
-
   /// \brief Determine which of the bits specified in \p Mask are known to be
   /// either zero or one and return them in the \p KnownZero and \p KnownOne
   /// bitsets.
-  virtual void computeMaskedBitsForTargetNode(const SDValue Op,
-                                              APInt &KnownZero,
-                                              APInt &KnownOne,
-                                              const SelectionDAG &DAG,
-                                              unsigned Depth = 0) const;
-
-  virtual bool getTgtMemIntrinsic(IntrinsicInfo &Info,
-                                  const CallInst &I, unsigned Intrinsic) const;
-
-  /// We want to mark f32/f64 floating point values as legal.
-  bool isFPImmLegal(const APFloat &Imm, EVT VT) const;
+  void computeKnownBitsForTargetNode(const SDValue Op,
+                                     APInt &KnownZero,
+                                     APInt &KnownOne,
+                                     const SelectionDAG &DAG,
+                                     unsigned Depth = 0) const override;
 
-  /// We don't want to shrink f64/f32 constants.
-  bool ShouldShrinkFPConstant(EVT VT) const;
+  virtual unsigned ComputeNumSignBitsForTargetNode(
+    SDValue Op,
+    const SelectionDAG &DAG,
+    unsigned Depth = 0) const override;
 
-private:
-  void InitAMDILLowering();
-  SDValue LowerSREM(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSREM8(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSREM16(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSREM32(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSREM64(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSDIV(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSDIV24(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSDIV32(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSDIV64(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const;
-  EVT genIntType(uint32_t size = 32, uint32_t numEle = 1) const;
-  SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const;
+  /// \brief Helper function that adds Reg to the LiveIn list of the DAG's
+  /// MachineFunction.
+  ///
+  /// \returns a RegisterSDNode representing Reg.
+  virtual SDValue CreateLiveInRegister(SelectionDAG &DAG,
+                                       const TargetRegisterClass *RC,
+                                       unsigned Reg, EVT VT) const;
 };
 
 namespace AMDGPUISD {
@@ -144,12 +170,15 @@ enum {
   FIRST_NUMBER = ISD::BUILTIN_OP_END,
   CALL,        // Function call based on a single integer
   UMUL,        // 32bit unsigned multiplication
-  DIV_INF,      // Divide with infinity returned on zero divisor
   RET_FLAG,
   BRANCH_COND,
   // End AMDIL ISD Opcodes
   DWORDADDR,
   FRACT,
+  CLAMP,
+
+  // SIN_HW, COS_HW - f32 for SI, 1 ULP max error, valid from -100 pi to 100 pi.
+  // Denormals handled on some parts.
   COS_HW,
   SIN_HW,
   FMAX,
@@ -159,7 +188,27 @@ enum {
   SMIN,
   UMIN,
   URECIP,
+  DIV_SCALE,
+  DIV_FMAS,
+  DIV_FIXUP,
+  TRIG_PREOP, // 1 ULP max error for f64
+
+  // RCP, RSQ - For f32, 1 ULP max error, no denormal handling.
+  //            For f64, max error 2^29 ULP, handles denormals.
+  RCP,
+  RSQ,
+  RSQ_LEGACY,
+  RSQ_CLAMPED,
   DOT4,
+  BFE_U32, // Extract range of bits with zero extension to 32-bits.
+  BFE_I32, // Extract range of bits with sign extension to 32-bits.
+  BFI, // (src0 & src1) | (~src0 & src2)
+  BFM, // Insert a range of bits into a 32-bit word.
+  BREV, // Reverse bits.
+  MUL_U24,
+  MUL_I24,
+  MAD_U24,
+  MAD_I24,
   TEXTURE_FETCH,
   EXPORT,
   CONST_ADDRESS,
@@ -170,6 +219,23 @@ enum {
   SAMPLEB,
   SAMPLED,
   SAMPLEL,
+
+  // These cvt_f32_ubyte* nodes need to remain consecutive and in order.
+  CVT_F32_UBYTE0,
+  CVT_F32_UBYTE1,
+  CVT_F32_UBYTE2,
+  CVT_F32_UBYTE3,
+  /// This node is for VLIW targets and it is used to represent a vector
+  /// that is stored in consecutive registers with the same channel.
+  /// For example:
+  ///   |X  |Y|Z|W|
+  /// T0|v.x| | | |
+  /// T1|v.y| | | |
+  /// T2|v.z| | | |
+  /// T3|v.w| | | |
+  BUILD_VERTICAL_VECTOR,
+  /// Pointer to the start of the shader's constant data.
+  CONST_DATA_PTR,
   FIRST_MEM_OPCODE_NUMBER = ISD::FIRST_TARGET_MEMORY_OPCODE,
   STORE_MSKOR,
   LOAD_CONSTANT,
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.cpp b/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.cpp
index 4f7084b..fef5b8c 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.cpp
@@ -20,19 +20,18 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 
+using namespace llvm;
+
 #define GET_INSTRINFO_CTOR_DTOR
 #define GET_INSTRINFO_NAMED_OPS
 #define GET_INSTRMAP_INFO
 #include "AMDGPUGenInstrInfo.inc"
 
-using namespace llvm;
-
-
 // Pin the vtable to this file.
 void AMDGPUInstrInfo::anchor() {}
 
-AMDGPUInstrInfo::AMDGPUInstrInfo(TargetMachine &tm)
-  : AMDGPUGenInstrInfo(-1,-1), RI(tm), TM(tm) { }
+AMDGPUInstrInfo::AMDGPUInstrInfo(const AMDGPUSubtarget &st)
+  : AMDGPUGenInstrInfo(-1,-1), RI(st), ST(st) { }
 
 const AMDGPURegisterInfo &AMDGPUInstrInfo::getRegisterInfo() const {
   return RI;
@@ -85,7 +84,7 @@ AMDGPUInstrInfo::convertToThreeAddress(MachineFunction::iterator &MFI,
                                       MachineBasicBlock::iterator &MBBI,
                                       LiveVariables *LV) const {
 // TODO: Implement this function
-  return NULL;
+  return nullptr;
 }
 bool AMDGPUInstrInfo::getNextBranchInstr(MachineBasicBlock::iterator &iter,
                                         MachineBasicBlock &MBB) const {
@@ -110,7 +109,7 @@ AMDGPUInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
                                     int FrameIndex,
                                     const TargetRegisterClass *RC,
                                     const TargetRegisterInfo *TRI) const {
-  assert(!"Not Implemented");
+  llvm_unreachable("Not Implemented");
 }
 
 void
@@ -119,22 +118,21 @@ AMDGPUInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
                                      unsigned DestReg, int FrameIndex,
                                      const TargetRegisterClass *RC,
                                      const TargetRegisterInfo *TRI) const {
-  assert(!"Not Implemented");
+  llvm_unreachable("Not Implemented");
 }
 
 bool AMDGPUInstrInfo::expandPostRAPseudo (MachineBasicBlock::iterator MI) const {
   MachineBasicBlock *MBB = MI->getParent();
-   int OffsetOpIdx =
-       AMDGPU::getNamedOperandIdx(MI->getOpcode(), AMDGPU::OpName::addr);
+  int OffsetOpIdx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
+                                               AMDGPU::OpName::addr);
    // addr is a custom operand with multiple MI operands, and only the
    // first MI operand is given a name.
   int RegOpIdx = OffsetOpIdx + 1;
-  int ChanOpIdx =
-      AMDGPU::getNamedOperandIdx(MI->getOpcode(), AMDGPU::OpName::chan);
-
+  int ChanOpIdx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
+                                             AMDGPU::OpName::chan);
   if (isRegisterLoad(*MI)) {
-    int DstOpIdx =
-        AMDGPU::getNamedOperandIdx(MI->getOpcode(), AMDGPU::OpName::dst);
+    int DstOpIdx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
+                                              AMDGPU::OpName::dst);
     unsigned RegIndex = MI->getOperand(RegOpIdx).getImm();
     unsigned Channel = MI->getOperand(ChanOpIdx).getImm();
     unsigned Address = calculateIndirectAddress(RegIndex, Channel);
@@ -147,8 +145,8 @@ bool AMDGPUInstrInfo::expandPostRAPseudo (MachineBasicBlock::iterator MI) const
                         Address, OffsetReg);
     }
   } else if (isRegisterStore(*MI)) {
-    int ValOpIdx =
-        AMDGPU::getNamedOperandIdx(MI->getOpcode(), AMDGPU::OpName::val);
+    int ValOpIdx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
+                                              AMDGPU::OpName::val);
     AMDGPU::getNamedOperandIdx(MI->getOpcode(), AMDGPU::OpName::dst);
     unsigned RegIndex = MI->getOperand(RegOpIdx).getImm();
     unsigned Channel = MI->getOperand(ChanOpIdx).getImm();
@@ -177,7 +175,7 @@ AMDGPUInstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
                                       const SmallVectorImpl<unsigned> &Ops,
                                       int FrameIndex) const {
 // TODO: Implement this function
-  return 0;
+  return nullptr;
 }
 MachineInstr*
 AMDGPUInstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
@@ -185,7 +183,7 @@ AMDGPUInstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
                                       const SmallVectorImpl<unsigned> &Ops,
                                       MachineInstr *LoadMI) const {
   // TODO: Implement this function
-  return 0;
+  return nullptr;
 }
 bool
 AMDGPUInstrInfo::canFoldMemoryOperand(const MachineInstr *MI,
@@ -322,33 +320,11 @@ int AMDGPUInstrInfo::getIndirectIndexEnd(const MachineFunction &MF) const {
     return -1;
   }
 
-  Offset = TM.getFrameLowering()->getFrameIndexOffset(MF, -1);
+  Offset = MF.getTarget().getFrameLowering()->getFrameIndexOffset(MF, -1);
 
   return getIndirectIndexBegin(MF) + Offset;
 }
 
-
-void AMDGPUInstrInfo::convertToISA(MachineInstr & MI, MachineFunction &MF,
-    DebugLoc DL) const {
-  MachineRegisterInfo &MRI = MF.getRegInfo();
-  const AMDGPURegisterInfo & RI = getRegisterInfo();
-
-  for (unsigned i = 0; i < MI.getNumOperands(); i++) {
-    MachineOperand &MO = MI.getOperand(i);
-    // Convert dst regclass to one that is supported by the ISA
-    if (MO.isReg() && MO.isDef()) {
-      if (TargetRegisterInfo::isVirtualRegister(MO.getReg())) {
-        const TargetRegisterClass * oldRegClass = MRI.getRegClass(MO.getReg());
-        const TargetRegisterClass * newRegClass = RI.getISARegClass(oldRegClass);
-
-        assert(newRegClass);
-
-        MRI.setRegClass(MO.getReg(), newRegClass);
-      }
-    }
-  }
-}
-
 int AMDGPUInstrInfo::getMaskedMIMGOp(uint16_t Opcode, unsigned Channels) const {
   switch (Channels) {
   default: return Opcode;
@@ -357,3 +333,14 @@ int AMDGPUInstrInfo::getMaskedMIMGOp(uint16_t Opcode, unsigned Channels) const {
   case 3: return AMDGPU::getMaskedMIMGOp(Opcode, AMDGPU::Channels_3);
   }
 }
+
+// Wrapper for Tablegen'd function.  enum Subtarget is not defined in any
+// header files, so we need to wrap it in a function that takes unsigned 
+// instead.
+namespace llvm {
+namespace AMDGPU {
+int getMCOpcode(uint16_t Opcode, unsigned Gen) {
+  return getMCOpcode(Opcode);
+}
+}
+}
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.h b/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.h
index ce5b58c..d5041f5 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.h
@@ -33,7 +33,7 @@
 
 namespace llvm {
 
-class AMDGPUTargetMachine;
+class AMDGPUSubtarget;
 class MachineFunction;
 class MachineInstr;
 class MachineInstrBuilder;
@@ -45,21 +45,22 @@ private:
                           MachineBasicBlock &MBB) const;
   virtual void anchor();
 protected:
-  TargetMachine &TM;
+  const AMDGPUSubtarget &ST;
 public:
-  explicit AMDGPUInstrInfo(TargetMachine &tm);
+  explicit AMDGPUInstrInfo(const AMDGPUSubtarget &st);
 
   virtual const AMDGPURegisterInfo &getRegisterInfo() const = 0;
 
   bool isCoalescableExtInstr(const MachineInstr &MI, unsigned &SrcReg,
-                             unsigned &DstReg, unsigned &SubIdx) const;
+                             unsigned &DstReg, unsigned &SubIdx) const override;
 
-  unsigned isLoadFromStackSlot(const MachineInstr *MI, int &FrameIndex) const;
+  unsigned isLoadFromStackSlot(const MachineInstr *MI,
+                               int &FrameIndex) const override;
   unsigned isLoadFromStackSlotPostFE(const MachineInstr *MI,
-                                     int &FrameIndex) const;
+                                     int &FrameIndex) const override;
   bool hasLoadFromStackSlot(const MachineInstr *MI,
                             const MachineMemOperand *&MMO,
-                            int &FrameIndex) const;
+                            int &FrameIndex) const override;
   unsigned isStoreFromStackSlot(const MachineInstr *MI, int &FrameIndex) const;
   unsigned isStoreFromStackSlotPostFE(const MachineInstr *MI,
                                       int &FrameIndex) const;
@@ -70,7 +71,7 @@ public:
   MachineInstr *
   convertToThreeAddress(MachineFunction::iterator &MFI,
                         MachineBasicBlock::iterator &MBBI,
-                        LiveVariables *LV) const;
+                        LiveVariables *LV) const override;
 
 
   virtual void copyPhysReg(MachineBasicBlock &MBB,
@@ -78,71 +79,64 @@ public:
                            unsigned DestReg, unsigned SrcReg,
                            bool KillSrc) const = 0;
 
+  bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const override;
+
   void storeRegToStackSlot(MachineBasicBlock &MBB,
                            MachineBasicBlock::iterator MI,
                            unsigned SrcReg, bool isKill, int FrameIndex,
                            const TargetRegisterClass *RC,
-                           const TargetRegisterInfo *TRI) const;
+                           const TargetRegisterInfo *TRI) const override;
   void loadRegFromStackSlot(MachineBasicBlock &MBB,
                             MachineBasicBlock::iterator MI,
                             unsigned DestReg, int FrameIndex,
                             const TargetRegisterClass *RC,
-                            const TargetRegisterInfo *TRI) const;
-  virtual bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const;
-
+                            const TargetRegisterInfo *TRI) const override;
 
 protected:
   MachineInstr *foldMemoryOperandImpl(MachineFunction &MF,
                                       MachineInstr *MI,
                                       const SmallVectorImpl<unsigned> &Ops,
-                                      int FrameIndex) const;
+                                      int FrameIndex) const override;
   MachineInstr *foldMemoryOperandImpl(MachineFunction &MF,
                                       MachineInstr *MI,
                                       const SmallVectorImpl<unsigned> &Ops,
-                                      MachineInstr *LoadMI) const;
+                                      MachineInstr *LoadMI) const override;
   /// \returns the smallest register index that will be accessed by an indirect
   /// read or write or -1 if indirect addressing is not used by this program.
-  virtual int getIndirectIndexBegin(const MachineFunction &MF) const;
+  int getIndirectIndexBegin(const MachineFunction &MF) const;
 
   /// \returns the largest register index that will be accessed by an indirect
   /// read or write or -1 if indirect addressing is not used by this program.
-  virtual int getIndirectIndexEnd(const MachineFunction &MF) const;
+  int getIndirectIndexEnd(const MachineFunction &MF) const;
 
 public:
   bool canFoldMemoryOperand(const MachineInstr *MI,
-                            const SmallVectorImpl<unsigned> &Ops) const;
+                           const SmallVectorImpl<unsigned> &Ops) const override;
   bool unfoldMemoryOperand(MachineFunction &MF, MachineInstr *MI,
-                           unsigned Reg, bool UnfoldLoad, bool UnfoldStore,
-                           SmallVectorImpl<MachineInstr *> &NewMIs) const;
+                        unsigned Reg, bool UnfoldLoad, bool UnfoldStore,
+                        SmallVectorImpl<MachineInstr *> &NewMIs) const override;
   bool unfoldMemoryOperand(SelectionDAG &DAG, SDNode *N,
-                           SmallVectorImpl<SDNode *> &NewNodes) const;
+                           SmallVectorImpl<SDNode *> &NewNodes) const override;
   unsigned getOpcodeAfterMemoryUnfold(unsigned Opc,
-                                      bool UnfoldLoad, bool UnfoldStore,
-                                      unsigned *LoadRegIndex = 0) const;
+                               bool UnfoldLoad, bool UnfoldStore,
+                               unsigned *LoadRegIndex = nullptr) const override;
   bool shouldScheduleLoadsNear(SDNode *Load1, SDNode *Load2,
                                int64_t Offset1, int64_t Offset2,
-                               unsigned NumLoads) const;
+                               unsigned NumLoads) const override;
 
-  bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const;
+  bool
+  ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const override;
   void insertNoop(MachineBasicBlock &MBB,
-                  MachineBasicBlock::iterator MI) const;
-  bool isPredicated(const MachineInstr *MI) const;
+                  MachineBasicBlock::iterator MI) const override;
+  bool isPredicated(const MachineInstr *MI) const override;
   bool SubsumesPredicate(const SmallVectorImpl<MachineOperand> &Pred1,
-                         const SmallVectorImpl<MachineOperand> &Pred2) const;
+                   const SmallVectorImpl<MachineOperand> &Pred2) const override;
   bool DefinesPredicate(MachineInstr *MI,
-                        std::vector<MachineOperand> &Pred) const;
-  bool isPredicable(MachineInstr *MI) const;
-  bool isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const;
+                        std::vector<MachineOperand> &Pred) const override;
+  bool isPredicable(MachineInstr *MI) const override;
+  bool isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const override;
 
   // Helper functions that check the opcode for status information
-  bool isLoadInst(llvm::MachineInstr *MI) const;
-  bool isExtLoadInst(llvm::MachineInstr *MI) const;
-  bool isSWSExtLoadInst(llvm::MachineInstr *MI) const;
-  bool isSExtLoadInst(llvm::MachineInstr *MI) const;
-  bool isZExtLoadInst(llvm::MachineInstr *MI) const;
-  bool isAExtLoadInst(llvm::MachineInstr *MI) const;
-  bool isStoreInst(llvm::MachineInstr *MI) const;
-  bool isTruncStoreInst(llvm::MachineInstr *MI) const;
   bool isRegisterStore(const MachineInstr &MI) const;
   bool isRegisterLoad(const MachineInstr &MI) const;
 
@@ -150,7 +144,6 @@ public:
 // Pure virtual funtions to be implemented by sub-classes.
 //===---------------------------------------------------------------------===//
 
-  virtual unsigned getIEQOpcode() const = 0;
   virtual bool isMov(unsigned opcode) const = 0;
 
   /// \brief Calculate the "Indirect Address" for the given \p RegIndex and
@@ -183,12 +176,6 @@ public:
                                     unsigned ValueReg, unsigned Address,
                                     unsigned OffsetReg) const = 0;
 
-
-  /// \brief Convert the AMDIL MachineInstr to a supported ISA
-  /// MachineInstr
-  virtual void convertToISA(MachineInstr & MI, MachineFunction &MF,
-    DebugLoc DL) const;
-
   /// \brief Build a MOV instruction.
   virtual MachineInstr *buildMovInstr(MachineBasicBlock *MBB,
                                       MachineBasicBlock::iterator I,
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.td b/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.td
index fccede0..820f1a8 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.td
+++ b/contrib/llvm/lib/Target/R600/AMDGPUInstrInfo.td
@@ -19,6 +19,14 @@ def AMDGPUDTIntTernaryOp : SDTypeProfile<1, 3, [
   SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisInt<0>, SDTCisInt<3>
 ]>;
 
+def AMDGPUTrigPreOp : SDTypeProfile<1, 2,
+  [SDTCisSameAs<0, 1>, SDTCisFP<0>, SDTCisInt<2>]
+>;
+
+def AMDGPUDivScaleOp : SDTypeProfile<2, 3,
+  [SDTCisFP<0>, SDTCisInt<1>, SDTCisSameAs<0, 2>, SDTCisSameAs<0, 3>, SDTCisSameAs<0, 4>]
+>;
+
 //===----------------------------------------------------------------------===//
 // AMDGPU DAG Nodes
 //
@@ -26,14 +34,31 @@ def AMDGPUDTIntTernaryOp : SDTypeProfile<1, 3, [
 // This argument to this node is a dword address.
 def AMDGPUdwordaddr : SDNode<"AMDGPUISD::DWORDADDR", SDTIntUnaryOp>;
 
+def AMDGPUcos : SDNode<"AMDGPUISD::COS_HW", SDTFPUnaryOp>;
+def AMDGPUsin : SDNode<"AMDGPUISD::SIN_HW", SDTFPUnaryOp>;
+
 // out = a - floor(a)
 def AMDGPUfract : SDNode<"AMDGPUISD::FRACT", SDTFPUnaryOp>;
 
+// out = 1.0 / a
+def AMDGPUrcp : SDNode<"AMDGPUISD::RCP", SDTFPUnaryOp>;
+
+// out = 1.0 / sqrt(a)
+def AMDGPUrsq : SDNode<"AMDGPUISD::RSQ", SDTFPUnaryOp>;
+
+// out = 1.0 / sqrt(a)
+def AMDGPUrsq_legacy : SDNode<"AMDGPUISD::RSQ_LEGACY", SDTFPUnaryOp>;
+
+// out = 1.0 / sqrt(a) result clamped to +/- max_float.
+def AMDGPUrsq_clamped : SDNode<"AMDGPUISD::RSQ_CLAMPED", SDTFPUnaryOp>;
+
 // out = max(a, b) a and b are floats
 def AMDGPUfmax : SDNode<"AMDGPUISD::FMAX", SDTFPBinOp,
   [SDNPCommutative, SDNPAssociative]
 >;
 
+def AMDGPUclamp : SDNode<"AMDGPUISD::CLAMP", SDTFPTernaryOp, []>;
+
 // out = max(a, b) a and b are signed ints
 def AMDGPUsmax : SDNode<"AMDGPUISD::SMAX", SDTIntBinOp,
   [SDNPCommutative, SDNPAssociative]
@@ -59,12 +84,38 @@ def AMDGPUumin : SDNode<"AMDGPUISD::UMIN", SDTIntBinOp,
   [SDNPCommutative, SDNPAssociative]
 >;
 
+
+def AMDGPUcvt_f32_ubyte0 : SDNode<"AMDGPUISD::CVT_F32_UBYTE0",
+  SDTIntToFPOp, []>;
+def AMDGPUcvt_f32_ubyte1 : SDNode<"AMDGPUISD::CVT_F32_UBYTE1",
+  SDTIntToFPOp, []>;
+def AMDGPUcvt_f32_ubyte2 : SDNode<"AMDGPUISD::CVT_F32_UBYTE2",
+  SDTIntToFPOp, []>;
+def AMDGPUcvt_f32_ubyte3 : SDNode<"AMDGPUISD::CVT_F32_UBYTE3",
+  SDTIntToFPOp, []>;
+
+
 // urecip - This operation is a helper for integer division, it returns the
 // result of 1 / a as a fractional unsigned integer.
 // out = (2^32 / a) + e
 // e is rounding error
 def AMDGPUurecip : SDNode<"AMDGPUISD::URECIP", SDTIntUnaryOp>;
 
+// Special case divide preop and flags.
+def AMDGPUdiv_scale : SDNode<"AMDGPUISD::DIV_SCALE", AMDGPUDivScaleOp>;
+
+//  Special case divide FMA with scale and flags (src0 = Quotient,
+//  src1 = Denominator, src2 = Numerator).
+def AMDGPUdiv_fmas : SDNode<"AMDGPUISD::DIV_FMAS", SDTFPTernaryOp>;
+
+// Single or double precision division fixup.
+// Special case divide fixup and flags(src0 = Quotient, src1 =
+// Denominator, src2 = Numerator).
+def AMDGPUdiv_fixup : SDNode<"AMDGPUISD::DIV_FIXUP", SDTFPTernaryOp>;
+
+// Look Up 2.0 / pi src0 with segment select src1[4:0]
+def AMDGPUtrig_preop : SDNode<"AMDGPUISD::TRIG_PREOP", AMDGPUTrigPreOp>;
+
 def AMDGPUregister_load : SDNode<"AMDGPUISD::REGISTER_LOAD",
                           SDTypeProfile<1, 2, [SDTCisPtrTy<1>, SDTCisInt<2>]>,
                           [SDNPHasChain, SDNPMayLoad]>;
@@ -86,3 +137,45 @@ def AMDGPUstore_mskor : SDNode<"AMDGPUISD::STORE_MSKOR",
 
 def AMDGPUround : SDNode<"ISD::FROUND",
                          SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>>;
+
+def AMDGPUbfe_u32 : SDNode<"AMDGPUISD::BFE_U32", AMDGPUDTIntTernaryOp>;
+def AMDGPUbfe_i32 : SDNode<"AMDGPUISD::BFE_I32", AMDGPUDTIntTernaryOp>;
+def AMDGPUbfi : SDNode<"AMDGPUISD::BFI", AMDGPUDTIntTernaryOp>;
+def AMDGPUbfm : SDNode<"AMDGPUISD::BFM", SDTIntBinOp>;
+
+def AMDGPUbrev : SDNode<"AMDGPUISD::BREV", SDTIntUnaryOp>;
+
+// Signed and unsigned 24-bit mulitply.  The highest 8-bits are ignore when
+// performing the mulitply.  The result is a 32-bit value.
+def AMDGPUmul_u24 : SDNode<"AMDGPUISD::MUL_U24", SDTIntBinOp,
+  [SDNPCommutative]
+>;
+def AMDGPUmul_i24 : SDNode<"AMDGPUISD::MUL_I24", SDTIntBinOp,
+  [SDNPCommutative]
+>;
+
+def AMDGPUmad_u24 : SDNode<"AMDGPUISD::MAD_U24", AMDGPUDTIntTernaryOp,
+  []
+>;
+def AMDGPUmad_i24 : SDNode<"AMDGPUISD::MAD_I24", AMDGPUDTIntTernaryOp,
+  []
+>;
+
+//===----------------------------------------------------------------------===//
+// Flow Control Profile Types
+//===----------------------------------------------------------------------===//
+// Branch instruction where second and third are basic blocks
+def SDTIL_BRCond : SDTypeProfile<0, 2, [
+    SDTCisVT<0, OtherVT>
+    ]>;
+
+//===----------------------------------------------------------------------===//
+// Flow Control DAG Nodes
+//===----------------------------------------------------------------------===//
+def IL_brcond      : SDNode<"AMDGPUISD::BRANCH_COND", SDTIL_BRCond, [SDNPHasChain]>;
+
+//===----------------------------------------------------------------------===//
+// Call/Return DAG Nodes
+//===----------------------------------------------------------------------===//
+def IL_retflag       : SDNode<"AMDGPUISD::RET_FLAG", SDTNone,
+    [SDNPHasChain, SDNPOptInGlue]>;
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUInstructions.td b/contrib/llvm/lib/Target/R600/AMDGPUInstructions.td
index 7acd673..cd35603 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUInstructions.td
+++ b/contrib/llvm/lib/Target/R600/AMDGPUInstructions.td
@@ -34,9 +34,34 @@ class AMDGPUShaderInst <dag outs, dag ins, string asm, list<dag> pattern>
 
 }
 
+def FP32Denormals : Predicate<"Subtarget.hasFP32Denormals()">;
+def FP64Denormals : Predicate<"Subtarget.hasFP64Denormals()">;
+def UnsafeFPMath : Predicate<"TM.Options.UnsafeFPMath">;
+
 def InstFlag : OperandWithDefaultOps <i32, (ops (i32 0))>;
 def ADDRIndirect : ComplexPattern<iPTR, 2, "SelectADDRIndirect", [], []>;
 
+let OperandType = "OPERAND_IMMEDIATE" in {
+
+def u32imm : Operand<i32> {
+  let PrintMethod = "printU32ImmOperand";
+}
+
+def u16imm : Operand<i16> {
+  let PrintMethod = "printU16ImmOperand";
+}
+
+def u8imm : Operand<i8> {
+  let PrintMethod = "printU8ImmOperand";
+}
+
+} // End OperandType = "OPERAND_IMMEDIATE"
+
+//===--------------------------------------------------------------------===//
+// Custom Operands
+//===--------------------------------------------------------------------===//
+def brtarget   : Operand<OtherVT>;
+
 //===----------------------------------------------------------------------===//
 // PatLeafs for floating-point comparisons
 //===----------------------------------------------------------------------===//
@@ -115,6 +140,43 @@ def COND_NULL : PatLeaf <
 // Load/Store Pattern Fragments
 //===----------------------------------------------------------------------===//
 
+class PrivateMemOp <dag ops, dag frag> : PatFrag <ops, frag, [{
+  return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUAS::PRIVATE_ADDRESS;
+}]>;
+
+class PrivateLoad <SDPatternOperator op> : PrivateMemOp <
+  (ops node:$ptr), (op node:$ptr)
+>;
+
+class PrivateStore <SDPatternOperator op> : PrivateMemOp <
+  (ops node:$value, node:$ptr), (op node:$value, node:$ptr)
+>;
+
+def extloadi8_private : PrivateLoad <extloadi8>;
+def sextloadi8_private : PrivateLoad <sextloadi8>;
+def extloadi16_private : PrivateLoad <extloadi16>;
+def sextloadi16_private : PrivateLoad <sextloadi16>;
+def load_private : PrivateLoad <load>;
+
+def truncstorei8_private : PrivateStore <truncstorei8>;
+def truncstorei16_private : PrivateStore <truncstorei16>;
+def store_private : PrivateStore <store>;
+
+def global_store : PatFrag<(ops node:$val, node:$ptr),
+    (store node:$val, node:$ptr), [{
+        return isGlobalStore(dyn_cast<StoreSDNode>(N));
+}]>;
+
+// Global address space loads
+def global_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+    return isGlobalLoad(dyn_cast<LoadSDNode>(N));
+}]>;
+
+// Constant address space loads
+def constant_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+    return isConstantLoad(dyn_cast<LoadSDNode>(N), -1);
+}]>;
+
 def az_extload : PatFrag<(ops node:$ptr), (unindexedload node:$ptr), [{
   LoadSDNode *L = cast<LoadSDNode>(N);
   return L->getExtensionType() == ISD::ZEXTLOAD ||
@@ -220,26 +282,55 @@ def local_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
     return isLocalLoad(dyn_cast<LoadSDNode>(N));
 }]>;
 
-def atomic_load_add_local : PatFrag<(ops node:$ptr, node:$value),
-                                    (atomic_load_add node:$ptr, node:$value), [{
-  return dyn_cast<MemSDNode>(N)->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS;
-}]>;
 
-def atomic_load_sub_local : PatFrag<(ops node:$ptr, node:$value),
-                                    (atomic_load_sub node:$ptr, node:$value), [{
-  return dyn_cast<MemSDNode>(N)->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS;
+class local_binary_atomic_op<SDNode atomic_op> :
+  PatFrag<(ops node:$ptr, node:$value),
+    (atomic_op node:$ptr, node:$value), [{
+  return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS;
 }]>;
 
+
+def atomic_swap_local : local_binary_atomic_op<atomic_swap>;
+def atomic_load_add_local : local_binary_atomic_op<atomic_load_add>;
+def atomic_load_sub_local : local_binary_atomic_op<atomic_load_sub>;
+def atomic_load_and_local : local_binary_atomic_op<atomic_load_and>;
+def atomic_load_or_local : local_binary_atomic_op<atomic_load_or>;
+def atomic_load_xor_local : local_binary_atomic_op<atomic_load_xor>;
+def atomic_load_nand_local : local_binary_atomic_op<atomic_load_nand>;
+def atomic_load_min_local : local_binary_atomic_op<atomic_load_min>;
+def atomic_load_max_local : local_binary_atomic_op<atomic_load_max>;
+def atomic_load_umin_local : local_binary_atomic_op<atomic_load_umin>;
+def atomic_load_umax_local : local_binary_atomic_op<atomic_load_umax>;
+
 def mskor_global : PatFrag<(ops node:$val, node:$ptr),
                             (AMDGPUstore_mskor node:$val, node:$ptr), [{
   return dyn_cast<MemSDNode>(N)->getAddressSpace() == AMDGPUAS::GLOBAL_ADDRESS;
 }]>;
 
+def atomic_cmp_swap_32_local :
+  PatFrag<(ops node:$ptr, node:$cmp, node:$swap),
+          (atomic_cmp_swap node:$ptr, node:$cmp, node:$swap), [{
+  AtomicSDNode *AN = cast<AtomicSDNode>(N);
+  return AN->getMemoryVT() == MVT::i32 &&
+         AN->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS;
+}]>;
+
+def atomic_cmp_swap_64_local :
+  PatFrag<(ops node:$ptr, node:$cmp, node:$swap),
+          (atomic_cmp_swap node:$ptr, node:$cmp, node:$swap), [{
+  AtomicSDNode *AN = cast<AtomicSDNode>(N);
+  return AN->getMemoryVT() == MVT::i64 &&
+         AN->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS;
+}]>;
+
+
 class Constants {
 int TWO_PI = 0x40c90fdb;
 int PI = 0x40490fdb;
 int TWO_PI_INV = 0x3e22f983;
 int FP_UINT_MAX_PLUS_1 = 0x4f800000;    // 1 << 32 in floating point encoding
+int FP32_NEG_ONE = 0xbf800000;
+int FP32_ONE = 0x3f800000;
 }
 def CONST : Constants;
 
@@ -253,9 +344,6 @@ def FP_ONE : PatLeaf <
   [{return N->isExactlyValue(1.0);}]
 >;
 
-def U24 : ComplexPattern<i32, 1, "SelectU24", [], []>;
-def I24 : ComplexPattern<i32, 1, "SelectI24", [], []>;
-
 let isCodeGenOnly = 1, isPseudo = 1 in {
 
 let usesCustomInserter = 1  in {
@@ -264,7 +352,7 @@ class CLAMP <RegisterClass rc> : AMDGPUShaderInst <
   (outs rc:$dst),
   (ins rc:$src0),
   "CLAMP $dst, $src0",
-  [(set f32:$dst, (int_AMDIL_clamp f32:$src0, (f32 FP_ZERO), (f32 FP_ONE)))]
+  [(set f32:$dst, (AMDGPUclamp f32:$src0, (f32 FP_ZERO), (f32 FP_ONE)))]
 >;
 
 class FABS <RegisterClass rc> : AMDGPUShaderInst <
@@ -322,7 +410,7 @@ class POW_Common <AMDGPUInst log_ieee, AMDGPUInst exp_ieee, AMDGPUInst mul>
 /* --------------------- */
 
 /* Extract element pattern */
-class Extract_Element <ValueType sub_type, ValueType vec_type, int sub_idx, 
+class Extract_Element <ValueType sub_type, ValueType vec_type, int sub_idx,
                        SubRegIndex sub_reg>
   : Pat<
   (sub_type (vector_extract vec_type:$src, sub_idx)),
@@ -337,12 +425,6 @@ class Insert_Element <ValueType elem_type, ValueType vec_type,
   (INSERT_SUBREG $vec, $elem, sub_reg)
 >;
 
-class Vector4_Build <ValueType vecType, ValueType elemType> : Pat <
-  (vecType (build_vector elemType:$x, elemType:$y, elemType:$z, elemType:$w)),
-  (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG
-    (vecType (IMPLICIT_DEF)), $x, sub0), $y, sub1), $z, sub2), $w, sub3)
->;
-
 // XXX: Convert to new syntax and use COPY_TO_REG, once the DFAPacketizer
 // can handle COPY instructions.
 // bitconvert pattern
@@ -360,7 +442,7 @@ class DwordAddrPat<ValueType vt, RegisterClass rc> : Pat <
 
 // BFI_INT patterns
 
-multiclass BFIPatterns <Instruction BFI_INT> {
+multiclass BFIPatterns <Instruction BFI_INT, Instruction LoadImm32> {
 
   // Definition from ISA doc:
   // (y & x) | (z & ~x)
@@ -376,6 +458,19 @@ multiclass BFIPatterns <Instruction BFI_INT> {
     (BFI_INT $x, $y, $z)
   >;
 
+  def : Pat <
+    (fcopysign f32:$src0, f32:$src1),
+    (BFI_INT (LoadImm32 0x7fffffff), $src0, $src1)
+  >;
+
+  def : Pat <
+    (f64 (fcopysign f64:$src0, f64:$src1)),
+      (INSERT_SUBREG (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
+      (i32 (EXTRACT_SUBREG $src0, sub0)), sub0),
+      (BFI_INT (LoadImm32 0x7fffffff),
+               (i32 (EXTRACT_SUBREG $src0, sub1)),
+               (i32 (EXTRACT_SUBREG $src1, sub1))), sub1)
+  >;
 }
 
 // SHA-256 Ma patterns
@@ -420,7 +515,61 @@ class UMUL24Pattern <Instruction UMUL24> : Pat <
 >;
 */
 
+class IMad24Pat<Instruction Inst> : Pat <
+  (add (AMDGPUmul_i24 i32:$src0, i32:$src1), i32:$src2),
+  (Inst $src0, $src1, $src2)
+>;
+
+class UMad24Pat<Instruction Inst> : Pat <
+  (add (AMDGPUmul_u24 i32:$src0, i32:$src1), i32:$src2),
+  (Inst $src0, $src1, $src2)
+>;
+
+multiclass Expand24IBitOps<Instruction MulInst, Instruction AddInst> {
+  def _expand_imad24 : Pat <
+    (AMDGPUmad_i24 i32:$src0, i32:$src1, i32:$src2),
+    (AddInst (MulInst $src0, $src1), $src2)
+  >;
+
+  def _expand_imul24 : Pat <
+    (AMDGPUmul_i24 i32:$src0, i32:$src1),
+    (MulInst $src0, $src1)
+  >;
+}
+
+multiclass Expand24UBitOps<Instruction MulInst, Instruction AddInst> {
+  def _expand_umad24 : Pat <
+    (AMDGPUmad_u24 i32:$src0, i32:$src1, i32:$src2),
+    (AddInst (MulInst $src0, $src1), $src2)
+  >;
+
+  def _expand_umul24 : Pat <
+    (AMDGPUmul_u24 i32:$src0, i32:$src1),
+    (MulInst $src0, $src1)
+  >;
+}
+
+class RcpPat<Instruction RcpInst, ValueType vt> : Pat <
+  (fdiv FP_ONE, vt:$src),
+  (RcpInst $src)
+>;
+
+multiclass RsqPat<Instruction RsqInst, ValueType vt> {
+  def : Pat <
+    (fdiv FP_ONE, (fsqrt vt:$src)),
+    (RsqInst $src)
+  >;
+
+  def : Pat <
+    (AMDGPUrcp (fsqrt vt:$src)),
+    (RsqInst $src)
+  >;
+}
+
 include "R600Instructions.td"
+include "R700Instructions.td"
+include "EvergreenInstructions.td"
+include "CaymanInstructions.td"
 
 include "SIInstrInfo.td"
 
diff --git a/contrib/llvm/lib/Target/R600/AMDILIntrinsicInfo.cpp b/contrib/llvm/lib/Target/R600/AMDGPUIntrinsicInfo.cpp
index 762ee39..58916a9 100644
--- a/contrib/llvm/lib/Target/R600/AMDILIntrinsicInfo.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUIntrinsicInfo.cpp
@@ -1,4 +1,4 @@
-//===- AMDILIntrinsicInfo.cpp - AMDGPU Intrinsic Information ------*- C++ -*-===//
+//===- AMDGPUIntrinsicInfo.cpp - AMDGPU Intrinsic Information ---*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -12,7 +12,7 @@
 //
 //===-----------------------------------------------------------------------===//
 
-#include "AMDILIntrinsicInfo.h"
+#include "AMDGPUIntrinsicInfo.h"
 #include "AMDGPUSubtarget.h"
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Intrinsics.h"
@@ -24,39 +24,37 @@ using namespace llvm;
 #include "AMDGPUGenIntrinsics.inc"
 #undef GET_LLVM_INTRINSIC_FOR_GCC_BUILTIN
 
-AMDGPUIntrinsicInfo::AMDGPUIntrinsicInfo(TargetMachine *tm) 
-  : TargetIntrinsicInfo() {
-}
+AMDGPUIntrinsicInfo::AMDGPUIntrinsicInfo(TargetMachine *tm)
+    : TargetIntrinsicInfo() {}
 
-std::string 
-AMDGPUIntrinsicInfo::getName(unsigned int IntrID, Type **Tys,
-    unsigned int numTys) const  {
-  static const char* const names[] = {
+std::string AMDGPUIntrinsicInfo::getName(unsigned IntrID, Type **Tys,
+                                         unsigned numTys) const {
+  static const char *const names[] = {
 #define GET_INTRINSIC_NAME_TABLE
 #include "AMDGPUGenIntrinsics.inc"
 #undef GET_INTRINSIC_NAME_TABLE
   };
 
   if (IntrID < Intrinsic::num_intrinsics) {
-    return 0;
+    return nullptr;
   }
-  assert(IntrID < AMDGPUIntrinsic::num_AMDGPU_intrinsics
-      && "Invalid intrinsic ID");
+  assert(IntrID < AMDGPUIntrinsic::num_AMDGPU_intrinsics &&
+         "Invalid intrinsic ID");
 
   std::string Result(names[IntrID - Intrinsic::num_intrinsics]);
   return Result;
 }
 
-unsigned int
-AMDGPUIntrinsicInfo::lookupName(const char *Name, unsigned int Len) const  {
+unsigned AMDGPUIntrinsicInfo::lookupName(const char *Name,
+                                         unsigned Len) const {
   if (!StringRef(Name, Len).startswith("llvm."))
     return 0; // All intrinsics start with 'llvm.'
 
 #define GET_FUNCTION_RECOGNIZER
 #include "AMDGPUGenIntrinsics.inc"
 #undef GET_FUNCTION_RECOGNIZER
-  AMDGPUIntrinsic::ID IntrinsicID
-    = (AMDGPUIntrinsic::ID)Intrinsic::not_intrinsic;
+  AMDGPUIntrinsic::ID IntrinsicID =
+      (AMDGPUIntrinsic::ID)Intrinsic::not_intrinsic;
   IntrinsicID = getIntrinsicForGCCBuiltin("AMDGPU", Name);
 
   if (IntrinsicID != (AMDGPUIntrinsic::ID)Intrinsic::not_intrinsic) {
@@ -65,17 +63,15 @@ AMDGPUIntrinsicInfo::lookupName(const char *Name, unsigned int Len) const  {
   return 0;
 }
 
-bool 
-AMDGPUIntrinsicInfo::isOverloaded(unsigned id) const  {
-  // Overload Table
+bool AMDGPUIntrinsicInfo::isOverloaded(unsigned id) const {
+// Overload Table
 #define GET_INTRINSIC_OVERLOAD_TABLE
 #include "AMDGPUGenIntrinsics.inc"
 #undef GET_INTRINSIC_OVERLOAD_TABLE
 }
 
-Function*
-AMDGPUIntrinsicInfo::getDeclaration(Module *M, unsigned IntrID,
-    Type **Tys,
-    unsigned numTys) const  {
+Function *AMDGPUIntrinsicInfo::getDeclaration(Module *M, unsigned IntrID,
+                                              Type **Tys,
+                                              unsigned numTys) const {
   llvm_unreachable("Not implemented");
 }
diff --git a/contrib/llvm/lib/Target/R600/AMDILIntrinsicInfo.h b/contrib/llvm/lib/Target/R600/AMDGPUIntrinsicInfo.h
index 35559e2..5be68a2 100644
--- a/contrib/llvm/lib/Target/R600/AMDILIntrinsicInfo.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPUIntrinsicInfo.h
@@ -1,4 +1,4 @@
-//===- AMDILIntrinsicInfo.h - AMDGPU Intrinsic Information ------*- C++ -*-===//
+//===- AMDGPUIntrinsicInfo.h - AMDGPU Intrinsic Information ------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -11,8 +11,8 @@
 /// \brief Interface for the AMDGPU Implementation of the Intrinsic Info class.
 //
 //===-----------------------------------------------------------------------===//
-#ifndef AMDIL_INTRINSICS_H
-#define AMDIL_INTRINSICS_H
+#ifndef AMDGPU_INTRINSICINFO_H
+#define AMDGPU_INTRINSICINFO_H
 
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/Target/TargetIntrinsicInfo.h"
@@ -34,16 +34,15 @@ enum ID {
 class AMDGPUIntrinsicInfo : public TargetIntrinsicInfo {
 public:
   AMDGPUIntrinsicInfo(TargetMachine *tm);
-  std::string getName(unsigned int IntrId, Type **Tys = 0,
-                      unsigned int numTys = 0) const;
-  unsigned int lookupName(const char *Name, unsigned int Len) const;
-  bool isOverloaded(unsigned int IID) const;
-  Function *getDeclaration(Module *M, unsigned int ID,
-                           Type **Tys = 0,
-                           unsigned int numTys = 0) const;
+  std::string getName(unsigned IntrId, Type **Tys = nullptr,
+                      unsigned numTys = 0) const override;
+  unsigned lookupName(const char *Name, unsigned Len) const override;
+  bool isOverloaded(unsigned IID) const override;
+  Function *getDeclaration(Module *M, unsigned ID,
+                           Type **Tys = nullptr,
+                           unsigned numTys = 0) const override;
 };
 
 } // end namespace llvm
 
-#endif // AMDIL_INTRINSICS_H
-
+#endif // AMDGPU_INTRINSICINFO_H
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUIntrinsics.td b/contrib/llvm/lib/Target/R600/AMDGPUIntrinsics.td
index 9f975bf..eee9c29 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUIntrinsics.td
+++ b/contrib/llvm/lib/Target/R600/AMDGPUIntrinsics.td
@@ -13,23 +13,28 @@
 
 let TargetPrefix = "AMDGPU", isTarget = 1 in {
 
-  def int_AMDGPU_load_const : Intrinsic<[llvm_float_ty], [llvm_i32_ty], [IntrNoMem]>;
-  def int_AMDGPU_load_imm : Intrinsic<[llvm_v4f32_ty], [llvm_i32_ty], [IntrNoMem]>;
-  def int_AMDGPU_reserve_reg : Intrinsic<[], [llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_store_output : Intrinsic<[], [llvm_float_ty, llvm_i32_ty], []>;
   def int_AMDGPU_swizzle : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty], [IntrNoMem]>;
-
+  def int_AMDGPU_abs : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_arl : Intrinsic<[llvm_i32_ty], [llvm_float_ty], [IntrNoMem]>;
   def int_AMDGPU_cndlt : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
   def int_AMDGPU_div : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_fract : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
+  def int_AMDGPU_clamp : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
+
+  // This is named backwards (instead of rsq_legacy) so we don't have
+  // to define it with the public builtins intrinsics. This is a
+  // workaround for how intrinsic names are parsed. If the name is
+  // llvm.AMDGPU.rsq.legacy, the parser assumes that you meant
+  // llvm.AMDGPU.rsq.{f32 | f64} and incorrectly mangled the name.
+  def int_AMDGPU_legacy_rsq : Intrinsic<[llvm_float_ty], [llvm_float_ty], [IntrNoMem]>;
+
   def int_AMDGPU_dp4 : Intrinsic<[llvm_float_ty], [llvm_v4f32_ty, llvm_v4f32_ty], [IntrNoMem]>;
   def int_AMDGPU_kill : Intrinsic<[], [llvm_float_ty], []>;
   def int_AMDGPU_kilp : Intrinsic<[], [], []>;
   def int_AMDGPU_lrp : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
   def int_AMDGPU_mul : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
   def int_AMDGPU_pow : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
-  def int_AMDGPU_rcp : Intrinsic<[llvm_float_ty], [llvm_float_ty], [IntrNoMem]>;
-  def int_AMDGPU_rsq : Intrinsic<[llvm_float_ty], [llvm_float_ty], [IntrNoMem]>;
   def int_AMDGPU_seq : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
   def int_AMDGPU_sgt : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
   def int_AMDGPU_sge : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
@@ -49,9 +54,31 @@ let TargetPrefix = "AMDGPU", isTarget = 1 in {
   def int_AMDGPU_imin : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_umax : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_umin : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_umul24 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_imul24 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_imad24 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_umad24 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_cvt_f32_ubyte0 : Intrinsic<[llvm_float_ty], [llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_cvt_f32_ubyte1 : Intrinsic<[llvm_float_ty], [llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_cvt_f32_ubyte2 : Intrinsic<[llvm_float_ty], [llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_cvt_f32_ubyte3 : Intrinsic<[llvm_float_ty], [llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_cube : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty], [IntrNoMem]>;
-
+  def int_AMDGPU_bfi : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_bfe_i32 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_bfe_u32 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_bfm : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_brev : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_barrier_local  : Intrinsic<[], [], []>;
+  def int_AMDGPU_barrier_global  : Intrinsic<[], [], []>;
+}
+
+// Legacy names for compatibility.
+let TargetPrefix = "AMDIL", isTarget = 1 in {
+  def int_AMDIL_abs : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>], [IntrNoMem]>;
+  def int_AMDIL_fraction : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
+  def int_AMDIL_clamp : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
+  def int_AMDIL_exp : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
+  def int_AMDIL_round_nearest : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
 }
 
 let TargetPrefix = "TGSI", isTarget = 1 in {
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUMCInstLower.cpp b/contrib/llvm/lib/Target/R600/AMDGPUMCInstLower.cpp
index 0ed598e..ce5c41c 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUMCInstLower.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUMCInstLower.cpp
@@ -15,12 +15,16 @@
 
 #include "AMDGPUMCInstLower.h"
 #include "AMDGPUAsmPrinter.h"
+#include "AMDGPUTargetMachine.h"
 #include "InstPrinter/AMDGPUInstPrinter.h"
 #include "R600InstrInfo.h"
+#include "SIInstrInfo.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/IR/Constants.h"
+#include "llvm/IR/GlobalVariable.h"
 #include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCObjectStreamer.h"
@@ -31,16 +35,30 @@
 
 using namespace llvm;
 
-AMDGPUMCInstLower::AMDGPUMCInstLower(MCContext &ctx):
-  Ctx(ctx)
+AMDGPUMCInstLower::AMDGPUMCInstLower(MCContext &ctx, const AMDGPUSubtarget &st):
+  Ctx(ctx), ST(st)
 { }
 
+enum AMDGPUMCInstLower::SISubtarget
+AMDGPUMCInstLower::AMDGPUSubtargetToSISubtarget(unsigned) const {
+  return AMDGPUMCInstLower::SI;
+}
+
+unsigned AMDGPUMCInstLower::getMCOpcode(unsigned MIOpcode) const {
+
+  int MCOpcode = AMDGPU::getMCOpcode(MIOpcode,
+                              AMDGPUSubtargetToSISubtarget(ST.getGeneration()));
+  if (MCOpcode == -1)
+    MCOpcode = MIOpcode;
+
+  return MCOpcode;
+}
+
 void AMDGPUMCInstLower::lower(const MachineInstr *MI, MCInst &OutMI) const {
-  OutMI.setOpcode(MI->getOpcode());
 
-  for (unsigned i = 0, e = MI->getNumExplicitOperands(); i != e; ++i) {
-    const MachineOperand &MO = MI->getOperand(i);
+  OutMI.setOpcode(getMCOpcode(MI->getOpcode()));
 
+  for (const MachineOperand &MO : MI->explicit_operands()) {
     MCOperand MCOp;
     switch (MO.getType()) {
     default:
@@ -61,14 +79,36 @@ void AMDGPUMCInstLower::lower(const MachineInstr *MI, MCInst &OutMI) const {
     case MachineOperand::MO_MachineBasicBlock:
       MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
                                    MO.getMBB()->getSymbol(), Ctx));
+      break;
+    case MachineOperand::MO_GlobalAddress: {
+      const GlobalValue *GV = MO.getGlobal();
+      MCSymbol *Sym = Ctx.GetOrCreateSymbol(StringRef(GV->getName()));
+      MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(Sym, Ctx));
+      break;
+    }
+    case MachineOperand::MO_TargetIndex: {
+      assert(MO.getIndex() == AMDGPU::TI_CONSTDATA_START);
+      MCSymbol *Sym = Ctx.GetOrCreateSymbol(StringRef(END_OF_TEXT_LABEL_NAME));
+      const MCSymbolRefExpr *Expr = MCSymbolRefExpr::Create(Sym, Ctx);
+      MCOp = MCOperand::CreateExpr(Expr);
+      break;
+    }
     }
     OutMI.addOperand(MCOp);
   }
 }
 
 void AMDGPUAsmPrinter::EmitInstruction(const MachineInstr *MI) {
-  AMDGPUMCInstLower MCInstLowering(OutContext);
+  AMDGPUMCInstLower MCInstLowering(OutContext,
+                               MF->getTarget().getSubtarget<AMDGPUSubtarget>());
 
+#ifdef _DEBUG
+  StringRef Err;
+  if (!TM.getInstrInfo()->verifyInstruction(MI, Err)) {
+    errs() << "Warning: Illegal instruction detected: " << Err << "\n";
+    MI->dump();
+  }
+#endif
   if (MI->isBundle()) {
     const MachineBasicBlock *MBB = MI->getParent();
     MachineBasicBlock::const_instr_iterator I = MI;
@@ -80,7 +120,7 @@ void AMDGPUAsmPrinter::EmitInstruction(const MachineInstr *MI) {
   } else {
     MCInst TmpInst;
     MCInstLowering.lower(MI, TmpInst);
-    OutStreamer.EmitInstruction(TmpInst);
+    EmitToStreamer(OutStreamer, TmpInst);
 
     if (DisasmEnabled) {
       // Disassemble instruction/operands to text.
@@ -99,7 +139,8 @@ void AMDGPUAsmPrinter::EmitInstruction(const MachineInstr *MI) {
 
       MCObjectStreamer &ObjStreamer = (MCObjectStreamer &)OutStreamer;
       MCCodeEmitter &InstEmitter = ObjStreamer.getAssembler().getEmitter();
-      InstEmitter.EncodeInstruction(TmpInst, CodeStream, Fixups);
+      InstEmitter.EncodeInstruction(TmpInst, CodeStream, Fixups,
+                                    TM.getSubtarget<MCSubtargetInfo>());
       CodeStream.flush();
 
       HexLines.resize(HexLines.size() + 1);
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUMCInstLower.h b/contrib/llvm/lib/Target/R600/AMDGPUMCInstLower.h
index d7d538e..58fe34d 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUMCInstLower.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPUMCInstLower.h
@@ -13,16 +13,30 @@
 
 namespace llvm {
 
-class MCInst;
-class MCContext;
+class AMDGPUSubtarget;
 class MachineInstr;
+class MCContext;
+class MCInst;
 
 class AMDGPUMCInstLower {
 
+  // This must be kept in sync with the SISubtarget class in SIInstrInfo.td
+  enum SISubtarget {
+    SI = 0
+  };
+
   MCContext &Ctx;
+  const AMDGPUSubtarget &ST;
+
+  /// Convert a member of the AMDGPUSubtarget::Generation enum to the
+  /// SISubtarget enum.
+  enum SISubtarget AMDGPUSubtargetToSISubtarget(unsigned Gen) const;
+
+  /// Get the MC opcode for this MachineInstr.
+  unsigned getMCOpcode(unsigned MIOpcode) const;
 
 public:
-  AMDGPUMCInstLower(MCContext &ctx);
+  AMDGPUMCInstLower(MCContext &ctx, const AMDGPUSubtarget &ST);
 
   /// \brief Lower a MachineInstr to an MCInst
   void lower(const MachineInstr *MI, MCInst &OutMI) const;
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUMachineFunction.cpp b/contrib/llvm/lib/Target/R600/AMDGPUMachineFunction.cpp
index 14171f4..90af801 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUMachineFunction.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUMachineFunction.cpp
@@ -10,9 +10,9 @@ static const char *const ShaderTypeAttribute = "ShaderType";
 void AMDGPUMachineFunction::anchor() {}
 
 AMDGPUMachineFunction::AMDGPUMachineFunction(const MachineFunction &MF) :
-    MachineFunctionInfo() {
-  ShaderType = ShaderType::COMPUTE;
-  LDSSize = 0;
+  MachineFunctionInfo(),
+  ShaderType(ShaderType::COMPUTE),
+  LDSSize(0) {
   AttributeSet Set = MF.getFunction()->getAttributes();
   Attribute A = Set.getAttribute(AttributeSet::FunctionIndex,
                                  ShaderTypeAttribute);
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUMachineFunction.h b/contrib/llvm/lib/Target/R600/AMDGPUMachineFunction.h
index fea0b39..0854d58 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUMachineFunction.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPUMachineFunction.h
@@ -20,14 +20,19 @@ namespace llvm {
 
 class AMDGPUMachineFunction : public MachineFunctionInfo {
   virtual void anchor();
+  unsigned ShaderType;
+
 public:
   AMDGPUMachineFunction(const MachineFunction &MF);
-  unsigned ShaderType;
   /// A map to keep track of local memory objects and their offsets within
   /// the local memory space.
   std::map<const GlobalValue *, unsigned> LocalMemoryObjects;
   /// Number of bytes in the LDS that are being used.
   unsigned LDSSize;
+
+  unsigned getShaderType() const {
+    return ShaderType;
+  }
 };
 
 }
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUPromoteAlloca.cpp b/contrib/llvm/lib/Target/R600/AMDGPUPromoteAlloca.cpp
new file mode 100644
index 0000000..218750d
--- /dev/null
+++ b/contrib/llvm/lib/Target/R600/AMDGPUPromoteAlloca.cpp
@@ -0,0 +1,387 @@
+//===-- AMDGPUPromoteAlloca.cpp - Promote Allocas -------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass eliminates allocas by either converting them into vectors or
+// by migrating them to local address space.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "AMDGPUSubtarget.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/InstVisitor.h"
+#include "llvm/Support/Debug.h"
+
+#define DEBUG_TYPE "amdgpu-promote-alloca"
+
+using namespace llvm;
+
+namespace {
+
+class AMDGPUPromoteAlloca : public FunctionPass,
+                       public InstVisitor<AMDGPUPromoteAlloca> {
+
+  static char ID;
+  Module *Mod;
+  const AMDGPUSubtarget &ST;
+  int LocalMemAvailable;
+
+public:
+  AMDGPUPromoteAlloca(const AMDGPUSubtarget &st) : FunctionPass(ID), ST(st),
+                                                   LocalMemAvailable(0) { }
+  virtual bool doInitialization(Module &M);
+  virtual bool runOnFunction(Function &F);
+  virtual const char *getPassName() const {
+    return "AMDGPU Promote Alloca";
+  }
+  void visitAlloca(AllocaInst &I);
+};
+
+} // End anonymous namespace
+
+char AMDGPUPromoteAlloca::ID = 0;
+
+bool AMDGPUPromoteAlloca::doInitialization(Module &M) {
+  Mod = &M;
+  return false;
+}
+
+bool AMDGPUPromoteAlloca::runOnFunction(Function &F) {
+
+  const FunctionType *FTy = F.getFunctionType();
+
+  LocalMemAvailable = ST.getLocalMemorySize();
+
+
+  // If the function has any arguments in the local address space, then it's
+  // possible these arguments require the entire local memory space, so
+  // we cannot use local memory in the pass.
+  for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i) {
+    const Type *ParamTy = FTy->getParamType(i);
+    if (ParamTy->isPointerTy() &&
+        ParamTy->getPointerAddressSpace() == AMDGPUAS::LOCAL_ADDRESS) {
+      LocalMemAvailable = 0;
+      DEBUG(dbgs() << "Function has local memory argument.  Promoting to "
+                      "local memory disabled.\n");
+      break;
+    }
+  }
+
+  if (LocalMemAvailable > 0) {
+    // Check how much local memory is being used by global objects
+    for (Module::global_iterator I = Mod->global_begin(),
+                                 E = Mod->global_end(); I != E; ++I) {
+      GlobalVariable *GV = I;
+      PointerType *GVTy = GV->getType();
+      if (GVTy->getAddressSpace() != AMDGPUAS::LOCAL_ADDRESS)
+        continue;
+      for (Value::use_iterator U = GV->use_begin(),
+                               UE = GV->use_end(); U != UE; ++U) {
+        Instruction *Use = dyn_cast<Instruction>(*U);
+        if (!Use)
+          continue;
+        if (Use->getParent()->getParent() == &F)
+          LocalMemAvailable -=
+              Mod->getDataLayout()->getTypeAllocSize(GVTy->getElementType());
+      }
+    }
+  }
+
+  LocalMemAvailable = std::max(0, LocalMemAvailable);
+  DEBUG(dbgs() << LocalMemAvailable << "bytes free in local memory.\n");
+
+  visit(F);
+
+  return false;
+}
+
+static VectorType *arrayTypeToVecType(const Type *ArrayTy) {
+  return VectorType::get(ArrayTy->getArrayElementType(),
+                         ArrayTy->getArrayNumElements());
+}
+
+static Value* calculateVectorIndex(Value *Ptr,
+                                  std::map<GetElementPtrInst*, Value*> GEPIdx) {
+  if (isa<AllocaInst>(Ptr))
+    return Constant::getNullValue(Type::getInt32Ty(Ptr->getContext()));
+
+  GetElementPtrInst *GEP = cast<GetElementPtrInst>(Ptr);
+
+  return GEPIdx[GEP];
+}
+
+static Value* GEPToVectorIndex(GetElementPtrInst *GEP) {
+  // FIXME we only support simple cases
+  if (GEP->getNumOperands() != 3)
+    return NULL;
+
+  ConstantInt *I0 = dyn_cast<ConstantInt>(GEP->getOperand(1));
+  if (!I0 || !I0->isZero())
+    return NULL;
+
+  return GEP->getOperand(2);
+}
+
+// Not an instruction handled below to turn into a vector.
+//
+// TODO: Check isTriviallyVectorizable for calls and handle other
+// instructions.
+static bool canVectorizeInst(Instruction *Inst) {
+  switch (Inst->getOpcode()) {
+  case Instruction::Load:
+  case Instruction::Store:
+  case Instruction::BitCast:
+  case Instruction::AddrSpaceCast:
+    return true;
+  default:
+    return false;
+  }
+}
+
+static bool tryPromoteAllocaToVector(AllocaInst *Alloca) {
+  Type *AllocaTy = Alloca->getAllocatedType();
+
+  DEBUG(dbgs() << "Alloca Candidate for vectorization \n");
+
+  // FIXME: There is no reason why we can't support larger arrays, we
+  // are just being conservative for now.
+  if (!AllocaTy->isArrayTy() ||
+      AllocaTy->getArrayElementType()->isVectorTy() ||
+      AllocaTy->getArrayNumElements() > 4) {
+
+    DEBUG(dbgs() << "  Cannot convert type to vector");
+    return false;
+  }
+
+  std::map<GetElementPtrInst*, Value*> GEPVectorIdx;
+  std::vector<Value*> WorkList;
+  for (User *AllocaUser : Alloca->users()) {
+    GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(AllocaUser);
+    if (!GEP) {
+      if (!canVectorizeInst(cast<Instruction>(AllocaUser)))
+        return false;
+
+      WorkList.push_back(AllocaUser);
+      continue;
+    }
+
+    Value *Index = GEPToVectorIndex(GEP);
+
+    // If we can't compute a vector index from this GEP, then we can't
+    // promote this alloca to vector.
+    if (!Index) {
+      DEBUG(dbgs() << "  Cannot compute vector index for GEP " << *GEP << '\n');
+      return false;
+    }
+
+    GEPVectorIdx[GEP] = Index;
+    for (User *GEPUser : AllocaUser->users()) {
+      if (!canVectorizeInst(cast<Instruction>(GEPUser)))
+        return false;
+
+      WorkList.push_back(GEPUser);
+    }
+  }
+
+  VectorType *VectorTy = arrayTypeToVecType(AllocaTy);
+
+  DEBUG(dbgs() << "  Converting alloca to vector "
+        << *AllocaTy << " -> " << *VectorTy << '\n');
+
+  for (std::vector<Value*>::iterator I = WorkList.begin(),
+                                     E = WorkList.end(); I != E; ++I) {
+    Instruction *Inst = cast<Instruction>(*I);
+    IRBuilder<> Builder(Inst);
+    switch (Inst->getOpcode()) {
+    case Instruction::Load: {
+      Value *Ptr = Inst->getOperand(0);
+      Value *Index = calculateVectorIndex(Ptr, GEPVectorIdx);
+      Value *BitCast = Builder.CreateBitCast(Alloca, VectorTy->getPointerTo(0));
+      Value *VecValue = Builder.CreateLoad(BitCast);
+      Value *ExtractElement = Builder.CreateExtractElement(VecValue, Index);
+      Inst->replaceAllUsesWith(ExtractElement);
+      Inst->eraseFromParent();
+      break;
+    }
+    case Instruction::Store: {
+      Value *Ptr = Inst->getOperand(1);
+      Value *Index = calculateVectorIndex(Ptr, GEPVectorIdx);
+      Value *BitCast = Builder.CreateBitCast(Alloca, VectorTy->getPointerTo(0));
+      Value *VecValue = Builder.CreateLoad(BitCast);
+      Value *NewVecValue = Builder.CreateInsertElement(VecValue,
+                                                       Inst->getOperand(0),
+                                                       Index);
+      Builder.CreateStore(NewVecValue, BitCast);
+      Inst->eraseFromParent();
+      break;
+    }
+    case Instruction::BitCast:
+    case Instruction::AddrSpaceCast:
+      break;
+
+    default:
+      Inst->dump();
+      llvm_unreachable("Inconsistency in instructions promotable to vector");
+    }
+  }
+  return true;
+}
+
+static void collectUsesWithPtrTypes(Value *Val, std::vector<Value*> &WorkList) {
+  for (User *User : Val->users()) {
+    if(std::find(WorkList.begin(), WorkList.end(), User) != WorkList.end())
+      continue;
+    if (isa<CallInst>(User)) {
+      WorkList.push_back(User);
+      continue;
+    }
+    if (!User->getType()->isPointerTy())
+      continue;
+    WorkList.push_back(User);
+    collectUsesWithPtrTypes(User, WorkList);
+  }
+}
+
+void AMDGPUPromoteAlloca::visitAlloca(AllocaInst &I) {
+  IRBuilder<> Builder(&I);
+
+  // First try to replace the alloca with a vector
+  Type *AllocaTy = I.getAllocatedType();
+
+  DEBUG(dbgs() << "Trying to promote " << I << '\n');
+
+  if (tryPromoteAllocaToVector(&I))
+    return;
+
+  DEBUG(dbgs() << " alloca is not a candidate for vectorization.\n");
+
+  // FIXME: This is the maximum work group size.  We should try to get
+  // value from the reqd_work_group_size function attribute if it is
+  // available.
+  unsigned WorkGroupSize = 256;
+  int AllocaSize = WorkGroupSize *
+      Mod->getDataLayout()->getTypeAllocSize(AllocaTy);
+
+  if (AllocaSize > LocalMemAvailable) {
+    DEBUG(dbgs() << " Not enough local memory to promote alloca.\n");
+    return;
+  }
+
+  DEBUG(dbgs() << "Promoting alloca to local memory\n");
+  LocalMemAvailable -= AllocaSize;
+
+  GlobalVariable *GV = new GlobalVariable(
+      *Mod, ArrayType::get(I.getAllocatedType(), 256), false,
+      GlobalValue::ExternalLinkage, 0, I.getName(), 0,
+      GlobalVariable::NotThreadLocal, AMDGPUAS::LOCAL_ADDRESS);
+
+  FunctionType *FTy = FunctionType::get(
+      Type::getInt32Ty(Mod->getContext()), false);
+  AttributeSet AttrSet;
+  AttrSet.addAttribute(Mod->getContext(), 0, Attribute::ReadNone);
+
+  Value *ReadLocalSizeY = Mod->getOrInsertFunction(
+      "llvm.r600.read.local.size.y", FTy, AttrSet);
+  Value *ReadLocalSizeZ = Mod->getOrInsertFunction(
+      "llvm.r600.read.local.size.z", FTy, AttrSet);
+  Value *ReadTIDIGX = Mod->getOrInsertFunction(
+      "llvm.r600.read.tidig.x", FTy, AttrSet);
+  Value *ReadTIDIGY = Mod->getOrInsertFunction(
+      "llvm.r600.read.tidig.y", FTy, AttrSet);
+  Value *ReadTIDIGZ = Mod->getOrInsertFunction(
+      "llvm.r600.read.tidig.z", FTy, AttrSet);
+
+
+  Value *TCntY = Builder.CreateCall(ReadLocalSizeY);
+  Value *TCntZ = Builder.CreateCall(ReadLocalSizeZ);
+  Value *TIdX  = Builder.CreateCall(ReadTIDIGX);
+  Value *TIdY  = Builder.CreateCall(ReadTIDIGY);
+  Value *TIdZ  = Builder.CreateCall(ReadTIDIGZ);
+
+  Value *Tmp0 = Builder.CreateMul(TCntY, TCntZ);
+  Tmp0 = Builder.CreateMul(Tmp0, TIdX);
+  Value *Tmp1 = Builder.CreateMul(TIdY, TCntZ);
+  Value *TID = Builder.CreateAdd(Tmp0, Tmp1);
+  TID = Builder.CreateAdd(TID, TIdZ);
+
+  std::vector<Value*> Indices;
+  Indices.push_back(Constant::getNullValue(Type::getInt32Ty(Mod->getContext())));
+  Indices.push_back(TID);
+
+  Value *Offset = Builder.CreateGEP(GV, Indices);
+  I.mutateType(Offset->getType());
+  I.replaceAllUsesWith(Offset);
+  I.eraseFromParent();
+
+  std::vector<Value*> WorkList;
+
+  collectUsesWithPtrTypes(Offset, WorkList);
+
+  for (std::vector<Value*>::iterator i = WorkList.begin(),
+                                     e = WorkList.end(); i != e; ++i) {
+    Value *V = *i;
+    CallInst *Call = dyn_cast<CallInst>(V);
+    if (!Call) {
+      Type *EltTy = V->getType()->getPointerElementType();
+      PointerType *NewTy = PointerType::get(EltTy, AMDGPUAS::LOCAL_ADDRESS);
+      V->mutateType(NewTy);
+      continue;
+    }
+
+    IntrinsicInst *Intr = dyn_cast<IntrinsicInst>(Call);
+    if (!Intr) {
+      std::vector<Type*> ArgTypes;
+      for (unsigned ArgIdx = 0, ArgEnd = Call->getNumArgOperands();
+                                ArgIdx != ArgEnd; ++ArgIdx) {
+        ArgTypes.push_back(Call->getArgOperand(ArgIdx)->getType());
+      }
+      Function *F = Call->getCalledFunction();
+      FunctionType *NewType = FunctionType::get(Call->getType(), ArgTypes,
+                                                F->isVarArg());
+      Constant *C = Mod->getOrInsertFunction(StringRef(F->getName().str() + ".local"), NewType,
+                                             F->getAttributes());
+      Function *NewF = cast<Function>(C);
+      Call->setCalledFunction(NewF);
+      continue;
+    }
+
+    Builder.SetInsertPoint(Intr);
+    switch (Intr->getIntrinsicID()) {
+    case Intrinsic::lifetime_start:
+    case Intrinsic::lifetime_end:
+      // These intrinsics are for address space 0 only
+      Intr->eraseFromParent();
+      continue;
+    case Intrinsic::memcpy: {
+      MemCpyInst *MemCpy = cast<MemCpyInst>(Intr);
+      Builder.CreateMemCpy(MemCpy->getRawDest(), MemCpy->getRawSource(),
+                           MemCpy->getLength(), MemCpy->getAlignment(),
+                           MemCpy->isVolatile());
+      Intr->eraseFromParent();
+      continue;
+    }
+    case Intrinsic::memset: {
+      MemSetInst *MemSet = cast<MemSetInst>(Intr);
+      Builder.CreateMemSet(MemSet->getRawDest(), MemSet->getValue(),
+                           MemSet->getLength(), MemSet->getAlignment(),
+                           MemSet->isVolatile());
+      Intr->eraseFromParent();
+      continue;
+    }
+    default:
+      Intr->dump();
+      llvm_unreachable("Don't know how to promote alloca intrinsic use.");
+    }
+  }
+}
+
+FunctionPass *llvm::createAMDGPUPromoteAlloca(const AMDGPUSubtarget &ST) {
+  return new AMDGPUPromoteAlloca(ST);
+}
diff --git a/contrib/llvm/lib/Target/R600/AMDGPURegisterInfo.cpp b/contrib/llvm/lib/Target/R600/AMDGPURegisterInfo.cpp
index 47617a7..3433280 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPURegisterInfo.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPURegisterInfo.cpp
@@ -17,9 +17,9 @@
 
 using namespace llvm;
 
-AMDGPURegisterInfo::AMDGPURegisterInfo(TargetMachine &tm)
+AMDGPURegisterInfo::AMDGPURegisterInfo(const AMDGPUSubtarget &st)
 : AMDGPUGenRegisterInfo(0),
-  TM(tm)
+  ST(st)
   { }
 
 //===----------------------------------------------------------------------===//
@@ -27,10 +27,10 @@ AMDGPURegisterInfo::AMDGPURegisterInfo(TargetMachine &tm)
 // they are not supported at this time.
 //===----------------------------------------------------------------------===//
 
-const uint16_t AMDGPURegisterInfo::CalleeSavedReg = AMDGPU::NoRegister;
+const MCPhysReg AMDGPURegisterInfo::CalleeSavedReg = AMDGPU::NoRegister;
 
-const uint16_t* AMDGPURegisterInfo::getCalleeSavedRegs(const MachineFunction *MF)
-                                                                         const {
+const MCPhysReg*
+AMDGPURegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
   return &CalleeSavedReg;
 }
 
@@ -38,7 +38,7 @@ void AMDGPURegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
                                              int SPAdj,
                                              unsigned FIOperandNum,
                                              RegScavenger *RS) const {
-  assert(!"Subroutines not supported yet");
+  llvm_unreachable("Subroutines not supported yet");
 }
 
 unsigned AMDGPURegisterInfo::getFrameRegister(const MachineFunction &MF) const {
@@ -54,7 +54,7 @@ unsigned AMDGPURegisterInfo::getSubRegFromChannel(unsigned Channel) const {
     AMDGPU::sub15
   };
 
-  assert (Channel < array_lengthof(SubRegs));
+  assert(Channel < array_lengthof(SubRegs));
   return SubRegs[Channel];
 }
 
diff --git a/contrib/llvm/lib/Target/R600/AMDGPURegisterInfo.h b/contrib/llvm/lib/Target/R600/AMDGPURegisterInfo.h
index 688e1a0..46aa7a1 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPURegisterInfo.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPURegisterInfo.h
@@ -25,29 +25,21 @@
 
 namespace llvm {
 
-class AMDGPUTargetMachine;
+class AMDGPUSubtarget;
 class TargetInstrInfo;
 
 struct AMDGPURegisterInfo : public AMDGPUGenRegisterInfo {
-  TargetMachine &TM;
-  static const uint16_t CalleeSavedReg;
+  static const MCPhysReg CalleeSavedReg;
+  const AMDGPUSubtarget &ST;
 
-  AMDGPURegisterInfo(TargetMachine &tm);
+  AMDGPURegisterInfo(const AMDGPUSubtarget &st);
 
-  virtual BitVector getReservedRegs(const MachineFunction &MF) const {
+  BitVector getReservedRegs(const MachineFunction &MF) const override {
     assert(!"Unimplemented");  return BitVector();
   }
 
-  /// \param RC is an AMDIL reg class.
-  ///
-  /// \returns The ISA reg class that is equivalent to \p RC.
-  virtual const TargetRegisterClass * getISARegClass(
-                                         const TargetRegisterClass * RC) const {
-    assert(!"Unimplemented"); return NULL;
-  }
-
   virtual const TargetRegisterClass* getCFGStructurizerRegClass(MVT VT) const {
-    assert(!"Unimplemented"); return NULL;
+    assert(!"Unimplemented"); return nullptr;
   }
 
   virtual unsigned getHWRegIndex(unsigned Reg) const {
@@ -58,11 +50,11 @@ struct AMDGPURegisterInfo : public AMDGPUGenRegisterInfo {
   /// (e.g. getSubRegFromChannel(0) -> AMDGPU::sub0)
   unsigned getSubRegFromChannel(unsigned Channel) const;
 
-  const uint16_t* getCalleeSavedRegs(const MachineFunction *MF) const;
-  void eliminateFrameIndex(MachineBasicBlock::iterator MI, int SPAdj,
+  const MCPhysReg* getCalleeSavedRegs(const MachineFunction *MF) const override;
+  virtual void eliminateFrameIndex(MachineBasicBlock::iterator MI, int SPAdj,
                            unsigned FIOperandNum,
-                           RegScavenger *RS) const;
-  unsigned getFrameRegister(const MachineFunction &MF) const;
+                           RegScavenger *RS) const override;
+  unsigned getFrameRegister(const MachineFunction &MF) const override;
 
   unsigned getIndirectSubReg(unsigned IndirectIndex) const;
 
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUSubtarget.cpp b/contrib/llvm/lib/Target/R600/AMDGPUSubtarget.cpp
index 061793a..e3c2a50 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUSubtarget.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUSubtarget.cpp
@@ -13,108 +13,77 @@
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPUSubtarget.h"
+#include "R600InstrInfo.h"
+#include "SIInstrInfo.h"
+#include "llvm/ADT/SmallString.h"
+
+#include "llvm/ADT/SmallString.h"
 
 using namespace llvm;
 
+#define DEBUG_TYPE "amdgpu-subtarget"
+
 #define GET_SUBTARGETINFO_ENUM
 #define GET_SUBTARGETINFO_TARGET_DESC
 #define GET_SUBTARGETINFO_CTOR
 #include "AMDGPUGenSubtargetInfo.inc"
 
-AMDGPUSubtarget::AMDGPUSubtarget(StringRef TT, StringRef CPU, StringRef FS) :
-  AMDGPUGenSubtargetInfo(TT, CPU, FS), DumpCode(false) {
-    InstrItins = getInstrItineraryForCPU(CPU);
+AMDGPUSubtarget::AMDGPUSubtarget(StringRef TT, StringRef GPU, StringRef FS) :
+  AMDGPUGenSubtargetInfo(TT, GPU, FS),
+  DevName(GPU),
+  Is64bit(false),
+  DumpCode(false),
+  R600ALUInst(false),
+  HasVertexCache(false),
+  TexVTXClauseSize(0),
+  Gen(AMDGPUSubtarget::R600),
+  FP64(false),
+  FP64Denormals(false),
+  FP32Denormals(false),
+  CaymanISA(false),
+  EnableIRStructurizer(true),
+  EnablePromoteAlloca(false),
+  EnableIfCvt(true),
+  WavefrontSize(0),
+  CFALUBug(false),
+  LocalMemorySize(0),
+  InstrItins(getInstrItineraryForCPU(GPU)) {
+  // On SI+, we want FP64 denormals to be on by default. FP32 denormals can be
+  // enabled, but some instructions do not respect them and they run at the
+  // double precision rate, so don't enable by default.
+  //
+  // We want to be able to turn these off, but making this a subtarget feature
+  // for SI has the unhelpful behavior that it unsets everything else if you
+  // disable it.
 
-  // Default card
-  StringRef GPU = CPU;
-  Is64bit = false;
-  DefaultSize[0] = 64;
-  DefaultSize[1] = 1;
-  DefaultSize[2] = 1;
-  HasVertexCache = false;
-  TexVTXClauseSize = 0;
-  Gen = AMDGPUSubtarget::R600;
-  FP64 = false;
-  CaymanISA = false;
-  EnableIRStructurizer = true;
-  EnableIfCvt = true;
-  ParseSubtargetFeatures(GPU, FS);
-  DevName = GPU;
-}
+  SmallString<256> FullFS("+promote-alloca,+fp64-denormals,");
+  FullFS += FS;
 
-bool
-AMDGPUSubtarget::is64bit() const  {
-  return Is64bit;
-}
-bool
-AMDGPUSubtarget::hasVertexCache() const {
-  return HasVertexCache;
-}
-short
-AMDGPUSubtarget::getTexVTXClauseSize() const {
-  return TexVTXClauseSize;
-}
-enum AMDGPUSubtarget::Generation
-AMDGPUSubtarget::getGeneration() const {
-  return Gen;
-}
-bool
-AMDGPUSubtarget::hasHWFP64() const {
-  return FP64;
-}
-bool
-AMDGPUSubtarget::hasCaymanISA() const {
-  return CaymanISA;
-}
-bool
-AMDGPUSubtarget::IsIRStructurizerEnabled() const {
-  return EnableIRStructurizer;
-}
-bool
-AMDGPUSubtarget::isIfCvtEnabled() const {
-  return EnableIfCvt;
-}
-bool
-AMDGPUSubtarget::isTargetELF() const {
-  return false;
-}
-size_t
-AMDGPUSubtarget::getDefaultSize(uint32_t dim) const {
-  if (dim > 3) {
-    return 1;
-  } else {
-    return DefaultSize[dim];
-  }
-}
-
-std::string
-AMDGPUSubtarget::getDataLayout() const {
-  std::string DataLayout = std::string(
-   "e"
-   "-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32"
-   "-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128"
-   "-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024-v2048:2048:2048"
-   "-n32:64"
-  );
+  ParseSubtargetFeatures(GPU, FullFS);
 
-  if (hasHWFP64()) {
-    DataLayout.append("-f64:64:64");
-  }
+  if (getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
+    InstrInfo.reset(new R600InstrInfo(*this));
 
-  if (is64bit()) {
-    DataLayout.append("-p:64:64:64");
+    // FIXME: I don't think think Evergreen has any useful support for
+    // denormals, but should be checked. Should we issue a warning somewhere if
+    // someone tries to enable these?
+    FP32Denormals = false;
+    FP64Denormals = false;
   } else {
-    DataLayout.append("-p:32:32:32");
-  }
-
-  if (Gen >= AMDGPUSubtarget::SOUTHERN_ISLANDS) {
-    DataLayout.append("-p3:32:32:32");
+    InstrInfo.reset(new SIInstrInfo(*this));
   }
-
-  return DataLayout;
 }
 
-std::string
-AMDGPUSubtarget::getDeviceName() const {
-  return DevName;
+unsigned AMDGPUSubtarget::getStackEntrySize() const {
+  assert(getGeneration() <= NORTHERN_ISLANDS);
+  switch(getWavefrontSize()) {
+  case 16:
+    return 8;
+  case 32:
+    return hasCaymanISA() ? 4 : 8;
+  case 64:
+    return 4;
+  default:
+    llvm_unreachable("Illegal wavefront size.");
+  }
 }
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUSubtarget.h b/contrib/llvm/lib/Target/R600/AMDGPUSubtarget.h
index 4288d27..a844b37 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUSubtarget.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPUSubtarget.h
@@ -15,6 +15,7 @@
 #ifndef AMDGPUSUBTARGET_H
 #define AMDGPUSUBTARGET_H
 #include "AMDGPU.h"
+#include "AMDGPUInstrInfo.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/Target/TargetSubtargetInfo.h"
@@ -27,6 +28,9 @@
 namespace llvm {
 
 class AMDGPUSubtarget : public AMDGPUGenSubtargetInfo {
+
+  std::unique_ptr<AMDGPUInstrInfo> InstrInfo;
+
 public:
   enum Generation {
     R600 = 0,
@@ -38,49 +42,156 @@ public:
   };
 
 private:
-  size_t DefaultSize[3];
   std::string DevName;
   bool Is64bit;
-  bool Is32on64bit;
   bool DumpCode;
   bool R600ALUInst;
   bool HasVertexCache;
   short TexVTXClauseSize;
-  enum Generation Gen;
+  Generation Gen;
   bool FP64;
+  bool FP64Denormals;
+  bool FP32Denormals;
   bool CaymanISA;
   bool EnableIRStructurizer;
+  bool EnablePromoteAlloca;
   bool EnableIfCvt;
+  unsigned WavefrontSize;
+  bool CFALUBug;
+  int LocalMemorySize;
 
   InstrItineraryData InstrItins;
 
 public:
   AMDGPUSubtarget(StringRef TT, StringRef CPU, StringRef FS);
 
-  const InstrItineraryData &getInstrItineraryData() const { return InstrItins; }
-  virtual void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
+  const AMDGPUInstrInfo *getInstrInfo() const {
+    return InstrInfo.get();
+  }
+
+  const InstrItineraryData &getInstrItineraryData() const {
+    return InstrItins;
+  }
+
+  void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
+
+  bool is64bit() const {
+    return Is64bit;
+  }
+
+  bool hasVertexCache() const {
+    return HasVertexCache;
+  }
+
+  short getTexVTXClauseSize() const {
+    return TexVTXClauseSize;
+  }
+
+  Generation getGeneration() const {
+    return Gen;
+  }
+
+  bool hasHWFP64() const {
+    return FP64;
+  }
+
+  bool hasCaymanISA() const {
+    return CaymanISA;
+  }
+
+  bool hasFP32Denormals() const {
+    return FP32Denormals;
+  }
+
+  bool hasFP64Denormals() const {
+    return FP64Denormals;
+  }
+
+  bool hasBFE() const {
+    return (getGeneration() >= EVERGREEN);
+  }
+
+  bool hasBFI() const {
+    return (getGeneration() >= EVERGREEN);
+  }
+
+  bool hasBFM() const {
+    return hasBFE();
+  }
+
+  bool hasBCNT(unsigned Size) const {
+    if (Size == 32)
+      return (getGeneration() >= EVERGREEN);
 
-  bool is64bit() const;
-  bool hasVertexCache() const;
-  short getTexVTXClauseSize() const;
-  enum Generation getGeneration() const;
-  bool hasHWFP64() const;
-  bool hasCaymanISA() const;
-  bool IsIRStructurizerEnabled() const;
-  bool isIfCvtEnabled() const;
+    if (Size == 64)
+      return (getGeneration() >= SOUTHERN_ISLANDS);
 
-  virtual bool enableMachineScheduler() const {
+    return false;
+  }
+
+  bool hasMulU24() const {
+    return (getGeneration() >= EVERGREEN);
+  }
+
+  bool hasMulI24() const {
+    return (getGeneration() >= SOUTHERN_ISLANDS ||
+            hasCaymanISA());
+  }
+
+  bool hasFFBL() const {
+    return (getGeneration() >= EVERGREEN);
+  }
+
+  bool hasFFBH() const {
+    return (getGeneration() >= EVERGREEN);
+  }
+
+  bool IsIRStructurizerEnabled() const {
+    return EnableIRStructurizer;
+  }
+
+  bool isPromoteAllocaEnabled() const {
+    return EnablePromoteAlloca;
+  }
+
+  bool isIfCvtEnabled() const {
+    return EnableIfCvt;
+  }
+
+  unsigned getWavefrontSize() const {
+    return WavefrontSize;
+  }
+
+  unsigned getStackEntrySize() const;
+
+  bool hasCFAluBug() const {
+    assert(getGeneration() <= NORTHERN_ISLANDS);
+    return CFALUBug;
+  }
+
+  int getLocalMemorySize() const {
+    return LocalMemorySize;
+  }
+
+  bool enableMachineScheduler() const override {
     return getGeneration() <= NORTHERN_ISLANDS;
   }
 
   // Helper functions to simplify if statements
-  bool isTargetELF() const;
-  std::string getDataLayout() const;
-  std::string getDeviceName() const;
-  virtual size_t getDefaultSize(uint32_t dim) const;
-  bool dumpCode() const { return DumpCode; }
-  bool r600ALUEncoding() const { return R600ALUInst; }
+  bool isTargetELF() const {
+    return false;
+  }
 
+  StringRef getDeviceName() const {
+    return DevName;
+  }
+
+  bool dumpCode() const {
+    return DumpCode;
+  }
+  bool r600ALUEncoding() const {
+    return R600ALUInst;
+  }
 };
 
 } // End namespace llvm
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUTargetMachine.cpp b/contrib/llvm/lib/Target/R600/AMDGPUTargetMachine.cpp
index bc4f5d7..56ba719 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUTargetMachine.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUTargetMachine.cpp
@@ -21,10 +21,10 @@
 #include "SIISelLowering.h"
 #include "SIInstrInfo.h"
 #include "llvm/Analysis/Passes.h"
-#include "llvm/Analysis/Verifier.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/Passes.h"
+#include "llvm/IR/Verifier.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/PassManager.h"
 #include "llvm/Support/TargetRegistry.h"
@@ -33,7 +33,6 @@
 #include "llvm/Transforms/Scalar.h"
 #include <llvm/CodeGen/Passes.h>
 
-
 using namespace llvm;
 
 extern "C" void LLVMInitializeR600Target() {
@@ -42,13 +41,27 @@ extern "C" void LLVMInitializeR600Target() {
 }
 
 static ScheduleDAGInstrs *createR600MachineScheduler(MachineSchedContext *C) {
-  return new ScheduleDAGMI(C, new R600SchedStrategy());
+  return new ScheduleDAGMILive(C, make_unique<R600SchedStrategy>());
 }
 
 static MachineSchedRegistry
 SchedCustomRegistry("r600", "Run R600's custom scheduler",
                     createR600MachineScheduler);
 
+static std::string computeDataLayout(const AMDGPUSubtarget &ST) {
+  std::string Ret = "e-p:32:32";
+
+  if (ST.is64bit()) {
+    // 32-bit local, and region pointers. 64-bit private, global, and constant.
+    Ret += "-p1:64:64-p2:64:64-p3:32:32-p4:64:64-p5:32:32-p24:64:64";
+  }
+
+  Ret += "-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256"
+         "-v512:512-v1024:1024-v2048:2048-n32:64";
+
+  return Ret;
+}
+
 AMDGPUTargetMachine::AMDGPUTargetMachine(const Target &T, StringRef TT,
     StringRef CPU, StringRef FS,
   TargetOptions Options,
@@ -58,7 +71,7 @@ AMDGPUTargetMachine::AMDGPUTargetMachine(const Target &T, StringRef TT,
 :
   LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OptLevel),
   Subtarget(TT, CPU, FS),
-  Layout(Subtarget.getDataLayout()),
+  Layout(computeDataLayout(Subtarget)),
   FrameLowering(TargetFrameLowering::StackGrowsUp,
                 64 * 16 // Maximum stack alignment (long16)
                , 0),
@@ -66,12 +79,11 @@ AMDGPUTargetMachine::AMDGPUTargetMachine(const Target &T, StringRef TT,
   InstrItins(&Subtarget.getInstrItineraryData()) {
   // TLInfo uses InstrInfo so it must be initialized after.
   if (Subtarget.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
-    InstrInfo.reset(new R600InstrInfo(*this));
     TLInfo.reset(new R600TargetLowering(*this));
   } else {
-    InstrInfo.reset(new SIInstrInfo(*this));
     TLInfo.reset(new SITargetLowering(*this));
   }
+  setRequiresStructuredCFG(true);
   initAsmInfo();
 }
 
@@ -88,20 +100,21 @@ public:
     return getTM<AMDGPUTargetMachine>();
   }
 
-  virtual ScheduleDAGInstrs *
-  createMachineScheduler(MachineSchedContext *C) const {
+  ScheduleDAGInstrs *
+  createMachineScheduler(MachineSchedContext *C) const override {
     const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
     if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS)
       return createR600MachineScheduler(C);
-    return 0;
+    return nullptr;
   }
 
-  virtual bool addPreISel();
-  virtual bool addInstSelector();
-  virtual bool addPreRegAlloc();
-  virtual bool addPostRegAlloc();
-  virtual bool addPreSched2();
-  virtual bool addPreEmitPass();
+  virtual void addCodeGenPrepare();
+  bool addPreISel() override;
+  bool addInstSelector() override;
+  bool addPreRegAlloc() override;
+  bool addPostRegAlloc() override;
+  bool addPreSched2() override;
+  bool addPreEmitPass() override;
 };
 } // End of anonymous namespace
 
@@ -121,13 +134,23 @@ void AMDGPUTargetMachine::addAnalysisPasses(PassManagerBase &PM) {
   PM.add(createAMDGPUTargetTransformInfoPass(this));
 }
 
+void AMDGPUPassConfig::addCodeGenPrepare() {
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+  if (ST.isPromoteAllocaEnabled()) {
+    addPass(createAMDGPUPromoteAlloca(ST));
+    addPass(createSROAPass());
+  }
+
+  TargetPassConfig::addCodeGenPrepare();
+}
+
 bool
 AMDGPUPassConfig::addPreISel() {
   const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
   addPass(createFlattenCFGPass());
   if (ST.IsIRStructurizerEnabled())
     addPass(createStructurizeCFGPass());
-  if (ST.getGeneration() > AMDGPUSubtarget::NORTHERN_ISLANDS) {
+  if (ST.getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS) {
     addPass(createSinkingPass());
     addPass(createSITypeRewriter());
     addPass(createSIAnnotateControlFlowPass());
@@ -139,17 +162,23 @@ AMDGPUPassConfig::addPreISel() {
 
 bool AMDGPUPassConfig::addInstSelector() {
   addPass(createAMDGPUISelDag(getAMDGPUTargetMachine()));
+  addPass(createSILowerI1CopiesPass());
   return false;
 }
 
 bool AMDGPUPassConfig::addPreRegAlloc() {
-  addPass(createAMDGPUConvertToISAPass(*TM));
   const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
 
   if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
     addPass(createR600VectorRegMerger(*TM));
   } else {
     addPass(createSIFixSGPRCopiesPass(*TM));
+    // SIFixSGPRCopies can generate a lot of duplicate instructions,
+    // so we need to run MachineCSE afterwards.
+    addPass(&MachineCSEID);
+    addPass(createSIShrinkInstructionsPass());
+    initializeSIFixSGPRLiveRangesPass(*PassRegistry::getPassRegistry());
+    insertPass(&RegisterCoalescerID, &SIFixSGPRLiveRangesID);
   }
   return false;
 }
@@ -157,6 +186,7 @@ bool AMDGPUPassConfig::addPreRegAlloc() {
 bool AMDGPUPassConfig::addPostRegAlloc() {
   const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
 
+  addPass(createSIShrinkInstructionsPass());
   if (ST.getGeneration() > AMDGPUSubtarget::NORTHERN_ISLANDS) {
     addPass(createSIInsertWaits(*TM));
   }
@@ -167,7 +197,7 @@ bool AMDGPUPassConfig::addPreSched2() {
   const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
 
   if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS)
-    addPass(createR600EmitClauseMarkers(*TM));
+    addPass(createR600EmitClauseMarkers());
   if (ST.isIfCvtEnabled())
     addPass(&IfConverterID);
   if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS)
@@ -178,7 +208,7 @@ bool AMDGPUPassConfig::addPreSched2() {
 bool AMDGPUPassConfig::addPreEmitPass() {
   const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
   if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
-    addPass(createAMDGPUCFGStructurizerPass(*TM));
+    addPass(createAMDGPUCFGStructurizerPass());
     addPass(createR600ExpandSpecialInstrsPass(*TM));
     addPass(&FinalizeMachineBundlesID);
     addPass(createR600Packetizer(*TM));
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUTargetMachine.h b/contrib/llvm/lib/Target/R600/AMDGPUTargetMachine.h
index f942614..3bb15be 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUTargetMachine.h
+++ b/contrib/llvm/lib/Target/R600/AMDGPUTargetMachine.h
@@ -17,10 +17,9 @@
 
 #include "AMDGPUFrameLowering.h"
 #include "AMDGPUInstrInfo.h"
+#include "AMDGPUIntrinsicInfo.h"
 #include "AMDGPUSubtarget.h"
-#include "AMDILIntrinsicInfo.h"
 #include "R600ISelLowering.h"
-#include "llvm/ADT/OwningPtr.h"
 #include "llvm/IR/DataLayout.h"
 
 namespace llvm {
@@ -31,8 +30,7 @@ class AMDGPUTargetMachine : public LLVMTargetMachine {
   const DataLayout Layout;
   AMDGPUFrameLowering FrameLowering;
   AMDGPUIntrinsicInfo IntrinsicInfo;
-  OwningPtr<AMDGPUInstrInfo> InstrInfo;
-  OwningPtr<AMDGPUTargetLowering> TLInfo;
+  std::unique_ptr<AMDGPUTargetLowering> TLInfo;
   const InstrItineraryData *InstrItins;
 
 public:
@@ -40,30 +38,32 @@ public:
                       StringRef CPU, TargetOptions Options, Reloc::Model RM,
                       CodeModel::Model CM, CodeGenOpt::Level OL);
   ~AMDGPUTargetMachine();
-  virtual const AMDGPUFrameLowering *getFrameLowering() const {
+  const AMDGPUFrameLowering *getFrameLowering() const override {
     return &FrameLowering;
   }
-  virtual const AMDGPUIntrinsicInfo *getIntrinsicInfo() const {
+  const AMDGPUIntrinsicInfo *getIntrinsicInfo() const override {
     return &IntrinsicInfo;
   }
-  virtual const AMDGPUInstrInfo *getInstrInfo() const {
-    return InstrInfo.get();
+  const AMDGPUInstrInfo *getInstrInfo() const override {
+    return getSubtargetImpl()->getInstrInfo();
   }
-  virtual const AMDGPUSubtarget *getSubtargetImpl() const { return &Subtarget; }
-  virtual const AMDGPURegisterInfo *getRegisterInfo() const {
-    return &InstrInfo->getRegisterInfo();
+  const AMDGPUSubtarget *getSubtargetImpl() const override {
+    return &Subtarget;
   }
-  virtual AMDGPUTargetLowering *getTargetLowering() const {
+  const AMDGPURegisterInfo *getRegisterInfo() const override {
+    return &getInstrInfo()->getRegisterInfo();
+  }
+  AMDGPUTargetLowering *getTargetLowering() const override {
     return TLInfo.get();
   }
-  virtual const InstrItineraryData *getInstrItineraryData() const {
+  const InstrItineraryData *getInstrItineraryData() const override {
     return InstrItins;
   }
-  virtual const DataLayout *getDataLayout() const { return &Layout; }
-  virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
+  const DataLayout *getDataLayout() const override { return &Layout; }
+  TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
 
   /// \brief Register R600 analysis passes with a pass manager.
-  virtual void addAnalysisPasses(PassManagerBase &PM);
+  void addAnalysisPasses(PassManagerBase &PM) override;
 };
 
 } // End namespace llvm
diff --git a/contrib/llvm/lib/Target/R600/AMDGPUTargetTransformInfo.cpp b/contrib/llvm/lib/Target/R600/AMDGPUTargetTransformInfo.cpp
index 8db319c..88934b6 100644
--- a/contrib/llvm/lib/Target/R600/AMDGPUTargetTransformInfo.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDGPUTargetTransformInfo.cpp
@@ -15,15 +15,18 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "AMDGPUtti"
 #include "AMDGPU.h"
 #include "AMDGPUTargetMachine.h"
+#include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
+#include "llvm/Analysis/ValueTracking.h"
 #include "llvm/Support/Debug.h"
-#include "llvm/Target/TargetLowering.h"
 #include "llvm/Target/CostTable.h"
+#include "llvm/Target/TargetLowering.h"
 using namespace llvm;
 
+#define DEBUG_TYPE "AMDGPUtti"
+
 // Declare the pass initialization routine locally as target-specific passes
 // don't have a target-wide initialization entry point, and so we rely on the
 // pass constructor initialization.
@@ -33,7 +36,7 @@ void initializeAMDGPUTTIPass(PassRegistry &);
 
 namespace {
 
-class AMDGPUTTI : public ImmutablePass, public TargetTransformInfo {
+class AMDGPUTTI final : public ImmutablePass, public TargetTransformInfo {
   const AMDGPUTargetMachine *TM;
   const AMDGPUSubtarget *ST;
   const AMDGPUTargetLowering *TLI;
@@ -43,7 +46,7 @@ class AMDGPUTTI : public ImmutablePass, public TargetTransformInfo {
   unsigned getScalarizationOverhead(Type *Ty, bool Insert, bool Extract) const;
 
 public:
-  AMDGPUTTI() : ImmutablePass(ID), TM(0), ST(0), TLI(0) {
+  AMDGPUTTI() : ImmutablePass(ID), TM(nullptr), ST(nullptr), TLI(nullptr) {
     llvm_unreachable("This pass cannot be directly constructed");
   }
 
@@ -53,11 +56,9 @@ public:
     initializeAMDGPUTTIPass(*PassRegistry::getPassRegistry());
   }
 
-  virtual void initializePass() { pushTTIStack(this); }
-
-  virtual void finalizePass() { popTTIStack(); }
+  void initializePass() override { pushTTIStack(this); }
 
-  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
     TargetTransformInfo::getAnalysisUsage(AU);
   }
 
@@ -65,13 +66,22 @@ public:
   static char ID;
 
   /// Provide necessary pointer adjustments for the two base classes.
-  virtual void *getAdjustedAnalysisPointer(const void *ID) {
+  void *getAdjustedAnalysisPointer(const void *ID) override {
     if (ID == &TargetTransformInfo::ID)
       return (TargetTransformInfo *)this;
     return this;
   }
 
-  virtual bool hasBranchDivergence() const;
+  bool hasBranchDivergence() const override;
+
+  void getUnrollingPreferences(Loop *L,
+                               UnrollingPreferences &UP) const override;
+
+  PopcntSupportKind getPopcntSupport(unsigned IntTyWidthInBit) const override;
+
+  unsigned getNumberOfRegisters(bool Vector) const override;
+  unsigned getRegisterBitWidth(bool Vector) const override;
+  unsigned getMaximumUnrollFactor() const override;
 
   /// @}
 };
@@ -88,3 +98,56 @@ llvm::createAMDGPUTargetTransformInfoPass(const AMDGPUTargetMachine *TM) {
 }
 
 bool AMDGPUTTI::hasBranchDivergence() const { return true; }
+
+void AMDGPUTTI::getUnrollingPreferences(Loop *L,
+                                        UnrollingPreferences &UP) const {
+  for (const BasicBlock *BB : L->getBlocks()) {
+    for (const Instruction &I : *BB) {
+      const GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(&I);
+      if (!GEP || GEP->getAddressSpace() != AMDGPUAS::PRIVATE_ADDRESS)
+        continue;
+
+      const Value *Ptr = GEP->getPointerOperand();
+      const AllocaInst *Alloca = dyn_cast<AllocaInst>(GetUnderlyingObject(Ptr));
+      if (Alloca) {
+        // We want to do whatever we can to limit the number of alloca
+        // instructions that make it through to the code generator.  allocas
+        // require us to use indirect addressing, which is slow and prone to
+        // compiler bugs.  If this loop does an address calculation on an
+        // alloca ptr, then we want to use a higher than normal loop unroll
+        // threshold. This will give SROA a better chance to eliminate these
+        // allocas.
+        //
+        // Don't use the maximum allowed value here as it will make some
+        // programs way too big.
+        UP.Threshold = 500;
+      }
+    }
+  }
+}
+
+AMDGPUTTI::PopcntSupportKind
+AMDGPUTTI::getPopcntSupport(unsigned TyWidth) const {
+  assert(isPowerOf2_32(TyWidth) && "Ty width must be power of 2");
+  return ST->hasBCNT(TyWidth) ? PSK_FastHardware : PSK_Software;
+}
+
+unsigned AMDGPUTTI::getNumberOfRegisters(bool Vec) const {
+  if (Vec)
+    return 0;
+
+  // Number of VGPRs on SI.
+  if (ST->getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS)
+    return 256;
+
+  return 4 * 128; // XXX - 4 channels. Should these count as vector instead?
+}
+
+unsigned AMDGPUTTI::getRegisterBitWidth(bool) const {
+  return 32;
+}
+
+unsigned AMDGPUTTI::getMaximumUnrollFactor() const {
+  // Semi-arbitrary large amount.
+  return 64;
+}
diff --git a/contrib/llvm/lib/Target/R600/AMDILBase.td b/contrib/llvm/lib/Target/R600/AMDILBase.td
deleted file mode 100644
index 5dcd478..0000000
--- a/contrib/llvm/lib/Target/R600/AMDILBase.td
+++ /dev/null
@@ -1,25 +0,0 @@
-//===- AMDIL.td - AMDIL Target Machine -------------*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-// Target-independent interfaces which we are implementing
-//===----------------------------------------------------------------------===//
-
-include "llvm/Target/Target.td"
-
-// Dummy Instruction itineraries for pseudo instructions
-def ALU_NULL : FuncUnit;
-def NullALU : InstrItinClass;
-
-//===----------------------------------------------------------------------===//
-// Register File, Calling Conv, Instruction Descriptions
-//===----------------------------------------------------------------------===//
-
-
-include "AMDILRegisterInfo.td"
-include "AMDILInstrInfo.td"
-
diff --git a/contrib/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp b/contrib/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp
index 507570f..f3a0391 100644
--- a/contrib/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp
+++ b/contrib/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp
@@ -8,19 +8,13 @@
 /// \file
 //==-----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "structcfg"
-
 #include "AMDGPU.h"
 #include "AMDGPUInstrInfo.h"
 #include "R600InstrInfo.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/DepthFirstIterator.h"
 #include "llvm/ADT/SCCIterator.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
-#include "llvm/ADT/DepthFirstIterator.h"
-#include "llvm/Analysis/DominatorInternals.h"
-#include "llvm/Analysis/Dominators.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
@@ -30,11 +24,16 @@
 #include "llvm/CodeGen/MachineLoopInfo.h"
 #include "llvm/CodeGen/MachinePostDominators.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetMachine.h"
 
 using namespace llvm;
 
+#define DEBUG_TYPE "structcfg"
+
 #define DEFAULT_VEC_SLOTS 8
 
 // TODO: move-begin.
@@ -54,6 +53,10 @@ STATISTIC(numLoopcontPatternMatch,  "CFGStructurizer number of loop-continue "
 STATISTIC(numClonedBlock,           "CFGStructurizer cloned blocks");
 STATISTIC(numClonedInstr,           "CFGStructurizer cloned instructions");
 
+namespace llvm {
+  void initializeAMDGPUCFGStructurizerPass(PassRegistry&);
+}
+
 //===----------------------------------------------------------------------===//
 //
 // Miscellaneous utility for CFGStructurizer.
@@ -131,16 +134,16 @@ public:
 
   static char ID;
 
-  AMDGPUCFGStructurizer(TargetMachine &tm) :
-      MachineFunctionPass(ID), TM(tm),
-      TII(static_cast<const R600InstrInfo *>(tm.getInstrInfo())),
-      TRI(&TII->getRegisterInfo()) { }
+  AMDGPUCFGStructurizer() :
+      MachineFunctionPass(ID), TII(nullptr), TRI(nullptr) {
+    initializeAMDGPUCFGStructurizerPass(*PassRegistry::getPassRegistry());
+  }
 
-   const char *getPassName() const {
-    return "AMD IL Control Flow Graph structurizer Pass";
+   const char *getPassName() const override {
+    return "AMDGPU Control Flow Graph structurizer Pass";
   }
 
-  void getAnalysisUsage(AnalysisUsage &AU) const {
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.addPreserved<MachineFunctionAnalysis>();
     AU.addRequired<MachineFunctionAnalysis>();
     AU.addRequired<MachineDominatorTree>();
@@ -156,14 +159,16 @@ public:
   /// sure all loops have an exit block
   bool prepare();
 
-  bool runOnMachineFunction(MachineFunction &MF) {
+  bool runOnMachineFunction(MachineFunction &MF) override {
+    TII = static_cast<const R600InstrInfo *>(MF.getTarget().getInstrInfo());
+    TRI = &TII->getRegisterInfo();
     DEBUG(MF.dump(););
     OrderedBlks.clear();
     FuncRep = &MF;
     MLI = &getAnalysis<MachineLoopInfo>();
     DEBUG(dbgs() << "LoopInfo:\n"; PrintLoopinfo(*MLI););
     MDT = &getAnalysis<MachineDominatorTree>();
-    DEBUG(MDT->print(dbgs(), (const llvm::Module*)0););
+    DEBUG(MDT->print(dbgs(), (const llvm::Module*)nullptr););
     PDT = &getAnalysis<MachinePostDominatorTree>();
     DEBUG(PDT->print(dbgs()););
     prepare();
@@ -173,7 +178,6 @@ public:
   }
 
 protected:
-  TargetMachine &TM;
   MachineDominatorTree *MDT;
   MachinePostDominatorTree *PDT;
   MachineLoopInfo *MLI;
@@ -220,7 +224,7 @@ protected:
   /// Compute the reversed DFS post order of Blocks
   void orderBlocks(MachineFunction *MF);
 
-  // Function originaly from CFGStructTraits
+  // Function originally from CFGStructTraits
   void insertInstrEnd(MachineBasicBlock *MBB, int NewOpcode,
       DebugLoc DL = DebugLoc());
   MachineInstr *insertInstrBefore(MachineBasicBlock *MBB, int NewOpcode,
@@ -330,7 +334,7 @@ protected:
       MachineBasicBlock *DstMBB, MachineBasicBlock::iterator I);
   void recordSccnum(MachineBasicBlock *MBB, int SCCNum);
   void retireBlock(MachineBasicBlock *MBB);
-  void setLoopLandBlock(MachineLoop *LoopRep, MachineBasicBlock *MBB = NULL);
+  void setLoopLandBlock(MachineLoop *LoopRep, MachineBasicBlock *MBB = nullptr);
 
   MachineBasicBlock *findNearestCommonPostDom(std::set<MachineBasicBlock *>&);
   /// This is work around solution for findNearestCommonDominator not avaiable
@@ -357,7 +361,7 @@ MachineBasicBlock *AMDGPUCFGStructurizer::getLoopLandInfo(MachineLoop *LoopRep)
     const {
   LoopLandInfoMap::const_iterator It = LLInfoMap.find(LoopRep);
   if (It == LLInfoMap.end())
-    return NULL;
+    return nullptr;
   return (*It).second;
 }
 
@@ -628,7 +632,7 @@ MachineInstr *AMDGPUCFGStructurizer::getNormalBlockBranchInstr(
   MachineInstr *MI = &*It;
   if (MI && (isCondBranch(MI) || isUncondBranch(MI)))
     return MI;
-  return NULL;
+  return nullptr;
 }
 
 MachineInstr *AMDGPUCFGStructurizer::getLoopendBlockBranchInstr(
@@ -644,7 +648,7 @@ MachineInstr *AMDGPUCFGStructurizer::getLoopendBlockBranchInstr(
         break;
     }
   }
-  return NULL;
+  return nullptr;
 }
 
 MachineInstr *AMDGPUCFGStructurizer::getReturnInstr(MachineBasicBlock *MBB) {
@@ -654,7 +658,7 @@ MachineInstr *AMDGPUCFGStructurizer::getReturnInstr(MachineBasicBlock *MBB) {
     if (instr->getOpcode() == AMDGPU::RETURN)
       return instr;
   }
-  return NULL;
+  return nullptr;
 }
 
 MachineInstr *AMDGPUCFGStructurizer::getContinueInstr(MachineBasicBlock *MBB) {
@@ -664,7 +668,7 @@ MachineInstr *AMDGPUCFGStructurizer::getContinueInstr(MachineBasicBlock *MBB) {
     if (MI->getOpcode() == AMDGPU::CONTINUE)
       return MI;
   }
-  return NULL;
+  return nullptr;
 }
 
 bool AMDGPUCFGStructurizer::isReturnBlock(MachineBasicBlock *MBB) {
@@ -786,7 +790,7 @@ bool AMDGPUCFGStructurizer::prepare() {
 bool AMDGPUCFGStructurizer::run() {
 
   //Assume reducible CFG...
-  DEBUG(dbgs() << "AMDGPUCFGStructurizer::run\n";FuncRep->viewCFG(););
+  DEBUG(dbgs() << "AMDGPUCFGStructurizer::run\n");
 
 #ifdef STRESSTEST
   //Use the worse block ordering to test the algorithm.
@@ -815,7 +819,7 @@ bool AMDGPUCFGStructurizer::run() {
 
     SmallVectorImpl<MachineBasicBlock *>::const_iterator SccBeginIter =
         It;
-    MachineBasicBlock *SccBeginMBB = NULL;
+    MachineBasicBlock *SccBeginMBB = nullptr;
     int SccNumBlk = 0;  // The number of active blocks, init to a
                         // maximum possible number.
     int SccNumIter;     // Number of iteration in this SCC.
@@ -858,8 +862,7 @@ bool AMDGPUCFGStructurizer::run() {
           ContNextScc = false;
           DEBUG(
             dbgs() << "repeat processing SCC" << getSCCNum(MBB)
-                   << "sccNumIter = " << SccNumIter << "\n";
-            FuncRep->viewCFG();
+                   << "sccNumIter = " << SccNumIter << '\n';
           );
         } else {
           // Finish the current scc.
@@ -871,7 +874,7 @@ bool AMDGPUCFGStructurizer::run() {
       }
 
       if (ContNextScc)
-        SccBeginMBB = NULL;
+        SccBeginMBB = nullptr;
     } //while, "one iteration" over the function.
 
     MachineBasicBlock *EntryMBB =
@@ -915,12 +918,10 @@ bool AMDGPUCFGStructurizer::run() {
   BlockInfoMap.clear();
   LLInfoMap.clear();
 
-  DEBUG(
-    FuncRep->viewCFG();
-  );
-
-  if (!Finish)
-    llvm_unreachable("IRREDUCIBL_CF");
+  if (!Finish) {
+    DEBUG(FuncRep->viewCFG());
+    llvm_unreachable("IRREDUCIBLE_CFG");
+  }
 
   return true;
 }
@@ -930,9 +931,9 @@ bool AMDGPUCFGStructurizer::run() {
 void AMDGPUCFGStructurizer::orderBlocks(MachineFunction *MF) {
   int SccNum = 0;
   MachineBasicBlock *MBB;
-  for (scc_iterator<MachineFunction *> It = scc_begin(MF), E = scc_end(MF);
-      It != E; ++It, ++SccNum) {
-    std::vector<MachineBasicBlock *> &SccNext = *It;
+  for (scc_iterator<MachineFunction *> It = scc_begin(MF); !It.isAtEnd();
+       ++It, ++SccNum) {
+    const std::vector<MachineBasicBlock *> &SccNext = *It;
     for (std::vector<MachineBasicBlock *>::const_iterator
          blockIter = SccNext.begin(), blockEnd = SccNext.end();
          blockIter != blockEnd; ++blockIter) {
@@ -1025,7 +1026,7 @@ int AMDGPUCFGStructurizer::ifPatternMatch(MachineBasicBlock *MBB) {
   } else if (TrueMBB->succ_size() == 1 && *TrueMBB->succ_begin() == FalseMBB) {
     // Triangle pattern, false is empty
     LandBlk = FalseMBB;
-    FalseMBB = NULL;
+    FalseMBB = nullptr;
   } else if (FalseMBB->succ_size() == 1
              && *FalseMBB->succ_begin() == TrueMBB) {
     // Triangle pattern, true is empty
@@ -1033,7 +1034,7 @@ int AMDGPUCFGStructurizer::ifPatternMatch(MachineBasicBlock *MBB) {
     std::swap(TrueMBB, FalseMBB);
     reversePredicateSetter(MBB->end());
     LandBlk = FalseMBB;
-    FalseMBB = NULL;
+    FalseMBB = nullptr;
   } else if (FalseMBB->succ_size() == 1
              && isSameloopDetachedContbreak(TrueMBB, FalseMBB)) {
     LandBlk = *FalseMBB->succ_begin();
@@ -1074,13 +1075,11 @@ int AMDGPUCFGStructurizer::ifPatternMatch(MachineBasicBlock *MBB) {
 
 int AMDGPUCFGStructurizer::loopendPatternMatch() {
   std::vector<MachineLoop *> NestedLoops;
-  for (MachineLoopInfo::iterator It = MLI->begin(), E = MLI->end();
-      It != E; ++It) {
-    df_iterator<MachineLoop *> LpIt = df_begin(*It),
-        LpE = df_end(*It);
-    for (; LpIt != LpE; ++LpIt)
-      NestedLoops.push_back(*LpIt);
-  }
+  for (MachineLoopInfo::iterator It = MLI->begin(), E = MLI->end(); It != E;
+       ++It)
+    for (MachineLoop *ML : depth_first(*It))
+      NestedLoops.push_back(ML);
+
   if (NestedLoops.size() == 0)
     return 0;
 
@@ -1234,7 +1233,7 @@ int AMDGPUCFGStructurizer::handleJumpintoIfImp(MachineBasicBlock *HeadMBB,
 
       numClonedBlock += Num;
       Num += serialPatternMatch(*HeadMBB->succ_begin());
-      Num += serialPatternMatch(*llvm::next(HeadMBB->succ_begin()));
+      Num += serialPatternMatch(*std::next(HeadMBB->succ_begin()));
       Num += ifPatternMatch(HeadMBB);
       assert(Num > 0);
 
@@ -1243,7 +1242,7 @@ int AMDGPUCFGStructurizer::handleJumpintoIfImp(MachineBasicBlock *HeadMBB,
     DEBUG(
       dbgs() << " not working\n";
     );
-    DownBlk = (DownBlk->succ_size() == 1) ? (*DownBlk->succ_begin()) : NULL;
+    DownBlk = (DownBlk->succ_size() == 1) ? (*DownBlk->succ_begin()) : nullptr;
   } // walk down the postDomTree
 
   return Num;
@@ -1722,11 +1721,11 @@ AMDGPUCFGStructurizer::normalizeInfiniteLoopExit(MachineLoop* LoopRep) {
   const TargetRegisterClass * I32RC = TRI->getCFGStructurizerRegClass(MVT::i32);
 
   if (!LoopHeader || !LoopLatch)
-    return NULL;
+    return nullptr;
   MachineInstr *BranchMI = getLoopendBlockBranchInstr(LoopLatch);
   // Is LoopRep an infinite loop ?
   if (!BranchMI || !isUncondBranch(BranchMI))
-    return NULL;
+    return nullptr;
 
   MachineBasicBlock *DummyExitBlk = FuncRep->CreateMachineBasicBlock();
   FuncRep->push_back(DummyExitBlk);  //insert to function
@@ -1763,7 +1762,7 @@ void AMDGPUCFGStructurizer::removeRedundantConditionalBranch(
   if (MBB->succ_size() != 2)
     return;
   MachineBasicBlock *MBB1 = *MBB->succ_begin();
-  MachineBasicBlock *MBB2 = *llvm::next(MBB->succ_begin());
+  MachineBasicBlock *MBB2 = *std::next(MBB->succ_begin());
   if (MBB1 != MBB2)
     return;
 
@@ -1859,7 +1858,7 @@ AMDGPUCFGStructurizer::findNearestCommonPostDom(MachineBasicBlock *MBB1,
     return findNearestCommonPostDom(MBB1, *MBB2->succ_begin());
 
   if (!Node1 || !Node2)
-    return NULL;
+    return nullptr;
 
   Node1 = Node1->getIDom();
   while (Node1) {
@@ -1868,7 +1867,7 @@ AMDGPUCFGStructurizer::findNearestCommonPostDom(MachineBasicBlock *MBB1,
     Node1 = Node1->getIDom();
   }
 
-  return NULL;
+  return nullptr;
 }
 
 MachineBasicBlock *
@@ -1899,6 +1898,14 @@ char AMDGPUCFGStructurizer::ID = 0;
 } // end anonymous namespace
 
 
-FunctionPass *llvm::createAMDGPUCFGStructurizerPass(TargetMachine &tm) {
-  return new AMDGPUCFGStructurizer(tm);
+INITIALIZE_PASS_BEGIN(AMDGPUCFGStructurizer, "amdgpustructurizer",
+                      "AMDGPU CFG Structurizer", false, false)
+INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
+INITIALIZE_PASS_DEPENDENCY(MachinePostDominatorTree)
+INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)
+INITIALIZE_PASS_END(AMDGPUCFGStructurizer, "amdgpustructurizer",
+                      "AMDGPU CFG Structurizer", false, false)
+
+FunctionPass *llvm::createAMDGPUCFGStructurizerPass() {
+  return new AMDGPUCFGStructurizer();
 }
diff --git a/contrib/llvm/lib/Target/R600/AMDILISelLowering.cpp b/contrib/llvm/lib/Target/R600/AMDILISelLowering.cpp
deleted file mode 100644
index 970787e..0000000
--- a/contrib/llvm/lib/Target/R600/AMDILISelLowering.cpp
+++ /dev/null
@@ -1,642 +0,0 @@
-//===-- AMDILISelLowering.cpp - AMDIL DAG Lowering Implementation ---------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//==-----------------------------------------------------------------------===//
-//
-/// \file
-/// \brief TargetLowering functions borrowed from AMDIL.
-//
-//===----------------------------------------------------------------------===//
-
-#include "AMDGPUISelLowering.h"
-#include "AMDGPURegisterInfo.h"
-#include "AMDGPUSubtarget.h"
-#include "AMDILIntrinsicInfo.h"
-#include "llvm/CodeGen/MachineFrameInfo.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/PseudoSourceValue.h"
-#include "llvm/CodeGen/SelectionDAG.h"
-#include "llvm/CodeGen/SelectionDAGNodes.h"
-#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
-#include "llvm/IR/CallingConv.h"
-#include "llvm/IR/DerivedTypes.h"
-#include "llvm/IR/Instructions.h"
-#include "llvm/IR/Intrinsics.h"
-#include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetOptions.h"
-
-using namespace llvm;
-//===----------------------------------------------------------------------===//
-// TargetLowering Implementation Help Functions End
-//===----------------------------------------------------------------------===//
-
-//===----------------------------------------------------------------------===//
-// TargetLowering Class Implementation Begins
-//===----------------------------------------------------------------------===//
-void AMDGPUTargetLowering::InitAMDILLowering() {
-  static const int types[] = {
-    (int)MVT::i8,
-    (int)MVT::i16,
-    (int)MVT::i32,
-    (int)MVT::f32,
-    (int)MVT::f64,
-    (int)MVT::i64,
-    (int)MVT::v2i8,
-    (int)MVT::v4i8,
-    (int)MVT::v2i16,
-    (int)MVT::v4i16,
-    (int)MVT::v4f32,
-    (int)MVT::v4i32,
-    (int)MVT::v2f32,
-    (int)MVT::v2i32,
-    (int)MVT::v2f64,
-    (int)MVT::v2i64
-  };
-
-  static const int IntTypes[] = {
-    (int)MVT::i8,
-    (int)MVT::i16,
-    (int)MVT::i32,
-    (int)MVT::i64
-  };
-
-  static const int FloatTypes[] = {
-    (int)MVT::f32,
-    (int)MVT::f64
-  };
-
-  static const int VectorTypes[] = {
-    (int)MVT::v2i8,
-    (int)MVT::v4i8,
-    (int)MVT::v2i16,
-    (int)MVT::v4i16,
-    (int)MVT::v4f32,
-    (int)MVT::v4i32,
-    (int)MVT::v2f32,
-    (int)MVT::v2i32,
-    (int)MVT::v2f64,
-    (int)MVT::v2i64
-  };
-  const size_t NumTypes = array_lengthof(types);
-  const size_t NumFloatTypes = array_lengthof(FloatTypes);
-  const size_t NumIntTypes = array_lengthof(IntTypes);
-  const size_t NumVectorTypes = array_lengthof(VectorTypes);
-
-  const AMDGPUSubtarget &STM = getTargetMachine().getSubtarget<AMDGPUSubtarget>();
-  // These are the current register classes that are
-  // supported
-
-  for (unsigned int x  = 0; x < NumTypes; ++x) {
-    MVT::SimpleValueType VT = (MVT::SimpleValueType)types[x];
-
-    //FIXME: SIGN_EXTEND_INREG is not meaningful for floating point types
-    // We cannot sextinreg, expand to shifts
-    setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Custom);
-    setOperationAction(ISD::SUBE, VT, Expand);
-    setOperationAction(ISD::SUBC, VT, Expand);
-    setOperationAction(ISD::ADDE, VT, Expand);
-    setOperationAction(ISD::ADDC, VT, Expand);
-    setOperationAction(ISD::BRCOND, VT, Custom);
-    setOperationAction(ISD::BR_JT, VT, Expand);
-    setOperationAction(ISD::BRIND, VT, Expand);
-    // TODO: Implement custom UREM/SREM routines
-    setOperationAction(ISD::SREM, VT, Expand);
-    setOperationAction(ISD::SMUL_LOHI, VT, Expand);
-    setOperationAction(ISD::UMUL_LOHI, VT, Expand);
-    if (VT != MVT::i64 && VT != MVT::v2i64) {
-      setOperationAction(ISD::SDIV, VT, Custom);
-    }
-  }
-  for (unsigned int x = 0; x < NumFloatTypes; ++x) {
-    MVT::SimpleValueType VT = (MVT::SimpleValueType)FloatTypes[x];
-
-    // IL does not have these operations for floating point types
-    setOperationAction(ISD::FP_ROUND_INREG, VT, Expand);
-    setOperationAction(ISD::SETOLT, VT, Expand);
-    setOperationAction(ISD::SETOGE, VT, Expand);
-    setOperationAction(ISD::SETOGT, VT, Expand);
-    setOperationAction(ISD::SETOLE, VT, Expand);
-    setOperationAction(ISD::SETULT, VT, Expand);
-    setOperationAction(ISD::SETUGE, VT, Expand);
-    setOperationAction(ISD::SETUGT, VT, Expand);
-    setOperationAction(ISD::SETULE, VT, Expand);
-  }
-
-  for (unsigned int x = 0; x < NumIntTypes; ++x) {
-    MVT::SimpleValueType VT = (MVT::SimpleValueType)IntTypes[x];
-
-    // GPU also does not have divrem function for signed or unsigned
-    setOperationAction(ISD::SDIVREM, VT, Expand);
-
-    // GPU does not have [S|U]MUL_LOHI functions as a single instruction
-    setOperationAction(ISD::SMUL_LOHI, VT, Expand);
-    setOperationAction(ISD::UMUL_LOHI, VT, Expand);
-
-    setOperationAction(ISD::BSWAP, VT, Expand);
-
-    // GPU doesn't have any counting operators
-    setOperationAction(ISD::CTPOP, VT, Expand);
-    setOperationAction(ISD::CTTZ, VT, Expand);
-    setOperationAction(ISD::CTLZ, VT, Expand);
-  }
-
-  for (unsigned int ii = 0; ii < NumVectorTypes; ++ii) {
-    MVT::SimpleValueType VT = (MVT::SimpleValueType)VectorTypes[ii];
-
-    setOperationAction(ISD::VECTOR_SHUFFLE, VT, Expand);
-    setOperationAction(ISD::SDIVREM, VT, Expand);
-    setOperationAction(ISD::SMUL_LOHI, VT, Expand);
-    // setOperationAction(ISD::VSETCC, VT, Expand);
-    setOperationAction(ISD::SELECT_CC, VT, Expand);
-
-  }
-  setOperationAction(ISD::MULHU, MVT::i64, Expand);
-  setOperationAction(ISD::MULHU, MVT::v2i64, Expand);
-  setOperationAction(ISD::MULHS, MVT::i64, Expand);
-  setOperationAction(ISD::MULHS, MVT::v2i64, Expand);
-  setOperationAction(ISD::ADD, MVT::v2i64, Expand);
-  setOperationAction(ISD::SREM, MVT::v2i64, Expand);
-  setOperationAction(ISD::Constant          , MVT::i64  , Legal);
-  setOperationAction(ISD::SDIV, MVT::v2i64, Expand);
-  setOperationAction(ISD::TRUNCATE, MVT::v2i64, Expand);
-  setOperationAction(ISD::SIGN_EXTEND, MVT::v2i64, Expand);
-  setOperationAction(ISD::ZERO_EXTEND, MVT::v2i64, Expand);
-  setOperationAction(ISD::ANY_EXTEND, MVT::v2i64, Expand);
-  if (STM.hasHWFP64()) {
-    // we support loading/storing v2f64 but not operations on the type
-    setOperationAction(ISD::FADD, MVT::v2f64, Expand);
-    setOperationAction(ISD::FSUB, MVT::v2f64, Expand);
-    setOperationAction(ISD::FMUL, MVT::v2f64, Expand);
-    setOperationAction(ISD::FP_ROUND_INREG, MVT::v2f64, Expand);
-    setOperationAction(ISD::FP_EXTEND, MVT::v2f64, Expand);
-    setOperationAction(ISD::ConstantFP        , MVT::f64  , Legal);
-    // We want to expand vector conversions into their scalar
-    // counterparts.
-    setOperationAction(ISD::TRUNCATE, MVT::v2f64, Expand);
-    setOperationAction(ISD::SIGN_EXTEND, MVT::v2f64, Expand);
-    setOperationAction(ISD::ZERO_EXTEND, MVT::v2f64, Expand);
-    setOperationAction(ISD::ANY_EXTEND, MVT::v2f64, Expand);
-    setOperationAction(ISD::FABS, MVT::f64, Expand);
-    setOperationAction(ISD::FABS, MVT::v2f64, Expand);
-  }
-  // TODO: Fix the UDIV24 algorithm so it works for these
-  // types correctly. This needs vector comparisons
-  // for this to work correctly.
-  setOperationAction(ISD::UDIV, MVT::v2i8, Expand);
-  setOperationAction(ISD::UDIV, MVT::v4i8, Expand);
-  setOperationAction(ISD::UDIV, MVT::v2i16, Expand);
-  setOperationAction(ISD::UDIV, MVT::v4i16, Expand);
-  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Custom);
-  setOperationAction(ISD::SUBC, MVT::Other, Expand);
-  setOperationAction(ISD::ADDE, MVT::Other, Expand);
-  setOperationAction(ISD::ADDC, MVT::Other, Expand);
-  setOperationAction(ISD::BRCOND, MVT::Other, Custom);
-  setOperationAction(ISD::BR_JT, MVT::Other, Expand);
-  setOperationAction(ISD::BRIND, MVT::Other, Expand);
-  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::Other, Expand);
-
-
-  // Use the default implementation.
-  setOperationAction(ISD::ConstantFP        , MVT::f32    , Legal);
-  setOperationAction(ISD::Constant          , MVT::i32    , Legal);
-
-  setSchedulingPreference(Sched::RegPressure);
-  setPow2DivIsCheap(false);
-  setSelectIsExpensive(true);
-  setJumpIsExpensive(true);
-
-  MaxStoresPerMemcpy  = 4096;
-  MaxStoresPerMemmove = 4096;
-  MaxStoresPerMemset  = 4096;
-
-}
-
-bool
-AMDGPUTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
-    const CallInst &I, unsigned Intrinsic) const {
-  return false;
-}
-
-// The backend supports 32 and 64 bit floating point immediates
-bool
-AMDGPUTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
-  if (VT.getScalarType().getSimpleVT().SimpleTy == MVT::f32
-      || VT.getScalarType().getSimpleVT().SimpleTy == MVT::f64) {
-    return true;
-  } else {
-    return false;
-  }
-}
-
-bool
-AMDGPUTargetLowering::ShouldShrinkFPConstant(EVT VT) const {
-  if (VT.getScalarType().getSimpleVT().SimpleTy == MVT::f32
-      || VT.getScalarType().getSimpleVT().SimpleTy == MVT::f64) {
-    return false;
-  } else {
-    return true;
-  }
-}
-
-
-// isMaskedValueZeroForTargetNode - Return true if 'Op & Mask' is known to
-// be zero. Op is expected to be a target specific node. Used by DAG
-// combiner.
-
-void
-AMDGPUTargetLowering::computeMaskedBitsForTargetNode(
-    const SDValue Op,
-    APInt &KnownZero,
-    APInt &KnownOne,
-    const SelectionDAG &DAG,
-    unsigned Depth) const {
-  APInt KnownZero2;
-  APInt KnownOne2;
-  KnownZero = KnownOne = APInt(KnownOne.getBitWidth(), 0); // Don't know anything
-  switch (Op.getOpcode()) {
-    default: break;
-    case ISD::SELECT_CC:
-             DAG.ComputeMaskedBits(
-                 Op.getOperand(1),
-                 KnownZero,
-                 KnownOne,
-                 Depth + 1
-                 );
-             DAG.ComputeMaskedBits(
-                 Op.getOperand(0),
-                 KnownZero2,
-                 KnownOne2
-                 );
-             assert((KnownZero & KnownOne) == 0
-                 && "Bits known to be one AND zero?");
-             assert((KnownZero2 & KnownOne2) == 0
-                 && "Bits known to be one AND zero?");
-             // Only known if known in both the LHS and RHS
-             KnownOne &= KnownOne2;
-             KnownZero &= KnownZero2;
-             break;
-  };
-}
-
-//===----------------------------------------------------------------------===//
-//                           Other Lowering Hooks
-//===----------------------------------------------------------------------===//
-
-SDValue
-AMDGPUTargetLowering::LowerSDIV(SDValue Op, SelectionDAG &DAG) const {
-  EVT OVT = Op.getValueType();
-  SDValue DST;
-  if (OVT.getScalarType() == MVT::i64) {
-    DST = LowerSDIV64(Op, DAG);
-  } else if (OVT.getScalarType() == MVT::i32) {
-    DST = LowerSDIV32(Op, DAG);
-  } else if (OVT.getScalarType() == MVT::i16
-      || OVT.getScalarType() == MVT::i8) {
-    DST = LowerSDIV24(Op, DAG);
-  } else {
-    DST = SDValue(Op.getNode(), 0);
-  }
-  return DST;
-}
-
-SDValue
-AMDGPUTargetLowering::LowerSREM(SDValue Op, SelectionDAG &DAG) const {
-  EVT OVT = Op.getValueType();
-  SDValue DST;
-  if (OVT.getScalarType() == MVT::i64) {
-    DST = LowerSREM64(Op, DAG);
-  } else if (OVT.getScalarType() == MVT::i32) {
-    DST = LowerSREM32(Op, DAG);
-  } else if (OVT.getScalarType() == MVT::i16) {
-    DST = LowerSREM16(Op, DAG);
-  } else if (OVT.getScalarType() == MVT::i8) {
-    DST = LowerSREM8(Op, DAG);
-  } else {
-    DST = SDValue(Op.getNode(), 0);
-  }
-  return DST;
-}
-
-SDValue
-AMDGPUTargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const {
-  SDValue Data = Op.getOperand(0);
-  VTSDNode *BaseType = cast<VTSDNode>(Op.getOperand(1));
-  SDLoc DL(Op);
-  EVT DVT = Data.getValueType();
-  EVT BVT = BaseType->getVT();
-  unsigned baseBits = BVT.getScalarType().getSizeInBits();
-  unsigned srcBits = DVT.isSimple() ? DVT.getScalarType().getSizeInBits() : 1;
-  unsigned shiftBits = srcBits - baseBits;
-  if (srcBits < 32) {
-    // If the op is less than 32 bits, then it needs to extend to 32bits
-    // so it can properly keep the upper bits valid.
-    EVT IVT = genIntType(32, DVT.isVector() ? DVT.getVectorNumElements() : 1);
-    Data = DAG.getNode(ISD::ZERO_EXTEND, DL, IVT, Data);
-    shiftBits = 32 - baseBits;
-    DVT = IVT;
-  }
-  SDValue Shift = DAG.getConstant(shiftBits, DVT);
-  // Shift left by 'Shift' bits.
-  Data = DAG.getNode(ISD::SHL, DL, DVT, Data, Shift);
-  // Signed shift Right by 'Shift' bits.
-  Data = DAG.getNode(ISD::SRA, DL, DVT, Data, Shift);
-  if (srcBits < 32) {
-    // Once the sign extension is done, the op needs to be converted to
-    // its original type.
-    Data = DAG.getSExtOrTrunc(Data, DL, Op.getOperand(0).getValueType());
-  }
-  return Data;
-}
-EVT
-AMDGPUTargetLowering::genIntType(uint32_t size, uint32_t numEle) const {
-  int iSize = (size * numEle);
-  int vEle = (iSize >> ((size == 64) ? 6 : 5));
-  if (!vEle) {
-    vEle = 1;
-  }
-  if (size == 64) {
-    if (vEle == 1) {
-      return EVT(MVT::i64);
-    } else {
-      return EVT(MVT::getVectorVT(MVT::i64, vEle));
-    }
-  } else {
-    if (vEle == 1) {
-      return EVT(MVT::i32);
-    } else {
-      return EVT(MVT::getVectorVT(MVT::i32, vEle));
-    }
-  }
-}
-
-SDValue
-AMDGPUTargetLowering::LowerBRCOND(SDValue Op, SelectionDAG &DAG) const {
-  SDValue Chain = Op.getOperand(0);
-  SDValue Cond  = Op.getOperand(1);
-  SDValue Jump  = Op.getOperand(2);
-  SDValue Result;
-  Result = DAG.getNode(
-      AMDGPUISD::BRANCH_COND,
-      SDLoc(Op),
-      Op.getValueType(),
-      Chain, Jump, Cond);
-  return Result;
-}
-
-SDValue
-AMDGPUTargetLowering::LowerSDIV24(SDValue Op, SelectionDAG &DAG) const {
-  SDLoc DL(Op);
-  EVT OVT = Op.getValueType();
-  SDValue LHS = Op.getOperand(0);
-  SDValue RHS = Op.getOperand(1);
-  MVT INTTY;
-  MVT FLTTY;
-  if (!OVT.isVector()) {
-    INTTY = MVT::i32;
-    FLTTY = MVT::f32;
-  } else if (OVT.getVectorNumElements() == 2) {
-    INTTY = MVT::v2i32;
-    FLTTY = MVT::v2f32;
-  } else if (OVT.getVectorNumElements() == 4) {
-    INTTY = MVT::v4i32;
-    FLTTY = MVT::v4f32;
-  }
-  unsigned bitsize = OVT.getScalarType().getSizeInBits();
-  // char|short jq = ia ^ ib;
-  SDValue jq = DAG.getNode(ISD::XOR, DL, OVT, LHS, RHS);
-
-  // jq = jq >> (bitsize - 2)
-  jq = DAG.getNode(ISD::SRA, DL, OVT, jq, DAG.getConstant(bitsize - 2, OVT)); 
-
-  // jq = jq | 0x1
-  jq = DAG.getNode(ISD::OR, DL, OVT, jq, DAG.getConstant(1, OVT));
-
-  // jq = (int)jq
-  jq = DAG.getSExtOrTrunc(jq, DL, INTTY);
-
-  // int ia = (int)LHS;
-  SDValue ia = DAG.getSExtOrTrunc(LHS, DL, INTTY);
-
-  // int ib, (int)RHS;
-  SDValue ib = DAG.getSExtOrTrunc(RHS, DL, INTTY);
-
-  // float fa = (float)ia;
-  SDValue fa = DAG.getNode(ISD::SINT_TO_FP, DL, FLTTY, ia);
-
-  // float fb = (float)ib;
-  SDValue fb = DAG.getNode(ISD::SINT_TO_FP, DL, FLTTY, ib);
-
-  // float fq = native_divide(fa, fb);
-  SDValue fq = DAG.getNode(AMDGPUISD::DIV_INF, DL, FLTTY, fa, fb);
-
-  // fq = trunc(fq);
-  fq = DAG.getNode(ISD::FTRUNC, DL, FLTTY, fq);
-
-  // float fqneg = -fq;
-  SDValue fqneg = DAG.getNode(ISD::FNEG, DL, FLTTY, fq);
-
-  // float fr = mad(fqneg, fb, fa);
-  SDValue fr = DAG.getNode(ISD::FADD, DL, FLTTY,
-      DAG.getNode(ISD::MUL, DL, FLTTY, fqneg, fb), fa);
-
-  // int iq = (int)fq;
-  SDValue iq = DAG.getNode(ISD::FP_TO_SINT, DL, INTTY, fq);
-
-  // fr = fabs(fr);
-  fr = DAG.getNode(ISD::FABS, DL, FLTTY, fr);
-
-  // fb = fabs(fb);
-  fb = DAG.getNode(ISD::FABS, DL, FLTTY, fb);
-
-  // int cv = fr >= fb;
-  SDValue cv;
-  if (INTTY == MVT::i32) {
-    cv = DAG.getSetCC(DL, INTTY, fr, fb, ISD::SETOGE);
-  } else {
-    cv = DAG.getSetCC(DL, INTTY, fr, fb, ISD::SETOGE);
-  }
-  // jq = (cv ? jq : 0);
-  jq = DAG.getNode(ISD::SELECT, DL, OVT, cv, jq, 
-      DAG.getConstant(0, OVT));
-  // dst = iq + jq;
-  iq = DAG.getSExtOrTrunc(iq, DL, OVT);
-  iq = DAG.getNode(ISD::ADD, DL, OVT, iq, jq);
-  return iq;
-}
-
-SDValue
-AMDGPUTargetLowering::LowerSDIV32(SDValue Op, SelectionDAG &DAG) const {
-  SDLoc DL(Op);
-  EVT OVT = Op.getValueType();
-  SDValue LHS = Op.getOperand(0);
-  SDValue RHS = Op.getOperand(1);
-  // The LowerSDIV32 function generates equivalent to the following IL.
-  // mov r0, LHS
-  // mov r1, RHS
-  // ilt r10, r0, 0
-  // ilt r11, r1, 0
-  // iadd r0, r0, r10
-  // iadd r1, r1, r11
-  // ixor r0, r0, r10
-  // ixor r1, r1, r11
-  // udiv r0, r0, r1
-  // ixor r10, r10, r11
-  // iadd r0, r0, r10
-  // ixor DST, r0, r10
-
-  // mov r0, LHS
-  SDValue r0 = LHS;
-
-  // mov r1, RHS
-  SDValue r1 = RHS;
-
-  // ilt r10, r0, 0
-  SDValue r10 = DAG.getSelectCC(DL,
-      r0, DAG.getConstant(0, OVT),
-      DAG.getConstant(-1, MVT::i32),
-      DAG.getConstant(0, MVT::i32),
-      ISD::SETLT);
-
-  // ilt r11, r1, 0
-  SDValue r11 = DAG.getSelectCC(DL,
-      r1, DAG.getConstant(0, OVT),
-      DAG.getConstant(-1, MVT::i32),
-      DAG.getConstant(0, MVT::i32),
-      ISD::SETLT);
-
-  // iadd r0, r0, r10
-  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
-
-  // iadd r1, r1, r11
-  r1 = DAG.getNode(ISD::ADD, DL, OVT, r1, r11);
-
-  // ixor r0, r0, r10
-  r0 = DAG.getNode(ISD::XOR, DL, OVT, r0, r10);
-
-  // ixor r1, r1, r11
-  r1 = DAG.getNode(ISD::XOR, DL, OVT, r1, r11);
-
-  // udiv r0, r0, r1
-  r0 = DAG.getNode(ISD::UDIV, DL, OVT, r0, r1);
-
-  // ixor r10, r10, r11
-  r10 = DAG.getNode(ISD::XOR, DL, OVT, r10, r11);
-
-  // iadd r0, r0, r10
-  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
-
-  // ixor DST, r0, r10
-  SDValue DST = DAG.getNode(ISD::XOR, DL, OVT, r0, r10); 
-  return DST;
-}
-
-SDValue
-AMDGPUTargetLowering::LowerSDIV64(SDValue Op, SelectionDAG &DAG) const {
-  return SDValue(Op.getNode(), 0);
-}
-
-SDValue
-AMDGPUTargetLowering::LowerSREM8(SDValue Op, SelectionDAG &DAG) const {
-  SDLoc DL(Op);
-  EVT OVT = Op.getValueType();
-  MVT INTTY = MVT::i32;
-  if (OVT == MVT::v2i8) {
-    INTTY = MVT::v2i32;
-  } else if (OVT == MVT::v4i8) {
-    INTTY = MVT::v4i32;
-  }
-  SDValue LHS = DAG.getSExtOrTrunc(Op.getOperand(0), DL, INTTY);
-  SDValue RHS = DAG.getSExtOrTrunc(Op.getOperand(1), DL, INTTY);
-  LHS = DAG.getNode(ISD::SREM, DL, INTTY, LHS, RHS);
-  LHS = DAG.getSExtOrTrunc(LHS, DL, OVT);
-  return LHS;
-}
-
-SDValue
-AMDGPUTargetLowering::LowerSREM16(SDValue Op, SelectionDAG &DAG) const {
-  SDLoc DL(Op);
-  EVT OVT = Op.getValueType();
-  MVT INTTY = MVT::i32;
-  if (OVT == MVT::v2i16) {
-    INTTY = MVT::v2i32;
-  } else if (OVT == MVT::v4i16) {
-    INTTY = MVT::v4i32;
-  }
-  SDValue LHS = DAG.getSExtOrTrunc(Op.getOperand(0), DL, INTTY);
-  SDValue RHS = DAG.getSExtOrTrunc(Op.getOperand(1), DL, INTTY);
-  LHS = DAG.getNode(ISD::SREM, DL, INTTY, LHS, RHS);
-  LHS = DAG.getSExtOrTrunc(LHS, DL, OVT);
-  return LHS;
-}
-
-SDValue
-AMDGPUTargetLowering::LowerSREM32(SDValue Op, SelectionDAG &DAG) const {
-  SDLoc DL(Op);
-  EVT OVT = Op.getValueType();
-  SDValue LHS = Op.getOperand(0);
-  SDValue RHS = Op.getOperand(1);
-  // The LowerSREM32 function generates equivalent to the following IL.
-  // mov r0, LHS
-  // mov r1, RHS
-  // ilt r10, r0, 0
-  // ilt r11, r1, 0
-  // iadd r0, r0, r10
-  // iadd r1, r1, r11
-  // ixor r0, r0, r10
-  // ixor r1, r1, r11
-  // udiv r20, r0, r1
-  // umul r20, r20, r1
-  // sub r0, r0, r20
-  // iadd r0, r0, r10
-  // ixor DST, r0, r10
-
-  // mov r0, LHS
-  SDValue r0 = LHS;
-
-  // mov r1, RHS
-  SDValue r1 = RHS;
-
-  // ilt r10, r0, 0
-  SDValue r10 = DAG.getSetCC(DL, OVT, r0, DAG.getConstant(0, OVT), ISD::SETLT);
-
-  // ilt r11, r1, 0
-  SDValue r11 = DAG.getSetCC(DL, OVT, r1, DAG.getConstant(0, OVT), ISD::SETLT);
-
-  // iadd r0, r0, r10
-  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
-
-  // iadd r1, r1, r11
-  r1 = DAG.getNode(ISD::ADD, DL, OVT, r1, r11);
-
-  // ixor r0, r0, r10
-  r0 = DAG.getNode(ISD::XOR, DL, OVT, r0, r10);
-
-  // ixor r1, r1, r11
-  r1 = DAG.getNode(ISD::XOR, DL, OVT, r1, r11);
-
-  // udiv r20, r0, r1
-  SDValue r20 = DAG.getNode(ISD::UREM, DL, OVT, r0, r1);
-
-  // umul r20, r20, r1
-  r20 = DAG.getNode(AMDGPUISD::UMUL, DL, OVT, r20, r1);
-
-  // sub r0, r0, r20
-  r0 = DAG.getNode(ISD::SUB, DL, OVT, r0, r20);
-
-  // iadd r0, r0, r10
-  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
-
-  // ixor DST, r0, r10
-  SDValue DST = DAG.getNode(ISD::XOR, DL, OVT, r0, r10); 
-  return DST;
-}
-
-SDValue
-AMDGPUTargetLowering::LowerSREM64(SDValue Op, SelectionDAG &DAG) const {
-  return SDValue(Op.getNode(), 0);
-}
diff --git a/contrib/llvm/lib/Target/R600/AMDILInstrInfo.td b/contrib/llvm/lib/Target/R600/AMDILInstrInfo.td
deleted file mode 100644
index 0f0c88d..0000000
--- a/contrib/llvm/lib/Target/R600/AMDILInstrInfo.td
+++ /dev/null
@@ -1,150 +0,0 @@
-//===------------ AMDILInstrInfo.td - AMDIL Target ------*-tablegen-*------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//==-----------------------------------------------------------------------===//
-//
-// This file describes the AMDIL instructions in TableGen format.
-//
-//===----------------------------------------------------------------------===//
-//===--------------------------------------------------------------------===//
-// Custom Operands
-//===--------------------------------------------------------------------===//
-def brtarget   : Operand<OtherVT>;
-
-//===--------------------------------------------------------------------===//
-// Custom Selection DAG Type Profiles
-//===--------------------------------------------------------------------===//
-//===----------------------------------------------------------------------===//
-// Generic Profile Types
-//===----------------------------------------------------------------------===//
-
-def SDTIL_GenBinaryOp : SDTypeProfile<1, 2, [
-    SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>
-    ]>;
-def SDTIL_GenTernaryOp : SDTypeProfile<1, 3, [
-    SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>, SDTCisSameAs<2, 3>
-    ]>;
-def SDTIL_GenVecBuild : SDTypeProfile<1, 1, [
-    SDTCisEltOfVec<1, 0>
-    ]>;
-
-//===----------------------------------------------------------------------===//
-// Flow Control Profile Types
-//===----------------------------------------------------------------------===//
-// Branch instruction where second and third are basic blocks
-def SDTIL_BRCond : SDTypeProfile<0, 2, [
-    SDTCisVT<0, OtherVT>
-    ]>;
-
-//===--------------------------------------------------------------------===//
-// Custom Selection DAG Nodes
-//===--------------------------------------------------------------------===//
-//===----------------------------------------------------------------------===//
-// Flow Control DAG Nodes
-//===----------------------------------------------------------------------===//
-def IL_brcond      : SDNode<"AMDGPUISD::BRANCH_COND", SDTIL_BRCond, [SDNPHasChain]>;
-
-//===----------------------------------------------------------------------===//
-// Call/Return DAG Nodes
-//===----------------------------------------------------------------------===//
-def IL_retflag       : SDNode<"AMDGPUISD::RET_FLAG", SDTNone,
-    [SDNPHasChain, SDNPOptInGlue]>;
-
-//===--------------------------------------------------------------------===//
-// Instructions
-//===--------------------------------------------------------------------===//
-// Floating point math functions
-def IL_div_inf      : SDNode<"AMDGPUISD::DIV_INF", SDTIL_GenBinaryOp>;
-
-//===----------------------------------------------------------------------===//
-// Integer functions
-//===----------------------------------------------------------------------===//
-def IL_umul        : SDNode<"AMDGPUISD::UMUL"    , SDTIntBinOp,
-    [SDNPCommutative, SDNPAssociative]>;
-
-//===--------------------------------------------------------------------===//
-// Custom Pattern DAG Nodes
-//===--------------------------------------------------------------------===//
-def global_store : PatFrag<(ops node:$val, node:$ptr),
-    (store node:$val, node:$ptr), [{
-        return isGlobalStore(dyn_cast<StoreSDNode>(N));
-}]>;
-
-//===----------------------------------------------------------------------===//
-// Load pattern fragments
-//===----------------------------------------------------------------------===//
-// Global address space loads
-def global_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
-    return isGlobalLoad(dyn_cast<LoadSDNode>(N));
-}]>;
-// Constant address space loads
-def constant_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
-    return isConstantLoad(dyn_cast<LoadSDNode>(N), -1);
-}]>;
-
-//===----------------------------------------------------------------------===//
-// Complex addressing mode patterns
-//===----------------------------------------------------------------------===//
-def ADDR : ComplexPattern<i32, 2, "SelectADDR", [], []>;
-def ADDRF : ComplexPattern<i32, 2, "SelectADDR", [frameindex], []>;
-def ADDR64 : ComplexPattern<i64, 2, "SelectADDR64", [], []>;
-def ADDR64F : ComplexPattern<i64, 2, "SelectADDR64", [frameindex], []>;
-
-//===----------------------------------------------------------------------===//
-// Instruction format classes
-//===----------------------------------------------------------------------===//
-class ILFormat<dag outs, dag ins, string asmstr, list<dag> pattern>
-: Instruction {
-
-     let Namespace = "AMDGPU";
-     dag OutOperandList = outs;
-     dag InOperandList = ins;
-     let Pattern = pattern;
-     let AsmString = !strconcat(asmstr, "\n");
-     let isPseudo = 1;
-     let Itinerary = NullALU;
-     bit hasIEEEFlag = 0;
-     bit hasZeroOpFlag = 0;
-     let mayLoad = 0;
-     let mayStore = 0;
-     let hasSideEffects = 0;
-}
-
-//===--------------------------------------------------------------------===//
-// Multiclass Instruction formats
-//===--------------------------------------------------------------------===//
-// Multiclass that handles branch instructions
-multiclass BranchConditional<SDNode Op, RegisterClass rci, RegisterClass rcf> {
-    def _i32 : ILFormat<(outs),
-  (ins brtarget:$target, rci:$src0),
-        "; i32 Pseudo branch instruction",
-  [(Op bb:$target, (i32 rci:$src0))]>;
-    def _f32 : ILFormat<(outs),
-  (ins brtarget:$target, rcf:$src0),
-        "; f32 Pseudo branch instruction",
-  [(Op bb:$target, (f32 rcf:$src0))]>;
-}
-
-// Only scalar types should generate flow control
-multiclass BranchInstr<string name> {
-  def _i32 : ILFormat<(outs), (ins GPRI32:$src),
-      !strconcat(name, " $src"), []>;
-  def _f32 : ILFormat<(outs), (ins GPRF32:$src),
-      !strconcat(name, " $src"), []>;
-}
-// Only scalar types should generate flow control
-multiclass BranchInstr2<string name> {
-  def _i32 : ILFormat<(outs), (ins GPRI32:$src0, GPRI32:$src1),
-      !strconcat(name, " $src0, $src1"), []>;
-  def _f32 : ILFormat<(outs), (ins GPRF32:$src0, GPRF32:$src1),
-      !strconcat(name, " $src0, $src1"), []>;
-}
-
-//===--------------------------------------------------------------------===//
-// Intrinsics support
-//===--------------------------------------------------------------------===//
-include "AMDILIntrinsics.td"
diff --git a/contrib/llvm/lib/Target/R600/AMDILIntrinsics.td b/contrib/llvm/lib/Target/R600/AMDILIntrinsics.td
deleted file mode 100644
index 6ec3559..0000000
--- a/contrib/llvm/lib/Target/R600/AMDILIntrinsics.td
+++ /dev/null
@@ -1,232 +0,0 @@
-//===- AMDILIntrinsics.td - Defines AMDIL Intrinscs -*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//==-----------------------------------------------------------------------===//
-//
-// This file defines all of the amdil-specific intrinsics
-//
-//===---------------------------------------------------------------===//
-//===--------------------------------------------------------------------===//
-// Intrinsic classes
-// Generic versions of the above classes but for Target specific intrinsics
-// instead of SDNode patterns.
-//===--------------------------------------------------------------------===//
-let TargetPrefix = "AMDIL", isTarget = 1 in {
-     class VoidIntLong :
-          Intrinsic<[llvm_i64_ty], [], []>;
-     class VoidIntInt :
-          Intrinsic<[llvm_i32_ty], [], []>;
-     class VoidIntBool :
-          Intrinsic<[llvm_i32_ty], [], []>;
-     class UnaryIntInt :
-          Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>], [IntrNoMem]>;
-     class UnaryIntFloat :
-          Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
-     class ConvertIntFTOI :
-          Intrinsic<[llvm_anyint_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
-     class ConvertIntITOF :
-          Intrinsic<[llvm_anyfloat_ty], [llvm_anyint_ty], [IntrNoMem]>;
-     class UnaryIntNoRetInt :
-          Intrinsic<[], [llvm_anyint_ty], []>;
-     class UnaryIntNoRetFloat :
-          Intrinsic<[], [llvm_anyfloat_ty], []>;
-     class BinaryIntInt :
-          Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
-     class BinaryIntFloat :
-          Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
-     class BinaryIntNoRetInt :
-          Intrinsic<[], [llvm_anyint_ty, LLVMMatchType<0>], []>;
-     class BinaryIntNoRetFloat :
-          Intrinsic<[], [llvm_anyfloat_ty, LLVMMatchType<0>], []>;
-     class TernaryIntInt :
-          Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>,
-          LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
-     class TernaryIntFloat :
-          Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>,
-          LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
-     class QuaternaryIntInt :
-          Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>,
-          LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
-     class UnaryAtomicInt :
-          Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty], [IntrReadWriteArgMem]>;
-     class BinaryAtomicInt :
-          Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty, llvm_i32_ty], [IntrReadWriteArgMem]>;
-     class TernaryAtomicInt :
-          Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty]>;
-     class UnaryAtomicIntNoRet :
-          Intrinsic<[], [llvm_ptr_ty, llvm_i32_ty], [IntrReadWriteArgMem]>;
-     class BinaryAtomicIntNoRet :
-          Intrinsic<[], [llvm_ptr_ty, llvm_i32_ty, llvm_i32_ty], [IntrReadWriteArgMem]>;
-     class TernaryAtomicIntNoRet :
-          Intrinsic<[], [llvm_ptr_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrReadWriteArgMem]>;
-}
-
-let TargetPrefix = "AMDIL", isTarget = 1 in {
-  def int_AMDIL_abs : GCCBuiltin<"__amdil_abs">, UnaryIntInt;
-
-  def int_AMDIL_bit_extract_i32 : GCCBuiltin<"__amdil_ibit_extract">,
-          TernaryIntInt;
-  def int_AMDIL_bit_extract_u32 : GCCBuiltin<"__amdil_ubit_extract">,
-          TernaryIntInt;
-  def int_AMDIL_bit_reverse_u32 : GCCBuiltin<"__amdil_ubit_reverse">,
-          UnaryIntInt;
-  def int_AMDIL_bit_count_i32 : GCCBuiltin<"__amdil_count_bits">,
-          UnaryIntInt;
-  def int_AMDIL_bit_find_first_lo : GCCBuiltin<"__amdil_ffb_lo">,
-          UnaryIntInt;
-  def int_AMDIL_bit_find_first_hi : GCCBuiltin<"__amdil_ffb_hi">,
-          UnaryIntInt;
-  def int_AMDIL_bit_find_first_sgn : GCCBuiltin<"__amdil_ffb_signed">,
-          UnaryIntInt;
-  def int_AMDIL_media_bitalign : GCCBuiltin<"__amdil_bitalign">,
-                    TernaryIntInt;
-  def int_AMDIL_media_bytealign : GCCBuiltin<"__amdil_bytealign">,
-                    TernaryIntInt;
-  def int_AMDIL_bit_insert_u32 : GCCBuiltin<"__amdil_ubit_insert">,
-                    QuaternaryIntInt;
-  def int_AMDIL_bfi : GCCBuiltin<"__amdil_bfi">,
-      TernaryIntInt;
-  def int_AMDIL_bfm : GCCBuiltin<"__amdil_bfm">,
-      BinaryIntInt;
-  def int_AMDIL_mulhi_i32 : GCCBuiltin<"__amdil_imul_high">,
-          BinaryIntInt;
-  def int_AMDIL_mulhi_u32 : GCCBuiltin<"__amdil_umul_high">,
-          BinaryIntInt;
-  def int_AMDIL_mul24_i32 : GCCBuiltin<"__amdil_imul24">,
-          BinaryIntInt;
-  def int_AMDIL_mul24_u32 : GCCBuiltin<"__amdil_umul24">,
-          BinaryIntInt;
-  def int_AMDIL_mulhi24_i32 : GCCBuiltin<"__amdil_imul24_high">,
-          BinaryIntInt;
-  def int_AMDIL_mulhi24_u32 : GCCBuiltin<"__amdil_umul24_high">,
-          BinaryIntInt;
-  def int_AMDIL_carry_i32 : GCCBuiltin<"__amdil_carry">,
-          BinaryIntInt;
-  def int_AMDIL_borrow_i32 : GCCBuiltin<"__amdil_borrow">,
-          BinaryIntInt;
-  def int_AMDIL_min_i32 : GCCBuiltin<"__amdil_imin">,
-          BinaryIntInt;
-  def int_AMDIL_min_u32 : GCCBuiltin<"__amdil_umin">,
-          BinaryIntInt;
-  def int_AMDIL_min     : GCCBuiltin<"__amdil_min">,
-          BinaryIntFloat;
-  def int_AMDIL_max_i32 : GCCBuiltin<"__amdil_imax">,
-          BinaryIntInt;
-  def int_AMDIL_max_u32 : GCCBuiltin<"__amdil_umax">,
-          BinaryIntInt;
-  def int_AMDIL_max     : GCCBuiltin<"__amdil_max">,
-          BinaryIntFloat;
-  def int_AMDIL_media_lerp_u4 : GCCBuiltin<"__amdil_u4lerp">,
-          TernaryIntInt;
-  def int_AMDIL_media_sad : GCCBuiltin<"__amdil_sad">,
-          TernaryIntInt;
-  def int_AMDIL_media_sad_hi : GCCBuiltin<"__amdil_sadhi">,
-          TernaryIntInt;
-  def int_AMDIL_fraction : GCCBuiltin<"__amdil_fraction">,
-          UnaryIntFloat;
-  def int_AMDIL_clamp : GCCBuiltin<"__amdil_clamp">,
-          TernaryIntFloat;
-  def int_AMDIL_pireduce : GCCBuiltin<"__amdil_pireduce">,
-          UnaryIntFloat;
-  def int_AMDIL_round_nearest : GCCBuiltin<"__amdil_round_nearest">,
-          UnaryIntFloat;
-  def int_AMDIL_round_neginf : GCCBuiltin<"__amdil_round_neginf">,
-          UnaryIntFloat;
-  def int_AMDIL_round_zero : GCCBuiltin<"__amdil_round_zero">,
-          UnaryIntFloat;
-  def int_AMDIL_acos : GCCBuiltin<"__amdil_acos">,
-          UnaryIntFloat;
-  def int_AMDIL_atan : GCCBuiltin<"__amdil_atan">,
-          UnaryIntFloat;
-  def int_AMDIL_asin : GCCBuiltin<"__amdil_asin">,
-          UnaryIntFloat;
-  def int_AMDIL_cos : GCCBuiltin<"__amdil_cos">,
-          UnaryIntFloat;
-  def int_AMDIL_cos_vec : GCCBuiltin<"__amdil_cos_vec">,
-          UnaryIntFloat;
-  def int_AMDIL_tan : GCCBuiltin<"__amdil_tan">,
-          UnaryIntFloat;
-  def int_AMDIL_sin : GCCBuiltin<"__amdil_sin">,
-          UnaryIntFloat;
-  def int_AMDIL_sin_vec : GCCBuiltin<"__amdil_sin_vec">,
-          UnaryIntFloat;
-  def int_AMDIL_pow : GCCBuiltin<"__amdil_pow">, BinaryIntFloat;
-  def int_AMDIL_div : GCCBuiltin<"__amdil_div">, BinaryIntFloat;
-  def int_AMDIL_udiv : GCCBuiltin<"__amdil_udiv">, BinaryIntInt;
-  def int_AMDIL_sqrt: GCCBuiltin<"__amdil_sqrt">,
-          UnaryIntFloat;
-  def int_AMDIL_sqrt_vec: GCCBuiltin<"__amdil_sqrt_vec">,
-          UnaryIntFloat;
-  def int_AMDIL_exp : GCCBuiltin<"__amdil_exp">,
-          UnaryIntFloat;
-  def int_AMDIL_exp_vec : GCCBuiltin<"__amdil_exp_vec">,
-          UnaryIntFloat;
-  def int_AMDIL_exn : GCCBuiltin<"__amdil_exn">,
-          UnaryIntFloat;
-  def int_AMDIL_log_vec : GCCBuiltin<"__amdil_log_vec">,
-          UnaryIntFloat;
-  def int_AMDIL_ln : GCCBuiltin<"__amdil_ln">,
-          UnaryIntFloat;
-  def int_AMDIL_sign: GCCBuiltin<"__amdil_sign">,
-          UnaryIntFloat;
-  def int_AMDIL_fma: GCCBuiltin<"__amdil_fma">,
-          TernaryIntFloat;
-  def int_AMDIL_rsq : GCCBuiltin<"__amdil_rsq">,
-          UnaryIntFloat;
-  def int_AMDIL_rsq_vec : GCCBuiltin<"__amdil_rsq_vec">,
-          UnaryIntFloat;
-  def int_AMDIL_length : GCCBuiltin<"__amdil_length">,
-          UnaryIntFloat;
-  def int_AMDIL_lerp : GCCBuiltin<"__amdil_lerp">,
-          TernaryIntFloat;
-  def int_AMDIL_media_sad4 : GCCBuiltin<"__amdil_sad4">,
-      Intrinsic<[llvm_i32_ty], [llvm_v4i32_ty,
-           llvm_v4i32_ty, llvm_i32_ty], []>;
-
-  def int_AMDIL_frexp_f64 : GCCBuiltin<"__amdil_frexp">,
-        Intrinsic<[llvm_v2i64_ty], [llvm_double_ty], []>;
- def int_AMDIL_ldexp : GCCBuiltin<"__amdil_ldexp">,
-    Intrinsic<[llvm_anyfloat_ty], [llvm_anyfloat_ty, llvm_anyint_ty], []>;
-  def int_AMDIL_drcp : GCCBuiltin<"__amdil_rcp">,
-      Intrinsic<[llvm_double_ty], [llvm_double_ty], []>;
-  def int_AMDIL_convert_f16_f32 : GCCBuiltin<"__amdil_half_to_float">,
-      ConvertIntITOF;
-  def int_AMDIL_convert_f32_f16 : GCCBuiltin<"__amdil_float_to_half">,
-      ConvertIntFTOI;
-  def int_AMDIL_convert_f32_i32_rpi : GCCBuiltin<"__amdil_float_to_int_rpi">,
-      ConvertIntFTOI;
-  def int_AMDIL_convert_f32_i32_flr : GCCBuiltin<"__amdil_float_to_int_flr">,
-      ConvertIntFTOI;
-  def int_AMDIL_convert_f32_f16_near : GCCBuiltin<"__amdil_float_to_half_near">,
-      ConvertIntFTOI;
-  def int_AMDIL_convert_f32_f16_neg_inf : GCCBuiltin<"__amdil_float_to_half_neg_inf">,
-      ConvertIntFTOI;
-  def int_AMDIL_convert_f32_f16_plus_inf : GCCBuiltin<"__amdil_float_to_half_plus_inf">,
-      ConvertIntFTOI;
- def int_AMDIL_media_convert_f2v4u8 : GCCBuiltin<"__amdil_f_2_u4">,
-      Intrinsic<[llvm_i32_ty], [llvm_v4f32_ty], []>;
-  def int_AMDIL_media_unpack_byte_0 : GCCBuiltin<"__amdil_unpack_0">,
-      ConvertIntITOF;
-  def int_AMDIL_media_unpack_byte_1 : GCCBuiltin<"__amdil_unpack_1">,
-      ConvertIntITOF;
-  def int_AMDIL_media_unpack_byte_2 : GCCBuiltin<"__amdil_unpack_2">,
-      ConvertIntITOF;
-  def int_AMDIL_media_unpack_byte_3 : GCCBuiltin<"__amdil_unpack_3">,
-      ConvertIntITOF;
-  def int_AMDIL_dp2_add : GCCBuiltin<"__amdil_dp2_add">,
-        Intrinsic<[llvm_float_ty], [llvm_v2f32_ty,
-          llvm_v2f32_ty, llvm_float_ty], []>;
-  def int_AMDIL_dp2 : GCCBuiltin<"__amdil_dp2">,
-        Intrinsic<[llvm_float_ty], [llvm_v2f32_ty,
-          llvm_v2f32_ty], []>;
-  def int_AMDIL_dp3 : GCCBuiltin<"__amdil_dp3">,
-        Intrinsic<[llvm_float_ty], [llvm_v4f32_ty,
-          llvm_v4f32_ty], []>;
-  def int_AMDIL_dp4 : GCCBuiltin<"__amdil_dp4">,
-        Intrinsic<[llvm_float_ty], [llvm_v4f32_ty,
-          llvm_v4f32_ty], []>;
-}
diff --git a/contrib/llvm/lib/Target/R600/AMDILRegisterInfo.td b/contrib/llvm/lib/Target/R600/AMDILRegisterInfo.td
deleted file mode 100644
index b9d0334..0000000
--- a/contrib/llvm/lib/Target/R600/AMDILRegisterInfo.td
+++ /dev/null
@@ -1,107 +0,0 @@
-//===- AMDILRegisterInfo.td - AMDIL Register defs ----------*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//==-----------------------------------------------------------------------===//
-//
-//  Declarations that describe the AMDIL register file
-//
-//===----------------------------------------------------------------------===//
-
-class AMDILReg<bits<16> num, string n> : Register<n> {
-  field bits<16> Value;
-  let Value = num;
-  let Namespace = "AMDGPU";
-}
-
-// We will start with 8 registers for each class before expanding to more
-// Since the swizzle is added based on the register class, we can leave it
-// off here and just specify different registers for different register classes
-def R1 : AMDILReg<1, "r1">, DwarfRegNum<[1]>;
-def R2 : AMDILReg<2, "r2">, DwarfRegNum<[2]>;
-def R3 : AMDILReg<3, "r3">, DwarfRegNum<[3]>;
-def R4 : AMDILReg<4, "r4">, DwarfRegNum<[4]>;
-def R5 : AMDILReg<5, "r5">, DwarfRegNum<[5]>;
-def R6 : AMDILReg<6, "r6">, DwarfRegNum<[6]>;
-def R7 : AMDILReg<7, "r7">, DwarfRegNum<[7]>;
-def R8 : AMDILReg<8, "r8">, DwarfRegNum<[8]>;
-def R9 : AMDILReg<9, "r9">, DwarfRegNum<[9]>;
-def R10 : AMDILReg<10, "r10">, DwarfRegNum<[10]>;
-def R11 : AMDILReg<11, "r11">, DwarfRegNum<[11]>;
-def R12 : AMDILReg<12, "r12">, DwarfRegNum<[12]>;
-def R13 : AMDILReg<13, "r13">, DwarfRegNum<[13]>;
-def R14 : AMDILReg<14, "r14">, DwarfRegNum<[14]>;
-def R15 : AMDILReg<15, "r15">, DwarfRegNum<[15]>;
-def R16 : AMDILReg<16, "r16">, DwarfRegNum<[16]>;
-def R17 : AMDILReg<17, "r17">, DwarfRegNum<[17]>;
-def R18 : AMDILReg<18, "r18">, DwarfRegNum<[18]>;
-def R19 : AMDILReg<19, "r19">, DwarfRegNum<[19]>;
-def R20 : AMDILReg<20, "r20">, DwarfRegNum<[20]>;
-
-// All registers between 1000 and 1024 are reserved and cannot be used
-// unless commented in this section
-// r1021-r1025 are used to dynamically calculate the local/group/thread/region/region_local ID's
-// r1020 is used to hold the frame index for local arrays
-// r1019 is used to hold the dynamic stack allocation pointer
-// r1018 is used as a temporary register for handwritten code
-// r1017 is used as a temporary register for handwritten code
-// r1016 is used as a temporary register for load/store code
-// r1015 is used as a temporary register for data segment offset
-// r1014 is used as a temporary register for store code
-// r1013 is used as the section data pointer register
-// r1012-r1010 and r1001-r1008 are used for temporary I/O registers
-// r1009 is used as the frame pointer register
-// r999 is used as the mem register.
-// r998 is used as the return address register.
-//def R1025 : AMDILReg<1025, "r1025">, DwarfRegNum<[1025]>;
-//def R1024 : AMDILReg<1024, "r1024">, DwarfRegNum<[1024]>;
-//def R1023 : AMDILReg<1023, "r1023">, DwarfRegNum<[1023]>;
-//def R1022 : AMDILReg<1022, "r1022">, DwarfRegNum<[1022]>;
-//def R1021 : AMDILReg<1021, "r1021">, DwarfRegNum<[1021]>;
-//def R1020 : AMDILReg<1020, "r1020">, DwarfRegNum<[1020]>;
-def SP : AMDILReg<1019, "r1019">, DwarfRegNum<[1019]>;
-def T1 : AMDILReg<1018, "r1018">, DwarfRegNum<[1018]>;
-def T2 : AMDILReg<1017, "r1017">, DwarfRegNum<[1017]>;
-def T3 : AMDILReg<1016, "r1016">, DwarfRegNum<[1016]>;
-def T4 : AMDILReg<1015, "r1015">, DwarfRegNum<[1015]>;
-def T5 : AMDILReg<1014, "r1014">, DwarfRegNum<[1014]>;
-def SDP : AMDILReg<1013, "r1013">, DwarfRegNum<[1013]>;
-def R1012: AMDILReg<1012, "r1012">, DwarfRegNum<[1012]>;
-def R1011: AMDILReg<1011, "r1011">, DwarfRegNum<[1011]>;
-def R1010: AMDILReg<1010, "r1010">, DwarfRegNum<[1010]>;
-def DFP : AMDILReg<1009, "r1009">, DwarfRegNum<[1009]>;
-def R1008: AMDILReg<1008, "r1008">, DwarfRegNum<[1008]>;
-def R1007: AMDILReg<1007, "r1007">, DwarfRegNum<[1007]>;
-def R1006: AMDILReg<1006, "r1006">, DwarfRegNum<[1006]>;
-def R1005: AMDILReg<1005, "r1005">, DwarfRegNum<[1005]>;
-def R1004: AMDILReg<1004, "r1004">, DwarfRegNum<[1004]>;
-def R1003: AMDILReg<1003, "r1003">, DwarfRegNum<[1003]>;
-def R1002: AMDILReg<1002, "r1002">, DwarfRegNum<[1002]>;
-def R1001: AMDILReg<1001, "r1001">, DwarfRegNum<[1001]>;
-def MEM : AMDILReg<999, "mem">, DwarfRegNum<[999]>;
-def RA : AMDILReg<998, "r998">, DwarfRegNum<[998]>;
-def FP : AMDILReg<997, "r997">, DwarfRegNum<[997]>;
-def GPRI16 : RegisterClass<"AMDGPU", [i16], 16,
-  (add (sequence "R%u", 1, 20), RA, SP, T1, T2, T3, T4, T5, SDP, R1010, R1011, R1001, R1002, R1003, R1004, R1005, R1006, R1007, R1008, MEM, R1012)> {
-        let AltOrders = [(add (sequence "R%u", 1, 20))];
-        let AltOrderSelect = [{
-          return 1;
-        }];
-    }
-def GPRI32 : RegisterClass<"AMDGPU", [i32], 32,
-  (add (sequence "R%u", 1, 20), RA, SP, T1, T2, T3, T4, T5, SDP, R1010, R1011, R1001, R1002, R1003, R1004, R1005, R1006, R1007, R1008, MEM, R1012)> {
-        let AltOrders = [(add (sequence "R%u", 1, 20))];
-        let AltOrderSelect = [{
-          return 1;
-        }];
-    }
-def GPRF32 : RegisterClass<"AMDGPU", [f32], 32,
-  (add (sequence "R%u", 1, 20), RA, SP, T1, T2, T3, T4, T5, SDP, R1010, R1011, R1001, R1002, R1003, R1004, R1005, R1006, R1007, R1008, MEM, R1012)> {
-        let AltOrders = [(add (sequence "R%u", 1, 20))];
-        let AltOrderSelect = [{
-          return 1;
-        }];
-    }
diff --git a/contrib/llvm/lib/Target/R600/CaymanInstructions.td b/contrib/llvm/lib/Target/R600/CaymanInstructions.td
new file mode 100644
index 0000000..2630345
--- /dev/null
+++ b/contrib/llvm/lib/Target/R600/CaymanInstructions.td
@@ -0,0 +1,224 @@
+//===-- CaymanInstructions.td - CM Instruction defs  -------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// TableGen definitions for instructions which are available only on Cayman
+// family GPUs.
+//
+//===----------------------------------------------------------------------===//
+
+def isCayman : Predicate<"Subtarget.hasCaymanISA()">;
+
+//===----------------------------------------------------------------------===//
+// Cayman Instructions
+//===----------------------------------------------------------------------===//
+
+let Predicates = [isCayman] in {
+
+def MULADD_INT24_cm : R600_3OP <0x08, "MULADD_INT24",
+  [(set i32:$dst, (AMDGPUmad_i24 i32:$src0, i32:$src1, i32:$src2))], VecALU
+>;
+def MUL_INT24_cm : R600_2OP <0x5B, "MUL_INT24",
+  [(set i32:$dst, (AMDGPUmul_i24 i32:$src0, i32:$src1))], VecALU
+>;
+
+def : IMad24Pat<MULADD_INT24_cm>;
+
+let isVector = 1 in {
+
+def RECIP_IEEE_cm : RECIP_IEEE_Common<0x86>;
+
+def MULLO_INT_cm : MULLO_INT_Common<0x8F>;
+def MULHI_INT_cm : MULHI_INT_Common<0x90>;
+def MULLO_UINT_cm : MULLO_UINT_Common<0x91>;
+def MULHI_UINT_cm : MULHI_UINT_Common<0x92>;
+def RECIPSQRT_CLAMPED_cm : RECIPSQRT_CLAMPED_Common<0x87>;
+def EXP_IEEE_cm : EXP_IEEE_Common<0x81>;
+def LOG_IEEE_cm : LOG_IEEE_Common<0x83>;
+def RECIP_CLAMPED_cm : RECIP_CLAMPED_Common<0x84>;
+def RECIPSQRT_IEEE_cm : RECIPSQRT_IEEE_Common<0x89>;
+def SIN_cm : SIN_Common<0x8D>;
+def COS_cm : COS_Common<0x8E>;
+} // End isVector = 1
+
+def : POW_Common <LOG_IEEE_cm, EXP_IEEE_cm, MUL>;
+
+defm DIV_cm : DIV_Common<RECIP_IEEE_cm>;
+defm : Expand24UBitOps<MULLO_UINT_cm, ADD_INT>;
+
+// RECIP_UINT emulation for Cayman
+// The multiplication scales from [0,1] to the unsigned integer range
+def : Pat <
+  (AMDGPUurecip i32:$src0),
+  (FLT_TO_UINT_eg (MUL_IEEE (RECIP_IEEE_cm (UINT_TO_FLT_eg $src0)),
+                            (MOV_IMM_I32 CONST.FP_UINT_MAX_PLUS_1)))
+>;
+
+  def CF_END_CM : CF_CLAUSE_EG<32, (ins), "CF_END"> {
+    let ADDR = 0;
+    let POP_COUNT = 0;
+    let COUNT = 0;
+  }
+
+
+def : Pat<(fsqrt f32:$src), (MUL R600_Reg32:$src, (RECIPSQRT_CLAMPED_cm $src))>;
+
+class RAT_STORE_DWORD <RegisterClass rc, ValueType vt, bits<4> mask> :
+  CF_MEM_RAT_CACHELESS <0x14, 0, mask,
+                        (ins rc:$rw_gpr, R600_TReg32_X:$index_gpr),
+                        "STORE_DWORD $rw_gpr, $index_gpr",
+                        [(global_store vt:$rw_gpr, i32:$index_gpr)]> {
+  let eop = 0; // This bit is not used on Cayman.
+}
+
+def RAT_STORE_DWORD32 : RAT_STORE_DWORD <R600_TReg32_X, i32, 0x1>;
+def RAT_STORE_DWORD64 : RAT_STORE_DWORD <R600_Reg64, v2i32, 0x3>;
+def RAT_STORE_DWORD128 : RAT_STORE_DWORD <R600_Reg128, v4i32, 0xf>;
+
+class VTX_READ_cm <string name, bits<8> buffer_id, dag outs, list<dag> pattern>
+    : VTX_WORD0_cm, VTX_READ<name, buffer_id, outs, pattern> {
+
+  // Static fields
+  let VC_INST = 0;
+  let FETCH_TYPE = 2;
+  let FETCH_WHOLE_QUAD = 0;
+  let BUFFER_ID = buffer_id;
+  let SRC_REL = 0;
+  // XXX: We can infer this field based on the SRC_GPR.  This would allow us
+  // to store vertex addresses in any channel, not just X.
+  let SRC_SEL_X = 0;
+  let SRC_SEL_Y = 0;
+  let STRUCTURED_READ = 0;
+  let LDS_REQ = 0;
+  let COALESCED_READ = 0;
+
+  let Inst{31-0} = Word0;
+}
+
+class VTX_READ_8_cm <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_cm <"VTX_READ_8 $dst_gpr, $src_gpr", buffer_id,
+                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+
+  let DST_SEL_X = 0;
+  let DST_SEL_Y = 7;   // Masked
+  let DST_SEL_Z = 7;   // Masked
+  let DST_SEL_W = 7;   // Masked
+  let DATA_FORMAT = 1; // FMT_8
+}
+
+class VTX_READ_16_cm <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_cm <"VTX_READ_16 $dst_gpr, $src_gpr", buffer_id,
+                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+  let DST_SEL_X = 0;
+  let DST_SEL_Y = 7;   // Masked
+  let DST_SEL_Z = 7;   // Masked
+  let DST_SEL_W = 7;   // Masked
+  let DATA_FORMAT = 5; // FMT_16
+
+}
+
+class VTX_READ_32_cm <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_cm <"VTX_READ_32 $dst_gpr, $src_gpr", buffer_id,
+                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+
+  let DST_SEL_X        = 0;
+  let DST_SEL_Y        = 7;   // Masked
+  let DST_SEL_Z        = 7;   // Masked
+  let DST_SEL_W        = 7;   // Masked
+  let DATA_FORMAT      = 0xD; // COLOR_32
+
+  // This is not really necessary, but there were some GPU hangs that appeared
+  // to be caused by ALU instructions in the next instruction group that wrote
+  // to the $src_gpr registers of the VTX_READ.
+  // e.g.
+  // %T3_X<def> = VTX_READ_PARAM_32_eg %T2_X<kill>, 24
+  // %T2_X<def> = MOV %ZERO
+  //Adding this constraint prevents this from happening.
+  let Constraints = "$src_gpr.ptr = $dst_gpr";
+}
+
+class VTX_READ_64_cm <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_cm <"VTX_READ_64 $dst_gpr, $src_gpr", buffer_id,
+                   (outs R600_Reg64:$dst_gpr), pattern> {
+
+  let DST_SEL_X        = 0;
+  let DST_SEL_Y        = 1;
+  let DST_SEL_Z        = 7;
+  let DST_SEL_W        = 7;
+  let DATA_FORMAT      = 0x1D; // COLOR_32_32
+}
+
+class VTX_READ_128_cm <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_cm <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr", buffer_id,
+                   (outs R600_Reg128:$dst_gpr), pattern> {
+
+  let DST_SEL_X        =  0;
+  let DST_SEL_Y        =  1;
+  let DST_SEL_Z        =  2;
+  let DST_SEL_W        =  3;
+  let DATA_FORMAT      =  0x22; // COLOR_32_32_32_32
+
+  // XXX: Need to force VTX_READ_128 instructions to write to the same register
+  // that holds its buffer address to avoid potential hangs.  We can't use
+  // the same constraint as VTX_READ_32_eg, because the $src_gpr.ptr and $dst
+  // registers are different sizes.
+}
+
+//===----------------------------------------------------------------------===//
+// VTX Read from parameter memory space
+//===----------------------------------------------------------------------===//
+def VTX_READ_PARAM_8_cm : VTX_READ_8_cm <0,
+  [(set i32:$dst_gpr, (load_param_exti8 ADDRVTX_READ:$src_gpr))]
+>;
+
+def VTX_READ_PARAM_16_cm : VTX_READ_16_cm <0,
+  [(set i32:$dst_gpr, (load_param_exti16 ADDRVTX_READ:$src_gpr))]
+>;
+
+def VTX_READ_PARAM_32_cm : VTX_READ_32_cm <0,
+  [(set i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
+>;
+
+def VTX_READ_PARAM_64_cm : VTX_READ_64_cm <0,
+  [(set v2i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
+>;
+
+def VTX_READ_PARAM_128_cm : VTX_READ_128_cm <0,
+  [(set v4i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
+>;
+
+//===----------------------------------------------------------------------===//
+// VTX Read from global memory space
+//===----------------------------------------------------------------------===//
+
+// 8-bit reads
+def VTX_READ_GLOBAL_8_cm : VTX_READ_8_cm <1,
+  [(set i32:$dst_gpr, (az_extloadi8_global ADDRVTX_READ:$src_gpr))]
+>;
+
+def VTX_READ_GLOBAL_16_cm : VTX_READ_16_cm <1,
+  [(set i32:$dst_gpr, (az_extloadi16_global ADDRVTX_READ:$src_gpr))]
+>;
+
+// 32-bit reads
+def VTX_READ_GLOBAL_32_cm : VTX_READ_32_cm <1,
+  [(set i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
+>;
+
+// 64-bit reads
+def VTX_READ_GLOBAL_64_cm : VTX_READ_64_cm <1,
+  [(set v2i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
+>;
+
+// 128-bit reads
+def VTX_READ_GLOBAL_128_cm : VTX_READ_128_cm <1,
+  [(set v4i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
+>;
+
+} // End isCayman
+
diff --git a/contrib/llvm/lib/Target/R600/EvergreenInstructions.td b/contrib/llvm/lib/Target/R600/EvergreenInstructions.td
new file mode 100644
index 0000000..484e522
--- /dev/null
+++ b/contrib/llvm/lib/Target/R600/EvergreenInstructions.td
@@ -0,0 +1,609 @@
+//===-- EvergreenInstructions.td - EG Instruction defs  ----*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// TableGen definitions for instructions which are:
+// - Available to Evergreen and newer VLIW4/VLIW5 GPUs
+// - Available only on Evergreen family GPUs.
+//
+//===----------------------------------------------------------------------===//
+
+def isEG : Predicate<
+  "Subtarget.getGeneration() >= AMDGPUSubtarget::EVERGREEN && "
+  "Subtarget.getGeneration() < AMDGPUSubtarget::SOUTHERN_ISLANDS && "
+  "!Subtarget.hasCaymanISA()"
+>;
+
+def isEGorCayman : Predicate<
+  "Subtarget.getGeneration() == AMDGPUSubtarget::EVERGREEN ||"
+  "Subtarget.getGeneration() ==AMDGPUSubtarget::NORTHERN_ISLANDS"
+>;
+
+//===----------------------------------------------------------------------===//
+// Evergreen / Cayman store instructions
+//===----------------------------------------------------------------------===//
+
+let Predicates = [isEGorCayman] in {
+
+class CF_MEM_RAT_CACHELESS <bits<6> rat_inst, bits<4> rat_id, bits<4> mask, dag ins,
+                           string name, list<dag> pattern>
+    : EG_CF_RAT <0x57, rat_inst, rat_id, mask, (outs), ins,
+                 "MEM_RAT_CACHELESS "#name, pattern>;
+
+class CF_MEM_RAT <bits<6> rat_inst, bits<4> rat_id, dag ins, string name,
+                  list<dag> pattern>
+    : EG_CF_RAT <0x56, rat_inst, rat_id, 0xf /* mask */, (outs), ins,
+                 "MEM_RAT "#name, pattern>;
+
+def RAT_MSKOR : CF_MEM_RAT <0x11, 0,
+  (ins R600_Reg128:$rw_gpr, R600_TReg32_X:$index_gpr),
+  "MSKOR $rw_gpr.XW, $index_gpr",
+  [(mskor_global v4i32:$rw_gpr, i32:$index_gpr)]
+> {
+  let eop = 0;
+}
+
+} // End let Predicates = [isEGorCayman]
+
+//===----------------------------------------------------------------------===//
+// Evergreen Only instructions
+//===----------------------------------------------------------------------===//
+
+let Predicates = [isEG] in {
+
+def RECIP_IEEE_eg : RECIP_IEEE_Common<0x86>;
+defm DIV_eg : DIV_Common<RECIP_IEEE_eg>;
+
+def MULLO_INT_eg : MULLO_INT_Common<0x8F>;
+def MULHI_INT_eg : MULHI_INT_Common<0x90>;
+def MULLO_UINT_eg : MULLO_UINT_Common<0x91>;
+def MULHI_UINT_eg : MULHI_UINT_Common<0x92>;
+def RECIP_UINT_eg : RECIP_UINT_Common<0x94>;
+def RECIPSQRT_CLAMPED_eg : RECIPSQRT_CLAMPED_Common<0x87>;
+def EXP_IEEE_eg : EXP_IEEE_Common<0x81>;
+def LOG_IEEE_eg : LOG_IEEE_Common<0x83>;
+def RECIP_CLAMPED_eg : RECIP_CLAMPED_Common<0x84>;
+def RECIPSQRT_IEEE_eg : RECIPSQRT_IEEE_Common<0x89>;
+def SIN_eg : SIN_Common<0x8D>;
+def COS_eg : COS_Common<0x8E>;
+
+def : POW_Common <LOG_IEEE_eg, EXP_IEEE_eg, MUL>;
+def : Pat<(fsqrt f32:$src), (MUL $src, (RECIPSQRT_CLAMPED_eg $src))>;
+
+defm : Expand24IBitOps<MULLO_INT_eg, ADD_INT>;
+
+//===----------------------------------------------------------------------===//
+// Memory read/write instructions
+//===----------------------------------------------------------------------===//
+
+let usesCustomInserter = 1 in {
+
+// 32-bit store
+def RAT_WRITE_CACHELESS_32_eg : CF_MEM_RAT_CACHELESS <0x2, 0, 0x1,
+  (ins R600_TReg32_X:$rw_gpr, R600_TReg32_X:$index_gpr, InstFlag:$eop),
+  "STORE_RAW $rw_gpr, $index_gpr, $eop",
+  [(global_store i32:$rw_gpr, i32:$index_gpr)]
+>;
+
+// 64-bit store
+def RAT_WRITE_CACHELESS_64_eg : CF_MEM_RAT_CACHELESS <0x2, 0, 0x3,
+  (ins R600_Reg64:$rw_gpr, R600_TReg32_X:$index_gpr, InstFlag:$eop),
+  "STORE_RAW $rw_gpr.XY, $index_gpr, $eop",
+  [(global_store v2i32:$rw_gpr, i32:$index_gpr)]
+>;
+
+//128-bit store
+def RAT_WRITE_CACHELESS_128_eg : CF_MEM_RAT_CACHELESS <0x2, 0, 0xf,
+  (ins R600_Reg128:$rw_gpr, R600_TReg32_X:$index_gpr, InstFlag:$eop),
+  "STORE_RAW $rw_gpr.XYZW, $index_gpr, $eop",
+  [(global_store v4i32:$rw_gpr, i32:$index_gpr)]
+>;
+
+} // End usesCustomInserter = 1
+
+class VTX_READ_eg <string name, bits<8> buffer_id, dag outs, list<dag> pattern>
+    : VTX_WORD0_eg, VTX_READ<name, buffer_id, outs, pattern> {
+
+  // Static fields
+  let VC_INST = 0;
+  let FETCH_TYPE = 2;
+  let FETCH_WHOLE_QUAD = 0;
+  let BUFFER_ID = buffer_id;
+  let SRC_REL = 0;
+  // XXX: We can infer this field based on the SRC_GPR.  This would allow us
+  // to store vertex addresses in any channel, not just X.
+  let SRC_SEL_X = 0;
+
+  let Inst{31-0} = Word0;
+}
+
+class VTX_READ_8_eg <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_eg <"VTX_READ_8 $dst_gpr, $src_gpr", buffer_id,
+                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+
+  let MEGA_FETCH_COUNT = 1;
+  let DST_SEL_X = 0;
+  let DST_SEL_Y = 7;   // Masked
+  let DST_SEL_Z = 7;   // Masked
+  let DST_SEL_W = 7;   // Masked
+  let DATA_FORMAT = 1; // FMT_8
+}
+
+class VTX_READ_16_eg <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_eg <"VTX_READ_16 $dst_gpr, $src_gpr", buffer_id,
+                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+  let MEGA_FETCH_COUNT = 2;
+  let DST_SEL_X = 0;
+  let DST_SEL_Y = 7;   // Masked
+  let DST_SEL_Z = 7;   // Masked
+  let DST_SEL_W = 7;   // Masked
+  let DATA_FORMAT = 5; // FMT_16
+
+}
+
+class VTX_READ_32_eg <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_eg <"VTX_READ_32 $dst_gpr, $src_gpr", buffer_id,
+                   (outs R600_TReg32_X:$dst_gpr), pattern> {
+
+  let MEGA_FETCH_COUNT = 4;
+  let DST_SEL_X        = 0;
+  let DST_SEL_Y        = 7;   // Masked
+  let DST_SEL_Z        = 7;   // Masked
+  let DST_SEL_W        = 7;   // Masked
+  let DATA_FORMAT      = 0xD; // COLOR_32
+
+  // This is not really necessary, but there were some GPU hangs that appeared
+  // to be caused by ALU instructions in the next instruction group that wrote
+  // to the $src_gpr registers of the VTX_READ.
+  // e.g.
+  // %T3_X<def> = VTX_READ_PARAM_32_eg %T2_X<kill>, 24
+  // %T2_X<def> = MOV %ZERO
+  //Adding this constraint prevents this from happening.
+  let Constraints = "$src_gpr.ptr = $dst_gpr";
+}
+
+class VTX_READ_64_eg <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_eg <"VTX_READ_64 $dst_gpr.XY, $src_gpr", buffer_id,
+                   (outs R600_Reg64:$dst_gpr), pattern> {
+
+  let MEGA_FETCH_COUNT = 8;
+  let DST_SEL_X        = 0;
+  let DST_SEL_Y        = 1;
+  let DST_SEL_Z        = 7;
+  let DST_SEL_W        = 7;
+  let DATA_FORMAT      = 0x1D; // COLOR_32_32
+}
+
+class VTX_READ_128_eg <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_eg <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr", buffer_id,
+                   (outs R600_Reg128:$dst_gpr), pattern> {
+
+  let MEGA_FETCH_COUNT = 16;
+  let DST_SEL_X        =  0;
+  let DST_SEL_Y        =  1;
+  let DST_SEL_Z        =  2;
+  let DST_SEL_W        =  3;
+  let DATA_FORMAT      =  0x22; // COLOR_32_32_32_32
+
+  // XXX: Need to force VTX_READ_128 instructions to write to the same register
+  // that holds its buffer address to avoid potential hangs.  We can't use
+  // the same constraint as VTX_READ_32_eg, because the $src_gpr.ptr and $dst
+  // registers are different sizes.
+}
+
+//===----------------------------------------------------------------------===//
+// VTX Read from parameter memory space
+//===----------------------------------------------------------------------===//
+
+def VTX_READ_PARAM_8_eg : VTX_READ_8_eg <0,
+  [(set i32:$dst_gpr, (load_param_exti8 ADDRVTX_READ:$src_gpr))]
+>;
+
+def VTX_READ_PARAM_16_eg : VTX_READ_16_eg <0,
+  [(set i32:$dst_gpr, (load_param_exti16 ADDRVTX_READ:$src_gpr))]
+>;
+
+def VTX_READ_PARAM_32_eg : VTX_READ_32_eg <0,
+  [(set i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
+>;
+
+def VTX_READ_PARAM_64_eg : VTX_READ_64_eg <0,
+  [(set v2i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
+>;
+
+def VTX_READ_PARAM_128_eg : VTX_READ_128_eg <0,
+  [(set v4i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
+>;
+
+//===----------------------------------------------------------------------===//
+// VTX Read from global memory space
+//===----------------------------------------------------------------------===//
+
+// 8-bit reads
+def VTX_READ_GLOBAL_8_eg : VTX_READ_8_eg <1,
+  [(set i32:$dst_gpr, (az_extloadi8_global ADDRVTX_READ:$src_gpr))]
+>;
+
+def VTX_READ_GLOBAL_16_eg : VTX_READ_16_eg <1,
+  [(set i32:$dst_gpr, (az_extloadi16_global ADDRVTX_READ:$src_gpr))]
+>;
+
+// 32-bit reads
+def VTX_READ_GLOBAL_32_eg : VTX_READ_32_eg <1,
+  [(set i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
+>;
+
+// 64-bit reads
+def VTX_READ_GLOBAL_64_eg : VTX_READ_64_eg <1,
+  [(set v2i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
+>;
+
+// 128-bit reads
+def VTX_READ_GLOBAL_128_eg : VTX_READ_128_eg <1,
+  [(set v4i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
+>;
+
+} // End Predicates = [isEG]
+
+//===----------------------------------------------------------------------===//
+// Evergreen / Cayman Instructions
+//===----------------------------------------------------------------------===//
+
+let Predicates = [isEGorCayman] in {
+
+// BFE_UINT - bit_extract, an optimization for mask and shift
+// Src0 = Input
+// Src1 = Offset
+// Src2 = Width
+//
+// bit_extract = (Input << (32 - Offset - Width)) >> (32 - Width)
+//
+// Example Usage:
+// (Offset, Width)
+//
+// (0, 8)  = (Input << 24) >> 24 = (Input &  0xff)       >> 0
+// (8, 8)  = (Input << 16) >> 24 = (Input &  0xffff)     >> 8
+// (16, 8) = (Input <<  8) >> 24 = (Input &  0xffffff)   >> 16
+// (24, 8) = (Input <<  0) >> 24 = (Input &  0xffffffff) >> 24
+def BFE_UINT_eg : R600_3OP <0x4, "BFE_UINT",
+  [(set i32:$dst, (AMDGPUbfe_u32 i32:$src0, i32:$src1, i32:$src2))],
+  VecALU
+>;
+
+def BFE_INT_eg : R600_3OP <0x5, "BFE_INT",
+  [(set i32:$dst, (AMDGPUbfe_i32 i32:$src0, i32:$src1, i32:$src2))],
+  VecALU
+>;
+
+// XXX: This pattern is broken, disabling for now.  See comment in
+// AMDGPUInstructions.td for more info.
+//  def : BFEPattern <BFE_UINT_eg>;
+def BFI_INT_eg : R600_3OP <0x06, "BFI_INT",
+  [(set i32:$dst, (AMDGPUbfi i32:$src0, i32:$src1, i32:$src2))],
+  VecALU
+>;
+
+def : Pat<(i32 (sext_inreg i32:$src, i1)),
+  (BFE_INT_eg i32:$src, (i32 ZERO), (i32 ONE_INT))>;
+def : Pat<(i32 (sext_inreg i32:$src, i8)),
+  (BFE_INT_eg i32:$src, (i32 ZERO), (MOV_IMM_I32 8))>;
+def : Pat<(i32 (sext_inreg i32:$src, i16)),
+  (BFE_INT_eg i32:$src, (i32 ZERO), (MOV_IMM_I32 16))>;
+
+defm : BFIPatterns <BFI_INT_eg, MOV_IMM_I32>;
+
+def BFM_INT_eg : R600_2OP <0xA0, "BFM_INT",
+  [(set i32:$dst, (AMDGPUbfm i32:$src0, i32:$src1))],
+  VecALU
+>;
+
+def MULADD_UINT24_eg : R600_3OP <0x10, "MULADD_UINT24",
+  [(set i32:$dst, (AMDGPUmad_u24 i32:$src0, i32:$src1, i32:$src2))], VecALU
+>;
+
+def : UMad24Pat<MULADD_UINT24_eg>;
+
+def BIT_ALIGN_INT_eg : R600_3OP <0xC, "BIT_ALIGN_INT", [], VecALU>;
+def : ROTRPattern <BIT_ALIGN_INT_eg>;
+def MULADD_eg : MULADD_Common<0x14>;
+def MULADD_IEEE_eg : MULADD_IEEE_Common<0x18>;
+def ASHR_eg : ASHR_Common<0x15>;
+def LSHR_eg : LSHR_Common<0x16>;
+def LSHL_eg : LSHL_Common<0x17>;
+def CNDE_eg : CNDE_Common<0x19>;
+def CNDGT_eg : CNDGT_Common<0x1A>;
+def CNDGE_eg : CNDGE_Common<0x1B>;
+def MUL_LIT_eg : MUL_LIT_Common<0x1F>;
+def LOG_CLAMPED_eg : LOG_CLAMPED_Common<0x82>;
+def MUL_UINT24_eg : R600_2OP <0xB5, "MUL_UINT24",
+  [(set i32:$dst, (AMDGPUmul_u24 i32:$src0, i32:$src1))], VecALU
+>;
+def DOT4_eg : DOT4_Common<0xBE>;
+defm CUBE_eg : CUBE_Common<0xC0>;
+
+def BCNT_INT : R600_1OP_Helper <0xAA, "BCNT_INT", ctpop, VecALU>;
+
+def FFBH_UINT : R600_1OP_Helper <0xAB, "FFBH_UINT", ctlz_zero_undef, VecALU>;
+def FFBL_INT : R600_1OP_Helper <0xAC, "FFBL_INT", cttz_zero_undef, VecALU>;
+
+let hasSideEffects = 1 in {
+  def MOVA_INT_eg : R600_1OP <0xCC, "MOVA_INT", [], VecALU>;
+}
+
+def TGSI_LIT_Z_eg : TGSI_LIT_Z_Common<MUL_LIT_eg, LOG_CLAMPED_eg, EXP_IEEE_eg>;
+
+def FLT_TO_INT_eg : FLT_TO_INT_Common<0x50> {
+  let Pattern = [];
+  let Itinerary = AnyALU;
+}
+
+def INT_TO_FLT_eg : INT_TO_FLT_Common<0x9B>;
+
+def FLT_TO_UINT_eg : FLT_TO_UINT_Common<0x9A> {
+  let Pattern = [];
+}
+
+def UINT_TO_FLT_eg : UINT_TO_FLT_Common<0x9C>;
+
+def GROUP_BARRIER : InstR600 <
+    (outs), (ins), "  GROUP_BARRIER", [(int_AMDGPU_barrier_local), (int_AMDGPU_barrier_global)], AnyALU>,
+    R600ALU_Word0,
+    R600ALU_Word1_OP2 <0x54> {
+
+  let dst = 0;
+  let dst_rel = 0;
+  let src0 = 0;
+  let src0_rel = 0;
+  let src0_neg = 0;
+  let src0_abs = 0;
+  let src1 = 0;
+  let src1_rel = 0;
+  let src1_neg = 0;
+  let src1_abs = 0;
+  let write = 0;
+  let omod = 0;
+  let clamp = 0;
+  let last = 1;
+  let bank_swizzle = 0;
+  let pred_sel = 0;
+  let update_exec_mask = 0;
+  let update_pred = 0;
+
+  let Inst{31-0}  = Word0;
+  let Inst{63-32} = Word1;
+
+  let ALUInst = 1;
+}
+
+def : Pat <
+	(int_AMDGPU_barrier_global),
+	(GROUP_BARRIER)
+>;
+
+//===----------------------------------------------------------------------===//
+// LDS Instructions
+//===----------------------------------------------------------------------===//
+class R600_LDS  <bits<6> op, dag outs, dag ins, string asm,
+                 list<dag> pattern = []> :
+
+    InstR600 <outs, ins, asm, pattern, XALU>,
+    R600_ALU_LDS_Word0,
+    R600LDS_Word1 {
+
+  bits<6>  offset = 0;
+  let lds_op = op;
+
+  let Word1{27} = offset{0};
+  let Word1{12} = offset{1};
+  let Word1{28} = offset{2};
+  let Word1{31} = offset{3};
+  let Word0{12} = offset{4};
+  let Word0{25} = offset{5};
+
+
+  let Inst{31-0}  = Word0;
+  let Inst{63-32} = Word1;
+
+  let ALUInst = 1;
+  let HasNativeOperands = 1;
+  let UseNamedOperandTable = 1;
+}
+
+class R600_LDS_1A <bits<6> lds_op, string name, list<dag> pattern> : R600_LDS <
+  lds_op,
+  (outs R600_Reg32:$dst),
+  (ins R600_Reg32:$src0, REL:$src0_rel, SEL:$src0_sel,
+       LAST:$last, R600_Pred:$pred_sel,
+       BANK_SWIZZLE:$bank_swizzle),
+  "  "#name#" $last OQAP, $src0$src0_rel $pred_sel",
+  pattern
+  > {
+
+  let src1 = 0;
+  let src1_rel = 0;
+  let src2 = 0;
+  let src2_rel = 0;
+
+  let usesCustomInserter = 1;
+  let LDS_1A = 1;
+  let DisableEncoding = "$dst";
+}
+
+class R600_LDS_1A1D <bits<6> lds_op, dag outs, string name, list<dag> pattern,
+                     string dst =""> :
+    R600_LDS <
+  lds_op, outs,
+  (ins R600_Reg32:$src0, REL:$src0_rel, SEL:$src0_sel,
+       R600_Reg32:$src1, REL:$src1_rel, SEL:$src1_sel,
+       LAST:$last, R600_Pred:$pred_sel,
+       BANK_SWIZZLE:$bank_swizzle),
+  "  "#name#" $last "#dst#"$src0$src0_rel, $src1$src1_rel, $pred_sel",
+  pattern
+  > {
+
+  field string BaseOp;
+
+  let src2 = 0;
+  let src2_rel = 0;
+  let LDS_1A1D = 1;
+}
+
+class R600_LDS_1A1D_NORET <bits<6> lds_op, string name, list<dag> pattern> :
+    R600_LDS_1A1D <lds_op, (outs), name, pattern> {
+  let BaseOp = name;
+}
+
+class R600_LDS_1A1D_RET <bits<6> lds_op, string name, list<dag> pattern> :
+    R600_LDS_1A1D <lds_op,  (outs R600_Reg32:$dst), name##"_RET", pattern, "OQAP, "> {
+
+  let BaseOp = name;
+  let usesCustomInserter = 1;
+  let DisableEncoding = "$dst";
+}
+
+class R600_LDS_1A2D <bits<6> lds_op, string name, list<dag> pattern> :
+    R600_LDS <
+  lds_op,
+  (outs),
+  (ins R600_Reg32:$src0, REL:$src0_rel, SEL:$src0_sel,
+       R600_Reg32:$src1, REL:$src1_rel, SEL:$src1_sel,
+       R600_Reg32:$src2, REL:$src2_rel, SEL:$src2_sel,
+       LAST:$last, R600_Pred:$pred_sel, BANK_SWIZZLE:$bank_swizzle),
+  "  "#name# "$last $src0$src0_rel, $src1$src1_rel, $src2$src2_rel, $pred_sel",
+  pattern> {
+  let LDS_1A2D = 1;
+}
+
+def LDS_ADD : R600_LDS_1A1D_NORET <0x0, "LDS_ADD", [] >;
+def LDS_SUB : R600_LDS_1A1D_NORET <0x1, "LDS_SUB", [] >;
+def LDS_WRITE : R600_LDS_1A1D_NORET <0xD, "LDS_WRITE",
+  [(local_store (i32 R600_Reg32:$src1), R600_Reg32:$src0)]
+>;
+def LDS_BYTE_WRITE : R600_LDS_1A1D_NORET<0x12, "LDS_BYTE_WRITE",
+  [(truncstorei8_local i32:$src1, i32:$src0)]
+>;
+def LDS_SHORT_WRITE : R600_LDS_1A1D_NORET<0x13, "LDS_SHORT_WRITE",
+  [(truncstorei16_local i32:$src1, i32:$src0)]
+>;
+def LDS_ADD_RET : R600_LDS_1A1D_RET <0x20, "LDS_ADD",
+  [(set i32:$dst, (atomic_load_add_local i32:$src0, i32:$src1))]
+>;
+def LDS_SUB_RET : R600_LDS_1A1D_RET <0x21, "LDS_SUB",
+  [(set i32:$dst, (atomic_load_sub_local i32:$src0, i32:$src1))]
+>;
+def LDS_READ_RET : R600_LDS_1A <0x32, "LDS_READ_RET",
+  [(set (i32 R600_Reg32:$dst), (local_load R600_Reg32:$src0))]
+>;
+def LDS_BYTE_READ_RET : R600_LDS_1A <0x36, "LDS_BYTE_READ_RET",
+  [(set i32:$dst, (sextloadi8_local i32:$src0))]
+>;
+def LDS_UBYTE_READ_RET : R600_LDS_1A <0x37, "LDS_UBYTE_READ_RET",
+  [(set i32:$dst, (az_extloadi8_local i32:$src0))]
+>;
+def LDS_SHORT_READ_RET : R600_LDS_1A <0x38, "LDS_SHORT_READ_RET",
+  [(set i32:$dst, (sextloadi16_local i32:$src0))]
+>;
+def LDS_USHORT_READ_RET : R600_LDS_1A <0x39, "LDS_USHORT_READ_RET",
+  [(set i32:$dst, (az_extloadi16_local i32:$src0))]
+>;
+
+// TRUNC is used for the FLT_TO_INT instructions to work around a
+// perceived problem where the rounding modes are applied differently
+// depending on the instruction and the slot they are in.
+// See:
+// https://bugs.freedesktop.org/show_bug.cgi?id=50232
+// Mesa commit: a1a0974401c467cb86ef818f22df67c21774a38c
+//
+// XXX: Lowering SELECT_CC will sometimes generate fp_to_[su]int nodes,
+// which do not need to be truncated since the fp values are 0.0f or 1.0f.
+// We should look into handling these cases separately.
+def : Pat<(fp_to_sint f32:$src0), (FLT_TO_INT_eg (TRUNC $src0))>;
+
+def : Pat<(fp_to_uint f32:$src0), (FLT_TO_UINT_eg (TRUNC $src0))>;
+
+// SHA-256 Patterns
+def : SHA256MaPattern <BFI_INT_eg, XOR_INT>;
+
+def : FROUNDPat <CNDGE_eg>;
+
+def EG_ExportSwz : ExportSwzInst {
+  let Word1{19-16} = 0; // BURST_COUNT
+  let Word1{20} = 0; // VALID_PIXEL_MODE
+  let Word1{21} = eop;
+  let Word1{29-22} = inst;
+  let Word1{30} = 0; // MARK
+  let Word1{31} = 1; // BARRIER
+}
+defm : ExportPattern<EG_ExportSwz, 83>;
+
+def EG_ExportBuf : ExportBufInst {
+  let Word1{19-16} = 0; // BURST_COUNT
+  let Word1{20} = 0; // VALID_PIXEL_MODE
+  let Word1{21} = eop;
+  let Word1{29-22} = inst;
+  let Word1{30} = 0; // MARK
+  let Word1{31} = 1; // BARRIER
+}
+defm : SteamOutputExportPattern<EG_ExportBuf, 0x40, 0x41, 0x42, 0x43>;
+
+def CF_TC_EG : CF_CLAUSE_EG<1, (ins i32imm:$ADDR, i32imm:$COUNT),
+  "TEX $COUNT @$ADDR"> {
+  let POP_COUNT = 0;
+}
+def CF_VC_EG : CF_CLAUSE_EG<2, (ins i32imm:$ADDR, i32imm:$COUNT),
+  "VTX $COUNT @$ADDR"> {
+  let POP_COUNT = 0;
+}
+def WHILE_LOOP_EG : CF_CLAUSE_EG<6, (ins i32imm:$ADDR),
+  "LOOP_START_DX10 @$ADDR"> {
+  let POP_COUNT = 0;
+  let COUNT = 0;
+}
+def END_LOOP_EG : CF_CLAUSE_EG<5, (ins i32imm:$ADDR), "END_LOOP @$ADDR"> {
+  let POP_COUNT = 0;
+  let COUNT = 0;
+}
+def LOOP_BREAK_EG : CF_CLAUSE_EG<9, (ins i32imm:$ADDR),
+  "LOOP_BREAK @$ADDR"> {
+  let POP_COUNT = 0;
+  let COUNT = 0;
+}
+def CF_CONTINUE_EG : CF_CLAUSE_EG<8, (ins i32imm:$ADDR),
+  "CONTINUE @$ADDR"> {
+  let POP_COUNT = 0;
+  let COUNT = 0;
+}
+def CF_JUMP_EG : CF_CLAUSE_EG<10, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
+  "JUMP @$ADDR POP:$POP_COUNT"> {
+  let COUNT = 0;
+}
+def CF_PUSH_EG : CF_CLAUSE_EG<11, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
+                              "PUSH @$ADDR POP:$POP_COUNT"> {
+  let COUNT = 0;
+}
+def CF_ELSE_EG : CF_CLAUSE_EG<13, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
+  "ELSE @$ADDR POP:$POP_COUNT"> {
+  let COUNT = 0;
+}
+def CF_CALL_FS_EG : CF_CLAUSE_EG<19, (ins), "CALL_FS"> {
+  let ADDR = 0;
+  let COUNT = 0;
+  let POP_COUNT = 0;
+}
+def POP_EG : CF_CLAUSE_EG<14, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
+  "POP @$ADDR POP:$POP_COUNT"> {
+  let COUNT = 0;
+}
+def CF_END_EG :  CF_CLAUSE_EG<0, (ins), "CF_END"> {
+  let COUNT = 0;
+  let POP_COUNT = 0;
+  let ADDR = 0;
+  let END_OF_PROGRAM = 1;
+}
+
+} // End Predicates = [isEGorCayman]
diff --git a/contrib/llvm/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.cpp b/contrib/llvm/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.cpp
index 99e1377..0927040 100644
--- a/contrib/llvm/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.cpp
+++ b/contrib/llvm/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.cpp
@@ -12,6 +12,8 @@
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/Support/MathExtras.h"
 
 using namespace llvm;
 
@@ -23,6 +25,21 @@ void AMDGPUInstPrinter::printInst(const MCInst *MI, raw_ostream &OS,
   printAnnotation(OS, Annot);
 }
 
+void AMDGPUInstPrinter::printU8ImmOperand(const MCInst *MI, unsigned OpNo,
+                                           raw_ostream &O) {
+  O << formatHex(MI->getOperand(OpNo).getImm() & 0xff);
+}
+
+void AMDGPUInstPrinter::printU16ImmOperand(const MCInst *MI, unsigned OpNo,
+                                           raw_ostream &O) {
+  O << formatHex(MI->getOperand(OpNo).getImm() & 0xffff);
+}
+
+void AMDGPUInstPrinter::printU32ImmOperand(const MCInst *MI, unsigned OpNo,
+                                           raw_ostream &O) {
+  O << formatHex(MI->getOperand(OpNo).getImm() & 0xffffffff);
+}
+
 void AMDGPUInstPrinter::printRegOperand(unsigned reg, raw_ostream &O) {
   switch (reg) {
   case AMDGPU::VCC:
@@ -41,43 +58,78 @@ void AMDGPUInstPrinter::printRegOperand(unsigned reg, raw_ostream &O) {
     break;
   }
 
-  // It's seems there's no way to use SIRegisterInfo here, and dealing with the
-  // giant enum of all the different shifted sets of registers is pretty
-  // unmanagable, so parse the name and reformat it to be prettier.
-  StringRef Name(getRegisterName(reg));
-
-  std::pair<StringRef, StringRef> Split = Name.split('_');
-  StringRef SubRegName = Split.first;
-  StringRef Rest = Split.second;
+  char Type;
+  unsigned NumRegs;
+
+  if (MRI.getRegClass(AMDGPU::VGPR_32RegClassID).contains(reg)) {
+    Type = 'v';
+    NumRegs = 1;
+  } else  if (MRI.getRegClass(AMDGPU::SGPR_32RegClassID).contains(reg)) {
+    Type = 's';
+    NumRegs = 1;
+  } else if (MRI.getRegClass(AMDGPU::VReg_64RegClassID).contains(reg)) {
+    Type = 'v';
+    NumRegs = 2;
+  } else  if (MRI.getRegClass(AMDGPU::SReg_64RegClassID).contains(reg)) {
+    Type = 's';
+    NumRegs = 2;
+  } else if (MRI.getRegClass(AMDGPU::VReg_128RegClassID).contains(reg)) {
+    Type = 'v';
+    NumRegs = 4;
+  } else  if (MRI.getRegClass(AMDGPU::SReg_128RegClassID).contains(reg)) {
+    Type = 's';
+    NumRegs = 4;
+  } else if (MRI.getRegClass(AMDGPU::VReg_96RegClassID).contains(reg)) {
+    Type = 'v';
+    NumRegs = 3;
+  } else if (MRI.getRegClass(AMDGPU::VReg_256RegClassID).contains(reg)) {
+    Type = 'v';
+    NumRegs = 8;
+  } else if (MRI.getRegClass(AMDGPU::SReg_256RegClassID).contains(reg)) {
+    Type = 's';
+    NumRegs = 8;
+  } else if (MRI.getRegClass(AMDGPU::VReg_512RegClassID).contains(reg)) {
+    Type = 'v';
+    NumRegs = 16;
+  } else if (MRI.getRegClass(AMDGPU::SReg_512RegClassID).contains(reg)) {
+    Type = 's';
+    NumRegs = 16;
+  } else {
+    O << getRegisterName(reg);
+    return;
+  }
 
-  if (SubRegName.size() <= 4) { // Must at least be as long as "SGPR"/"VGPR".
-    O << Name;
+  // The low 8 bits of the encoding value is the register index, for both VGPRs
+  // and SGPRs.
+  unsigned RegIdx = MRI.getEncodingValue(reg) & ((1 << 8) - 1);
+  if (NumRegs == 1) {
+    O << Type << RegIdx;
     return;
   }
 
-  unsigned RegIndex;
-  StringRef RegIndexStr = SubRegName.drop_front(4);
+  O << Type << '[' << RegIdx << ':' << (RegIdx + NumRegs - 1) << ']';
+}
 
-  if (RegIndexStr.getAsInteger(10, RegIndex)) {
-    O << Name;
+void AMDGPUInstPrinter::printImmediate(uint32_t Imm, raw_ostream &O) {
+  int32_t SImm = static_cast<int32_t>(Imm);
+  if (SImm >= -16 && SImm <= 64) {
+    O << SImm;
     return;
   }
 
-  if (SubRegName.front() == 'V')
-    O << 'v';
-  else if (SubRegName.front() == 'S')
-    O << 's';
-  else {
-    O << Name;
+  if (Imm == FloatToBits(1.0f) ||
+      Imm == FloatToBits(-1.0f) ||
+      Imm == FloatToBits(0.5f) ||
+      Imm == FloatToBits(-0.5f) ||
+      Imm == FloatToBits(2.0f) ||
+      Imm == FloatToBits(-2.0f) ||
+      Imm == FloatToBits(4.0f) ||
+      Imm == FloatToBits(-4.0f)) {
+    O << BitsToFloat(Imm);
     return;
   }
 
-  if (Rest.empty()) // Only 1 32-bit register
-    O << RegIndex;
-  else {
-    unsigned NumReg = Rest.count('_') + 2;
-    O << '[' << RegIndex << ':' << (RegIndex + NumReg - 1) << ']';
-  }
+  O << formatHex(static_cast<uint64_t>(Imm));
 }
 
 void AMDGPUInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
@@ -95,7 +147,7 @@ void AMDGPUInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
       break;
     }
   } else if (Op.isImm()) {
-    O << Op.getImm();
+    printImmediate(Op.getImm(), O);
   } else if (Op.isFPImm()) {
     O << Op.getFPImm();
   } else if (Op.isExpr()) {
@@ -106,6 +158,18 @@ void AMDGPUInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
   }
 }
 
+void AMDGPUInstPrinter::printOperandAndMods(const MCInst *MI, unsigned OpNo,
+                                            raw_ostream &O) {
+  unsigned InputModifiers = MI->getOperand(OpNo).getImm();
+  if (InputModifiers & 0x1)
+    O << "-";
+  if (InputModifiers & 0x2)
+    O << "|";
+  printOperand(MI, OpNo + 1, O);
+  if (InputModifiers & 0x2)
+    O << "|";
+}
+
 void AMDGPUInstPrinter::printInterpSlot(const MCInst *MI, unsigned OpNum,
                                         raw_ostream &O) {
   unsigned Imm = MI->getOperand(OpNum).getImm();
@@ -152,13 +216,8 @@ void AMDGPUInstPrinter::printClamp(const MCInst *MI, unsigned OpNo,
 
 void AMDGPUInstPrinter::printLiteral(const MCInst *MI, unsigned OpNo,
                                      raw_ostream &O) {
-  union Literal {
-    float f;
-    int32_t i;
-  } L;
-
-  L.i = MI->getOperand(OpNo).getImm();
-  O << L.i << "(" << L.f << ")";
+  int32_t Imm = MI->getOperand(OpNo).getImm();
+  O << Imm << '(' << BitsToFloat(Imm) << ')';
 }
 
 void AMDGPUInstPrinter::printLast(const MCInst *MI, unsigned OpNo,
@@ -316,6 +375,37 @@ void AMDGPUInstPrinter::printKCache(const MCInst *MI, unsigned OpNo,
   }
 }
 
+void AMDGPUInstPrinter::printSendMsg(const MCInst *MI, unsigned OpNo,
+                                     raw_ostream &O) {
+  unsigned SImm16 = MI->getOperand(OpNo).getImm();
+  unsigned Msg = SImm16 & 0xF;
+  if (Msg == 2 || Msg == 3) {
+    unsigned Op = (SImm16 >> 4) & 0xF;
+    if (Msg == 3)
+      O << "Gs_done(";
+    else
+      O << "Gs(";
+    if (Op == 0) {
+      O << "nop";
+    } else {
+      unsigned Stream = (SImm16 >> 8) & 0x3;
+      if (Op == 1)
+	O << "cut";
+      else if (Op == 2)
+	O << "emit";
+      else if (Op == 3)
+	O << "emit-cut";
+      O << " stream " << Stream;
+    }
+    O << "), [m0] ";
+  } else if (Msg == 1)
+    O << "interrupt ";
+  else if (Msg == 15)
+    O << "system ";
+  else
+    O << "unknown(" << Msg << ") ";
+}
+
 void AMDGPUInstPrinter::printWaitFlag(const MCInst *MI, unsigned OpNo,
                                       raw_ostream &O) {
   // Note: Mask values are taken from SIInsertWaits.cpp and not from ISA docs
diff --git a/contrib/llvm/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.h b/contrib/llvm/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.h
index 77af942..6ca7170 100644
--- a/contrib/llvm/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.h
+++ b/contrib/llvm/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.h
@@ -29,31 +29,38 @@ public:
   void printInstruction(const MCInst *MI, raw_ostream &O);
   static const char *getRegisterName(unsigned RegNo);
 
-  virtual void printInst(const MCInst *MI, raw_ostream &O, StringRef Annot);
+  void printInst(const MCInst *MI, raw_ostream &O, StringRef Annot) override;
 
 private:
+  void printU8ImmOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printU16ImmOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printU32ImmOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
   void printRegOperand(unsigned RegNo, raw_ostream &O);
+  void printImmediate(uint32_t Imm, raw_ostream &O);
   void printOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printInterpSlot(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+  void printOperandAndMods(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printInterpSlot(const MCInst *MI, unsigned OpNum, raw_ostream &O);
   void printMemOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printIfSet(const MCInst *MI, unsigned OpNo, raw_ostream &O,
-                  StringRef Asm, StringRef Default = "");
-  void printAbs(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printClamp(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printLiteral(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printLast(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printNeg(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printOMOD(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printRel(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printUpdateExecMask(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printUpdatePred(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printWrite(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printSel(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printBankSwizzle(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printRSel(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printCT(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printKCache(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printWaitFlag(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printIfSet(const MCInst *MI, unsigned OpNo, raw_ostream &O,
+                         StringRef Asm, StringRef Default = "");
+  static void printAbs(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printClamp(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printLiteral(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printLast(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printNeg(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printOMOD(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printRel(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printUpdateExecMask(const MCInst *MI, unsigned OpNo,
+                                  raw_ostream &O);
+  static void printUpdatePred(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printWrite(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printSel(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printBankSwizzle(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printRSel(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printCT(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printKCache(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printSendMsg(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  static void printWaitFlag(const MCInst *MI, unsigned OpNo, raw_ostream &O);
 };
 
 } // End namespace llvm
diff --git a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUAsmBackend.cpp b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUAsmBackend.cpp
index 29d0acf..d55f27b 100644
--- a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUAsmBackend.cpp
+++ b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUAsmBackend.cpp
@@ -9,9 +9,11 @@
 //===----------------------------------------------------------------------===//
 
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
+#include "MCTargetDesc/AMDGPUFixupKinds.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/MC/MCAsmBackend.h"
 #include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCFixupKindInfo.h"
 #include "llvm/MC/MCObjectWriter.h"
 #include "llvm/MC/MCValue.h"
 #include "llvm/Support/TargetRegistry.h"
@@ -23,19 +25,18 @@ namespace {
 class AMDGPUMCObjectWriter : public MCObjectWriter {
 public:
   AMDGPUMCObjectWriter(raw_ostream &OS) : MCObjectWriter(OS, true) { }
-  virtual void ExecutePostLayoutBinding(MCAssembler &Asm,
-                                        const MCAsmLayout &Layout) {
+  void ExecutePostLayoutBinding(MCAssembler &Asm,
+                                const MCAsmLayout &Layout) override {
     //XXX: Implement if necessary.
   }
-  virtual void RecordRelocation(const MCAssembler &Asm,
-                                const MCAsmLayout &Layout,
-                                const MCFragment *Fragment,
-                                const MCFixup &Fixup,
-                                MCValue Target, uint64_t &FixedValue) {
+  void RecordRelocation(const MCAssembler &Asm, const MCAsmLayout &Layout,
+                        const MCFragment *Fragment, const MCFixup &Fixup,
+                        MCValue Target, bool &IsPCRel,
+                        uint64_t &FixedValue) override {
     assert(!"Not implemented");
   }
 
-  virtual void WriteObject(MCAssembler &Asm, const MCAsmLayout &Layout);
+  void WriteObject(MCAssembler &Asm, const MCAsmLayout &Layout) override;
 
 };
 
@@ -44,21 +45,23 @@ public:
   AMDGPUAsmBackend(const Target &T)
     : MCAsmBackend() {}
 
-  virtual unsigned getNumFixupKinds() const { return 0; };
-  virtual void applyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize,
-                          uint64_t Value) const;
-  virtual bool fixupNeedsRelaxation(const MCFixup &Fixup, uint64_t Value,
-                                    const MCRelaxableFragment *DF,
-                                    const MCAsmLayout &Layout) const {
+  unsigned getNumFixupKinds() const override { return AMDGPU::NumTargetFixupKinds; };
+  void applyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize,
+                  uint64_t Value, bool IsPCRel) const override;
+  bool fixupNeedsRelaxation(const MCFixup &Fixup, uint64_t Value,
+                            const MCRelaxableFragment *DF,
+                            const MCAsmLayout &Layout) const override {
     return false;
   }
-  virtual void relaxInstruction(const MCInst &Inst, MCInst &Res) const {
+  void relaxInstruction(const MCInst &Inst, MCInst &Res) const override {
     assert(!"Not implemented");
   }
-  virtual bool mayNeedRelaxation(const MCInst &Inst) const { return false; }
-  virtual bool writeNopData(uint64_t Count, MCObjectWriter *OW) const {
+  bool mayNeedRelaxation(const MCInst &Inst) const override { return false; }
+  bool writeNopData(uint64_t Count, MCObjectWriter *OW) const override {
     return true;
   }
+
+  const MCFixupKindInfo &getFixupKindInfo(MCFixupKind Kind) const override;
 };
 
 } //End anonymous namespace
@@ -71,11 +74,46 @@ void AMDGPUMCObjectWriter::WriteObject(MCAssembler &Asm,
 }
 
 void AMDGPUAsmBackend::applyFixup(const MCFixup &Fixup, char *Data,
-                                  unsigned DataSize, uint64_t Value) const {
+                                  unsigned DataSize, uint64_t Value,
+                                  bool IsPCRel) const {
+
+  switch ((unsigned)Fixup.getKind()) {
+    default: llvm_unreachable("Unknown fixup kind");
+    case AMDGPU::fixup_si_sopp_br: {
+      uint16_t *Dst = (uint16_t*)(Data + Fixup.getOffset());
+      *Dst = (Value - 4) / 4;
+      break;
+    }
+
+    case AMDGPU::fixup_si_rodata: {
+      uint32_t *Dst = (uint32_t*)(Data + Fixup.getOffset());
+      *Dst = Value;
+      break;
+    }
+
+    case AMDGPU::fixup_si_end_of_text: {
+      uint32_t *Dst = (uint32_t*)(Data + Fixup.getOffset());
+      // The value points to the last instruction in the text section, so we
+      // need to add 4 bytes to get to the start of the constants.
+      *Dst = Value + 4;
+      break;
+    }
+  }
+}
+
+const MCFixupKindInfo &AMDGPUAsmBackend::getFixupKindInfo(
+                                                       MCFixupKind Kind) const {
+  const static MCFixupKindInfo Infos[AMDGPU::NumTargetFixupKinds] = {
+    // name                   offset bits  flags
+    { "fixup_si_sopp_br",     0,     16,   MCFixupKindInfo::FKF_IsPCRel },
+    { "fixup_si_rodata",      0,     32,   0 },
+    { "fixup_si_end_of_text", 0,     32,   MCFixupKindInfo::FKF_IsPCRel }
+  };
+
+  if (Kind < FirstTargetFixupKind)
+    return MCAsmBackend::getFixupKindInfo(Kind);
 
-  uint16_t *Dst = (uint16_t*)(Data + Fixup.getOffset());
-  assert(Fixup.getKind() == FK_PCRel_4);
-  *Dst = (Value - 4) / 4;
+  return Infos[Kind - FirstTargetFixupKind];
 }
 
 //===----------------------------------------------------------------------===//
@@ -88,7 +126,7 @@ class ELFAMDGPUAsmBackend : public AMDGPUAsmBackend {
 public:
   ELFAMDGPUAsmBackend(const Target &T) : AMDGPUAsmBackend(T) { }
 
-  MCObjectWriter *createObjectWriter(raw_ostream &OS) const {
+  MCObjectWriter *createObjectWriter(raw_ostream &OS) const override {
     return createAMDGPUELFObjectWriter(OS);
   }
 };
diff --git a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUELFObjectWriter.cpp b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUELFObjectWriter.cpp
index 48fac9f..5fb94d5 100644
--- a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUELFObjectWriter.cpp
+++ b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUELFObjectWriter.cpp
@@ -10,6 +10,7 @@
 
 #include "AMDGPUMCTargetDesc.h"
 #include "llvm/MC/MCELFObjectWriter.h"
+#include "llvm/MC/MCFixup.h"
 
 using namespace llvm;
 
@@ -19,10 +20,9 @@ class AMDGPUELFObjectWriter : public MCELFObjectTargetWriter {
 public:
   AMDGPUELFObjectWriter();
 protected:
-  virtual unsigned GetRelocType(const MCValue &Target, const MCFixup &Fixup,
-                                bool IsPCRel, bool IsRelocWithSymbol,
-                                int64_t Addend) const {
-    llvm_unreachable("Not implemented");
+  unsigned GetRelocType(const MCValue &Target, const MCFixup &Fixup,
+                        bool IsPCRel) const override {
+    return Fixup.getKind();
   }
 
 };
diff --git a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUFixupKinds.h b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUFixupKinds.h
new file mode 100644
index 0000000..4b12e54
--- /dev/null
+++ b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUFixupKinds.h
@@ -0,0 +1,34 @@
+//===-- AMDGPUFixupKinds.h - AMDGPU Specific Fixup Entries ------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_AMDGPUFIXUPKINDS_H
+#define LLVM_AMDGPUFIXUPKINDS_H
+
+#include "llvm/MC/MCFixup.h"
+
+namespace llvm {
+namespace AMDGPU {
+enum Fixups {
+  /// 16-bit PC relative fixup for SOPP branch instructions.
+  fixup_si_sopp_br = FirstTargetFixupKind,
+
+  /// fixup for global addresses with constant initializers
+  fixup_si_rodata,
+
+  /// fixup for offset from instruction to end of text section
+  fixup_si_end_of_text,
+
+  // Marker
+  LastTargetFixupKind,
+  NumTargetFixupKinds = LastTargetFixupKind - FirstTargetFixupKind
+};
+}
+}
+
+#endif // LLVM_AMDGPUFIXUPKINDS_H
diff --git a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.cpp b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.cpp
index 9b26af7..78bbe0a 100644
--- a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.cpp
+++ b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.cpp
@@ -21,12 +21,8 @@ AMDGPUMCAsmInfo::AMDGPUMCAsmInfo(StringRef &TT) : MCAsmInfo() {
   LinkerRequiresNonEmptyDwarfLines = true;
   MaxInstLength = 16;
   SeparatorString = "\n";
-  CommentColumn = 40;
   CommentString = ";";
   LabelSuffix = ":";
-  GlobalPrefix = "@";
-  PrivateGlobalPrefix = ";.";
-  LinkerPrivateGlobalPrefix = "!";
   InlineAsmStart = ";#ASMSTART";
   InlineAsmEnd = ";#ASMEND";
   AssemblerDialect = 0;
@@ -39,13 +35,11 @@ AMDGPUMCAsmInfo::AMDGPUMCAsmInfo(StringRef &TT) : MCAsmInfo() {
   Data16bitsDirective = ".short\t";
   Data32bitsDirective = ".long\t";
   Data64bitsDirective = ".quad\t";
-  GPRel32Directive = 0;
+  GPRel32Directive = nullptr;
   SunStyleELFSectionSwitchSyntax = true;
   UsesELFSectionDirectiveForBSS = true;
-  HasMicrosoftFastStdCallMangling = false;
 
   //===--- Alignment Information ----------------------------------------===//
-  AlignDirective = ".align\t";
   AlignmentIsInBytes = true;
   TextAlignFillValue = 0;
 
@@ -64,5 +58,5 @@ AMDGPUMCAsmInfo::AMDGPUMCAsmInfo(StringRef &TT) : MCAsmInfo() {
 
 const MCSection*
 AMDGPUMCAsmInfo::getNonexecutableStackSection(MCContext &CTX) const {
-  return 0;
+  return nullptr;
 }
diff --git a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.h b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.h
index 22afd63..59aebec 100644
--- a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.h
+++ b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.h
@@ -22,7 +22,7 @@ class StringRef;
 class AMDGPUMCAsmInfo : public MCAsmInfo {
 public:
   explicit AMDGPUMCAsmInfo(StringRef &TT);
-  const MCSection* getNonexecutableStackSection(MCContext &CTX) const;
+  const MCSection* getNonexecutableStackSection(MCContext &CTX) const override;
 };
 } // namespace llvm
 #endif // AMDGPUMCASMINFO_H
diff --git a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCCodeEmitter.h b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCCodeEmitter.h
index d8cf64a..d5e432d 100644
--- a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCCodeEmitter.h
+++ b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCCodeEmitter.h
@@ -1,4 +1,4 @@
-//===-- AMDGPUCodeEmitter.h - AMDGPU Code Emitter interface -----------------===//
+//===-- AMDGPUCodeEmitter.h - AMDGPU Code Emitter interface -----*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -22,16 +22,25 @@ namespace llvm {
 
 class MCInst;
 class MCOperand;
+class MCSubtargetInfo;
 
 class AMDGPUMCCodeEmitter : public MCCodeEmitter {
   virtual void anchor();
 public:
 
   uint64_t getBinaryCodeForInstr(const MCInst &MI,
-                                 SmallVectorImpl<MCFixup> &Fixups) const;
+                                 SmallVectorImpl<MCFixup> &Fixups,
+                                 const MCSubtargetInfo &STI) const;
 
   virtual uint64_t getMachineOpValue(const MCInst &MI, const MCOperand &MO,
-                                     SmallVectorImpl<MCFixup> &Fixups) const {
+                                     SmallVectorImpl<MCFixup> &Fixups,
+                                     const MCSubtargetInfo &STI) const {
+    return 0;
+  }
+
+  virtual unsigned getSOPPBrEncoding(const MCInst &MI, unsigned OpNo,
+                                     SmallVectorImpl<MCFixup> &Fixups,
+                                     const MCSubtargetInfo &STI) const {
     return 0;
   }
 };
diff --git a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCTargetDesc.cpp b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCTargetDesc.cpp
index a1bec28..38a2956 100644
--- a/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCTargetDesc.cpp
+++ b/contrib/llvm/lib/Target/R600/MCTargetDesc/AMDGPUMCTargetDesc.cpp
@@ -24,6 +24,8 @@
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
 
+using namespace llvm;
+
 #define GET_INSTRINFO_MC_DESC
 #include "AMDGPUGenInstrInfo.inc"
 
@@ -33,8 +35,6 @@
 #define GET_REGINFO_MC_DESC
 #include "AMDGPUGenRegisterInfo.inc"
 
-using namespace llvm;
-
 static MCInstrInfo *createAMDGPUMCInstrInfo() {
   MCInstrInfo *X = new MCInstrInfo();
   InitAMDGPUMCInstrInfo(X);
@@ -86,9 +86,10 @@ static MCStreamer *createMCStreamer(const Target &T, StringRef TT,
                                     MCContext &Ctx, MCAsmBackend &MAB,
                                     raw_ostream &_OS,
                                     MCCodeEmitter *_Emitter,
+                                    const MCSubtargetInfo &STI,
                                     bool RelaxAll,
                                     bool NoExecStack) {
-  return createELFStreamer(Ctx, 0, MAB, _OS, _Emitter, false, false);
+  return createELFStreamer(Ctx, MAB, _OS, _Emitter, false, false);
 }
 
 extern "C" void LLVMInitializeR600TargetMC() {
diff --git a/contrib/llvm/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp b/contrib/llvm/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp
index dd8df65..dc1344f 100644
--- a/contrib/llvm/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp
+++ b/contrib/llvm/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp
@@ -34,21 +34,21 @@ class R600MCCodeEmitter : public AMDGPUMCCodeEmitter {
   void operator=(const R600MCCodeEmitter &) LLVM_DELETED_FUNCTION;
   const MCInstrInfo &MCII;
   const MCRegisterInfo &MRI;
-  const MCSubtargetInfo &STI;
 
 public:
 
-  R600MCCodeEmitter(const MCInstrInfo &mcii, const MCRegisterInfo &mri,
-                    const MCSubtargetInfo &sti)
-    : MCII(mcii), MRI(mri), STI(sti) { }
+  R600MCCodeEmitter(const MCInstrInfo &mcii, const MCRegisterInfo &mri)
+    : MCII(mcii), MRI(mri) { }
 
   /// \brief Encode the instruction and write it to the OS.
-  virtual void EncodeInstruction(const MCInst &MI, raw_ostream &OS,
-                         SmallVectorImpl<MCFixup> &Fixups) const;
+  void EncodeInstruction(const MCInst &MI, raw_ostream &OS,
+                         SmallVectorImpl<MCFixup> &Fixups,
+                         const MCSubtargetInfo &STI) const override;
 
   /// \returns the encoding for an MCOperand.
-  virtual uint64_t getMachineOpValue(const MCInst &MI, const MCOperand &MO,
-                                     SmallVectorImpl<MCFixup> &Fixups) const;
+  uint64_t getMachineOpValue(const MCInst &MI, const MCOperand &MO,
+                             SmallVectorImpl<MCFixup> &Fixups,
+                             const MCSubtargetInfo &STI) const override;
 private:
 
   void EmitByte(unsigned int byte, raw_ostream &OS) const;
@@ -83,11 +83,12 @@ enum FCInstr {
 MCCodeEmitter *llvm::createR600MCCodeEmitter(const MCInstrInfo &MCII,
                                            const MCRegisterInfo &MRI,
                                            const MCSubtargetInfo &STI) {
-  return new R600MCCodeEmitter(MCII, MRI, STI);
+  return new R600MCCodeEmitter(MCII, MRI);
 }
 
 void R600MCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
-                                       SmallVectorImpl<MCFixup> &Fixups) const {
+                                       SmallVectorImpl<MCFixup> &Fixups,
+                                       const MCSubtargetInfo &STI) const {
   const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
   if (MI.getOpcode() == AMDGPU::RETURN ||
     MI.getOpcode() == AMDGPU::FETCH_CLAUSE ||
@@ -96,7 +97,7 @@ void R600MCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
     MI.getOpcode() == AMDGPU::KILL) {
     return;
   } else if (IS_VTX(Desc)) {
-    uint64_t InstWord01 = getBinaryCodeForInstr(MI, Fixups);
+    uint64_t InstWord01 = getBinaryCodeForInstr(MI, Fixups, STI);
     uint32_t InstWord2 = MI.getOperand(2).getImm(); // Offset
     if (!(STI.getFeatureBits() & AMDGPU::FeatureCaymanISA)) {
       InstWord2 |= 1 << 19; // Mega-Fetch bit
@@ -120,7 +121,7 @@ void R600MCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
         MI.getOperand(8).getImm() & 0x1F
       };
 
-      uint64_t Word01 = getBinaryCodeForInstr(MI, Fixups);
+      uint64_t Word01 = getBinaryCodeForInstr(MI, Fixups, STI);
       uint32_t Word2 = Sampler << 15 | SrcSelect[ELEMENT_X] << 20 |
           SrcSelect[ELEMENT_Y] << 23 | SrcSelect[ELEMENT_Z] << 26 |
           SrcSelect[ELEMENT_W] << 29 | Offsets[0] << 0 | Offsets[1] << 5 |
@@ -130,7 +131,7 @@ void R600MCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
       Emit(Word2, OS);
       Emit((uint32_t) 0, OS);
   } else {
-    uint64_t Inst = getBinaryCodeForInstr(MI, Fixups);
+    uint64_t Inst = getBinaryCodeForInstr(MI, Fixups, STI);
     if ((STI.getFeatureBits() & AMDGPU::FeatureR600ALUInst) &&
        ((Desc.TSFlags & R600_InstFlag::OP1) ||
          Desc.TSFlags & R600_InstFlag::OP2)) {
@@ -168,19 +169,16 @@ unsigned R600MCCodeEmitter::getHWReg(unsigned RegNo) const {
 
 uint64_t R600MCCodeEmitter::getMachineOpValue(const MCInst &MI,
                                               const MCOperand &MO,
-                                        SmallVectorImpl<MCFixup> &Fixup) const {
+                                        SmallVectorImpl<MCFixup> &Fixup,
+                                        const MCSubtargetInfo &STI) const {
   if (MO.isReg()) {
-    if (HAS_NATIVE_OPERANDS(MCII.get(MI.getOpcode()).TSFlags)) {
+    if (HAS_NATIVE_OPERANDS(MCII.get(MI.getOpcode()).TSFlags))
       return MRI.getEncodingValue(MO.getReg());
-    } else {
-      return getHWReg(MO.getReg());
-    }
-  } else if (MO.isImm()) {
-    return MO.getImm();
-  } else {
-    assert(0);
-    return 0;
+    return getHWReg(MO.getReg());
   }
+
+  assert(MO.isImm());
+  return MO.getImm();
 }
 
 #include "AMDGPUGenMCCodeEmitter.inc"
diff --git a/contrib/llvm/lib/Target/R600/MCTargetDesc/SIMCCodeEmitter.cpp b/contrib/llvm/lib/Target/R600/MCTargetDesc/SIMCCodeEmitter.cpp
index 5af8320..78776c1 100644
--- a/contrib/llvm/lib/Target/R600/MCTargetDesc/SIMCCodeEmitter.cpp
+++ b/contrib/llvm/lib/Target/R600/MCTargetDesc/SIMCCodeEmitter.cpp
@@ -13,8 +13,10 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "AMDGPU.h"
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "MCTargetDesc/AMDGPUMCCodeEmitter.h"
+#include "MCTargetDesc/AMDGPUFixupKinds.h"
 #include "llvm/MC/MCCodeEmitter.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCFixup.h"
@@ -39,6 +41,7 @@ class SIMCCodeEmitter : public  AMDGPUMCCodeEmitter {
   void operator=(const SIMCCodeEmitter &) LLVM_DELETED_FUNCTION;
   const MCInstrInfo &MCII;
   const MCRegisterInfo &MRI;
+  MCContext &Ctx;
 
   /// \brief Can this operand also contain immediate values?
   bool isSrcOperand(const MCInstrDesc &Desc, unsigned OpNo) const;
@@ -48,18 +51,26 @@ class SIMCCodeEmitter : public  AMDGPUMCCodeEmitter {
 
 public:
   SIMCCodeEmitter(const MCInstrInfo &mcii, const MCRegisterInfo &mri,
-                  const MCSubtargetInfo &sti, MCContext &ctx)
-    : MCII(mcii), MRI(mri) { }
+                  MCContext &ctx)
+    : MCII(mcii), MRI(mri), Ctx(ctx) { }
 
   ~SIMCCodeEmitter() { }
 
-  /// \breif Encode the instruction and write it to the OS.
-  virtual void EncodeInstruction(const MCInst &MI, raw_ostream &OS,
-                         SmallVectorImpl<MCFixup> &Fixups) const;
+  /// \brief Encode the instruction and write it to the OS.
+  void EncodeInstruction(const MCInst &MI, raw_ostream &OS,
+                         SmallVectorImpl<MCFixup> &Fixups,
+                         const MCSubtargetInfo &STI) const override;
 
   /// \returns the encoding for an MCOperand.
-  virtual uint64_t getMachineOpValue(const MCInst &MI, const MCOperand &MO,
-                                     SmallVectorImpl<MCFixup> &Fixups) const;
+  uint64_t getMachineOpValue(const MCInst &MI, const MCOperand &MO,
+                             SmallVectorImpl<MCFixup> &Fixups,
+                             const MCSubtargetInfo &STI) const override;
+
+  /// \brief Use a fixup to encode the simm16 field for SOPP branch
+  ///        instructions.
+  unsigned getSOPPBrEncoding(const MCInst &MI, unsigned OpNo,
+                             SmallVectorImpl<MCFixup> &Fixups,
+                             const MCSubtargetInfo &STI) const override;
 };
 
 } // End anonymous namespace
@@ -68,7 +79,7 @@ MCCodeEmitter *llvm::createSIMCCodeEmitter(const MCInstrInfo &MCII,
                                            const MCRegisterInfo &MRI,
                                            const MCSubtargetInfo &STI,
                                            MCContext &Ctx) {
-  return new SIMCCodeEmitter(MCII, MRI, STI, Ctx);
+  return new SIMCCodeEmitter(MCII, MRI, Ctx);
 }
 
 bool SIMCCodeEmitter::isSrcOperand(const MCInstrDesc &Desc,
@@ -88,6 +99,8 @@ uint32_t SIMCCodeEmitter::getLitEncoding(const MCOperand &MO) const {
     Imm.I = MO.getImm();
   else if (MO.isFPImm())
     Imm.F = MO.getFPImm();
+  else if (MO.isExpr())
+    return 255;
   else
     return ~0;
 
@@ -125,9 +138,10 @@ uint32_t SIMCCodeEmitter::getLitEncoding(const MCOperand &MO) const {
 }
 
 void SIMCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
-                                       SmallVectorImpl<MCFixup> &Fixups) const {
+                                       SmallVectorImpl<MCFixup> &Fixups,
+                                       const MCSubtargetInfo &STI) const {
 
-  uint64_t Encoding = getBinaryCodeForInstr(MI, Fixups);
+  uint64_t Encoding = getBinaryCodeForInstr(MI, Fixups, STI);
   const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
   unsigned bytes = Desc.getSize();
 
@@ -154,8 +168,13 @@ void SIMCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
     IntFloatUnion Imm;
     if (Op.isImm())
       Imm.I = Op.getImm();
-    else
+    else if (Op.isFPImm())
       Imm.F = Op.getFPImm();
+    else {
+      assert(Op.isExpr());
+      // This will be replaced with a fixup value.
+      Imm.I = 0;
+    }
 
     for (unsigned j = 0; j < 4; j++) {
       OS.write((uint8_t) ((Imm.I >> (8 * j)) & 0xff));
@@ -166,17 +185,42 @@ void SIMCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
   }
 }
 
+unsigned SIMCCodeEmitter::getSOPPBrEncoding(const MCInst &MI, unsigned OpNo,
+                                            SmallVectorImpl<MCFixup> &Fixups,
+                                            const MCSubtargetInfo &STI) const {
+  const MCOperand &MO = MI.getOperand(OpNo);
+
+  if (MO.isExpr()) {
+    const MCExpr *Expr = MO.getExpr();
+    MCFixupKind Kind = (MCFixupKind)AMDGPU::fixup_si_sopp_br;
+    Fixups.push_back(MCFixup::Create(0, Expr, Kind, MI.getLoc()));
+    return 0;
+  }
+
+  return getMachineOpValue(MI, MO, Fixups, STI);
+}
+
 uint64_t SIMCCodeEmitter::getMachineOpValue(const MCInst &MI,
                                             const MCOperand &MO,
-                                       SmallVectorImpl<MCFixup> &Fixups) const {
+                                       SmallVectorImpl<MCFixup> &Fixups,
+                                       const MCSubtargetInfo &STI) const {
   if (MO.isReg())
     return MRI.getEncodingValue(MO.getReg());
 
   if (MO.isExpr()) {
-    const MCExpr *Expr = MO.getExpr();
-    MCFixupKind Kind = MCFixupKind(FK_PCRel_4);
-    Fixups.push_back(MCFixup::Create(0, Expr, Kind, MI.getLoc()));
-    return 0;
+    const MCSymbolRefExpr *Expr = cast<MCSymbolRefExpr>(MO.getExpr());
+    MCFixupKind Kind;
+    const MCSymbol *Sym =
+        Ctx.GetOrCreateSymbol(StringRef(END_OF_TEXT_LABEL_NAME));
+
+    if (&Expr->getSymbol() == Sym) {
+      // Add the offset to the beginning of the constant values.
+      Kind = (MCFixupKind)AMDGPU::fixup_si_end_of_text;
+    } else {
+      // This is used for constant data stored in .rodata.
+     Kind = (MCFixupKind)AMDGPU::fixup_si_rodata;
+    }
+    Fixups.push_back(MCFixup::Create(4, Expr, Kind, MI.getLoc()));
   }
 
   // Figure out the operand number, needed for isSrcOperand check
diff --git a/contrib/llvm/lib/Target/R600/Processors.td b/contrib/llvm/lib/Target/R600/Processors.td
index ee190e4..ce17d7c 100644
--- a/contrib/llvm/lib/Target/R600/Processors.td
+++ b/contrib/llvm/lib/Target/R600/Processors.td
@@ -9,46 +9,102 @@
 
 class Proc<string Name, ProcessorItineraries itin, list<SubtargetFeature> Features>
 : Processor<Name, itin, Features>;
+
+//===----------------------------------------------------------------------===//
+// R600
+//===----------------------------------------------------------------------===//
 def : Proc<"",           R600_VLIW5_Itin,
     [FeatureR600, FeatureVertexCache]>;
+
 def : Proc<"r600",       R600_VLIW5_Itin,
-    [FeatureR600 , FeatureVertexCache]>;
+    [FeatureR600 , FeatureVertexCache, FeatureWavefrontSize64]>;
+
+def : Proc<"r630",       R600_VLIW5_Itin,
+    [FeatureR600, FeatureVertexCache, FeatureWavefrontSize32]>;
+
 def : Proc<"rs880",      R600_VLIW5_Itin,
-    [FeatureR600]>;
+    [FeatureR600, FeatureWavefrontSize16]>;
+
 def : Proc<"rv670",      R600_VLIW5_Itin,
-    [FeatureR600, FeatureFP64, FeatureVertexCache]>;
+    [FeatureR600, FeatureFP64, FeatureVertexCache, FeatureWavefrontSize64]>;
+
+//===----------------------------------------------------------------------===//
+// R700
+//===----------------------------------------------------------------------===//
+
 def : Proc<"rv710",      R600_VLIW5_Itin,
-    [FeatureR700, FeatureVertexCache]>;
+    [FeatureR700, FeatureVertexCache, FeatureWavefrontSize32]>;
+
 def : Proc<"rv730",      R600_VLIW5_Itin,
-    [FeatureR700, FeatureVertexCache]>;
+    [FeatureR700, FeatureVertexCache, FeatureWavefrontSize32]>;
+
 def : Proc<"rv770",      R600_VLIW5_Itin,
-    [FeatureR700, FeatureFP64, FeatureVertexCache]>;
+    [FeatureR700, FeatureFP64, FeatureVertexCache, FeatureWavefrontSize64]>;
+
+//===----------------------------------------------------------------------===//
+// Evergreen
+//===----------------------------------------------------------------------===//
+
 def : Proc<"cedar",      R600_VLIW5_Itin,
-    [FeatureEvergreen, FeatureVertexCache]>;
+    [FeatureEvergreen, FeatureVertexCache, FeatureWavefrontSize32,
+     FeatureCFALUBug]>;
+
 def : Proc<"redwood",    R600_VLIW5_Itin,
-    [FeatureEvergreen, FeatureVertexCache]>;
+    [FeatureEvergreen, FeatureVertexCache, FeatureWavefrontSize64,
+     FeatureCFALUBug]>;
+
 def : Proc<"sumo",       R600_VLIW5_Itin,
-    [FeatureEvergreen]>;
+    [FeatureEvergreen, FeatureWavefrontSize64, FeatureCFALUBug]>;
+
 def : Proc<"juniper",    R600_VLIW5_Itin,
-    [FeatureEvergreen, FeatureVertexCache]>;
+    [FeatureEvergreen, FeatureVertexCache, FeatureWavefrontSize64]>;
+
 def : Proc<"cypress",    R600_VLIW5_Itin,
-    [FeatureEvergreen, FeatureFP64, FeatureVertexCache]>;
+    [FeatureEvergreen, FeatureFP64, FeatureVertexCache,
+     FeatureWavefrontSize64]>;
+
+//===----------------------------------------------------------------------===//
+// Northern Islands
+//===----------------------------------------------------------------------===//
+
 def : Proc<"barts",      R600_VLIW5_Itin,
-    [FeatureNorthernIslands, FeatureVertexCache]>;
+    [FeatureNorthernIslands, FeatureVertexCache, FeatureCFALUBug]>;
+
 def : Proc<"turks",      R600_VLIW5_Itin,
-    [FeatureNorthernIslands, FeatureVertexCache]>;
+    [FeatureNorthernIslands, FeatureVertexCache, FeatureCFALUBug]>;
+
 def : Proc<"caicos",     R600_VLIW5_Itin,
-    [FeatureNorthernIslands]>;
+    [FeatureNorthernIslands, FeatureCFALUBug]>;
+
 def : Proc<"cayman",     R600_VLIW4_Itin,
     [FeatureNorthernIslands, FeatureFP64, FeatureCaymanISA]>;
 
+//===----------------------------------------------------------------------===//
+// Southern Islands
+//===----------------------------------------------------------------------===//
+
 def : Proc<"SI",         SI_Itin, [FeatureSouthernIslands]>;
+
 def : Proc<"tahiti",     SI_Itin, [FeatureSouthernIslands]>;
+
 def : Proc<"pitcairn",   SI_Itin, [FeatureSouthernIslands]>;
+
 def : Proc<"verde",      SI_Itin, [FeatureSouthernIslands]>;
+
 def : Proc<"oland",      SI_Itin, [FeatureSouthernIslands]>;
+
 def : Proc<"hainan",     SI_Itin, [FeatureSouthernIslands]>;
+
+//===----------------------------------------------------------------------===//
+// Sea Islands
+//===----------------------------------------------------------------------===//
+
 def : Proc<"bonaire",    SI_Itin, [FeatureSeaIslands]>;
+
 def : Proc<"kabini",     SI_Itin, [FeatureSeaIslands]>;
+
 def : Proc<"kaveri",     SI_Itin, [FeatureSeaIslands]>;
+
 def : Proc<"hawaii",     SI_Itin, [FeatureSeaIslands]>;
+
+def : Proc<"mullins",    SI_Itin, [FeatureSeaIslands]>;
diff --git a/contrib/llvm/lib/Target/R600/R600ClauseMergePass.cpp b/contrib/llvm/lib/Target/R600/R600ClauseMergePass.cpp
index 33d2ca3..92bf0df 100644
--- a/contrib/llvm/lib/Target/R600/R600ClauseMergePass.cpp
+++ b/contrib/llvm/lib/Target/R600/R600ClauseMergePass.cpp
@@ -13,7 +13,6 @@
 /// It needs to be called after IfCvt for best results.
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "r600mergeclause"
 #include "AMDGPU.h"
 #include "R600Defines.h"
 #include "R600InstrInfo.h"
@@ -27,6 +26,8 @@
 
 using namespace llvm;
 
+#define DEBUG_TYPE "r600mergeclause"
+
 namespace {
 
 static bool isCFAlu(const MachineInstr *MI) {
@@ -50,7 +51,7 @@ private:
 
   /// IfCvt pass can generate "disabled" ALU clause marker that need to be
   /// removed and their content affected to the previous alu clause.
-  /// This function parse instructions after CFAlu untill it find a disabled
+  /// This function parse instructions after CFAlu until it find a disabled
   /// CFAlu and merge the content, or an enabled CFAlu.
   void cleanPotentialDisabledCFAlu(MachineInstr *CFAlu) const;
 
@@ -62,9 +63,9 @@ private:
 public:
   R600ClauseMergePass(TargetMachine &tm) : MachineFunctionPass(ID) { }
 
-  virtual bool runOnMachineFunction(MachineFunction &MF);
+  bool runOnMachineFunction(MachineFunction &MF) override;
 
-  const char *getPassName() const;
+  const char *getPassName() const override;
 };
 
 char R600ClauseMergePass::ID = 0;
diff --git a/contrib/llvm/lib/Target/R600/R600ControlFlowFinalizer.cpp b/contrib/llvm/lib/Target/R600/R600ControlFlowFinalizer.cpp
index 2a8276b..e37767a 100644
--- a/contrib/llvm/lib/Target/R600/R600ControlFlowFinalizer.cpp
+++ b/contrib/llvm/lib/Target/R600/R600ControlFlowFinalizer.cpp
@@ -12,9 +12,9 @@
 /// computing their address on the fly ; it also sets STACK_SIZE info.
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "r600cf"
 #include "llvm/Support/Debug.h"
 #include "AMDGPU.h"
+#include "AMDGPUSubtarget.h"
 #include "R600Defines.h"
 #include "R600InstrInfo.h"
 #include "R600MachineFunctionInfo.h"
@@ -26,8 +26,176 @@
 
 using namespace llvm;
 
+#define DEBUG_TYPE "r600cf"
+
 namespace {
 
+struct CFStack {
+
+  enum StackItem {
+    ENTRY = 0,
+    SUB_ENTRY = 1,
+    FIRST_NON_WQM_PUSH = 2,
+    FIRST_NON_WQM_PUSH_W_FULL_ENTRY = 3
+  };
+
+  const AMDGPUSubtarget &ST;
+  std::vector<StackItem> BranchStack;
+  std::vector<StackItem> LoopStack;
+  unsigned MaxStackSize;
+  unsigned CurrentEntries;
+  unsigned CurrentSubEntries;
+
+  CFStack(const AMDGPUSubtarget &st, unsigned ShaderType) : ST(st),
+      // We need to reserve a stack entry for CALL_FS in vertex shaders.
+      MaxStackSize(ShaderType == ShaderType::VERTEX ? 1 : 0),
+      CurrentEntries(0), CurrentSubEntries(0) { }
+
+  unsigned getLoopDepth();
+  bool branchStackContains(CFStack::StackItem);
+  bool requiresWorkAroundForInst(unsigned Opcode);
+  unsigned getSubEntrySize(CFStack::StackItem Item);
+  void updateMaxStackSize();
+  void pushBranch(unsigned Opcode, bool isWQM = false);
+  void pushLoop();
+  void popBranch();
+  void popLoop();
+};
+
+unsigned CFStack::getLoopDepth() {
+  return LoopStack.size();
+}
+
+bool CFStack::branchStackContains(CFStack::StackItem Item) {
+  for (std::vector<CFStack::StackItem>::const_iterator I = BranchStack.begin(),
+       E = BranchStack.end(); I != E; ++I) {
+    if (*I == Item)
+      return true;
+  }
+  return false;
+}
+
+bool CFStack::requiresWorkAroundForInst(unsigned Opcode) {
+  if (Opcode == AMDGPU::CF_ALU_PUSH_BEFORE && ST.hasCaymanISA() &&
+      getLoopDepth() > 1)
+    return true;
+
+  if (!ST.hasCFAluBug())
+    return false;
+
+  switch(Opcode) {
+  default: return false;
+  case AMDGPU::CF_ALU_PUSH_BEFORE:
+  case AMDGPU::CF_ALU_ELSE_AFTER:
+  case AMDGPU::CF_ALU_BREAK:
+  case AMDGPU::CF_ALU_CONTINUE:
+    if (CurrentSubEntries == 0)
+      return false;
+    if (ST.getWavefrontSize() == 64) {
+      // We are being conservative here.  We only require this work-around if
+      // CurrentSubEntries > 3 &&
+      // (CurrentSubEntries % 4 == 3 || CurrentSubEntries % 4 == 0)
+      //
+      // We have to be conservative, because we don't know for certain that
+      // our stack allocation algorithm for Evergreen/NI is correct.  Applying this
+      // work-around when CurrentSubEntries > 3 allows us to over-allocate stack
+      // resources without any problems.
+      return CurrentSubEntries > 3;
+    } else {
+      assert(ST.getWavefrontSize() == 32);
+      // We are being conservative here.  We only require the work-around if
+      // CurrentSubEntries > 7 &&
+      // (CurrentSubEntries % 8 == 7 || CurrentSubEntries % 8 == 0)
+      // See the comment on the wavefront size == 64 case for why we are
+      // being conservative.
+      return CurrentSubEntries > 7;
+    }
+  }
+}
+
+unsigned CFStack::getSubEntrySize(CFStack::StackItem Item) {
+  switch(Item) {
+  default:
+    return 0;
+  case CFStack::FIRST_NON_WQM_PUSH:
+  assert(!ST.hasCaymanISA());
+  if (ST.getGeneration() <= AMDGPUSubtarget::R700) {
+    // +1 For the push operation.
+    // +2 Extra space required.
+    return 3;
+  } else {
+    // Some documentation says that this is not necessary on Evergreen,
+    // but experimentation has show that we need to allocate 1 extra
+    // sub-entry for the first non-WQM push.
+    // +1 For the push operation.
+    // +1 Extra space required.
+    return 2;
+  }
+  case CFStack::FIRST_NON_WQM_PUSH_W_FULL_ENTRY:
+    assert(ST.getGeneration() >= AMDGPUSubtarget::EVERGREEN);
+    // +1 For the push operation.
+    // +1 Extra space required.
+    return 2;
+  case CFStack::SUB_ENTRY:
+    return 1;
+  }
+}
+
+void CFStack::updateMaxStackSize() {
+  unsigned CurrentStackSize = CurrentEntries +
+                              (RoundUpToAlignment(CurrentSubEntries, 4) / 4);
+  MaxStackSize = std::max(CurrentStackSize, MaxStackSize);
+}
+
+void CFStack::pushBranch(unsigned Opcode, bool isWQM) {
+  CFStack::StackItem Item = CFStack::ENTRY;
+  switch(Opcode) {
+  case AMDGPU::CF_PUSH_EG:
+  case AMDGPU::CF_ALU_PUSH_BEFORE:
+    if (!isWQM) {
+      if (!ST.hasCaymanISA() && !branchStackContains(CFStack::FIRST_NON_WQM_PUSH))
+        Item = CFStack::FIRST_NON_WQM_PUSH;  // May not be required on Evergreen/NI
+                                             // See comment in
+                                             // CFStack::getSubEntrySize()
+      else if (CurrentEntries > 0 &&
+               ST.getGeneration() > AMDGPUSubtarget::EVERGREEN &&
+               !ST.hasCaymanISA() &&
+               !branchStackContains(CFStack::FIRST_NON_WQM_PUSH_W_FULL_ENTRY))
+        Item = CFStack::FIRST_NON_WQM_PUSH_W_FULL_ENTRY;
+      else
+        Item = CFStack::SUB_ENTRY;
+    } else
+      Item = CFStack::ENTRY;
+    break;
+  }
+  BranchStack.push_back(Item);
+  if (Item == CFStack::ENTRY)
+    CurrentEntries++;
+  else
+    CurrentSubEntries += getSubEntrySize(Item);
+  updateMaxStackSize();
+}
+
+void CFStack::pushLoop() {
+  LoopStack.push_back(CFStack::ENTRY);
+  CurrentEntries++;
+  updateMaxStackSize();
+}
+
+void CFStack::popBranch() {
+  CFStack::StackItem Top = BranchStack.back();
+  if (Top == CFStack::ENTRY)
+    CurrentEntries--;
+  else
+    CurrentSubEntries-= getSubEntrySize(Top);
+  BranchStack.pop_back();
+}
+
+void CFStack::popLoop() {
+  CurrentEntries--;
+  LoopStack.pop_back();
+}
+
 class R600ControlFlowFinalizer : public MachineFunctionPass {
 
 private:
@@ -300,51 +468,30 @@ private:
     }
   }
 
-  unsigned getHWStackSize(unsigned StackSubEntry, bool hasPush) const {
-    switch (ST.getGeneration()) {
-    case AMDGPUSubtarget::R600:
-    case AMDGPUSubtarget::R700:
-      if (hasPush)
-        StackSubEntry += 2;
-      break;
-    case AMDGPUSubtarget::EVERGREEN:
-      if (hasPush)
-        StackSubEntry ++;
-    case AMDGPUSubtarget::NORTHERN_ISLANDS:
-      StackSubEntry += 2;
-      break;
-    default: llvm_unreachable("Not a VLIW4/VLIW5 GPU");
-    }
-    return (StackSubEntry + 3)/4; // Need ceil value of StackSubEntry/4
-  }
-
 public:
   R600ControlFlowFinalizer(TargetMachine &tm) : MachineFunctionPass(ID),
-    TII (0), TRI(0),
+    TII (nullptr), TRI(nullptr),
     ST(tm.getSubtarget<AMDGPUSubtarget>()) {
       const AMDGPUSubtarget &ST = tm.getSubtarget<AMDGPUSubtarget>();
       MaxFetchInst = ST.getTexVTXClauseSize();
   }
 
-  virtual bool runOnMachineFunction(MachineFunction &MF) {
+  bool runOnMachineFunction(MachineFunction &MF) override {
     TII=static_cast<const R600InstrInfo *>(MF.getTarget().getInstrInfo());
     TRI=static_cast<const R600RegisterInfo *>(MF.getTarget().getRegisterInfo());
+    R600MachineFunctionInfo *MFI = MF.getInfo<R600MachineFunctionInfo>();
 
-    unsigned MaxStack = 0;
-    unsigned CurrentStack = 0;
-    bool HasPush = false;
+    CFStack CFStack(ST, MFI->getShaderType());
     for (MachineFunction::iterator MB = MF.begin(), ME = MF.end(); MB != ME;
         ++MB) {
       MachineBasicBlock &MBB = *MB;
       unsigned CfCount = 0;
       std::vector<std::pair<unsigned, std::set<MachineInstr *> > > LoopStack;
       std::vector<MachineInstr * > IfThenElseStack;
-      R600MachineFunctionInfo *MFI = MF.getInfo<R600MachineFunctionInfo>();
-      if (MFI->ShaderType == 1) {
+      if (MFI->getShaderType() == ShaderType::VERTEX) {
         BuildMI(MBB, MBB.begin(), MBB.findDebugLoc(MBB.begin()),
             getHWInstrDesc(CF_CALL_FS));
         CfCount++;
-        MaxStack = 1;
       }
       std::vector<ClauseFile> FetchClauses, AluClauses;
       std::vector<MachineInstr *> LastAlu(1);
@@ -356,21 +503,31 @@ public:
           DEBUG(dbgs() << CfCount << ":"; I->dump(););
           FetchClauses.push_back(MakeFetchClause(MBB, I));
           CfCount++;
-          LastAlu.back() = 0;
+          LastAlu.back() = nullptr;
           continue;
         }
 
         MachineBasicBlock::iterator MI = I;
         if (MI->getOpcode() != AMDGPU::ENDIF)
-          LastAlu.back() = 0;
+          LastAlu.back() = nullptr;
         if (MI->getOpcode() == AMDGPU::CF_ALU)
           LastAlu.back() = MI;
         I++;
+        bool RequiresWorkAround =
+            CFStack.requiresWorkAroundForInst(MI->getOpcode());
         switch (MI->getOpcode()) {
         case AMDGPU::CF_ALU_PUSH_BEFORE:
-          CurrentStack++;
-          MaxStack = std::max(MaxStack, CurrentStack);
-          HasPush = true;
+          if (RequiresWorkAround) {
+            DEBUG(dbgs() << "Applying bug work-around for ALU_PUSH_BEFORE\n");
+            BuildMI(MBB, MI, MBB.findDebugLoc(MI), TII->get(AMDGPU::CF_PUSH_EG))
+                .addImm(CfCount + 1)
+                .addImm(1);
+            MI->setDesc(TII->get(AMDGPU::CF_ALU));
+            CfCount++;
+            CFStack.pushBranch(AMDGPU::CF_PUSH_EG);
+          } else
+            CFStack.pushBranch(AMDGPU::CF_ALU_PUSH_BEFORE);
+
         case AMDGPU::CF_ALU:
           I = MI;
           AluClauses.push_back(MakeALUClause(MBB, I));
@@ -378,8 +535,7 @@ public:
           CfCount++;
           break;
         case AMDGPU::WHILELOOP: {
-          CurrentStack+=4;
-          MaxStack = std::max(MaxStack, CurrentStack);
+          CFStack.pushLoop();
           MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
               getHWInstrDesc(CF_WHILE_LOOP))
               .addImm(1);
@@ -392,7 +548,7 @@ public:
           break;
         }
         case AMDGPU::ENDLOOP: {
-          CurrentStack-=4;
+          CFStack.popLoop();
           std::pair<unsigned, std::set<MachineInstr *> > Pair =
               LoopStack.back();
           LoopStack.pop_back();
@@ -404,7 +560,7 @@ public:
           break;
         }
         case AMDGPU::IF_PREDICATE_SET: {
-          LastAlu.push_back(0);
+          LastAlu.push_back(nullptr);
           MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
               getHWInstrDesc(CF_JUMP))
               .addImm(0)
@@ -430,7 +586,7 @@ public:
           break;
         }
         case AMDGPU::ENDIF: {
-          CurrentStack--;
+          CFStack.popBranch();
           if (LastAlu.back()) {
             ToPopAfter.push_back(LastAlu.back());
           } else {
@@ -505,13 +661,13 @@ public:
             .addImm(Alu->getOperand(8).getImm());
         Alu->eraseFromParent();
       }
-      MFI->StackSize = getHWStackSize(MaxStack, HasPush);
+      MFI->StackSize = CFStack.MaxStackSize;
     }
 
     return false;
   }
 
-  const char *getPassName() const {
+  const char *getPassName() const override {
     return "R600 Control Flow Finalizer Pass";
   }
 };
diff --git a/contrib/llvm/lib/Target/R600/R600Defines.h b/contrib/llvm/lib/Target/R600/R600Defines.h
index 1781f2a..f2f28fe 100644
--- a/contrib/llvm/lib/Target/R600/R600Defines.h
+++ b/contrib/llvm/lib/Target/R600/R600Defines.h
@@ -52,7 +52,7 @@ namespace R600_InstFlag {
 
 #define HAS_NATIVE_OPERANDS(Flags) ((Flags) & R600_InstFlag::NATIVE_OPERANDS)
 
-/// \brief Defines for extracting register infomation from register encoding
+/// \brief Defines for extracting register information from register encoding
 #define HW_REG_MASK 0x1ff
 #define HW_CHAN_SHIFT 9
 
diff --git a/contrib/llvm/lib/Target/R600/R600EmitClauseMarkers.cpp b/contrib/llvm/lib/Target/R600/R600EmitClauseMarkers.cpp
index 1bbfd2b..38afebe 100644
--- a/contrib/llvm/lib/Target/R600/R600EmitClauseMarkers.cpp
+++ b/contrib/llvm/lib/Target/R600/R600EmitClauseMarkers.cpp
@@ -25,12 +25,15 @@
 
 using namespace llvm;
 
+namespace llvm {
+  void initializeR600EmitClauseMarkersPass(PassRegistry&);
+}
+
 namespace {
 
-class R600EmitClauseMarkersPass : public MachineFunctionPass {
+class R600EmitClauseMarkers : public MachineFunctionPass {
 
 private:
-  static char ID;
   const R600InstrInfo *TII;
   int Address;
 
@@ -287,10 +290,13 @@ private:
   }
 
 public:
-  R600EmitClauseMarkersPass(TargetMachine &tm) : MachineFunctionPass(ID),
-    TII(0), Address(0) { }
+  static char ID;
+  R600EmitClauseMarkers() : MachineFunctionPass(ID), TII(nullptr), Address(0) {
+
+    initializeR600EmitClauseMarkersPass(*PassRegistry::getPassRegistry());
+  }
 
-  virtual bool runOnMachineFunction(MachineFunction &MF) {
+  bool runOnMachineFunction(MachineFunction &MF) override {
     TII = static_cast<const R600InstrInfo *>(MF.getTarget().getInstrInfo());
 
     for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
@@ -309,17 +315,21 @@ public:
     return false;
   }
 
-  const char *getPassName() const {
+  const char *getPassName() const override {
     return "R600 Emit Clause Markers Pass";
   }
 };
 
-char R600EmitClauseMarkersPass::ID = 0;
+char R600EmitClauseMarkers::ID = 0;
 
 } // end anonymous namespace
 
+INITIALIZE_PASS_BEGIN(R600EmitClauseMarkers, "emitclausemarkers",
+                      "R600 Emit Clause Markters", false, false)
+INITIALIZE_PASS_END(R600EmitClauseMarkers, "emitclausemarkers",
+                      "R600 Emit Clause Markters", false, false)
 
-llvm::FunctionPass *llvm::createR600EmitClauseMarkers(TargetMachine &TM) {
-  return new R600EmitClauseMarkersPass(TM);
+llvm::FunctionPass *llvm::createR600EmitClauseMarkers() {
+  return new R600EmitClauseMarkers();
 }
 
diff --git a/contrib/llvm/lib/Target/R600/R600ExpandSpecialInstrs.cpp b/contrib/llvm/lib/Target/R600/R600ExpandSpecialInstrs.cpp
index aeee4aa..732b06d 100644
--- a/contrib/llvm/lib/Target/R600/R600ExpandSpecialInstrs.cpp
+++ b/contrib/llvm/lib/Target/R600/R600ExpandSpecialInstrs.cpp
@@ -33,16 +33,16 @@ private:
   static char ID;
   const R600InstrInfo *TII;
 
-  bool ExpandInputPerspective(MachineInstr& MI);
-  bool ExpandInputConstant(MachineInstr& MI);
+  void SetFlagInNewMI(MachineInstr *NewMI, const MachineInstr *OldMI,
+      unsigned Op);
 
 public:
   R600ExpandSpecialInstrsPass(TargetMachine &tm) : MachineFunctionPass(ID),
-    TII(0) { }
+    TII(nullptr) { }
 
-  virtual bool runOnMachineFunction(MachineFunction &MF);
+  bool runOnMachineFunction(MachineFunction &MF) override;
 
-  const char *getPassName() const {
+  const char *getPassName() const override {
     return "R600 Expand special instructions pass";
   }
 };
@@ -55,6 +55,15 @@ FunctionPass *llvm::createR600ExpandSpecialInstrsPass(TargetMachine &TM) {
   return new R600ExpandSpecialInstrsPass(TM);
 }
 
+void R600ExpandSpecialInstrsPass::SetFlagInNewMI(MachineInstr *NewMI,
+    const MachineInstr *OldMI, unsigned Op) {
+  int OpIdx = TII->getOperandIdx(*OldMI, Op);
+  if (OpIdx > -1) {
+    uint64_t Val = OldMI->getOperand(OpIdx).getImm();
+    TII->setImmOperand(NewMI, Op, Val);
+  }
+}
+
 bool R600ExpandSpecialInstrsPass::runOnMachineFunction(MachineFunction &MF) {
   TII = static_cast<const R600InstrInfo *>(MF.getTarget().getInstrInfo());
 
@@ -66,7 +75,7 @@ bool R600ExpandSpecialInstrsPass::runOnMachineFunction(MachineFunction &MF) {
     MachineBasicBlock::iterator I = MBB.begin();
     while (I != MBB.end()) {
       MachineInstr &MI = *I;
-      I = llvm::next(I);
+      I = std::next(I);
 
       // Expand LDS_*_RET instructions
       if (TII->isLDSRetInstr(MI.getOpcode())) {
@@ -325,6 +334,12 @@ bool R600ExpandSpecialInstrsPass::runOnMachineFunction(MachineFunction &MF) {
         if (NotLast) {
           TII->addFlag(NewMI, 0, MO_FLAG_NOT_LAST);
         }
+        SetFlagInNewMI(NewMI, &MI, AMDGPU::OpName::clamp);
+        SetFlagInNewMI(NewMI, &MI, AMDGPU::OpName::literal);
+        SetFlagInNewMI(NewMI, &MI, AMDGPU::OpName::src0_abs);
+        SetFlagInNewMI(NewMI, &MI, AMDGPU::OpName::src1_abs);
+        SetFlagInNewMI(NewMI, &MI, AMDGPU::OpName::src0_neg);
+        SetFlagInNewMI(NewMI, &MI, AMDGPU::OpName::src1_neg);
       }
       MI.eraseFromParent();
     }
diff --git a/contrib/llvm/lib/Target/R600/R600ISelLowering.cpp b/contrib/llvm/lib/Target/R600/R600ISelLowering.cpp
index 0fcb488..52315bf 100644
--- a/contrib/llvm/lib/Target/R600/R600ISelLowering.cpp
+++ b/contrib/llvm/lib/Target/R600/R600ISelLowering.cpp
@@ -13,9 +13,13 @@
 //===----------------------------------------------------------------------===//
 
 #include "R600ISelLowering.h"
+#include "AMDGPUFrameLowering.h"
+#include "AMDGPUIntrinsicInfo.h"
+#include "AMDGPUSubtarget.h"
 #include "R600Defines.h"
 #include "R600InstrInfo.h"
 #include "R600MachineFunctionInfo.h"
+#include "llvm/Analysis/ValueTracking.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
@@ -65,6 +69,7 @@ R600TargetLowering::R600TargetLowering(TargetMachine &TM) :
 
   setOperationAction(ISD::BR_CC, MVT::i32, Expand);
   setOperationAction(ISD::BR_CC, MVT::f32, Expand);
+  setOperationAction(ISD::BRCOND, MVT::Other, Custom);
 
   setOperationAction(ISD::FSUB, MVT::f32, Expand);
 
@@ -78,13 +83,37 @@ R600TargetLowering::R600TargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::SETCC, MVT::i32, Expand);
   setOperationAction(ISD::SETCC, MVT::f32, Expand);
   setOperationAction(ISD::FP_TO_UINT, MVT::i1, Custom);
+  setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
+  setOperationAction(ISD::FP_TO_UINT, MVT::i64, Custom);
 
   setOperationAction(ISD::SELECT, MVT::i32, Expand);
   setOperationAction(ISD::SELECT, MVT::f32, Expand);
   setOperationAction(ISD::SELECT, MVT::v2i32, Expand);
-  setOperationAction(ISD::SELECT, MVT::v2f32, Expand);
   setOperationAction(ISD::SELECT, MVT::v4i32, Expand);
-  setOperationAction(ISD::SELECT, MVT::v4f32, Expand);
+
+  // Expand sign extension of vectors
+  if (!Subtarget->hasBFE())
+    setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
+
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i1, Expand);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i1, Expand);
+
+  if (!Subtarget->hasBFE())
+    setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8, Expand);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i8, Expand);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i8, Expand);
+
+  if (!Subtarget->hasBFE())
+    setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16, Expand);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i16, Expand);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i16, Expand);
+
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i32, Legal);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i32, Expand);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i32, Expand);
+
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::Other, Expand);
+
 
   // Legalize loads and stores to the private address space.
   setOperationAction(ISD::LOAD, MVT::i32, Custom);
@@ -111,14 +140,47 @@ R600TargetLowering::R600TargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::LOAD, MVT::v4i32, Custom);
   setOperationAction(ISD::FrameIndex, MVT::i32, Custom);
 
+  setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2i32, Custom);
+  setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2f32, Custom);
+  setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v4i32, Custom);
+  setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v4f32, Custom);
+
+  setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v2i32, Custom);
+  setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v2f32, Custom);
+  setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4i32, Custom);
+  setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4f32, Custom);
+
   setTargetDAGCombine(ISD::FP_ROUND);
   setTargetDAGCombine(ISD::FP_TO_SINT);
   setTargetDAGCombine(ISD::EXTRACT_VECTOR_ELT);
   setTargetDAGCombine(ISD::SELECT_CC);
   setTargetDAGCombine(ISD::INSERT_VECTOR_ELT);
 
+  setOperationAction(ISD::SUB, MVT::i64, Expand);
+
+  // These should be replaced by UDVIREM, but it does not happen automatically
+  // during Type Legalization
+  setOperationAction(ISD::UDIV, MVT::i64, Custom);
+  setOperationAction(ISD::UREM, MVT::i64, Custom);
+  setOperationAction(ISD::SDIV, MVT::i64, Custom);
+  setOperationAction(ISD::SREM, MVT::i64, Custom);
+
+  // We don't have 64-bit shifts. Thus we need either SHX i64 or SHX_PARTS i32
+  //  to be Legal/Custom in order to avoid library calls.
+  setOperationAction(ISD::SHL_PARTS, MVT::i32, Custom);
+  setOperationAction(ISD::SRL_PARTS, MVT::i32, Custom);
+  setOperationAction(ISD::SRA_PARTS, MVT::i32, Custom);
+
   setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
 
+  const MVT ScalarIntVTs[] = { MVT::i32, MVT::i64 };
+  for (MVT VT : ScalarIntVTs) {
+    setOperationAction(ISD::ADDC, VT, Expand);
+    setOperationAction(ISD::SUBC, VT, Expand);
+    setOperationAction(ISD::ADDE, VT, Expand);
+    setOperationAction(ISD::SUBE, VT, Expand);
+  }
+
   setBooleanContents(ZeroOrNegativeOneBooleanContent);
   setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
   setSchedulingPreference(Sched::Source);
@@ -207,7 +269,7 @@ MachineBasicBlock * R600TargetLowering::EmitInstrWithCustomInserter(
   case AMDGPU::RAT_WRITE_CACHELESS_32_eg:
   case AMDGPU::RAT_WRITE_CACHELESS_64_eg:
   case AMDGPU::RAT_WRITE_CACHELESS_128_eg: {
-    unsigned EOP = (llvm::next(I)->getOpcode() == AMDGPU::RETURN) ? 1 : 0;
+    unsigned EOP = (std::next(I)->getOpcode() == AMDGPU::RETURN) ? 1 : 0;
 
     BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(MI->getOpcode()))
             .addOperand(MI->getOperand(0))
@@ -457,9 +519,9 @@ MachineBasicBlock * R600TargetLowering::EmitInstrWithCustomInserter(
     // Instruction is left unmodified if its not the last one of its type
     bool isLastInstructionOfItsType = true;
     unsigned InstExportType = MI->getOperand(1).getImm();
-    for (MachineBasicBlock::iterator NextExportInst = llvm::next(I),
+    for (MachineBasicBlock::iterator NextExportInst = std::next(I),
          EndBlock = BB->end(); NextExportInst != EndBlock;
-         NextExportInst = llvm::next(NextExportInst)) {
+         NextExportInst = std::next(NextExportInst)) {
       if (NextExportInst->getOpcode() == AMDGPU::EG_ExportSwz ||
           NextExportInst->getOpcode() == AMDGPU::R600_ExportSwz) {
         unsigned CurrentInstExportType = NextExportInst->getOperand(1)
@@ -470,7 +532,7 @@ MachineBasicBlock * R600TargetLowering::EmitInstrWithCustomInserter(
         }
       }
     }
-    bool EOP = (llvm::next(I)->getOpcode() == AMDGPU::RETURN)? 1 : 0;
+    bool EOP = (std::next(I)->getOpcode() == AMDGPU::RETURN) ? 1 : 0;
     if (!EOP && !isLastInstructionOfItsType)
       return BB;
     unsigned CfInst = (MI->getOpcode() == AMDGPU::EG_ExportSwz)? 84 : 40;
@@ -510,11 +572,24 @@ SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const
   R600MachineFunctionInfo *MFI = MF.getInfo<R600MachineFunctionInfo>();
   switch (Op.getOpcode()) {
   default: return AMDGPUTargetLowering::LowerOperation(Op, DAG);
+  case ISD::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR_ELT(Op, DAG);
+  case ISD::INSERT_VECTOR_ELT: return LowerINSERT_VECTOR_ELT(Op, DAG);
+  case ISD::SHL_PARTS: return LowerSHLParts(Op, DAG);
+  case ISD::SRA_PARTS:
+  case ISD::SRL_PARTS: return LowerSRXParts(Op, DAG);
   case ISD::FCOS:
   case ISD::FSIN: return LowerTrig(Op, DAG);
   case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
   case ISD::STORE: return LowerSTORE(Op, DAG);
-  case ISD::LOAD: return LowerLOAD(Op, DAG);
+  case ISD::LOAD: {
+    SDValue Result = LowerLOAD(Op, DAG);
+    assert((!Result.getNode() ||
+            Result.getNode()->getNumValues() == 2) &&
+           "Load should return a value and a chain");
+    return Result;
+  }
+
+  case ISD::BRCOND: return LowerBRCOND(Op, DAG);
   case ISD::GlobalAddress: return LowerGlobalAddress(MFI, Op, DAG);
   case ISD::INTRINSIC_VOID: {
     SDValue Chain = Op.getOperand(0);
@@ -538,8 +613,7 @@ SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const
         DAG.getConstant(2, MVT::i32), // SWZ_Z
         DAG.getConstant(3, MVT::i32) // SWZ_W
       };
-      return DAG.getNode(AMDGPUISD::EXPORT, SDLoc(Op), Op.getValueType(),
-          Args, 8);
+      return DAG.getNode(AMDGPUISD::EXPORT, SDLoc(Op), Op.getValueType(), Args);
     }
 
     // default for switch(IntrinsicID)
@@ -689,7 +763,7 @@ SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const
         Op.getOperand(9),
         Op.getOperand(10)
       };
-      return DAG.getNode(AMDGPUISD::TEXTURE_FETCH, DL, MVT::v4f32, TexArgs, 19);
+      return DAG.getNode(AMDGPUISD::TEXTURE_FETCH, DL, MVT::v4f32, TexArgs);
     }
     case AMDGPUIntrinsic::AMDGPU_dp4: {
       SDValue Args[8] = {
@@ -710,7 +784,7 @@ SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const
       DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32, Op.getOperand(2),
           DAG.getConstant(3, MVT::i32))
       };
-      return DAG.getNode(AMDGPUISD::DOT4, DL, MVT::f32, Args, 8);
+      return DAG.getNode(AMDGPUISD::DOT4, DL, MVT::f32, Args);
     }
 
     case Intrinsic::r600_read_ngroups_x:
@@ -750,6 +824,9 @@ SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const
     case Intrinsic::r600_read_tidig_z:
       return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
                                   AMDGPU::T0_Z, VT);
+    case Intrinsic::AMDGPU_rsq:
+      // XXX - I'm assuming SI's RSQ_LEGACY matches R600's behavior.
+      return DAG.getNode(AMDGPUISD::RSQ_LEGACY, DL, VT, Op.getOperand(1));
     }
     // break out of case ISD::INTRINSIC_WO_CHAIN in switch(Op.getOpcode())
     break;
@@ -762,23 +839,189 @@ void R600TargetLowering::ReplaceNodeResults(SDNode *N,
                                             SmallVectorImpl<SDValue> &Results,
                                             SelectionDAG &DAG) const {
   switch (N->getOpcode()) {
-  default: return;
-  case ISD::FP_TO_UINT: Results.push_back(LowerFPTOUINT(N->getOperand(0), DAG));
+  default:
+    AMDGPUTargetLowering::ReplaceNodeResults(N, Results, DAG);
     return;
-  case ISD::LOAD: {
-    SDNode *Node = LowerLOAD(SDValue(N, 0), DAG).getNode();
-    Results.push_back(SDValue(Node, 0));
-    Results.push_back(SDValue(Node, 1));
-    // XXX: LLVM seems not to replace Chain Value inside CustomWidenLowerNode
-    // function
-    DAG.ReplaceAllUsesOfValueWith(SDValue(N,1), SDValue(Node, 1));
+  case ISD::FP_TO_UINT:
+    if (N->getValueType(0) == MVT::i1) {
+      Results.push_back(LowerFPTOUINT(N->getOperand(0), DAG));
+      return;
+    }
+    // Fall-through. Since we don't care about out of bounds values
+    // we can use FP_TO_SINT for uints too. The DAGLegalizer code for uint
+    // considers some extra cases which are not necessary here.
+  case ISD::FP_TO_SINT: {
+    SDValue Result;
+    if (expandFP_TO_SINT(N, Result, DAG))
+      Results.push_back(Result);
     return;
   }
-  case ISD::STORE:
-    SDNode *Node = LowerSTORE(SDValue(N, 0), DAG).getNode();
-    Results.push_back(SDValue(Node, 0));
-    return;
+  case ISD::UDIV: {
+    SDValue Op = SDValue(N, 0);
+    SDLoc DL(Op);
+    EVT VT = Op.getValueType();
+    SDValue UDIVREM = DAG.getNode(ISD::UDIVREM, DL, DAG.getVTList(VT, VT),
+      N->getOperand(0), N->getOperand(1));
+    Results.push_back(UDIVREM);
+    break;
+  }
+  case ISD::UREM: {
+    SDValue Op = SDValue(N, 0);
+    SDLoc DL(Op);
+    EVT VT = Op.getValueType();
+    SDValue UDIVREM = DAG.getNode(ISD::UDIVREM, DL, DAG.getVTList(VT, VT),
+      N->getOperand(0), N->getOperand(1));
+    Results.push_back(UDIVREM.getValue(1));
+    break;
+  }
+  case ISD::SDIV: {
+    SDValue Op = SDValue(N, 0);
+    SDLoc DL(Op);
+    EVT VT = Op.getValueType();
+    SDValue SDIVREM = DAG.getNode(ISD::SDIVREM, DL, DAG.getVTList(VT, VT),
+      N->getOperand(0), N->getOperand(1));
+    Results.push_back(SDIVREM);
+    break;
+  }
+  case ISD::SREM: {
+    SDValue Op = SDValue(N, 0);
+    SDLoc DL(Op);
+    EVT VT = Op.getValueType();
+    SDValue SDIVREM = DAG.getNode(ISD::SDIVREM, DL, DAG.getVTList(VT, VT),
+      N->getOperand(0), N->getOperand(1));
+    Results.push_back(SDIVREM.getValue(1));
+    break;
+  }
+  case ISD::SDIVREM: {
+    SDValue Op = SDValue(N, 1);
+    SDValue RES = LowerSDIVREM(Op, DAG);
+    Results.push_back(RES);
+    Results.push_back(RES.getValue(1));
+    break;
+  }
+  case ISD::UDIVREM: {
+    SDValue Op = SDValue(N, 0);
+    SDLoc DL(Op);
+    EVT VT = Op.getValueType();
+    EVT HalfVT = VT.getHalfSizedIntegerVT(*DAG.getContext());
+
+    SDValue one = DAG.getConstant(1, HalfVT);
+    SDValue zero = DAG.getConstant(0, HalfVT);
+
+    //HiLo split
+    SDValue LHS = N->getOperand(0);
+    SDValue LHS_Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, LHS, zero);
+    SDValue LHS_Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, LHS, one);
+
+    SDValue RHS = N->getOperand(1);
+    SDValue RHS_Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, RHS, zero);
+    SDValue RHS_Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, RHS, one);
+
+    // Get Speculative values
+    SDValue DIV_Part = DAG.getNode(ISD::UDIV, DL, HalfVT, LHS_Hi, RHS_Lo);
+    SDValue REM_Part = DAG.getNode(ISD::UREM, DL, HalfVT, LHS_Hi, RHS_Lo);
+
+    SDValue REM_Hi = zero;
+    SDValue REM_Lo = DAG.getSelectCC(DL, RHS_Hi, zero, REM_Part, LHS_Hi, ISD::SETEQ);
+
+    SDValue DIV_Hi = DAG.getSelectCC(DL, RHS_Hi, zero, DIV_Part, zero, ISD::SETEQ);
+    SDValue DIV_Lo = zero;
+
+    const unsigned halfBitWidth = HalfVT.getSizeInBits();
+
+    for (unsigned i = 0; i < halfBitWidth; ++i) {
+      SDValue POS = DAG.getConstant(halfBitWidth - i - 1, HalfVT);
+      // Get Value of high bit
+      SDValue HBit;
+      if (halfBitWidth == 32 && Subtarget->hasBFE()) {
+        HBit = DAG.getNode(AMDGPUISD::BFE_U32, DL, HalfVT, LHS_Lo, POS, one);
+      } else {
+        HBit = DAG.getNode(ISD::SRL, DL, HalfVT, LHS_Lo, POS);
+        HBit = DAG.getNode(ISD::AND, DL, HalfVT, HBit, one);
+      }
+
+      SDValue Carry = DAG.getNode(ISD::SRL, DL, HalfVT, REM_Lo,
+        DAG.getConstant(halfBitWidth - 1, HalfVT));
+      REM_Hi = DAG.getNode(ISD::SHL, DL, HalfVT, REM_Hi, one);
+      REM_Hi = DAG.getNode(ISD::OR, DL, HalfVT, REM_Hi, Carry);
+
+      REM_Lo = DAG.getNode(ISD::SHL, DL, HalfVT, REM_Lo, one);
+      REM_Lo = DAG.getNode(ISD::OR, DL, HalfVT, REM_Lo, HBit);
+
+
+      SDValue REM = DAG.getNode(ISD::BUILD_PAIR, DL, VT, REM_Lo, REM_Hi);
+
+      SDValue BIT = DAG.getConstant(1 << (halfBitWidth - i - 1), HalfVT);
+      SDValue realBIT = DAG.getSelectCC(DL, REM, RHS, BIT, zero, ISD::SETGE);
+
+      DIV_Lo = DAG.getNode(ISD::OR, DL, HalfVT, DIV_Lo, realBIT);
+
+      // Update REM
+
+      SDValue REM_sub = DAG.getNode(ISD::SUB, DL, VT, REM, RHS);
+
+      REM = DAG.getSelectCC(DL, REM, RHS, REM_sub, REM, ISD::SETGE);
+      REM_Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, REM, zero);
+      REM_Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, REM, one);
+    }
+
+    SDValue REM = DAG.getNode(ISD::BUILD_PAIR, DL, VT, REM_Lo, REM_Hi);
+    SDValue DIV = DAG.getNode(ISD::BUILD_PAIR, DL, VT, DIV_Lo, DIV_Hi);
+    Results.push_back(DIV);
+    Results.push_back(REM);
+    break;
+  }
+  }
+}
+
+SDValue R600TargetLowering::vectorToVerticalVector(SelectionDAG &DAG,
+                                                   SDValue Vector) const {
+
+  SDLoc DL(Vector);
+  EVT VecVT = Vector.getValueType();
+  EVT EltVT = VecVT.getVectorElementType();
+  SmallVector<SDValue, 8> Args;
+
+  for (unsigned i = 0, e = VecVT.getVectorNumElements();
+                                                           i != e; ++i) {
+    Args.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltVT,
+                               Vector, DAG.getConstant(i, getVectorIdxTy())));
   }
+
+  return DAG.getNode(AMDGPUISD::BUILD_VERTICAL_VECTOR, DL, VecVT, Args);
+}
+
+SDValue R600TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
+                                                    SelectionDAG &DAG) const {
+
+  SDLoc DL(Op);
+  SDValue Vector = Op.getOperand(0);
+  SDValue Index = Op.getOperand(1);
+
+  if (isa<ConstantSDNode>(Index) ||
+      Vector.getOpcode() == AMDGPUISD::BUILD_VERTICAL_VECTOR)
+    return Op;
+
+  Vector = vectorToVerticalVector(DAG, Vector);
+  return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, Op.getValueType(),
+                     Vector, Index);
+}
+
+SDValue R600TargetLowering::LowerINSERT_VECTOR_ELT(SDValue Op,
+                                                   SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  SDValue Vector = Op.getOperand(0);
+  SDValue Value = Op.getOperand(1);
+  SDValue Index = Op.getOperand(2);
+
+  if (isa<ConstantSDNode>(Index) ||
+      Vector.getOpcode() == AMDGPUISD::BUILD_VERTICAL_VECTOR)
+    return Op;
+
+  Vector = vectorToVerticalVector(DAG, Vector);
+  SDValue Insert = DAG.getNode(ISD::INSERT_VECTOR_ELT, DL, Op.getValueType(),
+                               Vector, Value, Index);
+  return vectorToVerticalVector(DAG, Insert);
 }
 
 SDValue R600TargetLowering::LowerTrig(SDValue Op, SelectionDAG &DAG) const {
@@ -812,6 +1055,80 @@ SDValue R600TargetLowering::LowerTrig(SDValue Op, SelectionDAG &DAG) const {
       DAG.getConstantFP(3.14159265359, MVT::f32));
 }
 
+SDValue R600TargetLowering::LowerSHLParts(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  EVT VT = Op.getValueType();
+
+  SDValue Lo = Op.getOperand(0);
+  SDValue Hi = Op.getOperand(1);
+  SDValue Shift = Op.getOperand(2);
+  SDValue Zero = DAG.getConstant(0, VT);
+  SDValue One  = DAG.getConstant(1, VT);
+
+  SDValue Width  = DAG.getConstant(VT.getSizeInBits(), VT);
+  SDValue Width1 = DAG.getConstant(VT.getSizeInBits() - 1, VT);
+  SDValue BigShift  = DAG.getNode(ISD::SUB, DL, VT, Shift, Width);
+  SDValue CompShift = DAG.getNode(ISD::SUB, DL, VT, Width1, Shift);
+
+  // The dance around Width1 is necessary for 0 special case.
+  // Without it the CompShift might be 32, producing incorrect results in
+  // Overflow. So we do the shift in two steps, the alternative is to
+  // add a conditional to filter the special case.
+
+  SDValue Overflow = DAG.getNode(ISD::SRL, DL, VT, Lo, CompShift);
+  Overflow = DAG.getNode(ISD::SRL, DL, VT, Overflow, One);
+
+  SDValue HiSmall = DAG.getNode(ISD::SHL, DL, VT, Hi, Shift);
+  HiSmall = DAG.getNode(ISD::OR, DL, VT, HiSmall, Overflow);
+  SDValue LoSmall = DAG.getNode(ISD::SHL, DL, VT, Lo, Shift);
+
+  SDValue HiBig = DAG.getNode(ISD::SHL, DL, VT, Lo, BigShift);
+  SDValue LoBig = Zero;
+
+  Hi = DAG.getSelectCC(DL, Shift, Width, HiSmall, HiBig, ISD::SETULT);
+  Lo = DAG.getSelectCC(DL, Shift, Width, LoSmall, LoBig, ISD::SETULT);
+
+  return DAG.getNode(ISD::MERGE_VALUES, DL, DAG.getVTList(VT,VT), Lo, Hi);
+}
+
+SDValue R600TargetLowering::LowerSRXParts(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  EVT VT = Op.getValueType();
+
+  SDValue Lo = Op.getOperand(0);
+  SDValue Hi = Op.getOperand(1);
+  SDValue Shift = Op.getOperand(2);
+  SDValue Zero = DAG.getConstant(0, VT);
+  SDValue One  = DAG.getConstant(1, VT);
+
+  const bool SRA = Op.getOpcode() == ISD::SRA_PARTS;
+
+  SDValue Width  = DAG.getConstant(VT.getSizeInBits(), VT);
+  SDValue Width1 = DAG.getConstant(VT.getSizeInBits() - 1, VT);
+  SDValue BigShift  = DAG.getNode(ISD::SUB, DL, VT, Shift, Width);
+  SDValue CompShift = DAG.getNode(ISD::SUB, DL, VT, Width1, Shift);
+
+  // The dance around Width1 is necessary for 0 special case.
+  // Without it the CompShift might be 32, producing incorrect results in
+  // Overflow. So we do the shift in two steps, the alternative is to
+  // add a conditional to filter the special case.
+
+  SDValue Overflow = DAG.getNode(ISD::SHL, DL, VT, Hi, CompShift);
+  Overflow = DAG.getNode(ISD::SHL, DL, VT, Overflow, One);
+
+  SDValue HiSmall = DAG.getNode(SRA ? ISD::SRA : ISD::SRL, DL, VT, Hi, Shift);
+  SDValue LoSmall = DAG.getNode(ISD::SRL, DL, VT, Lo, Shift);
+  LoSmall = DAG.getNode(ISD::OR, DL, VT, LoSmall, Overflow);
+
+  SDValue LoBig = DAG.getNode(SRA ? ISD::SRA : ISD::SRL, DL, VT, Hi, BigShift);
+  SDValue HiBig = SRA ? DAG.getNode(ISD::SRA, DL, VT, Hi, Width1) : Zero;
+
+  Hi = DAG.getSelectCC(DL, Shift, Width, HiSmall, HiBig, ISD::SETULT);
+  Lo = DAG.getSelectCC(DL, Shift, Width, LoSmall, LoBig, ISD::SETULT);
+
+  return DAG.getNode(ISD::MERGE_VALUES, DL, DAG.getVTList(VT,VT), Lo, Hi);
+}
+
 SDValue R600TargetLowering::LowerFPTOUINT(SDValue Op, SelectionDAG &DAG) const {
   return DAG.getNode(
       ISD::SETCC,
@@ -958,13 +1275,6 @@ SDValue R600TargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const
     return DAG.getNode(ISD::BITCAST, DL, VT, SelectNode);
   }
 
-
-  // Possible Min/Max pattern
-  SDValue MinMax = LowerMinMax(Op, DAG);
-  if (MinMax.getNode()) {
-    return MinMax;
-  }
-
   // If we make it this for it means we have no native instructions to handle
   // this SELECT_CC, so we must lower it.
   SDValue HWTrue, HWFalse;
@@ -977,7 +1287,7 @@ SDValue R600TargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const
     HWFalse = DAG.getConstant(0, CompareVT);
   }
   else {
-    assert(!"Unhandled value type in LowerSELECT_CC");
+    llvm_unreachable("Unhandled value type in LowerSELECT_CC");
   }
 
   // Lower this unsupported SELECT_CC into a combination of two supported
@@ -990,7 +1300,7 @@ SDValue R600TargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const
       DAG.getCondCode(ISD::SETNE));
 }
 
-/// LLVM generates byte-addresed pointers.  For indirect addressing, we need to
+/// LLVM generates byte-addressed pointers.  For indirect addressing, we need to
 /// convert these pointers to a register index.  Each register holds
 /// 16 bytes, (4 x 32bit sub-register), but we need to take into account the
 /// \p StackWidth, which tells us how many of the 4 sub-registrers will be used
@@ -1086,10 +1396,10 @@ SDValue R600TargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
         DAG.getConstant(0, MVT::i32),
         Mask
       };
-      SDValue Input = DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v4i32, Src, 4);
+      SDValue Input = DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v4i32, Src);
       SDValue Args[3] = { Chain, Input, DWordAddr };
       return DAG.getMemIntrinsicNode(AMDGPUISD::STORE_MSKOR, DL,
-                                     Op->getVTList(), Args, 3, MemVT,
+                                     Op->getVTList(), Args, MemVT,
                                      StoreNode->getMemOperand());
     } else if (Ptr->getOpcode() != AMDGPUISD::DWORDADDR &&
                Value.getValueType().bitsGE(MVT::i32)) {
@@ -1099,7 +1409,7 @@ SDValue R600TargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
                                     Ptr, DAG.getConstant(2, MVT::i32)));
 
       if (StoreNode->isTruncatingStore() || StoreNode->isIndexed()) {
-        assert(!"Truncated and indexed stores not supported yet");
+        llvm_unreachable("Truncated and indexed stores not supported yet");
       } else {
         Chain = DAG.getStore(Chain, DL, Value, Ptr, StoreNode->getMemOperand());
       }
@@ -1113,6 +1423,10 @@ SDValue R600TargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
     return SDValue();
   }
 
+  SDValue Ret = AMDGPUTargetLowering::LowerSTORE(Op, DAG);
+  if (Ret.getNode()) {
+    return Ret;
+  }
   // Lowering for indirect addressing
 
   const MachineFunction &MF = DAG.getMachineFunction();
@@ -1125,7 +1439,7 @@ SDValue R600TargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
   if (ValueVT.isVector()) {
     unsigned NumElemVT = ValueVT.getVectorNumElements();
     EVT ElemVT = ValueVT.getVectorElementType();
-    SDValue Stores[4];
+    SmallVector<SDValue, 4> Stores(NumElemVT);
 
     assert(NumElemVT >= StackWidth && "Stack width cannot be greater than "
                                       "vector width in load");
@@ -1142,7 +1456,7 @@ SDValue R600TargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
                               Chain, Elem, Ptr,
                               DAG.getTargetConstant(Channel, MVT::i32));
     }
-     Chain =  DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Stores, NumElemVT);
+     Chain =  DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Stores);
    } else {
     if (ValueVT == MVT::i8) {
       Value = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i32, Value);
@@ -1204,12 +1518,35 @@ SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
   SDValue Ptr = Op.getOperand(1);
   SDValue LoweredLoad;
 
+  SDValue Ret = AMDGPUTargetLowering::LowerLOAD(Op, DAG);
+  if (Ret.getNode()) {
+    SDValue Ops[2] = {
+      Ret,
+      Chain
+    };
+    return DAG.getMergeValues(Ops, DL);
+  }
+
+  // Lower loads constant address space global variable loads
+  if (LoadNode->getAddressSpace() == AMDGPUAS::CONSTANT_ADDRESS &&
+      isa<GlobalVariable>(
+          GetUnderlyingObject(LoadNode->getMemOperand()->getValue()))) {
+
+    SDValue Ptr = DAG.getZExtOrTrunc(LoadNode->getBasePtr(), DL,
+        getPointerTy(AMDGPUAS::PRIVATE_ADDRESS));
+    Ptr = DAG.getNode(ISD::SRL, DL, MVT::i32, Ptr,
+        DAG.getConstant(2, MVT::i32));
+    return DAG.getNode(AMDGPUISD::REGISTER_LOAD, DL, Op->getVTList(),
+                       LoadNode->getChain(), Ptr,
+                       DAG.getTargetConstant(0, MVT::i32), Op.getOperand(2));
+  }
+
   if (LoadNode->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS && VT.isVector()) {
     SDValue MergedValues[2] = {
       SplitVectorLoad(Op, DAG),
       Chain
     };
-    return DAG.getMergeValues(MergedValues, 2, DL);
+    return DAG.getMergeValues(MergedValues, DL);
   }
 
   int ConstantBlock = ConstantAddressBlock(LoadNode->getAddressSpace());
@@ -1217,8 +1554,8 @@ SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
       ((LoadNode->getExtensionType() == ISD::NON_EXTLOAD) ||
        (LoadNode->getExtensionType() == ISD::ZEXTLOAD))) {
     SDValue Result;
-    if (isa<ConstantExpr>(LoadNode->getSrcValue()) ||
-        isa<Constant>(LoadNode->getSrcValue()) ||
+    if (isa<ConstantExpr>(LoadNode->getMemOperand()->getValue()) ||
+        isa<Constant>(LoadNode->getMemOperand()->getValue()) ||
         isa<ConstantSDNode>(Ptr)) {
       SDValue Slots[4];
       for (unsigned i = 0; i < 4; i++) {
@@ -1237,9 +1574,10 @@ SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
         NewVT = VT;
         NumElements = VT.getVectorNumElements();
       }
-      Result = DAG.getNode(ISD::BUILD_VECTOR, DL, NewVT, Slots, NumElements);
+      Result = DAG.getNode(ISD::BUILD_VECTOR, DL, NewVT,
+                           makeArrayRef(Slots, NumElements));
     } else {
-      // non constant ptr cant be folded, keeps it as a v4f32 load
+      // non-constant ptr can't be folded, keeps it as a v4f32 load
       Result = DAG.getNode(AMDGPUISD::CONST_ADDRESS, DL, MVT::v4i32,
           DAG.getNode(ISD::SRL, DL, MVT::i32, Ptr, DAG.getConstant(4, MVT::i32)),
           DAG.getConstant(LoadNode->getAddressSpace() -
@@ -1253,10 +1591,10 @@ SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
     }
 
     SDValue MergedValues[2] = {
-        Result,
-        Chain
+      Result,
+      Chain
     };
-    return DAG.getMergeValues(MergedValues, 2, DL);
+    return DAG.getMergeValues(MergedValues, DL);
   }
 
   // For most operations returning SDValue() will result in the node being
@@ -1280,7 +1618,7 @@ SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
     SDValue Sra = DAG.getNode(ISD::SRA, DL, VT, Shl, ShiftAmount);
 
     SDValue MergedValues[2] = { Sra, Chain };
-    return DAG.getMergeValues(MergedValues, 2, DL);
+    return DAG.getMergeValues(MergedValues, DL);
   }
 
   if (LoadNode->getAddressSpace() != AMDGPUAS::PRIVATE_ADDRESS) {
@@ -1317,7 +1655,7 @@ SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
       Loads[i] = DAG.getUNDEF(ElemVT);
     }
     EVT TargetVT = EVT::getVectorVT(*DAG.getContext(), ElemVT, 4);
-    LoweredLoad = DAG.getNode(ISD::BUILD_VECTOR, DL, TargetVT, Loads, 4);
+    LoweredLoad = DAG.getNode(ISD::BUILD_VECTOR, DL, TargetVT, Loads);
   } else {
     LoweredLoad = DAG.getNode(AMDGPUISD::REGISTER_LOAD, DL, VT,
                               Chain, Ptr,
@@ -1325,11 +1663,21 @@ SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
                               Op.getOperand(2));
   }
 
-  SDValue Ops[2];
-  Ops[0] = LoweredLoad;
-  Ops[1] = Chain;
+  SDValue Ops[2] = {
+    LoweredLoad,
+    Chain
+  };
 
-  return DAG.getMergeValues(Ops, 2, DL);
+  return DAG.getMergeValues(Ops, DL);
+}
+
+SDValue R600TargetLowering::LowerBRCOND(SDValue Op, SelectionDAG &DAG) const {
+  SDValue Chain = Op.getOperand(0);
+  SDValue Cond  = Op.getOperand(1);
+  SDValue Jump  = Op.getOperand(2);
+
+  return DAG.getNode(AMDGPUISD::BRANCH_COND, SDLoc(Op), Op.getValueType(),
+                     Chain, Jump, Cond);
 }
 
 /// XXX Only kernel functions are supported, so we can assume for now that
@@ -1346,12 +1694,11 @@ SDValue R600TargetLowering::LowerFormalArguments(
   CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
                  getTargetMachine(), ArgLocs, *DAG.getContext());
   MachineFunction &MF = DAG.getMachineFunction();
-  unsigned ShaderType = MF.getInfo<R600MachineFunctionInfo>()->ShaderType;
+  unsigned ShaderType = MF.getInfo<R600MachineFunctionInfo>()->getShaderType();
 
   SmallVector<ISD::InputArg, 8> LocalIns;
 
-  getOriginalFunctionArgs(DAG, DAG.getMachineFunction().getFunction(), Ins,
-                          LocalIns);
+  getOriginalFunctionArgs(DAG, MF.getFunction(), Ins, LocalIns);
 
   AnalyzeFormalArguments(CCInfo, LocalIns);
 
@@ -1370,34 +1717,45 @@ SDValue R600TargetLowering::LowerFormalArguments(
     PointerType *PtrTy = PointerType::get(VT.getTypeForEVT(*DAG.getContext()),
                                                    AMDGPUAS::CONSTANT_BUFFER_0);
 
+    // i64 isn't a legal type, so the register type used ends up as i32, which
+    // isn't expected here. It attempts to create this sextload, but it ends up
+    // being invalid. Somehow this seems to work with i64 arguments, but breaks
+    // for <1 x i64>.
+
     // The first 36 bytes of the input buffer contains information about
     // thread group and global sizes.
-    SDValue Arg = DAG.getExtLoad(ISD::SEXTLOAD, DL, VT, Chain,
+
+    // FIXME: This should really check the extload type, but the handling of
+    // extload vecto parameters seems to be broken.
+    //ISD::LoadExtType Ext = Ins[i].Flags.isSExt() ? ISD::SEXTLOAD : ISD::ZEXTLOAD;
+    ISD::LoadExtType Ext = ISD::SEXTLOAD;
+    SDValue Arg = DAG.getExtLoad(Ext, DL, VT, Chain,
                                  DAG.getConstant(36 + VA.getLocMemOffset(), MVT::i32),
                                  MachinePointerInfo(UndefValue::get(PtrTy)),
                                  MemVT, false, false, 4);
-                                 // 4 is the prefered alignment for
-                                 // the CONSTANT memory space.
+
+    // 4 is the preferred alignment for the CONSTANT memory space.
     InVals.push_back(Arg);
   }
   return Chain;
 }
 
 EVT R600TargetLowering::getSetCCResultType(LLVMContext &, EVT VT) const {
-   if (!VT.isVector()) return MVT::i32;
+   if (!VT.isVector())
+     return MVT::i32;
    return VT.changeVectorElementTypeToInteger();
 }
 
-static SDValue
-CompactSwizzlableVector(SelectionDAG &DAG, SDValue VectorEntry,
-                        DenseMap<unsigned, unsigned> &RemapSwizzle) {
+static SDValue CompactSwizzlableVector(
+  SelectionDAG &DAG, SDValue VectorEntry,
+  DenseMap<unsigned, unsigned> &RemapSwizzle) {
   assert(VectorEntry.getOpcode() == ISD::BUILD_VECTOR);
   assert(RemapSwizzle.empty());
   SDValue NewBldVec[4] = {
-      VectorEntry.getOperand(0),
-      VectorEntry.getOperand(1),
-      VectorEntry.getOperand(2),
-      VectorEntry.getOperand(3)
+    VectorEntry.getOperand(0),
+    VectorEntry.getOperand(1),
+    VectorEntry.getOperand(2),
+    VectorEntry.getOperand(3)
   };
 
   for (unsigned i = 0; i < 4; i++) {
@@ -1428,7 +1786,7 @@ CompactSwizzlableVector(SelectionDAG &DAG, SDValue VectorEntry,
   }
 
   return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(VectorEntry),
-      VectorEntry.getValueType(), NewBldVec, 4);
+                     VectorEntry.getValueType(), NewBldVec);
 }
 
 static SDValue ReorganizeVector(SelectionDAG &DAG, SDValue VectorEntry,
@@ -1442,17 +1800,20 @@ static SDValue ReorganizeVector(SelectionDAG &DAG, SDValue VectorEntry,
       VectorEntry.getOperand(3)
   };
   bool isUnmovable[4] = { false, false, false, false };
-  for (unsigned i = 0; i < 4; i++)
+  for (unsigned i = 0; i < 4; i++) {
     RemapSwizzle[i] = i;
+    if (NewBldVec[i].getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
+      unsigned Idx = dyn_cast<ConstantSDNode>(NewBldVec[i].getOperand(1))
+          ->getZExtValue();
+      if (i == Idx)
+        isUnmovable[Idx] = true;
+    }
+  }
 
   for (unsigned i = 0; i < 4; i++) {
     if (NewBldVec[i].getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
       unsigned Idx = dyn_cast<ConstantSDNode>(NewBldVec[i].getOperand(1))
           ->getZExtValue();
-      if (i == Idx) {
-        isUnmovable[Idx] = true;
-        continue;
-      }
       if (isUnmovable[Idx])
         continue;
       // Swap i and Idx
@@ -1463,7 +1824,7 @@ static SDValue ReorganizeVector(SelectionDAG &DAG, SDValue VectorEntry,
   }
 
   return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(VectorEntry),
-      VectorEntry.getValueType(), NewBldVec, 4);
+                     VectorEntry.getValueType(), NewBldVec);
 }
 
 
@@ -1501,6 +1862,7 @@ SDValue R600TargetLowering::PerformDAGCombine(SDNode *N,
   SelectionDAG &DAG = DCI.DAG;
 
   switch (N->getOpcode()) {
+  default: return AMDGPUTargetLowering::PerformDAGCombine(N, DCI);
   // (f32 fp_round (f64 uint_to_fp a)) -> (f32 uint_to_fp a)
   case ISD::FP_ROUND: {
       SDValue Arg = N->getOperand(0);
@@ -1590,8 +1952,7 @@ SDValue R600TargetLowering::PerformDAGCombine(SDNode *N,
     }
 
     // Return the new vector
-    return DAG.getNode(ISD::BUILD_VECTOR, dl,
-                       VT, &Ops[0], Ops.size());
+    return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Ops);
   }
 
   // Extract_vec (Build_vector) generated by custom lowering
@@ -1615,6 +1976,11 @@ SDValue R600TargetLowering::PerformDAGCombine(SDNode *N,
   }
 
   case ISD::SELECT_CC: {
+    // Try common optimizations
+    SDValue Ret = AMDGPUTargetLowering::PerformDAGCombine(N, DCI);
+    if (Ret.getNode())
+      return Ret;
+
     // fold selectcc (selectcc x, y, a, b, cc), b, a, b, seteq ->
     //      selectcc x, y, a, b, inv(cc)
     //
@@ -1674,7 +2040,7 @@ SDValue R600TargetLowering::PerformDAGCombine(SDNode *N,
     };
     SDLoc DL(N);
     NewArgs[1] = OptimizeSwizzle(N->getOperand(1), &NewArgs[4], DAG);
-    return DAG.getNode(AMDGPUISD::EXPORT, DL, N->getVTList(), NewArgs, 8);
+    return DAG.getNode(AMDGPUISD::EXPORT, DL, N->getVTList(), NewArgs);
   }
   case AMDGPUISD::TEXTURE_FETCH: {
     SDValue Arg = N->getOperand(1);
@@ -1704,10 +2070,11 @@ SDValue R600TargetLowering::PerformDAGCombine(SDNode *N,
     };
     NewArgs[1] = OptimizeSwizzle(N->getOperand(1), &NewArgs[2], DAG);
     return DAG.getNode(AMDGPUISD::TEXTURE_FETCH, SDLoc(N), N->getVTList(),
-        NewArgs, 19);
+        NewArgs);
   }
   }
-  return SDValue();
+
+  return AMDGPUTargetLowering::PerformDAGCombine(N, DCI);
 }
 
 static bool
@@ -1756,8 +2123,7 @@ FoldOperand(SDNode *ParentNode, unsigned SrcIdx, SDValue &Src, SDValue &Neg,
       TII->getOperandIdx(Opcode, AMDGPU::OpName::src1_W)
     };
     std::vector<unsigned> Consts;
-    for (unsigned i = 0; i < sizeof(SrcIndices) / sizeof(int); i++) {
-      int OtherSrcIdx = SrcIndices[i];
+    for (int OtherSrcIdx : SrcIndices) {
       int OtherSelIdx = TII->getSelIdx(Opcode, OtherSrcIdx);
       if (OtherSrcIdx < 0 || OtherSelIdx < 0)
         continue;
@@ -1768,14 +2134,14 @@ FoldOperand(SDNode *ParentNode, unsigned SrcIdx, SDValue &Src, SDValue &Neg,
       if (RegisterSDNode *Reg =
           dyn_cast<RegisterSDNode>(ParentNode->getOperand(OtherSrcIdx))) {
         if (Reg->getReg() == AMDGPU::ALU_CONST) {
-          ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(
-              ParentNode->getOperand(OtherSelIdx));
+          ConstantSDNode *Cst
+            = cast<ConstantSDNode>(ParentNode->getOperand(OtherSelIdx));
           Consts.push_back(Cst->getZExtValue());
         }
       }
     }
 
-    ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(CstOffset);
+    ConstantSDNode *Cst = cast<ConstantSDNode>(CstOffset);
     Consts.push_back(Cst->getZExtValue());
     if (!TII->fitsConstReadLimitations(Consts)) {
       return false;
@@ -1847,9 +2213,8 @@ SDNode *R600TargetLowering::PostISelFolding(MachineSDNode *Node,
   SDValue FakeOp;
 
   std::vector<SDValue> Ops;
-  for(SDNode::op_iterator I = Node->op_begin(), E = Node->op_end();
-              I != E; ++I)
-          Ops.push_back(*I);
+  for (const SDUse &I : Node->ops())
+    Ops.push_back(I);
 
   if (Opcode == AMDGPU::DOT_4) {
     int OperandIdx[] = {
diff --git a/contrib/llvm/lib/Target/R600/R600ISelLowering.h b/contrib/llvm/lib/Target/R600/R600ISelLowering.h
index c10257e..d22c8c9 100644
--- a/contrib/llvm/lib/Target/R600/R600ISelLowering.h
+++ b/contrib/llvm/lib/Target/R600/R600ISelLowering.h
@@ -24,26 +24,26 @@ class R600InstrInfo;
 class R600TargetLowering : public AMDGPUTargetLowering {
 public:
   R600TargetLowering(TargetMachine &TM);
-  virtual MachineBasicBlock * EmitInstrWithCustomInserter(MachineInstr *MI,
-      MachineBasicBlock * BB) const;
-  virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
-  virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
+  MachineBasicBlock * EmitInstrWithCustomInserter(MachineInstr *MI,
+      MachineBasicBlock * BB) const override;
+  SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
+  SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
   void ReplaceNodeResults(SDNode * N,
-      SmallVectorImpl<SDValue> &Results,
-      SelectionDAG &DAG) const;
-  virtual SDValue LowerFormalArguments(
-                                      SDValue Chain,
-                                      CallingConv::ID CallConv,
-                                      bool isVarArg,
-                                      const SmallVectorImpl<ISD::InputArg> &Ins,
-                                      SDLoc DL, SelectionDAG &DAG,
-                                      SmallVectorImpl<SDValue> &InVals) const;
-  virtual EVT getSetCCResultType(LLVMContext &, EVT VT) const;
+                          SmallVectorImpl<SDValue> &Results,
+                          SelectionDAG &DAG) const override;
+  SDValue LowerFormalArguments(
+                              SDValue Chain,
+                              CallingConv::ID CallConv,
+                              bool isVarArg,
+                              const SmallVectorImpl<ISD::InputArg> &Ins,
+                              SDLoc DL, SelectionDAG &DAG,
+                              SmallVectorImpl<SDValue> &InVals) const override;
+  EVT getSetCCResultType(LLVMContext &, EVT VT) const override;
 private:
   unsigned Gen;
   /// Each OpenCL kernel has nine implicit parameters that are stored in the
   /// first nine dwords of a Vertex Buffer.  These implicit parameters are
-  /// lowered to load instructions which retreive the values from the Vertex
+  /// lowered to load instructions which retrieve the values from the Vertex
   /// Buffer.
   SDValue LowerImplicitParameter(SelectionDAG &DAG, EVT VT,
                                  SDLoc DL, unsigned DwordOffset) const;
@@ -51,22 +51,25 @@ private:
   void lowerImplicitParameter(MachineInstr *MI, MachineBasicBlock &BB,
       MachineRegisterInfo & MRI, unsigned dword_offset) const;
   SDValue OptimizeSwizzle(SDValue BuildVector, SDValue Swz[], SelectionDAG &DAG) const;
+  SDValue vectorToVerticalVector(SelectionDAG &DAG, SDValue Vector) const;
 
-  /// \brief Lower ROTL opcode to BITALIGN
-  SDValue LowerROTL(SDValue Op, SelectionDAG &DAG) const;
-
+  SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFPTOUINT(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerTrig(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSHLParts(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSRXParts(SDValue Op, SelectionDAG &DAG) const;
 
   SDValue stackPtrToRegIndex(SDValue Ptr, unsigned StackWidth,
                                           SelectionDAG &DAG) const;
   void getStackAddress(unsigned StackWidth, unsigned ElemIdx,
                        unsigned &Channel, unsigned &PtrIncr) const;
   bool isZero(SDValue Op) const;
-  virtual SDNode *PostISelFolding(MachineSDNode *N, SelectionDAG &DAG) const;
+  SDNode *PostISelFolding(MachineSDNode *N, SelectionDAG &DAG) const override;
 };
 
 } // End namespace llvm;
diff --git a/contrib/llvm/lib/Target/R600/R600InstrInfo.cpp b/contrib/llvm/lib/Target/R600/R600InstrInfo.cpp
index 2eca6cf..99920b7 100644
--- a/contrib/llvm/lib/Target/R600/R600InstrInfo.cpp
+++ b/contrib/llvm/lib/Target/R600/R600InstrInfo.cpp
@@ -23,15 +23,14 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 
+using namespace llvm;
+
 #define GET_INSTRINFO_CTOR_DTOR
 #include "AMDGPUGenDFAPacketizer.inc"
 
-using namespace llvm;
-
-R600InstrInfo::R600InstrInfo(AMDGPUTargetMachine &tm)
-  : AMDGPUInstrInfo(tm),
-    RI(tm),
-    ST(tm.getSubtarget<AMDGPUSubtarget>())
+R600InstrInfo::R600InstrInfo(const AMDGPUSubtarget &st)
+  : AMDGPUInstrInfo(st),
+    RI(st)
   { }
 
 const R600RegisterInfo &R600InstrInfo::getRegisterInfo() const {
@@ -52,11 +51,15 @@ R600InstrInfo::copyPhysReg(MachineBasicBlock &MBB,
                            unsigned DestReg, unsigned SrcReg,
                            bool KillSrc) const {
   unsigned VectorComponents = 0;
-  if (AMDGPU::R600_Reg128RegClass.contains(DestReg) &&
-      AMDGPU::R600_Reg128RegClass.contains(SrcReg)) {
+  if ((AMDGPU::R600_Reg128RegClass.contains(DestReg) ||
+      AMDGPU::R600_Reg128VerticalRegClass.contains(DestReg)) &&
+      (AMDGPU::R600_Reg128RegClass.contains(SrcReg) ||
+       AMDGPU::R600_Reg128VerticalRegClass.contains(SrcReg))) {
     VectorComponents = 4;
-  } else if(AMDGPU::R600_Reg64RegClass.contains(DestReg) &&
-            AMDGPU::R600_Reg64RegClass.contains(SrcReg)) {
+  } else if((AMDGPU::R600_Reg64RegClass.contains(DestReg) ||
+            AMDGPU::R600_Reg64VerticalRegClass.contains(DestReg)) &&
+            (AMDGPU::R600_Reg64RegClass.contains(SrcReg) ||
+             AMDGPU::R600_Reg64VerticalRegClass.contains(SrcReg))) {
     VectorComponents = 2;
   }
 
@@ -89,10 +92,6 @@ bool R600InstrInfo::isLegalToSplitMBBAt(MachineBasicBlock &MBB,
   return true;
 }
 
-unsigned R600InstrInfo::getIEQOpcode() const {
-  return AMDGPU::SETE_INT;
-}
-
 bool R600InstrInfo::isMov(unsigned Opcode) const {
 
 
@@ -206,8 +205,10 @@ bool R600InstrInfo::usesVertexCache(unsigned Opcode) const {
 }
 
 bool R600InstrInfo::usesVertexCache(const MachineInstr *MI) const {
-  const R600MachineFunctionInfo *MFI = MI->getParent()->getParent()->getInfo<R600MachineFunctionInfo>();
-  return MFI->ShaderType != ShaderType::COMPUTE && usesVertexCache(MI->getOpcode());
+  const MachineFunction *MF = MI->getParent()->getParent();
+  const R600MachineFunctionInfo *MFI = MF->getInfo<R600MachineFunctionInfo>();
+  return MFI->getShaderType() != ShaderType::COMPUTE &&
+    usesVertexCache(MI->getOpcode());
 }
 
 bool R600InstrInfo::usesTextureCache(unsigned Opcode) const {
@@ -215,9 +216,11 @@ bool R600InstrInfo::usesTextureCache(unsigned Opcode) const {
 }
 
 bool R600InstrInfo::usesTextureCache(const MachineInstr *MI) const {
-  const R600MachineFunctionInfo *MFI = MI->getParent()->getParent()->getInfo<R600MachineFunctionInfo>();
-  return (MFI->ShaderType == ShaderType::COMPUTE && usesVertexCache(MI->getOpcode())) ||
-         usesTextureCache(MI->getOpcode());
+  const MachineFunction *MF = MI->getParent()->getParent();
+  const R600MachineFunctionInfo *MFI = MF->getInfo<R600MachineFunctionInfo>();
+  return (MFI->getShaderType() == ShaderType::COMPUTE &&
+          usesVertexCache(MI->getOpcode())) ||
+    usesTextureCache(MI->getOpcode());
 }
 
 bool R600InstrInfo::mustBeLastInClause(unsigned Opcode) const {
@@ -316,7 +319,7 @@ R600InstrInfo::getSrcs(MachineInstr *MI) const {
         Result.push_back(std::pair<MachineOperand *, int64_t>(&MO, Sel));
         continue;
       }
-      
+
     }
     return Result;
   }
@@ -677,7 +680,7 @@ findFirstPredicateSetterFrom(MachineBasicBlock &MBB,
       return MI;
   }
 
-  return NULL;
+  return nullptr;
 }
 
 static
@@ -717,8 +720,8 @@ R600InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
   }
 
   // Remove successive JUMP
-  while (I != MBB.begin() && llvm::prior(I)->getOpcode() == AMDGPU::JUMP) {
-      MachineBasicBlock::iterator PriorI = llvm::prior(I);
+  while (I != MBB.begin() && std::prev(I)->getOpcode() == AMDGPU::JUMP) {
+      MachineBasicBlock::iterator PriorI = std::prev(I);
       if (AllowModify)
         I->removeFromParent();
       I = PriorI;
@@ -768,23 +771,13 @@ R600InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
   return true;
 }
 
-int R600InstrInfo::getBranchInstr(const MachineOperand &op) const {
-  const MachineInstr *MI = op.getParent();
-
-  switch (MI->getDesc().OpInfo->RegClass) {
-  default: // FIXME: fallthrough??
-  case AMDGPU::GPRI32RegClassID: return AMDGPU::BRANCH_COND_i32;
-  case AMDGPU::GPRF32RegClassID: return AMDGPU::BRANCH_COND_f32;
-  };
-}
-
 static
 MachineBasicBlock::iterator FindLastAluClause(MachineBasicBlock &MBB) {
   for (MachineBasicBlock::reverse_iterator It = MBB.rbegin(), E = MBB.rend();
       It != E; ++It) {
     if (It->getOpcode() == AMDGPU::CF_ALU ||
         It->getOpcode() == AMDGPU::CF_ALU_PUSH_BEFORE)
-      return llvm::prior(It.base());
+      return std::prev(It.base());
   }
   return MBB.end();
 }
@@ -797,7 +790,7 @@ R600InstrInfo::InsertBranch(MachineBasicBlock &MBB,
                             DebugLoc DL) const {
   assert(TBB && "InsertBranch must not be told to insert a fallthrough");
 
-  if (FBB == 0) {
+  if (!FBB) {
     if (Cond.empty()) {
       BuildMI(&MBB, DL, get(AMDGPU::JUMP)).addMBB(TBB);
       return 1;
@@ -1064,10 +1057,34 @@ unsigned int R600InstrInfo::getInstrLatency(const InstrItineraryData *ItinData,
   return 2;
 }
 
+bool R600InstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
+
+  switch(MI->getOpcode()) {
+  default: return AMDGPUInstrInfo::expandPostRAPseudo(MI);
+  case AMDGPU::R600_EXTRACT_ELT_V2:
+  case AMDGPU::R600_EXTRACT_ELT_V4:
+    buildIndirectRead(MI->getParent(), MI, MI->getOperand(0).getReg(),
+                      RI.getHWRegIndex(MI->getOperand(1).getReg()), //  Address
+                      MI->getOperand(2).getReg(),
+                      RI.getHWRegChan(MI->getOperand(1).getReg()));
+    break;
+  case AMDGPU::R600_INSERT_ELT_V2:
+  case AMDGPU::R600_INSERT_ELT_V4:
+    buildIndirectWrite(MI->getParent(), MI, MI->getOperand(2).getReg(), // Value
+                       RI.getHWRegIndex(MI->getOperand(1).getReg()),  // Address
+                       MI->getOperand(3).getReg(),                    // Offset
+                       RI.getHWRegChan(MI->getOperand(1).getReg()));  // Channel
+    break;
+  }
+  MI->eraseFromParent();
+  return true;
+}
+
 void  R600InstrInfo::reserveIndirectRegisters(BitVector &Reserved,
                                              const MachineFunction &MF) const {
   const AMDGPUFrameLowering *TFL =
-                 static_cast<const AMDGPUFrameLowering*>(TM.getFrameLowering());
+    static_cast<const AMDGPUFrameLowering*>(
+    MF.getTarget().getFrameLowering());
 
   unsigned StackWidth = TFL->getStackWidth(MF);
   int End = getIndirectIndexEnd(MF);
@@ -1100,7 +1117,22 @@ MachineInstrBuilder R600InstrInfo::buildIndirectWrite(MachineBasicBlock *MBB,
                                        MachineBasicBlock::iterator I,
                                        unsigned ValueReg, unsigned Address,
                                        unsigned OffsetReg) const {
-  unsigned AddrReg = AMDGPU::R600_AddrRegClass.getRegister(Address);
+  return buildIndirectWrite(MBB, I, ValueReg, Address, OffsetReg, 0);
+}
+
+MachineInstrBuilder R600InstrInfo::buildIndirectWrite(MachineBasicBlock *MBB,
+                                       MachineBasicBlock::iterator I,
+                                       unsigned ValueReg, unsigned Address,
+                                       unsigned OffsetReg,
+                                       unsigned AddrChan) const {
+  unsigned AddrReg;
+  switch (AddrChan) {
+    default: llvm_unreachable("Invalid Channel");
+    case 0: AddrReg = AMDGPU::R600_AddrRegClass.getRegister(Address); break;
+    case 1: AddrReg = AMDGPU::R600_Addr_YRegClass.getRegister(Address); break;
+    case 2: AddrReg = AMDGPU::R600_Addr_ZRegClass.getRegister(Address); break;
+    case 3: AddrReg = AMDGPU::R600_Addr_WRegClass.getRegister(Address); break;
+  }
   MachineInstr *MOVA = buildDefaultInstruction(*MBB, I, AMDGPU::MOVA_INT_eg,
                                                AMDGPU::AR_X, OffsetReg);
   setImmOperand(MOVA, AMDGPU::OpName::write, 0);
@@ -1117,7 +1149,22 @@ MachineInstrBuilder R600InstrInfo::buildIndirectRead(MachineBasicBlock *MBB,
                                        MachineBasicBlock::iterator I,
                                        unsigned ValueReg, unsigned Address,
                                        unsigned OffsetReg) const {
-  unsigned AddrReg = AMDGPU::R600_AddrRegClass.getRegister(Address);
+  return buildIndirectRead(MBB, I, ValueReg, Address, OffsetReg, 0);
+}
+
+MachineInstrBuilder R600InstrInfo::buildIndirectRead(MachineBasicBlock *MBB,
+                                       MachineBasicBlock::iterator I,
+                                       unsigned ValueReg, unsigned Address,
+                                       unsigned OffsetReg,
+                                       unsigned AddrChan) const {
+  unsigned AddrReg;
+  switch (AddrChan) {
+    default: llvm_unreachable("Invalid Channel");
+    case 0: AddrReg = AMDGPU::R600_AddrRegClass.getRegister(Address); break;
+    case 1: AddrReg = AMDGPU::R600_Addr_YRegClass.getRegister(Address); break;
+    case 2: AddrReg = AMDGPU::R600_Addr_ZRegClass.getRegister(Address); break;
+    case 3: AddrReg = AMDGPU::R600_Addr_WRegClass.getRegister(Address); break;
+  }
   MachineInstr *MOVA = buildDefaultInstruction(*MBB, I, AMDGPU::MOVA_INT_eg,
                                                        AMDGPU::AR_X,
                                                        OffsetReg);
@@ -1220,7 +1267,6 @@ MachineInstr *R600InstrInfo::buildSlotOfVectorInstruction(
     const {
   assert (MI->getOpcode() == AMDGPU::DOT_4 && "Not Implemented");
   unsigned Opcode;
-  const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
   if (ST.getGeneration() <= AMDGPUSubtarget::R700)
     Opcode = AMDGPU::DOT4_r600;
   else
diff --git a/contrib/llvm/lib/Target/R600/R600InstrInfo.h b/contrib/llvm/lib/Target/R600/R600InstrInfo.h
index 13d9810..1c3cb63 100644
--- a/contrib/llvm/lib/Target/R600/R600InstrInfo.h
+++ b/contrib/llvm/lib/Target/R600/R600InstrInfo.h
@@ -32,12 +32,22 @@ namespace llvm {
   class R600InstrInfo : public AMDGPUInstrInfo {
   private:
   const R600RegisterInfo RI;
-  const AMDGPUSubtarget &ST;
 
-  int getBranchInstr(const MachineOperand &op) const;
   std::vector<std::pair<int, unsigned> >
   ExtractSrcs(MachineInstr *MI, const DenseMap<unsigned, unsigned> &PV, unsigned &ConstCount) const;
 
+
+  MachineInstrBuilder buildIndirectRead(MachineBasicBlock *MBB,
+                                        MachineBasicBlock::iterator I,
+                                        unsigned ValueReg, unsigned Address,
+                                        unsigned OffsetReg,
+                                        unsigned AddrChan) const;
+
+  MachineInstrBuilder buildIndirectWrite(MachineBasicBlock *MBB,
+                                        MachineBasicBlock::iterator I,
+                                        unsigned ValueReg, unsigned Address,
+                                        unsigned OffsetReg,
+                                        unsigned AddrChan) const;
   public:
   enum BankSwizzle {
     ALU_VEC_012_SCL_210 = 0,
@@ -48,15 +58,15 @@ namespace llvm {
     ALU_VEC_210
   };
 
-  explicit R600InstrInfo(AMDGPUTargetMachine &tm);
+  explicit R600InstrInfo(const AMDGPUSubtarget &st);
 
-  const R600RegisterInfo &getRegisterInfo() const;
-  virtual void copyPhysReg(MachineBasicBlock &MBB,
-                           MachineBasicBlock::iterator MI, DebugLoc DL,
-                           unsigned DestReg, unsigned SrcReg,
-                           bool KillSrc) const;
+  const R600RegisterInfo &getRegisterInfo() const override;
+  void copyPhysReg(MachineBasicBlock &MBB,
+                   MachineBasicBlock::iterator MI, DebugLoc DL,
+                   unsigned DestReg, unsigned SrcReg,
+                   bool KillSrc) const override;
   bool isLegalToSplitMBBAt(MachineBasicBlock &MBB,
-                           MachineBasicBlock::iterator MBBI) const;
+                           MachineBasicBlock::iterator MBBI) const override;
 
   bool isTrig(const MachineInstr &MI) const;
   bool isPlaceHolderOpcode(unsigned opcode) const;
@@ -138,83 +148,84 @@ namespace llvm {
   /// Same but using const index set instead of MI set.
   bool fitsConstReadLimitations(const std::vector<unsigned>&) const;
 
-  /// \breif Vector instructions are instructions that must fill all
+  /// \brief Vector instructions are instructions that must fill all
   /// instruction slots within an instruction group.
   bool isVector(const MachineInstr &MI) const;
 
-  virtual unsigned getIEQOpcode() const;
-  virtual bool isMov(unsigned Opcode) const;
+  bool isMov(unsigned Opcode) const override;
 
   DFAPacketizer *CreateTargetScheduleState(const TargetMachine *TM,
-                                           const ScheduleDAG *DAG) const;
+                                           const ScheduleDAG *DAG) const override;
 
-  bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const;
+  bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const override;
 
   bool AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB, MachineBasicBlock *&FBB,
-                     SmallVectorImpl<MachineOperand> &Cond, bool AllowModify) const;
+                     SmallVectorImpl<MachineOperand> &Cond, bool AllowModify) const override;
 
-  unsigned InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB, MachineBasicBlock *FBB, const SmallVectorImpl<MachineOperand> &Cond, DebugLoc DL) const;
+  unsigned InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB, MachineBasicBlock *FBB, const SmallVectorImpl<MachineOperand> &Cond, DebugLoc DL) const override;
 
-  unsigned RemoveBranch(MachineBasicBlock &MBB) const;
+  unsigned RemoveBranch(MachineBasicBlock &MBB) const override;
 
-  bool isPredicated(const MachineInstr *MI) const;
+  bool isPredicated(const MachineInstr *MI) const override;
 
-  bool isPredicable(MachineInstr *MI) const;
+  bool isPredicable(MachineInstr *MI) const override;
 
   bool
    isProfitableToDupForIfCvt(MachineBasicBlock &MBB, unsigned NumCyles,
-                             const BranchProbability &Probability) const;
+                             const BranchProbability &Probability) const override;
 
   bool isProfitableToIfCvt(MachineBasicBlock &MBB, unsigned NumCyles,
                            unsigned ExtraPredCycles,
-                           const BranchProbability &Probability) const ;
+                           const BranchProbability &Probability) const override ;
 
   bool
    isProfitableToIfCvt(MachineBasicBlock &TMBB,
                        unsigned NumTCycles, unsigned ExtraTCycles,
                        MachineBasicBlock &FMBB,
                        unsigned NumFCycles, unsigned ExtraFCycles,
-                       const BranchProbability &Probability) const;
+                       const BranchProbability &Probability) const override;
 
   bool DefinesPredicate(MachineInstr *MI,
-                                  std::vector<MachineOperand> &Pred) const;
+                                  std::vector<MachineOperand> &Pred) const override;
 
   bool SubsumesPredicate(const SmallVectorImpl<MachineOperand> &Pred1,
-                         const SmallVectorImpl<MachineOperand> &Pred2) const;
+                         const SmallVectorImpl<MachineOperand> &Pred2) const override;
 
   bool isProfitableToUnpredicate(MachineBasicBlock &TMBB,
-                                          MachineBasicBlock &FMBB) const;
+                                          MachineBasicBlock &FMBB) const override;
 
   bool PredicateInstruction(MachineInstr *MI,
-                        const SmallVectorImpl<MachineOperand> &Pred) const;
+                        const SmallVectorImpl<MachineOperand> &Pred) const override;
 
-  unsigned int getPredicationCost(const MachineInstr *) const;
+  unsigned int getPredicationCost(const MachineInstr *) const override;
 
   unsigned int getInstrLatency(const InstrItineraryData *ItinData,
                                const MachineInstr *MI,
-                               unsigned *PredCost = 0) const;
+                               unsigned *PredCost = nullptr) const override;
+
+  int getInstrLatency(const InstrItineraryData *ItinData,
+                      SDNode *Node) const override { return 1;}
 
-  virtual int getInstrLatency(const InstrItineraryData *ItinData,
-                              SDNode *Node) const { return 1;}
+  virtual bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const;
 
   /// \brief Reserve the registers that may be accesed using indirect addressing.
   void reserveIndirectRegisters(BitVector &Reserved,
                                 const MachineFunction &MF) const;
 
-  virtual unsigned calculateIndirectAddress(unsigned RegIndex,
-                                            unsigned Channel) const;
+  unsigned calculateIndirectAddress(unsigned RegIndex,
+                                    unsigned Channel) const override;
 
-  virtual const TargetRegisterClass *getIndirectAddrRegClass() const;
+  const TargetRegisterClass *getIndirectAddrRegClass() const override;
 
-  virtual MachineInstrBuilder buildIndirectWrite(MachineBasicBlock *MBB,
-                                  MachineBasicBlock::iterator I,
-                                  unsigned ValueReg, unsigned Address,
-                                  unsigned OffsetReg) const;
+  MachineInstrBuilder buildIndirectWrite(MachineBasicBlock *MBB,
+                          MachineBasicBlock::iterator I,
+                          unsigned ValueReg, unsigned Address,
+                          unsigned OffsetReg) const override;
 
-  virtual MachineInstrBuilder buildIndirectRead(MachineBasicBlock *MBB,
-                                  MachineBasicBlock::iterator I,
-                                  unsigned ValueReg, unsigned Address,
-                                  unsigned OffsetReg) const;
+  MachineInstrBuilder buildIndirectRead(MachineBasicBlock *MBB,
+                                        MachineBasicBlock::iterator I,
+                                        unsigned ValueReg, unsigned Address,
+                                        unsigned OffsetReg) const override;
 
   unsigned getMaxAlusPerClause() const;
 
@@ -244,7 +255,7 @@ namespace llvm {
 
   MachineInstr *buildMovInstr(MachineBasicBlock *MBB,
                               MachineBasicBlock::iterator I,
-                              unsigned DstReg, unsigned SrcReg) const;
+                              unsigned DstReg, unsigned SrcReg) const override;
 
   /// \brief Get the index of Op in the MachineInstr.
   ///
diff --git a/contrib/llvm/lib/Target/R600/R600Instructions.td b/contrib/llvm/lib/Target/R600/R600Instructions.td
index 74c65da..704507d 100644
--- a/contrib/llvm/lib/Target/R600/R600Instructions.td
+++ b/contrib/llvm/lib/Target/R600/R600Instructions.td
@@ -7,7 +7,8 @@
 //
 //===----------------------------------------------------------------------===//
 //
-// R600 Tablegen instruction definitions
+// TableGen definitions for instructions which are available on R600 family
+// GPUs.
 //
 //===----------------------------------------------------------------------===//
 
@@ -124,7 +125,7 @@ class R600_1OP <bits<11> inst, string opName, list<dag> pattern,
 class R600_1OP_Helper <bits<11> inst, string opName, SDPatternOperator node,
                     InstrItinClass itin = AnyALU> :
     R600_1OP <inst, opName,
-              [(set R600_Reg32:$dst, (node R600_Reg32:$src0))]
+              [(set R600_Reg32:$dst, (node R600_Reg32:$src0))], itin
 >;
 
 // If you add or change the operands for R600_2OP instructions, you must
@@ -160,10 +161,10 @@ class R600_2OP <bits<11> inst, string opName, list<dag> pattern,
 }
 
 class R600_2OP_Helper <bits<11> inst, string opName, SDPatternOperator node,
-                       InstrItinClass itim = AnyALU> :
+                       InstrItinClass itin = AnyALU> :
     R600_2OP <inst, opName,
               [(set R600_Reg32:$dst, (node R600_Reg32:$src0,
-                                           R600_Reg32:$src1))]
+                                           R600_Reg32:$src1))], itin
 >;
 
 // If you add our change the operands for R600_3OP instructions, you must
@@ -215,7 +216,7 @@ class R600_REDUCTION <bits<11> inst, dag ins, string asm, list<dag> pattern,
 def TEX_SHADOW : PatLeaf<
   (imm),
   [{uint32_t TType = (uint32_t)N->getZExtValue();
-    return (TType >= 6 && TType <= 8) || (TType >= 11 && TType <= 13);
+    return (TType >= 6 && TType <= 8) || TType == 13;
   }]
 >;
 
@@ -335,17 +336,6 @@ def load_param_exti8 : LoadParamFrag<az_extloadi8>;
 def load_param_exti16 : LoadParamFrag<az_extloadi16>;
 
 def isR600 : Predicate<"Subtarget.getGeneration() <= AMDGPUSubtarget::R700">;
-def isR700 : Predicate<"Subtarget.getGeneration() == AMDGPUSubtarget::R700">;
-def isEG : Predicate<
-  "Subtarget.getGeneration() >= AMDGPUSubtarget::EVERGREEN && "
-  "Subtarget.getGeneration() < AMDGPUSubtarget::SOUTHERN_ISLANDS && "
-  "!Subtarget.hasCaymanISA()">;
-
-def isCayman : Predicate<"Subtarget.hasCaymanISA()">;
-def isEGorCayman : Predicate<"Subtarget.getGeneration() == "
-                             "AMDGPUSubtarget::EVERGREEN"
-                            "|| Subtarget.getGeneration() =="
-                            "AMDGPUSubtarget::NORTHERN_ISLANDS">;
 
 def isR600toCayman : Predicate<
                      "Subtarget.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS">;
@@ -642,6 +632,9 @@ ins, AsmPrint, [] >, CF_WORD0_EG, CF_WORD1_EG {
 def CF_ALU : ALU_CLAUSE<8, "ALU">;
 def CF_ALU_PUSH_BEFORE : ALU_CLAUSE<9, "ALU_PUSH_BEFORE">;
 def CF_ALU_POP_AFTER : ALU_CLAUSE<10, "ALU_POP_AFTER">;
+def CF_ALU_CONTINUE : ALU_CLAUSE<13, "ALU_CONTINUE">;
+def CF_ALU_BREAK : ALU_CLAUSE<14, "ALU_BREAK">;
+def CF_ALU_ELSE_AFTER : ALU_CLAUSE<15, "ALU_ELSE_AFTER">;
 
 def FETCH_CLAUSE : AMDGPUInst <(outs),
 (ins i32imm:$addr), "Fetch clause starting at $addr:", [] > {
@@ -728,7 +721,7 @@ def SETNE_DX10 : R600_2OP <
 >;
 
 def FRACT : R600_1OP_Helper <0x10, "FRACT", AMDGPUfract>;
-def TRUNC : R600_1OP_Helper <0x11, "TRUNC", int_AMDGPU_trunc>;
+def TRUNC : R600_1OP_Helper <0x11, "TRUNC", ftrunc>;
 def CEIL : R600_1OP_Helper <0x12, "CEIL", fceil>;
 def RNDNE : R600_1OP_Helper <0x13, "RNDNE", frint>;
 def FLOOR : R600_1OP_Helper <0x14, "FLOOR", ffloor>;
@@ -1086,18 +1079,21 @@ class RECIP_UINT_Common <bits<11> inst> : R600_1OP_Helper <
   let Itinerary = TransALU;
 }
 
+// Clamped to maximum.
 class RECIPSQRT_CLAMPED_Common <bits<11> inst> : R600_1OP_Helper <
-  inst, "RECIPSQRT_CLAMPED", int_AMDGPU_rsq
+  inst, "RECIPSQRT_CLAMPED", AMDGPUrsq_clamped
 > {
   let Itinerary = TransALU;
 }
 
-class RECIPSQRT_IEEE_Common <bits<11> inst> : R600_1OP <
-  inst, "RECIPSQRT_IEEE", []
+class RECIPSQRT_IEEE_Common <bits<11> inst> : R600_1OP_Helper <
+  inst, "RECIPSQRT_IEEE", AMDGPUrsq_legacy
 > {
   let Itinerary = TransALU;
 }
 
+// TODO: There is also RECIPSQRT_FF which clamps to zero.
+
 class SIN_Common <bits<11> inst> : R600_1OP <
   inst, "SIN", [(set f32:$dst, (SIN_HW f32:$src0))]>{
   let Trig = 1;
@@ -1235,6 +1231,10 @@ let Predicates = [isR600] in {
   "JUMP @$ADDR POP:$POP_COUNT"> {
     let CNT = 0;
   }
+  def CF_PUSH_ELSE_R600 : CF_CLAUSE_R600<12, (ins i32imm:$ADDR),
+  "PUSH_ELSE @$ADDR"> {
+    let CNT = 0;
+  }
   def CF_ELSE_R600 : CF_CLAUSE_R600<13, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
   "ELSE @$ADDR POP:$POP_COUNT"> {
     let CNT = 0;
@@ -1257,561 +1257,6 @@ let Predicates = [isR600] in {
 
 }
 
-//===----------------------------------------------------------------------===//
-// R700 Only instructions
-//===----------------------------------------------------------------------===//
-
-let Predicates = [isR700] in {
-  def SIN_r700 : SIN_Common<0x6E>;
-  def COS_r700 : COS_Common<0x6F>;
-}
-
-//===----------------------------------------------------------------------===//
-// Evergreen / Cayman store instructions
-//===----------------------------------------------------------------------===//
-
-let Predicates = [isEGorCayman] in {
-
-class CF_MEM_RAT_CACHELESS <bits<6> rat_inst, bits<4> rat_id, bits<4> mask, dag ins,
-                           string name, list<dag> pattern>
-    : EG_CF_RAT <0x57, rat_inst, rat_id, mask, (outs), ins,
-                 "MEM_RAT_CACHELESS "#name, pattern>;
-
-class CF_MEM_RAT <bits<6> rat_inst, bits<4> rat_id, dag ins, string name,
-                  list<dag> pattern>
-    : EG_CF_RAT <0x56, rat_inst, rat_id, 0xf /* mask */, (outs), ins,
-                 "MEM_RAT "#name, pattern>;
-
-def RAT_MSKOR : CF_MEM_RAT <0x11, 0,
-  (ins R600_Reg128:$rw_gpr, R600_TReg32_X:$index_gpr),
-  "MSKOR $rw_gpr.XW, $index_gpr",
-  [(mskor_global v4i32:$rw_gpr, i32:$index_gpr)]
-> {
-  let eop = 0;
-}
-
-} // End Predicates = [isEGorCayman]
-
-
-//===----------------------------------------------------------------------===//
-// Evergreen Only instructions
-//===----------------------------------------------------------------------===//
-
-let Predicates = [isEG] in {
-
-def RECIP_IEEE_eg : RECIP_IEEE_Common<0x86>;
-defm DIV_eg : DIV_Common<RECIP_IEEE_eg>;
-
-def MULLO_INT_eg : MULLO_INT_Common<0x8F>;
-def MULHI_INT_eg : MULHI_INT_Common<0x90>;
-def MULLO_UINT_eg : MULLO_UINT_Common<0x91>;
-def MULHI_UINT_eg : MULHI_UINT_Common<0x92>;
-def RECIP_UINT_eg : RECIP_UINT_Common<0x94>;
-def RECIPSQRT_CLAMPED_eg : RECIPSQRT_CLAMPED_Common<0x87>;
-def EXP_IEEE_eg : EXP_IEEE_Common<0x81>;
-def LOG_IEEE_eg : LOG_IEEE_Common<0x83>;
-def RECIP_CLAMPED_eg : RECIP_CLAMPED_Common<0x84>;
-def RECIPSQRT_IEEE_eg : RECIPSQRT_IEEE_Common<0x89>;
-def SIN_eg : SIN_Common<0x8D>;
-def COS_eg : COS_Common<0x8E>;
-
-def : POW_Common <LOG_IEEE_eg, EXP_IEEE_eg, MUL>;
-def : Pat<(fsqrt f32:$src), (MUL $src, (RECIPSQRT_CLAMPED_eg $src))>;
-
-//===----------------------------------------------------------------------===//
-// Memory read/write instructions
-//===----------------------------------------------------------------------===//
-
-let usesCustomInserter = 1 in {
-
-// 32-bit store
-def RAT_WRITE_CACHELESS_32_eg : CF_MEM_RAT_CACHELESS <0x2, 0, 0x1,
-  (ins R600_TReg32_X:$rw_gpr, R600_TReg32_X:$index_gpr, InstFlag:$eop),
-  "STORE_RAW $rw_gpr, $index_gpr, $eop",
-  [(global_store i32:$rw_gpr, i32:$index_gpr)]
->;
-
-// 64-bit store
-def RAT_WRITE_CACHELESS_64_eg : CF_MEM_RAT_CACHELESS <0x2, 0, 0x3,
-  (ins R600_Reg64:$rw_gpr, R600_TReg32_X:$index_gpr, InstFlag:$eop),
-  "STORE_RAW $rw_gpr.XY, $index_gpr, $eop",
-  [(global_store v2i32:$rw_gpr, i32:$index_gpr)]
->;
-
-//128-bit store
-def RAT_WRITE_CACHELESS_128_eg : CF_MEM_RAT_CACHELESS <0x2, 0, 0xf,
-  (ins R600_Reg128:$rw_gpr, R600_TReg32_X:$index_gpr, InstFlag:$eop),
-  "STORE_RAW $rw_gpr.XYZW, $index_gpr, $eop",
-  [(global_store v4i32:$rw_gpr, i32:$index_gpr)]
->;
-
-} // End usesCustomInserter = 1
-
-class VTX_READ_eg <string name, bits<8> buffer_id, dag outs, list<dag> pattern>
-    : VTX_WORD0_eg, VTX_READ<name, buffer_id, outs, pattern> {
-
-  // Static fields
-  let VC_INST = 0;
-  let FETCH_TYPE = 2;
-  let FETCH_WHOLE_QUAD = 0;
-  let BUFFER_ID = buffer_id;
-  let SRC_REL = 0;
-  // XXX: We can infer this field based on the SRC_GPR.  This would allow us
-  // to store vertex addresses in any channel, not just X.
-  let SRC_SEL_X = 0;
-
-  let Inst{31-0} = Word0;
-}
-
-class VTX_READ_8_eg <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_eg <"VTX_READ_8 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
-
-  let MEGA_FETCH_COUNT = 1;
-  let DST_SEL_X = 0;
-  let DST_SEL_Y = 7;   // Masked
-  let DST_SEL_Z = 7;   // Masked
-  let DST_SEL_W = 7;   // Masked
-  let DATA_FORMAT = 1; // FMT_8
-}
-
-class VTX_READ_16_eg <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_eg <"VTX_READ_16 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
-  let MEGA_FETCH_COUNT = 2;
-  let DST_SEL_X = 0;
-  let DST_SEL_Y = 7;   // Masked
-  let DST_SEL_Z = 7;   // Masked
-  let DST_SEL_W = 7;   // Masked
-  let DATA_FORMAT = 5; // FMT_16
-
-}
-
-class VTX_READ_32_eg <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_eg <"VTX_READ_32 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
-
-  let MEGA_FETCH_COUNT = 4;
-  let DST_SEL_X        = 0;
-  let DST_SEL_Y        = 7;   // Masked
-  let DST_SEL_Z        = 7;   // Masked
-  let DST_SEL_W        = 7;   // Masked
-  let DATA_FORMAT      = 0xD; // COLOR_32
-
-  // This is not really necessary, but there were some GPU hangs that appeared
-  // to be caused by ALU instructions in the next instruction group that wrote
-  // to the $src_gpr registers of the VTX_READ.
-  // e.g.
-  // %T3_X<def> = VTX_READ_PARAM_32_eg %T2_X<kill>, 24
-  // %T2_X<def> = MOV %ZERO
-  //Adding this constraint prevents this from happening.
-  let Constraints = "$src_gpr.ptr = $dst_gpr";
-}
-
-class VTX_READ_64_eg <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_eg <"VTX_READ_64 $dst_gpr.XY, $src_gpr", buffer_id,
-                   (outs R600_Reg64:$dst_gpr), pattern> {
-
-  let MEGA_FETCH_COUNT = 8;
-  let DST_SEL_X        = 0;
-  let DST_SEL_Y        = 1;
-  let DST_SEL_Z        = 7;
-  let DST_SEL_W        = 7;
-  let DATA_FORMAT      = 0x1D; // COLOR_32_32
-}
-
-class VTX_READ_128_eg <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_eg <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr", buffer_id,
-                   (outs R600_Reg128:$dst_gpr), pattern> {
-
-  let MEGA_FETCH_COUNT = 16;
-  let DST_SEL_X        =  0;
-  let DST_SEL_Y        =  1;
-  let DST_SEL_Z        =  2;
-  let DST_SEL_W        =  3;
-  let DATA_FORMAT      =  0x22; // COLOR_32_32_32_32
-
-  // XXX: Need to force VTX_READ_128 instructions to write to the same register
-  // that holds its buffer address to avoid potential hangs.  We can't use
-  // the same constraint as VTX_READ_32_eg, because the $src_gpr.ptr and $dst
-  // registers are different sizes.
-}
-
-//===----------------------------------------------------------------------===//
-// VTX Read from parameter memory space
-//===----------------------------------------------------------------------===//
-
-def VTX_READ_PARAM_8_eg : VTX_READ_8_eg <0,
-  [(set i32:$dst_gpr, (load_param_exti8 ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_16_eg : VTX_READ_16_eg <0,
-  [(set i32:$dst_gpr, (load_param_exti16 ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_32_eg : VTX_READ_32_eg <0,
-  [(set i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_64_eg : VTX_READ_64_eg <0,
-  [(set v2i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_128_eg : VTX_READ_128_eg <0,
-  [(set v4i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
-
-//===----------------------------------------------------------------------===//
-// VTX Read from global memory space
-//===----------------------------------------------------------------------===//
-
-// 8-bit reads
-def VTX_READ_GLOBAL_8_eg : VTX_READ_8_eg <1,
-  [(set i32:$dst_gpr, (az_extloadi8_global ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_GLOBAL_16_eg : VTX_READ_16_eg <1,
-  [(set i32:$dst_gpr, (az_extloadi16_global ADDRVTX_READ:$src_gpr))]
->;
-
-// 32-bit reads
-def VTX_READ_GLOBAL_32_eg : VTX_READ_32_eg <1,
-  [(set i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 64-bit reads
-def VTX_READ_GLOBAL_64_eg : VTX_READ_64_eg <1,
-  [(set v2i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 128-bit reads
-def VTX_READ_GLOBAL_128_eg : VTX_READ_128_eg <1,
-  [(set v4i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
->;
-
-} // End Predicates = [isEG]
-
-//===----------------------------------------------------------------------===//
-// Evergreen / Cayman Instructions
-//===----------------------------------------------------------------------===//
-
-let Predicates = [isEGorCayman] in {
-
-  // BFE_UINT - bit_extract, an optimization for mask and shift
-  // Src0 = Input
-  // Src1 = Offset
-  // Src2 = Width
-  //
-  // bit_extract = (Input << (32 - Offset - Width)) >> (32 - Width)
-  //
-  // Example Usage:
-  // (Offset, Width)
-  //
-  // (0, 8)           = (Input << 24) >> 24  = (Input &  0xff)       >> 0
-  // (8, 8)           = (Input << 16) >> 24  = (Input &  0xffff)     >> 8
-  // (16,8)           = (Input <<  8) >> 24  = (Input &  0xffffff)   >> 16
-  // (24,8)           = (Input <<  0) >> 24  = (Input &  0xffffffff) >> 24
-  def BFE_UINT_eg : R600_3OP <0x4, "BFE_UINT",
-    [(set i32:$dst, (int_AMDIL_bit_extract_u32 i32:$src0, i32:$src1,
-                                               i32:$src2))],
-    VecALU
-  >;
-// XXX: This pattern is broken, disabling for now.  See comment in
-// AMDGPUInstructions.td for more info.
-//  def : BFEPattern <BFE_UINT_eg>;
-
-  def BFI_INT_eg : R600_3OP <0x06, "BFI_INT", [], VecALU>;
-  defm : BFIPatterns <BFI_INT_eg>;
-
-  def MULADD_UINT24_eg : R600_3OP <0x10, "MULADD_UINT24",
-    [(set i32:$dst, (add (mul U24:$src0, U24:$src1), i32:$src2))], VecALU
-  >;
-  def BIT_ALIGN_INT_eg : R600_3OP <0xC, "BIT_ALIGN_INT", [], VecALU>;
-  def : ROTRPattern <BIT_ALIGN_INT_eg>;
-
-  def MULADD_eg : MULADD_Common<0x14>;
-  def MULADD_IEEE_eg : MULADD_IEEE_Common<0x18>;
-  def ASHR_eg : ASHR_Common<0x15>;
-  def LSHR_eg : LSHR_Common<0x16>;
-  def LSHL_eg : LSHL_Common<0x17>;
-  def CNDE_eg : CNDE_Common<0x19>;
-  def CNDGT_eg : CNDGT_Common<0x1A>;
-  def CNDGE_eg : CNDGE_Common<0x1B>;
-  def MUL_LIT_eg : MUL_LIT_Common<0x1F>;
-  def LOG_CLAMPED_eg : LOG_CLAMPED_Common<0x82>;
-  def MUL_UINT24_eg : R600_2OP <0xB5, "MUL_UINT24",
-    [(set i32:$dst, (mul U24:$src0, U24:$src1))], VecALU
-  >;
-  def DOT4_eg : DOT4_Common<0xBE>;
-  defm CUBE_eg : CUBE_Common<0xC0>;
-
-let hasSideEffects = 1 in {
-  def MOVA_INT_eg : R600_1OP <0xCC, "MOVA_INT", []>;
-}
-
-  def TGSI_LIT_Z_eg : TGSI_LIT_Z_Common<MUL_LIT_eg, LOG_CLAMPED_eg, EXP_IEEE_eg>;
-
-  def FLT_TO_INT_eg : FLT_TO_INT_Common<0x50> {
-    let Pattern = [];
-    let Itinerary = AnyALU;
-  }
-
-  def INT_TO_FLT_eg : INT_TO_FLT_Common<0x9B>;
-
-  def FLT_TO_UINT_eg : FLT_TO_UINT_Common<0x9A> {
-    let Pattern = [];
-  }
-
-  def UINT_TO_FLT_eg : UINT_TO_FLT_Common<0x9C>;
-
-def GROUP_BARRIER : InstR600 <
-    (outs), (ins), "  GROUP_BARRIER", [(int_AMDGPU_barrier_local)], AnyALU>,
-    R600ALU_Word0,
-    R600ALU_Word1_OP2 <0x54> {
-
-  let dst = 0;
-  let dst_rel = 0;
-  let src0 = 0;
-  let src0_rel = 0;
-  let src0_neg = 0;
-  let src0_abs = 0;
-  let src1 = 0;
-  let src1_rel = 0;
-  let src1_neg = 0;
-  let src1_abs = 0;
-  let write = 0;
-  let omod = 0;
-  let clamp = 0;
-  let last = 1;
-  let bank_swizzle = 0;
-  let pred_sel = 0;
-  let update_exec_mask = 0;
-  let update_pred = 0;
-
-  let Inst{31-0}  = Word0;
-  let Inst{63-32} = Word1;
-
-  let ALUInst = 1;
-}
-
-//===----------------------------------------------------------------------===//
-// LDS Instructions
-//===----------------------------------------------------------------------===//
-class R600_LDS  <bits<6> op, dag outs, dag ins, string asm,
-                 list<dag> pattern = []> :
-
-    InstR600 <outs, ins, asm, pattern, XALU>,
-    R600_ALU_LDS_Word0,
-    R600LDS_Word1 {
-
-  bits<6>  offset = 0;
-  let lds_op = op;
-
-  let Word1{27} = offset{0};
-  let Word1{12} = offset{1};
-  let Word1{28} = offset{2};
-  let Word1{31} = offset{3};
-  let Word0{12} = offset{4};
-  let Word0{25} = offset{5};
-
-
-  let Inst{31-0}  = Word0;
-  let Inst{63-32} = Word1;
-
-  let ALUInst = 1;
-  let HasNativeOperands = 1;
-  let UseNamedOperandTable = 1;
-}
-
-class R600_LDS_1A <bits<6> lds_op, string name, list<dag> pattern> : R600_LDS <
-  lds_op,
-  (outs R600_Reg32:$dst),
-  (ins R600_Reg32:$src0, REL:$src0_rel, SEL:$src0_sel,
-       LAST:$last, R600_Pred:$pred_sel,
-       BANK_SWIZZLE:$bank_swizzle),
-  "  "#name#" $last OQAP, $src0$src0_rel $pred_sel",
-  pattern
-  > {
-
-  let src1 = 0;
-  let src1_rel = 0;
-  let src2 = 0;
-  let src2_rel = 0;
-
-  let usesCustomInserter = 1;
-  let LDS_1A = 1;
-  let DisableEncoding = "$dst";
-}
-
-class R600_LDS_1A1D <bits<6> lds_op, dag outs, string name, list<dag> pattern,
-                     string dst =""> :
-    R600_LDS <
-  lds_op, outs,
-  (ins R600_Reg32:$src0, REL:$src0_rel, SEL:$src0_sel,
-       R600_Reg32:$src1, REL:$src1_rel, SEL:$src1_sel,
-       LAST:$last, R600_Pred:$pred_sel,
-       BANK_SWIZZLE:$bank_swizzle),
-  "  "#name#" $last "#dst#"$src0$src0_rel, $src1$src1_rel, $pred_sel",
-  pattern
-  > {
-
-  field string BaseOp;
-
-  let src2 = 0;
-  let src2_rel = 0;
-  let LDS_1A1D = 1;
-}
-
-class R600_LDS_1A1D_NORET <bits<6> lds_op, string name, list<dag> pattern> :
-    R600_LDS_1A1D <lds_op, (outs), name, pattern> {
-  let BaseOp = name;
-}
-
-class R600_LDS_1A1D_RET <bits<6> lds_op, string name, list<dag> pattern> :
-    R600_LDS_1A1D <lds_op,  (outs R600_Reg32:$dst), name##"_RET", pattern, "OQAP, "> {
-
-  let BaseOp = name;
-  let usesCustomInserter = 1;
-  let DisableEncoding = "$dst";
-}
-
-class R600_LDS_1A2D <bits<6> lds_op, string name, list<dag> pattern> :
-    R600_LDS <
-  lds_op,
-  (outs),
-  (ins R600_Reg32:$src0, REL:$src0_rel, SEL:$src0_sel,
-       R600_Reg32:$src1, REL:$src1_rel, SEL:$src1_sel,
-       R600_Reg32:$src2, REL:$src2_rel, SEL:$src2_sel,
-       LAST:$last, R600_Pred:$pred_sel, BANK_SWIZZLE:$bank_swizzle),
-  "  "#name# "$last $src0$src0_rel, $src1$src1_rel, $src2$src2_rel, $pred_sel",
-  pattern> {
-  let LDS_1A2D = 1;
-}
-
-def LDS_ADD : R600_LDS_1A1D_NORET <0x0, "LDS_ADD", [] >;
-def LDS_SUB : R600_LDS_1A1D_NORET <0x1, "LDS_SUB", [] >;
-def LDS_WRITE : R600_LDS_1A1D_NORET <0xD, "LDS_WRITE",
-  [(local_store (i32 R600_Reg32:$src1), R600_Reg32:$src0)]
->;
-def LDS_BYTE_WRITE : R600_LDS_1A1D_NORET<0x12, "LDS_BYTE_WRITE",
-  [(truncstorei8_local i32:$src1, i32:$src0)]
->;
-def LDS_SHORT_WRITE : R600_LDS_1A1D_NORET<0x13, "LDS_SHORT_WRITE",
-  [(truncstorei16_local i32:$src1, i32:$src0)]
->;
-def LDS_ADD_RET : R600_LDS_1A1D_RET <0x20, "LDS_ADD",
-  [(set i32:$dst, (atomic_load_add_local i32:$src0, i32:$src1))]
->;
-def LDS_SUB_RET : R600_LDS_1A1D_RET <0x21, "LDS_SUB",
-  [(set i32:$dst, (atomic_load_sub_local i32:$src0, i32:$src1))]
->;
-def LDS_READ_RET : R600_LDS_1A <0x32, "LDS_READ_RET",
-  [(set (i32 R600_Reg32:$dst), (local_load R600_Reg32:$src0))]
->;
-def LDS_BYTE_READ_RET : R600_LDS_1A <0x36, "LDS_BYTE_READ_RET",
-  [(set i32:$dst, (sextloadi8_local i32:$src0))]
->;
-def LDS_UBYTE_READ_RET : R600_LDS_1A <0x37, "LDS_UBYTE_READ_RET",
-  [(set i32:$dst, (az_extloadi8_local i32:$src0))]
->;
-def LDS_SHORT_READ_RET : R600_LDS_1A <0x38, "LDS_SHORT_READ_RET",
-  [(set i32:$dst, (sextloadi16_local i32:$src0))]
->;
-def LDS_USHORT_READ_RET : R600_LDS_1A <0x39, "LDS_USHORT_READ_RET",
-  [(set i32:$dst, (az_extloadi16_local i32:$src0))]
->;
-
-  // TRUNC is used for the FLT_TO_INT instructions to work around a
-  // perceived problem where the rounding modes are applied differently
-  // depending on the instruction and the slot they are in.
-  // See:
-  // https://bugs.freedesktop.org/show_bug.cgi?id=50232
-  // Mesa commit: a1a0974401c467cb86ef818f22df67c21774a38c
-  //
-  // XXX: Lowering SELECT_CC will sometimes generate fp_to_[su]int nodes,
-  // which do not need to be truncated since the fp values are 0.0f or 1.0f.
-  // We should look into handling these cases separately.
-  def : Pat<(fp_to_sint f32:$src0), (FLT_TO_INT_eg (TRUNC $src0))>;
-
-  def : Pat<(fp_to_uint f32:$src0), (FLT_TO_UINT_eg (TRUNC $src0))>;
-
-  // SHA-256 Patterns
-  def : SHA256MaPattern <BFI_INT_eg, XOR_INT>;
-
-  def : FROUNDPat <CNDGE_eg>;
-
-  def EG_ExportSwz : ExportSwzInst {
-    let Word1{19-16} = 0; // BURST_COUNT
-    let Word1{20} = 0; // VALID_PIXEL_MODE
-    let Word1{21} = eop;
-    let Word1{29-22} = inst;
-    let Word1{30} = 0; // MARK
-    let Word1{31} = 1; // BARRIER
-  }
-  defm : ExportPattern<EG_ExportSwz, 83>;
-
-  def EG_ExportBuf : ExportBufInst {
-    let Word1{19-16} = 0; // BURST_COUNT
-    let Word1{20} = 0; // VALID_PIXEL_MODE
-    let Word1{21} = eop;
-    let Word1{29-22} = inst;
-    let Word1{30} = 0; // MARK
-    let Word1{31} = 1; // BARRIER
-  }
-  defm : SteamOutputExportPattern<EG_ExportBuf, 0x40, 0x41, 0x42, 0x43>;
-
-  def CF_TC_EG : CF_CLAUSE_EG<1, (ins i32imm:$ADDR, i32imm:$COUNT),
-  "TEX $COUNT @$ADDR"> {
-    let POP_COUNT = 0;
-  }
-  def CF_VC_EG : CF_CLAUSE_EG<2, (ins i32imm:$ADDR, i32imm:$COUNT),
-  "VTX $COUNT @$ADDR"> {
-    let POP_COUNT = 0;
-  }
-  def WHILE_LOOP_EG : CF_CLAUSE_EG<6, (ins i32imm:$ADDR),
-  "LOOP_START_DX10 @$ADDR"> {
-    let POP_COUNT = 0;
-    let COUNT = 0;
-  }
-  def END_LOOP_EG : CF_CLAUSE_EG<5, (ins i32imm:$ADDR), "END_LOOP @$ADDR"> {
-    let POP_COUNT = 0;
-    let COUNT = 0;
-  }
-  def LOOP_BREAK_EG : CF_CLAUSE_EG<9, (ins i32imm:$ADDR),
-  "LOOP_BREAK @$ADDR"> {
-    let POP_COUNT = 0;
-    let COUNT = 0;
-  }
-  def CF_CONTINUE_EG : CF_CLAUSE_EG<8, (ins i32imm:$ADDR),
-  "CONTINUE @$ADDR"> {
-    let POP_COUNT = 0;
-    let COUNT = 0;
-  }
-  def CF_JUMP_EG : CF_CLAUSE_EG<10, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
-  "JUMP @$ADDR POP:$POP_COUNT"> {
-    let COUNT = 0;
-  }
-  def CF_ELSE_EG : CF_CLAUSE_EG<13, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
-  "ELSE @$ADDR POP:$POP_COUNT"> {
-    let COUNT = 0;
-  }
-  def CF_CALL_FS_EG : CF_CLAUSE_EG<19, (ins), "CALL_FS"> {
-    let ADDR = 0;
-    let COUNT = 0;
-    let POP_COUNT = 0;
-  }
-  def POP_EG : CF_CLAUSE_EG<14, (ins i32imm:$ADDR, i32imm:$POP_COUNT),
-  "POP @$ADDR POP:$POP_COUNT"> {
-    let COUNT = 0;
-  }
-  def CF_END_EG :  CF_CLAUSE_EG<0, (ins), "CF_END"> {
-    let COUNT = 0;
-    let POP_COUNT = 0;
-    let ADDR = 0;
-    let END_OF_PROGRAM = 1;
-  }
-
-} // End Predicates = [isEGorCayman]
 
 //===----------------------------------------------------------------------===//
 // Regist loads and stores - for indirect addressing
@@ -1819,217 +1264,6 @@ def LDS_USHORT_READ_RET : R600_LDS_1A <0x39, "LDS_USHORT_READ_RET",
 
 defm R600_ : RegisterLoadStore <R600_Reg32, FRAMEri, ADDRIndirect>;
 
-//===----------------------------------------------------------------------===//
-// Cayman Instructions
-//===----------------------------------------------------------------------===//
-
-let Predicates = [isCayman] in {
-
-def MULADD_INT24_cm : R600_3OP <0x08, "MULADD_INT24",
-  [(set i32:$dst, (add (mul I24:$src0, I24:$src1), i32:$src2))], VecALU
->;
-def MUL_INT24_cm : R600_2OP <0x5B, "MUL_INT24",
-  [(set i32:$dst, (mul I24:$src0, I24:$src1))], VecALU
->;
-
-let isVector = 1 in {
-
-def RECIP_IEEE_cm : RECIP_IEEE_Common<0x86>;
-
-def MULLO_INT_cm : MULLO_INT_Common<0x8F>;
-def MULHI_INT_cm : MULHI_INT_Common<0x90>;
-def MULLO_UINT_cm : MULLO_UINT_Common<0x91>;
-def MULHI_UINT_cm : MULHI_UINT_Common<0x92>;
-def RECIPSQRT_CLAMPED_cm : RECIPSQRT_CLAMPED_Common<0x87>;
-def EXP_IEEE_cm : EXP_IEEE_Common<0x81>;
-def LOG_IEEE_cm : LOG_IEEE_Common<0x83>;
-def RECIP_CLAMPED_cm : RECIP_CLAMPED_Common<0x84>;
-def RECIPSQRT_IEEE_cm : RECIPSQRT_IEEE_Common<0x89>;
-def SIN_cm : SIN_Common<0x8D>;
-def COS_cm : COS_Common<0x8E>;
-} // End isVector = 1
-
-def : POW_Common <LOG_IEEE_cm, EXP_IEEE_cm, MUL>;
-
-defm DIV_cm : DIV_Common<RECIP_IEEE_cm>;
-
-// RECIP_UINT emulation for Cayman
-// The multiplication scales from [0,1] to the unsigned integer range
-def : Pat <
-  (AMDGPUurecip i32:$src0),
-  (FLT_TO_UINT_eg (MUL_IEEE (RECIP_IEEE_cm (UINT_TO_FLT_eg $src0)),
-                            (MOV_IMM_I32 CONST.FP_UINT_MAX_PLUS_1)))
->;
-
-  def CF_END_CM : CF_CLAUSE_EG<32, (ins), "CF_END"> {
-    let ADDR = 0;
-    let POP_COUNT = 0;
-    let COUNT = 0;
-  }
-
-def : Pat<(fsqrt f32:$src), (MUL R600_Reg32:$src, (RECIPSQRT_CLAMPED_cm $src))>;
-
-class RAT_STORE_DWORD <RegisterClass rc, ValueType vt, bits<4> mask> :
-  CF_MEM_RAT_CACHELESS <0x14, 0, mask,
-                        (ins rc:$rw_gpr, R600_TReg32_X:$index_gpr),
-                        "STORE_DWORD $rw_gpr, $index_gpr",
-                        [(global_store vt:$rw_gpr, i32:$index_gpr)]> {
-  let eop = 0; // This bit is not used on Cayman.
-}
-
-def RAT_STORE_DWORD32 : RAT_STORE_DWORD <R600_TReg32_X, i32, 0x1>;
-def RAT_STORE_DWORD64 : RAT_STORE_DWORD <R600_Reg64, v2i32, 0x3>;
-def RAT_STORE_DWORD128 : RAT_STORE_DWORD <R600_Reg128, v4i32, 0xf>;
-
-class VTX_READ_cm <string name, bits<8> buffer_id, dag outs, list<dag> pattern>
-    : VTX_WORD0_cm, VTX_READ<name, buffer_id, outs, pattern> {
-
-  // Static fields
-  let VC_INST = 0;
-  let FETCH_TYPE = 2;
-  let FETCH_WHOLE_QUAD = 0;
-  let BUFFER_ID = buffer_id;
-  let SRC_REL = 0;
-  // XXX: We can infer this field based on the SRC_GPR.  This would allow us
-  // to store vertex addresses in any channel, not just X.
-  let SRC_SEL_X = 0;
-  let SRC_SEL_Y = 0;
-  let STRUCTURED_READ = 0;
-  let LDS_REQ = 0;
-  let COALESCED_READ = 0;
-
-  let Inst{31-0} = Word0;
-}
-
-class VTX_READ_8_cm <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_cm <"VTX_READ_8 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
-
-  let DST_SEL_X = 0;
-  let DST_SEL_Y = 7;   // Masked
-  let DST_SEL_Z = 7;   // Masked
-  let DST_SEL_W = 7;   // Masked
-  let DATA_FORMAT = 1; // FMT_8
-}
-
-class VTX_READ_16_cm <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_cm <"VTX_READ_16 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
-  let DST_SEL_X = 0;
-  let DST_SEL_Y = 7;   // Masked
-  let DST_SEL_Z = 7;   // Masked
-  let DST_SEL_W = 7;   // Masked
-  let DATA_FORMAT = 5; // FMT_16
-
-}
-
-class VTX_READ_32_cm <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_cm <"VTX_READ_32 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_TReg32_X:$dst_gpr), pattern> {
-
-  let DST_SEL_X        = 0;
-  let DST_SEL_Y        = 7;   // Masked
-  let DST_SEL_Z        = 7;   // Masked
-  let DST_SEL_W        = 7;   // Masked
-  let DATA_FORMAT      = 0xD; // COLOR_32
-
-  // This is not really necessary, but there were some GPU hangs that appeared
-  // to be caused by ALU instructions in the next instruction group that wrote
-  // to the $src_gpr registers of the VTX_READ.
-  // e.g.
-  // %T3_X<def> = VTX_READ_PARAM_32_eg %T2_X<kill>, 24
-  // %T2_X<def> = MOV %ZERO
-  //Adding this constraint prevents this from happening.
-  let Constraints = "$src_gpr.ptr = $dst_gpr";
-}
-
-class VTX_READ_64_cm <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_cm <"VTX_READ_64 $dst_gpr, $src_gpr", buffer_id,
-                   (outs R600_Reg64:$dst_gpr), pattern> {
-
-  let DST_SEL_X        = 0;
-  let DST_SEL_Y        = 1;
-  let DST_SEL_Z        = 7;
-  let DST_SEL_W        = 7;
-  let DATA_FORMAT      = 0x1D; // COLOR_32_32
-}
-
-class VTX_READ_128_cm <bits<8> buffer_id, list<dag> pattern>
-    : VTX_READ_cm <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr", buffer_id,
-                   (outs R600_Reg128:$dst_gpr), pattern> {
-
-  let DST_SEL_X        =  0;
-  let DST_SEL_Y        =  1;
-  let DST_SEL_Z        =  2;
-  let DST_SEL_W        =  3;
-  let DATA_FORMAT      =  0x22; // COLOR_32_32_32_32
-
-  // XXX: Need to force VTX_READ_128 instructions to write to the same register
-  // that holds its buffer address to avoid potential hangs.  We can't use
-  // the same constraint as VTX_READ_32_eg, because the $src_gpr.ptr and $dst
-  // registers are different sizes.
-}
-
-//===----------------------------------------------------------------------===//
-// VTX Read from parameter memory space
-//===----------------------------------------------------------------------===//
-def VTX_READ_PARAM_8_cm : VTX_READ_8_cm <0,
-  [(set i32:$dst_gpr, (load_param_exti8 ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_16_cm : VTX_READ_16_cm <0,
-  [(set i32:$dst_gpr, (load_param_exti16 ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_32_cm : VTX_READ_32_cm <0,
-  [(set i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_64_cm : VTX_READ_64_cm <0,
-  [(set v2i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_PARAM_128_cm : VTX_READ_128_cm <0,
-  [(set v4i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
->;
-
-//===----------------------------------------------------------------------===//
-// VTX Read from global memory space
-//===----------------------------------------------------------------------===//
-
-// 8-bit reads
-def VTX_READ_GLOBAL_8_cm : VTX_READ_8_cm <1,
-  [(set i32:$dst_gpr, (az_extloadi8_global ADDRVTX_READ:$src_gpr))]
->;
-
-def VTX_READ_GLOBAL_16_cm : VTX_READ_16_cm <1,
-  [(set i32:$dst_gpr, (az_extloadi16_global ADDRVTX_READ:$src_gpr))]
->;
-
-// 32-bit reads
-def VTX_READ_GLOBAL_32_cm : VTX_READ_32_cm <1,
-  [(set i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 64-bit reads
-def VTX_READ_GLOBAL_64_cm : VTX_READ_64_cm <1,
-  [(set v2i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
->;
-
-// 128-bit reads
-def VTX_READ_GLOBAL_128_cm : VTX_READ_128_cm <1,
-  [(set v4i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
->;
-
-} // End isCayman
-
-//===----------------------------------------------------------------------===//
-// Branch Instructions
-//===----------------------------------------------------------------------===//
-
-
-def IF_PREDICATE_SET  : ILFormat<(outs), (ins GPRI32:$src),
-  "IF_PREDICATE_SET $src", []>;
 
 //===----------------------------------------------------------------------===//
 // Pseudo instructions
@@ -2104,15 +1338,6 @@ def TXD_SHADOW: InstR600 <
 } // End isPseudo = 1
 } // End usesCustomInserter = 1
 
-//===---------------------------------------------------------------------===//
-// Return instruction
-//===---------------------------------------------------------------------===//
-let isTerminator = 1, isReturn = 1, hasCtrlDep = 1,
-    usesCustomInserter = 1 in {
-  def RETURN          : ILFormat<(outs), (ins variable_ops),
-      "RETURN", [(IL_retflag)]>;
-}
-
 
 //===----------------------------------------------------------------------===//
 // Constant Buffer Addressing Support
@@ -2239,14 +1464,55 @@ let Inst{63-32} = Word1;
   let VTXInst = 1;
 }
 
+//===---------------------------------------------------------------------===//
+// Flow and Program control Instructions
+//===---------------------------------------------------------------------===//
+class ILFormat<dag outs, dag ins, string asmstr, list<dag> pattern>
+: Instruction {
+
+     let Namespace = "AMDGPU";
+     dag OutOperandList = outs;
+     dag InOperandList = ins;
+     let Pattern = pattern;
+     let AsmString = !strconcat(asmstr, "\n");
+     let isPseudo = 1;
+     let Itinerary = NullALU;
+     bit hasIEEEFlag = 0;
+     bit hasZeroOpFlag = 0;
+     let mayLoad = 0;
+     let mayStore = 0;
+     let hasSideEffects = 0;
+}
+
+multiclass BranchConditional<SDNode Op, RegisterClass rci, RegisterClass rcf> {
+    def _i32 : ILFormat<(outs),
+  (ins brtarget:$target, rci:$src0),
+        "; i32 Pseudo branch instruction",
+  [(Op bb:$target, (i32 rci:$src0))]>;
+    def _f32 : ILFormat<(outs),
+  (ins brtarget:$target, rcf:$src0),
+        "; f32 Pseudo branch instruction",
+  [(Op bb:$target, (f32 rcf:$src0))]>;
+}
+
+// Only scalar types should generate flow control
+multiclass BranchInstr<string name> {
+  def _i32 : ILFormat<(outs), (ins R600_Reg32:$src),
+      !strconcat(name, " $src"), []>;
+  def _f32 : ILFormat<(outs), (ins R600_Reg32:$src),
+      !strconcat(name, " $src"), []>;
+}
+// Only scalar types should generate flow control
+multiclass BranchInstr2<string name> {
+  def _i32 : ILFormat<(outs), (ins R600_Reg32:$src0, R600_Reg32:$src1),
+      !strconcat(name, " $src0, $src1"), []>;
+  def _f32 : ILFormat<(outs), (ins R600_Reg32:$src0, R600_Reg32:$src1),
+      !strconcat(name, " $src0, $src1"), []>;
+}
 
-
-//===--------------------------------------------------------------------===//
-// Instructions support
-//===--------------------------------------------------------------------===//
 //===---------------------------------------------------------------------===//
 // Custom Inserter for Branches and returns, this eventually will be a
-// seperate pass
+// separate pass
 //===---------------------------------------------------------------------===//
 let isTerminator = 1, usesCustomInserter = 1, isBranch = 1, isBarrier = 1 in {
   def BRANCH : ILFormat<(outs), (ins brtarget:$target),
@@ -2256,13 +1522,22 @@ let isTerminator = 1, usesCustomInserter = 1, isBranch = 1, isBarrier = 1 in {
 }
 
 //===---------------------------------------------------------------------===//
-// Flow and Program control Instructions
+// Return instruction
 //===---------------------------------------------------------------------===//
+let isTerminator = 1, isReturn = 1, hasCtrlDep = 1,
+    usesCustomInserter = 1 in {
+  def RETURN          : ILFormat<(outs), (ins variable_ops),
+      "RETURN", [(IL_retflag)]>;
+}
+
+//===----------------------------------------------------------------------===//
+// Branch Instructions
+//===----------------------------------------------------------------------===//
+
+def IF_PREDICATE_SET  : ILFormat<(outs), (ins R600_Reg32:$src),
+  "IF_PREDICATE_SET $src", []>;
+
 let isTerminator=1 in {
-  def SWITCH      : ILFormat< (outs), (ins GPRI32:$src),
-  !strconcat("SWITCH", " $src"), []>;
-  def CASE        : ILFormat< (outs), (ins GPRI32:$src),
-      !strconcat("CASE", " $src"), []>;
   def BREAK       : ILFormat< (outs), (ins),
       "BREAK", []>;
   def CONTINUE    : ILFormat< (outs), (ins),
@@ -2307,6 +1582,60 @@ let isTerminator=1 in {
 }
 
 //===----------------------------------------------------------------------===//
+// Indirect addressing pseudo instructions
+//===----------------------------------------------------------------------===//
+
+let isPseudo = 1 in {
+
+class ExtractVertical <RegisterClass vec_rc> : InstR600 <
+  (outs R600_Reg32:$dst),
+  (ins vec_rc:$vec, R600_Reg32:$index), "",
+  [],
+  AnyALU
+>;
+
+let Constraints = "$dst = $vec" in {
+
+class InsertVertical <RegisterClass vec_rc> : InstR600 <
+  (outs vec_rc:$dst),
+  (ins vec_rc:$vec, R600_Reg32:$value, R600_Reg32:$index), "",
+  [],
+  AnyALU
+>;
+
+} // End Constraints = "$dst = $vec"
+
+} // End isPseudo = 1
+
+def R600_EXTRACT_ELT_V2 : ExtractVertical <R600_Reg64Vertical>;
+def R600_EXTRACT_ELT_V4 : ExtractVertical <R600_Reg128Vertical>;
+
+def R600_INSERT_ELT_V2 : InsertVertical <R600_Reg64Vertical>;
+def R600_INSERT_ELT_V4 : InsertVertical <R600_Reg128Vertical>;
+
+class ExtractVerticalPat <Instruction inst, ValueType vec_ty,
+                          ValueType scalar_ty> : Pat <
+  (scalar_ty (extractelt vec_ty:$vec, i32:$index)),
+  (inst $vec, $index)
+>;
+
+def : ExtractVerticalPat <R600_EXTRACT_ELT_V2, v2i32, i32>;
+def : ExtractVerticalPat <R600_EXTRACT_ELT_V2, v2f32, f32>;
+def : ExtractVerticalPat <R600_EXTRACT_ELT_V4, v4i32, i32>;
+def : ExtractVerticalPat <R600_EXTRACT_ELT_V4, v4f32, f32>;
+
+class InsertVerticalPat <Instruction inst, ValueType vec_ty,
+                         ValueType scalar_ty> : Pat <
+  (vec_ty (insertelt vec_ty:$vec, scalar_ty:$value, i32:$index)),
+  (inst $vec, $value, $index)
+>;
+
+def : InsertVerticalPat <R600_INSERT_ELT_V2, v2i32, i32>;
+def : InsertVerticalPat <R600_INSERT_ELT_V2, v2f32, f32>;
+def : InsertVerticalPat <R600_INSERT_ELT_V4, v4i32, i32>;
+def : InsertVerticalPat <R600_INSERT_ELT_V4, v4f32, f32>;
+
+//===----------------------------------------------------------------------===//
 // ISel Patterns
 //===----------------------------------------------------------------------===//
 
@@ -2358,9 +1687,6 @@ def : Insert_Element <i32, v4i32, 1, sub1>;
 def : Insert_Element <i32, v4i32, 2, sub2>;
 def : Insert_Element <i32, v4i32, 3, sub3>;
 
-def : Vector4_Build <v4f32, f32>;
-def : Vector4_Build <v4i32, i32>;
-
 def : Extract_Element <f32, v2f32, 0, sub0>;
 def : Extract_Element <f32, v2f32, 1, sub1>;
 
@@ -2387,6 +1713,12 @@ def : DwordAddrPat  <i32, R600_Reg32>;
 
 } // End isR600toCayman Predicate
 
+let Predicates = [isR600] in {
+// Intrinsic patterns
+defm : Expand24IBitOps<MULLO_INT_r600, ADD_INT>;
+defm : Expand24UBitOps<MULLO_UINT_r600, ADD_INT>;
+} // End isR600
+
 def getLDSNoRetOp : InstrMapping {
   let FilterClass = "R600_LDS_1A1D";
   let RowFields = ["BaseOp"];
diff --git a/contrib/llvm/lib/Target/R600/R600MachineFunctionInfo.h b/contrib/llvm/lib/Target/R600/R600MachineFunctionInfo.h
index c1bec0a..b0ae22e 100644
--- a/contrib/llvm/lib/Target/R600/R600MachineFunctionInfo.h
+++ b/contrib/llvm/lib/Target/R600/R600MachineFunctionInfo.h
@@ -21,7 +21,7 @@
 namespace llvm {
 
 class R600MachineFunctionInfo : public AMDGPUMachineFunction {
-  virtual void anchor();
+  void anchor() override;
 public:
   R600MachineFunctionInfo(const MachineFunction &MF);
   SmallVector<unsigned, 4> LiveOuts;
diff --git a/contrib/llvm/lib/Target/R600/R600MachineScheduler.cpp b/contrib/llvm/lib/Target/R600/R600MachineScheduler.cpp
index da2a4d8..7ea654c 100644
--- a/contrib/llvm/lib/Target/R600/R600MachineScheduler.cpp
+++ b/contrib/llvm/lib/Target/R600/R600MachineScheduler.cpp
@@ -12,9 +12,8 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "misched"
-
 #include "R600MachineScheduler.h"
+#include "AMDGPUSubtarget.h"
 #include "llvm/CodeGen/LiveIntervalAnalysis.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Pass.h"
@@ -23,9 +22,11 @@
 
 using namespace llvm;
 
-void R600SchedStrategy::initialize(ScheduleDAGMI *dag) {
+#define DEBUG_TYPE "misched"
 
-  DAG = dag;
+void R600SchedStrategy::initialize(ScheduleDAGMI *dag) {
+  assert(dag->hasVRegLiveness() && "R600SchedStrategy needs vreg liveness");
+  DAG = static_cast<ScheduleDAGMILive*>(dag);
   TII = static_cast<const R600InstrInfo*>(DAG->TII);
   TRI = static_cast<const R600RegisterInfo*>(DAG->TRI);
   VLIW5 = !DAG->MF.getTarget().getSubtarget<AMDGPUSubtarget>().hasCaymanISA();
@@ -56,7 +57,7 @@ unsigned getWFCountLimitedByGPR(unsigned GPRCount) {
 }
 
 SUnit* R600SchedStrategy::pickNode(bool &IsTopNode) {
-  SUnit *SU = 0;
+  SUnit *SU = nullptr;
   NextInstKind = IDOther;
 
   IsTopNode = false;
@@ -72,7 +73,7 @@ SUnit* R600SchedStrategy::pickNode(bool &IsTopNode) {
     // OpenCL Programming Guide :
     // The approx. number of WF that allows TEX inst to hide ALU inst is :
     // 500 (cycles for TEX) / (AluFetchRatio * 8 (cycles for ALU))
-    float ALUFetchRationEstimate = 
+    float ALUFetchRationEstimate =
         (AluInstCount + AvailablesAluCount() + Pending[IDAlu].size()) /
         (FetchInstCount + Available[IDFetch].size());
     unsigned NeededWF = 62.5f / ALUFetchRationEstimate;
@@ -316,7 +317,7 @@ int R600SchedStrategy::getInstKind(SUnit* SU) {
 
 SUnit *R600SchedStrategy::PopInst(std::vector<SUnit *> &Q, bool AnyALU) {
   if (Q.empty())
-    return NULL;
+    return nullptr;
   for (std::vector<SUnit *>::reverse_iterator It = Q.rbegin(), E = Q.rend();
       It != E; ++It) {
     SUnit *SU = *It;
@@ -331,7 +332,7 @@ SUnit *R600SchedStrategy::PopInst(std::vector<SUnit *> &Q, bool AnyALU) {
       InstructionsGroupCandidate.pop_back();
     }
   }
-  return NULL;
+  return nullptr;
 }
 
 void R600SchedStrategy::LoadAlu() {
@@ -448,11 +449,11 @@ SUnit* R600SchedStrategy::pickAlu() {
     }
     PrepareNextSlot();
   }
-  return NULL;
+  return nullptr;
 }
 
 SUnit* R600SchedStrategy::pickOther(int QID) {
-  SUnit *SU = 0;
+  SUnit *SU = nullptr;
   std::vector<SUnit *> &AQ = Available[QID];
 
   if (AQ.empty()) {
@@ -464,4 +465,3 @@ SUnit* R600SchedStrategy::pickOther(int QID) {
   }
   return SU;
 }
-
diff --git a/contrib/llvm/lib/Target/R600/R600MachineScheduler.h b/contrib/llvm/lib/Target/R600/R600MachineScheduler.h
index 97c8cde..fd475af 100644
--- a/contrib/llvm/lib/Target/R600/R600MachineScheduler.h
+++ b/contrib/llvm/lib/Target/R600/R600MachineScheduler.h
@@ -26,7 +26,7 @@ namespace llvm {
 
 class R600SchedStrategy : public MachineSchedStrategy {
 
-  const ScheduleDAGMI *DAG;
+  const ScheduleDAGMILive *DAG;
   const R600InstrInfo *TII;
   const R600RegisterInfo *TRI;
   MachineRegisterInfo *MRI;
@@ -68,17 +68,16 @@ class R600SchedStrategy : public MachineSchedStrategy {
 
 public:
   R600SchedStrategy() :
-    DAG(0), TII(0), TRI(0), MRI(0) {
+    DAG(nullptr), TII(nullptr), TRI(nullptr), MRI(nullptr) {
   }
 
-  virtual ~R600SchedStrategy() {
-  }
+  virtual ~R600SchedStrategy() {}
 
-  virtual void initialize(ScheduleDAGMI *dag);
-  virtual SUnit *pickNode(bool &IsTopNode);
-  virtual void schedNode(SUnit *SU, bool IsTopNode);
-  virtual void releaseTopNode(SUnit *SU);
-  virtual void releaseBottomNode(SUnit *SU);
+  void initialize(ScheduleDAGMI *dag) override;
+  SUnit *pickNode(bool &IsTopNode) override;
+  void schedNode(SUnit *SU, bool IsTopNode) override;
+  void releaseTopNode(SUnit *SU) override;
+  void releaseBottomNode(SUnit *SU) override;
 
 private:
   std::vector<MachineInstr *> InstructionsGroupCandidate;
diff --git a/contrib/llvm/lib/Target/R600/R600OptimizeVectorRegisters.cpp b/contrib/llvm/lib/Target/R600/R600OptimizeVectorRegisters.cpp
index cf719c0..2314136 100644
--- a/contrib/llvm/lib/Target/R600/R600OptimizeVectorRegisters.cpp
+++ b/contrib/llvm/lib/Target/R600/R600OptimizeVectorRegisters.cpp
@@ -27,27 +27,28 @@
 /// to reduce MOV count.
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "vec-merger"
 #include "llvm/Support/Debug.h"
 #include "AMDGPU.h"
 #include "R600InstrInfo.h"
 #include "llvm/CodeGen/DFAPacketizer.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineLoopInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/Passes.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
 
 using namespace llvm;
 
+#define DEBUG_TYPE "vec-merger"
+
 namespace {
 
 static bool
 isImplicitlyDef(MachineRegisterInfo &MRI, unsigned Reg) {
-  for (MachineRegisterInfo::def_iterator It = MRI.def_begin(Reg),
-      E = MRI.def_end(); It != E; ++It) {
+  for (MachineRegisterInfo::def_instr_iterator It = MRI.def_instr_begin(Reg),
+      E = MRI.def_instr_end(); It != E; ++It) {
     return (*It).isImplicitDef();
   }
   if (MRI.isReserved(Reg)) {
@@ -63,7 +64,7 @@ public:
   DenseMap<unsigned, unsigned> RegToChan;
   std::vector<unsigned> UndefReg;
   RegSeqInfo(MachineRegisterInfo &MRI, MachineInstr *MI) : Instr(MI) {
-    assert (MI->getOpcode() == AMDGPU::REG_SEQUENCE);
+    assert(MI->getOpcode() == AMDGPU::REG_SEQUENCE);
     for (unsigned i = 1, e = Instr->getNumOperands(); i < e; i+=2) {
       MachineOperand &MO = Instr->getOperand(i);
       unsigned Chan = Instr->getOperand(i + 1).getImm();
@@ -107,9 +108,9 @@ private:
 public:
   static char ID;
   R600VectorRegMerger(TargetMachine &tm) : MachineFunctionPass(ID),
-  TII(0) { }
+  TII(nullptr) { }
 
-  void getAnalysisUsage(AnalysisUsage &AU) const {
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.setPreservesCFG();
     AU.addRequired<MachineDominatorTree>();
     AU.addPreserved<MachineDominatorTree>();
@@ -118,11 +119,11 @@ public:
     MachineFunctionPass::getAnalysisUsage(AU);
   }
 
-  const char *getPassName() const {
+  const char *getPassName() const override {
     return "R600 Vector Registers Merge Pass";
   }
 
-  bool runOnMachineFunction(MachineFunction &Fn);
+  bool runOnMachineFunction(MachineFunction &Fn) override;
 };
 
 char R600VectorRegMerger::ID = 0;
@@ -213,8 +214,8 @@ MachineInstr *R600VectorRegMerger::RebuildVector(
   DEBUG(dbgs() << "    ->"; Pos->dump(););
 
   DEBUG(dbgs() << "  Updating Swizzle:\n");
-  for (MachineRegisterInfo::use_iterator It = MRI->use_begin(Reg),
-      E = MRI->use_end(); It != E; ++It) {
+  for (MachineRegisterInfo::use_instr_iterator It = MRI->use_instr_begin(Reg),
+      E = MRI->use_instr_end(); It != E; ++It) {
     DEBUG(dbgs() << "    ";(*It).dump(); dbgs() << "    ->");
     SwizzleInput(*It, RemapChan);
     DEBUG((*It).dump());
@@ -261,8 +262,8 @@ void R600VectorRegMerger::SwizzleInput(MachineInstr &MI,
 }
 
 bool R600VectorRegMerger::areAllUsesSwizzeable(unsigned Reg) const {
-  for (MachineRegisterInfo::use_iterator It = MRI->use_begin(Reg),
-      E = MRI->use_end(); It != E; ++It) {
+  for (MachineRegisterInfo::use_instr_iterator It = MRI->use_instr_begin(Reg),
+      E = MRI->use_instr_end(); It != E; ++It) {
     if (!canSwizzle(*It))
       return false;
   }
@@ -328,8 +329,9 @@ bool R600VectorRegMerger::runOnMachineFunction(MachineFunction &Fn) {
       if (MI->getOpcode() != AMDGPU::REG_SEQUENCE) {
         if (TII->get(MI->getOpcode()).TSFlags & R600_InstFlag::TEX_INST) {
           unsigned Reg = MI->getOperand(1).getReg();
-          for (MachineRegisterInfo::def_iterator It = MRI->def_begin(Reg),
-              E = MRI->def_end(); It != E; ++It) {
+          for (MachineRegisterInfo::def_instr_iterator
+               It = MRI->def_instr_begin(Reg), E = MRI->def_instr_end();
+               It != E; ++It) {
             RemoveMI(&(*It));
           }
         }
diff --git a/contrib/llvm/lib/Target/R600/R600Packetizer.cpp b/contrib/llvm/lib/Target/R600/R600Packetizer.cpp
index cd9b6ea..74cf309 100644
--- a/contrib/llvm/lib/Target/R600/R600Packetizer.cpp
+++ b/contrib/llvm/lib/Target/R600/R600Packetizer.cpp
@@ -14,9 +14,9 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "packets"
 #include "llvm/Support/Debug.h"
 #include "AMDGPU.h"
+#include "AMDGPUSubtarget.h"
 #include "R600InstrInfo.h"
 #include "llvm/CodeGen/DFAPacketizer.h"
 #include "llvm/CodeGen/MachineDominators.h"
@@ -28,6 +28,8 @@
 
 using namespace llvm;
 
+#define DEBUG_TYPE "packets"
+
 namespace {
 
 class R600Packetizer : public MachineFunctionPass {
@@ -36,7 +38,7 @@ public:
   static char ID;
   R600Packetizer(const TargetMachine &TM) : MachineFunctionPass(ID) {}
 
-  void getAnalysisUsage(AnalysisUsage &AU) const {
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.setPreservesCFG();
     AU.addRequired<MachineDominatorTree>();
     AU.addPreserved<MachineDominatorTree>();
@@ -45,11 +47,11 @@ public:
     MachineFunctionPass::getAnalysisUsage(AU);
   }
 
-  const char *getPassName() const {
+  const char *getPassName() const override {
     return "R600 Packetizer";
   }
 
-  bool runOnMachineFunction(MachineFunction &Fn);
+  bool runOnMachineFunction(MachineFunction &Fn) override;
 };
 char R600Packetizer::ID = 0;
 
@@ -66,7 +68,7 @@ private:
   }
 
   /// \returns register to PV chan mapping for bundle/single instructions that
-  /// immediatly precedes I.
+  /// immediately precedes I.
   DenseMap<unsigned, unsigned> getPreviousVector(MachineBasicBlock::iterator I)
       const {
     DenseMap<unsigned, unsigned> Result;
@@ -155,18 +157,19 @@ public:
   }
 
   // initPacketizerState - initialize some internal flags.
-  void initPacketizerState() {
+  void initPacketizerState() override {
     ConsideredInstUsesAlreadyWrittenVectorElement = false;
   }
 
   // ignorePseudoInstruction - Ignore bundling of pseudo instructions.
-  bool ignorePseudoInstruction(MachineInstr *MI, MachineBasicBlock *MBB) {
+  bool ignorePseudoInstruction(MachineInstr *MI,
+                               MachineBasicBlock *MBB) override {
     return false;
   }
 
   // isSoloInstruction - return true if instruction MI can not be packetized
   // with any other instruction, which means that MI itself is a packet.
-  bool isSoloInstruction(MachineInstr *MI) {
+  bool isSoloInstruction(MachineInstr *MI) override {
     if (TII->isVector(*MI))
       return true;
     if (!TII->isALUInstr(MI->getOpcode()))
@@ -182,7 +185,7 @@ public:
 
   // isLegalToPacketizeTogether - Is it legal to packetize SUI and SUJ
   // together.
-  bool isLegalToPacketizeTogether(SUnit *SUI, SUnit *SUJ) {
+  bool isLegalToPacketizeTogether(SUnit *SUI, SUnit *SUJ) override {
     MachineInstr *MII = SUI->getInstr(), *MIJ = SUJ->getInstr();
     if (getSlot(MII) == getSlot(MIJ))
       ConsideredInstUsesAlreadyWrittenVectorElement = true;
@@ -219,7 +222,9 @@ public:
 
   // isLegalToPruneDependencies - Is it legal to prune dependece between SUI
   // and SUJ.
-  bool isLegalToPruneDependencies(SUnit *SUI, SUnit *SUJ) {return false;}
+  bool isLegalToPruneDependencies(SUnit *SUI, SUnit *SUJ) override {
+    return false;
+  }
 
   void setIsLastBit(MachineInstr *MI, unsigned Bit) const {
     unsigned LastOp = TII->getOperandIdx(MI->getOpcode(), AMDGPU::OpName::last);
@@ -288,7 +293,7 @@ public:
     return true;
   }
 
-  MachineBasicBlock::iterator addToPacket(MachineInstr *MI) {
+  MachineBasicBlock::iterator addToPacket(MachineInstr *MI) override {
     MachineBasicBlock::iterator FirstInBundle =
         CurrentPacketMIs.empty() ? MI : CurrentPacketMIs.front();
     const DenseMap<unsigned, unsigned> &PV =
@@ -311,7 +316,7 @@ public:
       substitutePV(MI, PV);
       MachineBasicBlock::iterator It = VLIWPacketizerList::addToPacket(MI);
       if (isTransSlot) {
-        endPacket(llvm::next(It)->getParent(), llvm::next(It));
+        endPacket(std::next(It)->getParent(), std::next(It));
       }
       return It;
     }
@@ -371,20 +376,20 @@ bool R600Packetizer::runOnMachineFunction(MachineFunction &Fn) {
       // instruction stream until we find the nearest boundary.
       MachineBasicBlock::iterator I = RegionEnd;
       for(;I != MBB->begin(); --I, --RemainingCount) {
-        if (TII->isSchedulingBoundary(llvm::prior(I), MBB, Fn))
+        if (TII->isSchedulingBoundary(std::prev(I), MBB, Fn))
           break;
       }
       I = MBB->begin();
 
       // Skip empty scheduling regions.
       if (I == RegionEnd) {
-        RegionEnd = llvm::prior(RegionEnd);
+        RegionEnd = std::prev(RegionEnd);
         --RemainingCount;
         continue;
       }
       // Skip regions with one instruction.
-      if (I == llvm::prior(RegionEnd)) {
-        RegionEnd = llvm::prior(RegionEnd);
+      if (I == std::prev(RegionEnd)) {
+        RegionEnd = std::prev(RegionEnd);
         continue;
       }
 
diff --git a/contrib/llvm/lib/Target/R600/R600RegisterInfo.cpp b/contrib/llvm/lib/Target/R600/R600RegisterInfo.cpp
index f3bb88b..dc95675 100644
--- a/contrib/llvm/lib/Target/R600/R600RegisterInfo.cpp
+++ b/contrib/llvm/lib/Target/R600/R600RegisterInfo.cpp
@@ -20,15 +20,14 @@
 
 using namespace llvm;
 
-R600RegisterInfo::R600RegisterInfo(AMDGPUTargetMachine &tm)
-: AMDGPURegisterInfo(tm),
-  TM(tm)
+R600RegisterInfo::R600RegisterInfo(const AMDGPUSubtarget &st)
+: AMDGPURegisterInfo(st)
   { RCW.RegWeight = 0; RCW.WeightLimit = 0;}
 
 BitVector R600RegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   BitVector Reserved(getNumRegs());
 
-  const R600InstrInfo *TII = static_cast<const R600InstrInfo*>(TM.getInstrInfo());
+  const R600InstrInfo *TII = static_cast<const R600InstrInfo*>(ST.getInstrInfo());
 
   Reserved.set(AMDGPU::ZERO);
   Reserved.set(AMDGPU::HALF);
@@ -55,16 +54,6 @@ BitVector R600RegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   return Reserved;
 }
 
-const TargetRegisterClass *
-R600RegisterInfo::getISARegClass(const TargetRegisterClass * rc) const {
-  switch (rc->getID()) {
-  case AMDGPU::GPRF32RegClassID:
-  case AMDGPU::GPRI32RegClassID:
-    return &AMDGPU::R600_Reg32RegClass;
-  default: return rc;
-  }
-}
-
 unsigned R600RegisterInfo::getHWRegChan(unsigned reg) const {
   return this->getEncodingValue(reg) >> HW_CHAN_SHIFT;
 }
diff --git a/contrib/llvm/lib/Target/R600/R600RegisterInfo.h b/contrib/llvm/lib/Target/R600/R600RegisterInfo.h
index c74c49e..247808b 100644
--- a/contrib/llvm/lib/Target/R600/R600RegisterInfo.h
+++ b/contrib/llvm/lib/Target/R600/R600RegisterInfo.h
@@ -16,39 +16,32 @@
 #define R600REGISTERINFO_H_
 
 #include "AMDGPURegisterInfo.h"
-#include "AMDGPUTargetMachine.h"
 
 namespace llvm {
 
-class R600TargetMachine;
+class AMDGPUSubtarget;
 
 struct R600RegisterInfo : public AMDGPURegisterInfo {
-  AMDGPUTargetMachine &TM;
   RegClassWeight RCW;
 
-  R600RegisterInfo(AMDGPUTargetMachine &tm);
+  R600RegisterInfo(const AMDGPUSubtarget &st);
 
-  virtual BitVector getReservedRegs(const MachineFunction &MF) const;
-
-  /// \param RC is an AMDIL reg class.
-  ///
-  /// \returns the R600 reg class that is equivalent to \p RC.
-  virtual const TargetRegisterClass *getISARegClass(
-    const TargetRegisterClass *RC) const;
+  BitVector getReservedRegs(const MachineFunction &MF) const override;
 
   /// \brief get the HW encoding for a register's channel.
   unsigned getHWRegChan(unsigned reg) const;
 
-  virtual unsigned getHWRegIndex(unsigned Reg) const;
+  unsigned getHWRegIndex(unsigned Reg) const override;
 
   /// \brief get the register class of the specified type to use in the
   /// CFGStructurizer
-  virtual const TargetRegisterClass * getCFGStructurizerRegClass(MVT VT) const;
+  const TargetRegisterClass * getCFGStructurizerRegClass(MVT VT) const override;
 
-  virtual const RegClassWeight &getRegClassWeight(const TargetRegisterClass *RC) const;
+  const RegClassWeight &
+    getRegClassWeight(const TargetRegisterClass *RC) const override;
 
   // \returns true if \p Reg can be defined in one ALU caluse and used in another.
-  virtual bool isPhysRegLiveAcrossClauses(unsigned Reg) const;
+  bool isPhysRegLiveAcrossClauses(unsigned Reg) const;
 };
 
 } // End namespace llvm
diff --git a/contrib/llvm/lib/Target/R600/R600RegisterInfo.td b/contrib/llvm/lib/Target/R600/R600RegisterInfo.td
index 68bcd20..cc667d9 100644
--- a/contrib/llvm/lib/Target/R600/R600RegisterInfo.td
+++ b/contrib/llvm/lib/Target/R600/R600RegisterInfo.td
@@ -18,18 +18,28 @@ class R600RegWithChan <string name, bits<9> sel, string chan> :
 
 class R600Reg_128<string n, list<Register> subregs, bits<16> encoding> :
     RegisterWithSubRegs<n, subregs> {
+  field bits<2> chan_encoding = 0;
   let Namespace = "AMDGPU";
   let SubRegIndices = [sub0, sub1, sub2, sub3];
-  let HWEncoding = encoding;
+  let HWEncoding{8-0} = encoding{8-0};
+  let HWEncoding{10-9} = chan_encoding;
 }
 
 class R600Reg_64<string n, list<Register> subregs, bits<16> encoding> :
     RegisterWithSubRegs<n, subregs> {
+  field bits<2> chan_encoding = 0;
   let Namespace = "AMDGPU";
   let SubRegIndices = [sub0, sub1];
   let HWEncoding = encoding;
+  let HWEncoding{8-0} = encoding{8-0};
+  let HWEncoding{10-9} = chan_encoding;
 }
 
+class R600Reg_64Vertical<int lo, int hi, string chan> : R600Reg_64 <
+  "V"#lo#hi#"_"#chan,
+  [!cast<Register>("T"#lo#"_"#chan), !cast<Register>("T"#hi#"_"#chan)],
+  lo
+>;
 
 foreach Index = 0-127 in {
   foreach Chan = [ "X", "Y", "Z", "W" ] in {
@@ -54,6 +64,24 @@ foreach Index = 0-127 in {
                                    Index>;
 }
 
+foreach Chan = [ "X", "Y", "Z", "W"] in {
+
+  let chan_encoding = !if(!eq(Chan, "X"), 0,
+                      !if(!eq(Chan, "Y"), 1,
+                      !if(!eq(Chan, "Z"), 2,
+                      !if(!eq(Chan, "W"), 3, 0)))) in {
+    def V0123_#Chan : R600Reg_128 <"V0123_"#Chan,
+                                   [!cast<Register>("T0_"#Chan),
+                                    !cast<Register>("T1_"#Chan),
+                                    !cast<Register>("T2_"#Chan),
+                                    !cast<Register>("T3_"#Chan)],
+                                    0>;
+    def V01_#Chan : R600Reg_64Vertical<0, 1, Chan>;
+    def V23_#Chan : R600Reg_64Vertical<2, 3, Chan>;
+  }
+}
+
+
 // KCACHE_BANK0
 foreach Index = 159-128 in {
   foreach Chan = [ "X", "Y", "Z", "W" ] in {
@@ -130,8 +158,14 @@ def ALU_PARAM : R600Reg<"Param", 0>;
 
 let isAllocatable = 0 in {
 
-// XXX: Only use the X channel, until we support wider stack widths
-def R600_Addr : RegisterClass <"AMDGPU", [i32], 127, (add (sequence "Addr%u_X", 0, 127))>;
+def R600_Addr : RegisterClass <"AMDGPU", [i32], 32, (add (sequence "Addr%u_X", 0, 127))>;
+
+// We only use Addr_[YZW] for vertical vectors.
+// FIXME if we add more vertical vector registers we will need to ad more
+// registers to these classes.
+def R600_Addr_Y : RegisterClass <"AMDGPU", [i32], 32, (add Addr0_Y)>;
+def R600_Addr_Z : RegisterClass <"AMDGPU", [i32], 32, (add Addr0_Z)>;
+def R600_Addr_W : RegisterClass <"AMDGPU", [i32], 32, (add Addr0_W)>;
 
 def R600_LDS_SRC_REG : RegisterClass<"AMDGPU", [i32], 32,
   (add OQA, OQB, OQAP, OQBP, LDS_DIRECT_A, LDS_DIRECT_B)>;
@@ -206,5 +240,13 @@ def R600_Reg128 : RegisterClass<"AMDGPU", [v4f32, v4i32], 128,
   let CopyCost = -1;
 }
 
+def R600_Reg128Vertical : RegisterClass<"AMDGPU", [v4f32, v4i32], 128,
+  (add V0123_W, V0123_Z, V0123_Y, V0123_X)
+>;
+
 def R600_Reg64 : RegisterClass<"AMDGPU", [v2f32, v2i32], 64,
                                 (add (sequence "T%u_XY", 0, 63))>;
+
+def R600_Reg64Vertical : RegisterClass<"AMDGPU", [v2f32, v2i32], 64,
+                                      (add V01_X, V01_Y, V01_Z, V01_W,
+                                           V23_X, V23_Y, V23_Z, V23_W)>;
diff --git a/contrib/llvm/lib/Target/R600/R600TextureIntrinsicsReplacer.cpp b/contrib/llvm/lib/Target/R600/R600TextureIntrinsicsReplacer.cpp
index 3258894..419ec8b 100644
--- a/contrib/llvm/lib/Target/R600/R600TextureIntrinsicsReplacer.cpp
+++ b/contrib/llvm/lib/Target/R600/R600TextureIntrinsicsReplacer.cpp
@@ -18,7 +18,7 @@
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalValue.h"
 #include "llvm/IR/IRBuilder.h"
-#include "llvm/InstVisitor.h"
+#include "llvm/IR/InstVisitor.h"
 
 using namespace llvm;
 
@@ -209,7 +209,7 @@ public:
     FunctionPass(ID) {
   }
 
-  virtual bool doInitialization(Module &M) {
+  bool doInitialization(Module &M) override {
     LLVMContext &Ctx = M.getContext();
     Mod = &M;
     FloatType = Type::getFloatTy(Ctx);
@@ -245,16 +245,16 @@ public:
     return false;
   }
 
-  virtual bool runOnFunction(Function &F) {
+  bool runOnFunction(Function &F) override {
     visit(F);
     return false;
   }
 
-  virtual const char *getPassName() const {
+  const char *getPassName() const override {
     return "R600 Texture Intrinsics Replacer";
   }
 
-  void getAnalysisUsage(AnalysisUsage &AU) const {
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
   }
 
   void visitCallInst(CallInst &I) {
diff --git a/contrib/llvm/lib/Target/R600/R700Instructions.td b/contrib/llvm/lib/Target/R600/R700Instructions.td
new file mode 100644
index 0000000..9aad85d
--- /dev/null
+++ b/contrib/llvm/lib/Target/R600/R700Instructions.td
@@ -0,0 +1,21 @@
+//===-- R700Instructions.td - R700 Instruction defs  -------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// TableGen definitions for instructions which are:
+// - Available to R700 and newer VLIW4/VLIW5 GPUs
+// - Available only on R700 family GPUs.
+//
+//===----------------------------------------------------------------------===//
+
+def isR700 : Predicate<"Subtarget.getGeneration() == AMDGPUSubtarget::R700">;
+
+let Predicates = [isR700] in {
+  def SIN_r700 : SIN_Common<0x6E>;
+  def COS_r700 : COS_Common<0x6F>;
+}
diff --git a/contrib/llvm/lib/Target/R600/SIAnnotateControlFlow.cpp b/contrib/llvm/lib/Target/R600/SIAnnotateControlFlow.cpp
index 6bbdf59..91eb60b 100644
--- a/contrib/llvm/lib/Target/R600/SIAnnotateControlFlow.cpp
+++ b/contrib/llvm/lib/Target/R600/SIAnnotateControlFlow.cpp
@@ -14,8 +14,8 @@
 
 #include "AMDGPU.h"
 #include "llvm/ADT/DepthFirstIterator.h"
-#include "llvm/Analysis/Dominators.h"
 #include "llvm/IR/Constants.h"
+#include "llvm/IR/Dominators.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Pass.h"
@@ -24,6 +24,8 @@
 
 using namespace llvm;
 
+#define DEBUG_TYPE "si-annotate-control-flow"
+
 namespace {
 
 // Complex types used in this pass
@@ -63,7 +65,6 @@ class SIAnnotateControlFlow : public FunctionPass {
 
   DominatorTree *DT;
   StackVector Stack;
-  SSAUpdater PhiInserter;
 
   bool isTopOfStack(BasicBlock *BB);
 
@@ -79,7 +80,7 @@ class SIAnnotateControlFlow : public FunctionPass {
 
   void insertElse(BranchInst *Term);
 
-  void handleLoopCondition(Value *Cond);
+  Value *handleLoopCondition(Value *Cond, PHINode *Broken);
 
   void handleLoop(BranchInst *Term);
 
@@ -89,17 +90,17 @@ public:
   SIAnnotateControlFlow():
     FunctionPass(ID) { }
 
-  virtual bool doInitialization(Module &M);
+  bool doInitialization(Module &M) override;
 
-  virtual bool runOnFunction(Function &F);
+  bool runOnFunction(Function &F) override;
 
-  virtual const char *getPassName() const {
+  const char *getPassName() const override {
     return "SI annotate control flow";
   }
 
-  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
-    AU.addRequired<DominatorTree>();
-    AU.addPreserved<DominatorTree>();
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.addRequired<DominatorTreeWrapperPass>();
+    AU.addPreserved<DominatorTreeWrapperPass>();
     FunctionPass::getAnalysisUsage(AU);
   }
 
@@ -116,7 +117,7 @@ bool SIAnnotateControlFlow::doInitialization(Module &M) {
   Void = Type::getVoidTy(Context);
   Boolean = Type::getInt1Ty(Context);
   Int64 = Type::getInt64Ty(Context);
-  ReturnStruct = StructType::get(Boolean, Int64, (Type *)0);
+  ReturnStruct = StructType::get(Boolean, Int64, (Type *)nullptr);
 
   BoolTrue = ConstantInt::getTrue(Context);
   BoolFalse = ConstantInt::getFalse(Context);
@@ -124,25 +125,25 @@ bool SIAnnotateControlFlow::doInitialization(Module &M) {
   Int64Zero = ConstantInt::get(Int64, 0);
 
   If = M.getOrInsertFunction(
-    IfIntrinsic, ReturnStruct, Boolean, (Type *)0);
+    IfIntrinsic, ReturnStruct, Boolean, (Type *)nullptr);
 
   Else = M.getOrInsertFunction(
-    ElseIntrinsic, ReturnStruct, Int64, (Type *)0);
+    ElseIntrinsic, ReturnStruct, Int64, (Type *)nullptr);
 
   Break = M.getOrInsertFunction(
-    BreakIntrinsic, Int64, Int64, (Type *)0);
+    BreakIntrinsic, Int64, Int64, (Type *)nullptr);
 
   IfBreak = M.getOrInsertFunction(
-    IfBreakIntrinsic, Int64, Boolean, Int64, (Type *)0);
+    IfBreakIntrinsic, Int64, Boolean, Int64, (Type *)nullptr);
 
   ElseBreak = M.getOrInsertFunction(
-    ElseBreakIntrinsic, Int64, Int64, Int64, (Type *)0);
+    ElseBreakIntrinsic, Int64, Int64, Int64, (Type *)nullptr);
 
   Loop = M.getOrInsertFunction(
-    LoopIntrinsic, Boolean, Int64, (Type *)0);
+    LoopIntrinsic, Boolean, Int64, (Type *)nullptr);
 
   EndCf = M.getOrInsertFunction(
-    EndCfIntrinsic, Void, Int64, (Type *)0);
+    EndCfIntrinsic, Void, Int64, (Type *)nullptr);
 
   return false;
 }
@@ -175,7 +176,7 @@ bool SIAnnotateControlFlow::isElse(PHINode *Phi) {
     } else {
       if (Phi->getIncomingValue(i) != BoolFalse)
         return false;
- 
+
     }
   }
   return true;
@@ -202,20 +203,26 @@ void SIAnnotateControlFlow::insertElse(BranchInst *Term) {
 }
 
 /// \brief Recursively handle the condition leading to a loop
-void SIAnnotateControlFlow::handleLoopCondition(Value *Cond) {
+Value *SIAnnotateControlFlow::handleLoopCondition(Value *Cond, PHINode *Broken) {
   if (PHINode *Phi = dyn_cast<PHINode>(Cond)) {
+    BasicBlock *Parent = Phi->getParent();
+    PHINode *NewPhi = PHINode::Create(Int64, 0, "", &Parent->front());
+    Value *Ret = NewPhi;
 
-    // Handle all non constant incoming values first
+    // Handle all non-constant incoming values first
     for (unsigned i = 0, e = Phi->getNumIncomingValues(); i != e; ++i) {
       Value *Incoming = Phi->getIncomingValue(i);
-      if (isa<ConstantInt>(Incoming))
+      BasicBlock *From = Phi->getIncomingBlock(i);
+      if (isa<ConstantInt>(Incoming)) {
+        NewPhi->addIncoming(Broken, From);
         continue;
+      }
 
       Phi->setIncomingValue(i, BoolFalse);
-      handleLoopCondition(Incoming);
+      Value *PhiArg = handleLoopCondition(Incoming, Broken);
+      NewPhi->addIncoming(PhiArg, From);
     }
 
-    BasicBlock *Parent = Phi->getParent();
     BasicBlock *IDom = DT->getNode(Parent)->getIDom()->getBlock();
 
     for (unsigned i = 0, e = Phi->getNumIncomingValues(); i != e; ++i) {
@@ -228,33 +235,28 @@ void SIAnnotateControlFlow::handleLoopCondition(Value *Cond) {
       if (From == IDom) {
         CallInst *OldEnd = dyn_cast<CallInst>(Parent->getFirstInsertionPt());
         if (OldEnd && OldEnd->getCalledFunction() == EndCf) {
-          Value *Args[] = {
-            OldEnd->getArgOperand(0),
-            PhiInserter.GetValueAtEndOfBlock(Parent)
-          };
-          Value *Ret = CallInst::Create(ElseBreak, Args, "", OldEnd);
-          PhiInserter.AddAvailableValue(Parent, Ret);
+          Value *Args[] = { OldEnd->getArgOperand(0), NewPhi };
+          Ret = CallInst::Create(ElseBreak, Args, "", OldEnd);
           continue;
         }
       }
-
       TerminatorInst *Insert = From->getTerminator();
-      Value *Arg = PhiInserter.GetValueAtEndOfBlock(From);
-      Value *Ret = CallInst::Create(Break, Arg, "", Insert);
-      PhiInserter.AddAvailableValue(From, Ret);
+      Value *PhiArg = CallInst::Create(Break, Broken, "", Insert);
+      NewPhi->setIncomingValue(i, PhiArg);
     }
     eraseIfUnused(Phi);
+    return Ret;
 
   } else if (Instruction *Inst = dyn_cast<Instruction>(Cond)) {
     BasicBlock *Parent = Inst->getParent();
     TerminatorInst *Insert = Parent->getTerminator();
-    Value *Args[] = { Cond, PhiInserter.GetValueAtEndOfBlock(Parent) };
-    Value *Ret = CallInst::Create(IfBreak, Args, "", Insert);
-    PhiInserter.AddAvailableValue(Parent, Ret);
+    Value *Args[] = { Cond, Broken };
+    return CallInst::Create(IfBreak, Args, "", Insert);
 
   } else {
-    assert(0 && "Unhandled loop condition!");
+    llvm_unreachable("Unhandled loop condition!");
   }
+  return 0;
 }
 
 /// \brief Handle a back edge (loop)
@@ -262,15 +264,11 @@ void SIAnnotateControlFlow::handleLoop(BranchInst *Term) {
   BasicBlock *Target = Term->getSuccessor(1);
   PHINode *Broken = PHINode::Create(Int64, 0, "", &Target->front());
 
-  PhiInserter.Initialize(Int64, "");
-  PhiInserter.AddAvailableValue(Target, Broken);
-
   Value *Cond = Term->getCondition();
   Term->setCondition(BoolTrue);
-  handleLoopCondition(Cond);
+  Value *Arg = handleLoopCondition(Cond, Broken);
 
   BasicBlock *BB = Term->getParent();
-  Value *Arg = PhiInserter.GetValueAtEndOfBlock(BB);
   for (pred_iterator PI = pred_begin(Target), PE = pred_end(Target);
        PI != PE; ++PI) {
 
@@ -289,7 +287,7 @@ void SIAnnotateControlFlow::closeControlFlow(BasicBlock *BB) {
 /// \brief Annotate the control flow with intrinsics so the backend can
 /// recognize if/then/else and loops.
 bool SIAnnotateControlFlow::runOnFunction(Function &F) {
-  DT = &getAnalysis<DominatorTree>();
+  DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
 
   for (df_iterator<BasicBlock *> I = df_begin(&F.getEntryBlock()),
        E = df_end(&F.getEntryBlock()); I != E; ++I) {
diff --git a/contrib/llvm/lib/Target/R600/SIDefines.h b/contrib/llvm/lib/Target/R600/SIDefines.h
index 2cbce28..b7e7a2d 100644
--- a/contrib/llvm/lib/Target/R600/SIDefines.h
+++ b/contrib/llvm/lib/Target/R600/SIDefines.h
@@ -32,7 +32,61 @@ enum {
 #define   S_00B028_VGPRS(x)                                           (((x) & 0x3F) << 0)
 #define   S_00B028_SGPRS(x)                                           (((x) & 0x0F) << 6)
 #define R_00B84C_COMPUTE_PGM_RSRC2                                      0x00B84C
+#define   S_00B02C_SCRATCH_EN(x)                                      (((x) & 0x1) << 0)
 #define   S_00B84C_LDS_SIZE(x)                                        (((x) & 0x1FF) << 15)
 #define R_0286CC_SPI_PS_INPUT_ENA                                       0x0286CC
 
+
+#define R_00B848_COMPUTE_PGM_RSRC1                                      0x00B848
+#define   S_00B848_VGPRS(x)                                           (((x) & 0x3F) << 0)
+#define   G_00B848_VGPRS(x)                                           (((x) >> 0) & 0x3F)
+#define   C_00B848_VGPRS                                              0xFFFFFFC0
+#define   S_00B848_SGPRS(x)                                           (((x) & 0x0F) << 6)
+#define   G_00B848_SGPRS(x)                                           (((x) >> 6) & 0x0F)
+#define   C_00B848_SGPRS                                              0xFFFFFC3F
+#define   S_00B848_PRIORITY(x)                                        (((x) & 0x03) << 10)
+#define   G_00B848_PRIORITY(x)                                        (((x) >> 10) & 0x03)
+#define   C_00B848_PRIORITY                                           0xFFFFF3FF
+#define   S_00B848_FLOAT_MODE(x)                                      (((x) & 0xFF) << 12)
+#define   G_00B848_FLOAT_MODE(x)                                      (((x) >> 12) & 0xFF)
+#define   C_00B848_FLOAT_MODE                                         0xFFF00FFF
+#define   S_00B848_PRIV(x)                                            (((x) & 0x1) << 20)
+#define   G_00B848_PRIV(x)                                            (((x) >> 20) & 0x1)
+#define   C_00B848_PRIV                                               0xFFEFFFFF
+#define   S_00B848_DX10_CLAMP(x)                                      (((x) & 0x1) << 21)
+#define   G_00B848_DX10_CLAMP(x)                                      (((x) >> 21) & 0x1)
+#define   C_00B848_DX10_CLAMP                                         0xFFDFFFFF
+#define   S_00B848_DEBUG_MODE(x)                                      (((x) & 0x1) << 22)
+#define   G_00B848_DEBUG_MODE(x)                                      (((x) >> 22) & 0x1)
+#define   C_00B848_DEBUG_MODE                                         0xFFBFFFFF
+#define   S_00B848_IEEE_MODE(x)                                       (((x) & 0x1) << 23)
+#define   G_00B848_IEEE_MODE(x)                                       (((x) >> 23) & 0x1)
+#define   C_00B848_IEEE_MODE                                          0xFF7FFFFF
+
+
+// Helpers for setting FLOAT_MODE
+#define FP_ROUND_ROUND_TO_NEAREST 0
+#define FP_ROUND_ROUND_TO_INF 1
+#define FP_ROUND_ROUND_TO_NEGINF 2
+#define FP_ROUND_ROUND_TO_ZERO 3
+
+// Bits 3:0 control rounding mode. 1:0 control single precision, 3:2 double
+// precision.
+#define FP_ROUND_MODE_SP(x) ((x) & 0x3)
+#define FP_ROUND_MODE_DP(x) (((x) & 0x3) << 2)
+
+#define FP_DENORM_FLUSH_IN_FLUSH_OUT 0
+#define FP_DENORM_FLUSH_OUT 1
+#define FP_DENORM_FLUSH_IN 2
+#define FP_DENORM_FLUSH_NONE 3
+
+
+// Bits 7:4 control denormal handling. 5:4 control single precision, 6:7 double
+// precision.
+#define FP_DENORM_MODE_SP(x) (((x) & 0x3) << 4)
+#define FP_DENORM_MODE_DP(x) (((x) & 0x3) << 6)
+
+#define R_00B860_COMPUTE_TMPRING_SIZE                                   0x00B860
+#define   S_00B860_WAVESIZE(x)                                        (((x) & 0x1FFF) << 12)
+
 #endif // SIDEFINES_H_
diff --git a/contrib/llvm/lib/Target/R600/SIFixSGPRCopies.cpp b/contrib/llvm/lib/Target/R600/SIFixSGPRCopies.cpp
index f0065ea..5f71453 100644
--- a/contrib/llvm/lib/Target/R600/SIFixSGPRCopies.cpp
+++ b/contrib/llvm/lib/Target/R600/SIFixSGPRCopies.cpp
@@ -65,7 +65,6 @@
 /// ultimately led to the creation of an illegal COPY.
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "sgpr-copies"
 #include "AMDGPU.h"
 #include "SIInstrInfo.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
@@ -77,6 +76,8 @@
 
 using namespace llvm;
 
+#define DEBUG_TYPE "sgpr-copies"
+
 namespace {
 
 class SIFixSGPRCopies : public MachineFunctionPass {
@@ -97,9 +98,9 @@ private:
 public:
   SIFixSGPRCopies(TargetMachine &tm) : MachineFunctionPass(ID) { }
 
-  virtual bool runOnMachineFunction(MachineFunction &MF);
+  bool runOnMachineFunction(MachineFunction &MF) override;
 
-  const char *getPassName() const {
+  const char *getPassName() const override {
     return "SI Fix SGPR copies";
   }
 
@@ -141,8 +142,8 @@ const TargetRegisterClass *SIFixSGPRCopies::inferRegClassFromUses(
 
   const TargetRegisterClass *RC = MRI.getRegClass(Reg);
   RC = TRI->getSubRegClass(RC, SubReg);
-  for (MachineRegisterInfo::use_iterator I = MRI.use_begin(Reg),
-                                         E = MRI.use_end(); I != E; ++I) {
+  for (MachineRegisterInfo::use_instr_iterator
+       I = MRI.use_instr_begin(Reg), E = MRI.use_instr_end(); I != E; ++I) {
     switch (I->getOpcode()) {
     case AMDGPU::COPY:
       RC = TRI->getCommonSubClass(RC, inferRegClassFromUses(TRI, MRI,
@@ -184,7 +185,8 @@ bool SIFixSGPRCopies::isVGPRToSGPRCopy(const MachineInstr &Copy,
   const TargetRegisterClass *SrcRC;
 
   if (!TargetRegisterInfo::isVirtualRegister(SrcReg) ||
-      DstRC == &AMDGPU::M0RegRegClass)
+      DstRC == &AMDGPU::M0RegRegClass ||
+      MRI.getRegClass(SrcReg) == &AMDGPU::VReg_1RegClass)
     return false;
 
   SrcRC = TRI->getSubRegClass(MRI.getRegClass(SrcReg), SrcSubReg);
@@ -256,6 +258,19 @@ bool SIFixSGPRCopies::runOnMachineFunction(MachineFunction &MF) {
         TII->moveToVALU(MI);
         break;
       }
+      case AMDGPU::INSERT_SUBREG: {
+        const TargetRegisterClass *DstRC, *Src0RC, *Src1RC;
+        DstRC = MRI.getRegClass(MI.getOperand(0).getReg());
+        Src0RC = MRI.getRegClass(MI.getOperand(1).getReg());
+        Src1RC = MRI.getRegClass(MI.getOperand(2).getReg());
+        if (TRI->isSGPRClass(DstRC) &&
+            (TRI->hasVGPRs(Src0RC) || TRI->hasVGPRs(Src1RC))) {
+          DEBUG(dbgs() << " Fixing INSERT_SUBREG:\n");
+          DEBUG(MI.print(dbgs()));
+          TII->moveToVALU(MI);
+        }
+        break;
+      }
       }
     }
   }
diff --git a/contrib/llvm/lib/Target/R600/SIFixSGPRLiveRanges.cpp b/contrib/llvm/lib/Target/R600/SIFixSGPRLiveRanges.cpp
new file mode 100644
index 0000000..7d116ee
--- /dev/null
+++ b/contrib/llvm/lib/Target/R600/SIFixSGPRLiveRanges.cpp
@@ -0,0 +1,110 @@
+//===-- SIFixSGPRLiveRanges.cpp - Fix SGPR live ranges ----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// SALU instructions ignore control flow, so we need to modify the live ranges
+/// of the registers they define.
+///
+/// The strategy is to view the entire program as if it were a single basic
+/// block and calculate the intervals accordingly.  We implement this
+/// by walking this list of segments for each LiveRange and setting the
+/// end of each segment equal to the start of the segment that immediately
+/// follows it.
+
+#include "AMDGPU.h"
+#include "SIRegisterInfo.h"
+#include "llvm/CodeGen/LiveIntervalAnalysis.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Target/TargetMachine.h"
+
+using namespace llvm;
+
+#define DEBUG_TYPE "si-fix-sgpr-live-ranges"
+
+namespace {
+
+class SIFixSGPRLiveRanges : public MachineFunctionPass {
+public:
+  static char ID;
+
+public:
+  SIFixSGPRLiveRanges() : MachineFunctionPass(ID) {
+    initializeSIFixSGPRLiveRangesPass(*PassRegistry::getPassRegistry());
+  }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF) override;
+
+  virtual const char *getPassName() const override {
+    return "SI Fix SGPR live ranges";
+  }
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.addRequired<LiveIntervals>();
+    AU.addPreserved<LiveIntervals>();
+    AU.addPreserved<SlotIndexes>();
+    AU.setPreservesCFG();
+    MachineFunctionPass::getAnalysisUsage(AU);
+  }
+};
+
+} // End anonymous namespace.
+
+INITIALIZE_PASS_BEGIN(SIFixSGPRLiveRanges, DEBUG_TYPE,
+                      "SI Fix SGPR Live Ranges", false, false)
+INITIALIZE_PASS_DEPENDENCY(LiveIntervals)
+INITIALIZE_PASS_END(SIFixSGPRLiveRanges, DEBUG_TYPE,
+                    "SI Fix SGPR Live Ranges", false, false)
+
+char SIFixSGPRLiveRanges::ID = 0;
+
+char &llvm::SIFixSGPRLiveRangesID = SIFixSGPRLiveRanges::ID;
+
+FunctionPass *llvm::createSIFixSGPRLiveRangesPass() {
+  return new SIFixSGPRLiveRanges();
+}
+
+bool SIFixSGPRLiveRanges::runOnMachineFunction(MachineFunction &MF) {
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+  const SIRegisterInfo *TRI = static_cast<const SIRegisterInfo *>(
+      MF.getTarget().getRegisterInfo());
+  LiveIntervals *LIS = &getAnalysis<LiveIntervals>();
+
+  for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
+                                                  BI != BE; ++BI) {
+
+    MachineBasicBlock &MBB = *BI;
+    for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
+                                                      I != E; ++I) {
+      MachineInstr &MI = *I;
+      MachineOperand *ExecUse = MI.findRegisterUseOperand(AMDGPU::EXEC);
+      if (ExecUse)
+        continue;
+
+      for (const MachineOperand &Def : MI.operands()) {
+        if (!Def.isReg() || !Def.isDef() ||!TargetRegisterInfo::isVirtualRegister(Def.getReg()))
+          continue;
+
+        const TargetRegisterClass *RC = MRI.getRegClass(Def.getReg());
+
+        if (!TRI->isSGPRClass(RC))
+          continue;
+        LiveInterval &LI = LIS->getInterval(Def.getReg());
+        for (unsigned i = 0, e = LI.size() - 1; i != e; ++i) {
+          LiveRange::Segment &Seg = LI.segments[i];
+          LiveRange::Segment &Next = LI.segments[i + 1];
+          Seg.end = Next.start;
+        }
+      }
+    }
+  }
+
+  return false;
+}
diff --git a/contrib/llvm/lib/Target/R600/SIISelLowering.cpp b/contrib/llvm/lib/Target/R600/SIISelLowering.cpp
index d5d2b68..5a148a2 100644
--- a/contrib/llvm/lib/Target/R600/SIISelLowering.cpp
+++ b/contrib/llvm/lib/Target/R600/SIISelLowering.cpp
@@ -12,9 +12,16 @@
 //
 //===----------------------------------------------------------------------===//
 
+#ifdef _MSC_VER
+// Provide M_PI.
+#define _USE_MATH_DEFINES
+#include <cmath>
+#endif
+
 #include "SIISelLowering.h"
 #include "AMDGPU.h"
-#include "AMDILIntrinsicInfo.h"
+#include "AMDGPUIntrinsicInfo.h"
+#include "AMDGPUSubtarget.h"
 #include "SIInstrInfo.h"
 #include "SIMachineFunctionInfo.h"
 #include "SIRegisterInfo.h"
@@ -23,30 +30,27 @@
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/IR/Function.h"
-
-const uint64_t RSRC_DATA_FORMAT = 0xf00000000000LL;
+#include "llvm/ADT/SmallString.h"
 
 using namespace llvm;
 
 SITargetLowering::SITargetLowering(TargetMachine &TM) :
     AMDGPUTargetLowering(TM) {
-
-  addRegisterClass(MVT::i1, &AMDGPU::SReg_64RegClass);
-  addRegisterClass(MVT::i64, &AMDGPU::VSrc_64RegClass);
+  addRegisterClass(MVT::i1, &AMDGPU::VReg_1RegClass);
+  addRegisterClass(MVT::i64, &AMDGPU::SReg_64RegClass);
 
   addRegisterClass(MVT::v32i8, &AMDGPU::SReg_256RegClass);
   addRegisterClass(MVT::v64i8, &AMDGPU::SReg_512RegClass);
 
-  addRegisterClass(MVT::i32, &AMDGPU::VSrc_32RegClass);
-  addRegisterClass(MVT::f32, &AMDGPU::VSrc_32RegClass);
+  addRegisterClass(MVT::i32, &AMDGPU::SReg_32RegClass);
+  addRegisterClass(MVT::f32, &AMDGPU::VReg_32RegClass);
 
-  addRegisterClass(MVT::f64, &AMDGPU::VSrc_64RegClass);
-  addRegisterClass(MVT::v2i32, &AMDGPU::VSrc_64RegClass);
-  addRegisterClass(MVT::v2f32, &AMDGPU::VSrc_64RegClass);
+  addRegisterClass(MVT::f64, &AMDGPU::VReg_64RegClass);
+  addRegisterClass(MVT::v2i32, &AMDGPU::SReg_64RegClass);
+  addRegisterClass(MVT::v2f32, &AMDGPU::VReg_64RegClass);
 
-  addRegisterClass(MVT::v4i32, &AMDGPU::VReg_128RegClass);
+  addRegisterClass(MVT::v4i32, &AMDGPU::SReg_128RegClass);
   addRegisterClass(MVT::v4f32, &AMDGPU::VReg_128RegClass);
-  addRegisterClass(MVT::i128, &AMDGPU::SReg_128RegClass);
 
   addRegisterClass(MVT::v8i32, &AMDGPU::VReg_256RegClass);
   addRegisterClass(MVT::v8f32, &AMDGPU::VReg_256RegClass);
@@ -76,15 +80,16 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v16i32, Expand);
   setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v16f32, Expand);
 
-  setOperationAction(ISD::ADD, MVT::i64, Legal);
   setOperationAction(ISD::ADD, MVT::i32, Legal);
   setOperationAction(ISD::ADDC, MVT::i32, Legal);
   setOperationAction(ISD::ADDE, MVT::i32, Legal);
+  setOperationAction(ISD::SUBC, MVT::i32, Legal);
+  setOperationAction(ISD::SUBE, MVT::i32, Legal);
 
-  setOperationAction(ISD::BITCAST, MVT::i128, Legal);
+  setOperationAction(ISD::FSIN, MVT::f32, Custom);
+  setOperationAction(ISD::FCOS, MVT::f32, Custom);
 
   // We need to custom lower vector stores from local memory
-  setOperationAction(ISD::LOAD, MVT::v2i32, Custom);
   setOperationAction(ISD::LOAD, MVT::v4i32, Custom);
   setOperationAction(ISD::LOAD, MVT::v8i32, Custom);
   setOperationAction(ISD::LOAD, MVT::v16i32, Custom);
@@ -92,30 +97,40 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::STORE, MVT::v8i32, Custom);
   setOperationAction(ISD::STORE, MVT::v16i32, Custom);
 
-  // We need to custom lower loads/stores from private memory
-  setOperationAction(ISD::LOAD, MVT::i32, Custom);
-  setOperationAction(ISD::LOAD, MVT::i64, Custom);
-  setOperationAction(ISD::LOAD, MVT::v2i32, Custom);
-  setOperationAction(ISD::LOAD, MVT::v4i32, Custom);
-
+  setOperationAction(ISD::STORE, MVT::i1, Custom);
   setOperationAction(ISD::STORE, MVT::i32, Custom);
-  setOperationAction(ISD::STORE, MVT::i64, Custom);
-  setOperationAction(ISD::STORE, MVT::i128, Custom);
   setOperationAction(ISD::STORE, MVT::v2i32, Custom);
   setOperationAction(ISD::STORE, MVT::v4i32, Custom);
 
+  setOperationAction(ISD::SELECT, MVT::f32, Promote);
+  AddPromotedToType(ISD::SELECT, MVT::f32, MVT::i32);
+  setOperationAction(ISD::SELECT, MVT::i64, Custom);
+  setOperationAction(ISD::SELECT, MVT::f64, Promote);
+  AddPromotedToType(ISD::SELECT, MVT::f64, MVT::i64);
 
-  setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
-  setOperationAction(ISD::SELECT_CC, MVT::i32, Custom);
-
-  setOperationAction(ISD::SELECT_CC, MVT::Other, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::i32, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::i64, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
 
   setOperationAction(ISD::SETCC, MVT::v2i1, Expand);
   setOperationAction(ISD::SETCC, MVT::v4i1, Expand);
 
-  setOperationAction(ISD::ANY_EXTEND, MVT::i64, Custom);
-  setOperationAction(ISD::SIGN_EXTEND, MVT::i64, Custom);
-  setOperationAction(ISD::ZERO_EXTEND, MVT::i64, Custom);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Legal);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i1, Custom);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i1, Custom);
+
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8, Legal);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i8, Custom);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i8, Custom);
+
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16, Legal);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i16, Custom);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i16, Custom);
+
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i32, Custom);
+
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::Other, Custom);
 
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::f32, Custom);
@@ -123,26 +138,101 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::v4f32, Custom);
 
   setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
+  setOperationAction(ISD::BRCOND, MVT::Other, Custom);
 
+  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
+  setLoadExtAction(ISD::SEXTLOAD, MVT::i8, Custom);
+  setLoadExtAction(ISD::SEXTLOAD, MVT::i16, Custom);
   setLoadExtAction(ISD::SEXTLOAD, MVT::i32, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::i32, Expand);
   setLoadExtAction(ISD::SEXTLOAD, MVT::v8i16, Expand);
   setLoadExtAction(ISD::SEXTLOAD, MVT::v16i16, Expand);
 
+  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
+  setLoadExtAction(ISD::ZEXTLOAD, MVT::i8, Custom);
+  setLoadExtAction(ISD::ZEXTLOAD, MVT::i16, Custom);
+  setLoadExtAction(ISD::ZEXTLOAD, MVT::i32, Expand);
+
+  setLoadExtAction(ISD::EXTLOAD, MVT::i1, Promote);
+  setLoadExtAction(ISD::EXTLOAD, MVT::i8, Custom);
+  setLoadExtAction(ISD::EXTLOAD, MVT::i16, Custom);
+  setLoadExtAction(ISD::EXTLOAD, MVT::i32, Expand);
   setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
+
+  setTruncStoreAction(MVT::i32, MVT::i8, Custom);
+  setTruncStoreAction(MVT::i32, MVT::i16, Custom);
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
   setTruncStoreAction(MVT::i64, MVT::i32, Expand);
-  setTruncStoreAction(MVT::i128, MVT::i64, Expand);
   setTruncStoreAction(MVT::v8i32, MVT::v8i16, Expand);
   setTruncStoreAction(MVT::v16i32, MVT::v16i16, Expand);
 
+  setOperationAction(ISD::LOAD, MVT::i1, Custom);
+
+  setOperationAction(ISD::FP_TO_SINT, MVT::i64, Expand);
+  setOperationAction(ISD::FP_TO_UINT, MVT::i64, Expand);
+
   setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
-  setOperationAction(ISD::FrameIndex, MVT::i64, Custom);
+  setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
+  setOperationAction(ISD::FrameIndex, MVT::i32, Custom);
 
-  setTargetDAGCombine(ISD::SELECT_CC);
+  // These should use UDIVREM, so set them to expand
+  setOperationAction(ISD::UDIV, MVT::i64, Expand);
+  setOperationAction(ISD::UREM, MVT::i64, Expand);
+
+  // We only support LOAD/STORE and vector manipulation ops for vectors
+  // with > 4 elements.
+  MVT VecTypes[] = {
+    MVT::v8i32, MVT::v8f32, MVT::v16i32, MVT::v16f32
+  };
+
+  setOperationAction(ISD::SELECT_CC, MVT::i1, Expand);
+  setOperationAction(ISD::SELECT, MVT::i1, Promote);
+
+  for (MVT VT : VecTypes) {
+    for (unsigned Op = 0; Op < ISD::BUILTIN_OP_END; ++Op) {
+      switch(Op) {
+      case ISD::LOAD:
+      case ISD::STORE:
+      case ISD::BUILD_VECTOR:
+      case ISD::BITCAST:
+      case ISD::EXTRACT_VECTOR_ELT:
+      case ISD::INSERT_VECTOR_ELT:
+      case ISD::CONCAT_VECTORS:
+      case ISD::INSERT_SUBVECTOR:
+      case ISD::EXTRACT_SUBVECTOR:
+        break;
+      default:
+        setOperationAction(Op, VT, Expand);
+        break;
+      }
+    }
+  }
+
+  for (int I = MVT::v1f64; I <= MVT::v8f64; ++I) {
+    MVT::SimpleValueType VT = static_cast<MVT::SimpleValueType>(I);
+    setOperationAction(ISD::FTRUNC, VT, Expand);
+    setOperationAction(ISD::FCEIL, VT, Expand);
+    setOperationAction(ISD::FFLOOR, VT, Expand);
+  }
+
+  if (Subtarget->getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS) {
+    setOperationAction(ISD::FTRUNC, MVT::f64, Legal);
+    setOperationAction(ISD::FCEIL, MVT::f64, Legal);
+    setOperationAction(ISD::FFLOOR, MVT::f64, Legal);
+    setOperationAction(ISD::FRINT, MVT::f64, Legal);
+  }
+
+  // FIXME: These should be removed and handled the same was as f32 fneg. Source
+  // modifiers also work for the double instructions.
+  setOperationAction(ISD::FNEG, MVT::f64, Expand);
+  setOperationAction(ISD::FABS, MVT::f64, Expand);
 
+  setOperationAction(ISD::FDIV, MVT::f32, Custom);
+
+  setTargetDAGCombine(ISD::SELECT_CC);
   setTargetDAGCombine(ISD::SETCC);
 
+  setTargetDAGCombine(ISD::UINT_TO_FP);
+
   setSchedulingPreference(Sched::RegPressure);
 }
 
@@ -151,21 +241,55 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
 //===----------------------------------------------------------------------===//
 
 bool SITargetLowering::allowsUnalignedMemoryAccesses(EVT  VT,
+                                                     unsigned AddrSpace,
                                                      bool *IsFast) const {
+  if (IsFast)
+    *IsFast = false;
+
   // XXX: This depends on the address space and also we may want to revist
   // the alignment values we specify in the DataLayout.
+
+  // TODO: I think v3i32 should allow unaligned accesses on CI with DS_READ_B96,
+  // which isn't a simple VT.
   if (!VT.isSimple() || VT == MVT::Other)
     return false;
+
+  // XXX - CI changes say "Support for unaligned memory accesses" but I don't
+  // see what for specifically. The wording everywhere else seems to be the
+  // same.
+
+  // XXX - The only mention I see of this in the ISA manual is for LDS direct
+  // reads the "byte address and must be dword aligned". Is it also true for the
+  // normal loads and stores?
+  if (AddrSpace == AMDGPUAS::LOCAL_ADDRESS)
+    return false;
+
+  // 8.1.6 - For Dword or larger reads or writes, the two LSBs of the
+  // byte-address are ignored, thus forcing Dword alignment.
+  // This applies to private, global, and constant memory.
+  if (IsFast)
+    *IsFast = true;
   return VT.bitsGT(MVT::i32);
 }
 
-bool SITargetLowering::shouldSplitVectorElementType(EVT VT) const {
-  return VT.bitsLE(MVT::i16);
+TargetLoweringBase::LegalizeTypeAction
+SITargetLowering::getPreferredVectorAction(EVT VT) const {
+  if (VT.getVectorNumElements() != 1 && VT.getScalarType().bitsLE(MVT::i16))
+    return TypeSplitVector;
+
+  return TargetLoweringBase::getPreferredVectorAction(VT);
+}
+
+bool SITargetLowering::shouldConvertConstantLoadToIntImm(const APInt &Imm,
+                                                         Type *Ty) const {
+  const SIInstrInfo *TII =
+    static_cast<const SIInstrInfo*>(getTargetMachine().getInstrInfo());
+  return TII->isInlineConstant(Imm);
 }
 
 SDValue SITargetLowering::LowerParameter(SelectionDAG &DAG, EVT VT, EVT MemVT,
                                          SDLoc DL, SDValue Chain,
-                                         unsigned Offset) const {
+                                         unsigned Offset, bool Signed) const {
   MachineRegisterInfo &MRI = DAG.getMachineFunction().getRegInfo();
   PointerType *PtrTy = PointerType::get(VT.getTypeForEVT(*DAG.getContext()),
                                             AMDGPUAS::CONSTANT_ADDRESS);
@@ -173,7 +297,7 @@ SDValue SITargetLowering::LowerParameter(SelectionDAG &DAG, EVT VT, EVT MemVT,
                            MRI.getLiveInVirtReg(AMDGPU::SGPR0_SGPR1), MVT::i64);
   SDValue Ptr = DAG.getNode(ISD::ADD, DL, MVT::i64, BasePtr,
                                              DAG.getConstant(Offset, MVT::i64));
-  return DAG.getExtLoad(ISD::SEXTLOAD, DL, VT, Chain, Ptr,
+  return DAG.getExtLoad(Signed ? ISD::SEXTLOAD : ISD::ZEXTLOAD, DL, VT, Chain, Ptr,
                             MachinePointerInfo(UndefValue::get(PtrTy)), MemVT,
                             false, false, MemVT.getSizeInBits() >> 3);
 
@@ -202,7 +326,7 @@ SDValue SITargetLowering::LowerFormalArguments(
     const ISD::InputArg &Arg = Ins[i];
 
     // First check if it's a PS input addr
-    if (Info->ShaderType == ShaderType::PIXEL && !Arg.Flags.isInReg() &&
+    if (Info->getShaderType() == ShaderType::PIXEL && !Arg.Flags.isInReg() &&
         !Arg.Flags.isByVal()) {
 
       assert((PSInputNum <= 15) && "Too many PS inputs!");
@@ -218,7 +342,7 @@ SDValue SITargetLowering::LowerFormalArguments(
     }
 
     // Second split vertices into their elements
-    if (Info->ShaderType != ShaderType::COMPUTE && Arg.VT.isVector()) {
+    if (Info->getShaderType() != ShaderType::COMPUTE && Arg.VT.isVector()) {
       ISD::InputArg NewArg = Arg;
       NewArg.Flags.setSplit();
       NewArg.VT = Arg.VT.getVectorElementType();
@@ -234,7 +358,7 @@ SDValue SITargetLowering::LowerFormalArguments(
         NewArg.PartOffset += NewArg.VT.getStoreSize();
       }
 
-    } else if (Info->ShaderType != ShaderType::COMPUTE) {
+    } else if (Info->getShaderType() != ShaderType::COMPUTE) {
       Splits.push_back(Arg);
     }
   }
@@ -244,20 +368,26 @@ SDValue SITargetLowering::LowerFormalArguments(
                  getTargetMachine(), ArgLocs, *DAG.getContext());
 
   // At least one interpolation mode must be enabled or else the GPU will hang.
-  if (Info->ShaderType == ShaderType::PIXEL && (Info->PSInputAddr & 0x7F) == 0) {
+  if (Info->getShaderType() == ShaderType::PIXEL &&
+      (Info->PSInputAddr & 0x7F) == 0) {
     Info->PSInputAddr |= 1;
     CCInfo.AllocateReg(AMDGPU::VGPR0);
     CCInfo.AllocateReg(AMDGPU::VGPR1);
   }
 
   // The pointer to the list of arguments is stored in SGPR0, SGPR1
-  if (Info->ShaderType == ShaderType::COMPUTE) {
+	// The pointer to the scratch buffer is stored in SGPR2, SGPR3
+  if (Info->getShaderType() == ShaderType::COMPUTE) {
+    Info->NumUserSGPRs = 4;
     CCInfo.AllocateReg(AMDGPU::SGPR0);
     CCInfo.AllocateReg(AMDGPU::SGPR1);
+    CCInfo.AllocateReg(AMDGPU::SGPR2);
+    CCInfo.AllocateReg(AMDGPU::SGPR3);
     MF.addLiveIn(AMDGPU::SGPR0_SGPR1, &AMDGPU::SReg_64RegClass);
+    MF.addLiveIn(AMDGPU::SGPR2_SGPR3, &AMDGPU::SReg_64RegClass);
   }
 
-  if (Info->ShaderType == ShaderType::COMPUTE) {
+  if (Info->getShaderType() == ShaderType::COMPUTE) {
     getOriginalFunctionArgs(DAG, DAG.getMachineFunction().getFunction(), Ins,
                             Splits);
   }
@@ -281,7 +411,8 @@ SDValue SITargetLowering::LowerFormalArguments(
       // The first 36 bytes of the input buffer contains information about
       // thread group and global sizes.
       SDValue Arg = LowerParameter(DAG, VT, MemVT,  DL, DAG.getRoot(),
-                                   36 + VA.getLocMemOffset());
+                                   36 + VA.getLocMemOffset(),
+                                   Ins[i].Flags.isSExt());
       InVals.push_back(Arg);
       continue;
     }
@@ -322,8 +453,7 @@ SDValue SITargetLowering::LowerFormalArguments(
       for (unsigned j = 0; j != NumElements; ++j)
         Regs.push_back(DAG.getUNDEF(VT));
 
-      InVals.push_back(DAG.getNode(ISD::BUILD_VECTOR, DL, Arg.VT,
-                                   Regs.data(), Regs.size()));
+      InVals.push_back(DAG.getNode(ISD::BUILD_VECTOR, DL, Arg.VT, Regs));
       continue;
     }
 
@@ -336,26 +466,26 @@ MachineBasicBlock * SITargetLowering::EmitInstrWithCustomInserter(
     MachineInstr * MI, MachineBasicBlock * BB) const {
 
   MachineBasicBlock::iterator I = *MI;
+  const SIInstrInfo *TII =
+    static_cast<const SIInstrInfo*>(getTargetMachine().getInstrInfo());
+  MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
 
   switch (MI->getOpcode()) {
   default:
     return AMDGPUTargetLowering::EmitInstrWithCustomInserter(MI, BB);
   case AMDGPU::BRANCH: return BB;
   case AMDGPU::SI_ADDR64_RSRC: {
-    const SIInstrInfo *TII =
-      static_cast<const SIInstrInfo*>(getTargetMachine().getInstrInfo());
-    MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
     unsigned SuperReg = MI->getOperand(0).getReg();
-    unsigned SubRegLo = MRI.createVirtualRegister(&AMDGPU::SReg_64RegClass);
-    unsigned SubRegHi = MRI.createVirtualRegister(&AMDGPU::SReg_64RegClass);
-    unsigned SubRegHiHi = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
-    unsigned SubRegHiLo = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
+    unsigned SubRegLo = MRI.createVirtualRegister(&AMDGPU::SGPR_64RegClass);
+    unsigned SubRegHi = MRI.createVirtualRegister(&AMDGPU::SGPR_64RegClass);
+    unsigned SubRegHiHi = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+    unsigned SubRegHiLo = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
     BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::S_MOV_B64), SubRegLo)
             .addOperand(MI->getOperand(1));
     BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::S_MOV_B32), SubRegHiLo)
             .addImm(0);
     BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::S_MOV_B32), SubRegHiHi)
-            .addImm(RSRC_DATA_FORMAT >> 32);
+            .addImm(AMDGPU::RSRC_DATA_FORMAT >> 32);
     BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::REG_SEQUENCE), SubRegHi)
             .addReg(SubRegHiLo)
             .addImm(AMDGPU::sub0)
@@ -369,25 +499,52 @@ MachineBasicBlock * SITargetLowering::EmitInstrWithCustomInserter(
     MI->eraseFromParent();
     break;
   }
+  case AMDGPU::SI_BUFFER_RSRC: {
+    unsigned SuperReg = MI->getOperand(0).getReg();
+    unsigned Args[4];
+    for (unsigned i = 0, e = 4; i < e; ++i) {
+      MachineOperand &Arg = MI->getOperand(i + 1);
+
+      if (Arg.isReg()) {
+        Args[i] = Arg.getReg();
+        continue;
+      }
+
+      assert(Arg.isImm());
+      unsigned Reg = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+      BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::S_MOV_B32), Reg)
+              .addImm(Arg.getImm());
+      Args[i] = Reg;
+    }
+    BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::REG_SEQUENCE),
+            SuperReg)
+            .addReg(Args[0])
+            .addImm(AMDGPU::sub0)
+            .addReg(Args[1])
+            .addImm(AMDGPU::sub1)
+            .addReg(Args[2])
+            .addImm(AMDGPU::sub2)
+            .addReg(Args[3])
+            .addImm(AMDGPU::sub3);
+    MI->eraseFromParent();
+    break;
+  }
   case AMDGPU::V_SUB_F64: {
-    const SIInstrInfo *TII =
-      static_cast<const SIInstrInfo*>(getTargetMachine().getInstrInfo());
-    BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::V_ADD_F64),
-            MI->getOperand(0).getReg())
-            .addReg(MI->getOperand(1).getReg())
-            .addReg(MI->getOperand(2).getReg())
-            .addImm(0)  /* src2 */
-            .addImm(0)  /* ABS */
-            .addImm(0)  /* CLAMP */
-            .addImm(0)  /* OMOD */
-            .addImm(2); /* NEG */
+    unsigned DestReg = MI->getOperand(0).getReg();
+    BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::V_ADD_F64), DestReg)
+      .addImm(0)  // SRC0 modifiers
+      .addReg(MI->getOperand(1).getReg())
+      .addImm(1)  // SRC1 modifiers
+      .addReg(MI->getOperand(2).getReg())
+      .addImm(0)  // SRC2 modifiers
+      .addImm(0)  // src2
+      .addImm(0)  // CLAMP
+      .addImm(0); // OMOD
     MI->eraseFromParent();
     break;
   }
   case AMDGPU::SI_RegisterStorePseudo: {
     MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
-    const SIInstrInfo *TII =
-      static_cast<const SIInstrInfo*>(getTargetMachine().getInstrInfo());
     unsigned Reg = MRI.createVirtualRegister(&AMDGPU::SReg_64RegClass);
     MachineInstrBuilder MIB =
         BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::SI_RegisterStore),
@@ -396,6 +553,50 @@ MachineBasicBlock * SITargetLowering::EmitInstrWithCustomInserter(
       MIB.addOperand(MI->getOperand(i));
 
     MI->eraseFromParent();
+    break;
+  }
+  case AMDGPU::FABS_SI: {
+    MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+    const SIInstrInfo *TII =
+      static_cast<const SIInstrInfo*>(getTargetMachine().getInstrInfo());
+    unsigned Reg = MRI.createVirtualRegister(&AMDGPU::VReg_32RegClass);
+    BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::V_MOV_B32_e32),
+            Reg)
+            .addImm(0x7fffffff);
+    BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::V_AND_B32_e32),
+            MI->getOperand(0).getReg())
+            .addReg(MI->getOperand(1).getReg())
+            .addReg(Reg);
+    MI->eraseFromParent();
+    break;
+  }
+  case AMDGPU::FNEG_SI: {
+    MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+    const SIInstrInfo *TII =
+      static_cast<const SIInstrInfo*>(getTargetMachine().getInstrInfo());
+    unsigned Reg = MRI.createVirtualRegister(&AMDGPU::VReg_32RegClass);
+    BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::V_MOV_B32_e32),
+            Reg)
+            .addImm(0x80000000);
+    BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::V_XOR_B32_e32),
+            MI->getOperand(0).getReg())
+            .addReg(MI->getOperand(1).getReg())
+            .addReg(Reg);
+    MI->eraseFromParent();
+    break;
+  }
+  case AMDGPU::FCLAMP_SI: {
+    const SIInstrInfo *TII =
+      static_cast<const SIInstrInfo*>(getTargetMachine().getInstrInfo());
+    BuildMI(*BB, I, MI->getDebugLoc(), TII->get(AMDGPU::V_ADD_F32_e64),
+            MI->getOperand(0).getReg())
+            .addImm(0) // SRC0 modifiers
+            .addOperand(MI->getOperand(1))
+            .addImm(0) // SRC1 modifiers
+            .addImm(0) // SRC1
+            .addImm(1) // CLAMP
+            .addImm(0); // OMOD
+    MI->eraseFromParent();
   }
   }
   return BB;
@@ -439,65 +640,57 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
   switch (Op.getOpcode()) {
   default: return AMDGPUTargetLowering::LowerOperation(Op, DAG);
-  case ISD::ADD: return LowerADD(Op, DAG);
+  case ISD::FrameIndex: return LowerFrameIndex(Op, DAG);
   case ISD::BRCOND: return LowerBRCOND(Op, DAG);
   case ISD::LOAD: {
-    LoadSDNode *Load = dyn_cast<LoadSDNode>(Op);
-    if ((Load->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS ||
-         Load->getAddressSpace() == AMDGPUAS::PRIVATE_ADDRESS) &&
-        Op.getValueType().isVector()) {
-      SDValue MergedValues[2] = {
-        SplitVectorLoad(Op, DAG),
-        Load->getChain()
-      };
-      return DAG.getMergeValues(MergedValues, 2, SDLoc(Op));
-    } else {
-      return LowerLOAD(Op, DAG);
-    }
+    SDValue Result = LowerLOAD(Op, DAG);
+    assert((!Result.getNode() ||
+            Result.getNode()->getNumValues() == 2) &&
+           "Load should return a value and a chain");
+    return Result;
   }
 
-  case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
-  case ISD::SIGN_EXTEND: return LowerSIGN_EXTEND(Op, DAG);
+  case ISD::FSIN:
+  case ISD::FCOS:
+    return LowerTrig(Op, DAG);
+  case ISD::SELECT: return LowerSELECT(Op, DAG);
+  case ISD::FDIV: return LowerFDIV(Op, DAG);
   case ISD::STORE: return LowerSTORE(Op, DAG);
-  case ISD::ANY_EXTEND: // Fall-through
-  case ISD::ZERO_EXTEND: return LowerZERO_EXTEND(Op, DAG);
   case ISD::GlobalAddress: return LowerGlobalAddress(MFI, Op, DAG);
   case ISD::INTRINSIC_WO_CHAIN: {
     unsigned IntrinsicID =
                          cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
     EVT VT = Op.getValueType();
     SDLoc DL(Op);
-    //XXX: Hardcoded we only use two to store the pointer to the parameters.
-    unsigned NumUserSGPRs = 2;
     switch (IntrinsicID) {
     default: return AMDGPUTargetLowering::LowerOperation(Op, DAG);
     case Intrinsic::r600_read_ngroups_x:
-      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 0);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 0, false);
     case Intrinsic::r600_read_ngroups_y:
-      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 4);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 4, false);
     case Intrinsic::r600_read_ngroups_z:
-      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 8);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 8, false);
     case Intrinsic::r600_read_global_size_x:
-      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 12);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 12, false);
     case Intrinsic::r600_read_global_size_y:
-      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 16);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 16, false);
     case Intrinsic::r600_read_global_size_z:
-      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 20);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 20, false);
     case Intrinsic::r600_read_local_size_x:
-      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 24);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 24, false);
     case Intrinsic::r600_read_local_size_y:
-      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 28);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 28, false);
     case Intrinsic::r600_read_local_size_z:
-      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 32);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 32, false);
     case Intrinsic::r600_read_tgid_x:
       return CreateLiveInRegister(DAG, &AMDGPU::SReg_32RegClass,
-                     AMDGPU::SReg_32RegClass.getRegister(NumUserSGPRs + 0), VT);
+                     AMDGPU::SReg_32RegClass.getRegister(MFI->NumUserSGPRs + 0), VT);
     case Intrinsic::r600_read_tgid_y:
       return CreateLiveInRegister(DAG, &AMDGPU::SReg_32RegClass,
-                     AMDGPU::SReg_32RegClass.getRegister(NumUserSGPRs + 1), VT);
+                     AMDGPU::SReg_32RegClass.getRegister(MFI->NumUserSGPRs + 1), VT);
     case Intrinsic::r600_read_tgid_z:
       return CreateLiveInRegister(DAG, &AMDGPU::SReg_32RegClass,
-                     AMDGPU::SReg_32RegClass.getRegister(NumUserSGPRs + 2), VT);
+                     AMDGPU::SReg_32RegClass.getRegister(MFI->NumUserSGPRs + 2), VT);
     case Intrinsic::r600_read_tidig_x:
       return CreateLiveInRegister(DAG, &AMDGPU::VReg_32RegClass,
                                   AMDGPU::VGPR0, VT);
@@ -509,7 +702,7 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
                                   AMDGPU::VGPR2, VT);
     case AMDGPUIntrinsic::SI_load_const: {
       SDValue Ops [] = {
-        ResourceDescriptorToi128(Op.getOperand(1), DAG),
+        Op.getOperand(1),
         Op.getOperand(2)
       };
 
@@ -518,7 +711,7 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
           MachineMemOperand::MOLoad | MachineMemOperand::MOInvariant,
           VT.getSizeInBits() / 8, 4);
       return DAG.getMemIntrinsicNode(AMDGPUISD::LOAD_CONSTANT, DL,
-                                     Op->getVTList(), Ops, 2, VT, MMO);
+                                     Op->getVTList(), Ops, VT, MMO);
     }
     case AMDGPUIntrinsic::SI_sample:
       return LowerSampleIntrinsic(AMDGPUISD::SAMPLE, Op, DAG);
@@ -530,7 +723,7 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
       return LowerSampleIntrinsic(AMDGPUISD::SAMPLEL, Op, DAG);
     case AMDGPUIntrinsic::SI_vs_load_input:
       return DAG.getNode(AMDGPUISD::LOAD_INPUT, DL, VT,
-                         ResourceDescriptorToi128(Op.getOperand(1), DAG),
+                         Op.getOperand(1),
                          Op.getOperand(2),
                          Op.getOperand(3));
     }
@@ -545,7 +738,7 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
         SDLoc DL(Op);
         SDValue Ops [] = {
           Chain,
-          ResourceDescriptorToi128(Op.getOperand(2), DAG),
+          Op.getOperand(2),
           Op.getOperand(3),
           Op.getOperand(4),
           Op.getOperand(5),
@@ -566,8 +759,7 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
             MachineMemOperand::MOStore,
             VT.getSizeInBits() / 8, 4);
         return DAG.getMemIntrinsicNode(AMDGPUISD::TBUFFER_STORE_FORMAT, DL,
-                                       Op->getVTList(), Ops,
-                                       sizeof(Ops)/sizeof(Ops[0]), VT, MMO);
+                                       Op->getVTList(), Ops, VT, MMO);
       }
       default:
         break;
@@ -576,33 +768,6 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   return SDValue();
 }
 
-SDValue SITargetLowering::LowerADD(SDValue Op,
-                                   SelectionDAG &DAG) const {
-  if (Op.getValueType() != MVT::i64)
-    return SDValue();
-
-  SDLoc DL(Op);
-  SDValue LHS = Op.getOperand(0);
-  SDValue RHS = Op.getOperand(1);
-
-  SDValue Zero = DAG.getConstant(0, MVT::i32);
-  SDValue One = DAG.getConstant(1, MVT::i32);
-
-  SDValue Lo0 = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, LHS, Zero);
-  SDValue Hi0 = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, LHS, One);
-
-  SDValue Lo1 = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, RHS, Zero);
-  SDValue Hi1 = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, RHS, One);
-
-  SDVTList VTList = DAG.getVTList(MVT::i32, MVT::Glue);
-
-  SDValue AddLo = DAG.getNode(ISD::ADDC, DL, VTList, Lo0, Lo1);
-  SDValue Carry = AddLo.getValue(1);
-  SDValue AddHi = DAG.getNode(ISD::ADDE, DL, VTList, Hi0, Hi1, Carry);
-
-  return DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, AddLo, AddHi.getValue(0));
-}
-
 /// \brief Helper function for LowerBRCOND
 static SDNode *findUser(SDValue Value, unsigned Opcode) {
 
@@ -616,7 +781,22 @@ static SDNode *findUser(SDValue Value, unsigned Opcode) {
     if (I->getOpcode() == Opcode)
       return *I;
   }
-  return 0;
+  return nullptr;
+}
+
+SDValue SITargetLowering::LowerFrameIndex(SDValue Op, SelectionDAG &DAG) const {
+
+  MachineFunction &MF = DAG.getMachineFunction();
+  const SIInstrInfo *TII =
+    static_cast<const SIInstrInfo*>(getTargetMachine().getInstrInfo());
+  const SIRegisterInfo &TRI = TII->getRegisterInfo();
+  FrameIndexSDNode *FINode = cast<FrameIndexSDNode>(Op);
+  unsigned FrameIndex = FINode->getIndex();
+
+  CreateLiveInRegister(DAG, &AMDGPU::SReg_32RegClass,
+    TRI.getPreloadedValue(MF, SIRegisterInfo::SCRATCH_WAVE_OFFSET), MVT::i32);
+
+  return DAG.getTargetFrameIndex(FrameIndex, MVT::i32);
 }
 
 /// This transforms the control flow intrinsics to get the branch destination as
@@ -628,7 +808,7 @@ SDValue SITargetLowering::LowerBRCOND(SDValue BRCOND,
 
   SDNode *Intr = BRCOND.getOperand(1).getNode();
   SDValue Target = BRCOND.getOperand(2);
-  SDNode *BR = 0;
+  SDNode *BR = nullptr;
 
   if (Intr->getOpcode() == ISD::SETCC) {
     // As long as we negate the condition everything is fine
@@ -661,7 +841,7 @@ SDValue SITargetLowering::LowerBRCOND(SDValue BRCOND,
   // build the new intrinsic call
   SDNode *Result = DAG.getNode(
     Res.size() > 1 ? ISD::INTRINSIC_W_CHAIN : ISD::INTRINSIC_VOID, DL,
-    DAG.getVTList(Res.data(), Res.size()), Ops.data(), Ops.size()).getNode();
+    DAG.getVTList(Res), Ops).getNode();
 
   if (BR) {
     // Give the branch instruction our target
@@ -669,7 +849,7 @@ SDValue SITargetLowering::LowerBRCOND(SDValue BRCOND,
       BR->getOperand(0),
       BRCOND.getOperand(2)
     };
-    DAG.MorphNodeTo(BR, ISD::BR, BR->getVTList(), Ops, 2);
+    DAG.MorphNodeTo(BR, ISD::BR, BR->getVTList(), Ops);
   }
 
   SDValue Chain = SDValue(Result, Result->getNumValues() - 1);
@@ -697,42 +877,57 @@ SDValue SITargetLowering::LowerBRCOND(SDValue BRCOND,
   return Chain;
 }
 
-SDValue SITargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
-  SDLoc DL(Op);
-  LoadSDNode *Load = cast<LoadSDNode>(Op);
+SDValue SITargetLowering::LowerGlobalAddress(AMDGPUMachineFunction *MFI,
+                                             SDValue Op,
+                                             SelectionDAG &DAG) const {
+  GlobalAddressSDNode *GSD = cast<GlobalAddressSDNode>(Op);
 
-  if (Load->getAddressSpace() != AMDGPUAS::PRIVATE_ADDRESS)
-    return SDValue();
+  if (GSD->getAddressSpace() != AMDGPUAS::CONSTANT_ADDRESS)
+    return AMDGPUTargetLowering::LowerGlobalAddress(MFI, Op, DAG);
 
-  SDValue TruncPtr = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32,
-                                 Load->getBasePtr(), DAG.getConstant(0, MVT::i32));
-  SDValue Ptr = DAG.getNode(ISD::SRL, DL, MVT::i32, TruncPtr,
-                            DAG.getConstant(2, MVT::i32));
-
-  SDValue Ret = DAG.getNode(AMDGPUISD::REGISTER_LOAD, DL, Op.getValueType(),
-                            Load->getChain(), Ptr,
-                            DAG.getTargetConstant(0, MVT::i32),
-                            Op.getOperand(2));
-  SDValue MergedValues[2] = {
-    Ret,
-    Load->getChain()
-  };
-  return DAG.getMergeValues(MergedValues, 2, DL);
+  SDLoc DL(GSD);
+  const GlobalValue *GV = GSD->getGlobal();
+  MVT PtrVT = getPointerTy(GSD->getAddressSpace());
+
+  SDValue Ptr = DAG.getNode(AMDGPUISD::CONST_DATA_PTR, DL, PtrVT);
+  SDValue GA = DAG.getTargetGlobalAddress(GV, DL, MVT::i32);
+
+  SDValue PtrLo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, Ptr,
+                              DAG.getConstant(0, MVT::i32));
+  SDValue PtrHi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, Ptr,
+                              DAG.getConstant(1, MVT::i32));
 
+  SDValue Lo = DAG.getNode(ISD::ADDC, DL, DAG.getVTList(MVT::i32, MVT::Glue),
+                           PtrLo, GA);
+  SDValue Hi = DAG.getNode(ISD::ADDE, DL, DAG.getVTList(MVT::i32, MVT::Glue),
+                           PtrHi, DAG.getConstant(0, MVT::i32),
+                           SDValue(Lo.getNode(), 1));
+  return DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Lo, Hi);
 }
 
-SDValue SITargetLowering::ResourceDescriptorToi128(SDValue Op,
-                                             SelectionDAG &DAG) const {
+SDValue SITargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  LoadSDNode *Load = cast<LoadSDNode>(Op);
 
-  if (Op.getValueType() == MVT::i128) {
-    return Op;
+  if (Op.getValueType().isVector()) {
+    assert(Op.getValueType().getVectorElementType() == MVT::i32 &&
+           "Custom lowering for non-i32 vectors hasn't been implemented.");
+    unsigned NumElements = Op.getValueType().getVectorNumElements();
+    assert(NumElements != 2 && "v2 loads are supported for all address spaces.");
+    switch (Load->getAddressSpace()) {
+      default: break;
+      case AMDGPUAS::GLOBAL_ADDRESS:
+      case AMDGPUAS::PRIVATE_ADDRESS:
+        // v4 loads are supported for private and global memory.
+        if (NumElements <= 4)
+          break;
+        // fall-through
+      case AMDGPUAS::LOCAL_ADDRESS:
+        return SplitVectorLoad(Op, DAG);
+    }
   }
 
-  assert(Op.getOpcode() == ISD::UNDEF);
-
-  return DAG.getNode(ISD::BUILD_PAIR, SDLoc(Op), MVT::i128,
-                     DAG.getConstant(0, MVT::i64),
-                     DAG.getConstant(0, MVT::i64));
+  return AMDGPUTargetLowering::LowerLOAD(Op, DAG);
 }
 
 SDValue SITargetLowering::LowerSampleIntrinsic(unsigned Opcode,
@@ -740,42 +935,129 @@ SDValue SITargetLowering::LowerSampleIntrinsic(unsigned Opcode,
                                                SelectionDAG &DAG) const {
   return DAG.getNode(Opcode, SDLoc(Op), Op.getValueType(), Op.getOperand(1),
                      Op.getOperand(2),
-                     ResourceDescriptorToi128(Op.getOperand(3), DAG),
+                     Op.getOperand(3),
                      Op.getOperand(4));
 }
 
-SDValue SITargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const {
+SDValue SITargetLowering::LowerSELECT(SDValue Op, SelectionDAG &DAG) const {
+  if (Op.getValueType() != MVT::i64)
+    return SDValue();
+
+  SDLoc DL(Op);
+  SDValue Cond = Op.getOperand(0);
+
+  SDValue Zero = DAG.getConstant(0, MVT::i32);
+  SDValue One = DAG.getConstant(1, MVT::i32);
+
+  SDValue LHS = DAG.getNode(ISD::BITCAST, DL, MVT::v2i32, Op.getOperand(1));
+  SDValue RHS = DAG.getNode(ISD::BITCAST, DL, MVT::v2i32, Op.getOperand(2));
+
+  SDValue Lo0 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::i32, LHS, Zero);
+  SDValue Lo1 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::i32, RHS, Zero);
+
+  SDValue Lo = DAG.getSelect(DL, MVT::i32, Cond, Lo0, Lo1);
+
+  SDValue Hi0 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::i32, LHS, One);
+  SDValue Hi1 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::i32, RHS, One);
+
+  SDValue Hi = DAG.getSelect(DL, MVT::i32, Cond, Hi0, Hi1);
+
+  SDValue Res = DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v2i32, Lo, Hi);
+  return DAG.getNode(ISD::BITCAST, DL, MVT::i64, Res);
+}
+
+// Catch division cases where we can use shortcuts with rcp and rsq
+// instructions.
+SDValue SITargetLowering::LowerFastFDIV(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc SL(Op);
   SDValue LHS = Op.getOperand(0);
   SDValue RHS = Op.getOperand(1);
-  SDValue True = Op.getOperand(2);
-  SDValue False = Op.getOperand(3);
-  SDValue CC = Op.getOperand(4);
   EVT VT = Op.getValueType();
-  SDLoc DL(Op);
+  bool Unsafe = DAG.getTarget().Options.UnsafeFPMath;
+
+  if (const ConstantFPSDNode *CLHS = dyn_cast<ConstantFPSDNode>(LHS)) {
+    if ((Unsafe || (VT == MVT::f32 && !Subtarget->hasFP32Denormals())) &&
+        CLHS->isExactlyValue(1.0)) {
+      // v_rcp_f32 and v_rsq_f32 do not support denormals, and according to
+      // the CI documentation has a worst case error of 1 ulp.
+      // OpenCL requires <= 2.5 ulp for 1.0 / x, so it should always be OK to
+      // use it as long as we aren't trying to use denormals.
+
+      // 1.0 / sqrt(x) -> rsq(x)
+      //
+      // XXX - Is UnsafeFPMath sufficient to do this for f64? The maximum ULP
+      // error seems really high at 2^29 ULP.
+      if (RHS.getOpcode() == ISD::FSQRT)
+        return DAG.getNode(AMDGPUISD::RSQ, SL, VT, RHS.getOperand(0));
+
+      // 1.0 / x -> rcp(x)
+      return DAG.getNode(AMDGPUISD::RCP, SL, VT, RHS);
+    }
+  }
 
-  // Possible Min/Max pattern
-  SDValue MinMax = LowerMinMax(Op, DAG);
-  if (MinMax.getNode()) {
-    return MinMax;
+  if (Unsafe) {
+    // Turn into multiply by the reciprocal.
+    // x / y -> x * (1.0 / y)
+    SDValue Recip = DAG.getNode(AMDGPUISD::RCP, SL, VT, RHS);
+    return DAG.getNode(ISD::FMUL, SL, VT, LHS, Recip);
   }
 
-  SDValue Cond = DAG.getNode(ISD::SETCC, DL, MVT::i1, LHS, RHS, CC);
-  return DAG.getNode(ISD::SELECT, DL, VT, Cond, True, False);
+  return SDValue();
 }
 
-SDValue SITargetLowering::LowerSIGN_EXTEND(SDValue Op,
-                                           SelectionDAG &DAG) const {
-  EVT VT = Op.getValueType();
-  SDLoc DL(Op);
+SDValue SITargetLowering::LowerFDIV32(SDValue Op, SelectionDAG &DAG) const {
+  SDValue FastLowered = LowerFastFDIV(Op, DAG);
+  if (FastLowered.getNode())
+    return FastLowered;
 
-  if (VT != MVT::i64) {
+  // This uses v_rcp_f32 which does not handle denormals. Let this hit a
+  // selection error for now rather than do something incorrect.
+  if (Subtarget->hasFP32Denormals())
     return SDValue();
-  }
 
-  SDValue Hi = DAG.getNode(ISD::SRA, DL, MVT::i32, Op.getOperand(0),
-                                                 DAG.getConstant(31, MVT::i32));
+  SDLoc SL(Op);
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+
+  SDValue r1 = DAG.getNode(ISD::FABS, SL, MVT::f32, RHS);
 
-  return DAG.getNode(ISD::BUILD_PAIR, DL, VT, Op.getOperand(0), Hi);
+  const APFloat K0Val(BitsToFloat(0x6f800000));
+  const SDValue K0 = DAG.getConstantFP(K0Val, MVT::f32);
+
+  const APFloat K1Val(BitsToFloat(0x2f800000));
+  const SDValue K1 = DAG.getConstantFP(K1Val, MVT::f32);
+
+  const SDValue One = DAG.getTargetConstantFP(1.0, MVT::f32);
+
+  EVT SetCCVT = getSetCCResultType(*DAG.getContext(), MVT::f32);
+
+  SDValue r2 = DAG.getSetCC(SL, SetCCVT, r1, K0, ISD::SETOGT);
+
+  SDValue r3 = DAG.getNode(ISD::SELECT, SL, MVT::f32, r2, K1, One);
+
+  r1 = DAG.getNode(ISD::FMUL, SL, MVT::f32, RHS, r3);
+
+  SDValue r0 = DAG.getNode(AMDGPUISD::RCP, SL, MVT::f32, r1);
+
+  SDValue Mul = DAG.getNode(ISD::FMUL, SL, MVT::f32, LHS, r0);
+
+  return DAG.getNode(ISD::FMUL, SL, MVT::f32, r3, Mul);
+}
+
+SDValue SITargetLowering::LowerFDIV64(SDValue Op, SelectionDAG &DAG) const {
+  return SDValue();
+}
+
+SDValue SITargetLowering::LowerFDIV(SDValue Op, SelectionDAG &DAG) const {
+  EVT VT = Op.getValueType();
+
+  if (VT == MVT::f32)
+    return LowerFDIV32(Op, DAG);
+
+  if (VT == MVT::f64)
+    return LowerFDIV64(Op, DAG);
+
+  llvm_unreachable("Unexpected type for fdiv");
 }
 
 SDValue SITargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
@@ -783,6 +1065,18 @@ SDValue SITargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
   StoreSDNode *Store = cast<StoreSDNode>(Op);
   EVT VT = Store->getMemoryVT();
 
+  // These stores are legal.
+  if (Store->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS &&
+      VT.isVector() && VT.getVectorNumElements() == 2 &&
+      VT.getVectorElementType() == MVT::i32)
+    return SDValue();
+
+  if (Store->getAddressSpace() == AMDGPUAS::PRIVATE_ADDRESS) {
+    if (VT.isVector() && VT.getVectorNumElements() > 4)
+      return SplitVectorStore(Op, DAG);
+    return SDValue();
+  }
+
   SDValue Ret = AMDGPUTargetLowering::LowerSTORE(Op, DAG);
   if (Ret.getNode())
     return Ret;
@@ -790,61 +1084,125 @@ SDValue SITargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
   if (VT.isVector() && VT.getVectorNumElements() >= 8)
       return SplitVectorStore(Op, DAG);
 
-  if (Store->getAddressSpace() != AMDGPUAS::PRIVATE_ADDRESS)
+  if (VT == MVT::i1)
+    return DAG.getTruncStore(Store->getChain(), DL,
+                        DAG.getSExtOrTrunc(Store->getValue(), DL, MVT::i32),
+                        Store->getBasePtr(), MVT::i1, Store->getMemOperand());
+
+  return SDValue();
+}
+
+SDValue SITargetLowering::LowerTrig(SDValue Op, SelectionDAG &DAG) const {
+  EVT VT = Op.getValueType();
+  SDValue Arg = Op.getOperand(0);
+  SDValue FractPart = DAG.getNode(AMDGPUISD::FRACT, SDLoc(Op), VT,
+        DAG.getNode(ISD::FMUL, SDLoc(Op), VT, Arg,
+          DAG.getConstantFP(0.5 / M_PI, VT)));
+
+  switch (Op.getOpcode()) {
+  case ISD::FCOS:
+    return DAG.getNode(AMDGPUISD::COS_HW, SDLoc(Op), VT, FractPart);
+  case ISD::FSIN:
+    return DAG.getNode(AMDGPUISD::SIN_HW, SDLoc(Op), VT, FractPart);
+  default:
+    llvm_unreachable("Wrong trig opcode");
+  }
+}
+
+//===----------------------------------------------------------------------===//
+// Custom DAG optimizations
+//===----------------------------------------------------------------------===//
+
+SDValue SITargetLowering::performUCharToFloatCombine(SDNode *N,
+                                                     DAGCombinerInfo &DCI) {
+  EVT VT = N->getValueType(0);
+  EVT ScalarVT = VT.getScalarType();
+  if (ScalarVT != MVT::f32)
     return SDValue();
 
-  SDValue TruncPtr = DAG.getZExtOrTrunc(Store->getBasePtr(), DL, MVT::i32);
-  SDValue Ptr = DAG.getNode(ISD::SRL, DL, MVT::i32, TruncPtr,
-                            DAG.getConstant(2, MVT::i32));
-  SDValue Chain = Store->getChain();
-  SmallVector<SDValue, 8> Values;
+  SelectionDAG &DAG = DCI.DAG;
+  SDLoc DL(N);
 
-  if (VT == MVT::i64) {
-    for (unsigned i = 0; i < 2; ++i) {
-      Values.push_back(DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32,
-                       Store->getValue(), DAG.getConstant(i, MVT::i32)));
-    }
-  } else if (VT == MVT::i128) {
-    for (unsigned i = 0; i < 2; ++i) {
-      for (unsigned j = 0; j < 2; ++j) {
-        Values.push_back(DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32,
-                           DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i64,
-                           Store->getValue(), DAG.getConstant(i, MVT::i32)),
-                         DAG.getConstant(j, MVT::i32)));
-      }
+  SDValue Src = N->getOperand(0);
+  EVT SrcVT = Src.getValueType();
+
+  // TODO: We could try to match extracting the higher bytes, which would be
+  // easier if i8 vectors weren't promoted to i32 vectors, particularly after
+  // types are legalized. v4i8 -> v4f32 is probably the only case to worry
+  // about in practice.
+  if (DCI.isAfterLegalizeVectorOps() && SrcVT == MVT::i32) {
+    if (DAG.MaskedValueIsZero(Src, APInt::getHighBitsSet(32, 24))) {
+      SDValue Cvt = DAG.getNode(AMDGPUISD::CVT_F32_UBYTE0, DL, VT, Src);
+      DCI.AddToWorklist(Cvt.getNode());
+      return Cvt;
     }
-  } else {
-    Values.push_back(Store->getValue());
   }
 
-  for (unsigned i = 0; i < Values.size(); ++i) {
-    SDValue PartPtr = DAG.getNode(ISD::ADD, DL, MVT::i32,
-                                  Ptr, DAG.getConstant(i, MVT::i32));
-    Chain = DAG.getNode(AMDGPUISD::REGISTER_STORE, DL, MVT::Other,
-                        Chain, Values[i], PartPtr,
-                        DAG.getTargetConstant(0, MVT::i32));
+  // We are primarily trying to catch operations on illegal vector types
+  // before they are expanded.
+  // For scalars, we can use the more flexible method of checking masked bits
+  // after legalization.
+  if (!DCI.isBeforeLegalize() ||
+      !SrcVT.isVector() ||
+      SrcVT.getVectorElementType() != MVT::i8) {
+    return SDValue();
   }
-  return Chain;
-}
-
 
-SDValue SITargetLowering::LowerZERO_EXTEND(SDValue Op,
-                                           SelectionDAG &DAG) const {
-  EVT VT = Op.getValueType();
-  SDLoc DL(Op);
+  assert(DCI.isBeforeLegalize() && "Unexpected legal type");
 
-  if (VT != MVT::i64) {
+  // Weird sized vectors are a pain to handle, but we know 3 is really the same
+  // size as 4.
+  unsigned NElts = SrcVT.getVectorNumElements();
+  if (!SrcVT.isSimple() && NElts != 3)
     return SDValue();
+
+  // Handle v4i8 -> v4f32 extload. Replace the v4i8 with a legal i32 load to
+  // prevent a mess from expanding to v4i32 and repacking.
+  if (ISD::isNormalLoad(Src.getNode()) && Src.hasOneUse()) {
+    EVT LoadVT = getEquivalentMemType(*DAG.getContext(), SrcVT);
+    EVT RegVT = getEquivalentLoadRegType(*DAG.getContext(), SrcVT);
+    EVT FloatVT = EVT::getVectorVT(*DAG.getContext(), MVT::f32, NElts);
+
+    LoadSDNode *Load = cast<LoadSDNode>(Src);
+    SDValue NewLoad = DAG.getExtLoad(ISD::ZEXTLOAD, DL, RegVT,
+                                     Load->getChain(),
+                                     Load->getBasePtr(),
+                                     LoadVT,
+                                     Load->getMemOperand());
+
+    // Make sure successors of the original load stay after it by updating
+    // them to use the new Chain.
+    DAG.ReplaceAllUsesOfValueWith(SDValue(Load, 1), NewLoad.getValue(1));
+
+    SmallVector<SDValue, 4> Elts;
+    if (RegVT.isVector())
+      DAG.ExtractVectorElements(NewLoad, Elts);
+    else
+      Elts.push_back(NewLoad);
+
+    SmallVector<SDValue, 4> Ops;
+
+    unsigned EltIdx = 0;
+    for (SDValue Elt : Elts) {
+      unsigned ComponentsInElt = std::min(4u, NElts - 4 * EltIdx);
+      for (unsigned I = 0; I < ComponentsInElt; ++I) {
+        unsigned Opc = AMDGPUISD::CVT_F32_UBYTE0 + I;
+        SDValue Cvt = DAG.getNode(Opc, DL, MVT::f32, Elt);
+        DCI.AddToWorklist(Cvt.getNode());
+        Ops.push_back(Cvt);
+      }
+
+      ++EltIdx;
+    }
+
+    assert(Ops.size() == NElts);
+
+    return DAG.getNode(ISD::BUILD_VECTOR, DL, FloatVT, Ops);
   }
 
-  return DAG.getNode(ISD::BUILD_PAIR, DL, VT, Op.getOperand(0),
-                                              DAG.getConstant(0, MVT::i32));
+  return SDValue();
 }
 
-//===----------------------------------------------------------------------===//
-// Custom DAG optimizations
-//===----------------------------------------------------------------------===//
-
 SDValue SITargetLowering::PerformDAGCombine(SDNode *N,
                                             DAGCombinerInfo &DCI) const {
   SelectionDAG &DAG = DCI.DAG;
@@ -852,26 +1210,12 @@ SDValue SITargetLowering::PerformDAGCombine(SDNode *N,
   EVT VT = N->getValueType(0);
 
   switch (N->getOpcode()) {
-    default: break;
-    case ISD::SELECT_CC: {
-      ConstantSDNode *True, *False;
-      // i1 selectcc(l, r, -1, 0, cc) -> i1 setcc(l, r, cc)
-      if ((True = dyn_cast<ConstantSDNode>(N->getOperand(2)))
-          && (False = dyn_cast<ConstantSDNode>(N->getOperand(3)))
-          && True->isAllOnesValue()
-          && False->isNullValue()
-          && VT == MVT::i1) {
-        return DAG.getNode(ISD::SETCC, DL, VT, N->getOperand(0),
-                           N->getOperand(1), N->getOperand(4));
-
-      }
-      break;
-    }
+    default: return AMDGPUTargetLowering::PerformDAGCombine(N, DCI);
     case ISD::SETCC: {
       SDValue Arg0 = N->getOperand(0);
       SDValue Arg1 = N->getOperand(1);
       SDValue CC = N->getOperand(2);
-      ConstantSDNode * C = NULL;
+      ConstantSDNode * C = nullptr;
       ISD::CondCode CCOp = dyn_cast<CondCodeSDNode>(CC)->get();
 
       // i1 setcc (sext(i1), 0, setne) -> i1 setcc(i1, 0, setne)
@@ -886,8 +1230,34 @@ SDValue SITargetLowering::PerformDAGCombine(SDNode *N,
       }
       break;
     }
+
+  case AMDGPUISD::CVT_F32_UBYTE0:
+  case AMDGPUISD::CVT_F32_UBYTE1:
+  case AMDGPUISD::CVT_F32_UBYTE2:
+  case AMDGPUISD::CVT_F32_UBYTE3: {
+    unsigned Offset = N->getOpcode() - AMDGPUISD::CVT_F32_UBYTE0;
+
+    SDValue Src = N->getOperand(0);
+    APInt Demanded = APInt::getBitsSet(32, 8 * Offset, 8 * Offset + 8);
+
+    APInt KnownZero, KnownOne;
+    TargetLowering::TargetLoweringOpt TLO(DAG, !DCI.isBeforeLegalize(),
+                                          !DCI.isBeforeLegalizeOps());
+    const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+    if (TLO.ShrinkDemandedConstant(Src, Demanded) ||
+        TLI.SimplifyDemandedBits(Src, Demanded, KnownZero, KnownOne, TLO)) {
+      DCI.CommitTargetLoweringOpt(TLO);
+    }
+
+    break;
   }
-  return SDValue();
+
+  case ISD::UINT_TO_FP: {
+    return performUCharToFloatCombine(N, DCI);
+  }
+  }
+
+  return AMDGPUTargetLowering::PerformDAGCombine(N, DCI);
 }
 
 /// \brief Test if RegClass is one of the VSrc classes
@@ -918,9 +1288,11 @@ int32_t SITargetLowering::analyzeImmediate(const SDNode *N) const {
         return -1;
     }
     Imm.I = Node->getSExtValue();
-  } else if (const ConstantFPSDNode *Node = dyn_cast<ConstantFPSDNode>(N))
+  } else if (const ConstantFPSDNode *Node = dyn_cast<ConstantFPSDNode>(N)) {
+    if (N->getValueType(0) != MVT::f32)
+      return -1;
     Imm.F = Node->getValueAPF().convertToFloat();
-  else
+  } else
     return -1; // It isn't an immediate
 
   if ((Imm.I >= -16 && Imm.I <= 64) ||
@@ -940,7 +1312,7 @@ bool SITargetLowering::foldImm(SDValue &Operand, int32_t &Immediate,
   MachineSDNode *Mov = dyn_cast<MachineSDNode>(Operand);
   const SIInstrInfo *TII =
     static_cast<const SIInstrInfo*>(getTargetMachine().getInstrInfo());
-  if (Mov == 0 || !TII->isMov(Mov->getMachineOpcode()))
+  if (!Mov || !TII->isMov(Mov->getMachineOpcode()))
     return false;
 
   const SDValue &Op = Mov->getOperand(0);
@@ -987,7 +1359,7 @@ const TargetRegisterClass *SITargetLowering::getRegClassForNode(
       }
       return TRI.getPhysRegClass(Reg);
     }
-    default:  return NULL;
+    default:  return nullptr;
     }
   }
   const MCInstrDesc &Desc = TII->get(Op->getMachineOpcode());
@@ -1047,7 +1419,7 @@ void SITargetLowering::ensureSRegLimit(SelectionDAG &DAG, SDValue &Operand,
   else
     return;
 
-  // Nothing todo if they fit naturaly
+  // Nothing to do if they fit naturally
   if (fitsRegClass(DAG, Operand, RegClass))
     return;
 
@@ -1059,9 +1431,19 @@ void SITargetLowering::ensureSRegLimit(SelectionDAG &DAG, SDValue &Operand,
 
   // This is a conservative aproach. It is possible that we can't determine the
   // correct register class and copy too often, but better safe than sorry.
-  SDValue RC = DAG.getTargetConstant(RegClass, MVT::i32);
-  SDNode *Node = DAG.getMachineNode(TargetOpcode::COPY_TO_REGCLASS, SDLoc(),
-                                    Operand.getValueType(), Operand, RC);
+
+  SDNode *Node;
+  // We can't use COPY_TO_REGCLASS with FrameIndex arguments.
+  if (isa<FrameIndexSDNode>(Operand)) {
+    unsigned Opcode = Operand.getValueType() == MVT::i32 ?
+                      AMDGPU::S_MOV_B32 : AMDGPU::S_MOV_B64;
+    Node = DAG.getMachineNode(Opcode, SDLoc(), Operand.getValueType(),
+                              Operand);
+  } else {
+    SDValue RC = DAG.getTargetConstant(RegClass, MVT::i32);
+    Node = DAG.getMachineNode(TargetOpcode::COPY_TO_REGCLASS, SDLoc(),
+                              Operand.getValueType(), Operand, RC);
+  }
   Operand = SDValue(Node, 0);
 }
 
@@ -1091,22 +1473,22 @@ SDNode *SITargetLowering::foldOperands(MachineSDNode *Node,
 
   // Commuted opcode if available
   int OpcodeRev = Desc->isCommutable() ? TII->commuteOpcode(Opcode) : -1;
-  const MCInstrDesc *DescRev = OpcodeRev == -1 ? 0 : &TII->get(OpcodeRev);
+  const MCInstrDesc *DescRev = OpcodeRev == -1 ? nullptr : &TII->get(OpcodeRev);
 
   assert(!DescRev || DescRev->getNumDefs() == NumDefs);
   assert(!DescRev || DescRev->getNumOperands() == NumOps);
 
   // e64 version if available, -1 otherwise
   int OpcodeE64 = AMDGPU::getVOPe64(Opcode);
-  const MCInstrDesc *DescE64 = OpcodeE64 == -1 ? 0 : &TII->get(OpcodeE64);
+  const MCInstrDesc *DescE64 = OpcodeE64 == -1 ? nullptr : &TII->get(OpcodeE64);
+  int InputModifiers[3] = {0};
 
   assert(!DescE64 || DescE64->getNumDefs() == NumDefs);
-  assert(!DescE64 || DescE64->getNumOperands() == (NumOps + 4));
 
   int32_t Immediate = Desc->getSize() == 4 ? 0 : -1;
   bool HaveVSrc = false, HaveSSrc = false;
 
-  // First figure out what we alread have in this instruction
+  // First figure out what we already have in this instruction.
   for (unsigned i = 0, e = Node->getNumOperands(), Op = NumDefs;
        i != e && Op < NumOps; ++i, ++Op) {
 
@@ -1125,7 +1507,7 @@ SDNode *SITargetLowering::foldOperands(MachineSDNode *Node,
     }
   }
 
-  // If we neither have VSrc nor SSrc it makes no sense to continue
+  // If we neither have VSrc nor SSrc, it makes no sense to continue.
   if (!HaveVSrc && !HaveSSrc)
     return Node;
 
@@ -1141,20 +1523,28 @@ SDNode *SITargetLowering::foldOperands(MachineSDNode *Node,
     const SDValue &Operand = Node->getOperand(i);
     Ops.push_back(Operand);
 
-    // Already folded immediate ?
+    // Already folded immediate?
     if (isa<ConstantSDNode>(Operand.getNode()) ||
         isa<ConstantFPSDNode>(Operand.getNode()))
       continue;
 
-    // Is this a VSrc or SSrc operand ?
+    // Is this a VSrc or SSrc operand?
     unsigned RegClass = Desc->OpInfo[Op].RegClass;
     if (isVSrc(RegClass) || isSSrc(RegClass)) {
       // Try to fold the immediates
       if (!foldImm(Ops[i], Immediate, ScalarSlotUsed)) {
-        // Folding didn't worked, make sure we don't hit the SReg limit
+        // Folding didn't work, make sure we don't hit the SReg limit.
         ensureSRegLimit(DAG, Ops[i], RegClass, ScalarSlotUsed);
       }
       continue;
+    } else {
+      // If it's not a VSrc or SSrc operand check if we have a GlobalAddress.
+      // These will be lowered to immediates, so we will need to insert a MOV.
+      if (isa<GlobalAddressSDNode>(Ops[i])) {
+        SDNode *Node = DAG.getMachineNode(AMDGPU::V_MOV_B32_e32, SDLoc(),
+                                    Operand.getValueType(), Operand);
+        Ops[i] = SDValue(Node, 0);
+      }
     }
 
     if (i == 1 && DescRev && fitsRegClass(DAG, Ops[0], RegClass)) {
@@ -1168,18 +1558,18 @@ SDNode *SITargetLowering::foldOperands(MachineSDNode *Node,
            fitsRegClass(DAG, Ops[1], OtherRegClass))) {
 
         // Swap commutable operands
-        SDValue Tmp = Ops[1];
-        Ops[1] = Ops[0];
-        Ops[0] = Tmp;
+        std::swap(Ops[0], Ops[1]);
 
         Desc = DescRev;
-        DescRev = 0;
+        DescRev = nullptr;
         continue;
       }
     }
 
-    if (DescE64 && !Immediate) {
+    if (Immediate)
+      continue;
 
+    if (DescE64) {
       // Test if it makes sense to switch to e64 encoding
       unsigned OtherRegClass = DescE64->OpInfo[Op].RegClass;
       if (!isVSrc(OtherRegClass) && !isSSrc(OtherRegClass))
@@ -1194,14 +1584,46 @@ SDNode *SITargetLowering::foldOperands(MachineSDNode *Node,
         Immediate = -1;
         Promote2e64 = true;
         Desc = DescE64;
-        DescE64 = 0;
+        DescE64 = nullptr;
       }
     }
+
+    if (!DescE64 && !Promote2e64)
+      continue;
+    if (!Operand.isMachineOpcode())
+      continue;
+    if (Operand.getMachineOpcode() == AMDGPU::FNEG_SI) {
+      Ops.pop_back();
+      Ops.push_back(Operand.getOperand(0));
+      InputModifiers[i] = 1;
+      Promote2e64 = true;
+      if (!DescE64)
+        continue;
+      Desc = DescE64;
+      DescE64 = nullptr;
+    }
+    else if (Operand.getMachineOpcode() == AMDGPU::FABS_SI) {
+      Ops.pop_back();
+      Ops.push_back(Operand.getOperand(0));
+      InputModifiers[i] = 2;
+      Promote2e64 = true;
+      if (!DescE64)
+        continue;
+      Desc = DescE64;
+      DescE64 = nullptr;
+    }
   }
 
   if (Promote2e64) {
+    std::vector<SDValue> OldOps(Ops);
+    Ops.clear();
+    for (unsigned i = 0; i < OldOps.size(); ++i) {
+      // src_modifier
+      Ops.push_back(DAG.getTargetConstant(InputModifiers[i], MVT::i32));
+      Ops.push_back(OldOps[i]);
+    }
     // Add the modifier flags while promoting
-    for (unsigned i = 0; i < 4; ++i)
+    for (unsigned i = 0; i < 2; ++i)
       Ops.push_back(DAG.getTargetConstant(0, MVT::i32));
   }
 
@@ -1279,7 +1701,7 @@ void SITargetLowering::adjustWritemask(MachineSDNode *&Node,
   Ops.push_back(DAG.getTargetConstant(NewDmask, MVT::i32));
   for (unsigned i = 1, e = Node->getNumOperands(); i != e; ++i)
     Ops.push_back(Node->getOperand(i));
-  Node = (MachineSDNode*)DAG.UpdateNodeOperands(Node, Ops.data(), Ops.size());
+  Node = (MachineSDNode*)DAG.UpdateNodeOperands(Node, Ops);
 
   // If we only got one lane, replace it with a copy
   // (if NewDmask has only one bit set...)
@@ -1311,7 +1733,7 @@ void SITargetLowering::adjustWritemask(MachineSDNode *&Node,
   }
 }
 
-/// \brief Fold the instructions after slecting them
+/// \brief Fold the instructions after selecting them.
 SDNode *SITargetLowering::PostISelFolding(MachineSDNode *Node,
                                           SelectionDAG &DAG) const {
   const SIInstrInfo *TII =
diff --git a/contrib/llvm/lib/Target/R600/SIISelLowering.h b/contrib/llvm/lib/Target/R600/SIISelLowering.h
index 9933ece..d106d4a 100644
--- a/contrib/llvm/lib/Target/R600/SIISelLowering.h
+++ b/contrib/llvm/lib/Target/R600/SIISelLowering.h
@@ -22,18 +22,22 @@ namespace llvm {
 
 class SITargetLowering : public AMDGPUTargetLowering {
   SDValue LowerParameter(SelectionDAG &DAG, EVT VT, EVT MemVT, SDLoc DL,
-                         SDValue Chain, unsigned Offset) const;
+                         SDValue Chain, unsigned Offset, bool Signed) const;
   SDValue LowerSampleIntrinsic(unsigned Opcode, const SDValue &Op,
                                SelectionDAG &DAG) const;
+  SDValue LowerGlobalAddress(AMDGPUMachineFunction *MFI, SDValue Op,
+                             SelectionDAG &DAG) const override;
+  SDValue LowerFrameIndex(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerSIGN_EXTEND(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFastFDIV(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFDIV32(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFDIV64(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFDIV(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerZERO_EXTEND(SDValue Op, SelectionDAG &DAG) const;
-  SDValue LowerADD(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerTrig(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
 
-  SDValue ResourceDescriptorToi128(SDValue Op, SelectionDAG &DAG) const;
   bool foldImm(SDValue &Operand, int32_t &Immediate,
                bool &ScalarSlotUsed) const;
   const TargetRegisterClass *getRegClassForNode(SelectionDAG &DAG,
@@ -47,31 +51,40 @@ class SITargetLowering : public AMDGPUTargetLowering {
   void adjustWritemask(MachineSDNode *&N, SelectionDAG &DAG) const;
   MachineSDNode *AdjustRegClass(MachineSDNode *N, SelectionDAG &DAG) const;
 
+  static SDValue performUCharToFloatCombine(SDNode *N,
+                                            DAGCombinerInfo &DCI);
+
 public:
   SITargetLowering(TargetMachine &tm);
-  bool allowsUnalignedMemoryAccesses(EVT  VT, bool *IsFast) const;
-  virtual bool shouldSplitVectorElementType(EVT VT) const;
+  bool allowsUnalignedMemoryAccesses(EVT VT, unsigned AS,
+                                     bool *IsFast) const override;
+
+  TargetLoweringBase::LegalizeTypeAction
+  getPreferredVectorAction(EVT VT) const override;
+
+  bool shouldConvertConstantLoadToIntImm(const APInt &Imm,
+                                        Type *Ty) const override;
 
   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 MachineBasicBlock * EmitInstrWithCustomInserter(MachineInstr * MI,
-                                              MachineBasicBlock * BB) const;
-  virtual EVT getSetCCResultType(LLVMContext &Context, EVT VT) const;
-  virtual MVT getScalarShiftAmountTy(EVT VT) const;
-  virtual bool isFMAFasterThanFMulAndFAdd(EVT VT) const;
-  virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
-  virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
-  virtual SDNode *PostISelFolding(MachineSDNode *N, SelectionDAG &DAG) const;
-  virtual void AdjustInstrPostInstrSelection(MachineInstr *MI,
-                                             SDNode *Node) const;
+  MachineBasicBlock * EmitInstrWithCustomInserter(MachineInstr * MI,
+                                      MachineBasicBlock * BB) const override;
+  EVT getSetCCResultType(LLVMContext &Context, EVT VT) const override;
+  MVT getScalarShiftAmountTy(EVT VT) const override;
+  bool isFMAFasterThanFMulAndFAdd(EVT VT) const override;
+  SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
+  SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
+  SDNode *PostISelFolding(MachineSDNode *N, SelectionDAG &DAG) const override;
+  void AdjustInstrPostInstrSelection(MachineInstr *MI,
+                                     SDNode *Node) const override;
 
   int32_t analyzeImmediate(const SDNode *N) const;
   SDValue CreateLiveInRegister(SelectionDAG &DAG, const TargetRegisterClass *RC,
-                               unsigned Reg, EVT VT) const;
+                               unsigned Reg, EVT VT) const override;
 };
 
 } // End namespace llvm
diff --git a/contrib/llvm/lib/Target/R600/SIInsertWaits.cpp b/contrib/llvm/lib/Target/R600/SIInsertWaits.cpp
index 695ec40..7dfc31b 100644
--- a/contrib/llvm/lib/Target/R600/SIInsertWaits.cpp
+++ b/contrib/llvm/lib/Target/R600/SIInsertWaits.cpp
@@ -97,13 +97,13 @@ private:
 public:
   SIInsertWaits(TargetMachine &tm) :
     MachineFunctionPass(ID),
-    TII(0),
-    TRI(0),
+    TII(nullptr),
+    TRI(nullptr),
     ExpInstrTypesSeen(0) { }
 
-  virtual bool runOnMachineFunction(MachineFunction &MF);
+  bool runOnMachineFunction(MachineFunction &MF) override;
 
-  const char *getPassName() const {
+  const char *getPassName() const override {
     return "SI insert wait  instructions";
   }
 
@@ -273,17 +273,17 @@ bool SIInsertWaits::insertWait(MachineBasicBlock &MBB,
       continue;
 
     NeedWait = true;
-    
+
     if (Ordered[i]) {
       unsigned Value = LastIssued.Array[i] - Required.Array[i];
 
-      // adjust the value to the real hardware posibilities
+      // Adjust the value to the real hardware possibilities.
       Counts.Array[i] = std::min(Value, WaitCounts.Array[i]);
 
     } else
       Counts.Array[i] = 0;
 
-    // Remember on what we have waited on
+    // Remember on what we have waited on.
     WaitedOn.Array[i] = LastIssued.Array[i] - Counts.Array[i];
   }
 
@@ -341,6 +341,8 @@ Counters SIInsertWaits::handleOperands(MachineInstr &MI) {
   return Result;
 }
 
+// FIXME: Insert waits listed in Table 4.2 "Required User-Inserted Wait States"
+// around other non-memory instructions.
 bool SIInsertWaits::runOnMachineFunction(MachineFunction &MF) {
   bool Changes = false;
 
diff --git a/contrib/llvm/lib/Target/R600/SIInstrFormats.td b/contrib/llvm/lib/Target/R600/SIInstrFormats.td
index 53ebaaf..00e69dd 100644
--- a/contrib/llvm/lib/Target/R600/SIInstrFormats.td
+++ b/contrib/llvm/lib/Target/R600/SIInstrFormats.td
@@ -12,7 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 class InstSI <dag outs, dag ins, string asm, list<dag> pattern> :
-    AMDGPUInst<outs, ins, asm, pattern> {
+    AMDGPUInst<outs, ins, asm, pattern>, PredicateControl {
 
   field bits<1> VM_CNT = 0;
   field bits<1> EXP_CNT = 0;
@@ -37,26 +37,35 @@ class InstSI <dag outs, dag ins, string asm, list<dag> pattern> :
   let TSFlags{9} = SALU;
 }
 
-class Enc32 <dag outs, dag ins, string asm, list<dag> pattern> :
-    InstSI <outs, ins, asm, pattern> {
+class Enc32 {
 
   field bits<32> Inst;
-  let Size = 4;
+  int Size = 4;
 }
 
-class Enc64 <dag outs, dag ins, string asm, list<dag> pattern> :
-    InstSI <outs, ins, asm, pattern> {
+class Enc64 {
 
   field bits<64> Inst;
-  let Size = 8;
+  int Size = 8;
+}
+
+class VOP3Common <dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI <outs, ins, asm, pattern> {
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+  let UseNamedOperandTable = 1;
+  let VOP3 = 1;
+
+  int Size = 8;
 }
 
 //===----------------------------------------------------------------------===//
 // Scalar operations
 //===----------------------------------------------------------------------===//
 
-class SOP1 <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc32<outs, ins, asm, pattern> {
+class SOP1e <bits<8> op> : Enc32 {
 
   bits<7> SDST;
   bits<8> SSRC0;
@@ -65,16 +74,10 @@ class SOP1 <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
   let Inst{15-8} = op;
   let Inst{22-16} = SDST;
   let Inst{31-23} = 0x17d; //encoding;
-
-  let mayLoad = 0;
-  let mayStore = 0;
-  let hasSideEffects = 0;
-  let SALU = 1;
 }
 
-class SOP2 <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc32 <outs, ins, asm, pattern> {
-  
+class SOP2e <bits<7> op> : Enc32 {
+
   bits<7> SDST;
   bits<8> SSRC0;
   bits<8> SSRC1;
@@ -84,15 +87,9 @@ class SOP2 <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
   let Inst{22-16} = SDST;
   let Inst{29-23} = op;
   let Inst{31-30} = 0x2; // encoding
-
-  let mayLoad = 0;
-  let mayStore = 0;
-  let hasSideEffects = 0;
-  let SALU = 1;
 }
 
-class SOPC <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
-  Enc32<outs, ins, asm, pattern> {
+class SOPCe <bits<7> op> : Enc32 {
 
   bits<8> SSRC0;
   bits<8> SSRC1;
@@ -101,62 +98,90 @@ class SOPC <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
   let Inst{15-8} = SSRC1;
   let Inst{22-16} = op;
   let Inst{31-23} = 0x17e;
-
-  let DisableEncoding = "$dst";
-  let mayLoad = 0;
-  let mayStore = 0;
-  let hasSideEffects = 0;
-  let SALU = 1;
 }
 
-class SOPK <bits<5> op, dag outs, dag ins, string asm, list<dag> pattern> :
-   Enc32 <outs, ins , asm, pattern> {
+class SOPKe <bits<5> op> : Enc32 {
 
   bits <7> SDST;
   bits <16> SIMM16;
-  
+
   let Inst{15-0} = SIMM16;
   let Inst{22-16} = SDST;
   let Inst{27-23} = op;
   let Inst{31-28} = 0xb; //encoding
-
-  let mayLoad = 0;
-  let mayStore = 0;
-  let hasSideEffects = 0;
-  let SALU = 1;
 }
 
-class SOPP <bits<7> op, dag ins, string asm, list<dag> pattern> : Enc32 <
-  (outs),
-  ins,
-  asm,
-  pattern > {
+class SOPPe <bits<7> op> : Enc32 {
 
-  bits <16> SIMM16;
+  bits <16> simm16;
 
-  let Inst{15-0} = SIMM16;
+  let Inst{15-0} = simm16;
   let Inst{22-16} = op;
   let Inst{31-23} = 0x17f; // encoding
-
-  let mayLoad = 0;
-  let mayStore = 0;
-  let hasSideEffects = 0;
-  let SALU = 1;
 }
 
-class SMRD <bits<5> op, bits<1> imm, dag outs, dag ins, string asm,
-            list<dag> pattern> : Enc32<outs, ins, asm, pattern> {
+class SMRDe <bits<5> op, bits<1> imm> : Enc32 {
 
   bits<7> SDST;
   bits<7> SBASE;
   bits<8> OFFSET;
-  
+
   let Inst{7-0} = OFFSET;
   let Inst{8} = imm;
   let Inst{14-9} = SBASE{6-1};
   let Inst{21-15} = SDST;
   let Inst{26-22} = op;
   let Inst{31-27} = 0x18; //encoding
+}
+
+class SOP1 <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI<outs, ins, asm, pattern>, SOP1e <op> {
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+  let SALU = 1;
+}
+
+class SOP2 <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI <outs, ins, asm, pattern>, SOP2e<op> {
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+  let SALU = 1;
+}
+
+class SOPC <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
+  InstSI<outs, ins, asm, pattern>, SOPCe <op> {
+
+  let DisableEncoding = "$dst";
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+  let SALU = 1;
+}
+
+class SOPK <bits<5> op, dag outs, dag ins, string asm, list<dag> pattern> :
+   InstSI <outs, ins , asm, pattern>, SOPKe<op> {
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+  let SALU = 1;
+}
+
+class SOPP <bits<7> op, dag ins, string asm, list<dag> pattern> :
+		InstSI <(outs), ins, asm, pattern >, SOPPe <op> {
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+  let SALU = 1;
+}
+
+class SMRD <bits<5> op, bits<1> imm, dag outs, dag ins, string asm,
+            list<dag> pattern> : InstSI<outs, ins, asm, pattern>, SMRDe<op, imm> {
 
   let LGKM_CNT = 1;
   let SMRD = 1;
@@ -165,61 +190,47 @@ class SMRD <bits<5> op, bits<1> imm, dag outs, dag ins, string asm,
 //===----------------------------------------------------------------------===//
 // Vector ALU operations
 //===----------------------------------------------------------------------===//
-    
-let Uses = [EXEC] in {
 
-class VOP1 <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc32 <outs, ins, asm, pattern> {
+class VOP1e <bits<8> op> : Enc32 {
 
   bits<8> VDST;
   bits<9> SRC0;
-  
+
   let Inst{8-0} = SRC0;
   let Inst{16-9} = op;
   let Inst{24-17} = VDST;
   let Inst{31-25} = 0x3f; //encoding
-  
-  let mayLoad = 0;
-  let mayStore = 0;
-  let hasSideEffects = 0;
-  let UseNamedOperandTable = 1;
-  let VOP1 = 1;
 }
 
-class VOP2 <bits<6> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc32 <outs, ins, asm, pattern> {
+class VOP2e <bits<6> op> : Enc32 {
 
   bits<8> VDST;
   bits<9> SRC0;
   bits<8> VSRC1;
-  
+
   let Inst{8-0} = SRC0;
   let Inst{16-9} = VSRC1;
   let Inst{24-17} = VDST;
   let Inst{30-25} = op;
   let Inst{31} = 0x0; //encoding
-  
-  let mayLoad = 0;
-  let mayStore = 0;
-  let hasSideEffects = 0;
-  let UseNamedOperandTable = 1;
-  let VOP2 = 1;
 }
 
-class VOP3 <bits<9> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc64 <outs, ins, asm, pattern> {
+class VOP3e <bits<9> op> : Enc64 {
 
   bits<8> dst;
+  bits<2> src0_modifiers;
   bits<9> src0;
+  bits<2> src1_modifiers;
   bits<9> src1;
+  bits<2> src2_modifiers;
   bits<9> src2;
-  bits<3> abs;
   bits<1> clamp;
   bits<2> omod;
-  bits<3> neg;
 
   let Inst{7-0} = dst;
-  let Inst{10-8} = abs;
+  let Inst{8} = src0_modifiers{1};
+  let Inst{9} = src1_modifiers{1};
+  let Inst{10} = src2_modifiers{1};
   let Inst{11} = clamp;
   let Inst{25-17} = op;
   let Inst{31-26} = 0x34; //encoding
@@ -227,25 +238,22 @@ class VOP3 <bits<9> op, dag outs, dag ins, string asm, list<dag> pattern> :
   let Inst{49-41} = src1;
   let Inst{58-50} = src2;
   let Inst{60-59} = omod;
-  let Inst{63-61} = neg;
-  
-  let mayLoad = 0;
-  let mayStore = 0;
-  let hasSideEffects = 0;
-  let UseNamedOperandTable = 1;
-  let VOP3 = 1;
+  let Inst{61} = src0_modifiers{0};
+  let Inst{62} = src1_modifiers{0};
+  let Inst{63} = src2_modifiers{0};
 }
 
-class VOP3b <bits<9> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc64 <outs, ins, asm, pattern> {
+class VOP3be <bits<9> op> : Enc64 {
 
   bits<8> dst;
+  bits<2> src0_modifiers;
   bits<9> src0;
+  bits<2> src1_modifiers;
   bits<9> src1;
+  bits<2> src2_modifiers;
   bits<9> src2;
   bits<7> sdst;
   bits<2> omod;
-  bits<3> neg;
 
   let Inst{7-0} = dst;
   let Inst{14-8} = sdst;
@@ -255,17 +263,12 @@ class VOP3b <bits<9> op, dag outs, dag ins, string asm, list<dag> pattern> :
   let Inst{49-41} = src1;
   let Inst{58-50} = src2;
   let Inst{60-59} = omod;
-  let Inst{63-61} = neg;
-
-  let mayLoad = 0;
-  let mayStore = 0;
-  let hasSideEffects = 0;
-  let UseNamedOperandTable = 1;
-  let VOP3 = 1;
+  let Inst{61} = src0_modifiers{0};
+  let Inst{62} = src1_modifiers{0};
+  let Inst{63} = src2_modifiers{0};
 }
 
-class VOPC <bits<8> op, dag ins, string asm, list<dag> pattern> :
-    Enc32 <(outs VCCReg:$dst), ins, asm, pattern> {
+class VOPCe <bits<8> op> : Enc32 {
 
   bits<9> SRC0;
   bits<8> VSRC1;
@@ -274,16 +277,9 @@ class VOPC <bits<8> op, dag ins, string asm, list<dag> pattern> :
   let Inst{16-9} = VSRC1;
   let Inst{24-17} = op;
   let Inst{31-25} = 0x3e;
- 
-  let DisableEncoding = "$dst";
-  let mayLoad = 0;
-  let mayStore = 0;
-  let hasSideEffects = 0;
-  let VOPC = 1;
 }
 
-class VINTRP <bits <2> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc32 <outs, ins, asm, pattern> {
+class VINTRPe <bits<2> op> : Enc32 {
 
   bits<8> VDST;
   bits<8> VSRC;
@@ -296,22 +292,9 @@ class VINTRP <bits <2> op, dag outs, dag ins, string asm, list<dag> pattern> :
   let Inst{17-16} = op;
   let Inst{25-18} = VDST;
   let Inst{31-26} = 0x32; // encoding
-
-  let neverHasSideEffects = 1;
-  let mayLoad = 1;
-  let mayStore = 0;
 }
 
-} // End Uses = [EXEC]
-
-//===----------------------------------------------------------------------===//
-// Vector I/O operations
-//===----------------------------------------------------------------------===//
-
-let Uses = [EXEC] in {
-
-class DS <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc64 <outs, ins, asm, pattern> {
+class DSe <bits<8> op> : Enc64 {
 
   bits<8> vdst;
   bits<1> gds;
@@ -330,12 +313,9 @@ class DS <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
   let Inst{47-40} = data0;
   let Inst{55-48} = data1;
   let Inst{63-56} = vdst;
-
-  let LGKM_CNT = 1;
 }
 
-class MUBUF <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc64<outs, ins, asm, pattern> {
+class MUBUFe <bits<7> op> : Enc64 {
 
   bits<12> offset;
   bits<1> offen;
@@ -364,15 +344,9 @@ class MUBUF <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
   let Inst{54} = slc;
   let Inst{55} = tfe;
   let Inst{63-56} = soffset;
-
-  let VM_CNT = 1;
-  let EXP_CNT = 1;
-
-  let neverHasSideEffects = 1;
 }
 
-class MTBUF <bits<3> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc64<outs, ins, asm, pattern> {
+class MTBUFe <bits<3> op> : Enc64 {
 
   bits<8> VDATA;
   bits<12> OFFSET;
@@ -403,15 +377,9 @@ class MTBUF <bits<3> op, dag outs, dag ins, string asm, list<dag> pattern> :
   let Inst{54} = SLC;
   let Inst{55} = TFE;
   let Inst{63-56} = SOFFSET;
-
-  let VM_CNT = 1;
-  let EXP_CNT = 1;
-
-  let neverHasSideEffects = 1;
 }
 
-class MIMG <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
-    Enc64 <outs, ins, asm, pattern> {
+class MIMGe <bits<7> op> : Enc64 {
 
   bits<8> VDATA;
   bits<4> DMASK;
@@ -424,7 +392,7 @@ class MIMG <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
   bits<1> SLC;
   bits<8> VADDR;
   bits<7> SRSRC;
-  bits<7> SSAMP; 
+  bits<7> SSAMP;
 
   let Inst{11-8} = DMASK;
   let Inst{12} = UNORM;
@@ -440,18 +408,9 @@ class MIMG <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
   let Inst{47-40} = VDATA;
   let Inst{52-48} = SRSRC{6-2};
   let Inst{57-53} = SSAMP{6-2};
-
-  let VM_CNT = 1;
-  let EXP_CNT = 1;
-  let MIMG = 1;
 }
 
-def EXP : Enc64<
-  (outs),
-  (ins i32imm:$en, i32imm:$tgt, i32imm:$compr, i32imm:$done, i32imm:$vm,
-       VReg_32:$src0, VReg_32:$src1, VReg_32:$src2, VReg_32:$src3),
-  "EXP $en, $tgt, $compr, $done, $vm, $src0, $src1, $src2, $src3",
-  [] > {
+class EXPe : Enc64 {
 
   bits<4> EN;
   bits<6> TGT;
@@ -473,6 +432,102 @@ def EXP : Enc64<
   let Inst{47-40} = VSRC1;
   let Inst{55-48} = VSRC2;
   let Inst{63-56} = VSRC3;
+}
+
+let Uses = [EXEC] in {
+
+class VOP1 <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI <outs, ins, asm, pattern>, VOP1e<op> {
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+  let UseNamedOperandTable = 1;
+  let VOP1 = 1;
+}
+
+class VOP2 <bits<6> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI <outs, ins, asm, pattern>, VOP2e<op> {
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+  let UseNamedOperandTable = 1;
+  let VOP2 = 1;
+}
+
+class VOP3 <bits<9> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    VOP3Common <outs, ins, asm, pattern>, VOP3e<op>;
+
+class VOP3b <bits<9> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    VOP3Common <outs, ins, asm, pattern>, VOP3be<op>;
+
+class VOPC <bits<8> op, dag ins, string asm, list<dag> pattern> :
+    InstSI <(outs VCCReg:$dst), ins, asm, pattern>, VOPCe <op> {
+
+  let DisableEncoding = "$dst";
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+  let UseNamedOperandTable = 1;
+  let VOPC = 1;
+}
+
+class VINTRP <bits <2> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI <outs, ins, asm, pattern>, VINTRPe<op> {
+
+  let neverHasSideEffects = 1;
+  let mayLoad = 1;
+  let mayStore = 0;
+}
+
+} // End Uses = [EXEC]
+
+//===----------------------------------------------------------------------===//
+// Vector I/O operations
+//===----------------------------------------------------------------------===//
+
+let Uses = [EXEC] in {
+
+class DS <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI <outs, ins, asm, pattern> , DSe<op> {
+
+  let LGKM_CNT = 1;
+}
+
+class MUBUF <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI<outs, ins, asm, pattern>, MUBUFe <op> {
+
+  let VM_CNT = 1;
+  let EXP_CNT = 1;
+
+  let neverHasSideEffects = 1;
+  let UseNamedOperandTable = 1;
+}
+
+class MTBUF <bits<3> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI<outs, ins, asm, pattern>, MTBUFe <op> {
+
+  let VM_CNT = 1;
+  let EXP_CNT = 1;
+
+  let neverHasSideEffects = 1;
+}
+
+class MIMG <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI <outs, ins, asm, pattern>, MIMGe <op> {
+
+  let VM_CNT = 1;
+  let EXP_CNT = 1;
+  let MIMG = 1;
+}
+
+def EXP : InstSI<
+  (outs),
+  (ins i32imm:$en, i32imm:$tgt, i32imm:$compr, i32imm:$done, i32imm:$vm,
+       VReg_32:$src0, VReg_32:$src1, VReg_32:$src2, VReg_32:$src3),
+  "EXP $en, $tgt, $compr, $done, $vm, $src0, $src1, $src2, $src3",
+  [] >, EXPe {
 
   let EXP_CNT = 1;
 }
diff --git a/contrib/llvm/lib/Target/R600/SIInstrInfo.cpp b/contrib/llvm/lib/Target/R600/SIInstrInfo.cpp
index ab55c1b..51f4532 100644
--- a/contrib/llvm/lib/Target/R600/SIInstrInfo.cpp
+++ b/contrib/llvm/lib/Target/R600/SIInstrInfo.cpp
@@ -16,20 +16,17 @@
 #include "SIInstrInfo.h"
 #include "AMDGPUTargetMachine.h"
 #include "SIDefines.h"
+#include "SIMachineFunctionInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/Function.h"
 #include "llvm/MC/MCInstrDesc.h"
 
 using namespace llvm;
 
-SIInstrInfo::SIInstrInfo(AMDGPUTargetMachine &tm)
-  : AMDGPUInstrInfo(tm),
-    RI(tm)
-    { }
-
-const SIRegisterInfo &SIInstrInfo::getRegisterInfo() const {
-  return RI;
-}
+SIInstrInfo::SIInstrInfo(const AMDGPUSubtarget &st)
+  : AMDGPUInstrInfo(st),
+    RI(st) { }
 
 //===----------------------------------------------------------------------===//
 // TargetInstrInfo callbacks
@@ -185,23 +182,226 @@ unsigned SIInstrInfo::commuteOpcode(unsigned Opcode) const {
   return Opcode;
 }
 
+void SIInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
+                                      MachineBasicBlock::iterator MI,
+                                      unsigned SrcReg, bool isKill,
+                                      int FrameIndex,
+                                      const TargetRegisterClass *RC,
+                                      const TargetRegisterInfo *TRI) const {
+  MachineFunction *MF = MBB.getParent();
+  SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
+  MachineRegisterInfo &MRI = MF->getRegInfo();
+  DebugLoc DL = MBB.findDebugLoc(MI);
+  unsigned KillFlag = isKill ? RegState::Kill : 0;
+
+  if (RI.hasVGPRs(RC)) {
+    LLVMContext &Ctx = MF->getFunction()->getContext();
+    Ctx.emitError("SIInstrInfo::storeRegToStackSlot - Can't spill VGPR!");
+    BuildMI(MBB, MI, DL, get(AMDGPU::V_MOV_B32_e32), AMDGPU::VGPR0)
+            .addReg(SrcReg);
+  } else if (TRI->getCommonSubClass(RC, &AMDGPU::SGPR_32RegClass)) {
+    unsigned Lane = MFI->SpillTracker.reserveLanes(MRI, MF);
+    unsigned TgtReg = MFI->SpillTracker.LaneVGPR;
+
+    BuildMI(MBB, MI, DL, get(AMDGPU::V_WRITELANE_B32), TgtReg)
+            .addReg(SrcReg, KillFlag)
+            .addImm(Lane);
+    MFI->SpillTracker.addSpilledReg(FrameIndex, TgtReg, Lane);
+  } else if (RI.isSGPRClass(RC)) {
+    // We are only allowed to create one new instruction when spilling
+    // registers, so we need to use pseudo instruction for vector
+    // registers.
+    //
+    // Reserve a spot in the spill tracker for each sub-register of
+    // the vector register.
+    unsigned NumSubRegs = RC->getSize() / 4;
+    unsigned FirstLane = MFI->SpillTracker.reserveLanes(MRI, MF, NumSubRegs);
+    MFI->SpillTracker.addSpilledReg(FrameIndex, MFI->SpillTracker.LaneVGPR,
+                                    FirstLane);
+
+    unsigned Opcode;
+    switch (RC->getSize() * 8) {
+    case 64:  Opcode = AMDGPU::SI_SPILL_S64_SAVE;  break;
+    case 128: Opcode = AMDGPU::SI_SPILL_S128_SAVE; break;
+    case 256: Opcode = AMDGPU::SI_SPILL_S256_SAVE; break;
+    case 512: Opcode = AMDGPU::SI_SPILL_S512_SAVE; break;
+    default: llvm_unreachable("Cannot spill register class");
+    }
+
+    BuildMI(MBB, MI, DL, get(Opcode), MFI->SpillTracker.LaneVGPR)
+            .addReg(SrcReg)
+            .addImm(FrameIndex);
+  } else {
+    llvm_unreachable("VGPR spilling not supported");
+  }
+}
+
+void SIInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
+                                       MachineBasicBlock::iterator MI,
+                                       unsigned DestReg, int FrameIndex,
+                                       const TargetRegisterClass *RC,
+                                       const TargetRegisterInfo *TRI) const {
+  MachineFunction *MF = MBB.getParent();
+  SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
+  DebugLoc DL = MBB.findDebugLoc(MI);
+
+  if (RI.hasVGPRs(RC)) {
+    LLVMContext &Ctx = MF->getFunction()->getContext();
+    Ctx.emitError("SIInstrInfo::loadRegToStackSlot - Can't retrieve spilled VGPR!");
+    BuildMI(MBB, MI, DL, get(AMDGPU::V_MOV_B32_e32), DestReg)
+            .addImm(0);
+  } else if (RI.isSGPRClass(RC)){
+    unsigned Opcode;
+    switch(RC->getSize() * 8) {
+    case 32:  Opcode = AMDGPU::SI_SPILL_S32_RESTORE; break;
+    case 64:  Opcode = AMDGPU::SI_SPILL_S64_RESTORE;  break;
+    case 128: Opcode = AMDGPU::SI_SPILL_S128_RESTORE; break;
+    case 256: Opcode = AMDGPU::SI_SPILL_S256_RESTORE; break;
+    case 512: Opcode = AMDGPU::SI_SPILL_S512_RESTORE; break;
+    default: llvm_unreachable("Cannot spill register class");
+    }
+
+    SIMachineFunctionInfo::SpilledReg Spill =
+        MFI->SpillTracker.getSpilledReg(FrameIndex);
+
+    BuildMI(MBB, MI, DL, get(Opcode), DestReg)
+            .addReg(Spill.VGPR)
+            .addImm(FrameIndex);
+  } else {
+    llvm_unreachable("VGPR spilling not supported");
+  }
+}
+
+static unsigned getNumSubRegsForSpillOp(unsigned Op) {
+
+  switch (Op) {
+  case AMDGPU::SI_SPILL_S512_SAVE:
+  case AMDGPU::SI_SPILL_S512_RESTORE:
+    return 16;
+  case AMDGPU::SI_SPILL_S256_SAVE:
+  case AMDGPU::SI_SPILL_S256_RESTORE:
+    return 8;
+  case AMDGPU::SI_SPILL_S128_SAVE:
+  case AMDGPU::SI_SPILL_S128_RESTORE:
+    return 4;
+  case AMDGPU::SI_SPILL_S64_SAVE:
+  case AMDGPU::SI_SPILL_S64_RESTORE:
+    return 2;
+  case AMDGPU::SI_SPILL_S32_RESTORE:
+    return 1;
+  default: llvm_unreachable("Invalid spill opcode");
+  }
+}
+
+void SIInstrInfo::insertNOPs(MachineBasicBlock::iterator MI,
+                             int Count) const {
+  while (Count > 0) {
+    int Arg;
+    if (Count >= 8)
+      Arg = 7;
+    else
+      Arg = Count - 1;
+    Count -= 8;
+    BuildMI(*MI->getParent(), MI, MI->getDebugLoc(), get(AMDGPU::S_NOP))
+            .addImm(Arg);
+  }
+}
+
+bool SIInstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
+  SIMachineFunctionInfo *MFI =
+      MI->getParent()->getParent()->getInfo<SIMachineFunctionInfo>();
+  MachineBasicBlock &MBB = *MI->getParent();
+  DebugLoc DL = MBB.findDebugLoc(MI);
+  switch (MI->getOpcode()) {
+  default: return AMDGPUInstrInfo::expandPostRAPseudo(MI);
+
+  // SGPR register spill
+  case AMDGPU::SI_SPILL_S512_SAVE:
+  case AMDGPU::SI_SPILL_S256_SAVE:
+  case AMDGPU::SI_SPILL_S128_SAVE:
+  case AMDGPU::SI_SPILL_S64_SAVE: {
+    unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
+    unsigned FrameIndex = MI->getOperand(2).getImm();
+
+    for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
+      SIMachineFunctionInfo::SpilledReg Spill;
+      unsigned SubReg = RI.getPhysRegSubReg(MI->getOperand(1).getReg(),
+                                            &AMDGPU::SGPR_32RegClass, i);
+      Spill = MFI->SpillTracker.getSpilledReg(FrameIndex);
+
+      BuildMI(MBB, MI, DL, get(AMDGPU::V_WRITELANE_B32),
+              MI->getOperand(0).getReg())
+              .addReg(SubReg)
+              .addImm(Spill.Lane + i);
+    }
+    MI->eraseFromParent();
+    break;
+  }
+
+  // SGPR register restore
+  case AMDGPU::SI_SPILL_S512_RESTORE:
+  case AMDGPU::SI_SPILL_S256_RESTORE:
+  case AMDGPU::SI_SPILL_S128_RESTORE:
+  case AMDGPU::SI_SPILL_S64_RESTORE:
+  case AMDGPU::SI_SPILL_S32_RESTORE: {
+    unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
+
+    for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
+      SIMachineFunctionInfo::SpilledReg Spill;
+      unsigned FrameIndex = MI->getOperand(2).getImm();
+      unsigned SubReg = RI.getPhysRegSubReg(MI->getOperand(0).getReg(),
+                                   &AMDGPU::SGPR_32RegClass, i);
+      Spill = MFI->SpillTracker.getSpilledReg(FrameIndex);
+
+      BuildMI(MBB, MI, DL, get(AMDGPU::V_READLANE_B32), SubReg)
+              .addReg(MI->getOperand(1).getReg())
+              .addImm(Spill.Lane + i);
+    }
+    insertNOPs(MI, 3);
+    MI->eraseFromParent();
+    break;
+  }
+  case AMDGPU::SI_CONSTDATA_PTR: {
+    unsigned Reg = MI->getOperand(0).getReg();
+    unsigned RegLo = RI.getSubReg(Reg, AMDGPU::sub0);
+    unsigned RegHi = RI.getSubReg(Reg, AMDGPU::sub1);
+
+    BuildMI(MBB, MI, DL, get(AMDGPU::S_GETPC_B64), Reg);
+
+    // Add 32-bit offset from this instruction to the start of the constant data.
+    BuildMI(MBB, MI, DL, get(AMDGPU::S_ADD_I32), RegLo)
+            .addReg(RegLo)
+            .addTargetIndex(AMDGPU::TI_CONSTDATA_START)
+            .addReg(AMDGPU::SCC, RegState::Define | RegState::Implicit);
+    BuildMI(MBB, MI, DL, get(AMDGPU::S_ADDC_U32), RegHi)
+            .addReg(RegHi)
+            .addImm(0)
+            .addReg(AMDGPU::SCC, RegState::Define | RegState::Implicit)
+            .addReg(AMDGPU::SCC, RegState::Implicit);
+    MI->eraseFromParent();
+    break;
+  }
+  }
+  return true;
+}
+
 MachineInstr *SIInstrInfo::commuteInstruction(MachineInstr *MI,
                                               bool NewMI) const {
 
   MachineRegisterInfo &MRI = MI->getParent()->getParent()->getRegInfo();
   if (MI->getNumOperands() < 3 || !MI->getOperand(1).isReg())
-    return 0;
+    return nullptr;
 
   // Cannot commute VOP2 if src0 is SGPR.
   if (isVOP2(MI->getOpcode()) && MI->getOperand(1).isReg() &&
       RI.isSGPRClass(MRI.getRegClass(MI->getOperand(1).getReg())))
-   return 0;
+   return nullptr;
 
   if (!MI->getOperand(2).isReg()) {
     // XXX: Commute instructions with FPImm operands
     if (NewMI || MI->getOperand(2).isFPImm() ||
        (!isVOP2(MI->getOpcode()) && !isVOP3(MI->getOpcode()))) {
-      return 0;
+      return nullptr;
     }
 
     // XXX: Commute VOP3 instructions with abs and neg set.
@@ -210,11 +410,13 @@ MachineInstr *SIInstrInfo::commuteInstruction(MachineInstr *MI,
                         AMDGPU::OpName::abs)).getImm() ||
          MI->getOperand(AMDGPU::getNamedOperandIdx(MI->getOpcode(),
                         AMDGPU::OpName::neg)).getImm()))
-      return 0;
+      return nullptr;
 
     unsigned Reg = MI->getOperand(1).getReg();
+    unsigned SubReg = MI->getOperand(1).getSubReg();
     MI->getOperand(1).ChangeToImmediate(MI->getOperand(2).getImm());
     MI->getOperand(2).ChangeToRegister(Reg, false);
+    MI->getOperand(2).setSubReg(SubReg);
   } else {
     MI = TargetInstrInfo::commuteInstruction(MI, NewMI);
   }
@@ -249,6 +451,30 @@ SIInstrInfo::isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const {
   return RC != &AMDGPU::EXECRegRegClass;
 }
 
+bool
+SIInstrInfo::isTriviallyReMaterializable(const MachineInstr *MI,
+                                         AliasAnalysis *AA) const {
+  switch(MI->getOpcode()) {
+  default: return AMDGPUInstrInfo::isTriviallyReMaterializable(MI, AA);
+  case AMDGPU::S_MOV_B32:
+  case AMDGPU::S_MOV_B64:
+  case AMDGPU::V_MOV_B32_e32:
+    return MI->getOperand(1).isImm();
+  }
+}
+
+namespace llvm {
+namespace AMDGPU {
+// Helper function generated by tablegen.  We are wrapping this with
+// an SIInstrInfo function that reutrns bool rather than int.
+int isDS(uint16_t Opcode);
+}
+}
+
+bool SIInstrInfo::isDS(uint16_t Opcode) const {
+  return ::AMDGPU::isDS(Opcode) != -1;
+}
+
 int SIInstrInfo::isMIMG(uint16_t Opcode) const {
   return get(Opcode).TSFlags & SIInstrFlags::MIMG;
 }
@@ -277,21 +503,40 @@ bool SIInstrInfo::isSALUInstr(const MachineInstr &MI) const {
   return get(MI.getOpcode()).TSFlags & SIInstrFlags::SALU;
 }
 
+bool SIInstrInfo::isInlineConstant(const APInt &Imm) const {
+  int32_t Val = Imm.getSExtValue();
+  if (Val >= -16 && Val <= 64)
+    return true;
+
+  // The actual type of the operand does not seem to matter as long
+  // as the bits match one of the inline immediate values.  For example:
+  //
+  // -nan has the hexadecimal encoding of 0xfffffffe which is -2 in decimal,
+  // so it is a legal inline immediate.
+  //
+  // 1065353216 has the hexadecimal encoding 0x3f800000 which is 1.0f in
+  // floating-point, so it is a legal inline immediate.
+
+  return (APInt::floatToBits(0.0f) == Imm) ||
+         (APInt::floatToBits(1.0f) == Imm) ||
+         (APInt::floatToBits(-1.0f) == Imm) ||
+         (APInt::floatToBits(0.5f) == Imm) ||
+         (APInt::floatToBits(-0.5f) == Imm) ||
+         (APInt::floatToBits(2.0f) == Imm) ||
+         (APInt::floatToBits(-2.0f) == Imm) ||
+         (APInt::floatToBits(4.0f) == Imm) ||
+         (APInt::floatToBits(-4.0f) == Imm);
+}
+
 bool SIInstrInfo::isInlineConstant(const MachineOperand &MO) const {
-  if(MO.isImm()) {
-    return MO.getImm() >= -16 && MO.getImm() <= 64;
-  }
+  if (MO.isImm())
+    return isInlineConstant(APInt(32, MO.getImm(), true));
+
   if (MO.isFPImm()) {
-    return MO.getFPImm()->isExactlyValue(0.0)  ||
-           MO.getFPImm()->isExactlyValue(0.5)  ||
-           MO.getFPImm()->isExactlyValue(-0.5) ||
-           MO.getFPImm()->isExactlyValue(1.0)  ||
-           MO.getFPImm()->isExactlyValue(-1.0) ||
-           MO.getFPImm()->isExactlyValue(2.0)  ||
-           MO.getFPImm()->isExactlyValue(-2.0) ||
-           MO.getFPImm()->isExactlyValue(4.0)  ||
-           MO.getFPImm()->isExactlyValue(-4.0);
+    APFloat FpImm = MO.getFPImm()->getValueAPF();
+    return isInlineConstant(FpImm.bitcastToAPInt());
   }
+
   return false;
 }
 
@@ -299,6 +544,42 @@ bool SIInstrInfo::isLiteralConstant(const MachineOperand &MO) const {
   return (MO.isImm() || MO.isFPImm()) && !isInlineConstant(MO);
 }
 
+static bool compareMachineOp(const MachineOperand &Op0,
+                             const MachineOperand &Op1) {
+  if (Op0.getType() != Op1.getType())
+    return false;
+
+  switch (Op0.getType()) {
+  case MachineOperand::MO_Register:
+    return Op0.getReg() == Op1.getReg();
+  case MachineOperand::MO_Immediate:
+    return Op0.getImm() == Op1.getImm();
+  case MachineOperand::MO_FPImmediate:
+    return Op0.getFPImm() == Op1.getFPImm();
+  default:
+    llvm_unreachable("Didn't expect to be comparing these operand types");
+  }
+}
+
+bool SIInstrInfo::isImmOperandLegal(const MachineInstr *MI, unsigned OpNo,
+                                 const MachineOperand &MO) const {
+  const MCOperandInfo &OpInfo = get(MI->getOpcode()).OpInfo[OpNo];
+
+  assert(MO.isImm() || MO.isFPImm());
+
+  if (OpInfo.OperandType == MCOI::OPERAND_IMMEDIATE)
+    return true;
+
+  if (OpInfo.RegClass < 0)
+    return false;
+
+  return RI.regClassCanUseImmediate(OpInfo.RegClass);
+}
+
+bool SIInstrInfo::hasVALU32BitEncoding(unsigned Opcode) const {
+  return AMDGPU::getVOPe32(Opcode) != -1;
+}
+
 bool SIInstrInfo::verifyInstruction(const MachineInstr *MI,
                                     StringRef &ErrInfo) const {
   uint16_t Opcode = MI->getOpcode();
@@ -306,6 +587,58 @@ bool SIInstrInfo::verifyInstruction(const MachineInstr *MI,
   int Src1Idx = AMDGPU::getNamedOperandIdx(Opcode, AMDGPU::OpName::src1);
   int Src2Idx = AMDGPU::getNamedOperandIdx(Opcode, AMDGPU::OpName::src2);
 
+  // Make sure the number of operands is correct.
+  const MCInstrDesc &Desc = get(Opcode);
+  if (!Desc.isVariadic() &&
+      Desc.getNumOperands() != MI->getNumExplicitOperands()) {
+     ErrInfo = "Instruction has wrong number of operands.";
+     return false;
+  }
+
+  // Make sure the register classes are correct
+  for (unsigned i = 0, e = Desc.getNumOperands(); i != e; ++i) {
+    switch (Desc.OpInfo[i].OperandType) {
+    case MCOI::OPERAND_REGISTER: {
+      int RegClass = Desc.OpInfo[i].RegClass;
+      if (!RI.regClassCanUseImmediate(RegClass) &&
+          (MI->getOperand(i).isImm() || MI->getOperand(i).isFPImm())) {
+        ErrInfo = "Expected register, but got immediate";
+        return false;
+      }
+    }
+      break;
+    case MCOI::OPERAND_IMMEDIATE:
+      // Check if this operand is an immediate.
+      // FrameIndex operands will be replaced by immediates, so they are
+      // allowed.
+      if (!MI->getOperand(i).isImm() && !MI->getOperand(i).isFPImm() &&
+          !MI->getOperand(i).isFI()) {
+        ErrInfo = "Expected immediate, but got non-immediate";
+        return false;
+      }
+      // Fall-through
+    default:
+      continue;
+    }
+
+    if (!MI->getOperand(i).isReg())
+      continue;
+
+    int RegClass = Desc.OpInfo[i].RegClass;
+    if (RegClass != -1) {
+      unsigned Reg = MI->getOperand(i).getReg();
+      if (TargetRegisterInfo::isVirtualRegister(Reg))
+        continue;
+
+      const TargetRegisterClass *RC = RI.getRegClass(RegClass);
+      if (!RC->contains(Reg)) {
+        ErrInfo = "Operand has incorrect register class.";
+        return false;
+      }
+    }
+  }
+
+
   // Verify VOP*
   if (isVOP1(Opcode) || isVOP2(Opcode) || isVOP3(Opcode) || isVOPC(Opcode)) {
     unsigned ConstantBusCount = 0;
@@ -364,6 +697,24 @@ bool SIInstrInfo::verifyInstruction(const MachineInstr *MI,
       return false;
     }
   }
+
+  // Verify misc. restrictions on specific instructions.
+  if (Desc.getOpcode() == AMDGPU::V_DIV_SCALE_F32 ||
+      Desc.getOpcode() == AMDGPU::V_DIV_SCALE_F64) {
+    MI->dump();
+
+    const MachineOperand &Src0 = MI->getOperand(2);
+    const MachineOperand &Src1 = MI->getOperand(3);
+    const MachineOperand &Src2 = MI->getOperand(4);
+    if (Src0.isReg() && Src1.isReg() && Src2.isReg()) {
+      if (!compareMachineOp(Src0, Src1) &&
+          !compareMachineOp(Src0, Src2)) {
+        ErrInfo = "v_div_scale_{f32|f64} require src0 = src1 or src2";
+        return false;
+      }
+    }
+  }
+
   return true;
 }
 
@@ -373,16 +724,49 @@ unsigned SIInstrInfo::getVALUOp(const MachineInstr &MI) {
   case AMDGPU::REG_SEQUENCE: return AMDGPU::REG_SEQUENCE;
   case AMDGPU::COPY: return AMDGPU::COPY;
   case AMDGPU::PHI: return AMDGPU::PHI;
+  case AMDGPU::INSERT_SUBREG: return AMDGPU::INSERT_SUBREG;
+  case AMDGPU::S_MOV_B32:
+    return MI.getOperand(1).isReg() ?
+           AMDGPU::COPY : AMDGPU::V_MOV_B32_e32;
   case AMDGPU::S_ADD_I32: return AMDGPU::V_ADD_I32_e32;
   case AMDGPU::S_ADDC_U32: return AMDGPU::V_ADDC_U32_e32;
   case AMDGPU::S_SUB_I32: return AMDGPU::V_SUB_I32_e32;
   case AMDGPU::S_SUBB_U32: return AMDGPU::V_SUBB_U32_e32;
+  case AMDGPU::S_AND_B32: return AMDGPU::V_AND_B32_e32;
+  case AMDGPU::S_OR_B32: return AMDGPU::V_OR_B32_e32;
+  case AMDGPU::S_XOR_B32: return AMDGPU::V_XOR_B32_e32;
+  case AMDGPU::S_MIN_I32: return AMDGPU::V_MIN_I32_e32;
+  case AMDGPU::S_MIN_U32: return AMDGPU::V_MIN_U32_e32;
+  case AMDGPU::S_MAX_I32: return AMDGPU::V_MAX_I32_e32;
+  case AMDGPU::S_MAX_U32: return AMDGPU::V_MAX_U32_e32;
   case AMDGPU::S_ASHR_I32: return AMDGPU::V_ASHR_I32_e32;
   case AMDGPU::S_ASHR_I64: return AMDGPU::V_ASHR_I64;
   case AMDGPU::S_LSHL_B32: return AMDGPU::V_LSHL_B32_e32;
   case AMDGPU::S_LSHL_B64: return AMDGPU::V_LSHL_B64;
   case AMDGPU::S_LSHR_B32: return AMDGPU::V_LSHR_B32_e32;
   case AMDGPU::S_LSHR_B64: return AMDGPU::V_LSHR_B64;
+  case AMDGPU::S_SEXT_I32_I8: return AMDGPU::V_BFE_I32;
+  case AMDGPU::S_SEXT_I32_I16: return AMDGPU::V_BFE_I32;
+  case AMDGPU::S_BFE_U32: return AMDGPU::V_BFE_U32;
+  case AMDGPU::S_BFE_I32: return AMDGPU::V_BFE_I32;
+  case AMDGPU::S_BREV_B32: return AMDGPU::V_BFREV_B32_e32;
+  case AMDGPU::S_NOT_B32: return AMDGPU::V_NOT_B32_e32;
+  case AMDGPU::S_NOT_B64: return AMDGPU::V_NOT_B32_e32;
+  case AMDGPU::S_CMP_EQ_I32: return AMDGPU::V_CMP_EQ_I32_e32;
+  case AMDGPU::S_CMP_LG_I32: return AMDGPU::V_CMP_NE_I32_e32;
+  case AMDGPU::S_CMP_GT_I32: return AMDGPU::V_CMP_GT_I32_e32;
+  case AMDGPU::S_CMP_GE_I32: return AMDGPU::V_CMP_GE_I32_e32;
+  case AMDGPU::S_CMP_LT_I32: return AMDGPU::V_CMP_LT_I32_e32;
+  case AMDGPU::S_CMP_LE_I32: return AMDGPU::V_CMP_LE_I32_e32;
+  case AMDGPU::S_LOAD_DWORD_IMM:
+  case AMDGPU::S_LOAD_DWORD_SGPR: return AMDGPU::BUFFER_LOAD_DWORD_ADDR64;
+  case AMDGPU::S_LOAD_DWORDX2_IMM:
+  case AMDGPU::S_LOAD_DWORDX2_SGPR: return AMDGPU::BUFFER_LOAD_DWORDX2_ADDR64;
+  case AMDGPU::S_LOAD_DWORDX4_IMM:
+  case AMDGPU::S_LOAD_DWORDX4_SGPR: return AMDGPU::BUFFER_LOAD_DWORDX4_ADDR64;
+  case AMDGPU::S_BCNT1_I32_B32: return AMDGPU::V_BCNT_U32_B32_e32;
+  case AMDGPU::S_FF1_I32_B32: return AMDGPU::V_FFBL_B32_e32;
+  case AMDGPU::S_FLBIT_I32_B32: return AMDGPU::V_FFBH_U32_e32;
   }
 }
 
@@ -406,6 +790,8 @@ bool SIInstrInfo::canReadVGPR(const MachineInstr &MI, unsigned OpNo) const {
   switch (MI.getOpcode()) {
   case AMDGPU::COPY:
   case AMDGPU::REG_SEQUENCE:
+  case AMDGPU::PHI:
+  case AMDGPU::INSERT_SUBREG:
     return RI.hasVGPRs(getOpRegClass(MI, 0));
   default:
     return RI.hasVGPRs(getOpRegClass(MI, OpNo));
@@ -432,6 +818,84 @@ void SIInstrInfo::legalizeOpWithMove(MachineInstr *MI, unsigned OpIdx) const {
   MO.ChangeToRegister(Reg, false);
 }
 
+unsigned SIInstrInfo::buildExtractSubReg(MachineBasicBlock::iterator MI,
+                                         MachineRegisterInfo &MRI,
+                                         MachineOperand &SuperReg,
+                                         const TargetRegisterClass *SuperRC,
+                                         unsigned SubIdx,
+                                         const TargetRegisterClass *SubRC)
+                                         const {
+  assert(SuperReg.isReg());
+
+  unsigned NewSuperReg = MRI.createVirtualRegister(SuperRC);
+  unsigned SubReg = MRI.createVirtualRegister(SubRC);
+
+  // Just in case the super register is itself a sub-register, copy it to a new
+  // value so we don't need to worry about merging its subreg index with the
+  // SubIdx passed to this function. The register coalescer should be able to
+  // eliminate this extra copy.
+  BuildMI(*MI->getParent(), MI, MI->getDebugLoc(), get(TargetOpcode::COPY),
+          NewSuperReg)
+          .addOperand(SuperReg);
+
+  BuildMI(*MI->getParent(), MI, MI->getDebugLoc(), get(TargetOpcode::COPY),
+          SubReg)
+          .addReg(NewSuperReg, 0, SubIdx);
+  return SubReg;
+}
+
+MachineOperand SIInstrInfo::buildExtractSubRegOrImm(
+  MachineBasicBlock::iterator MII,
+  MachineRegisterInfo &MRI,
+  MachineOperand &Op,
+  const TargetRegisterClass *SuperRC,
+  unsigned SubIdx,
+  const TargetRegisterClass *SubRC) const {
+  if (Op.isImm()) {
+    // XXX - Is there a better way to do this?
+    if (SubIdx == AMDGPU::sub0)
+      return MachineOperand::CreateImm(Op.getImm() & 0xFFFFFFFF);
+    if (SubIdx == AMDGPU::sub1)
+      return MachineOperand::CreateImm(Op.getImm() >> 32);
+
+    llvm_unreachable("Unhandled register index for immediate");
+  }
+
+  unsigned SubReg = buildExtractSubReg(MII, MRI, Op, SuperRC,
+                                       SubIdx, SubRC);
+  return MachineOperand::CreateReg(SubReg, false);
+}
+
+unsigned SIInstrInfo::split64BitImm(SmallVectorImpl<MachineInstr *> &Worklist,
+                                    MachineBasicBlock::iterator MI,
+                                    MachineRegisterInfo &MRI,
+                                    const TargetRegisterClass *RC,
+                                    const MachineOperand &Op) const {
+  MachineBasicBlock *MBB = MI->getParent();
+  DebugLoc DL = MI->getDebugLoc();
+  unsigned LoDst = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+  unsigned HiDst = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+  unsigned Dst = MRI.createVirtualRegister(RC);
+
+  MachineInstr *Lo = BuildMI(*MBB, MI, DL, get(AMDGPU::S_MOV_B32),
+                             LoDst)
+    .addImm(Op.getImm() & 0xFFFFFFFF);
+  MachineInstr *Hi = BuildMI(*MBB, MI, DL, get(AMDGPU::S_MOV_B32),
+                             HiDst)
+    .addImm(Op.getImm() >> 32);
+
+  BuildMI(*MBB, MI, DL, get(TargetOpcode::REG_SEQUENCE), Dst)
+    .addReg(LoDst)
+    .addImm(AMDGPU::sub0)
+    .addReg(HiDst)
+    .addImm(AMDGPU::sub1);
+
+  Worklist.push_back(Lo);
+  Worklist.push_back(Hi);
+
+  return Dst;
+}
+
 void SIInstrInfo::legalizeOperands(MachineInstr *MI) const {
   MachineRegisterInfo &MRI = MI->getParent()->getParent()->getRegInfo();
   int Src0Idx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
@@ -506,11 +970,12 @@ void SIInstrInfo::legalizeOperands(MachineInstr *MI) const {
     }
   }
 
-  // Legalize REG_SEQUENCE
+  // Legalize REG_SEQUENCE and PHI
   // The register class of the operands much be the same type as the register
   // class of the output.
-  if (MI->getOpcode() == AMDGPU::REG_SEQUENCE) {
-    const TargetRegisterClass *RC = NULL, *SRC = NULL, *VRC = NULL;
+  if (MI->getOpcode() == AMDGPU::REG_SEQUENCE ||
+      MI->getOpcode() == AMDGPU::PHI) {
+    const TargetRegisterClass *RC = nullptr, *SRC = nullptr, *VRC = nullptr;
     for (unsigned i = 1, e = MI->getNumOperands(); i != e; i+=2) {
       if (!MI->getOperand(i).isReg() ||
           !TargetRegisterInfo::isVirtualRegister(MI->getOperand(i).getReg()))
@@ -543,12 +1008,209 @@ void SIInstrInfo::legalizeOperands(MachineInstr *MI) const {
           !TargetRegisterInfo::isVirtualRegister(MI->getOperand(i).getReg()))
         continue;
       unsigned DstReg = MRI.createVirtualRegister(RC);
-      BuildMI(*MI->getParent(), MI, MI->getDebugLoc(),
+      MachineBasicBlock *InsertBB;
+      MachineBasicBlock::iterator Insert;
+      if (MI->getOpcode() == AMDGPU::REG_SEQUENCE) {
+        InsertBB = MI->getParent();
+        Insert = MI;
+      } else {
+        // MI is a PHI instruction.
+        InsertBB = MI->getOperand(i + 1).getMBB();
+        Insert = InsertBB->getFirstTerminator();
+      }
+      BuildMI(*InsertBB, Insert, MI->getDebugLoc(),
               get(AMDGPU::COPY), DstReg)
               .addOperand(MI->getOperand(i));
       MI->getOperand(i).setReg(DstReg);
     }
   }
+
+  // Legalize INSERT_SUBREG
+  // src0 must have the same register class as dst
+  if (MI->getOpcode() == AMDGPU::INSERT_SUBREG) {
+    unsigned Dst = MI->getOperand(0).getReg();
+    unsigned Src0 = MI->getOperand(1).getReg();
+    const TargetRegisterClass *DstRC = MRI.getRegClass(Dst);
+    const TargetRegisterClass *Src0RC = MRI.getRegClass(Src0);
+    if (DstRC != Src0RC) {
+      MachineBasicBlock &MBB = *MI->getParent();
+      unsigned NewSrc0 = MRI.createVirtualRegister(DstRC);
+      BuildMI(MBB, MI, MI->getDebugLoc(), get(AMDGPU::COPY), NewSrc0)
+              .addReg(Src0);
+      MI->getOperand(1).setReg(NewSrc0);
+    }
+    return;
+  }
+
+  // Legalize MUBUF* instructions
+  // FIXME: If we start using the non-addr64 instructions for compute, we
+  // may need to legalize them here.
+
+  int SRsrcIdx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
+                                            AMDGPU::OpName::srsrc);
+  int VAddrIdx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
+                                             AMDGPU::OpName::vaddr);
+  if (SRsrcIdx != -1 && VAddrIdx != -1) {
+    const TargetRegisterClass *VAddrRC =
+        RI.getRegClass(get(MI->getOpcode()).OpInfo[VAddrIdx].RegClass);
+
+    if(VAddrRC->getSize() == 8 &&
+       MRI.getRegClass(MI->getOperand(SRsrcIdx).getReg()) != VAddrRC) {
+      // We have a MUBUF instruction that uses a 64-bit vaddr register and
+      // srsrc has the incorrect register class.  In order to fix this, we
+      // need to extract the pointer from the resource descriptor (srsrc),
+      // add it to the value of vadd,  then store the result in the vaddr
+      // operand.  Then, we need to set the pointer field of the resource
+      // descriptor to zero.
+
+      MachineBasicBlock &MBB = *MI->getParent();
+      MachineOperand &SRsrcOp = MI->getOperand(SRsrcIdx);
+      MachineOperand &VAddrOp = MI->getOperand(VAddrIdx);
+      unsigned SRsrcPtrLo, SRsrcPtrHi, VAddrLo, VAddrHi;
+      unsigned NewVAddrLo = MRI.createVirtualRegister(&AMDGPU::VReg_32RegClass);
+      unsigned NewVAddrHi = MRI.createVirtualRegister(&AMDGPU::VReg_32RegClass);
+      unsigned NewVAddr = MRI.createVirtualRegister(&AMDGPU::VReg_64RegClass);
+      unsigned Zero64 = MRI.createVirtualRegister(&AMDGPU::SReg_64RegClass);
+      unsigned SRsrcFormatLo = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+      unsigned SRsrcFormatHi = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+      unsigned NewSRsrc = MRI.createVirtualRegister(&AMDGPU::SReg_128RegClass);
+
+      // SRsrcPtrLo = srsrc:sub0
+      SRsrcPtrLo = buildExtractSubReg(MI, MRI, SRsrcOp,
+          &AMDGPU::VReg_128RegClass, AMDGPU::sub0, &AMDGPU::VReg_32RegClass);
+
+      // SRsrcPtrHi = srsrc:sub1
+      SRsrcPtrHi = buildExtractSubReg(MI, MRI, SRsrcOp,
+          &AMDGPU::VReg_128RegClass, AMDGPU::sub1, &AMDGPU::VReg_32RegClass);
+
+      // VAddrLo = vaddr:sub0
+      VAddrLo = buildExtractSubReg(MI, MRI, VAddrOp,
+          &AMDGPU::VReg_64RegClass, AMDGPU::sub0, &AMDGPU::VReg_32RegClass);
+
+      // VAddrHi = vaddr:sub1
+      VAddrHi = buildExtractSubReg(MI, MRI, VAddrOp,
+          &AMDGPU::VReg_64RegClass, AMDGPU::sub1, &AMDGPU::VReg_32RegClass);
+
+      // NewVaddrLo = SRsrcPtrLo + VAddrLo
+      BuildMI(MBB, MI, MI->getDebugLoc(), get(AMDGPU::V_ADD_I32_e32),
+              NewVAddrLo)
+              .addReg(SRsrcPtrLo)
+              .addReg(VAddrLo)
+              .addReg(AMDGPU::VCC, RegState::Define | RegState::Implicit);
+
+      // NewVaddrHi = SRsrcPtrHi + VAddrHi
+      BuildMI(MBB, MI, MI->getDebugLoc(), get(AMDGPU::V_ADDC_U32_e32),
+              NewVAddrHi)
+              .addReg(SRsrcPtrHi)
+              .addReg(VAddrHi)
+              .addReg(AMDGPU::VCC, RegState::ImplicitDefine)
+              .addReg(AMDGPU::VCC, RegState::Implicit);
+
+      // NewVaddr = {NewVaddrHi, NewVaddrLo}
+      BuildMI(MBB, MI, MI->getDebugLoc(), get(AMDGPU::REG_SEQUENCE),
+              NewVAddr)
+              .addReg(NewVAddrLo)
+              .addImm(AMDGPU::sub0)
+              .addReg(NewVAddrHi)
+              .addImm(AMDGPU::sub1);
+
+      // Zero64 = 0
+      BuildMI(MBB, MI, MI->getDebugLoc(), get(AMDGPU::S_MOV_B64),
+              Zero64)
+              .addImm(0);
+
+      // SRsrcFormatLo = RSRC_DATA_FORMAT{31-0}
+      BuildMI(MBB, MI, MI->getDebugLoc(), get(AMDGPU::S_MOV_B32),
+              SRsrcFormatLo)
+              .addImm(AMDGPU::RSRC_DATA_FORMAT & 0xFFFFFFFF);
+
+      // SRsrcFormatHi = RSRC_DATA_FORMAT{63-32}
+      BuildMI(MBB, MI, MI->getDebugLoc(), get(AMDGPU::S_MOV_B32),
+              SRsrcFormatHi)
+              .addImm(AMDGPU::RSRC_DATA_FORMAT >> 32);
+
+      // NewSRsrc = {Zero64, SRsrcFormat}
+      BuildMI(MBB, MI, MI->getDebugLoc(), get(AMDGPU::REG_SEQUENCE),
+              NewSRsrc)
+              .addReg(Zero64)
+              .addImm(AMDGPU::sub0_sub1)
+              .addReg(SRsrcFormatLo)
+              .addImm(AMDGPU::sub2)
+              .addReg(SRsrcFormatHi)
+              .addImm(AMDGPU::sub3);
+
+      // Update the instruction to use NewVaddr
+      MI->getOperand(VAddrIdx).setReg(NewVAddr);
+      // Update the instruction to use NewSRsrc
+      MI->getOperand(SRsrcIdx).setReg(NewSRsrc);
+    }
+  }
+}
+
+void SIInstrInfo::moveSMRDToVALU(MachineInstr *MI, MachineRegisterInfo &MRI) const {
+  MachineBasicBlock *MBB = MI->getParent();
+  switch (MI->getOpcode()) {
+    case AMDGPU::S_LOAD_DWORD_IMM:
+    case AMDGPU::S_LOAD_DWORD_SGPR:
+    case AMDGPU::S_LOAD_DWORDX2_IMM:
+    case AMDGPU::S_LOAD_DWORDX2_SGPR:
+    case AMDGPU::S_LOAD_DWORDX4_IMM:
+    case AMDGPU::S_LOAD_DWORDX4_SGPR:
+      unsigned NewOpcode = getVALUOp(*MI);
+      unsigned RegOffset;
+      unsigned ImmOffset;
+
+      if (MI->getOperand(2).isReg()) {
+        RegOffset = MI->getOperand(2).getReg();
+        ImmOffset = 0;
+      } else {
+        assert(MI->getOperand(2).isImm());
+        // SMRD instructions take a dword offsets and MUBUF instructions
+        // take a byte offset.
+        ImmOffset = MI->getOperand(2).getImm() << 2;
+        RegOffset = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+        if (isUInt<12>(ImmOffset)) {
+          BuildMI(*MBB, MI, MI->getDebugLoc(), get(AMDGPU::S_MOV_B32),
+                  RegOffset)
+                  .addImm(0);
+        } else {
+          BuildMI(*MBB, MI, MI->getDebugLoc(), get(AMDGPU::S_MOV_B32),
+                  RegOffset)
+                  .addImm(ImmOffset);
+          ImmOffset = 0;
+        }
+      }
+
+      unsigned SRsrc = MRI.createVirtualRegister(&AMDGPU::SReg_128RegClass);
+      unsigned DWord0 = RegOffset;
+      unsigned DWord1 = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+      unsigned DWord2 = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+      unsigned DWord3 = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+
+      BuildMI(*MBB, MI, MI->getDebugLoc(), get(AMDGPU::S_MOV_B32), DWord1)
+              .addImm(0);
+      BuildMI(*MBB, MI, MI->getDebugLoc(), get(AMDGPU::S_MOV_B32), DWord2)
+              .addImm(AMDGPU::RSRC_DATA_FORMAT & 0xFFFFFFFF);
+      BuildMI(*MBB, MI, MI->getDebugLoc(), get(AMDGPU::S_MOV_B32), DWord3)
+              .addImm(AMDGPU::RSRC_DATA_FORMAT >> 32);
+      BuildMI(*MBB, MI, MI->getDebugLoc(), get(AMDGPU::REG_SEQUENCE), SRsrc)
+              .addReg(DWord0)
+              .addImm(AMDGPU::sub0)
+              .addReg(DWord1)
+              .addImm(AMDGPU::sub1)
+              .addReg(DWord2)
+              .addImm(AMDGPU::sub2)
+              .addReg(DWord3)
+              .addImm(AMDGPU::sub3);
+     MI->setDesc(get(NewOpcode));
+     if (MI->getOperand(2).isReg()) {
+       MI->getOperand(2).setReg(MI->getOperand(1).getReg());
+     } else {
+       MI->getOperand(2).ChangeToRegister(MI->getOperand(1).getReg(), false);
+     }
+     MI->getOperand(1).setReg(SRsrc);
+     MI->addOperand(*MBB->getParent(), MachineOperand::CreateImm(ImmOffset));
+  }
 }
 
 void SIInstrInfo::moveToVALU(MachineInstr &TopInst) const {
@@ -557,11 +1219,80 @@ void SIInstrInfo::moveToVALU(MachineInstr &TopInst) const {
 
   while (!Worklist.empty()) {
     MachineInstr *Inst = Worklist.pop_back_val();
+    MachineBasicBlock *MBB = Inst->getParent();
+    MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
+
+    unsigned Opcode = Inst->getOpcode();
     unsigned NewOpcode = getVALUOp(*Inst);
-    if (NewOpcode == AMDGPU::INSTRUCTION_LIST_END)
+
+    // Handle some special cases
+    switch (Opcode) {
+    default:
+      if (isSMRD(Inst->getOpcode())) {
+        moveSMRDToVALU(Inst, MRI);
+      }
+      break;
+    case AMDGPU::S_MOV_B64: {
+      DebugLoc DL = Inst->getDebugLoc();
+
+      // If the source operand is a register we can replace this with a
+      // copy.
+      if (Inst->getOperand(1).isReg()) {
+        MachineInstr *Copy = BuildMI(*MBB, Inst, DL, get(TargetOpcode::COPY))
+          .addOperand(Inst->getOperand(0))
+          .addOperand(Inst->getOperand(1));
+        Worklist.push_back(Copy);
+      } else {
+        // Otherwise, we need to split this into two movs, because there is
+        // no 64-bit VALU move instruction.
+        unsigned Reg = Inst->getOperand(0).getReg();
+        unsigned Dst = split64BitImm(Worklist,
+                                     Inst,
+                                     MRI,
+                                     MRI.getRegClass(Reg),
+                                     Inst->getOperand(1));
+        MRI.replaceRegWith(Reg, Dst);
+      }
+      Inst->eraseFromParent();
+      continue;
+    }
+    case AMDGPU::S_AND_B64:
+      splitScalar64BitBinaryOp(Worklist, Inst, AMDGPU::S_AND_B32);
+      Inst->eraseFromParent();
+      continue;
+
+    case AMDGPU::S_OR_B64:
+      splitScalar64BitBinaryOp(Worklist, Inst, AMDGPU::S_OR_B32);
+      Inst->eraseFromParent();
+      continue;
+
+    case AMDGPU::S_XOR_B64:
+      splitScalar64BitBinaryOp(Worklist, Inst, AMDGPU::S_XOR_B32);
+      Inst->eraseFromParent();
       continue;
 
-    MachineRegisterInfo &MRI = Inst->getParent()->getParent()->getRegInfo();
+    case AMDGPU::S_NOT_B64:
+      splitScalar64BitUnaryOp(Worklist, Inst, AMDGPU::S_NOT_B32);
+      Inst->eraseFromParent();
+      continue;
+
+    case AMDGPU::S_BCNT1_I32_B64:
+      splitScalar64BitBCNT(Worklist, Inst);
+      Inst->eraseFromParent();
+      continue;
+
+    case AMDGPU::S_BFE_U64:
+    case AMDGPU::S_BFE_I64:
+    case AMDGPU::S_BFM_B64:
+      llvm_unreachable("Moving this op to VALU not implemented");
+    }
+
+    if (NewOpcode == AMDGPU::INSTRUCTION_LIST_END) {
+      // We cannot move this instruction to the VALU, so we should try to
+      // legalize its operands instead.
+      legalizeOperands(Inst);
+      continue;
+    }
 
     // Use the new VALU Opcode.
     const MCInstrDesc &NewDesc = get(NewOpcode);
@@ -576,27 +1307,56 @@ void SIInstrInfo::moveToVALU(MachineInstr &TopInst) const {
         Inst->RemoveOperand(i);
     }
 
-    // Add the implict and explicit register definitions.
-    if (NewDesc.ImplicitUses) {
-      for (unsigned i = 0; NewDesc.ImplicitUses[i]; ++i) {
-        unsigned Reg = NewDesc.ImplicitUses[i];
-        Inst->addOperand(MachineOperand::CreateReg(Reg, false, true));
-      }
+    if (Opcode == AMDGPU::S_SEXT_I32_I8 || Opcode == AMDGPU::S_SEXT_I32_I16) {
+      // We are converting these to a BFE, so we need to add the missing
+      // operands for the size and offset.
+      unsigned Size = (Opcode == AMDGPU::S_SEXT_I32_I8) ? 8 : 16;
+      Inst->addOperand(Inst->getOperand(1));
+      Inst->getOperand(1).ChangeToImmediate(0);
+      Inst->addOperand(MachineOperand::CreateImm(0));
+      Inst->addOperand(MachineOperand::CreateImm(0));
+      Inst->addOperand(MachineOperand::CreateImm(0));
+      Inst->addOperand(MachineOperand::CreateImm(Size));
+
+      // XXX - Other pointless operands. There are 4, but it seems you only need
+      // 3 to not hit an assertion later in MCInstLower.
+      Inst->addOperand(MachineOperand::CreateImm(0));
+      Inst->addOperand(MachineOperand::CreateImm(0));
+    } else if (Opcode == AMDGPU::S_BCNT1_I32_B32) {
+      // The VALU version adds the second operand to the result, so insert an
+      // extra 0 operand.
+      Inst->addOperand(MachineOperand::CreateImm(0));
     }
 
-    if (NewDesc.ImplicitDefs) {
-      for (unsigned i = 0; NewDesc.ImplicitDefs[i]; ++i) {
-        unsigned Reg = NewDesc.ImplicitDefs[i];
-        Inst->addOperand(MachineOperand::CreateReg(Reg, true, true));
-      }
+    addDescImplicitUseDef(NewDesc, Inst);
+
+    if (Opcode == AMDGPU::S_BFE_I32 || Opcode == AMDGPU::S_BFE_U32) {
+      const MachineOperand &OffsetWidthOp = Inst->getOperand(2);
+      // If we need to move this to VGPRs, we need to unpack the second operand
+      // back into the 2 separate ones for bit offset and width.
+      assert(OffsetWidthOp.isImm() &&
+             "Scalar BFE is only implemented for constant width and offset");
+      uint32_t Imm = OffsetWidthOp.getImm();
+
+      uint32_t Offset = Imm & 0x3f; // Extract bits [5:0].
+      uint32_t BitWidth = (Imm & 0x7f0000) >> 16; // Extract bits [22:16].
+
+      Inst->RemoveOperand(2); // Remove old immediate.
+      Inst->addOperand(Inst->getOperand(1));
+      Inst->getOperand(1).ChangeToImmediate(0);
+      Inst->addOperand(MachineOperand::CreateImm(0));
+      Inst->addOperand(MachineOperand::CreateImm(Offset));
+      Inst->addOperand(MachineOperand::CreateImm(0));
+      Inst->addOperand(MachineOperand::CreateImm(BitWidth));
+      Inst->addOperand(MachineOperand::CreateImm(0));
+      Inst->addOperand(MachineOperand::CreateImm(0));
     }
 
-    legalizeOperands(Inst);
-
     // Update the destination register class.
+
     const TargetRegisterClass *NewDstRC = getOpRegClass(*Inst, 0);
 
-    switch (Inst->getOpcode()) {
+    switch (Opcode) {
       // For target instructions, getOpRegClass just returns the virtual
       // register class associated with the operand, so we need to find an
       // equivalent VGPR register class in order to move the instruction to the
@@ -604,6 +1364,7 @@ void SIInstrInfo::moveToVALU(MachineInstr &TopInst) const {
     case AMDGPU::COPY:
     case AMDGPU::PHI:
     case AMDGPU::REG_SEQUENCE:
+    case AMDGPU::INSERT_SUBREG:
       if (RI.hasVGPRs(NewDstRC))
         continue;
       NewDstRC = RI.getEquivalentVGPRClass(NewDstRC);
@@ -618,9 +1379,12 @@ void SIInstrInfo::moveToVALU(MachineInstr &TopInst) const {
     unsigned NewDstReg = MRI.createVirtualRegister(NewDstRC);
     MRI.replaceRegWith(DstReg, NewDstReg);
 
+    // Legalize the operands
+    legalizeOperands(Inst);
+
     for (MachineRegisterInfo::use_iterator I = MRI.use_begin(NewDstReg),
            E = MRI.use_end(); I != E; ++I) {
-      MachineInstr &UseMI = *I;
+      MachineInstr &UseMI = *I->getParent();
       if (!canReadVGPR(UseMI, I.getOperandNo())) {
         Worklist.push_back(&UseMI);
       }
@@ -642,6 +1406,180 @@ const TargetRegisterClass *SIInstrInfo::getIndirectAddrRegClass() const {
   return &AMDGPU::VReg_32RegClass;
 }
 
+void SIInstrInfo::splitScalar64BitUnaryOp(
+  SmallVectorImpl<MachineInstr *> &Worklist,
+  MachineInstr *Inst,
+  unsigned Opcode) const {
+  MachineBasicBlock &MBB = *Inst->getParent();
+  MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
+
+  MachineOperand &Dest = Inst->getOperand(0);
+  MachineOperand &Src0 = Inst->getOperand(1);
+  DebugLoc DL = Inst->getDebugLoc();
+
+  MachineBasicBlock::iterator MII = Inst;
+
+  const MCInstrDesc &InstDesc = get(Opcode);
+  const TargetRegisterClass *Src0RC = Src0.isReg() ?
+    MRI.getRegClass(Src0.getReg()) :
+    &AMDGPU::SGPR_32RegClass;
+
+  const TargetRegisterClass *Src0SubRC = RI.getSubRegClass(Src0RC, AMDGPU::sub0);
+
+  MachineOperand SrcReg0Sub0 = buildExtractSubRegOrImm(MII, MRI, Src0, Src0RC,
+                                                       AMDGPU::sub0, Src0SubRC);
+
+  const TargetRegisterClass *DestRC = MRI.getRegClass(Dest.getReg());
+  const TargetRegisterClass *DestSubRC = RI.getSubRegClass(DestRC, AMDGPU::sub0);
+
+  unsigned DestSub0 = MRI.createVirtualRegister(DestRC);
+  MachineInstr *LoHalf = BuildMI(MBB, MII, DL, InstDesc, DestSub0)
+    .addOperand(SrcReg0Sub0);
+
+  MachineOperand SrcReg0Sub1 = buildExtractSubRegOrImm(MII, MRI, Src0, Src0RC,
+                                                       AMDGPU::sub1, Src0SubRC);
+
+  unsigned DestSub1 = MRI.createVirtualRegister(DestSubRC);
+  MachineInstr *HiHalf = BuildMI(MBB, MII, DL, InstDesc, DestSub1)
+    .addOperand(SrcReg0Sub1);
+
+  unsigned FullDestReg = MRI.createVirtualRegister(DestRC);
+  BuildMI(MBB, MII, DL, get(TargetOpcode::REG_SEQUENCE), FullDestReg)
+    .addReg(DestSub0)
+    .addImm(AMDGPU::sub0)
+    .addReg(DestSub1)
+    .addImm(AMDGPU::sub1);
+
+  MRI.replaceRegWith(Dest.getReg(), FullDestReg);
+
+  // Try to legalize the operands in case we need to swap the order to keep it
+  // valid.
+  Worklist.push_back(LoHalf);
+  Worklist.push_back(HiHalf);
+}
+
+void SIInstrInfo::splitScalar64BitBinaryOp(
+  SmallVectorImpl<MachineInstr *> &Worklist,
+  MachineInstr *Inst,
+  unsigned Opcode) const {
+  MachineBasicBlock &MBB = *Inst->getParent();
+  MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
+
+  MachineOperand &Dest = Inst->getOperand(0);
+  MachineOperand &Src0 = Inst->getOperand(1);
+  MachineOperand &Src1 = Inst->getOperand(2);
+  DebugLoc DL = Inst->getDebugLoc();
+
+  MachineBasicBlock::iterator MII = Inst;
+
+  const MCInstrDesc &InstDesc = get(Opcode);
+  const TargetRegisterClass *Src0RC = Src0.isReg() ?
+    MRI.getRegClass(Src0.getReg()) :
+    &AMDGPU::SGPR_32RegClass;
+
+  const TargetRegisterClass *Src0SubRC = RI.getSubRegClass(Src0RC, AMDGPU::sub0);
+  const TargetRegisterClass *Src1RC = Src1.isReg() ?
+    MRI.getRegClass(Src1.getReg()) :
+    &AMDGPU::SGPR_32RegClass;
+
+  const TargetRegisterClass *Src1SubRC = RI.getSubRegClass(Src1RC, AMDGPU::sub0);
+
+  MachineOperand SrcReg0Sub0 = buildExtractSubRegOrImm(MII, MRI, Src0, Src0RC,
+                                                       AMDGPU::sub0, Src0SubRC);
+  MachineOperand SrcReg1Sub0 = buildExtractSubRegOrImm(MII, MRI, Src1, Src1RC,
+                                                       AMDGPU::sub0, Src1SubRC);
+
+  const TargetRegisterClass *DestRC = MRI.getRegClass(Dest.getReg());
+  const TargetRegisterClass *DestSubRC = RI.getSubRegClass(DestRC, AMDGPU::sub0);
+
+  unsigned DestSub0 = MRI.createVirtualRegister(DestRC);
+  MachineInstr *LoHalf = BuildMI(MBB, MII, DL, InstDesc, DestSub0)
+    .addOperand(SrcReg0Sub0)
+    .addOperand(SrcReg1Sub0);
+
+  MachineOperand SrcReg0Sub1 = buildExtractSubRegOrImm(MII, MRI, Src0, Src0RC,
+                                                       AMDGPU::sub1, Src0SubRC);
+  MachineOperand SrcReg1Sub1 = buildExtractSubRegOrImm(MII, MRI, Src1, Src1RC,
+                                                       AMDGPU::sub1, Src1SubRC);
+
+  unsigned DestSub1 = MRI.createVirtualRegister(DestSubRC);
+  MachineInstr *HiHalf = BuildMI(MBB, MII, DL, InstDesc, DestSub1)
+    .addOperand(SrcReg0Sub1)
+    .addOperand(SrcReg1Sub1);
+
+  unsigned FullDestReg = MRI.createVirtualRegister(DestRC);
+  BuildMI(MBB, MII, DL, get(TargetOpcode::REG_SEQUENCE), FullDestReg)
+    .addReg(DestSub0)
+    .addImm(AMDGPU::sub0)
+    .addReg(DestSub1)
+    .addImm(AMDGPU::sub1);
+
+  MRI.replaceRegWith(Dest.getReg(), FullDestReg);
+
+  // Try to legalize the operands in case we need to swap the order to keep it
+  // valid.
+  Worklist.push_back(LoHalf);
+  Worklist.push_back(HiHalf);
+}
+
+void SIInstrInfo::splitScalar64BitBCNT(SmallVectorImpl<MachineInstr *> &Worklist,
+                                       MachineInstr *Inst) const {
+  MachineBasicBlock &MBB = *Inst->getParent();
+  MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
+
+  MachineBasicBlock::iterator MII = Inst;
+  DebugLoc DL = Inst->getDebugLoc();
+
+  MachineOperand &Dest = Inst->getOperand(0);
+  MachineOperand &Src = Inst->getOperand(1);
+
+  const MCInstrDesc &InstDesc = get(AMDGPU::V_BCNT_U32_B32_e32);
+  const TargetRegisterClass *SrcRC = Src.isReg() ?
+    MRI.getRegClass(Src.getReg()) :
+    &AMDGPU::SGPR_32RegClass;
+
+  unsigned MidReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+  unsigned ResultReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+
+  const TargetRegisterClass *SrcSubRC = RI.getSubRegClass(SrcRC, AMDGPU::sub0);
+
+  MachineOperand SrcRegSub0 = buildExtractSubRegOrImm(MII, MRI, Src, SrcRC,
+                                                      AMDGPU::sub0, SrcSubRC);
+  MachineOperand SrcRegSub1 = buildExtractSubRegOrImm(MII, MRI, Src, SrcRC,
+                                                      AMDGPU::sub1, SrcSubRC);
+
+  MachineInstr *First = BuildMI(MBB, MII, DL, InstDesc, MidReg)
+    .addOperand(SrcRegSub0)
+    .addImm(0);
+
+  MachineInstr *Second = BuildMI(MBB, MII, DL, InstDesc, ResultReg)
+    .addOperand(SrcRegSub1)
+    .addReg(MidReg);
+
+  MRI.replaceRegWith(Dest.getReg(), ResultReg);
+
+  Worklist.push_back(First);
+  Worklist.push_back(Second);
+}
+
+void SIInstrInfo::addDescImplicitUseDef(const MCInstrDesc &NewDesc,
+                                        MachineInstr *Inst) const {
+  // Add the implict and explicit register definitions.
+  if (NewDesc.ImplicitUses) {
+    for (unsigned i = 0; NewDesc.ImplicitUses[i]; ++i) {
+      unsigned Reg = NewDesc.ImplicitUses[i];
+      Inst->addOperand(MachineOperand::CreateReg(Reg, false, true));
+    }
+  }
+
+  if (NewDesc.ImplicitDefs) {
+    for (unsigned i = 0; NewDesc.ImplicitDefs[i]; ++i) {
+      unsigned Reg = NewDesc.ImplicitDefs[i];
+      Inst->addOperand(MachineOperand::CreateReg(Reg, true, true));
+    }
+  }
+}
+
 MachineInstrBuilder SIInstrInfo::buildIndirectWrite(
                                    MachineBasicBlock *MBB,
                                    MachineBasicBlock::iterator I,
@@ -705,3 +1643,12 @@ void SIInstrInfo::reserveIndirectRegisters(BitVector &Reserved,
   for (int Index = std::max(0, Begin - 15); Index <= End; ++Index)
     Reserved.set(AMDGPU::VReg_512RegClass.getRegister(Index));
 }
+
+const MachineOperand *SIInstrInfo::getNamedOperand(const MachineInstr& MI,
+                                                   unsigned OperandName) const {
+  int Idx = AMDGPU::getNamedOperandIdx(MI.getOpcode(), OperandName);
+  if (Idx == -1)
+    return nullptr;
+
+  return &MI.getOperand(Idx);
+}
diff --git a/contrib/llvm/lib/Target/R600/SIInstrInfo.h b/contrib/llvm/lib/Target/R600/SIInstrInfo.h
index 4af6348..4687539 100644
--- a/contrib/llvm/lib/Target/R600/SIInstrInfo.h
+++ b/contrib/llvm/lib/Target/R600/SIInstrInfo.h
@@ -25,50 +25,100 @@ class SIInstrInfo : public AMDGPUInstrInfo {
 private:
   const SIRegisterInfo RI;
 
-  MachineInstrBuilder buildIndirectIndexLoop(MachineBasicBlock &MBB,
-                                             MachineBasicBlock::iterator I,
-                                             unsigned OffsetVGPR,
-                                             unsigned MovRelOp,
-                                             unsigned Dst,
-                                             unsigned Src0) const;
-  // If you add or remove instructions from this function, you will
+  unsigned buildExtractSubReg(MachineBasicBlock::iterator MI,
+                              MachineRegisterInfo &MRI,
+                              MachineOperand &SuperReg,
+                              const TargetRegisterClass *SuperRC,
+                              unsigned SubIdx,
+                              const TargetRegisterClass *SubRC) const;
+  MachineOperand buildExtractSubRegOrImm(MachineBasicBlock::iterator MI,
+                                         MachineRegisterInfo &MRI,
+                                         MachineOperand &SuperReg,
+                                         const TargetRegisterClass *SuperRC,
+                                         unsigned SubIdx,
+                                         const TargetRegisterClass *SubRC) const;
+
+  unsigned split64BitImm(SmallVectorImpl<MachineInstr *> &Worklist,
+                         MachineBasicBlock::iterator MI,
+                         MachineRegisterInfo &MRI,
+                         const TargetRegisterClass *RC,
+                         const MachineOperand &Op) const;
+
+  void splitScalar64BitUnaryOp(SmallVectorImpl<MachineInstr *> &Worklist,
+                               MachineInstr *Inst, unsigned Opcode) const;
+
+  void splitScalar64BitBinaryOp(SmallVectorImpl<MachineInstr *> &Worklist,
+                                MachineInstr *Inst, unsigned Opcode) const;
+
+  void splitScalar64BitBCNT(SmallVectorImpl<MachineInstr *> &Worklist,
+                            MachineInstr *Inst) const;
+
+  void addDescImplicitUseDef(const MCInstrDesc &Desc, MachineInstr *MI) const;
 
 public:
-  explicit SIInstrInfo(AMDGPUTargetMachine &tm);
+  explicit SIInstrInfo(const AMDGPUSubtarget &st);
 
-  const SIRegisterInfo &getRegisterInfo() const;
+  const SIRegisterInfo &getRegisterInfo() const override {
+    return RI;
+  }
 
-  virtual void copyPhysReg(MachineBasicBlock &MBB,
-                           MachineBasicBlock::iterator MI, DebugLoc DL,
-                           unsigned DestReg, unsigned SrcReg,
-                           bool KillSrc) const;
+  void copyPhysReg(MachineBasicBlock &MBB,
+                   MachineBasicBlock::iterator MI, DebugLoc DL,
+                   unsigned DestReg, unsigned SrcReg,
+                   bool KillSrc) const override;
+
+  void storeRegToStackSlot(MachineBasicBlock &MBB,
+                           MachineBasicBlock::iterator MI,
+                           unsigned SrcReg, bool isKill, int FrameIndex,
+                           const TargetRegisterClass *RC,
+                           const TargetRegisterInfo *TRI) const override;
+
+  void loadRegFromStackSlot(MachineBasicBlock &MBB,
+                            MachineBasicBlock::iterator MI,
+                            unsigned DestReg, int FrameIndex,
+                            const TargetRegisterClass *RC,
+                            const TargetRegisterInfo *TRI) const override;
+
+  virtual bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const;
 
   unsigned commuteOpcode(unsigned Opcode) const;
 
-  virtual MachineInstr *commuteInstruction(MachineInstr *MI,
-                                           bool NewMI=false) const;
+  MachineInstr *commuteInstruction(MachineInstr *MI,
+                                   bool NewMI=false) const override;
+
+  bool isTriviallyReMaterializable(const MachineInstr *MI,
+                                   AliasAnalysis *AA = nullptr) const;
 
-  virtual unsigned getIEQOpcode() const { assert(!"Implement"); return 0;}
   MachineInstr *buildMovInstr(MachineBasicBlock *MBB,
                               MachineBasicBlock::iterator I,
-                              unsigned DstReg, unsigned SrcReg) const;
-  virtual bool isMov(unsigned Opcode) const;
+                              unsigned DstReg, unsigned SrcReg) const override;
+  bool isMov(unsigned Opcode) const override;
 
-  virtual bool isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const;
+  bool isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const override;
+  bool isDS(uint16_t Opcode) const;
   int isMIMG(uint16_t Opcode) const;
   int isSMRD(uint16_t Opcode) const;
   bool isVOP1(uint16_t Opcode) const;
   bool isVOP2(uint16_t Opcode) const;
   bool isVOP3(uint16_t Opcode) const;
   bool isVOPC(uint16_t Opcode) const;
+  bool isInlineConstant(const APInt &Imm) const;
   bool isInlineConstant(const MachineOperand &MO) const;
   bool isLiteralConstant(const MachineOperand &MO) const;
 
-  virtual bool verifyInstruction(const MachineInstr *MI,
-                                 StringRef &ErrInfo) const;
+  bool isImmOperandLegal(const MachineInstr *MI, unsigned OpNo,
+                         const MachineOperand &MO) const;
+
+  /// \brief Return true if this 64-bit VALU instruction has a 32-bit encoding.
+  /// This function will return false if you pass it a 32-bit instruction.
+  bool hasVALU32BitEncoding(unsigned Opcode) const;
+
+  bool verifyInstruction(const MachineInstr *MI,
+                         StringRef &ErrInfo) const override;
 
   bool isSALUInstr(const MachineInstr &MI) const;
   static unsigned getVALUOp(const MachineInstr &MI);
+
   bool isSALUOpSupportedOnVALU(const MachineInstr &MI) const;
 
   /// \brief Return the correct register class for \p OpNo.  For target-specific
@@ -98,39 +148,53 @@ public:
   /// create new instruction and insert them before \p MI.
   void legalizeOperands(MachineInstr *MI) const;
 
+  void moveSMRDToVALU(MachineInstr *MI, MachineRegisterInfo &MRI) const;
+
   /// \brief Replace this instruction's opcode with the equivalent VALU
   /// opcode.  This function will also move the users of \p MI to the
   /// VALU if necessary.
   void moveToVALU(MachineInstr &MI) const;
 
-  virtual unsigned calculateIndirectAddress(unsigned RegIndex,
-                                            unsigned Channel) const;
+  unsigned calculateIndirectAddress(unsigned RegIndex,
+                                    unsigned Channel) const override;
 
-  virtual const TargetRegisterClass *getIndirectAddrRegClass() const;
+  const TargetRegisterClass *getIndirectAddrRegClass() const override;
 
-  virtual MachineInstrBuilder buildIndirectWrite(MachineBasicBlock *MBB,
-                                                 MachineBasicBlock::iterator I,
-                                                 unsigned ValueReg,
-                                                 unsigned Address,
-                                                 unsigned OffsetReg) const;
+  MachineInstrBuilder buildIndirectWrite(MachineBasicBlock *MBB,
+                                         MachineBasicBlock::iterator I,
+                                         unsigned ValueReg,
+                                         unsigned Address,
+                                         unsigned OffsetReg) const override;
 
-  virtual MachineInstrBuilder buildIndirectRead(MachineBasicBlock *MBB,
-                                                MachineBasicBlock::iterator I,
-                                                unsigned ValueReg,
-                                                unsigned Address,
-                                                unsigned OffsetReg) const;
+  MachineInstrBuilder buildIndirectRead(MachineBasicBlock *MBB,
+                                        MachineBasicBlock::iterator I,
+                                        unsigned ValueReg,
+                                        unsigned Address,
+                                        unsigned OffsetReg) const override;
   void reserveIndirectRegisters(BitVector &Reserved,
                                 const MachineFunction &MF) const;
 
   void LoadM0(MachineInstr *MoveRel, MachineBasicBlock::iterator I,
               unsigned SavReg, unsigned IndexReg) const;
+
+  void insertNOPs(MachineBasicBlock::iterator MI, int Count) const;
+
+  /// \brief Returns the operand named \p Op.  If \p MI does not have an
+  /// operand named \c Op, this function returns nullptr.
+  const MachineOperand *getNamedOperand(const MachineInstr& MI,
+                                        unsigned OperandName) const;
 };
 
 namespace AMDGPU {
 
   int getVOPe64(uint16_t Opcode);
+  int getVOPe32(uint16_t Opcode);
   int getCommuteRev(uint16_t Opcode);
   int getCommuteOrig(uint16_t Opcode);
+  int getMCOpcode(uint16_t Opcode, unsigned Gen);
+
+  const uint64_t RSRC_DATA_FORMAT = 0xf00000000000LL;
+  const uint64_t RSRC_TID_ENABLE = 1LL << 55;
 
 } // End namespace AMDGPU
 
diff --git a/contrib/llvm/lib/Target/R600/SIInstrInfo.td b/contrib/llvm/lib/Target/R600/SIInstrInfo.td
index b7879c6..b0ac20f 100644
--- a/contrib/llvm/lib/Target/R600/SIInstrInfo.td
+++ b/contrib/llvm/lib/Target/R600/SIInstrInfo.td
@@ -7,23 +7,25 @@
 //
 //===----------------------------------------------------------------------===//
 
+// Execpt for the NONE field, this must be kept in sync with the SISubtarget enum
+// in AMDGPUMCInstLower.h
+def SISubtarget {
+  int NONE = -1;
+  int SI = 0;
+}
+
 //===----------------------------------------------------------------------===//
 // SI DAG Nodes
 //===----------------------------------------------------------------------===//
 
-// SMRD takes a 64bit memory address and can only add an 32bit offset
-def SIadd64bit32bit : SDNode<"ISD::ADD",
-  SDTypeProfile<1, 2, [SDTCisSameAs<0, 1>, SDTCisVT<0, i64>, SDTCisVT<2, i32>]>
->;
-
 def SIload_constant : SDNode<"AMDGPUISD::LOAD_CONSTANT",
-  SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisVT<1, i128>, SDTCisVT<2, i32>]>,
+  SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisVT<1, v4i32>, SDTCisVT<2, i32>]>,
                       [SDNPMayLoad, SDNPMemOperand]
 >;
 
 def SItbuffer_store : SDNode<"AMDGPUISD::TBUFFER_STORE_FORMAT",
   SDTypeProfile<0, 13,
-    [SDTCisVT<0, i128>,   // rsrc(SGPR)
+    [SDTCisVT<0, v4i32>,   // rsrc(SGPR)
      SDTCisVT<1, iAny>,   // vdata(VGPR)
      SDTCisVT<2, i32>,    // num_channels(imm)
      SDTCisVT<3, i32>,    // vaddr(VGPR)
@@ -41,13 +43,13 @@ def SItbuffer_store : SDNode<"AMDGPUISD::TBUFFER_STORE_FORMAT",
 >;
 
 def SIload_input : SDNode<"AMDGPUISD::LOAD_INPUT",
-  SDTypeProfile<1, 3, [SDTCisVT<0, v4f32>, SDTCisVT<1, i128>, SDTCisVT<2, i16>,
+  SDTypeProfile<1, 3, [SDTCisVT<0, v4f32>, SDTCisVT<1, v4i32>, SDTCisVT<2, i16>,
                        SDTCisVT<3, i32>]>
 >;
 
 class SDSample<string opcode> : SDNode <opcode,
   SDTypeProfile<1, 4, [SDTCisVT<0, v4f32>, SDTCisVT<2, v32i8>,
-                       SDTCisVT<3, i128>, SDTCisVT<4, i32>]>
+                       SDTCisVT<3, v4i32>, SDTCisVT<4, i32>]>
 >;
 
 def SIsample : SDSample<"AMDGPUISD::SAMPLE">;
@@ -55,6 +57,10 @@ def SIsampleb : SDSample<"AMDGPUISD::SAMPLEB">;
 def SIsampled : SDSample<"AMDGPUISD::SAMPLED">;
 def SIsamplel : SDSample<"AMDGPUISD::SAMPLEL">;
 
+def SIconstdata_ptr : SDNode<
+  "AMDGPUISD::CONST_DATA_PTR", SDTypeProfile <1, 0, [SDTCisVT<0, i64>]>
+>;
+
 // Transformation function, extract the lower 32bit of a 64bit immediate
 def LO32 : SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(N->getZExtValue() & 0xffffffff, MVT::i32);
@@ -75,15 +81,14 @@ def HI32f : SDNodeXForm<fpimm, [{
   return CurDAG->getTargetConstantFP(APFloat(APFloat::IEEEsingle, V), MVT::f32);
 }]>;
 
-def IMM8bitDWORD : ImmLeaf <
-  i32, [{
-    return (Imm & ~0x3FC) == 0;
-  }], SDNodeXForm<imm, [{
-    return CurDAG->getTargetConstant(
-      N->getZExtValue() >> 2, MVT::i32);
-  }]>
+def IMM8bitDWORD : PatLeaf <(imm),
+  [{return (N->getZExtValue() & ~0x3FC) == 0;}]
 >;
 
+def as_dword_i32imm : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(N->getZExtValue() >> 2, MVT::i32);
+}]>;
+
 def as_i1imm : SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(N->getZExtValue(), MVT::i1);
 }]>;
@@ -96,13 +101,33 @@ def as_i16imm : SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(N->getSExtValue(), MVT::i16);
 }]>;
 
+def as_i32imm: SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(N->getSExtValue(), MVT::i32);
+}]>;
+
+def IMM8bit : PatLeaf <(imm),
+  [{return isUInt<8>(N->getZExtValue());}]
+>;
+
 def IMM12bit : PatLeaf <(imm),
   [{return isUInt<12>(N->getZExtValue());}]
 >;
 
+def IMM16bit : PatLeaf <(imm),
+  [{return isUInt<16>(N->getZExtValue());}]
+>;
+
+def IMM32bit : PatLeaf <(imm),
+  [{return isUInt<32>(N->getZExtValue());}]
+>;
+
+def mubuf_vaddr_offset : PatFrag<
+  (ops node:$ptr, node:$offset, node:$imm_offset),
+  (add (add node:$ptr, node:$offset), node:$imm_offset)
+>;
+
 class InlineImm <ValueType vt> : PatLeaf <(vt imm), [{
-  return
-    (*(const SITargetLowering *)getTargetLowering()).analyzeImmediate(N) == 0;
+  return isInlineImmediate(N);
 }]>;
 
 class SGPRImm <dag frag> : PatLeaf<frag, [{
@@ -121,10 +146,27 @@ class SGPRImm <dag frag> : PatLeaf<frag, [{
   return false;
 }]>;
 
-def FRAMEri64 : Operand<iPTR> {
-  let MIOperandInfo = (ops SReg_32:$ptr, i32imm:$index);
+//===----------------------------------------------------------------------===//
+// Custom Operands
+//===----------------------------------------------------------------------===//
+
+def FRAMEri32 : Operand<iPTR> {
+  let MIOperandInfo = (ops i32:$ptr, i32imm:$index);
 }
 
+def sopp_brtarget : Operand<OtherVT> {
+  let EncoderMethod = "getSOPPBrEncoding";
+  let OperandType = "OPERAND_PCREL";
+}
+
+//===----------------------------------------------------------------------===//
+// Complex patterns
+//===----------------------------------------------------------------------===//
+
+def MUBUFAddr32 : ComplexPattern<i64, 9, "SelectMUBUFAddr32">;
+def MUBUFAddr64 : ComplexPattern<i64, 3, "SelectMUBUFAddr64">;
+def MUBUFScratch : ComplexPattern<i64, 4, "SelectMUBUFScratch">;
+
 //===----------------------------------------------------------------------===//
 // SI assembler operands
 //===----------------------------------------------------------------------===//
@@ -166,6 +208,12 @@ class SOP1_64 <bits<8> op, string opName, list<dag> pattern> : SOP1 <
   opName#" $dst, $src0", pattern
 >;
 
+// 64-bit input, 32-bit output.
+class SOP1_32_64 <bits<8> op, string opName, list<dag> pattern> : SOP1 <
+  op, (outs SReg_32:$dst), (ins SSrc_64:$src0),
+  opName#" $dst, $src0", pattern
+>;
+
 class SOP2_32 <bits<7> op, string opName, list<dag> pattern> : SOP2 <
   op, (outs SReg_32:$dst), (ins SSrc_32:$src0, SSrc_32:$src1),
   opName#" $dst, $src0, $src1", pattern
@@ -181,15 +229,17 @@ class SOP2_SHIFT_64 <bits<7> op, string opName, list<dag> pattern> : SOP2 <
   opName#" $dst, $src0, $src1", pattern
 >;
 
-class SOPC_32 <bits<7> op, string opName, list<dag> pattern> : SOPC <
-  op, (outs SCCReg:$dst), (ins SSrc_32:$src0, SSrc_32:$src1),
-  opName#" $dst, $src0, $src1", pattern
->;
 
-class SOPC_64 <bits<7> op, string opName, list<dag> pattern> : SOPC <
-  op, (outs SCCReg:$dst), (ins SSrc_64:$src0, SSrc_64:$src1),
-  opName#" $dst, $src0, $src1", pattern
->;
+class SOPC_Helper <bits<7> op, RegisterClass rc, ValueType vt,
+                    string opName, PatLeaf cond> : SOPC <
+  op, (outs SCCReg:$dst), (ins rc:$src0, rc:$src1),
+  opName#" $dst, $src0, $src1", []>;
+
+class SOPC_32<bits<7> op, string opName, PatLeaf cond = COND_NULL>
+  : SOPC_Helper<op, SSrc_32, i32, opName, cond>;
+
+class SOPC_64<bits<7> op, string opName, PatLeaf cond = COND_NULL>
+  : SOPC_Helper<op, SSrc_64, i64, opName, cond>;
 
 class SOPK_32 <bits<5> op, string opName, list<dag> pattern> : SOPK <
   op, (outs SReg_32:$dst), (ins i16imm:$src0),
@@ -205,7 +255,7 @@ multiclass SMRD_Helper <bits<5> op, string asm, RegisterClass baseClass,
                         RegisterClass dstClass> {
   def _IMM : SMRD <
     op, 1, (outs dstClass:$dst),
-    (ins baseClass:$sbase, i32imm:$offset),
+    (ins baseClass:$sbase, u32imm:$offset),
     asm#" $dst, $sbase, $offset", []
   >;
 
@@ -229,6 +279,66 @@ class VOP2_REV <string revOp, bit isOrig> {
   bit IsOrig = isOrig;
 }
 
+class SIMCInstr <string pseudo, int subtarget> {
+  string PseudoInstr = pseudo;
+  int Subtarget = subtarget;
+}
+
+class VOP3_Pseudo <dag outs, dag ins, list<dag> pattern, string opName> :
+  VOP3Common <outs, ins, "", pattern>,
+  VOP <opName>,
+  SIMCInstr<opName, SISubtarget.NONE> {
+  let isPseudo = 1;
+}
+
+class VOP3_Real_si <bits<9> op, dag outs, dag ins, string asm, string opName> :
+  VOP3 <op, outs, ins, asm, []>,
+  SIMCInstr<opName, SISubtarget.SI>;
+
+multiclass VOP3_m <bits<9> op, dag outs, dag ins, string asm, list<dag> pattern,
+                   string opName> {
+
+  def "" : VOP3_Pseudo <outs, ins, pattern, opName>;
+
+  def _si : VOP3_Real_si <op, outs, ins, asm, opName>;
+
+}
+
+multiclass VOP3_1_m <bits<8> op, dag outs, dag ins, string asm,
+                     list<dag> pattern, string opName> {
+
+  def "" : VOP3_Pseudo <outs, ins, pattern, opName>;
+
+  let src1 = 0, src1_modifiers = 0, src2 = 0, src2_modifiers = 0 in {
+
+    def _si : VOP3_Real_si <
+      {1, 1, op{6}, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
+      outs, ins, asm, opName
+    >;
+
+  } // src1 = 0, src1_modifiers = 0, src2 = 0, src2_modifiers = 0
+}
+
+multiclass VOP3_2_m <bits<6> op, dag outs, dag ins, string asm,
+                     list<dag> pattern, string opName, string revOp> {
+
+  def "" : VOP3_Pseudo <outs, ins, pattern, opName>;
+
+  let src2 = 0, src2_modifiers = 0 in {
+
+    def _si : VOP3_Real_si <
+        {1, 0, 0, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
+        outs, ins, asm, opName>,
+        VOP2_REV<revOp#"_e64", !eq(revOp, opName)>;
+
+  } // src2 = 0, src2_modifiers = 0
+}
+
+// This must always be right before the operand being input modified.
+def InputMods : OperandWithDefaultOps <i32, (ops (i32 0))> {
+  let PrintMethod = "printOperandAndMods";
+}
+
 multiclass VOP1_Helper <bits<8> op, RegisterClass drc, RegisterClass src,
                         string opName, list<dag> pattern> {
 
@@ -237,17 +347,11 @@ multiclass VOP1_Helper <bits<8> op, RegisterClass drc, RegisterClass src,
     opName#"_e32 $dst, $src0", pattern
   >, VOP <opName>;
 
-  def _e64 : VOP3 <
-    {1, 1, op{6}, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
+  defm _e64 : VOP3_1_m <
+    op,
     (outs drc:$dst),
-    (ins src:$src0,
-         i32imm:$abs, i32imm:$clamp,
-         i32imm:$omod, i32imm:$neg),
-    opName#"_e64 $dst, $src0, $abs, $clamp, $omod, $neg", []
-  >, VOP <opName> {
-    let src1 = SIOperand.ZERO;
-    let src2 = SIOperand.ZERO;
-  }
+    (ins InputMods:$src0_modifiers, src:$src0, i32imm:$clamp, i32imm:$omod),
+    opName#"_e64 $dst, $src0_modifiers, $clamp, $omod", [], opName>;
 }
 
 multiclass VOP1_32 <bits<8> op, string opName, list<dag> pattern>
@@ -269,16 +373,14 @@ multiclass VOP2_Helper <bits<6> op, RegisterClass vrc, RegisterClass arc,
     opName#"_e32 $dst, $src0, $src1", pattern
   >, VOP <opName>, VOP2_REV<revOp#"_e32", !eq(revOp, opName)>;
 
-  def _e64 : VOP3 <
-    {1, 0, 0, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
+  defm _e64 : VOP3_2_m <
+    op,
     (outs vrc:$dst),
-    (ins arc:$src0, arc:$src1,
-         i32imm:$abs, i32imm:$clamp,
-         i32imm:$omod, i32imm:$neg),
-    opName#"_e64 $dst, $src0, $src1, $abs, $clamp, $omod, $neg", []
-  >, VOP <opName>, VOP2_REV<revOp#"_e64", !eq(revOp, opName)> {
-    let src2 = SIOperand.ZERO;
-  }
+    (ins InputMods:$src0_modifiers, arc:$src0,
+         InputMods:$src1_modifiers, arc:$src1,
+         i32imm:$clamp, i32imm:$omod),
+    opName#"_e64 $dst, $src0_modifiers, $src1_modifiers, $clamp, $omod", [],
+    opName, revOp>;
 }
 
 multiclass VOP2_32 <bits<6> op, string opName, list<dag> pattern,
@@ -300,12 +402,13 @@ multiclass VOP2b_32 <bits<6> op, string opName, list<dag> pattern,
   def _e64 : VOP3b <
     {1, 0, 0, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
     (outs VReg_32:$dst),
-    (ins VSrc_32:$src0, VSrc_32:$src1,
-         i32imm:$abs, i32imm:$clamp,
-         i32imm:$omod, i32imm:$neg),
-    opName#"_e64 $dst, $src0, $src1, $abs, $clamp, $omod, $neg", []
+    (ins InputMods: $src0_modifiers, VSrc_32:$src0,
+         InputMods:$src1_modifiers, VSrc_32:$src1,
+         i32imm:$clamp, i32imm:$omod),
+    opName#"_e64 $dst, $src0_modifiers, $src1_modifiers, $clamp, $omod", []
   >, VOP <opName>, VOP2_REV<revOp#"_e64", !eq(revOp, opName)> {
-    let src2 = SIOperand.ZERO;
+    let src2 = 0;
+    let src2_modifiers = 0;
     /* the VOP2 variant puts the carry out into VCC, the VOP3 variant
        can write it into any SGPR. We currently don't use the carry out,
        so for now hardcode it to VCC as well */
@@ -314,25 +417,28 @@ multiclass VOP2b_32 <bits<6> op, string opName, list<dag> pattern,
 }
 
 multiclass VOPC_Helper <bits<8> op, RegisterClass vrc, RegisterClass arc,
-                        string opName, ValueType vt, PatLeaf cond> {
-
+                        string opName, ValueType vt, PatLeaf cond, bit defExec = 0> {
   def _e32 : VOPC <
     op, (ins arc:$src0, vrc:$src1),
     opName#"_e32 $dst, $src0, $src1", []
-  >, VOP <opName>;
+  >, VOP <opName> {
+    let Defs = !if(defExec, [EXEC], []);
+  }
 
   def _e64 : VOP3 <
     {0, op{7}, op{6}, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
     (outs SReg_64:$dst),
-    (ins arc:$src0, arc:$src1,
-         InstFlag:$abs, InstFlag:$clamp,
-         InstFlag:$omod, InstFlag:$neg),
-    opName#"_e64 $dst, $src0, $src1, $abs, $clamp, $omod, $neg",
+    (ins InputMods:$src0_modifiers, arc:$src0,
+         InputMods:$src1_modifiers, arc:$src1,
+         InstFlag:$clamp, InstFlag:$omod),
+    opName#"_e64 $dst, $src0_modifiers, $src1_modifiers, $clamp, $omod",
     !if(!eq(!cast<string>(cond), "COND_NULL"), []<dag>,
       [(set SReg_64:$dst, (i1 (setcc (vt arc:$src0), arc:$src1, cond)))]
     )
   >, VOP <opName> {
-    let src2 = SIOperand.ZERO;
+    let Defs = !if(defExec, [EXEC], []);
+    let src2 = 0;
+    let src2_modifiers = 0;
   }
 }
 
@@ -344,76 +450,172 @@ multiclass VOPC_64 <bits<8> op, string opName,
   ValueType vt = untyped, PatLeaf cond = COND_NULL>
   : VOPC_Helper <op, VReg_64, VSrc_64, opName, vt, cond>;
 
-class VOP3_32 <bits<9> op, string opName, list<dag> pattern> : VOP3 <
+multiclass VOPCX_32 <bits<8> op, string opName,
+  ValueType vt = untyped, PatLeaf cond = COND_NULL>
+  : VOPC_Helper <op, VReg_32, VSrc_32, opName, vt, cond, 1>;
+
+multiclass VOPCX_64 <bits<8> op, string opName,
+  ValueType vt = untyped, PatLeaf cond = COND_NULL>
+  : VOPC_Helper <op, VReg_64, VSrc_64, opName, vt, cond, 1>;
+
+multiclass VOP3_32 <bits<9> op, string opName, list<dag> pattern> : VOP3_m <
   op, (outs VReg_32:$dst),
-  (ins VSrc_32:$src0, VSrc_32:$src1, VSrc_32:$src2,
-   InstFlag:$abs, InstFlag:$clamp, InstFlag:$omod, InstFlag:$neg),
-  opName#" $dst, $src0, $src1, $src2, $abs, $clamp, $omod, $neg", pattern
->, VOP <opName>;
+  (ins InputMods: $src0_modifiers, VSrc_32:$src0, InputMods:$src1_modifiers,
+   VSrc_32:$src1, InputMods:$src2_modifiers, VSrc_32:$src2,
+   InstFlag:$clamp, InstFlag:$omod),
+  opName#" $dst, $src0_modifiers, $src1, $src2, $clamp, $omod", pattern, opName
+>;
 
-class VOP3_64_Shift <bits <9> op, string opName, list<dag> pattern> : VOP3 <
+class VOP3_64_32 <bits <9> op, string opName, list<dag> pattern> : VOP3 <
   op, (outs VReg_64:$dst),
   (ins VSrc_64:$src0, VSrc_32:$src1),
   opName#" $dst, $src0, $src1", pattern
 >, VOP <opName> {
 
-  let src2 = SIOperand.ZERO;
-  let abs = 0;
+  let src2 = 0;
+  let src2_modifiers = 0;
+  let src0_modifiers = 0;
   let clamp = 0;
   let omod = 0;
-  let neg = 0;
 }
 
 class VOP3_64 <bits<9> op, string opName, list<dag> pattern> : VOP3 <
   op, (outs VReg_64:$dst),
-  (ins VSrc_64:$src0, VSrc_64:$src1, VSrc_64:$src2,
+  (ins InputMods:$src0_modifiers, VSrc_64:$src0,
+       InputMods:$src1_modifiers, VSrc_64:$src1,
+       InputMods:$src2_modifiers, VSrc_64:$src2,
+       InstFlag:$clamp, InstFlag:$omod),
+  opName#" $dst, $src0_modifiers, $src1_modifiers, $src2_modifiers, $clamp, $omod", pattern
+>, VOP <opName>;
+
+
+class VOP3b_Helper <bits<9> op, RegisterClass vrc, RegisterClass arc,
+                    string opName, list<dag> pattern> : VOP3 <
+  op, (outs vrc:$dst0, SReg_64:$dst1),
+  (ins arc:$src0, arc:$src1, arc:$src2,
    InstFlag:$abs, InstFlag:$clamp, InstFlag:$omod, InstFlag:$neg),
-  opName#" $dst, $src0, $src1, $src2, $abs, $clamp, $omod, $neg", pattern
+  opName#" $dst0, $dst1, $src0, $src1, $src2, $abs, $clamp, $omod, $neg", pattern
 >, VOP <opName>;
 
+
+class VOP3b_64 <bits<9> op, string opName, list<dag> pattern> :
+  VOP3b_Helper <op, VReg_64, VSrc_64, opName, pattern>;
+
+class VOP3b_32 <bits<9> op, string opName, list<dag> pattern> :
+  VOP3b_Helper <op, VReg_32, VSrc_32, opName, pattern>;
+
 //===----------------------------------------------------------------------===//
 // Vector I/O classes
 //===----------------------------------------------------------------------===//
 
-class DS_Load_Helper <bits<8> op, string asm, RegisterClass regClass> : DS <
+class DS_1A <bits<8> op, dag outs, dag ins, string asm, list<dag> pat> :
+    DS <op, outs, ins, asm, pat> {
+  bits<16> offset;
+
+  // Single load interpret the 2 i8imm operands as a single i16 offset.
+  let offset0 = offset{7-0};
+  let offset1 = offset{15-8};
+}
+
+class DS_Load_Helper <bits<8> op, string asm, RegisterClass regClass> : DS_1A <
   op,
   (outs regClass:$vdst),
-  (ins i1imm:$gds, VReg_32:$addr, VReg_32:$data0, VReg_32:$data1,
-       i8imm:$offset0, i8imm:$offset1),
-  asm#" $vdst, $gds, $addr, $data0, $data1, $offset0, $offset1, [M0]",
+  (ins i1imm:$gds, VReg_32:$addr, u16imm:$offset),
+  asm#" $vdst, $addr, $offset, [M0]",
   []> {
+  let data0 = 0;
+  let data1 = 0;
+  let mayLoad = 1;
+  let mayStore = 0;
+}
+
+class DS_Load2_Helper <bits<8> op, string asm, RegisterClass regClass> : DS <
+  op,
+  (outs regClass:$vdst),
+  (ins i1imm:$gds, VReg_32:$addr, u8imm:$offset0, u8imm:$offset1),
+  asm#" $gds, $vdst, $addr, $offset0, $offset1, [M0]",
+  []> {
+  let data0 = 0;
+  let data1 = 0;
   let mayLoad = 1;
   let mayStore = 0;
 }
 
-class DS_Store_Helper <bits<8> op, string asm, RegisterClass regClass> : DS <
+class DS_Store_Helper <bits<8> op, string asm, RegisterClass regClass> : DS_1A <
   op,
   (outs),
-  (ins i1imm:$gds, VReg_32:$addr, VReg_32:$data0, VReg_32:$data1,
-       i8imm:$offset0, i8imm:$offset1),
-  asm#" $gds, $addr, $data0, $data1, $offset0, $offset1, [M0]",
+  (ins i1imm:$gds, VReg_32:$addr, regClass:$data0, u16imm:$offset),
+  asm#" $addr, $data0, $offset [M0]",
+  []> {
+  let data1 = 0;
+  let mayStore = 1;
+  let mayLoad = 0;
+  let vdst = 0;
+}
+
+class DS_Store2_Helper <bits<8> op, string asm, RegisterClass regClass> : DS_1A <
+  op,
+  (outs),
+  (ins i1imm:$gds, VReg_32:$addr, regClass:$data0, u8imm:$offset0, u8imm:$offset1),
+  asm#" $addr, $data0, $data1, $offset0, $offset1 [M0]",
   []> {
   let mayStore = 1;
   let mayLoad = 0;
   let vdst = 0;
 }
 
-class DS_1A1D_RET <bits<8> op, string asm, RegisterClass rc> : DS <
+// 1 address, 1 data.
+class DS_1A1D_RET <bits<8> op, string asm, RegisterClass rc> : DS_1A <
+  op,
+  (outs rc:$vdst),
+  (ins i1imm:$gds, VReg_32:$addr, rc:$data0, u16imm:$offset),
+  asm#" $vdst, $addr, $data0, $offset, [M0]",
+  []> {
+
+  let data1 = 0;
+  let mayStore = 1;
+  let mayLoad = 1;
+}
+
+// 1 address, 2 data.
+class DS_1A2D_RET <bits<8> op, string asm, RegisterClass rc> : DS_1A <
   op,
   (outs rc:$vdst),
-  (ins i1imm:$gds, VReg_32:$addr, VReg_32:$data0, i8imm:$offset0,
-       i8imm:$offset1),
-  asm#" $gds, $vdst, $addr, $data0, $offset0, $offset1, [M0]",
+  (ins i1imm:$gds, VReg_32:$addr, rc:$data0, rc:$data1, u16imm:$offset),
+  asm#" $vdst, $addr, $data0, $data1, $offset, [M0]",
   []> {
   let mayStore = 1;
   let mayLoad = 1;
+}
+
+// 1 address, 2 data.
+class DS_1A2D_NORET <bits<8> op, string asm, RegisterClass rc> : DS_1A <
+  op,
+  (outs),
+  (ins i1imm:$gds, VReg_32:$addr, rc:$data0, rc:$data1, u16imm:$offset),
+  asm#" $addr, $data0, $data1, $offset, [M0]",
+  []> {
+  let mayStore = 1;
+  let mayLoad = 1;
+}
+
+// 1 address, 1 data.
+class DS_1A1D_NORET <bits<8> op, string asm, RegisterClass rc> : DS_1A <
+  op,
+  (outs),
+  (ins i1imm:$gds, VReg_32:$addr, rc:$data0, u16imm:$offset),
+  asm#" $addr, $data0, $offset, [M0]",
+  []> {
+
   let data1 = 0;
+  let mayStore = 1;
+  let mayLoad = 1;
 }
 
 class MTBUF_Store_Helper <bits<3> op, string asm, RegisterClass regClass> : MTBUF <
   op,
   (outs),
-  (ins regClass:$vdata, i16imm:$offset, i1imm:$offen, i1imm:$idxen, i1imm:$glc,
+  (ins regClass:$vdata, u16imm:$offset, i1imm:$offen, i1imm:$idxen, i1imm:$glc,
    i1imm:$addr64, i8imm:$dfmt, i8imm:$nfmt, VReg_32:$vaddr,
    SReg_128:$srsrc, i1imm:$slc, i1imm:$tfe, SSrc_32:$soffset),
   asm#" $vdata, $offset, $offen, $idxen, $glc, $addr64, $dfmt,"
@@ -423,32 +625,34 @@ class MTBUF_Store_Helper <bits<3> op, string asm, RegisterClass regClass> : MTBU
   let mayLoad = 0;
 }
 
-multiclass MUBUF_Load_Helper <bits<7> op, string asm, RegisterClass regClass> {
+multiclass MUBUF_Load_Helper <bits<7> op, string asm, RegisterClass regClass,
+                              ValueType load_vt = i32,
+                              SDPatternOperator ld = null_frag> {
 
   let lds = 0, mayLoad = 1 in {
 
     let addr64 = 0 in {
 
-      let offen = 0, idxen = 0 in {
+      let offen = 0, idxen = 0, vaddr = 0 in {
         def _OFFSET : MUBUF <op, (outs regClass:$vdata),
-                             (ins SReg_128:$srsrc, VReg_32:$vaddr,
-                             i16imm:$offset, SSrc_32:$soffset, i1imm:$glc,
+                             (ins SReg_128:$srsrc,
+                             u16imm:$offset, SSrc_32:$soffset, i1imm:$glc,
                              i1imm:$slc, i1imm:$tfe),
                              asm#" $vdata, $srsrc + $offset + $soffset, glc=$glc, slc=$slc, tfe=$tfe", []>;
       }
 
-      let offen = 1, idxen = 0, offset = 0 in {
+      let offen = 1, idxen = 0  in {
         def _OFFEN  : MUBUF <op, (outs regClass:$vdata),
                              (ins SReg_128:$srsrc, VReg_32:$vaddr,
-                             SSrc_32:$soffset, i1imm:$glc, i1imm:$slc,
+                             SSrc_32:$soffset, u16imm:$offset, i1imm:$glc, i1imm:$slc,
                              i1imm:$tfe),
-                             asm#" $vdata, $srsrc + $vaddr + $soffset, glc=$glc, slc=$slc, tfe=$tfe", []>;
+                             asm#" $vdata, $srsrc + $vaddr + $soffset + $offset, glc=$glc, slc=$slc, tfe=$tfe", []>;
       }
 
       let offen = 0, idxen = 1 in {
         def _IDXEN  : MUBUF <op, (outs regClass:$vdata),
                              (ins SReg_128:$srsrc, VReg_32:$vaddr,
-                             i16imm:$offset, SSrc_32:$soffset, i1imm:$glc,
+                             u16imm:$offset, SSrc_32:$soffset, i1imm:$glc,
                              i1imm:$slc, i1imm:$tfe),
                              asm#" $vdata, $srsrc[$vaddr] + $offset + $soffset, glc=$glc, slc=$slc, tfe=$tfe", []>;
       }
@@ -464,36 +668,54 @@ multiclass MUBUF_Load_Helper <bits<7> op, string asm, RegisterClass regClass> {
 
     let offen = 0, idxen = 0, addr64 = 1, glc = 0, slc = 0, tfe = 0, soffset = 128 /* ZERO */ in {
       def _ADDR64 : MUBUF <op, (outs regClass:$vdata),
-                           (ins SReg_128:$srsrc, VReg_64:$vaddr, i16imm:$offset),
-                           asm#" $vdata, $srsrc + $vaddr + $offset", []>;
+                           (ins SReg_128:$srsrc, VReg_64:$vaddr, u16imm:$offset),
+                           asm#" $vdata, $srsrc + $vaddr + $offset",
+                           [(set load_vt:$vdata, (ld (MUBUFAddr64 v4i32:$srsrc,
+                                                  i64:$vaddr, u16imm:$offset)))]>;
     }
   }
 }
 
-class MUBUF_Store_Helper <bits<7> op, string name, RegisterClass vdataClass> :
-    MUBUF <op, (outs), (ins vdataClass:$vdata, SReg_128:$srsrc, VReg_64:$vaddr,
-                            i16imm:$offset),
-          name#" $vdata, $srsrc + $vaddr + $offset",
-         []> {
-
-  let mayLoad = 0;
-  let mayStore = 1;
+multiclass MUBUF_Store_Helper <bits<7> op, string name, RegisterClass vdataClass,
+                          ValueType store_vt, SDPatternOperator st> {
+
+  def "" : MUBUF <
+    op, (outs),
+    (ins vdataClass:$vdata, SReg_128:$srsrc, VReg_32:$vaddr, SSrc_32:$soffset,
+         u16imm:$offset, i1imm:$offen, i1imm:$idxen, i1imm:$glc, i1imm:$slc,
+         i1imm:$tfe),
+    name#" $vdata, $srsrc, $vaddr, $soffset, $offset $offen $idxen $glc $slc $tfe",
+    []
+  > {
+    let addr64 = 0;
+  }
 
-  // Encoding
-  let offen = 0;
-  let idxen = 0;
-  let glc = 0;
-  let addr64 = 1;
-  let lds = 0;
-  let slc = 0;
-  let tfe = 0;
-  let soffset = 128; // ZERO
+  def _ADDR64 : MUBUF <
+    op, (outs),
+    (ins vdataClass:$vdata, SReg_128:$srsrc, VReg_64:$vaddr, u16imm:$offset),
+    name#" $vdata, $srsrc + $vaddr + $offset",
+    [(st store_vt:$vdata,
+     (MUBUFAddr64 v4i32:$srsrc, i64:$vaddr, u16imm:$offset))]> {
+
+      let mayLoad = 0;
+      let mayStore = 1;
+
+      // Encoding
+      let offen = 0;
+      let idxen = 0;
+      let glc = 0;
+      let addr64 = 1;
+      let lds = 0;
+      let slc = 0;
+      let tfe = 0;
+      let soffset = 128; // ZERO
+   }
 }
 
 class MTBUF_Load_Helper <bits<3> op, string asm, RegisterClass regClass> : MTBUF <
   op,
   (outs regClass:$dst),
-  (ins i16imm:$offset, i1imm:$offen, i1imm:$idxen, i1imm:$glc, i1imm:$addr64,
+  (ins u16imm:$offset, i1imm:$offen, i1imm:$idxen, i1imm:$glc, i1imm:$addr64,
        i8imm:$dfmt, i8imm:$nfmt, VReg_32:$vaddr, SReg_128:$srsrc,
        i1imm:$slc, i1imm:$tfe, SSrc_32:$soffset),
   asm#" $dst, $offset, $offen, $idxen, $glc, $addr64, $dfmt,"
@@ -581,6 +803,53 @@ multiclass MIMG_Sampler <bits<7> op, string asm> {
   defm _V4 : MIMG_Sampler_Src_Helper<op, asm, VReg_128, 4>;
 }
 
+class MIMG_Gather_Helper <bits<7> op, string asm,
+                          RegisterClass dst_rc,
+                          RegisterClass src_rc> : MIMG <
+  op,
+  (outs dst_rc:$vdata),
+  (ins i32imm:$dmask, i1imm:$unorm, i1imm:$glc, i1imm:$da, i1imm:$r128,
+       i1imm:$tfe, i1imm:$lwe, i1imm:$slc, src_rc:$vaddr,
+       SReg_256:$srsrc, SReg_128:$ssamp),
+  asm#" $vdata, $dmask, $unorm, $glc, $da, $r128,"
+     #" $tfe, $lwe, $slc, $vaddr, $srsrc, $ssamp",
+  []> {
+  let mayLoad = 1;
+  let mayStore = 0;
+
+  // DMASK was repurposed for GATHER4. 4 components are always
+  // returned and DMASK works like a swizzle - it selects
+  // the component to fetch. The only useful DMASK values are
+  // 1=red, 2=green, 4=blue, 8=alpha. (e.g. 1 returns
+  // (red,red,red,red) etc.) The ISA document doesn't mention
+  // this.
+  // Therefore, disable all code which updates DMASK by setting these two:
+  let MIMG = 0;
+  let hasPostISelHook = 0;
+}
+
+multiclass MIMG_Gather_Src_Helper <bits<7> op, string asm,
+                                    RegisterClass dst_rc,
+                                    int channels> {
+  def _V1 : MIMG_Gather_Helper <op, asm, dst_rc, VReg_32>,
+            MIMG_Mask<asm#"_V1", channels>;
+  def _V2 : MIMG_Gather_Helper <op, asm, dst_rc, VReg_64>,
+            MIMG_Mask<asm#"_V2", channels>;
+  def _V4 : MIMG_Gather_Helper <op, asm, dst_rc, VReg_128>,
+            MIMG_Mask<asm#"_V4", channels>;
+  def _V8 : MIMG_Gather_Helper <op, asm, dst_rc, VReg_256>,
+            MIMG_Mask<asm#"_V8", channels>;
+  def _V16 : MIMG_Gather_Helper <op, asm, dst_rc, VReg_512>,
+            MIMG_Mask<asm#"_V16", channels>;
+}
+
+multiclass MIMG_Gather <bits<7> op, string asm> {
+  defm _V1 : MIMG_Gather_Src_Helper<op, asm, VReg_32, 1>;
+  defm _V2 : MIMG_Gather_Src_Helper<op, asm, VReg_64, 2>;
+  defm _V3 : MIMG_Gather_Src_Helper<op, asm, VReg_96, 3>;
+  defm _V4 : MIMG_Gather_Src_Helper<op, asm, VReg_128, 4>;
+}
+
 //===----------------------------------------------------------------------===//
 // Vector instruction mappings
 //===----------------------------------------------------------------------===//
@@ -594,6 +863,15 @@ def getVOPe64 : InstrMapping {
   let ValueCols = [["8"]];
 }
 
+// Maps an opcode in e64 form to its e32 equivalent
+def getVOPe32 : InstrMapping {
+  let FilterClass = "VOP";
+  let RowFields = ["OpName"];
+  let ColFields = ["Size"];
+  let KeyCol = ["8"];
+  let ValueCols = [["4"]];
+}
+
 // Maps an original opcode to its commuted version
 def getCommuteRev : InstrMapping {
   let FilterClass = "VOP2_REV";
@@ -620,4 +898,20 @@ def getCommuteOrig : InstrMapping {
   let ValueCols = [["1"]];
 }
 
+def isDS : InstrMapping {
+  let FilterClass = "DS";
+  let RowFields = ["Inst"];
+  let ColFields = ["Size"];
+  let KeyCol = ["8"];
+  let ValueCols = [["8"]];
+}
+
+def getMCOpcode : InstrMapping {
+  let FilterClass = "SIMCInstr";
+  let RowFields = ["PseudoInstr"];
+  let ColFields = ["Subtarget"];
+  let KeyCol = [!cast<string>(SISubtarget.NONE)];
+  let ValueCols = [[!cast<string>(SISubtarget.SI)]];
+}
+
 include "SIInstructions.td"
diff --git a/contrib/llvm/lib/Target/R600/SIInstructions.td b/contrib/llvm/lib/Target/R600/SIInstructions.td
index 2ca6a95..aecd847 100644
--- a/contrib/llvm/lib/Target/R600/SIInstructions.td
+++ b/contrib/llvm/lib/Target/R600/SIInstructions.td
@@ -22,14 +22,66 @@ def InterpSlot : Operand<i32> {
   let PrintMethod = "printInterpSlot";
 }
 
-def SendMsgImm : Operand<i32>;
+def SendMsgImm : Operand<i32> {
+  let PrintMethod = "printSendMsg";
+}
 
 def isSI : Predicate<"Subtarget.getGeneration() "
                       ">= AMDGPUSubtarget::SOUTHERN_ISLANDS">;
 
+def isCI : Predicate<"Subtarget.getGeneration() "
+                      ">= AMDGPUSubtarget::SEA_ISLANDS">;
+
+def isCFDepth0 : Predicate<"isCFDepth0()">;
+
 def WAIT_FLAG : InstFlag<"printWaitFlag">;
 
-let Predicates = [isSI] in {
+let SubtargetPredicate = isSI in {
+let OtherPredicates  = [isCFDepth0] in {
+
+//===----------------------------------------------------------------------===//
+// SMRD Instructions
+//===----------------------------------------------------------------------===//
+
+let mayLoad = 1 in {
+
+// We are using the SGPR_32 and not the SReg_32 register class for 32-bit
+// SMRD instructions, because the SGPR_32 register class does not include M0
+// and writing to M0 from an SMRD instruction will hang the GPU.
+defm S_LOAD_DWORD : SMRD_Helper <0x00, "S_LOAD_DWORD", SReg_64, SGPR_32>;
+defm S_LOAD_DWORDX2 : SMRD_Helper <0x01, "S_LOAD_DWORDX2", SReg_64, SReg_64>;
+defm S_LOAD_DWORDX4 : SMRD_Helper <0x02, "S_LOAD_DWORDX4", SReg_64, SReg_128>;
+defm S_LOAD_DWORDX8 : SMRD_Helper <0x03, "S_LOAD_DWORDX8", SReg_64, SReg_256>;
+defm S_LOAD_DWORDX16 : SMRD_Helper <0x04, "S_LOAD_DWORDX16", SReg_64, SReg_512>;
+
+defm S_BUFFER_LOAD_DWORD : SMRD_Helper <
+  0x08, "S_BUFFER_LOAD_DWORD", SReg_128, SGPR_32
+>;
+
+defm S_BUFFER_LOAD_DWORDX2 : SMRD_Helper <
+  0x09, "S_BUFFER_LOAD_DWORDX2", SReg_128, SReg_64
+>;
+
+defm S_BUFFER_LOAD_DWORDX4 : SMRD_Helper <
+  0x0a, "S_BUFFER_LOAD_DWORDX4", SReg_128, SReg_128
+>;
+
+defm S_BUFFER_LOAD_DWORDX8 : SMRD_Helper <
+  0x0b, "S_BUFFER_LOAD_DWORDX8", SReg_128, SReg_256
+>;
+
+defm S_BUFFER_LOAD_DWORDX16 : SMRD_Helper <
+  0x0c, "S_BUFFER_LOAD_DWORDX16", SReg_128, SReg_512
+>;
+
+} // mayLoad = 1
+
+//def S_MEMTIME : SMRD_ <0x0000001e, "S_MEMTIME", []>;
+//def S_DCACHE_INV : SMRD_ <0x0000001f, "S_DCACHE_INV", []>;
+
+//===----------------------------------------------------------------------===//
+// SOP1 Instructions
+//===----------------------------------------------------------------------===//
 
 let neverHasSideEffects = 1 in {
 
@@ -40,33 +92,58 @@ def S_CMOV_B32 : SOP1_32 <0x00000005, "S_CMOV_B32", []>;
 def S_CMOV_B64 : SOP1_64 <0x00000006, "S_CMOV_B64", []>;
 } // End isMoveImm = 1
 
-def S_NOT_B32 : SOP1_32 <0x00000007, "S_NOT_B32", []>;
-def S_NOT_B64 : SOP1_64 <0x00000008, "S_NOT_B64", []>;
+def S_NOT_B32 : SOP1_32 <0x00000007, "S_NOT_B32",
+  [(set i32:$dst, (not i32:$src0))]
+>;
+
+def S_NOT_B64 : SOP1_64 <0x00000008, "S_NOT_B64",
+  [(set i64:$dst, (not i64:$src0))]
+>;
 def S_WQM_B32 : SOP1_32 <0x00000009, "S_WQM_B32", []>;
 def S_WQM_B64 : SOP1_64 <0x0000000a, "S_WQM_B64", []>;
-def S_BREV_B32 : SOP1_32 <0x0000000b, "S_BREV_B32", []>;
+def S_BREV_B32 : SOP1_32 <0x0000000b, "S_BREV_B32",
+  [(set i32:$dst, (AMDGPUbrev i32:$src0))]
+>;
 def S_BREV_B64 : SOP1_64 <0x0000000c, "S_BREV_B64", []>;
 } // End neverHasSideEffects = 1
 
 ////def S_BCNT0_I32_B32 : SOP1_BCNT0 <0x0000000d, "S_BCNT0_I32_B32", []>;
 ////def S_BCNT0_I32_B64 : SOP1_BCNT0 <0x0000000e, "S_BCNT0_I32_B64", []>;
-////def S_BCNT1_I32_B32 : SOP1_BCNT1 <0x0000000f, "S_BCNT1_I32_B32", []>;
-////def S_BCNT1_I32_B64 : SOP1_BCNT1 <0x00000010, "S_BCNT1_I32_B64", []>;
-////def S_FF0_I32_B32 : SOP1_FF0 <0x00000011, "S_FF0_I32_B32", []>;
+def S_BCNT1_I32_B32 : SOP1_32 <0x0000000f, "S_BCNT1_I32_B32",
+  [(set i32:$dst, (ctpop i32:$src0))]
+>;
+def S_BCNT1_I32_B64 : SOP1_32_64 <0x00000010, "S_BCNT1_I32_B64", []>;
+
+////def S_FF0_I32_B32 : SOP1_32 <0x00000011, "S_FF0_I32_B32", []>;
 ////def S_FF0_I32_B64 : SOP1_FF0 <0x00000012, "S_FF0_I32_B64", []>;
-////def S_FF1_I32_B32 : SOP1_FF1 <0x00000013, "S_FF1_I32_B32", []>;
+def S_FF1_I32_B32 : SOP1_32 <0x00000013, "S_FF1_I32_B32",
+  [(set i32:$dst, (cttz_zero_undef i32:$src0))]
+>;
 ////def S_FF1_I32_B64 : SOP1_FF1 <0x00000014, "S_FF1_I32_B64", []>;
-//def S_FLBIT_I32_B32 : SOP1_32 <0x00000015, "S_FLBIT_I32_B32", []>;
+
+def S_FLBIT_I32_B32 : SOP1_32 <0x00000015, "S_FLBIT_I32_B32",
+  [(set i32:$dst, (ctlz_zero_undef i32:$src0))]
+>;
+
 //def S_FLBIT_I32_B64 : SOP1_32 <0x00000016, "S_FLBIT_I32_B64", []>;
 def S_FLBIT_I32 : SOP1_32 <0x00000017, "S_FLBIT_I32", []>;
 //def S_FLBIT_I32_I64 : SOP1_32 <0x00000018, "S_FLBIT_I32_I64", []>;
-//def S_SEXT_I32_I8 : SOP1_32 <0x00000019, "S_SEXT_I32_I8", []>;
-//def S_SEXT_I32_I16 : SOP1_32 <0x0000001a, "S_SEXT_I32_I16", []>;
+def S_SEXT_I32_I8 : SOP1_32 <0x00000019, "S_SEXT_I32_I8",
+  [(set i32:$dst, (sext_inreg i32:$src0, i8))]
+>;
+def S_SEXT_I32_I16 : SOP1_32 <0x0000001a, "S_SEXT_I32_I16",
+  [(set i32:$dst, (sext_inreg i32:$src0, i16))]
+>;
+
 ////def S_BITSET0_B32 : SOP1_BITSET0 <0x0000001b, "S_BITSET0_B32", []>;
 ////def S_BITSET0_B64 : SOP1_BITSET0 <0x0000001c, "S_BITSET0_B64", []>;
 ////def S_BITSET1_B32 : SOP1_BITSET1 <0x0000001d, "S_BITSET1_B32", []>;
 ////def S_BITSET1_B64 : SOP1_BITSET1 <0x0000001e, "S_BITSET1_B64", []>;
-def S_GETPC_B64 : SOP1_64 <0x0000001f, "S_GETPC_B64", []>;
+def S_GETPC_B64 : SOP1 <
+  0x0000001f, (outs SReg_64:$dst), (ins), "S_GETPC_B64 $dst", []
+> {
+  let SSRC0 = 0;
+}
 def S_SETPC_B64 : SOP1_64 <0x00000020, "S_SETPC_B64", []>;
 def S_SWAPPC_B64 : SOP1_64 <0x00000021, "S_SWAPPC_B64", []>;
 def S_RFE_B64 : SOP1_64 <0x00000022, "S_RFE_B64", []>;
@@ -94,6 +171,150 @@ def S_MOVRELD_B64 : SOP1_64 <0x00000031, "S_MOVRELD_B64", []>;
 def S_MOV_REGRD_B32 : SOP1_32 <0x00000033, "S_MOV_REGRD_B32", []>;
 def S_ABS_I32 : SOP1_32 <0x00000034, "S_ABS_I32", []>;
 def S_MOV_FED_B32 : SOP1_32 <0x00000035, "S_MOV_FED_B32", []>;
+
+//===----------------------------------------------------------------------===//
+// SOP2 Instructions
+//===----------------------------------------------------------------------===//
+
+let Defs = [SCC] in { // Carry out goes to SCC
+let isCommutable = 1 in {
+def S_ADD_U32 : SOP2_32 <0x00000000, "S_ADD_U32", []>;
+def S_ADD_I32 : SOP2_32 <0x00000002, "S_ADD_I32",
+  [(set i32:$dst, (add SSrc_32:$src0, SSrc_32:$src1))]
+>;
+} // End isCommutable = 1
+
+def S_SUB_U32 : SOP2_32 <0x00000001, "S_SUB_U32", []>;
+def S_SUB_I32 : SOP2_32 <0x00000003, "S_SUB_I32",
+  [(set i32:$dst, (sub SSrc_32:$src0, SSrc_32:$src1))]
+>;
+
+let Uses = [SCC] in { // Carry in comes from SCC
+let isCommutable = 1 in {
+def S_ADDC_U32 : SOP2_32 <0x00000004, "S_ADDC_U32",
+  [(set i32:$dst, (adde (i32 SSrc_32:$src0), (i32 SSrc_32:$src1)))]>;
+} // End isCommutable = 1
+
+def S_SUBB_U32 : SOP2_32 <0x00000005, "S_SUBB_U32",
+  [(set i32:$dst, (sube (i32 SSrc_32:$src0), (i32 SSrc_32:$src1)))]>;
+} // End Uses = [SCC]
+} // End Defs = [SCC]
+
+def S_MIN_I32 : SOP2_32 <0x00000006, "S_MIN_I32",
+  [(set i32:$dst, (AMDGPUsmin i32:$src0, i32:$src1))]
+>;
+def S_MIN_U32 : SOP2_32 <0x00000007, "S_MIN_U32",
+  [(set i32:$dst, (AMDGPUumin i32:$src0, i32:$src1))]
+>;
+def S_MAX_I32 : SOP2_32 <0x00000008, "S_MAX_I32",
+  [(set i32:$dst, (AMDGPUsmax i32:$src0, i32:$src1))]
+>;
+def S_MAX_U32 : SOP2_32 <0x00000009, "S_MAX_U32",
+  [(set i32:$dst, (AMDGPUumax i32:$src0, i32:$src1))]
+>;
+
+def S_CSELECT_B32 : SOP2 <
+  0x0000000a, (outs SReg_32:$dst),
+  (ins SReg_32:$src0, SReg_32:$src1, SCCReg:$scc), "S_CSELECT_B32",
+  []
+>;
+
+def S_CSELECT_B64 : SOP2_64 <0x0000000b, "S_CSELECT_B64", []>;
+
+def S_AND_B32 : SOP2_32 <0x0000000e, "S_AND_B32",
+  [(set i32:$dst, (and i32:$src0, i32:$src1))]
+>;
+
+def S_AND_B64 : SOP2_64 <0x0000000f, "S_AND_B64",
+  [(set i64:$dst, (and i64:$src0, i64:$src1))]
+>;
+
+def S_OR_B32 : SOP2_32 <0x00000010, "S_OR_B32",
+  [(set i32:$dst, (or i32:$src0, i32:$src1))]
+>;
+
+def S_OR_B64 : SOP2_64 <0x00000011, "S_OR_B64",
+  [(set i64:$dst, (or i64:$src0, i64:$src1))]
+>;
+
+def S_XOR_B32 : SOP2_32 <0x00000012, "S_XOR_B32",
+  [(set i32:$dst, (xor i32:$src0, i32:$src1))]
+>;
+
+def S_XOR_B64 : SOP2_64 <0x00000013, "S_XOR_B64",
+  [(set i64:$dst, (xor i64:$src0, i64:$src1))]
+>;
+def S_ANDN2_B32 : SOP2_32 <0x00000014, "S_ANDN2_B32", []>;
+def S_ANDN2_B64 : SOP2_64 <0x00000015, "S_ANDN2_B64", []>;
+def S_ORN2_B32 : SOP2_32 <0x00000016, "S_ORN2_B32", []>;
+def S_ORN2_B64 : SOP2_64 <0x00000017, "S_ORN2_B64", []>;
+def S_NAND_B32 : SOP2_32 <0x00000018, "S_NAND_B32", []>;
+def S_NAND_B64 : SOP2_64 <0x00000019, "S_NAND_B64", []>;
+def S_NOR_B32 : SOP2_32 <0x0000001a, "S_NOR_B32", []>;
+def S_NOR_B64 : SOP2_64 <0x0000001b, "S_NOR_B64", []>;
+def S_XNOR_B32 : SOP2_32 <0x0000001c, "S_XNOR_B32", []>;
+def S_XNOR_B64 : SOP2_64 <0x0000001d, "S_XNOR_B64", []>;
+
+// Use added complexity so these patterns are preferred to the VALU patterns.
+let AddedComplexity = 1 in {
+
+def S_LSHL_B32 : SOP2_32 <0x0000001e, "S_LSHL_B32",
+  [(set i32:$dst, (shl i32:$src0, i32:$src1))]
+>;
+def S_LSHL_B64 : SOP2_SHIFT_64 <0x0000001f, "S_LSHL_B64",
+  [(set i64:$dst, (shl i64:$src0, i32:$src1))]
+>;
+def S_LSHR_B32 : SOP2_32 <0x00000020, "S_LSHR_B32",
+  [(set i32:$dst, (srl i32:$src0, i32:$src1))]
+>;
+def S_LSHR_B64 : SOP2_SHIFT_64 <0x00000021, "S_LSHR_B64",
+  [(set i64:$dst, (srl i64:$src0, i32:$src1))]
+>;
+def S_ASHR_I32 : SOP2_32 <0x00000022, "S_ASHR_I32",
+  [(set i32:$dst, (sra i32:$src0, i32:$src1))]
+>;
+def S_ASHR_I64 : SOP2_SHIFT_64 <0x00000023, "S_ASHR_I64",
+  [(set i64:$dst, (sra i64:$src0, i32:$src1))]
+>;
+
+} // End AddedComplexity = 1
+
+def S_BFM_B32 : SOP2_32 <0x00000024, "S_BFM_B32", []>;
+def S_BFM_B64 : SOP2_64 <0x00000025, "S_BFM_B64", []>;
+def S_MUL_I32 : SOP2_32 <0x00000026, "S_MUL_I32", []>;
+def S_BFE_U32 : SOP2_32 <0x00000027, "S_BFE_U32", []>;
+def S_BFE_I32 : SOP2_32 <0x00000028, "S_BFE_I32", []>;
+def S_BFE_U64 : SOP2_64 <0x00000029, "S_BFE_U64", []>;
+def S_BFE_I64 : SOP2_64 <0x0000002a, "S_BFE_I64", []>;
+//def S_CBRANCH_G_FORK : SOP2_ <0x0000002b, "S_CBRANCH_G_FORK", []>;
+def S_ABSDIFF_I32 : SOP2_32 <0x0000002c, "S_ABSDIFF_I32", []>;
+
+//===----------------------------------------------------------------------===//
+// SOPC Instructions
+//===----------------------------------------------------------------------===//
+
+def S_CMP_EQ_I32 : SOPC_32 <0x00000000, "S_CMP_EQ_I32">;
+def S_CMP_LG_I32 : SOPC_32 <0x00000001, "S_CMP_LG_I32">;
+def S_CMP_GT_I32 : SOPC_32 <0x00000002, "S_CMP_GT_I32">;
+def S_CMP_GE_I32 : SOPC_32 <0x00000003, "S_CMP_GE_I32">;
+def S_CMP_LT_I32 : SOPC_32 <0x00000004, "S_CMP_LT_I32">;
+def S_CMP_LE_I32 : SOPC_32 <0x00000005, "S_CMP_LE_I32">;
+def S_CMP_EQ_U32 : SOPC_32 <0x00000006, "S_CMP_EQ_U32">;
+def S_CMP_LG_U32 : SOPC_32 <0x00000007, "S_CMP_LG_U32">;
+def S_CMP_GT_U32 : SOPC_32 <0x00000008, "S_CMP_GT_U32">;
+def S_CMP_GE_U32 : SOPC_32 <0x00000009, "S_CMP_GE_U32">;
+def S_CMP_LT_U32 : SOPC_32 <0x0000000a, "S_CMP_LT_U32">;
+def S_CMP_LE_U32 : SOPC_32 <0x0000000b, "S_CMP_LE_U32">;
+////def S_BITCMP0_B32 : SOPC_BITCMP0 <0x0000000c, "S_BITCMP0_B32", []>;
+////def S_BITCMP1_B32 : SOPC_BITCMP1 <0x0000000d, "S_BITCMP1_B32", []>;
+////def S_BITCMP0_B64 : SOPC_BITCMP0 <0x0000000e, "S_BITCMP0_B64", []>;
+////def S_BITCMP1_B64 : SOPC_BITCMP1 <0x0000000f, "S_BITCMP1_B64", []>;
+//def S_SETVSKIP : SOPC_ <0x00000010, "S_SETVSKIP", []>;
+
+//===----------------------------------------------------------------------===//
+// SOPK Instructions
+//===----------------------------------------------------------------------===//
+
 def S_MOVK_I32 : SOPK_32 <0x00000000, "S_MOVK_I32", []>;
 def S_CMOVK_I32 : SOPK_32 <0x00000002, "S_CMOVK_I32", []>;
 
@@ -116,7 +337,7 @@ def S_CMPK_EQ_I32 : SOPK <
 >;
 */
 
-let isCompare = 1 in {
+let isCompare = 1, Defs = [SCC] in {
 def S_CMPK_LG_I32 : SOPK_32 <0x00000004, "S_CMPK_LG_I32", []>;
 def S_CMPK_GT_I32 : SOPK_32 <0x00000005, "S_CMPK_GT_I32", []>;
 def S_CMPK_GE_I32 : SOPK_32 <0x00000006, "S_CMPK_GE_I32", []>;
@@ -128,7 +349,7 @@ def S_CMPK_GT_U32 : SOPK_32 <0x0000000b, "S_CMPK_GT_U32", []>;
 def S_CMPK_GE_U32 : SOPK_32 <0x0000000c, "S_CMPK_GE_U32", []>;
 def S_CMPK_LT_U32 : SOPK_32 <0x0000000d, "S_CMPK_LT_U32", []>;
 def S_CMPK_LE_U32 : SOPK_32 <0x0000000e, "S_CMPK_LE_U32", []>;
-} // End isCompare = 1
+} // End isCompare = 1, Defs = [SCC]
 
 let Defs = [SCC], isCommutable = 1 in {
   def S_ADDK_I32 : SOPK_32 <0x0000000f, "S_ADDK_I32", []>;
@@ -142,6 +363,108 @@ def S_GETREG_REGRD_B32 : SOPK_32 <0x00000014, "S_GETREG_REGRD_B32", []>;
 //def S_SETREG_IMM32_B32 : SOPK_32 <0x00000015, "S_SETREG_IMM32_B32", []>;
 //def EXP : EXP_ <0x00000000, "EXP", []>;
 
+} // End let OtherPredicates = [isCFDepth0]
+
+//===----------------------------------------------------------------------===//
+// SOPP Instructions
+//===----------------------------------------------------------------------===//
+
+def S_NOP : SOPP <0x00000000, (ins i16imm:$simm16), "S_NOP $simm16", []>;
+
+let isTerminator = 1 in {
+
+def S_ENDPGM : SOPP <0x00000001, (ins), "S_ENDPGM",
+  [(IL_retflag)]> {
+  let simm16 = 0;
+  let isBarrier = 1;
+  let hasCtrlDep = 1;
+}
+
+let isBranch = 1 in {
+def S_BRANCH : SOPP <
+  0x00000002, (ins sopp_brtarget:$simm16), "S_BRANCH $simm16",
+  [(br bb:$simm16)]> {
+  let isBarrier = 1;
+}
+
+let DisableEncoding = "$scc" in {
+def S_CBRANCH_SCC0 : SOPP <
+  0x00000004, (ins sopp_brtarget:$simm16, SCCReg:$scc),
+  "S_CBRANCH_SCC0 $simm16", []
+>;
+def S_CBRANCH_SCC1 : SOPP <
+  0x00000005, (ins sopp_brtarget:$simm16, SCCReg:$scc),
+  "S_CBRANCH_SCC1 $simm16",
+  []
+>;
+} // End DisableEncoding = "$scc"
+
+def S_CBRANCH_VCCZ : SOPP <
+  0x00000006, (ins sopp_brtarget:$simm16, VCCReg:$vcc),
+  "S_CBRANCH_VCCZ $simm16",
+  []
+>;
+def S_CBRANCH_VCCNZ : SOPP <
+  0x00000007, (ins sopp_brtarget:$simm16, VCCReg:$vcc),
+  "S_CBRANCH_VCCNZ $simm16",
+  []
+>;
+
+let DisableEncoding = "$exec" in {
+def S_CBRANCH_EXECZ : SOPP <
+  0x00000008, (ins sopp_brtarget:$simm16, EXECReg:$exec),
+  "S_CBRANCH_EXECZ $simm16",
+  []
+>;
+def S_CBRANCH_EXECNZ : SOPP <
+  0x00000009, (ins sopp_brtarget:$simm16, EXECReg:$exec),
+  "S_CBRANCH_EXECNZ $simm16",
+  []
+>;
+} // End DisableEncoding = "$exec"
+
+
+} // End isBranch = 1
+} // End isTerminator = 1
+
+let hasSideEffects = 1 in {
+def S_BARRIER : SOPP <0x0000000a, (ins), "S_BARRIER",
+  [(int_AMDGPU_barrier_local)]
+> {
+  let simm16 = 0;
+  let isBarrier = 1;
+  let hasCtrlDep = 1;
+  let mayLoad = 1;
+  let mayStore = 1;
+}
+
+def S_WAITCNT : SOPP <0x0000000c, (ins WAIT_FLAG:$simm16), "S_WAITCNT $simm16",
+  []
+>;
+//def S_SETHALT : SOPP_ <0x0000000d, "S_SETHALT", []>;
+//def S_SLEEP : SOPP_ <0x0000000e, "S_SLEEP", []>;
+//def S_SETPRIO : SOPP_ <0x0000000f, "S_SETPRIO", []>;
+
+let Uses = [EXEC] in {
+  def S_SENDMSG : SOPP <0x00000010, (ins SendMsgImm:$simm16, M0Reg:$m0), "S_SENDMSG $simm16",
+      [(int_SI_sendmsg imm:$simm16, M0Reg:$m0)]
+  > {
+    let DisableEncoding = "$m0";
+  }
+} // End Uses = [EXEC]
+
+//def S_SENDMSGHALT : SOPP_ <0x00000011, "S_SENDMSGHALT", []>;
+//def S_TRAP : SOPP_ <0x00000012, "S_TRAP", []>;
+//def S_ICACHE_INV : SOPP_ <0x00000013, "S_ICACHE_INV", []>;
+//def S_INCPERFLEVEL : SOPP_ <0x00000014, "S_INCPERFLEVEL", []>;
+//def S_DECPERFLEVEL : SOPP_ <0x00000015, "S_DECPERFLEVEL", []>;
+//def S_TTRACEDATA : SOPP_ <0x00000016, "S_TTRACEDATA", []>;
+} // End hasSideEffects
+
+//===----------------------------------------------------------------------===//
+// VOPC Instructions
+//===----------------------------------------------------------------------===//
+
 let isCompare = 1 in {
 
 defm V_CMP_F_F32 : VOPC_32 <0x00000000, "V_CMP_F_F32">;
@@ -161,26 +484,26 @@ defm V_CMP_NEQ_F32 : VOPC_32 <0x0000000d, "V_CMP_NEQ_F32", f32, COND_UNE>;
 defm V_CMP_NLT_F32 : VOPC_32 <0x0000000e, "V_CMP_NLT_F32">;
 defm V_CMP_TRU_F32 : VOPC_32 <0x0000000f, "V_CMP_TRU_F32">;
 
-let hasSideEffects = 1, Defs = [EXEC] in {
+let hasSideEffects = 1 in {
 
-defm V_CMPX_F_F32 : VOPC_32 <0x00000010, "V_CMPX_F_F32">;
-defm V_CMPX_LT_F32 : VOPC_32 <0x00000011, "V_CMPX_LT_F32">;
-defm V_CMPX_EQ_F32 : VOPC_32 <0x00000012, "V_CMPX_EQ_F32">;
-defm V_CMPX_LE_F32 : VOPC_32 <0x00000013, "V_CMPX_LE_F32">;
-defm V_CMPX_GT_F32 : VOPC_32 <0x00000014, "V_CMPX_GT_F32">;
-defm V_CMPX_LG_F32 : VOPC_32 <0x00000015, "V_CMPX_LG_F32">;
-defm V_CMPX_GE_F32 : VOPC_32 <0x00000016, "V_CMPX_GE_F32">;
-defm V_CMPX_O_F32 : VOPC_32 <0x00000017, "V_CMPX_O_F32">;
-defm V_CMPX_U_F32 : VOPC_32 <0x00000018, "V_CMPX_U_F32">;
-defm V_CMPX_NGE_F32 : VOPC_32 <0x00000019, "V_CMPX_NGE_F32">;
-defm V_CMPX_NLG_F32 : VOPC_32 <0x0000001a, "V_CMPX_NLG_F32">;
-defm V_CMPX_NGT_F32 : VOPC_32 <0x0000001b, "V_CMPX_NGT_F32">;
-defm V_CMPX_NLE_F32 : VOPC_32 <0x0000001c, "V_CMPX_NLE_F32">;
-defm V_CMPX_NEQ_F32 : VOPC_32 <0x0000001d, "V_CMPX_NEQ_F32">;
-defm V_CMPX_NLT_F32 : VOPC_32 <0x0000001e, "V_CMPX_NLT_F32">;
-defm V_CMPX_TRU_F32 : VOPC_32 <0x0000001f, "V_CMPX_TRU_F32">;
+defm V_CMPX_F_F32 : VOPCX_32 <0x00000010, "V_CMPX_F_F32">;
+defm V_CMPX_LT_F32 : VOPCX_32 <0x00000011, "V_CMPX_LT_F32">;
+defm V_CMPX_EQ_F32 : VOPCX_32 <0x00000012, "V_CMPX_EQ_F32">;
+defm V_CMPX_LE_F32 : VOPCX_32 <0x00000013, "V_CMPX_LE_F32">;
+defm V_CMPX_GT_F32 : VOPCX_32 <0x00000014, "V_CMPX_GT_F32">;
+defm V_CMPX_LG_F32 : VOPCX_32 <0x00000015, "V_CMPX_LG_F32">;
+defm V_CMPX_GE_F32 : VOPCX_32 <0x00000016, "V_CMPX_GE_F32">;
+defm V_CMPX_O_F32 : VOPCX_32 <0x00000017, "V_CMPX_O_F32">;
+defm V_CMPX_U_F32 : VOPCX_32 <0x00000018, "V_CMPX_U_F32">;
+defm V_CMPX_NGE_F32 : VOPCX_32 <0x00000019, "V_CMPX_NGE_F32">;
+defm V_CMPX_NLG_F32 : VOPCX_32 <0x0000001a, "V_CMPX_NLG_F32">;
+defm V_CMPX_NGT_F32 : VOPCX_32 <0x0000001b, "V_CMPX_NGT_F32">;
+defm V_CMPX_NLE_F32 : VOPCX_32 <0x0000001c, "V_CMPX_NLE_F32">;
+defm V_CMPX_NEQ_F32 : VOPCX_32 <0x0000001d, "V_CMPX_NEQ_F32">;
+defm V_CMPX_NLT_F32 : VOPCX_32 <0x0000001e, "V_CMPX_NLT_F32">;
+defm V_CMPX_TRU_F32 : VOPCX_32 <0x0000001f, "V_CMPX_TRU_F32">;
 
-} // End hasSideEffects = 1, Defs = [EXEC]
+} // End hasSideEffects = 1
 
 defm V_CMP_F_F64 : VOPC_64 <0x00000020, "V_CMP_F_F64">;
 defm V_CMP_LT_F64 : VOPC_64 <0x00000021, "V_CMP_LT_F64", f64, COND_OLT>;
@@ -199,26 +522,26 @@ defm V_CMP_NEQ_F64 : VOPC_64 <0x0000002d, "V_CMP_NEQ_F64", f64, COND_UNE>;
 defm V_CMP_NLT_F64 : VOPC_64 <0x0000002e, "V_CMP_NLT_F64">;
 defm V_CMP_TRU_F64 : VOPC_64 <0x0000002f, "V_CMP_TRU_F64">;
 
-let hasSideEffects = 1, Defs = [EXEC] in {
+let hasSideEffects = 1 in {
 
-defm V_CMPX_F_F64 : VOPC_64 <0x00000030, "V_CMPX_F_F64">;
-defm V_CMPX_LT_F64 : VOPC_64 <0x00000031, "V_CMPX_LT_F64">;
-defm V_CMPX_EQ_F64 : VOPC_64 <0x00000032, "V_CMPX_EQ_F64">;
-defm V_CMPX_LE_F64 : VOPC_64 <0x00000033, "V_CMPX_LE_F64">;
-defm V_CMPX_GT_F64 : VOPC_64 <0x00000034, "V_CMPX_GT_F64">;
-defm V_CMPX_LG_F64 : VOPC_64 <0x00000035, "V_CMPX_LG_F64">;
-defm V_CMPX_GE_F64 : VOPC_64 <0x00000036, "V_CMPX_GE_F64">;
-defm V_CMPX_O_F64 : VOPC_64 <0x00000037, "V_CMPX_O_F64">;
-defm V_CMPX_U_F64 : VOPC_64 <0x00000038, "V_CMPX_U_F64">;
-defm V_CMPX_NGE_F64 : VOPC_64 <0x00000039, "V_CMPX_NGE_F64">;
-defm V_CMPX_NLG_F64 : VOPC_64 <0x0000003a, "V_CMPX_NLG_F64">;
-defm V_CMPX_NGT_F64 : VOPC_64 <0x0000003b, "V_CMPX_NGT_F64">;
-defm V_CMPX_NLE_F64 : VOPC_64 <0x0000003c, "V_CMPX_NLE_F64">;
-defm V_CMPX_NEQ_F64 : VOPC_64 <0x0000003d, "V_CMPX_NEQ_F64">;
-defm V_CMPX_NLT_F64 : VOPC_64 <0x0000003e, "V_CMPX_NLT_F64">;
-defm V_CMPX_TRU_F64 : VOPC_64 <0x0000003f, "V_CMPX_TRU_F64">;
+defm V_CMPX_F_F64 : VOPCX_64 <0x00000030, "V_CMPX_F_F64">;
+defm V_CMPX_LT_F64 : VOPCX_64 <0x00000031, "V_CMPX_LT_F64">;
+defm V_CMPX_EQ_F64 : VOPCX_64 <0x00000032, "V_CMPX_EQ_F64">;
+defm V_CMPX_LE_F64 : VOPCX_64 <0x00000033, "V_CMPX_LE_F64">;
+defm V_CMPX_GT_F64 : VOPCX_64 <0x00000034, "V_CMPX_GT_F64">;
+defm V_CMPX_LG_F64 : VOPCX_64 <0x00000035, "V_CMPX_LG_F64">;
+defm V_CMPX_GE_F64 : VOPCX_64 <0x00000036, "V_CMPX_GE_F64">;
+defm V_CMPX_O_F64 : VOPCX_64 <0x00000037, "V_CMPX_O_F64">;
+defm V_CMPX_U_F64 : VOPCX_64 <0x00000038, "V_CMPX_U_F64">;
+defm V_CMPX_NGE_F64 : VOPCX_64 <0x00000039, "V_CMPX_NGE_F64">;
+defm V_CMPX_NLG_F64 : VOPCX_64 <0x0000003a, "V_CMPX_NLG_F64">;
+defm V_CMPX_NGT_F64 : VOPCX_64 <0x0000003b, "V_CMPX_NGT_F64">;
+defm V_CMPX_NLE_F64 : VOPCX_64 <0x0000003c, "V_CMPX_NLE_F64">;
+defm V_CMPX_NEQ_F64 : VOPCX_64 <0x0000003d, "V_CMPX_NEQ_F64">;
+defm V_CMPX_NLT_F64 : VOPCX_64 <0x0000003e, "V_CMPX_NLT_F64">;
+defm V_CMPX_TRU_F64 : VOPCX_64 <0x0000003f, "V_CMPX_TRU_F64">;
 
-} // End hasSideEffects = 1, Defs = [EXEC]
+} // End hasSideEffects = 1
 
 defm V_CMPS_F_F32 : VOPC_32 <0x00000040, "V_CMPS_F_F32">;
 defm V_CMPS_LT_F32 : VOPC_32 <0x00000041, "V_CMPS_LT_F32">;
@@ -237,26 +560,26 @@ defm V_CMPS_NEQ_F32 : VOPC_32 <0x0000004d, "V_CMPS_NEQ_F32">;
 defm V_CMPS_NLT_F32 : VOPC_32 <0x0000004e, "V_CMPS_NLT_F32">;
 defm V_CMPS_TRU_F32 : VOPC_32 <0x0000004f, "V_CMPS_TRU_F32">;
 
-let hasSideEffects = 1, Defs = [EXEC] in {
+let hasSideEffects = 1 in {
 
-defm V_CMPSX_F_F32 : VOPC_32 <0x00000050, "V_CMPSX_F_F32">;
-defm V_CMPSX_LT_F32 : VOPC_32 <0x00000051, "V_CMPSX_LT_F32">;
-defm V_CMPSX_EQ_F32 : VOPC_32 <0x00000052, "V_CMPSX_EQ_F32">;
-defm V_CMPSX_LE_F32 : VOPC_32 <0x00000053, "V_CMPSX_LE_F32">;
-defm V_CMPSX_GT_F32 : VOPC_32 <0x00000054, "V_CMPSX_GT_F32">;
-defm V_CMPSX_LG_F32 : VOPC_32 <0x00000055, "V_CMPSX_LG_F32">;
-defm V_CMPSX_GE_F32 : VOPC_32 <0x00000056, "V_CMPSX_GE_F32">;
-defm V_CMPSX_O_F32 : VOPC_32 <0x00000057, "V_CMPSX_O_F32">;
-defm V_CMPSX_U_F32 : VOPC_32 <0x00000058, "V_CMPSX_U_F32">;
-defm V_CMPSX_NGE_F32 : VOPC_32 <0x00000059, "V_CMPSX_NGE_F32">;
-defm V_CMPSX_NLG_F32 : VOPC_32 <0x0000005a, "V_CMPSX_NLG_F32">;
-defm V_CMPSX_NGT_F32 : VOPC_32 <0x0000005b, "V_CMPSX_NGT_F32">;
-defm V_CMPSX_NLE_F32 : VOPC_32 <0x0000005c, "V_CMPSX_NLE_F32">;
-defm V_CMPSX_NEQ_F32 : VOPC_32 <0x0000005d, "V_CMPSX_NEQ_F32">;
-defm V_CMPSX_NLT_F32 : VOPC_32 <0x0000005e, "V_CMPSX_NLT_F32">;
-defm V_CMPSX_TRU_F32 : VOPC_32 <0x0000005f, "V_CMPSX_TRU_F32">;
+defm V_CMPSX_F_F32 : VOPCX_32 <0x00000050, "V_CMPSX_F_F32">;
+defm V_CMPSX_LT_F32 : VOPCX_32 <0x00000051, "V_CMPSX_LT_F32">;
+defm V_CMPSX_EQ_F32 : VOPCX_32 <0x00000052, "V_CMPSX_EQ_F32">;
+defm V_CMPSX_LE_F32 : VOPCX_32 <0x00000053, "V_CMPSX_LE_F32">;
+defm V_CMPSX_GT_F32 : VOPCX_32 <0x00000054, "V_CMPSX_GT_F32">;
+defm V_CMPSX_LG_F32 : VOPCX_32 <0x00000055, "V_CMPSX_LG_F32">;
+defm V_CMPSX_GE_F32 : VOPCX_32 <0x00000056, "V_CMPSX_GE_F32">;
+defm V_CMPSX_O_F32 : VOPCX_32 <0x00000057, "V_CMPSX_O_F32">;
+defm V_CMPSX_U_F32 : VOPCX_32 <0x00000058, "V_CMPSX_U_F32">;
+defm V_CMPSX_NGE_F32 : VOPCX_32 <0x00000059, "V_CMPSX_NGE_F32">;
+defm V_CMPSX_NLG_F32 : VOPCX_32 <0x0000005a, "V_CMPSX_NLG_F32">;
+defm V_CMPSX_NGT_F32 : VOPCX_32 <0x0000005b, "V_CMPSX_NGT_F32">;
+defm V_CMPSX_NLE_F32 : VOPCX_32 <0x0000005c, "V_CMPSX_NLE_F32">;
+defm V_CMPSX_NEQ_F32 : VOPCX_32 <0x0000005d, "V_CMPSX_NEQ_F32">;
+defm V_CMPSX_NLT_F32 : VOPCX_32 <0x0000005e, "V_CMPSX_NLT_F32">;
+defm V_CMPSX_TRU_F32 : VOPCX_32 <0x0000005f, "V_CMPSX_TRU_F32">;
 
-} // End hasSideEffects = 1, Defs = [EXEC]
+} // End hasSideEffects = 1
 
 defm V_CMPS_F_F64 : VOPC_64 <0x00000060, "V_CMPS_F_F64">;
 defm V_CMPS_LT_F64 : VOPC_64 <0x00000061, "V_CMPS_LT_F64">;
@@ -305,18 +628,18 @@ defm V_CMP_NE_I32 : VOPC_32 <0x00000085, "V_CMP_NE_I32", i32, COND_NE>;
 defm V_CMP_GE_I32 : VOPC_32 <0x00000086, "V_CMP_GE_I32", i32, COND_SGE>;
 defm V_CMP_T_I32 : VOPC_32 <0x00000087, "V_CMP_T_I32">;
 
-let hasSideEffects = 1, Defs = [EXEC] in {
+let hasSideEffects = 1 in {
 
-defm V_CMPX_F_I32 : VOPC_32 <0x00000090, "V_CMPX_F_I32">;
-defm V_CMPX_LT_I32 : VOPC_32 <0x00000091, "V_CMPX_LT_I32">;
-defm V_CMPX_EQ_I32 : VOPC_32 <0x00000092, "V_CMPX_EQ_I32">;
-defm V_CMPX_LE_I32 : VOPC_32 <0x00000093, "V_CMPX_LE_I32">;
-defm V_CMPX_GT_I32 : VOPC_32 <0x00000094, "V_CMPX_GT_I32">;
-defm V_CMPX_NE_I32 : VOPC_32 <0x00000095, "V_CMPX_NE_I32">;
-defm V_CMPX_GE_I32 : VOPC_32 <0x00000096, "V_CMPX_GE_I32">;
-defm V_CMPX_T_I32 : VOPC_32 <0x00000097, "V_CMPX_T_I32">;
+defm V_CMPX_F_I32 : VOPCX_32 <0x00000090, "V_CMPX_F_I32">;
+defm V_CMPX_LT_I32 : VOPCX_32 <0x00000091, "V_CMPX_LT_I32">;
+defm V_CMPX_EQ_I32 : VOPCX_32 <0x00000092, "V_CMPX_EQ_I32">;
+defm V_CMPX_LE_I32 : VOPCX_32 <0x00000093, "V_CMPX_LE_I32">;
+defm V_CMPX_GT_I32 : VOPCX_32 <0x00000094, "V_CMPX_GT_I32">;
+defm V_CMPX_NE_I32 : VOPCX_32 <0x00000095, "V_CMPX_NE_I32">;
+defm V_CMPX_GE_I32 : VOPCX_32 <0x00000096, "V_CMPX_GE_I32">;
+defm V_CMPX_T_I32 : VOPCX_32 <0x00000097, "V_CMPX_T_I32">;
 
-} // End hasSideEffects = 1, Defs = [EXEC]
+} // End hasSideEffects = 1
 
 defm V_CMP_F_I64 : VOPC_64 <0x000000a0, "V_CMP_F_I64">;
 defm V_CMP_LT_I64 : VOPC_64 <0x000000a1, "V_CMP_LT_I64", i64, COND_SLT>;
@@ -327,18 +650,18 @@ defm V_CMP_NE_I64 : VOPC_64 <0x000000a5, "V_CMP_NE_I64", i64, COND_NE>;
 defm V_CMP_GE_I64 : VOPC_64 <0x000000a6, "V_CMP_GE_I64", i64, COND_SGE>;
 defm V_CMP_T_I64 : VOPC_64 <0x000000a7, "V_CMP_T_I64">;
 
-let hasSideEffects = 1, Defs = [EXEC] in {
+let hasSideEffects = 1 in {
 
-defm V_CMPX_F_I64 : VOPC_64 <0x000000b0, "V_CMPX_F_I64">;
-defm V_CMPX_LT_I64 : VOPC_64 <0x000000b1, "V_CMPX_LT_I64">;
-defm V_CMPX_EQ_I64 : VOPC_64 <0x000000b2, "V_CMPX_EQ_I64">;
-defm V_CMPX_LE_I64 : VOPC_64 <0x000000b3, "V_CMPX_LE_I64">;
-defm V_CMPX_GT_I64 : VOPC_64 <0x000000b4, "V_CMPX_GT_I64">;
-defm V_CMPX_NE_I64 : VOPC_64 <0x000000b5, "V_CMPX_NE_I64">;
-defm V_CMPX_GE_I64 : VOPC_64 <0x000000b6, "V_CMPX_GE_I64">;
-defm V_CMPX_T_I64 : VOPC_64 <0x000000b7, "V_CMPX_T_I64">;
+defm V_CMPX_F_I64 : VOPCX_64 <0x000000b0, "V_CMPX_F_I64">;
+defm V_CMPX_LT_I64 : VOPCX_64 <0x000000b1, "V_CMPX_LT_I64">;
+defm V_CMPX_EQ_I64 : VOPCX_64 <0x000000b2, "V_CMPX_EQ_I64">;
+defm V_CMPX_LE_I64 : VOPCX_64 <0x000000b3, "V_CMPX_LE_I64">;
+defm V_CMPX_GT_I64 : VOPCX_64 <0x000000b4, "V_CMPX_GT_I64">;
+defm V_CMPX_NE_I64 : VOPCX_64 <0x000000b5, "V_CMPX_NE_I64">;
+defm V_CMPX_GE_I64 : VOPCX_64 <0x000000b6, "V_CMPX_GE_I64">;
+defm V_CMPX_T_I64 : VOPCX_64 <0x000000b7, "V_CMPX_T_I64">;
 
-} // End hasSideEffects = 1, Defs = [EXEC]
+} // End hasSideEffects = 1
 
 defm V_CMP_F_U32 : VOPC_32 <0x000000c0, "V_CMP_F_U32">;
 defm V_CMP_LT_U32 : VOPC_32 <0x000000c1, "V_CMP_LT_U32", i32, COND_ULT>;
@@ -349,18 +672,18 @@ defm V_CMP_NE_U32 : VOPC_32 <0x000000c5, "V_CMP_NE_U32", i32, COND_NE>;
 defm V_CMP_GE_U32 : VOPC_32 <0x000000c6, "V_CMP_GE_U32", i32, COND_UGE>;
 defm V_CMP_T_U32 : VOPC_32 <0x000000c7, "V_CMP_T_U32">;
 
-let hasSideEffects = 1, Defs = [EXEC] in {
+let hasSideEffects = 1 in {
 
-defm V_CMPX_F_U32 : VOPC_32 <0x000000d0, "V_CMPX_F_U32">;
-defm V_CMPX_LT_U32 : VOPC_32 <0x000000d1, "V_CMPX_LT_U32">;
-defm V_CMPX_EQ_U32 : VOPC_32 <0x000000d2, "V_CMPX_EQ_U32">;
-defm V_CMPX_LE_U32 : VOPC_32 <0x000000d3, "V_CMPX_LE_U32">;
-defm V_CMPX_GT_U32 : VOPC_32 <0x000000d4, "V_CMPX_GT_U32">;
-defm V_CMPX_NE_U32 : VOPC_32 <0x000000d5, "V_CMPX_NE_U32">;
-defm V_CMPX_GE_U32 : VOPC_32 <0x000000d6, "V_CMPX_GE_U32">;
-defm V_CMPX_T_U32 : VOPC_32 <0x000000d7, "V_CMPX_T_U32">;
+defm V_CMPX_F_U32 : VOPCX_32 <0x000000d0, "V_CMPX_F_U32">;
+defm V_CMPX_LT_U32 : VOPCX_32 <0x000000d1, "V_CMPX_LT_U32">;
+defm V_CMPX_EQ_U32 : VOPCX_32 <0x000000d2, "V_CMPX_EQ_U32">;
+defm V_CMPX_LE_U32 : VOPCX_32 <0x000000d3, "V_CMPX_LE_U32">;
+defm V_CMPX_GT_U32 : VOPCX_32 <0x000000d4, "V_CMPX_GT_U32">;
+defm V_CMPX_NE_U32 : VOPCX_32 <0x000000d5, "V_CMPX_NE_U32">;
+defm V_CMPX_GE_U32 : VOPCX_32 <0x000000d6, "V_CMPX_GE_U32">;
+defm V_CMPX_T_U32 : VOPCX_32 <0x000000d7, "V_CMPX_T_U32">;
 
-} // End hasSideEffects = 1, Defs = [EXEC]
+} // End hasSideEffects = 1
 
 defm V_CMP_F_U64 : VOPC_64 <0x000000e0, "V_CMP_F_U64">;
 defm V_CMP_LT_U64 : VOPC_64 <0x000000e1, "V_CMP_LT_U64", i64, COND_ULT>;
@@ -371,43 +694,153 @@ defm V_CMP_NE_U64 : VOPC_64 <0x000000e5, "V_CMP_NE_U64", i64, COND_NE>;
 defm V_CMP_GE_U64 : VOPC_64 <0x000000e6, "V_CMP_GE_U64", i64, COND_UGE>;
 defm V_CMP_T_U64 : VOPC_64 <0x000000e7, "V_CMP_T_U64">;
 
-let hasSideEffects = 1, Defs = [EXEC] in {
+let hasSideEffects = 1 in {
 
-defm V_CMPX_F_U64 : VOPC_64 <0x000000f0, "V_CMPX_F_U64">;
-defm V_CMPX_LT_U64 : VOPC_64 <0x000000f1, "V_CMPX_LT_U64">;
-defm V_CMPX_EQ_U64 : VOPC_64 <0x000000f2, "V_CMPX_EQ_U64">;
-defm V_CMPX_LE_U64 : VOPC_64 <0x000000f3, "V_CMPX_LE_U64">;
-defm V_CMPX_GT_U64 : VOPC_64 <0x000000f4, "V_CMPX_GT_U64">;
-defm V_CMPX_NE_U64 : VOPC_64 <0x000000f5, "V_CMPX_NE_U64">;
-defm V_CMPX_GE_U64 : VOPC_64 <0x000000f6, "V_CMPX_GE_U64">;
-defm V_CMPX_T_U64 : VOPC_64 <0x000000f7, "V_CMPX_T_U64">;
+defm V_CMPX_F_U64 : VOPCX_64 <0x000000f0, "V_CMPX_F_U64">;
+defm V_CMPX_LT_U64 : VOPCX_64 <0x000000f1, "V_CMPX_LT_U64">;
+defm V_CMPX_EQ_U64 : VOPCX_64 <0x000000f2, "V_CMPX_EQ_U64">;
+defm V_CMPX_LE_U64 : VOPCX_64 <0x000000f3, "V_CMPX_LE_U64">;
+defm V_CMPX_GT_U64 : VOPCX_64 <0x000000f4, "V_CMPX_GT_U64">;
+defm V_CMPX_NE_U64 : VOPCX_64 <0x000000f5, "V_CMPX_NE_U64">;
+defm V_CMPX_GE_U64 : VOPCX_64 <0x000000f6, "V_CMPX_GE_U64">;
+defm V_CMPX_T_U64 : VOPCX_64 <0x000000f7, "V_CMPX_T_U64">;
 
-} // End hasSideEffects = 1, Defs = [EXEC]
+} // End hasSideEffects = 1
 
 defm V_CMP_CLASS_F32 : VOPC_32 <0x00000088, "V_CMP_CLASS_F32">;
 
-let hasSideEffects = 1, Defs = [EXEC] in {
-defm V_CMPX_CLASS_F32 : VOPC_32 <0x00000098, "V_CMPX_CLASS_F32">;
-} // End hasSideEffects = 1, Defs = [EXEC]
+let hasSideEffects = 1 in {
+defm V_CMPX_CLASS_F32 : VOPCX_32 <0x00000098, "V_CMPX_CLASS_F32">;
+} // End hasSideEffects = 1
 
 defm V_CMP_CLASS_F64 : VOPC_64 <0x000000a8, "V_CMP_CLASS_F64">;
 
-let hasSideEffects = 1, Defs = [EXEC] in {
-defm V_CMPX_CLASS_F64 : VOPC_64 <0x000000b8, "V_CMPX_CLASS_F64">;
-} // End hasSideEffects = 1, Defs = [EXEC]
+let hasSideEffects = 1 in {
+defm V_CMPX_CLASS_F64 : VOPCX_64 <0x000000b8, "V_CMPX_CLASS_F64">;
+} // End hasSideEffects = 1
 
 } // End isCompare = 1
 
-def DS_ADD_U32_RTN : DS_1A1D_RET <0x20, "DS_ADD_U32_RTN", VReg_32>;
-def DS_SUB_U32_RTN : DS_1A1D_RET <0x21, "DS_SUB_U32_RTN", VReg_32>;
+//===----------------------------------------------------------------------===//
+// DS Instructions
+//===----------------------------------------------------------------------===//
+
+
+def DS_ADD_U32 : DS_1A1D_NORET <0x0, "DS_ADD_U32", VReg_32>;
+def DS_SUB_U32 : DS_1A1D_NORET <0x1, "DS_SUB_U32", VReg_32>;
+def DS_RSUB_U32 : DS_1A1D_NORET <0x2, "DS_RSUB_U32", VReg_32>;
+def DS_INC_U32 : DS_1A1D_NORET <0x3, "DS_INC_U32", VReg_32>;
+def DS_DEC_U32 : DS_1A1D_NORET <0x4, "DS_DEC_U32", VReg_32>;
+def DS_MIN_I32 : DS_1A1D_NORET <0x5, "DS_MIN_I32", VReg_32>;
+def DS_MAX_I32 : DS_1A1D_NORET <0x6, "DS_MAX_I32", VReg_32>;
+def DS_MIN_U32 : DS_1A1D_NORET <0x7, "DS_MIN_U32", VReg_32>;
+def DS_MAX_U32 : DS_1A1D_NORET <0x8, "DS_MAX_U32", VReg_32>;
+def DS_AND_B32 : DS_1A1D_NORET <0x9, "DS_AND_B32", VReg_32>;
+def DS_OR_B32 : DS_1A1D_NORET <0xa, "DS_OR_B32", VReg_32>;
+def DS_XOR_B32 : DS_1A1D_NORET <0xb, "DS_XOR_B32", VReg_32>;
+def DS_MSKOR_B32 : DS_1A1D_NORET <0xc, "DS_MSKOR_B32", VReg_32>;
+def DS_CMPST_B32 : DS_1A2D_NORET <0x10, "DS_CMPST_B32", VReg_32>;
+def DS_CMPST_F32 : DS_1A2D_NORET <0x11, "DS_CMPST_F32", VReg_32>;
+def DS_MIN_F32 : DS_1A1D_NORET <0x12, "DS_MIN_F32", VReg_32>;
+def DS_MAX_F32 : DS_1A1D_NORET <0x13, "DS_MAX_F32", VReg_32>;
+
+def DS_ADD_RTN_U32 : DS_1A1D_RET <0x20, "DS_ADD_RTN_U32", VReg_32>;
+def DS_SUB_RTN_U32 : DS_1A1D_RET <0x21, "DS_SUB_RTN_U32", VReg_32>;
+def DS_RSUB_RTN_U32 : DS_1A1D_RET <0x22, "DS_RSUB_RTN_U32", VReg_32>;
+def DS_INC_RTN_U32 : DS_1A1D_RET <0x23, "DS_INC_RTN_U32", VReg_32>;
+def DS_DEC_RTN_U32 : DS_1A1D_RET <0x24, "DS_DEC_RTN_U32", VReg_32>;
+def DS_MIN_RTN_I32 : DS_1A1D_RET <0x25, "DS_MIN_RTN_I32", VReg_32>;
+def DS_MAX_RTN_I32 : DS_1A1D_RET <0x26, "DS_MAX_RTN_I32", VReg_32>;
+def DS_MIN_RTN_U32 : DS_1A1D_RET <0x27, "DS_MIN_RTN_U32", VReg_32>;
+def DS_MAX_RTN_U32 : DS_1A1D_RET <0x28, "DS_MAX_RTN_U32", VReg_32>;
+def DS_AND_RTN_B32 : DS_1A1D_RET <0x29, "DS_AND_RTN_B32", VReg_32>;
+def DS_OR_RTN_B32 : DS_1A1D_RET <0x2a, "DS_OR_RTN_B32", VReg_32>;
+def DS_XOR_RTN_B32 : DS_1A1D_RET <0x2b, "DS_XOR_RTN_B32", VReg_32>;
+def DS_MSKOR_RTN_B32 : DS_1A1D_RET <0x2c, "DS_MSKOR_RTN_B32", VReg_32>;
+def DS_WRXCHG_RTN_B32 : DS_1A1D_RET <0x2d, "DS_WRXCHG_RTN_B32", VReg_32>;
+//def DS_WRXCHG2_RTN_B32 : DS_2A0D_RET <0x2e, "DS_WRXCHG2_RTN_B32", VReg_32>;
+//def DS_WRXCHG2ST64_RTN_B32 : DS_2A0D_RET <0x2f, "DS_WRXCHG2_RTN_B32", VReg_32>;
+def DS_CMPST_RTN_B32 : DS_1A2D_RET <0x30, "DS_CMPST_RTN_B32", VReg_32>;
+def DS_CMPST_RTN_F32 : DS_1A2D_RET <0x31, "DS_CMPST_RTN_F32", VReg_32>;
+def DS_MIN_RTN_F32 : DS_1A1D_RET <0x32, "DS_MIN_RTN_F32", VReg_32>;
+def DS_MAX_RTN_F32 : DS_1A1D_RET <0x33, "DS_MAX_RTN_F32", VReg_32>;
+
+let SubtargetPredicate = isCI in {
+def DS_WRAP_RTN_F32 : DS_1A1D_RET <0x34, "DS_WRAP_RTN_F32", VReg_32>;
+} // End isCI
+
+
+def DS_ADD_U64 : DS_1A1D_NORET <0x40, "DS_ADD_U64", VReg_32>;
+def DS_SUB_U64 : DS_1A1D_NORET <0x41, "DS_SUB_U64", VReg_32>;
+def DS_RSUB_U64 : DS_1A1D_NORET <0x42, "DS_RSUB_U64", VReg_32>;
+def DS_INC_U64 : DS_1A1D_NORET <0x43, "DS_INC_U64", VReg_32>;
+def DS_DEC_U64 : DS_1A1D_NORET <0x44, "DS_DEC_U64", VReg_32>;
+def DS_MIN_I64 : DS_1A1D_NORET <0x45, "DS_MIN_I64", VReg_64>;
+def DS_MAX_I64 : DS_1A1D_NORET <0x46, "DS_MAX_I64", VReg_64>;
+def DS_MIN_U64 : DS_1A1D_NORET <0x47, "DS_MIN_U64", VReg_64>;
+def DS_MAX_U64 : DS_1A1D_NORET <0x48, "DS_MAX_U64", VReg_64>;
+def DS_AND_B64 : DS_1A1D_NORET <0x49, "DS_AND_B64", VReg_64>;
+def DS_OR_B64 : DS_1A1D_NORET <0x4a, "DS_OR_B64", VReg_64>;
+def DS_XOR_B64 : DS_1A1D_NORET <0x4b, "DS_XOR_B64", VReg_64>;
+def DS_MSKOR_B64 : DS_1A1D_NORET <0x4c, "DS_MSKOR_B64", VReg_64>;
+def DS_CMPST_B64 : DS_1A2D_NORET <0x50, "DS_CMPST_B64", VReg_64>;
+def DS_CMPST_F64 : DS_1A2D_NORET <0x51, "DS_CMPST_F64", VReg_64>;
+def DS_MIN_F64 : DS_1A1D_NORET <0x52, "DS_MIN_F64", VReg_64>;
+def DS_MAX_F64 : DS_1A1D_NORET <0x53, "DS_MAX_F64", VReg_64>;
+
+def DS_ADD_RTN_U64 : DS_1A1D_RET <0x60, "DS_ADD_RTN_U64", VReg_64>;
+def DS_SUB_RTN_U64 : DS_1A1D_RET <0x61, "DS_SUB_RTN_U64", VReg_64>;
+def DS_RSUB_RTN_U64 : DS_1A1D_RET <0x62, "DS_RSUB_RTN_U64", VReg_64>;
+def DS_INC_RTN_U64 : DS_1A1D_RET <0x63, "DS_INC_RTN_U64", VReg_64>;
+def DS_DEC_RTN_U64 : DS_1A1D_RET <0x64, "DS_DEC_RTN_U64", VReg_64>;
+def DS_MIN_RTN_I64 : DS_1A1D_RET <0x65, "DS_MIN_RTN_I64", VReg_64>;
+def DS_MAX_RTN_I64 : DS_1A1D_RET <0x66, "DS_MAX_RTN_I64", VReg_64>;
+def DS_MIN_RTN_U64 : DS_1A1D_RET <0x67, "DS_MIN_RTN_U64", VReg_64>;
+def DS_MAX_RTN_U64 : DS_1A1D_RET <0x68, "DS_MAX_RTN_U64", VReg_64>;
+def DS_AND_RTN_B64 : DS_1A1D_RET <0x69, "DS_AND_RTN_B64", VReg_64>;
+def DS_OR_RTN_B64 : DS_1A1D_RET <0x6a, "DS_OR_RTN_B64", VReg_64>;
+def DS_XOR_RTN_B64 : DS_1A1D_RET <0x6b, "DS_XOR_RTN_B64", VReg_64>;
+def DS_MSKOR_RTN_B64 : DS_1A1D_RET <0x6c, "DS_MSKOR_RTN_B64", VReg_64>;
+def DS_WRXCHG_RTN_B64 : DS_1A1D_RET <0x6d, "DS_WRXCHG_RTN_B64", VReg_64>;
+//def DS_WRXCHG2_RTN_B64 : DS_2A0D_RET <0x6e, "DS_WRXCHG2_RTN_B64", VReg_64>;
+//def DS_WRXCHG2ST64_RTN_B64 : DS_2A0D_RET <0x6f, "DS_WRXCHG2_RTN_B64", VReg_64>;
+def DS_CMPST_RTN_B64 : DS_1A2D_RET <0x70, "DS_CMPST_RTN_B64", VReg_64>;
+def DS_CMPST_RTN_F64 : DS_1A2D_RET <0x71, "DS_CMPST_RTN_F64", VReg_64>;
+def DS_MIN_RTN_F64 : DS_1A1D_RET <0x72, "DS_MIN_F64", VReg_64>;
+def DS_MAX_RTN_F64 : DS_1A1D_RET <0x73, "DS_MAX_F64", VReg_64>;
+
+//let SubtargetPredicate = isCI in {
+// DS_CONDXCHG32_RTN_B64
+// DS_CONDXCHG32_RTN_B128
+//} // End isCI
+
+// TODO: _SRC2_* forms
+
 def DS_WRITE_B32 : DS_Store_Helper <0x0000000d, "DS_WRITE_B32", VReg_32>;
 def DS_WRITE_B8 : DS_Store_Helper <0x00000001e, "DS_WRITE_B8", VReg_32>;
 def DS_WRITE_B16 : DS_Store_Helper <0x00000001f, "DS_WRITE_B16", VReg_32>;
+def DS_WRITE_B64 : DS_Store_Helper <0x00000004d, "DS_WRITE_B64", VReg_64>;
+
 def DS_READ_B32 : DS_Load_Helper <0x00000036, "DS_READ_B32", VReg_32>;
 def DS_READ_I8 : DS_Load_Helper <0x00000039, "DS_READ_I8", VReg_32>;
 def DS_READ_U8 : DS_Load_Helper <0x0000003a, "DS_READ_U8", VReg_32>;
 def DS_READ_I16 : DS_Load_Helper <0x0000003b, "DS_READ_I16", VReg_32>;
 def DS_READ_U16 : DS_Load_Helper <0x0000003c, "DS_READ_U16", VReg_32>;
+def DS_READ_B64 : DS_Load_Helper <0x00000076, "DS_READ_B64", VReg_64>;
+
+// 2 forms.
+def DS_WRITE2_B32 : DS_Load2_Helper <0x0000000E, "DS_WRITE2_B32", VReg_64>;
+def DS_WRITE2_B64 : DS_Load2_Helper <0x0000004E, "DS_WRITE2_B64", VReg_128>;
+
+def DS_READ2_B32 : DS_Load2_Helper <0x00000037, "DS_READ2_B32", VReg_64>;
+def DS_READ2_B64 : DS_Load2_Helper <0x00000075, "DS_READ2_B64", VReg_128>;
+
+// TODO: DS_READ2ST64_B32, DS_READ2ST64_B64,
+// DS_WRITE2ST64_B32, DS_WRITE2ST64_B64
+
+//===----------------------------------------------------------------------===//
+// MUBUF Instructions
+//===----------------------------------------------------------------------===//
 
 //def BUFFER_LOAD_FORMAT_X : MUBUF_ <0x00000000, "BUFFER_LOAD_FORMAT_X", []>;
 //def BUFFER_LOAD_FORMAT_XY : MUBUF_ <0x00000001, "BUFFER_LOAD_FORMAT_XY", []>;
@@ -417,32 +850,46 @@ defm BUFFER_LOAD_FORMAT_XYZW : MUBUF_Load_Helper <0x00000003, "BUFFER_LOAD_FORMA
 //def BUFFER_STORE_FORMAT_XY : MUBUF_ <0x00000005, "BUFFER_STORE_FORMAT_XY", []>;
 //def BUFFER_STORE_FORMAT_XYZ : MUBUF_ <0x00000006, "BUFFER_STORE_FORMAT_XYZ", []>;
 //def BUFFER_STORE_FORMAT_XYZW : MUBUF_ <0x00000007, "BUFFER_STORE_FORMAT_XYZW", []>;
-defm BUFFER_LOAD_UBYTE : MUBUF_Load_Helper <0x00000008, "BUFFER_LOAD_UBYTE", VReg_32>;
-defm BUFFER_LOAD_SBYTE : MUBUF_Load_Helper <0x00000009, "BUFFER_LOAD_SBYTE", VReg_32>;
-defm BUFFER_LOAD_USHORT : MUBUF_Load_Helper <0x0000000a, "BUFFER_LOAD_USHORT", VReg_32>;
-defm BUFFER_LOAD_SSHORT : MUBUF_Load_Helper <0x0000000b, "BUFFER_LOAD_SSHORT", VReg_32>;
-defm BUFFER_LOAD_DWORD : MUBUF_Load_Helper <0x0000000c, "BUFFER_LOAD_DWORD", VReg_32>;
-defm BUFFER_LOAD_DWORDX2 : MUBUF_Load_Helper <0x0000000d, "BUFFER_LOAD_DWORDX2", VReg_64>;
-defm BUFFER_LOAD_DWORDX4 : MUBUF_Load_Helper <0x0000000e, "BUFFER_LOAD_DWORDX4", VReg_128>;
+defm BUFFER_LOAD_UBYTE : MUBUF_Load_Helper <
+  0x00000008, "BUFFER_LOAD_UBYTE", VReg_32, i32, az_extloadi8_global
+>;
+defm BUFFER_LOAD_SBYTE : MUBUF_Load_Helper <
+  0x00000009, "BUFFER_LOAD_SBYTE", VReg_32, i32, sextloadi8_global
+>;
+defm BUFFER_LOAD_USHORT : MUBUF_Load_Helper <
+  0x0000000a, "BUFFER_LOAD_USHORT", VReg_32, i32, az_extloadi16_global
+>;
+defm BUFFER_LOAD_SSHORT : MUBUF_Load_Helper <
+  0x0000000b, "BUFFER_LOAD_SSHORT", VReg_32, i32, sextloadi16_global
+>;
+defm BUFFER_LOAD_DWORD : MUBUF_Load_Helper <
+  0x0000000c, "BUFFER_LOAD_DWORD", VReg_32, i32, global_load
+>;
+defm BUFFER_LOAD_DWORDX2 : MUBUF_Load_Helper <
+  0x0000000d, "BUFFER_LOAD_DWORDX2", VReg_64, v2i32, global_load
+>;
+defm BUFFER_LOAD_DWORDX4 : MUBUF_Load_Helper <
+  0x0000000e, "BUFFER_LOAD_DWORDX4", VReg_128, v4i32, global_load
+>;
 
-def BUFFER_STORE_BYTE : MUBUF_Store_Helper <
-  0x00000018, "BUFFER_STORE_BYTE", VReg_32
+defm BUFFER_STORE_BYTE : MUBUF_Store_Helper <
+  0x00000018, "BUFFER_STORE_BYTE", VReg_32, i32, truncstorei8_global
 >;
 
-def BUFFER_STORE_SHORT : MUBUF_Store_Helper <
-  0x0000001a, "BUFFER_STORE_SHORT", VReg_32
+defm BUFFER_STORE_SHORT : MUBUF_Store_Helper <
+  0x0000001a, "BUFFER_STORE_SHORT", VReg_32, i32, truncstorei16_global
 >;
 
-def BUFFER_STORE_DWORD : MUBUF_Store_Helper <
-  0x0000001c, "BUFFER_STORE_DWORD", VReg_32
+defm BUFFER_STORE_DWORD : MUBUF_Store_Helper <
+  0x0000001c, "BUFFER_STORE_DWORD", VReg_32, i32, global_store
 >;
 
-def BUFFER_STORE_DWORDX2 : MUBUF_Store_Helper <
-  0x0000001d, "BUFFER_STORE_DWORDX2", VReg_64
+defm BUFFER_STORE_DWORDX2 : MUBUF_Store_Helper <
+  0x0000001d, "BUFFER_STORE_DWORDX2", VReg_64, v2i32, global_store
 >;
 
-def BUFFER_STORE_DWORDX4 : MUBUF_Store_Helper <
-  0x0000001e, "BUFFER_STORE_DWORDX4", VReg_128
+defm BUFFER_STORE_DWORDX4 : MUBUF_Store_Helper <
+  0x0000001e, "BUFFER_STORE_DWORDX4", VReg_128, v4i32, global_store
 >;
 //def BUFFER_ATOMIC_SWAP : MUBUF_ <0x00000030, "BUFFER_ATOMIC_SWAP", []>;
 //def BUFFER_ATOMIC_CMPSWAP : MUBUF_ <0x00000031, "BUFFER_ATOMIC_CMPSWAP", []>;
@@ -480,6 +927,11 @@ def BUFFER_STORE_DWORDX4 : MUBUF_Store_Helper <
 //def BUFFER_ATOMIC_FMAX_X2 : MUBUF_X2 <0x00000060, "BUFFER_ATOMIC_FMAX_X2", []>;
 //def BUFFER_WBINVL1_SC : MUBUF_WBINVL1 <0x00000070, "BUFFER_WBINVL1_SC", []>;
 //def BUFFER_WBINVL1 : MUBUF_WBINVL1 <0x00000071, "BUFFER_WBINVL1", []>;
+
+//===----------------------------------------------------------------------===//
+// MTBUF Instructions
+//===----------------------------------------------------------------------===//
+
 //def TBUFFER_LOAD_FORMAT_X : MTBUF_ <0x00000000, "TBUFFER_LOAD_FORMAT_X", []>;
 //def TBUFFER_LOAD_FORMAT_XY : MTBUF_ <0x00000001, "TBUFFER_LOAD_FORMAT_XY", []>;
 //def TBUFFER_LOAD_FORMAT_XYZ : MTBUF_ <0x00000002, "TBUFFER_LOAD_FORMAT_XYZ", []>;
@@ -489,41 +941,10 @@ def TBUFFER_STORE_FORMAT_XY : MTBUF_Store_Helper <0x00000005, "TBUFFER_STORE_FOR
 def TBUFFER_STORE_FORMAT_XYZ : MTBUF_Store_Helper <0x00000006, "TBUFFER_STORE_FORMAT_XYZ", VReg_128>;
 def TBUFFER_STORE_FORMAT_XYZW : MTBUF_Store_Helper <0x00000007, "TBUFFER_STORE_FORMAT_XYZW", VReg_128>;
 
-let mayLoad = 1 in {
-
-// We are using the SGPR_32 and not the SReg_32 register class for 32-bit
-// SMRD instructions, because the SGPR_32 register class does not include M0
-// and writing to M0 from an SMRD instruction will hang the GPU.
-defm S_LOAD_DWORD : SMRD_Helper <0x00, "S_LOAD_DWORD", SReg_64, SGPR_32>;
-defm S_LOAD_DWORDX2 : SMRD_Helper <0x01, "S_LOAD_DWORDX2", SReg_64, SReg_64>;
-defm S_LOAD_DWORDX4 : SMRD_Helper <0x02, "S_LOAD_DWORDX4", SReg_64, SReg_128>;
-defm S_LOAD_DWORDX8 : SMRD_Helper <0x03, "S_LOAD_DWORDX8", SReg_64, SReg_256>;
-defm S_LOAD_DWORDX16 : SMRD_Helper <0x04, "S_LOAD_DWORDX16", SReg_64, SReg_512>;
-
-defm S_BUFFER_LOAD_DWORD : SMRD_Helper <
-  0x08, "S_BUFFER_LOAD_DWORD", SReg_128, SGPR_32
->;
-
-defm S_BUFFER_LOAD_DWORDX2 : SMRD_Helper <
-  0x09, "S_BUFFER_LOAD_DWORDX2", SReg_128, SReg_64
->;
-
-defm S_BUFFER_LOAD_DWORDX4 : SMRD_Helper <
-  0x0a, "S_BUFFER_LOAD_DWORDX4", SReg_128, SReg_128
->;
-
-defm S_BUFFER_LOAD_DWORDX8 : SMRD_Helper <
-  0x0b, "S_BUFFER_LOAD_DWORDX8", SReg_128, SReg_256
->;
-
-defm S_BUFFER_LOAD_DWORDX16 : SMRD_Helper <
-  0x0c, "S_BUFFER_LOAD_DWORDX16", SReg_128, SReg_512
->;
-
-} // mayLoad = 1
+//===----------------------------------------------------------------------===//
+// MIMG Instructions
+//===----------------------------------------------------------------------===//
 
-//def S_MEMTIME : SMRD_ <0x0000001e, "S_MEMTIME", []>;
-//def S_DCACHE_INV : SMRD_ <0x0000001f, "S_DCACHE_INV", []>;
 defm IMAGE_LOAD : MIMG_NoSampler <0x00000000, "IMAGE_LOAD">;
 defm IMAGE_LOAD_MIP : MIMG_NoSampler <0x00000001, "IMAGE_LOAD_MIP">;
 //def IMAGE_LOAD_PCK : MIMG_NoPattern_ <"IMAGE_LOAD_PCK", 0x00000002>;
@@ -552,81 +973,96 @@ defm IMAGE_GET_RESINFO : MIMG_NoSampler <0x0000000e, "IMAGE_GET_RESINFO">;
 //def IMAGE_ATOMIC_FCMPSWAP : MIMG_NoPattern_ <"IMAGE_ATOMIC_FCMPSWAP", 0x0000001d>;
 //def IMAGE_ATOMIC_FMIN : MIMG_NoPattern_ <"IMAGE_ATOMIC_FMIN", 0x0000001e>;
 //def IMAGE_ATOMIC_FMAX : MIMG_NoPattern_ <"IMAGE_ATOMIC_FMAX", 0x0000001f>;
-defm IMAGE_SAMPLE : MIMG_Sampler <0x00000020, "IMAGE_SAMPLE">;
-//def IMAGE_SAMPLE_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_CL", 0x00000021>;
-defm IMAGE_SAMPLE_D : MIMG_Sampler <0x00000022, "IMAGE_SAMPLE_D">;
-//def IMAGE_SAMPLE_D_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_D_CL", 0x00000023>;
-defm IMAGE_SAMPLE_L : MIMG_Sampler <0x00000024, "IMAGE_SAMPLE_L">;
-defm IMAGE_SAMPLE_B : MIMG_Sampler <0x00000025, "IMAGE_SAMPLE_B">;
-//def IMAGE_SAMPLE_B_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_B_CL", 0x00000026>;
-//def IMAGE_SAMPLE_LZ : MIMG_NoPattern_ <"IMAGE_SAMPLE_LZ", 0x00000027>;
-defm IMAGE_SAMPLE_C : MIMG_Sampler <0x00000028, "IMAGE_SAMPLE_C">;
-//def IMAGE_SAMPLE_C_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CL", 0x00000029>;
-defm IMAGE_SAMPLE_C_D : MIMG_Sampler <0x0000002a, "IMAGE_SAMPLE_C_D">;
-//def IMAGE_SAMPLE_C_D_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D_CL", 0x0000002b>;
-defm IMAGE_SAMPLE_C_L : MIMG_Sampler <0x0000002c, "IMAGE_SAMPLE_C_L">;
-defm IMAGE_SAMPLE_C_B : MIMG_Sampler <0x0000002d, "IMAGE_SAMPLE_C_B">;
-//def IMAGE_SAMPLE_C_B_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_B_CL", 0x0000002e>;
-//def IMAGE_SAMPLE_C_LZ : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_LZ", 0x0000002f>;
-//def IMAGE_SAMPLE_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_O", 0x00000030>;
-//def IMAGE_SAMPLE_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_CL_O", 0x00000031>;
-//def IMAGE_SAMPLE_D_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_D_O", 0x00000032>;
-//def IMAGE_SAMPLE_D_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_D_CL_O", 0x00000033>;
-//def IMAGE_SAMPLE_L_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_L_O", 0x00000034>;
-//def IMAGE_SAMPLE_B_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_B_O", 0x00000035>;
-//def IMAGE_SAMPLE_B_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_B_CL_O", 0x00000036>;
-//def IMAGE_SAMPLE_LZ_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_LZ_O", 0x00000037>;
-//def IMAGE_SAMPLE_C_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_O", 0x00000038>;
-//def IMAGE_SAMPLE_C_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CL_O", 0x00000039>;
-//def IMAGE_SAMPLE_C_D_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D_O", 0x0000003a>;
-//def IMAGE_SAMPLE_C_D_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D_CL_O", 0x0000003b>;
-//def IMAGE_SAMPLE_C_L_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_L_O", 0x0000003c>;
-//def IMAGE_SAMPLE_C_B_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_B_O", 0x0000003d>;
-//def IMAGE_SAMPLE_C_B_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_B_CL_O", 0x0000003e>;
-//def IMAGE_SAMPLE_C_LZ_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_LZ_O", 0x0000003f>;
-//def IMAGE_GATHER4 : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4", 0x00000040>;
-//def IMAGE_GATHER4_CL : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_CL", 0x00000041>;
-//def IMAGE_GATHER4_L : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_L", 0x00000044>;
-//def IMAGE_GATHER4_B : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_B", 0x00000045>;
-//def IMAGE_GATHER4_B_CL : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_B_CL", 0x00000046>;
-//def IMAGE_GATHER4_LZ : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_LZ", 0x00000047>;
-//def IMAGE_GATHER4_C : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C", 0x00000048>;
-//def IMAGE_GATHER4_C_CL : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_CL", 0x00000049>;
-//def IMAGE_GATHER4_C_L : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_L", 0x0000004c>;
-//def IMAGE_GATHER4_C_B : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_B", 0x0000004d>;
-//def IMAGE_GATHER4_C_B_CL : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_B_CL", 0x0000004e>;
-//def IMAGE_GATHER4_C_LZ : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_LZ", 0x0000004f>;
-//def IMAGE_GATHER4_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_O", 0x00000050>;
-//def IMAGE_GATHER4_CL_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_CL_O", 0x00000051>;
-//def IMAGE_GATHER4_L_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_L_O", 0x00000054>;
-//def IMAGE_GATHER4_B_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_B_O", 0x00000055>;
-//def IMAGE_GATHER4_B_CL_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_B_CL_O", 0x00000056>;
-//def IMAGE_GATHER4_LZ_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_LZ_O", 0x00000057>;
-//def IMAGE_GATHER4_C_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_O", 0x00000058>;
-//def IMAGE_GATHER4_C_CL_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_CL_O", 0x00000059>;
-//def IMAGE_GATHER4_C_L_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_L_O", 0x0000005c>;
-//def IMAGE_GATHER4_C_B_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_B_O", 0x0000005d>;
-//def IMAGE_GATHER4_C_B_CL_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_B_CL_O", 0x0000005e>;
-//def IMAGE_GATHER4_C_LZ_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_LZ_O", 0x0000005f>;
-//def IMAGE_GET_LOD : MIMG_NoPattern_ <"IMAGE_GET_LOD", 0x00000060>;
-//def IMAGE_SAMPLE_CD : MIMG_NoPattern_ <"IMAGE_SAMPLE_CD", 0x00000068>;
-//def IMAGE_SAMPLE_CD_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_CD_CL", 0x00000069>;
-//def IMAGE_SAMPLE_C_CD : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CD", 0x0000006a>;
-//def IMAGE_SAMPLE_C_CD_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CD_CL", 0x0000006b>;
-//def IMAGE_SAMPLE_CD_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_CD_O", 0x0000006c>;
-//def IMAGE_SAMPLE_CD_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_CD_CL_O", 0x0000006d>;
-//def IMAGE_SAMPLE_C_CD_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CD_O", 0x0000006e>;
-//def IMAGE_SAMPLE_C_CD_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CD_CL_O", 0x0000006f>;
+defm IMAGE_SAMPLE           : MIMG_Sampler <0x00000020, "IMAGE_SAMPLE">;
+defm IMAGE_SAMPLE_CL        : MIMG_Sampler <0x00000021, "IMAGE_SAMPLE_CL">;
+defm IMAGE_SAMPLE_D         : MIMG_Sampler <0x00000022, "IMAGE_SAMPLE_D">;
+defm IMAGE_SAMPLE_D_CL      : MIMG_Sampler <0x00000023, "IMAGE_SAMPLE_D_CL">;
+defm IMAGE_SAMPLE_L         : MIMG_Sampler <0x00000024, "IMAGE_SAMPLE_L">;
+defm IMAGE_SAMPLE_B         : MIMG_Sampler <0x00000025, "IMAGE_SAMPLE_B">;
+defm IMAGE_SAMPLE_B_CL      : MIMG_Sampler <0x00000026, "IMAGE_SAMPLE_B_CL">;
+defm IMAGE_SAMPLE_LZ        : MIMG_Sampler <0x00000027, "IMAGE_SAMPLE_LZ">;
+defm IMAGE_SAMPLE_C         : MIMG_Sampler <0x00000028, "IMAGE_SAMPLE_C">;
+defm IMAGE_SAMPLE_C_CL      : MIMG_Sampler <0x00000029, "IMAGE_SAMPLE_C_CL">;
+defm IMAGE_SAMPLE_C_D       : MIMG_Sampler <0x0000002a, "IMAGE_SAMPLE_C_D">;
+defm IMAGE_SAMPLE_C_D_CL    : MIMG_Sampler <0x0000002b, "IMAGE_SAMPLE_C_D_CL">;
+defm IMAGE_SAMPLE_C_L       : MIMG_Sampler <0x0000002c, "IMAGE_SAMPLE_C_L">;
+defm IMAGE_SAMPLE_C_B       : MIMG_Sampler <0x0000002d, "IMAGE_SAMPLE_C_B">;
+defm IMAGE_SAMPLE_C_B_CL    : MIMG_Sampler <0x0000002e, "IMAGE_SAMPLE_C_B_CL">;
+defm IMAGE_SAMPLE_C_LZ      : MIMG_Sampler <0x0000002f, "IMAGE_SAMPLE_C_LZ">;
+defm IMAGE_SAMPLE_O         : MIMG_Sampler <0x00000030, "IMAGE_SAMPLE_O">;
+defm IMAGE_SAMPLE_CL_O      : MIMG_Sampler <0x00000031, "IMAGE_SAMPLE_CL_O">;
+defm IMAGE_SAMPLE_D_O       : MIMG_Sampler <0x00000032, "IMAGE_SAMPLE_D_O">;
+defm IMAGE_SAMPLE_D_CL_O    : MIMG_Sampler <0x00000033, "IMAGE_SAMPLE_D_CL_O">;
+defm IMAGE_SAMPLE_L_O       : MIMG_Sampler <0x00000034, "IMAGE_SAMPLE_L_O">;
+defm IMAGE_SAMPLE_B_O       : MIMG_Sampler <0x00000035, "IMAGE_SAMPLE_B_O">;
+defm IMAGE_SAMPLE_B_CL_O    : MIMG_Sampler <0x00000036, "IMAGE_SAMPLE_B_CL_O">;
+defm IMAGE_SAMPLE_LZ_O      : MIMG_Sampler <0x00000037, "IMAGE_SAMPLE_LZ_O">;
+defm IMAGE_SAMPLE_C_O       : MIMG_Sampler <0x00000038, "IMAGE_SAMPLE_C_O">;
+defm IMAGE_SAMPLE_C_CL_O    : MIMG_Sampler <0x00000039, "IMAGE_SAMPLE_C_CL_O">;
+defm IMAGE_SAMPLE_C_D_O     : MIMG_Sampler <0x0000003a, "IMAGE_SAMPLE_C_D_O">;
+defm IMAGE_SAMPLE_C_D_CL_O  : MIMG_Sampler <0x0000003b, "IMAGE_SAMPLE_C_D_CL_O">;
+defm IMAGE_SAMPLE_C_L_O     : MIMG_Sampler <0x0000003c, "IMAGE_SAMPLE_C_L_O">;
+defm IMAGE_SAMPLE_C_B_O     : MIMG_Sampler <0x0000003d, "IMAGE_SAMPLE_C_B_O">;
+defm IMAGE_SAMPLE_C_B_CL_O  : MIMG_Sampler <0x0000003e, "IMAGE_SAMPLE_C_B_CL_O">;
+defm IMAGE_SAMPLE_C_LZ_O    : MIMG_Sampler <0x0000003f, "IMAGE_SAMPLE_C_LZ_O">;
+defm IMAGE_GATHER4          : MIMG_Gather <0x00000040, "IMAGE_GATHER4">;
+defm IMAGE_GATHER4_CL       : MIMG_Gather <0x00000041, "IMAGE_GATHER4_CL">;
+defm IMAGE_GATHER4_L        : MIMG_Gather <0x00000044, "IMAGE_GATHER4_L">;
+defm IMAGE_GATHER4_B        : MIMG_Gather <0x00000045, "IMAGE_GATHER4_B">;
+defm IMAGE_GATHER4_B_CL     : MIMG_Gather <0x00000046, "IMAGE_GATHER4_B_CL">;
+defm IMAGE_GATHER4_LZ       : MIMG_Gather <0x00000047, "IMAGE_GATHER4_LZ">;
+defm IMAGE_GATHER4_C        : MIMG_Gather <0x00000048, "IMAGE_GATHER4_C">;
+defm IMAGE_GATHER4_C_CL     : MIMG_Gather <0x00000049, "IMAGE_GATHER4_C_CL">;
+defm IMAGE_GATHER4_C_L      : MIMG_Gather <0x0000004c, "IMAGE_GATHER4_C_L">;
+defm IMAGE_GATHER4_C_B      : MIMG_Gather <0x0000004d, "IMAGE_GATHER4_C_B">;
+defm IMAGE_GATHER4_C_B_CL   : MIMG_Gather <0x0000004e, "IMAGE_GATHER4_C_B_CL">;
+defm IMAGE_GATHER4_C_LZ     : MIMG_Gather <0x0000004f, "IMAGE_GATHER4_C_LZ">;
+defm IMAGE_GATHER4_O        : MIMG_Gather <0x00000050, "IMAGE_GATHER4_O">;
+defm IMAGE_GATHER4_CL_O     : MIMG_Gather <0x00000051, "IMAGE_GATHER4_CL_O">;
+defm IMAGE_GATHER4_L_O      : MIMG_Gather <0x00000054, "IMAGE_GATHER4_L_O">;
+defm IMAGE_GATHER4_B_O      : MIMG_Gather <0x00000055, "IMAGE_GATHER4_B_O">;
+defm IMAGE_GATHER4_B_CL_O   : MIMG_Gather <0x00000056, "IMAGE_GATHER4_B_CL_O">;
+defm IMAGE_GATHER4_LZ_O     : MIMG_Gather <0x00000057, "IMAGE_GATHER4_LZ_O">;
+defm IMAGE_GATHER4_C_O      : MIMG_Gather <0x00000058, "IMAGE_GATHER4_C_O">;
+defm IMAGE_GATHER4_C_CL_O   : MIMG_Gather <0x00000059, "IMAGE_GATHER4_C_CL_O">;
+defm IMAGE_GATHER4_C_L_O    : MIMG_Gather <0x0000005c, "IMAGE_GATHER4_C_L_O">;
+defm IMAGE_GATHER4_C_B_O    : MIMG_Gather <0x0000005d, "IMAGE_GATHER4_C_B_O">;
+defm IMAGE_GATHER4_C_B_CL_O : MIMG_Gather <0x0000005e, "IMAGE_GATHER4_C_B_CL_O">;
+defm IMAGE_GATHER4_C_LZ_O   : MIMG_Gather <0x0000005f, "IMAGE_GATHER4_C_LZ_O">;
+defm IMAGE_GET_LOD          : MIMG_Sampler <0x00000060, "IMAGE_GET_LOD">;
+defm IMAGE_SAMPLE_CD        : MIMG_Sampler <0x00000068, "IMAGE_SAMPLE_CD">;
+defm IMAGE_SAMPLE_CD_CL     : MIMG_Sampler <0x00000069, "IMAGE_SAMPLE_CD_CL">;
+defm IMAGE_SAMPLE_C_CD      : MIMG_Sampler <0x0000006a, "IMAGE_SAMPLE_C_CD">;
+defm IMAGE_SAMPLE_C_CD_CL   : MIMG_Sampler <0x0000006b, "IMAGE_SAMPLE_C_CD_CL">;
+defm IMAGE_SAMPLE_CD_O      : MIMG_Sampler <0x0000006c, "IMAGE_SAMPLE_CD_O">;
+defm IMAGE_SAMPLE_CD_CL_O   : MIMG_Sampler <0x0000006d, "IMAGE_SAMPLE_CD_CL_O">;
+defm IMAGE_SAMPLE_C_CD_O    : MIMG_Sampler <0x0000006e, "IMAGE_SAMPLE_C_CD_O">;
+defm IMAGE_SAMPLE_C_CD_CL_O : MIMG_Sampler <0x0000006f, "IMAGE_SAMPLE_C_CD_CL_O">;
 //def IMAGE_RSRC256 : MIMG_NoPattern_RSRC256 <"IMAGE_RSRC256", 0x0000007e>;
 //def IMAGE_SAMPLER : MIMG_NoPattern_ <"IMAGE_SAMPLER", 0x0000007f>;
-//def V_NOP : VOP1_ <0x00000000, "V_NOP", []>;
 
+//===----------------------------------------------------------------------===//
+// VOP1 Instructions
+//===----------------------------------------------------------------------===//
+
+//def V_NOP : VOP1_ <0x00000000, "V_NOP", []>;
 
 let neverHasSideEffects = 1, isMoveImm = 1 in {
 defm V_MOV_B32 : VOP1_32 <0x00000001, "V_MOV_B32", []>;
 } // End neverHasSideEffects = 1, isMoveImm = 1
 
-defm V_READFIRSTLANE_B32 : VOP1_32 <0x00000002, "V_READFIRSTLANE_B32", []>;
+let Uses = [EXEC] in {
+
+def V_READFIRSTLANE_B32 : VOP1 <
+  0x00000002,
+  (outs SReg_32:$vdst),
+  (ins VReg_32:$src0),
+  "V_READFIRSTLANE_B32 $vdst, $src0",
+  []
+>;
+
+}
+
 defm V_CVT_I32_F64 : VOP1_32_64 <0x00000003, "V_CVT_I32_F64",
   [(set i32:$dst, (fp_to_sint f64:$src0))]
 >;
@@ -646,8 +1082,12 @@ defm V_CVT_I32_F32 : VOP1_32 <0x00000008, "V_CVT_I32_F32",
   [(set i32:$dst, (fp_to_sint f32:$src0))]
 >;
 defm V_MOV_FED_B32 : VOP1_32 <0x00000009, "V_MOV_FED_B32", []>;
-////def V_CVT_F16_F32 : VOP1_F16 <0x0000000a, "V_CVT_F16_F32", []>;
-//defm V_CVT_F32_F16 : VOP1_32 <0x0000000b, "V_CVT_F32_F16", []>;
+defm V_CVT_F16_F32 : VOP1_32 <0x0000000a, "V_CVT_F16_F32",
+  [(set i32:$dst, (fp_to_f16 f32:$src0))]
+>;
+defm V_CVT_F32_F16 : VOP1_32 <0x0000000b, "V_CVT_F32_F16",
+  [(set f32:$dst, (f16_to_fp i32:$src0))]
+>;
 //defm V_CVT_RPI_I32_F32 : VOP1_32 <0x0000000c, "V_CVT_RPI_I32_F32", []>;
 //defm V_CVT_FLR_I32_F32 : VOP1_32 <0x0000000d, "V_CVT_FLR_I32_F32", []>;
 //defm V_CVT_OFF_F32_I4 : VOP1_32 <0x0000000e, "V_CVT_OFF_F32_I4", []>;
@@ -657,17 +1097,30 @@ defm V_CVT_F32_F64 : VOP1_32_64 <0x0000000f, "V_CVT_F32_F64",
 defm V_CVT_F64_F32 : VOP1_64_32 <0x00000010, "V_CVT_F64_F32",
   [(set f64:$dst, (fextend f32:$src0))]
 >;
-//defm V_CVT_F32_UBYTE0 : VOP1_32 <0x00000011, "V_CVT_F32_UBYTE0", []>;
-//defm V_CVT_F32_UBYTE1 : VOP1_32 <0x00000012, "V_CVT_F32_UBYTE1", []>;
-//defm V_CVT_F32_UBYTE2 : VOP1_32 <0x00000013, "V_CVT_F32_UBYTE2", []>;
-//defm V_CVT_F32_UBYTE3 : VOP1_32 <0x00000014, "V_CVT_F32_UBYTE3", []>;
-//defm V_CVT_U32_F64 : VOP1_32 <0x00000015, "V_CVT_U32_F64", []>;
-//defm V_CVT_F64_U32 : VOP1_64 <0x00000016, "V_CVT_F64_U32", []>;
+defm V_CVT_F32_UBYTE0 : VOP1_32 <0x00000011, "V_CVT_F32_UBYTE0",
+  [(set f32:$dst, (AMDGPUcvt_f32_ubyte0 i32:$src0))]
+>;
+defm V_CVT_F32_UBYTE1 : VOP1_32 <0x00000012, "V_CVT_F32_UBYTE1",
+  [(set f32:$dst, (AMDGPUcvt_f32_ubyte1 i32:$src0))]
+>;
+defm V_CVT_F32_UBYTE2 : VOP1_32 <0x00000013, "V_CVT_F32_UBYTE2",
+  [(set f32:$dst, (AMDGPUcvt_f32_ubyte2 i32:$src0))]
+>;
+defm V_CVT_F32_UBYTE3 : VOP1_32 <0x00000014, "V_CVT_F32_UBYTE3",
+  [(set f32:$dst, (AMDGPUcvt_f32_ubyte3 i32:$src0))]
+>;
+defm V_CVT_U32_F64 : VOP1_32_64 <0x00000015, "V_CVT_U32_F64",
+  [(set i32:$dst, (fp_to_uint f64:$src0))]
+>;
+defm V_CVT_F64_U32 : VOP1_64_32 <0x00000016, "V_CVT_F64_U32",
+  [(set f64:$dst, (uint_to_fp i32:$src0))]
+>;
+
 defm V_FRACT_F32 : VOP1_32 <0x00000020, "V_FRACT_F32",
   [(set f32:$dst, (AMDGPUfract f32:$src0))]
 >;
 defm V_TRUNC_F32 : VOP1_32 <0x00000021, "V_TRUNC_F32",
-  [(set f32:$dst, (int_AMDGPU_trunc f32:$src0))]
+  [(set f32:$dst, (ftrunc f32:$src0))]
 >;
 defm V_CEIL_F32 : VOP1_32 <0x00000022, "V_CEIL_F32",
   [(set f32:$dst, (fceil f32:$src0))]
@@ -685,32 +1138,45 @@ defm V_LOG_CLAMP_F32 : VOP1_32 <0x00000026, "V_LOG_CLAMP_F32", []>;
 defm V_LOG_F32 : VOP1_32 <0x00000027, "V_LOG_F32",
   [(set f32:$dst, (flog2 f32:$src0))]
 >;
+
 defm V_RCP_CLAMP_F32 : VOP1_32 <0x00000028, "V_RCP_CLAMP_F32", []>;
 defm V_RCP_LEGACY_F32 : VOP1_32 <0x00000029, "V_RCP_LEGACY_F32", []>;
 defm V_RCP_F32 : VOP1_32 <0x0000002a, "V_RCP_F32",
-  [(set f32:$dst, (fdiv FP_ONE, f32:$src0))]
+  [(set f32:$dst, (AMDGPUrcp f32:$src0))]
 >;
 defm V_RCP_IFLAG_F32 : VOP1_32 <0x0000002b, "V_RCP_IFLAG_F32", []>;
-defm V_RSQ_CLAMP_F32 : VOP1_32 <0x0000002c, "V_RSQ_CLAMP_F32", []>;
+defm V_RSQ_CLAMP_F32 : VOP1_32 <0x0000002c, "V_RSQ_CLAMP_F32",
+  [(set f32:$dst, (AMDGPUrsq_clamped f32:$src0))]
+>;
 defm V_RSQ_LEGACY_F32 : VOP1_32 <
   0x0000002d, "V_RSQ_LEGACY_F32",
-  [(set f32:$dst, (int_AMDGPU_rsq f32:$src0))]
+  [(set f32:$dst, (AMDGPUrsq_legacy f32:$src0))]
+>;
+defm V_RSQ_F32 : VOP1_32 <0x0000002e, "V_RSQ_F32",
+  [(set f32:$dst, (AMDGPUrsq f32:$src0))]
 >;
-defm V_RSQ_F32 : VOP1_32 <0x0000002e, "V_RSQ_F32", []>;
 defm V_RCP_F64 : VOP1_64 <0x0000002f, "V_RCP_F64",
-  [(set f64:$dst, (fdiv FP_ONE, f64:$src0))]
+  [(set f64:$dst, (AMDGPUrcp f64:$src0))]
 >;
 defm V_RCP_CLAMP_F64 : VOP1_64 <0x00000030, "V_RCP_CLAMP_F64", []>;
-defm V_RSQ_F64 : VOP1_64 <0x00000031, "V_RSQ_F64", []>;
-defm V_RSQ_CLAMP_F64 : VOP1_64 <0x00000032, "V_RSQ_CLAMP_F64", []>;
+defm V_RSQ_F64 : VOP1_64 <0x00000031, "V_RSQ_F64",
+  [(set f64:$dst, (AMDGPUrsq f64:$src0))]
+>;
+defm V_RSQ_CLAMP_F64 : VOP1_64 <0x00000032, "V_RSQ_CLAMP_F64",
+  [(set f64:$dst, (AMDGPUrsq_clamped f64:$src0))]
+>;
 defm V_SQRT_F32 : VOP1_32 <0x00000033, "V_SQRT_F32",
   [(set f32:$dst, (fsqrt f32:$src0))]
 >;
 defm V_SQRT_F64 : VOP1_64 <0x00000034, "V_SQRT_F64",
   [(set f64:$dst, (fsqrt f64:$src0))]
 >;
-defm V_SIN_F32 : VOP1_32 <0x00000035, "V_SIN_F32", []>;
-defm V_COS_F32 : VOP1_32 <0x00000036, "V_COS_F32", []>;
+defm V_SIN_F32 : VOP1_32 <0x00000035, "V_SIN_F32",
+  [(set f32:$dst, (AMDGPUsin f32:$src0))]
+>;
+defm V_COS_F32 : VOP1_32 <0x00000036, "V_COS_F32",
+  [(set f32:$dst, (AMDGPUcos f32:$src0))]
+>;
 defm V_NOT_B32 : VOP1_32 <0x00000037, "V_NOT_B32", []>;
 defm V_BFREV_B32 : VOP1_32 <0x00000038, "V_BFREV_B32", []>;
 defm V_FFBH_U32 : VOP1_32 <0x00000039, "V_FFBH_U32", []>;
@@ -726,6 +1192,11 @@ defm V_MOVRELD_B32 : VOP1_32 <0x00000042, "V_MOVRELD_B32", []>;
 defm V_MOVRELS_B32 : VOP1_32 <0x00000043, "V_MOVRELS_B32", []>;
 defm V_MOVRELSD_B32 : VOP1_32 <0x00000044, "V_MOVRELSD_B32", []>;
 
+
+//===----------------------------------------------------------------------===//
+// VINTRP Instructions
+//===----------------------------------------------------------------------===//
+
 def V_INTERP_P1_F32 : VINTRP <
   0x00000000,
   (outs VReg_32:$dst),
@@ -756,97 +1227,9 @@ def V_INTERP_MOV_F32 : VINTRP <
   let DisableEncoding = "$m0";
 }
 
-//def S_NOP : SOPP_ <0x00000000, "S_NOP", []>;
-
-let isTerminator = 1 in {
-
-def S_ENDPGM : SOPP <0x00000001, (ins), "S_ENDPGM",
-  [(IL_retflag)]> {
-  let SIMM16 = 0;
-  let isBarrier = 1;
-  let hasCtrlDep = 1;
-}
-
-let isBranch = 1 in {
-def S_BRANCH : SOPP <
-  0x00000002, (ins brtarget:$target), "S_BRANCH $target",
-  [(br bb:$target)]> {
-  let isBarrier = 1;
-}
-
-let DisableEncoding = "$scc" in {
-def S_CBRANCH_SCC0 : SOPP <
-  0x00000004, (ins brtarget:$target, SCCReg:$scc),
-  "S_CBRANCH_SCC0 $target", []
->;
-def S_CBRANCH_SCC1 : SOPP <
-  0x00000005, (ins brtarget:$target, SCCReg:$scc),
-  "S_CBRANCH_SCC1 $target",
-  []
->;
-} // End DisableEncoding = "$scc"
-
-def S_CBRANCH_VCCZ : SOPP <
-  0x00000006, (ins brtarget:$target, VCCReg:$vcc),
-  "S_CBRANCH_VCCZ $target",
-  []
->;
-def S_CBRANCH_VCCNZ : SOPP <
-  0x00000007, (ins brtarget:$target, VCCReg:$vcc),
-  "S_CBRANCH_VCCNZ $target",
-  []
->;
-
-let DisableEncoding = "$exec" in {
-def S_CBRANCH_EXECZ : SOPP <
-  0x00000008, (ins brtarget:$target, EXECReg:$exec),
-  "S_CBRANCH_EXECZ $target",
-  []
->;
-def S_CBRANCH_EXECNZ : SOPP <
-  0x00000009, (ins brtarget:$target, EXECReg:$exec),
-  "S_CBRANCH_EXECNZ $target",
-  []
->;
-} // End DisableEncoding = "$exec"
-
-
-} // End isBranch = 1
-} // End isTerminator = 1
-
-let hasSideEffects = 1 in {
-def S_BARRIER : SOPP <0x0000000a, (ins), "S_BARRIER",
-  [(int_AMDGPU_barrier_local)]
-> {
-  let SIMM16 = 0;
-  let isBarrier = 1;
-  let hasCtrlDep = 1;
-  let mayLoad = 1;
-  let mayStore = 1;
-}
-
-def S_WAITCNT : SOPP <0x0000000c, (ins WAIT_FLAG:$simm16), "S_WAITCNT $simm16",
-  []
->;
-//def S_SETHALT : SOPP_ <0x0000000d, "S_SETHALT", []>;
-//def S_SLEEP : SOPP_ <0x0000000e, "S_SLEEP", []>;
-//def S_SETPRIO : SOPP_ <0x0000000f, "S_SETPRIO", []>;
-
-let Uses = [EXEC] in {
-  def S_SENDMSG : SOPP <0x00000010, (ins SendMsgImm:$simm16, M0Reg:$m0), "S_SENDMSG $simm16",
-      [(int_SI_sendmsg imm:$simm16, M0Reg:$m0)]
-  > {
-    let DisableEncoding = "$m0";
-  }
-} // End Uses = [EXEC]
-
-//def S_SENDMSGHALT : SOPP_ <0x00000011, "S_SENDMSGHALT", []>;
-//def S_TRAP : SOPP_ <0x00000012, "S_TRAP", []>;
-//def S_ICACHE_INV : SOPP_ <0x00000013, "S_ICACHE_INV", []>;
-//def S_INCPERFLEVEL : SOPP_ <0x00000014, "S_INCPERFLEVEL", []>;
-//def S_DECPERFLEVEL : SOPP_ <0x00000015, "S_DECPERFLEVEL", []>;
-//def S_TTRACEDATA : SOPP_ <0x00000016, "S_TTRACEDATA", []>;
-} // End hasSideEffects
+//===----------------------------------------------------------------------===//
+// VOP2 Instructions
+//===----------------------------------------------------------------------===//
 
 def V_CNDMASK_B32_e32 : VOP2 <0x00000000, (outs VReg_32:$dst),
   (ins VSrc_32:$src0, VReg_32:$src1, VCCReg:$vcc),
@@ -861,34 +1244,28 @@ def V_CNDMASK_B32_e64 : VOP3 <0x00000100, (outs VReg_32:$dst),
    InstFlag:$abs, InstFlag:$clamp, InstFlag:$omod, InstFlag:$neg),
   "V_CNDMASK_B32_e64 $dst, $src0, $src1, $src2, $abs, $clamp, $omod, $neg",
   [(set i32:$dst, (select i1:$src2, i32:$src1, i32:$src0))]
->;
-
-//f32 pattern for V_CNDMASK_B32_e64
-def : Pat <
-  (f32 (select i1:$src2, f32:$src1, f32:$src0)),
-  (V_CNDMASK_B32_e64 $src0, $src1, $src2)
->;
+> {
+  let src0_modifiers = 0;
+  let src1_modifiers = 0;
+  let src2_modifiers = 0;
+}
 
-def : Pat <
-  (i32 (trunc i64:$val)),
-  (EXTRACT_SUBREG $val, sub0)
+def V_READLANE_B32 : VOP2 <
+  0x00000001,
+  (outs SReg_32:$vdst),
+  (ins VReg_32:$src0, SSrc_32:$vsrc1),
+  "V_READLANE_B32 $vdst, $src0, $vsrc1",
+  []
 >;
 
-//use two V_CNDMASK_B32_e64 instructions for f64
-def : Pat <
-  (f64 (select i1:$src2, f64:$src1, f64:$src0)),
-  (INSERT_SUBREG (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
-  (V_CNDMASK_B32_e64 (EXTRACT_SUBREG $src0, sub0),
-                     (EXTRACT_SUBREG $src1, sub0),
-                     $src2), sub0),
-  (V_CNDMASK_B32_e64 (EXTRACT_SUBREG $src0, sub1),
-                     (EXTRACT_SUBREG $src1, sub1),
-                     $src2), sub1)
+def V_WRITELANE_B32 : VOP2 <
+  0x00000002,
+  (outs VReg_32:$vdst),
+  (ins SReg_32:$src0, SSrc_32:$vsrc1),
+  "V_WRITELANE_B32 $vdst, $src0, $vsrc1",
+  []
 >;
 
-defm V_READLANE_B32 : VOP2_32 <0x00000001, "V_READLANE_B32", []>;
-defm V_WRITELANE_B32 : VOP2_32 <0x00000002, "V_WRITELANE_B32", []>;
-
 let isCommutable = 1 in {
 defm V_ADD_F32 : VOP2_32 <0x00000003, "V_ADD_F32",
   [(set f32:$dst, (fadd f32:$src0, f32:$src1))]
@@ -915,11 +1292,11 @@ defm V_MUL_F32 : VOP2_32 <0x00000008, "V_MUL_F32",
 
 
 defm V_MUL_I32_I24 : VOP2_32 <0x00000009, "V_MUL_I32_I24",
-  [(set i32:$dst, (mul I24:$src0, I24:$src1))]
+  [(set i32:$dst, (AMDGPUmul_i24 i32:$src0, i32:$src1))]
 >;
 //defm V_MUL_HI_I32_I24 : VOP2_32 <0x0000000a, "V_MUL_HI_I32_I24", []>;
 defm V_MUL_U32_U24 : VOP2_32 <0x0000000b, "V_MUL_U32_U24",
-  [(set i32:$dst, (mul U24:$src0, U24:$src1))]
+  [(set i32:$dst, (AMDGPUmul_u24 i32:$src0, i32:$src1))]
 >;
 //defm V_MUL_HI_U32_U24 : VOP2_32 <0x0000000c, "V_MUL_HI_U32_U24", []>;
 
@@ -935,21 +1312,18 @@ defm V_MAX_LEGACY_F32 : VOP2_32 <0x0000000e, "V_MAX_LEGACY_F32",
 defm V_MIN_F32 : VOP2_32 <0x0000000f, "V_MIN_F32", []>;
 defm V_MAX_F32 : VOP2_32 <0x00000010, "V_MAX_F32", []>;
 defm V_MIN_I32 : VOP2_32 <0x00000011, "V_MIN_I32",
-  [(set i32:$dst, (AMDGPUsmin i32:$src0, i32:$src1))]
->;
+  [(set i32:$dst, (AMDGPUsmin i32:$src0, i32:$src1))]>;
 defm V_MAX_I32 : VOP2_32 <0x00000012, "V_MAX_I32",
-  [(set i32:$dst, (AMDGPUsmax i32:$src0, i32:$src1))]
->;
+  [(set i32:$dst, (AMDGPUsmax i32:$src0, i32:$src1))]>;
 defm V_MIN_U32 : VOP2_32 <0x00000013, "V_MIN_U32",
-  [(set i32:$dst, (AMDGPUumin i32:$src0, i32:$src1))]
->;
+  [(set i32:$dst, (AMDGPUumin i32:$src0, i32:$src1))]>;
 defm V_MAX_U32 : VOP2_32 <0x00000014, "V_MAX_U32",
-  [(set i32:$dst, (AMDGPUumax i32:$src0, i32:$src1))]
->;
+  [(set i32:$dst, (AMDGPUumax i32:$src0, i32:$src1))]>;
 
 defm V_LSHR_B32 : VOP2_32 <0x00000015, "V_LSHR_B32",
   [(set i32:$dst, (srl i32:$src0, i32:$src1))]
 >;
+
 defm V_LSHRREV_B32 : VOP2_32 <0x00000016, "V_LSHRREV_B32", [], "V_LSHR_B32">;
 
 defm V_ASHR_I32 : VOP2_32 <0x00000017, "V_ASHR_I32",
@@ -967,8 +1341,7 @@ defm V_LSHL_B32 : VOP2_32 <0x00000019, "V_LSHL_B32",
 defm V_LSHLREV_B32 : VOP2_32 <0x0000001a, "V_LSHLREV_B32", [], "V_LSHL_B32">;
 
 defm V_AND_B32 : VOP2_32 <0x0000001b, "V_AND_B32",
-  [(set i32:$dst, (and i32:$src0, i32:$src1))]
->;
+  [(set i32:$dst, (and i32:$src0, i32:$src1))]>;
 defm V_OR_B32 : VOP2_32 <0x0000001c, "V_OR_B32",
   [(set i32:$dst, (or i32:$src0, i32:$src1))]
 >;
@@ -978,25 +1351,30 @@ defm V_XOR_B32 : VOP2_32 <0x0000001d, "V_XOR_B32",
 
 } // End isCommutable = 1
 
-defm V_BFM_B32 : VOP2_32 <0x0000001e, "V_BFM_B32", []>;
+defm V_BFM_B32 : VOP2_32 <0x0000001e, "V_BFM_B32",
+  [(set i32:$dst, (AMDGPUbfm i32:$src0, i32:$src1))]>;
 defm V_MAC_F32 : VOP2_32 <0x0000001f, "V_MAC_F32", []>;
 defm V_MADMK_F32 : VOP2_32 <0x00000020, "V_MADMK_F32", []>;
 defm V_MADAK_F32 : VOP2_32 <0x00000021, "V_MADAK_F32", []>;
-//defm V_BCNT_U32_B32 : VOP2_32 <0x00000022, "V_BCNT_U32_B32", []>;
+defm V_BCNT_U32_B32 : VOP2_32 <0x00000022, "V_BCNT_U32_B32", []>;
 defm V_MBCNT_LO_U32_B32 : VOP2_32 <0x00000023, "V_MBCNT_LO_U32_B32", []>;
 defm V_MBCNT_HI_U32_B32 : VOP2_32 <0x00000024, "V_MBCNT_HI_U32_B32", []>;
 
 let isCommutable = 1, Defs = [VCC] in { // Carry-out goes to VCC
 // No patterns so that the scalar instructions are always selected.
 // The scalar versions will be replaced with vector when needed later.
-defm V_ADD_I32 : VOP2b_32 <0x00000025, "V_ADD_I32", [], VSrc_32>;
-defm V_SUB_I32 : VOP2b_32 <0x00000026, "V_SUB_I32", [], VSrc_32>;
+defm V_ADD_I32 : VOP2b_32 <0x00000025, "V_ADD_I32",
+  [(set i32:$dst, (add i32:$src0, i32:$src1))], VSrc_32>;
+defm V_SUB_I32 : VOP2b_32 <0x00000026, "V_SUB_I32",
+  [(set i32:$dst, (sub i32:$src0, i32:$src1))], VSrc_32>;
 defm V_SUBREV_I32 : VOP2b_32 <0x00000027, "V_SUBREV_I32", [], VSrc_32,
                               "V_SUB_I32">;
 
 let Uses = [VCC] in { // Carry-in comes from VCC
-defm V_ADDC_U32 : VOP2b_32 <0x00000028, "V_ADDC_U32", [], VReg_32>;
-defm V_SUBB_U32 : VOP2b_32 <0x00000029, "V_SUBB_U32", [], VReg_32>;
+defm V_ADDC_U32 : VOP2b_32 <0x00000028, "V_ADDC_U32",
+  [(set i32:$dst, (adde i32:$src0, i32:$src1))], VReg_32>;
+defm V_SUBB_U32 : VOP2b_32 <0x00000029, "V_SUBB_U32",
+  [(set i32:$dst, (sube i32:$src0, i32:$src1))], VReg_32>;
 defm V_SUBBREV_U32 : VOP2b_32 <0x0000002a, "V_SUBBREV_U32", [], VReg_32,
                                "V_SUBB_U32">;
 } // End Uses = [VCC]
@@ -1011,56 +1389,51 @@ defm V_CVT_PKRTZ_F16_F32 : VOP2_32 <0x0000002f, "V_CVT_PKRTZ_F16_F32",
 >;
 ////def V_CVT_PK_U16_U32 : VOP2_U16 <0x00000030, "V_CVT_PK_U16_U32", []>;
 ////def V_CVT_PK_I16_I32 : VOP2_I16 <0x00000031, "V_CVT_PK_I16_I32", []>;
-def S_CMP_EQ_I32 : SOPC_32 <0x00000000, "S_CMP_EQ_I32", []>;
-def S_CMP_LG_I32 : SOPC_32 <0x00000001, "S_CMP_LG_I32", []>;
-def S_CMP_GT_I32 : SOPC_32 <0x00000002, "S_CMP_GT_I32", []>;
-def S_CMP_GE_I32 : SOPC_32 <0x00000003, "S_CMP_GE_I32", []>;
-def S_CMP_LT_I32 : SOPC_32 <0x00000004, "S_CMP_LT_I32", []>;
-def S_CMP_LE_I32 : SOPC_32 <0x00000005, "S_CMP_LE_I32", []>;
-def S_CMP_EQ_U32 : SOPC_32 <0x00000006, "S_CMP_EQ_U32", []>;
-def S_CMP_LG_U32 : SOPC_32 <0x00000007, "S_CMP_LG_U32", []>;
-def S_CMP_GT_U32 : SOPC_32 <0x00000008, "S_CMP_GT_U32", []>;
-def S_CMP_GE_U32 : SOPC_32 <0x00000009, "S_CMP_GE_U32", []>;
-def S_CMP_LT_U32 : SOPC_32 <0x0000000a, "S_CMP_LT_U32", []>;
-def S_CMP_LE_U32 : SOPC_32 <0x0000000b, "S_CMP_LE_U32", []>;
-////def S_BITCMP0_B32 : SOPC_BITCMP0 <0x0000000c, "S_BITCMP0_B32", []>;
-////def S_BITCMP1_B32 : SOPC_BITCMP1 <0x0000000d, "S_BITCMP1_B32", []>;
-////def S_BITCMP0_B64 : SOPC_BITCMP0 <0x0000000e, "S_BITCMP0_B64", []>;
-////def S_BITCMP1_B64 : SOPC_BITCMP1 <0x0000000f, "S_BITCMP1_B64", []>;
-//def S_SETVSKIP : SOPC_ <0x00000010, "S_SETVSKIP", []>;
+
+//===----------------------------------------------------------------------===//
+// VOP3 Instructions
+//===----------------------------------------------------------------------===//
 
 let neverHasSideEffects = 1 in {
 
-def V_MAD_LEGACY_F32 : VOP3_32 <0x00000140, "V_MAD_LEGACY_F32", []>;
-def V_MAD_F32 : VOP3_32 <0x00000141, "V_MAD_F32", []>;
-def V_MAD_I32_I24 : VOP3_32 <0x00000142, "V_MAD_I32_I24",
-  [(set i32:$dst, (add (mul I24:$src0, I24:$src1), i32:$src2))]
+defm V_MAD_LEGACY_F32 : VOP3_32 <0x00000140, "V_MAD_LEGACY_F32", []>;
+defm V_MAD_F32 : VOP3_32 <0x00000141, "V_MAD_F32",
+  [(set f32:$dst, (fadd (fmul f32:$src0, f32:$src1), f32:$src2))]
 >;
-def V_MAD_U32_U24 : VOP3_32 <0x00000143, "V_MAD_U32_U24",
-  [(set i32:$dst, (add (mul U24:$src0, U24:$src1), i32:$src2))]
+defm V_MAD_I32_I24 : VOP3_32 <0x00000142, "V_MAD_I32_I24",
+  [(set i32:$dst, (AMDGPUmad_i24 i32:$src0, i32:$src1, i32:$src2))]
+>;
+defm V_MAD_U32_U24 : VOP3_32 <0x00000143, "V_MAD_U32_U24",
+  [(set i32:$dst, (AMDGPUmad_u24 i32:$src0, i32:$src1, i32:$src2))]
 >;
 
 } // End neverHasSideEffects
-def V_CUBEID_F32 : VOP3_32 <0x00000144, "V_CUBEID_F32", []>;
-def V_CUBESC_F32 : VOP3_32 <0x00000145, "V_CUBESC_F32", []>;
-def V_CUBETC_F32 : VOP3_32 <0x00000146, "V_CUBETC_F32", []>;
-def V_CUBEMA_F32 : VOP3_32 <0x00000147, "V_CUBEMA_F32", []>;
-def V_BFE_U32 : VOP3_32 <0x00000148, "V_BFE_U32", []>;
-def V_BFE_I32 : VOP3_32 <0x00000149, "V_BFE_I32", []>;
-def V_BFI_B32 : VOP3_32 <0x0000014a, "V_BFI_B32", []>;
-defm : BFIPatterns <V_BFI_B32>;
-def V_FMA_F32 : VOP3_32 <0x0000014b, "V_FMA_F32",
+
+defm V_CUBEID_F32 : VOP3_32 <0x00000144, "V_CUBEID_F32", []>;
+defm V_CUBESC_F32 : VOP3_32 <0x00000145, "V_CUBESC_F32", []>;
+defm V_CUBETC_F32 : VOP3_32 <0x00000146, "V_CUBETC_F32", []>;
+defm V_CUBEMA_F32 : VOP3_32 <0x00000147, "V_CUBEMA_F32", []>;
+
+let neverHasSideEffects = 1, mayLoad = 0, mayStore = 0 in {
+defm V_BFE_U32 : VOP3_32 <0x00000148, "V_BFE_U32",
+  [(set i32:$dst, (AMDGPUbfe_u32 i32:$src0, i32:$src1, i32:$src2))]>;
+defm V_BFE_I32 : VOP3_32 <0x00000149, "V_BFE_I32",
+  [(set i32:$dst, (AMDGPUbfe_i32 i32:$src0, i32:$src1, i32:$src2))]>;
+}
+
+defm V_BFI_B32 : VOP3_32 <0x0000014a, "V_BFI_B32",
+  [(set i32:$dst, (AMDGPUbfi i32:$src0, i32:$src1, i32:$src2))]>;
+defm V_FMA_F32 : VOP3_32 <0x0000014b, "V_FMA_F32",
   [(set f32:$dst, (fma f32:$src0, f32:$src1, f32:$src2))]
 >;
 def V_FMA_F64 : VOP3_64 <0x0000014c, "V_FMA_F64",
   [(set f64:$dst, (fma f64:$src0, f64:$src1, f64:$src2))]
 >;
 //def V_LERP_U8 : VOP3_U8 <0x0000014d, "V_LERP_U8", []>;
-def V_ALIGNBIT_B32 : VOP3_32 <0x0000014e, "V_ALIGNBIT_B32", []>;
-def : ROTRPattern <V_ALIGNBIT_B32>;
+defm V_ALIGNBIT_B32 : VOP3_32 <0x0000014e, "V_ALIGNBIT_B32", []>;
 
-def V_ALIGNBYTE_B32 : VOP3_32 <0x0000014f, "V_ALIGNBYTE_B32", []>;
-def V_MULLIT_F32 : VOP3_32 <0x00000150, "V_MULLIT_F32", []>;
+defm V_ALIGNBYTE_B32 : VOP3_32 <0x0000014f, "V_ALIGNBYTE_B32", []>;
+defm V_MULLIT_F32 : VOP3_32 <0x00000150, "V_MULLIT_F32", []>;
 ////def V_MIN3_F32 : VOP3_MIN3 <0x00000151, "V_MIN3_F32", []>;
 ////def V_MIN3_I32 : VOP3_MIN3 <0x00000152, "V_MIN3_I32", []>;
 ////def V_MIN3_U32 : VOP3_MIN3 <0x00000153, "V_MIN3_U32", []>;
@@ -1073,18 +1446,22 @@ def V_MULLIT_F32 : VOP3_32 <0x00000150, "V_MULLIT_F32", []>;
 //def V_SAD_U8 : VOP3_U8 <0x0000015a, "V_SAD_U8", []>;
 //def V_SAD_HI_U8 : VOP3_U8 <0x0000015b, "V_SAD_HI_U8", []>;
 //def V_SAD_U16 : VOP3_U16 <0x0000015c, "V_SAD_U16", []>;
-def V_SAD_U32 : VOP3_32 <0x0000015d, "V_SAD_U32", []>;
+defm V_SAD_U32 : VOP3_32 <0x0000015d, "V_SAD_U32", []>;
 ////def V_CVT_PK_U8_F32 : VOP3_U8 <0x0000015e, "V_CVT_PK_U8_F32", []>;
-def V_DIV_FIXUP_F32 : VOP3_32 <0x0000015f, "V_DIV_FIXUP_F32", []>;
-def V_DIV_FIXUP_F64 : VOP3_64 <0x00000160, "V_DIV_FIXUP_F64", []>;
+defm V_DIV_FIXUP_F32 : VOP3_32 <0x0000015f, "V_DIV_FIXUP_F32",
+  [(set f32:$dst, (AMDGPUdiv_fixup f32:$src0, f32:$src1, f32:$src2))]
+>;
+def V_DIV_FIXUP_F64 : VOP3_64 <0x00000160, "V_DIV_FIXUP_F64",
+  [(set f64:$dst, (AMDGPUdiv_fixup f64:$src0, f64:$src1, f64:$src2))]
+>;
 
-def V_LSHL_B64 : VOP3_64_Shift <0x00000161, "V_LSHL_B64",
+def V_LSHL_B64 : VOP3_64_32 <0x00000161, "V_LSHL_B64",
   [(set i64:$dst, (shl i64:$src0, i32:$src1))]
 >;
-def V_LSHR_B64 : VOP3_64_Shift <0x00000162, "V_LSHR_B64",
+def V_LSHR_B64 : VOP3_64_32 <0x00000162, "V_LSHR_B64",
   [(set i64:$dst, (srl i64:$src0, i32:$src1))]
 >;
-def V_ASHR_I64 : VOP3_64_Shift <0x00000163, "V_ASHR_I64",
+def V_ASHR_I64 : VOP3_64_32 <0x00000163, "V_ASHR_I64",
   [(set i64:$dst, (sra i64:$src0, i32:$src1))]
 >;
 
@@ -1097,162 +1474,61 @@ def V_MAX_F64 : VOP3_64 <0x00000167, "V_MAX_F64", []>;
 
 } // isCommutable = 1
 
-def : Pat <
-  (fadd f64:$src0, f64:$src1),
-  (V_ADD_F64 $src0, $src1, (i64 0))
->;
-
-def : Pat <
-  (fmul f64:$src0, f64:$src1),
-  (V_MUL_F64 $src0, $src1, (i64 0))
->;
-
 def V_LDEXP_F64 : VOP3_64 <0x00000168, "V_LDEXP_F64", []>;
 
 let isCommutable = 1 in {
 
-def V_MUL_LO_U32 : VOP3_32 <0x00000169, "V_MUL_LO_U32", []>;
-def V_MUL_HI_U32 : VOP3_32 <0x0000016a, "V_MUL_HI_U32", []>;
-def V_MUL_LO_I32 : VOP3_32 <0x0000016b, "V_MUL_LO_I32", []>;
-def V_MUL_HI_I32 : VOP3_32 <0x0000016c, "V_MUL_HI_I32", []>;
+defm V_MUL_LO_U32 : VOP3_32 <0x00000169, "V_MUL_LO_U32", []>;
+defm V_MUL_HI_U32 : VOP3_32 <0x0000016a, "V_MUL_HI_U32", []>;
+defm V_MUL_LO_I32 : VOP3_32 <0x0000016b, "V_MUL_LO_I32", []>;
+defm V_MUL_HI_I32 : VOP3_32 <0x0000016c, "V_MUL_HI_I32", []>;
 
 } // isCommutable = 1
 
-def : Pat <
-  (mul i32:$src0, i32:$src1),
-  (V_MUL_LO_I32 $src0, $src1, (i32 0))
->;
+def V_DIV_SCALE_F32 : VOP3b_32 <0x0000016d, "V_DIV_SCALE_F32", []>;
 
-def : Pat <
-  (mulhu i32:$src0, i32:$src1),
-  (V_MUL_HI_U32 $src0, $src1, (i32 0))
->;
+// Double precision division pre-scale.
+def V_DIV_SCALE_F64 : VOP3b_64 <0x0000016e, "V_DIV_SCALE_F64", []>;
 
-def : Pat <
-  (mulhs i32:$src0, i32:$src1),
-  (V_MUL_HI_I32 $src0, $src1, (i32 0))
+defm V_DIV_FMAS_F32 : VOP3_32 <0x0000016f, "V_DIV_FMAS_F32",
+  [(set f32:$dst, (AMDGPUdiv_fmas f32:$src0, f32:$src1, f32:$src2))]
+>;
+def V_DIV_FMAS_F64 : VOP3_64 <0x00000170, "V_DIV_FMAS_F64",
+  [(set f64:$dst, (AMDGPUdiv_fmas f64:$src0, f64:$src1, f64:$src2))]
 >;
-
-def V_DIV_SCALE_F32 : VOP3_32 <0x0000016d, "V_DIV_SCALE_F32", []>;
-def V_DIV_SCALE_F64 : VOP3_64 <0x0000016e, "V_DIV_SCALE_F64", []>;
-def V_DIV_FMAS_F32 : VOP3_32 <0x0000016f, "V_DIV_FMAS_F32", []>;
-def V_DIV_FMAS_F64 : VOP3_64 <0x00000170, "V_DIV_FMAS_F64", []>;
 //def V_MSAD_U8 : VOP3_U8 <0x00000171, "V_MSAD_U8", []>;
 //def V_QSAD_U8 : VOP3_U8 <0x00000172, "V_QSAD_U8", []>;
 //def V_MQSAD_U8 : VOP3_U8 <0x00000173, "V_MQSAD_U8", []>;
-def V_TRIG_PREOP_F64 : VOP3_64 <0x00000174, "V_TRIG_PREOP_F64", []>;
-
-let Defs = [SCC] in { // Carry out goes to SCC
-let isCommutable = 1 in {
-def S_ADD_U32 : SOP2_32 <0x00000000, "S_ADD_U32", []>;
-def S_ADD_I32 : SOP2_32 <0x00000002, "S_ADD_I32",
-  [(set i32:$dst, (add SSrc_32:$src0, SSrc_32:$src1))]
+def V_TRIG_PREOP_F64 : VOP3_64_32 <0x00000174, "V_TRIG_PREOP_F64",
+  [(set f64:$dst, (AMDGPUtrig_preop f64:$src0, i32:$src1))]
 >;
-} // End isCommutable = 1
 
-def S_SUB_U32 : SOP2_32 <0x00000001, "S_SUB_U32", []>;
-def S_SUB_I32 : SOP2_32 <0x00000003, "S_SUB_I32",
-  [(set i32:$dst, (sub SSrc_32:$src0, SSrc_32:$src1))]
->;
-
-let Uses = [SCC] in { // Carry in comes from SCC
-let isCommutable = 1 in {
-def S_ADDC_U32 : SOP2_32 <0x00000004, "S_ADDC_U32",
-  [(set i32:$dst, (adde (i32 SSrc_32:$src0), (i32 SSrc_32:$src1)))]>;
-} // End isCommutable = 1
-
-def S_SUBB_U32 : SOP2_32 <0x00000005, "S_SUBB_U32",
-  [(set i32:$dst, (sube (i32 SSrc_32:$src0), (i32 SSrc_32:$src1)))]>;
-} // End Uses = [SCC]
-} // End Defs = [SCC]
+//===----------------------------------------------------------------------===//
+// Pseudo Instructions
+//===----------------------------------------------------------------------===//
 
-def S_MIN_I32 : SOP2_32 <0x00000006, "S_MIN_I32", []>;
-def S_MIN_U32 : SOP2_32 <0x00000007, "S_MIN_U32", []>;
-def S_MAX_I32 : SOP2_32 <0x00000008, "S_MAX_I32", []>;
-def S_MAX_U32 : SOP2_32 <0x00000009, "S_MAX_U32", []>;
+let isCodeGenOnly = 1, isPseudo = 1 in {
 
-def S_CSELECT_B32 : SOP2 <
-  0x0000000a, (outs SReg_32:$dst),
-  (ins SReg_32:$src0, SReg_32:$src1, SCCReg:$scc), "S_CSELECT_B32",
-  []
+def V_MOV_I1 : InstSI <
+  (outs VReg_1:$dst),
+  (ins i1imm:$src),
+  "", [(set i1:$dst, (imm:$src))]
 >;
 
-def S_CSELECT_B64 : SOP2_64 <0x0000000b, "S_CSELECT_B64", []>;
-
-def S_AND_B32 : SOP2_32 <0x0000000e, "S_AND_B32", []>;
-
-def S_AND_B64 : SOP2_64 <0x0000000f, "S_AND_B64",
-  [(set i64:$dst, (and i64:$src0, i64:$src1))]
+def V_AND_I1 : InstSI <
+   (outs VReg_1:$dst), (ins VReg_1:$src0, VReg_1:$src1), "",
+   [(set i1:$dst, (and i1:$src0, i1:$src1))]
 >;
 
-def : Pat <
-  (i1 (and i1:$src0, i1:$src1)),
-  (S_AND_B64 $src0, $src1)
+def V_OR_I1 : InstSI <
+   (outs VReg_1:$dst), (ins VReg_1:$src0, VReg_1:$src1), "",
+   [(set i1:$dst, (or i1:$src0, i1:$src1))]
 >;
 
-def S_OR_B32 : SOP2_32 <0x00000010, "S_OR_B32", []>;
-def S_OR_B64 : SOP2_64 <0x00000011, "S_OR_B64", []>;
-def : Pat <
-  (i1 (or i1:$src0, i1:$src1)),
-  (S_OR_B64 $src0, $src1)
->;
-def S_XOR_B32 : SOP2_32 <0x00000012, "S_XOR_B32", []>;
-def S_XOR_B64 : SOP2_64 <0x00000013, "S_XOR_B64",
+def V_XOR_I1 : InstSI <
+  (outs VReg_1:$dst), (ins VReg_1:$src0, VReg_1:$src1), "",
   [(set i1:$dst, (xor i1:$src0, i1:$src1))]
 >;
-def S_ANDN2_B32 : SOP2_32 <0x00000014, "S_ANDN2_B32", []>;
-def S_ANDN2_B64 : SOP2_64 <0x00000015, "S_ANDN2_B64", []>;
-def S_ORN2_B32 : SOP2_32 <0x00000016, "S_ORN2_B32", []>;
-def S_ORN2_B64 : SOP2_64 <0x00000017, "S_ORN2_B64", []>;
-def S_NAND_B32 : SOP2_32 <0x00000018, "S_NAND_B32", []>;
-def S_NAND_B64 : SOP2_64 <0x00000019, "S_NAND_B64", []>;
-def S_NOR_B32 : SOP2_32 <0x0000001a, "S_NOR_B32", []>;
-def S_NOR_B64 : SOP2_64 <0x0000001b, "S_NOR_B64", []>;
-def S_XNOR_B32 : SOP2_32 <0x0000001c, "S_XNOR_B32", []>;
-def S_XNOR_B64 : SOP2_64 <0x0000001d, "S_XNOR_B64", []>;
-
-// Use added complexity so these patterns are preferred to the VALU patterns.
-let AddedComplexity = 1 in {
-
-def S_LSHL_B32 : SOP2_32 <0x0000001e, "S_LSHL_B32",
-  [(set i32:$dst, (shl i32:$src0, i32:$src1))]
->;
-def S_LSHL_B64 : SOP2_SHIFT_64 <0x0000001f, "S_LSHL_B64",
-  [(set i64:$dst, (shl i64:$src0, i32:$src1))]
->;
-def S_LSHR_B32 : SOP2_32 <0x00000020, "S_LSHR_B32",
-  [(set i32:$dst, (srl i32:$src0, i32:$src1))]
->;
-def S_LSHR_B64 : SOP2_SHIFT_64 <0x00000021, "S_LSHR_B64",
-  [(set i64:$dst, (srl i64:$src0, i32:$src1))]
->;
-def S_ASHR_I32 : SOP2_32 <0x00000022, "S_ASHR_I32",
-  [(set i32:$dst, (sra i32:$src0, i32:$src1))]
->;
-def S_ASHR_I64 : SOP2_SHIFT_64 <0x00000023, "S_ASHR_I64",
-  [(set i64:$dst, (sra i64:$src0, i32:$src1))]
->;
-
-} // End AddedComplexity = 1
-
-def S_BFM_B32 : SOP2_32 <0x00000024, "S_BFM_B32", []>;
-def S_BFM_B64 : SOP2_64 <0x00000025, "S_BFM_B64", []>;
-def S_MUL_I32 : SOP2_32 <0x00000026, "S_MUL_I32", []>;
-def S_BFE_U32 : SOP2_32 <0x00000027, "S_BFE_U32", []>;
-def S_BFE_I32 : SOP2_32 <0x00000028, "S_BFE_I32", []>;
-def S_BFE_U64 : SOP2_64 <0x00000029, "S_BFE_U64", []>;
-def S_BFE_I64 : SOP2_64 <0x0000002a, "S_BFE_I64", []>;
-//def S_CBRANCH_G_FORK : SOP2_ <0x0000002b, "S_CBRANCH_G_FORK", []>;
-def S_ABSDIFF_I32 : SOP2_32 <0x0000002c, "S_ABSDIFF_I32", []>;
-
-let isCodeGenOnly = 1, isPseudo = 1 in {
-
-def LOAD_CONST : AMDGPUShaderInst <
-  (outs GPRF32:$dst),
-  (ins i32imm:$src),
-  "LOAD_CONST $dst, $src",
-  [(set GPRF32:$dst, (int_AMDGPU_load_const imm:$src))]
->;
 
 // SI pseudo instructions. These are used by the CFG structurizer pass
 // and should be lowered to ISA instructions prior to codegen.
@@ -1262,19 +1538,19 @@ let mayLoad = 1, mayStore = 1, hasSideEffects = 1,
 
 let isBranch = 1, isTerminator = 1 in {
 
-def SI_IF : InstSI <
+def SI_IF: InstSI <
   (outs SReg_64:$dst),
   (ins SReg_64:$vcc, brtarget:$target),
-  "SI_IF $dst, $vcc, $target",
+  "",
   [(set i64:$dst, (int_SI_if i1:$vcc, bb:$target))]
 >;
 
 def SI_ELSE : InstSI <
   (outs SReg_64:$dst),
   (ins SReg_64:$src, brtarget:$target),
-  "SI_ELSE $dst, $src, $target",
-  [(set i64:$dst, (int_SI_else i64:$src, bb:$target))]> {
-
+  "",
+  [(set i64:$dst, (int_SI_else i64:$src, bb:$target))]
+> {
   let Constraints = "$src = $dst";
 }
 
@@ -1317,8 +1593,8 @@ def SI_END_CF : InstSI <
 
 def SI_KILL : InstSI <
   (outs),
-  (ins VReg_32:$src),
-  "SI_KIL $src",
+  (ins VSrc_32:$src),
+  "SI_KILL $src",
   [(int_AMDGPU_kill f32:$src)]
 >;
 
@@ -1327,22 +1603,22 @@ def SI_KILL : InstSI <
 
 let Uses = [EXEC], Defs = [EXEC,VCC,M0] in {
 
-//defm SI_ : RegisterLoadStore <VReg_32, FRAMEri64, ADDRIndirect>;
+//defm SI_ : RegisterLoadStore <VReg_32, FRAMEri, ADDRIndirect>;
 
 let UseNamedOperandTable = 1 in {
 
-def SI_RegisterLoad : AMDGPUShaderInst <
+def SI_RegisterLoad : InstSI <
   (outs VReg_32:$dst, SReg_64:$temp),
-  (ins FRAMEri64:$addr, i32imm:$chan),
+  (ins FRAMEri32:$addr, i32imm:$chan),
   "", []
 > {
   let isRegisterLoad = 1;
   let mayLoad = 1;
 }
 
-class SIRegStore<dag outs> : AMDGPUShaderInst <
+class SIRegStore<dag outs> : InstSI <
   outs,
-  (ins VReg_32:$val, FRAMEri64:$addr, i32imm:$chan),
+  (ins VReg_32:$val, FRAMEri32:$addr, i32imm:$chan),
   "", []
 > {
   let isRegisterStore = 1;
@@ -1387,7 +1663,13 @@ let usesCustomInserter = 1 in {
 // constant that can be used with the ADDR64 MUBUF instructions.
 def SI_ADDR64_RSRC : InstSI <
   (outs SReg_128:$srsrc),
-  (ins SReg_64:$ptr),
+  (ins SSrc_64:$ptr),
+  "", []
+>;
+
+def SI_BUFFER_RSRC : InstSI <
+  (outs SReg_128:$srsrc),
+  (ins SReg_32:$ptr_lo, SReg_32:$ptr_hi, SSrc_32:$data_lo, SSrc_32:$data_hi),
   "", []
 >;
 
@@ -1395,13 +1677,49 @@ def V_SUB_F64 : InstSI <
   (outs VReg_64:$dst),
   (ins VReg_64:$src0, VReg_64:$src1),
   "V_SUB_F64 $dst, $src0, $src1",
-  []
+  [(set f64:$dst, (fsub f64:$src0, f64:$src1))]
 >;
 
 } // end usesCustomInserter
 
+multiclass SI_SPILL_SGPR <RegisterClass sgpr_class> {
+
+  def _SAVE : InstSI <
+    (outs VReg_32:$dst),
+    (ins sgpr_class:$src, i32imm:$frame_idx),
+    "", []
+  >;
+
+  def _RESTORE : InstSI <
+    (outs sgpr_class:$dst),
+    (ins VReg_32:$src, i32imm:$frame_idx),
+    "", []
+  >;
+
+}
+
+defm SI_SPILL_S32  : SI_SPILL_SGPR <SReg_32>;
+defm SI_SPILL_S64  : SI_SPILL_SGPR <SReg_64>;
+defm SI_SPILL_S128 : SI_SPILL_SGPR <SReg_128>;
+defm SI_SPILL_S256 : SI_SPILL_SGPR <SReg_256>;
+defm SI_SPILL_S512 : SI_SPILL_SGPR <SReg_512>;
+
+let Defs = [SCC] in {
+
+def SI_CONSTDATA_PTR : InstSI <
+  (outs SReg_64:$dst),
+  (ins),
+  "", [(set SReg_64:$dst, (i64 SIconstdata_ptr))]
+>;
+
+} // End Defs = [SCC]
+
 } // end IsCodeGenOnly, isPseudo
 
+} // end SubtargetPredicate = SI
+
+let Predicates = [isSI] in {
+
 def : Pat<
   (int_AMDGPU_cndlt f32:$src0, f32:$src1, f32:$src2),
   (V_CNDMASK_B32_e64 $src2, $src1, (V_CMP_GT_F32_e64 0, $src0))
@@ -1409,12 +1727,12 @@ def : Pat<
 
 def : Pat <
   (int_AMDGPU_kilp),
-  (SI_KILL (V_MOV_B32_e32 0xbf800000))
+  (SI_KILL 0xbf800000)
 >;
 
 /* int_SI_vs_load_input */
 def : Pat<
-  (SIload_input i128:$tlst, IMM12bit:$attr_offset, i32:$buf_idx_vgpr),
+  (SIload_input v4i32:$tlst, imm:$attr_offset, i32:$buf_idx_vgpr),
   (BUFFER_LOAD_FORMAT_XYZW_IDXEN $tlst, $buf_idx_vgpr, imm:$attr_offset, 0, 0, 0, 0)
 >;
 
@@ -1426,45 +1744,312 @@ def : Pat <
        $src0, $src1, $src2, $src3)
 >;
 
+//===----------------------------------------------------------------------===//
+// SMRD Patterns
+//===----------------------------------------------------------------------===//
+
+multiclass SMRD_Pattern <SMRD Instr_IMM, SMRD Instr_SGPR, ValueType vt> {
+
+  // 1. Offset as 8bit DWORD immediate
+  def : Pat <
+    (constant_load (add i64:$sbase, (i64 IMM8bitDWORD:$offset))),
+    (vt (Instr_IMM $sbase, (as_dword_i32imm $offset)))
+  >;
+
+  // 2. Offset loaded in an 32bit SGPR
+  def : Pat <
+    (constant_load (add i64:$sbase, (i64 IMM32bit:$offset))),
+    (vt (Instr_SGPR $sbase, (S_MOV_B32 (i32 (as_i32imm $offset)))))
+  >;
+
+  // 3. No offset at all
+  def : Pat <
+    (constant_load i64:$sbase),
+    (vt (Instr_IMM $sbase, 0))
+  >;
+}
+
+defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, f32>;
+defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, i32>;
+defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, v2i32>;
+defm : SMRD_Pattern <S_LOAD_DWORDX4_IMM, S_LOAD_DWORDX4_SGPR, v4i32>;
+defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v32i8>;
+defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v8i32>;
+defm : SMRD_Pattern <S_LOAD_DWORDX16_IMM, S_LOAD_DWORDX16_SGPR, v16i32>;
+
+// 1. Offset as 8bit DWORD immediate
 def : Pat <
-  (f64 (fsub f64:$src0, f64:$src1)),
-  (V_SUB_F64 $src0, $src1)
+  (SIload_constant v4i32:$sbase, IMM8bitDWORD:$offset),
+  (S_BUFFER_LOAD_DWORD_IMM $sbase, (as_dword_i32imm $offset))
+>;
+
+// 2. Offset loaded in an 32bit SGPR
+def : Pat <
+  (SIload_constant v4i32:$sbase, imm:$offset),
+  (S_BUFFER_LOAD_DWORD_SGPR $sbase, (S_MOV_B32 imm:$offset))
+>;
+
+} // Predicates = [isSI] in {
+
+//===----------------------------------------------------------------------===//
+// SOP1 Patterns
+//===----------------------------------------------------------------------===//
+
+let Predicates = [isSI, isCFDepth0] in {
+
+def : Pat <
+  (i64 (ctpop i64:$src)),
+  (INSERT_SUBREG (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
+    (S_BCNT1_I32_B64 $src), sub0),
+    (S_MOV_B32 0), sub1)
+>;
+
+//===----------------------------------------------------------------------===//
+// SOP2 Patterns
+//===----------------------------------------------------------------------===//
+
+// V_ADD_I32_e32/S_ADD_I32 produces carry in VCC/SCC. For the vector
+// case, the sgpr-copies pass will fix this to use the vector version.
+def : Pat <
+  (i32 (addc i32:$src0, i32:$src1)),
+  (S_ADD_I32 $src0, $src1)
+>;
+
+} // Predicates = [isSI, isCFDepth0]
+
+let  Predicates = [isSI] in {
+
+//===----------------------------------------------------------------------===//
+// SOPP Patterns
+//===----------------------------------------------------------------------===//
+
+def : Pat <
+  (int_AMDGPU_barrier_global),
+  (S_BARRIER)
+>;
+
+//===----------------------------------------------------------------------===//
+// VOP1 Patterns
+//===----------------------------------------------------------------------===//
+
+let Predicates = [UnsafeFPMath] in {
+def : RcpPat<V_RCP_F64_e32, f64>;
+defm : RsqPat<V_RSQ_F64_e32, f64>;
+defm : RsqPat<V_RSQ_F32_e32, f32>;
+}
+
+//===----------------------------------------------------------------------===//
+// VOP2 Patterns
+//===----------------------------------------------------------------------===//
+
+class BinOp64Pat <SDNode node, Instruction inst> : Pat <
+  (node i64:$src0, i64:$src1),
+  (INSERT_SUBREG (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
+    (inst  (EXTRACT_SUBREG i64:$src0, sub0),
+                  (EXTRACT_SUBREG i64:$src1, sub0)), sub0),
+    (inst (EXTRACT_SUBREG i64:$src0, sub1),
+                  (EXTRACT_SUBREG i64:$src1, sub1)), sub1)
+>;
+
+def : BinOp64Pat <or, V_OR_B32_e32>;
+def : BinOp64Pat <xor, V_XOR_B32_e32>;
+
+class SextInReg <ValueType vt, int ShiftAmt> : Pat <
+  (sext_inreg i32:$src0, vt),
+  (V_ASHRREV_I32_e32 ShiftAmt, (V_LSHLREV_B32_e32 ShiftAmt, $src0))
+>;
+
+def : SextInReg <i8, 24>;
+def : SextInReg <i16, 16>;
+
+def : Pat <
+  (i32 (add (i32 (ctpop i32:$popcnt)), i32:$val)),
+  (V_BCNT_U32_B32_e32 $popcnt, $val)
+>;
+
+def : Pat <
+   (i32 (ctpop i32:$popcnt)),
+   (V_BCNT_U32_B32_e64 $popcnt, 0, 0, 0)
+>;
+
+def : Pat <
+  (i64 (ctpop i64:$src)),
+  (INSERT_SUBREG
+    (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
+      (V_BCNT_U32_B32_e32 (EXTRACT_SUBREG $src, sub1),
+        (V_BCNT_U32_B32_e64 (EXTRACT_SUBREG $src, sub0), 0, 0, 0)),
+      sub0),
+    (V_MOV_B32_e32 0), sub1)
+>;
+
+def : Pat <
+  (addc i32:$src0, i32:$src1),
+  (V_ADD_I32_e32 $src0, $src1)
 >;
 
 /********** ======================= **********/
 /********** Image sampling patterns **********/
 /********** ======================= **********/
 
+// Image + sampler
+class SampleRawPattern<SDPatternOperator name, MIMG opcode, ValueType vt> : Pat <
+  (name vt:$addr, v8i32:$rsrc, v4i32:$sampler, i32:$dmask, i32:$unorm,
+        i32:$r128, i32:$da, i32:$glc, i32:$slc, i32:$tfe, i32:$lwe),
+  (opcode (as_i32imm $dmask), (as_i1imm $unorm), (as_i1imm $glc), (as_i1imm $da),
+          (as_i1imm $r128), (as_i1imm $tfe), (as_i1imm $lwe), (as_i1imm $slc),
+          $addr, $rsrc, $sampler)
+>;
+
+multiclass SampleRawPatterns<SDPatternOperator name, string opcode> {
+  def : SampleRawPattern<name, !cast<MIMG>(opcode # _V4_V1), i32>;
+  def : SampleRawPattern<name, !cast<MIMG>(opcode # _V4_V2), v2i32>;
+  def : SampleRawPattern<name, !cast<MIMG>(opcode # _V4_V4), v4i32>;
+  def : SampleRawPattern<name, !cast<MIMG>(opcode # _V4_V8), v8i32>;
+  def : SampleRawPattern<name, !cast<MIMG>(opcode # _V4_V16), v16i32>;
+}
+
+// Image only
+class ImagePattern<SDPatternOperator name, MIMG opcode, ValueType vt> : Pat <
+  (name vt:$addr, v8i32:$rsrc, i32:$dmask, i32:$unorm,
+        i32:$r128, i32:$da, i32:$glc, i32:$slc, i32:$tfe, i32:$lwe),
+  (opcode (as_i32imm $dmask), (as_i1imm $unorm), (as_i1imm $glc), (as_i1imm $da),
+          (as_i1imm $r128), (as_i1imm $tfe), (as_i1imm $lwe), (as_i1imm $slc),
+          $addr, $rsrc)
+>;
+
+multiclass ImagePatterns<SDPatternOperator name, string opcode> {
+  def : ImagePattern<name, !cast<MIMG>(opcode # _V4_V1), i32>;
+  def : ImagePattern<name, !cast<MIMG>(opcode # _V4_V2), v2i32>;
+  def : ImagePattern<name, !cast<MIMG>(opcode # _V4_V4), v4i32>;
+}
+
+// Basic sample
+defm : SampleRawPatterns<int_SI_image_sample,           "IMAGE_SAMPLE">;
+defm : SampleRawPatterns<int_SI_image_sample_cl,        "IMAGE_SAMPLE_CL">;
+defm : SampleRawPatterns<int_SI_image_sample_d,         "IMAGE_SAMPLE_D">;
+defm : SampleRawPatterns<int_SI_image_sample_d_cl,      "IMAGE_SAMPLE_D_CL">;
+defm : SampleRawPatterns<int_SI_image_sample_l,         "IMAGE_SAMPLE_L">;
+defm : SampleRawPatterns<int_SI_image_sample_b,         "IMAGE_SAMPLE_B">;
+defm : SampleRawPatterns<int_SI_image_sample_b_cl,      "IMAGE_SAMPLE_B_CL">;
+defm : SampleRawPatterns<int_SI_image_sample_lz,        "IMAGE_SAMPLE_LZ">;
+defm : SampleRawPatterns<int_SI_image_sample_cd,        "IMAGE_SAMPLE_CD">;
+defm : SampleRawPatterns<int_SI_image_sample_cd_cl,     "IMAGE_SAMPLE_CD_CL">;
+
+// Sample with comparison
+defm : SampleRawPatterns<int_SI_image_sample_c,         "IMAGE_SAMPLE_C">;
+defm : SampleRawPatterns<int_SI_image_sample_c_cl,      "IMAGE_SAMPLE_C_CL">;
+defm : SampleRawPatterns<int_SI_image_sample_c_d,       "IMAGE_SAMPLE_C_D">;
+defm : SampleRawPatterns<int_SI_image_sample_c_d_cl,    "IMAGE_SAMPLE_C_D_CL">;
+defm : SampleRawPatterns<int_SI_image_sample_c_l,       "IMAGE_SAMPLE_C_L">;
+defm : SampleRawPatterns<int_SI_image_sample_c_b,       "IMAGE_SAMPLE_C_B">;
+defm : SampleRawPatterns<int_SI_image_sample_c_b_cl,    "IMAGE_SAMPLE_C_B_CL">;
+defm : SampleRawPatterns<int_SI_image_sample_c_lz,      "IMAGE_SAMPLE_C_LZ">;
+defm : SampleRawPatterns<int_SI_image_sample_c_cd,      "IMAGE_SAMPLE_C_CD">;
+defm : SampleRawPatterns<int_SI_image_sample_c_cd_cl,   "IMAGE_SAMPLE_C_CD_CL">;
+
+// Sample with offsets
+defm : SampleRawPatterns<int_SI_image_sample_o,         "IMAGE_SAMPLE_O">;
+defm : SampleRawPatterns<int_SI_image_sample_cl_o,      "IMAGE_SAMPLE_CL_O">;
+defm : SampleRawPatterns<int_SI_image_sample_d_o,       "IMAGE_SAMPLE_D_O">;
+defm : SampleRawPatterns<int_SI_image_sample_d_cl_o,    "IMAGE_SAMPLE_D_CL_O">;
+defm : SampleRawPatterns<int_SI_image_sample_l_o,       "IMAGE_SAMPLE_L_O">;
+defm : SampleRawPatterns<int_SI_image_sample_b_o,       "IMAGE_SAMPLE_B_O">;
+defm : SampleRawPatterns<int_SI_image_sample_b_cl_o,    "IMAGE_SAMPLE_B_CL_O">;
+defm : SampleRawPatterns<int_SI_image_sample_lz_o,      "IMAGE_SAMPLE_LZ_O">;
+defm : SampleRawPatterns<int_SI_image_sample_cd_o,      "IMAGE_SAMPLE_CD_O">;
+defm : SampleRawPatterns<int_SI_image_sample_cd_cl_o,   "IMAGE_SAMPLE_CD_CL_O">;
+
+// Sample with comparison and offsets
+defm : SampleRawPatterns<int_SI_image_sample_c_o,       "IMAGE_SAMPLE_C_O">;
+defm : SampleRawPatterns<int_SI_image_sample_c_cl_o,    "IMAGE_SAMPLE_C_CL_O">;
+defm : SampleRawPatterns<int_SI_image_sample_c_d_o,     "IMAGE_SAMPLE_C_D_O">;
+defm : SampleRawPatterns<int_SI_image_sample_c_d_cl_o,  "IMAGE_SAMPLE_C_D_CL_O">;
+defm : SampleRawPatterns<int_SI_image_sample_c_l_o,     "IMAGE_SAMPLE_C_L_O">;
+defm : SampleRawPatterns<int_SI_image_sample_c_b_o,     "IMAGE_SAMPLE_C_B_O">;
+defm : SampleRawPatterns<int_SI_image_sample_c_b_cl_o,  "IMAGE_SAMPLE_C_B_CL_O">;
+defm : SampleRawPatterns<int_SI_image_sample_c_lz_o,    "IMAGE_SAMPLE_C_LZ_O">;
+defm : SampleRawPatterns<int_SI_image_sample_c_cd_o,    "IMAGE_SAMPLE_C_CD_O">;
+defm : SampleRawPatterns<int_SI_image_sample_c_cd_cl_o, "IMAGE_SAMPLE_C_CD_CL_O">;
+
+// Gather opcodes
+// Only the variants which make sense are defined.
+def : SampleRawPattern<int_SI_gather4,           IMAGE_GATHER4_V4_V2,        v2i32>;
+def : SampleRawPattern<int_SI_gather4,           IMAGE_GATHER4_V4_V4,        v4i32>;
+def : SampleRawPattern<int_SI_gather4_cl,        IMAGE_GATHER4_CL_V4_V4,     v4i32>;
+def : SampleRawPattern<int_SI_gather4_l,         IMAGE_GATHER4_L_V4_V4,      v4i32>;
+def : SampleRawPattern<int_SI_gather4_b,         IMAGE_GATHER4_B_V4_V4,      v4i32>;
+def : SampleRawPattern<int_SI_gather4_b_cl,      IMAGE_GATHER4_B_CL_V4_V4,   v4i32>;
+def : SampleRawPattern<int_SI_gather4_b_cl,      IMAGE_GATHER4_B_CL_V4_V8,   v8i32>;
+def : SampleRawPattern<int_SI_gather4_lz,        IMAGE_GATHER4_LZ_V4_V2,     v2i32>;
+def : SampleRawPattern<int_SI_gather4_lz,        IMAGE_GATHER4_LZ_V4_V4,     v4i32>;
+
+def : SampleRawPattern<int_SI_gather4_c,         IMAGE_GATHER4_C_V4_V4,      v4i32>;
+def : SampleRawPattern<int_SI_gather4_c_cl,      IMAGE_GATHER4_C_CL_V4_V4,   v4i32>;
+def : SampleRawPattern<int_SI_gather4_c_cl,      IMAGE_GATHER4_C_CL_V4_V8,   v8i32>;
+def : SampleRawPattern<int_SI_gather4_c_l,       IMAGE_GATHER4_C_L_V4_V4,    v4i32>;
+def : SampleRawPattern<int_SI_gather4_c_l,       IMAGE_GATHER4_C_L_V4_V8,    v8i32>;
+def : SampleRawPattern<int_SI_gather4_c_b,       IMAGE_GATHER4_C_B_V4_V4,    v4i32>;
+def : SampleRawPattern<int_SI_gather4_c_b,       IMAGE_GATHER4_C_B_V4_V8,    v8i32>;
+def : SampleRawPattern<int_SI_gather4_c_b_cl,    IMAGE_GATHER4_C_B_CL_V4_V8, v8i32>;
+def : SampleRawPattern<int_SI_gather4_c_lz,      IMAGE_GATHER4_C_LZ_V4_V4,   v4i32>;
+
+def : SampleRawPattern<int_SI_gather4_o,         IMAGE_GATHER4_O_V4_V4,      v4i32>;
+def : SampleRawPattern<int_SI_gather4_cl_o,      IMAGE_GATHER4_CL_O_V4_V4,   v4i32>;
+def : SampleRawPattern<int_SI_gather4_cl_o,      IMAGE_GATHER4_CL_O_V4_V8,   v8i32>;
+def : SampleRawPattern<int_SI_gather4_l_o,       IMAGE_GATHER4_L_O_V4_V4,    v4i32>;
+def : SampleRawPattern<int_SI_gather4_l_o,       IMAGE_GATHER4_L_O_V4_V8,    v8i32>;
+def : SampleRawPattern<int_SI_gather4_b_o,       IMAGE_GATHER4_B_O_V4_V4,    v4i32>;
+def : SampleRawPattern<int_SI_gather4_b_o,       IMAGE_GATHER4_B_O_V4_V8,    v8i32>;
+def : SampleRawPattern<int_SI_gather4_b_cl_o,    IMAGE_GATHER4_B_CL_O_V4_V8, v8i32>;
+def : SampleRawPattern<int_SI_gather4_lz_o,      IMAGE_GATHER4_LZ_O_V4_V4,   v4i32>;
+
+def : SampleRawPattern<int_SI_gather4_c_o,       IMAGE_GATHER4_C_O_V4_V4,    v4i32>;
+def : SampleRawPattern<int_SI_gather4_c_o,       IMAGE_GATHER4_C_O_V4_V8,    v8i32>;
+def : SampleRawPattern<int_SI_gather4_c_cl_o,    IMAGE_GATHER4_C_CL_O_V4_V8, v8i32>;
+def : SampleRawPattern<int_SI_gather4_c_l_o,     IMAGE_GATHER4_C_L_O_V4_V8,  v8i32>;
+def : SampleRawPattern<int_SI_gather4_c_b_o,     IMAGE_GATHER4_C_B_O_V4_V8,  v8i32>;
+def : SampleRawPattern<int_SI_gather4_c_b_cl_o,  IMAGE_GATHER4_C_B_CL_O_V4_V8, v8i32>;
+def : SampleRawPattern<int_SI_gather4_c_lz_o,    IMAGE_GATHER4_C_LZ_O_V4_V4, v4i32>;
+def : SampleRawPattern<int_SI_gather4_c_lz_o,    IMAGE_GATHER4_C_LZ_O_V4_V8, v8i32>;
+
+def : SampleRawPattern<int_SI_getlod, IMAGE_GET_LOD_V4_V1, i32>;
+def : SampleRawPattern<int_SI_getlod, IMAGE_GET_LOD_V4_V2, v2i32>;
+def : SampleRawPattern<int_SI_getlod, IMAGE_GET_LOD_V4_V4, v4i32>;
+
+def : ImagePattern<int_SI_getresinfo, IMAGE_GET_RESINFO_V4_V1, i32>;
+defm : ImagePatterns<int_SI_image_load, "IMAGE_LOAD">;
+defm : ImagePatterns<int_SI_image_load_mip, "IMAGE_LOAD_MIP">;
+
 /* SIsample for simple 1D texture lookup */
 def : Pat <
-  (SIsample i32:$addr, v32i8:$rsrc, i128:$sampler, imm),
+  (SIsample i32:$addr, v32i8:$rsrc, v4i32:$sampler, imm),
   (IMAGE_SAMPLE_V4_V1 0xf, 0, 0, 0, 0, 0, 0, 0, $addr, $rsrc, $sampler)
 >;
 
 class SamplePattern<SDNode name, MIMG opcode, ValueType vt> : Pat <
-    (name vt:$addr, v32i8:$rsrc, i128:$sampler, imm),
+    (name vt:$addr, v32i8:$rsrc, v4i32:$sampler, imm),
     (opcode 0xf, 0, 0, 0, 0, 0, 0, 0, $addr, $rsrc, $sampler)
 >;
 
 class SampleRectPattern<SDNode name, MIMG opcode, ValueType vt> : Pat <
-    (name vt:$addr, v32i8:$rsrc, i128:$sampler, TEX_RECT),
+    (name vt:$addr, v32i8:$rsrc, v4i32:$sampler, TEX_RECT),
     (opcode 0xf, 1, 0, 0, 0, 0, 0, 0, $addr, $rsrc, $sampler)
 >;
 
 class SampleArrayPattern<SDNode name, MIMG opcode, ValueType vt> : Pat <
-    (name vt:$addr, v32i8:$rsrc, i128:$sampler, TEX_ARRAY),
+    (name vt:$addr, v32i8:$rsrc, v4i32:$sampler, TEX_ARRAY),
     (opcode 0xf, 0, 0, 1, 0, 0, 0, 0, $addr, $rsrc, $sampler)
 >;
 
 class SampleShadowPattern<SDNode name, MIMG opcode,
                           ValueType vt> : Pat <
-    (name vt:$addr, v32i8:$rsrc, i128:$sampler, TEX_SHADOW),
+    (name vt:$addr, v32i8:$rsrc, v4i32:$sampler, TEX_SHADOW),
     (opcode 0xf, 0, 0, 0, 0, 0, 0, 0, $addr, $rsrc, $sampler)
 >;
 
 class SampleShadowArrayPattern<SDNode name, MIMG opcode,
                                ValueType vt> : Pat <
-    (name vt:$addr, v32i8:$rsrc, i128:$sampler, TEX_SHADOW_ARRAY),
+    (name vt:$addr, v32i8:$rsrc, v4i32:$sampler, TEX_SHADOW_ARRAY),
     (opcode 0xf, 0, 0, 1, 0, 0, 0, 0, $addr, $rsrc, $sampler)
 >;
 
@@ -1649,25 +2234,42 @@ def : BitConvert <f64, i64, VReg_64>;
 def : BitConvert <v2f32, v2i32, VReg_64>;
 def : BitConvert <v2i32, v2f32, VReg_64>;
 def : BitConvert <v2i32, i64, VReg_64>;
-
+def : BitConvert <i64, v2i32, VReg_64>;
+def : BitConvert <v2f32, i64, VReg_64>;
+def : BitConvert <i64, v2f32, VReg_64>;
+def : BitConvert <v2i32, f64, VReg_64>;
+def : BitConvert <f64, v2i32, VReg_64>;
 def : BitConvert <v4f32, v4i32, VReg_128>;
 def : BitConvert <v4i32, v4f32, VReg_128>;
-def : BitConvert <v4i32, i128,  VReg_128>;
-def : BitConvert <i128, v4i32,  VReg_128>;
 
+def : BitConvert <v8f32, v8i32, SReg_256>;
+def : BitConvert <v8i32, v8f32, SReg_256>;
 def : BitConvert <v8i32, v32i8, SReg_256>;
 def : BitConvert <v32i8, v8i32, SReg_256>;
 def : BitConvert <v8i32, v32i8, VReg_256>;
+def : BitConvert <v8i32, v8f32, VReg_256>;
+def : BitConvert <v8f32, v8i32, VReg_256>;
 def : BitConvert <v32i8, v8i32, VReg_256>;
 
+def : BitConvert <v16i32, v16f32, VReg_512>;
+def : BitConvert <v16f32, v16i32, VReg_512>;
+
 /********** =================== **********/
 /********** Src & Dst modifiers **********/
 /********** =================== **********/
 
+def FCLAMP_SI : AMDGPUShaderInst <
+  (outs VReg_32:$dst),
+  (ins VSrc_32:$src0),
+  "FCLAMP_SI $dst, $src0",
+  []
+> {
+  let usesCustomInserter = 1;
+}
+
 def : Pat <
-  (int_AMDIL_clamp f32:$src, (f32 FP_ZERO), (f32 FP_ONE)),
-  (V_ADD_F32_e64 $src, (i32 0 /* SRC1 */),
-   0 /* ABS */, 1 /* CLAMP */, 0 /* OMOD */, 0 /* NEG */)
+  (AMDGPUclamp f32:$src, (f32 FP_ZERO), (f32 FP_ONE)),
+  (FCLAMP_SI f32:$src)
 >;
 
 /********** ================================ **********/
@@ -1686,14 +2288,32 @@ def : Pat <
   (V_OR_B32_e32 $src, (V_MOV_B32_e32 0x80000000)) /* Set sign bit */
 >;
 
+def FABS_SI : AMDGPUShaderInst <
+  (outs VReg_32:$dst),
+  (ins VSrc_32:$src0),
+  "FABS_SI $dst, $src0",
+  []
+> {
+  let usesCustomInserter = 1;
+}
+
 def : Pat <
   (fabs f32:$src),
-  (V_AND_B32_e32 $src, (V_MOV_B32_e32 0x7fffffff)) /* Clear sign bit */
+  (FABS_SI f32:$src)
 >;
 
+def FNEG_SI : AMDGPUShaderInst <
+  (outs VReg_32:$dst),
+  (ins VSrc_32:$src0),
+  "FNEG_SI $dst, $src0",
+  []
+> {
+  let usesCustomInserter = 1;
+}
+
 def : Pat <
   (fneg f32:$src),
-  (V_XOR_B32_e32 $src, (V_MOV_B32_e32 0x80000000)) /* Toggle sign bit */
+  (FNEG_SI f32:$src)
 >;
 
 /********** ================== **********/
@@ -1721,30 +2341,10 @@ def : Pat <
 >;
 
 def : Pat <
-  (i1 imm:$imm),
-  (S_MOV_B64 imm:$imm)
->;
-
-def : Pat <
   (i64 InlineImm<i64>:$imm),
   (S_MOV_B64 InlineImm<i64>:$imm)
 >;
 
-// i64 immediates aren't supported in hardware, split it into two 32bit values
-def : Pat <
-  (i64 imm:$imm),
-  (INSERT_SUBREG (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
-    (S_MOV_B32 (i32 (LO32 imm:$imm))), sub0),
-    (S_MOV_B32 (i32 (HI32 imm:$imm))), sub1)
->;
-
-def : Pat <
-  (f64 fpimm:$imm),
-  (INSERT_SUBREG (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
-    (V_MOV_B32_e32 (f32 (LO32f fpimm:$imm))), sub0),
-    (V_MOV_B32_e32 (f32 (HI32f fpimm:$imm))), sub1)
->;
-
 /********** ===================== **********/
 /********** Interpolation Paterns **********/
 /********** ===================== **********/
@@ -1775,26 +2375,11 @@ def : Pat <
 >;
 
 def : Pat<
-  (fdiv f32:$src0, f32:$src1),
-  (V_MUL_F32_e32 $src0, (V_RCP_F32_e32 $src1))
->;
-
-def : Pat<
   (fdiv f64:$src0, f64:$src1),
   (V_MUL_F64 $src0, (V_RCP_F64_e32 $src1), (i64 0))
 >;
 
 def : Pat <
-  (fcos f32:$src0),
-  (V_COS_F32_e32 (V_MUL_F32_e32 $src0, (V_MOV_B32_e32 CONST.TWO_PI_INV)))
->;
-
-def : Pat <
-  (fsin f32:$src0),
-  (V_SIN_F32_e32 (V_MUL_F32_e32 $src0, (V_MOV_B32_e32 CONST.TWO_PI_INV)))
->;
-
-def : Pat <
   (int_AMDGPU_cube v4f32:$src),
   (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
     (V_CUBETC_F32 (EXTRACT_SUBREG $src, sub0),
@@ -1820,27 +2405,18 @@ def : Pat <
   (V_CNDMASK_B32_e64 (i32 0), (i32 -1), $src0)
 >;
 
-def : Pat <
-  (i32 (zext i1:$src0)),
+class Ext32Pat <SDNode ext> : Pat <
+  (i32 (ext i1:$src0)),
   (V_CNDMASK_B32_e64 (i32 0), (i32 1), $src0)
 >;
 
-// 1. Offset as 8bit DWORD immediate
-def : Pat <
-  (SIload_constant i128:$sbase, IMM8bitDWORD:$offset),
-  (S_BUFFER_LOAD_DWORD_IMM $sbase, IMM8bitDWORD:$offset)
->;
-
-// 2. Offset loaded in an 32bit SGPR
-def : Pat <
-  (SIload_constant i128:$sbase, imm:$offset),
-  (S_BUFFER_LOAD_DWORD_SGPR $sbase, (S_MOV_B32 imm:$offset))
->;
+def : Ext32Pat <zext>;
+def : Ext32Pat <anyext>;
 
-// 3. Offset in an 32Bit VGPR
+// Offset in an 32Bit VGPR
 def : Pat <
-  (SIload_constant i128:$sbase, i32:$voff),
-  (BUFFER_LOAD_DWORD_OFFEN $sbase, $voff, 0, 0, 0, 0)
+  (SIload_constant v4i32:$sbase, i32:$voff),
+  (BUFFER_LOAD_DWORD_OFFEN $sbase, $voff, 0, 0, 0, 0, 0)
 >;
 
 // The multiplication scales from [0,1] to the unsigned integer range
@@ -1854,175 +2430,228 @@ def : Pat <
 def : Pat <
   (int_SI_tid),
   (V_MBCNT_HI_U32_B32_e32 0xffffffff,
-                          (V_MBCNT_LO_U32_B32_e64 0xffffffff, 0, 0, 0, 0, 0))
+                          (V_MBCNT_LO_U32_B32_e64 0xffffffff, 0, 0, 0))
 >;
 
-/********** ================== **********/
-/**********   VOP3 Patterns    **********/
-/********** ================== **********/
+//===----------------------------------------------------------------------===//
+// VOP3 Patterns
+//===----------------------------------------------------------------------===//
+
+def : IMad24Pat<V_MAD_I32_I24>;
+def : UMad24Pat<V_MAD_U32_U24>;
 
 def : Pat <
-  (f32 (fadd (fmul f32:$src0, f32:$src1), f32:$src2)),
-  (V_MAD_F32 $src0, $src1, $src2)
+  (fadd f64:$src0, f64:$src1),
+  (V_ADD_F64 $src0, $src1, (i64 0))
 >;
 
-/********** ======================= **********/
-/**********   Load/Store Patterns   **********/
-/********** ======================= **********/
-
-class DSReadPat <DS inst, ValueType vt, PatFrag frag> : Pat <
-  (frag i32:$src0),
-  (vt (inst 0, $src0, $src0, $src0, 0, 0))
+def : Pat <
+  (fmul f64:$src0, f64:$src1),
+  (V_MUL_F64 $src0, $src1, (i64 0))
 >;
 
-def : DSReadPat <DS_READ_I8,  i32, sextloadi8_local>;
-def : DSReadPat <DS_READ_U8,  i32, az_extloadi8_local>;
-def : DSReadPat <DS_READ_I16, i32, sextloadi16_local>;
-def : DSReadPat <DS_READ_U16, i32, az_extloadi16_local>;
-def : DSReadPat <DS_READ_B32, i32, local_load>;
 def : Pat <
-    (local_load i32:$src0),
-    (i32 (DS_READ_B32 0, $src0, $src0, $src0, 0, 0))
+  (mul i32:$src0, i32:$src1),
+  (V_MUL_LO_I32 $src0, $src1, (i32 0))
 >;
 
-class DSWritePat <DS inst, ValueType vt, PatFrag frag> : Pat <
-  (frag i32:$src1, i32:$src0),
-  (inst 0, $src0, $src1, $src1, 0, 0)
+def : Pat <
+  (mulhu i32:$src0, i32:$src1),
+  (V_MUL_HI_U32 $src0, $src1, (i32 0))
 >;
 
-def : DSWritePat <DS_WRITE_B8, i32, truncstorei8_local>;
-def : DSWritePat <DS_WRITE_B16, i32, truncstorei16_local>;
-def : DSWritePat <DS_WRITE_B32, i32, local_store>;
-
-def : Pat <(atomic_load_add_local i32:$ptr, i32:$val),
-           (DS_ADD_U32_RTN 0, $ptr, $val, 0, 0)>;
+def : Pat <
+  (mulhs i32:$src0, i32:$src1),
+  (V_MUL_HI_I32 $src0, $src1, (i32 0))
+>;
 
-def : Pat <(atomic_load_sub_local i32:$ptr, i32:$val),
-           (DS_SUB_U32_RTN 0, $ptr, $val, 0, 0)>;
+defm : BFIPatterns <V_BFI_B32, S_MOV_B32>;
+def : ROTRPattern <V_ALIGNBIT_B32>;
 
-/********** ================== **********/
-/**********   SMRD Patterns    **********/
-/********** ================== **********/
+/********** ======================= **********/
+/**********   Load/Store Patterns   **********/
+/********** ======================= **********/
 
-multiclass SMRD_Pattern <SMRD Instr_IMM, SMRD Instr_SGPR, ValueType vt> {
+multiclass DSReadPat <DS inst, ValueType vt, PatFrag frag> {
+  def : Pat <
+    (vt (frag (add i32:$ptr, (i32 IMM16bit:$offset)))),
+    (inst (i1 0), $ptr, (as_i16imm $offset))
+  >;
 
-  // 1. Offset as 8bit DWORD immediate
   def : Pat <
-    (constant_load (SIadd64bit32bit i64:$sbase, IMM8bitDWORD:$offset)),
-    (vt (Instr_IMM $sbase, IMM8bitDWORD:$offset))
+    (frag i32:$src0),
+    (vt (inst 0, $src0, 0))
   >;
+}
 
-  // 2. Offset loaded in an 32bit SGPR
+defm : DSReadPat <DS_READ_I8,  i32, sextloadi8_local>;
+defm : DSReadPat <DS_READ_U8,  i32, az_extloadi8_local>;
+defm : DSReadPat <DS_READ_I16, i32, sextloadi16_local>;
+defm : DSReadPat <DS_READ_U16, i32, az_extloadi16_local>;
+defm : DSReadPat <DS_READ_B32, i32, local_load>;
+defm : DSReadPat <DS_READ_B64, v2i32, local_load>;
+
+multiclass DSWritePat <DS inst, ValueType vt, PatFrag frag> {
   def : Pat <
-    (constant_load (SIadd64bit32bit i64:$sbase, imm:$offset)),
-    (vt (Instr_SGPR $sbase, (S_MOV_B32 imm:$offset)))
+    (frag vt:$value, (add i32:$ptr, (i32 IMM16bit:$offset))),
+    (inst (i1 0), $ptr, $value, (as_i16imm $offset))
   >;
 
-  // 3. No offset at all
   def : Pat <
-    (constant_load i64:$sbase),
-    (vt (Instr_IMM $sbase, 0))
+    (frag vt:$val, i32:$ptr),
+    (inst 0, $ptr, $val, 0)
   >;
 }
 
-defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, f32>;
-defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, i32>;
-defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, i64>;
-defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, v2i32>;
-defm : SMRD_Pattern <S_LOAD_DWORDX4_IMM, S_LOAD_DWORDX4_SGPR, i128>;
-defm : SMRD_Pattern <S_LOAD_DWORDX4_IMM, S_LOAD_DWORDX4_SGPR, v4i32>;
-defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v32i8>;
-defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v8i32>;
-defm : SMRD_Pattern <S_LOAD_DWORDX16_IMM, S_LOAD_DWORDX16_SGPR, v16i32>;
+defm : DSWritePat <DS_WRITE_B8, i32, truncstorei8_local>;
+defm : DSWritePat <DS_WRITE_B16, i32, truncstorei16_local>;
+defm : DSWritePat <DS_WRITE_B32, i32, local_store>;
+defm : DSWritePat <DS_WRITE_B64, v2i32, local_store>;
 
-//===----------------------------------------------------------------------===//
-// MUBUF Patterns
-//===----------------------------------------------------------------------===//
-
-multiclass MUBUFLoad_Pattern <MUBUF Instr_ADDR64, ValueType vt,
-                              PatFrag global_ld, PatFrag constant_ld> {
+multiclass DSAtomicRetPat<DS inst, ValueType vt, PatFrag frag> {
   def : Pat <
-    (vt (global_ld (add i64:$ptr, (i64 IMM12bit:$offset)))),
-    (Instr_ADDR64 (SI_ADDR64_RSRC (i64 0)), $ptr, (as_i16imm $offset))
+    (frag (add i32:$ptr, (i32 IMM16bit:$offset)), vt:$value),
+    (inst (i1 0), $ptr, $value, (as_i16imm $offset))
   >;
 
   def : Pat <
-    (vt (global_ld i64:$ptr)),
-    (Instr_ADDR64 (SI_ADDR64_RSRC (i64 0)), $ptr, 0)
+    (frag i32:$ptr, vt:$val),
+    (inst 0, $ptr, $val, 0)
   >;
+}
 
+// Special case of DSAtomicRetPat for add / sub 1 -> inc / dec
+//
+// We need to use something for the data0, so we set a register to
+// -1. For the non-rtn variants, the manual says it does
+// DS[A] = (DS[A] >= D0) ? 0 : DS[A] + 1, and setting D0 to uint_max
+// will always do the increment so I'm assuming it's the same.
+//
+// We also load this -1 with s_mov_b32 / s_mov_b64 even though this
+// needs to be a VGPR. The SGPR copy pass will fix this, and it's
+// easier since there is no v_mov_b64.
+multiclass DSAtomicIncRetPat<DS inst, ValueType vt,
+                             Instruction LoadImm, PatFrag frag> {
   def : Pat <
-     (vt (global_ld (add i64:$ptr, i64:$offset))),
-     (Instr_ADDR64 (SI_ADDR64_RSRC $ptr), $offset, 0)
+    (frag (add i32:$ptr, (i32 IMM16bit:$offset)), (vt 1)),
+    (inst (i1 0), $ptr, (LoadImm (vt -1)), (as_i16imm $offset))
   >;
 
   def : Pat <
-     (vt (constant_ld (add i64:$ptr, i64:$offset))),
-     (Instr_ADDR64 (SI_ADDR64_RSRC $ptr), $offset, 0)
+    (frag i32:$ptr, (vt 1)),
+    (inst 0, $ptr, (LoadImm (vt -1)), 0)
   >;
 }
 
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_SBYTE_ADDR64, i32,
-                          sextloadi8_global, sextloadi8_constant>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_ADDR64, i32,
-                          az_extloadi8_global, az_extloadi8_constant>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_SSHORT_ADDR64, i32,
-                          sextloadi16_global, sextloadi16_constant>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_USHORT_ADDR64, i32,
-                          az_extloadi16_global, az_extloadi16_constant>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORD_ADDR64, i32,
-                          global_load, constant_load>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, i64,
-                          global_load, constant_load>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, i64,
-                          az_extloadi32_global, az_extloadi32_constant>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, v2i32,
-                          global_load, constant_load>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX4_ADDR64, v4i32,
-                          global_load, constant_load>;
-
-multiclass MUBUFStore_Pattern <MUBUF Instr, ValueType vt, PatFrag st> {
+multiclass DSAtomicCmpXChg <DS inst, ValueType vt, PatFrag frag> {
+  def : Pat <
+    (frag (add i32:$ptr, (i32 IMM16bit:$offset)), vt:$cmp, vt:$swap),
+    (inst (i1 0), $ptr, $cmp, $swap, (as_i16imm $offset))
+  >;
 
   def : Pat <
-    (st vt:$value, i64:$ptr),
-    (Instr $value, (SI_ADDR64_RSRC (i64 0)), $ptr, 0)
+    (frag i32:$ptr, vt:$cmp, vt:$swap),
+    (inst 0, $ptr, $cmp, $swap, 0)
   >;
+}
+
 
+// 32-bit atomics.
+defm : DSAtomicIncRetPat<DS_INC_RTN_U32, i32,
+                         S_MOV_B32, atomic_load_add_local>;
+defm : DSAtomicIncRetPat<DS_DEC_RTN_U32, i32,
+                         S_MOV_B32, atomic_load_sub_local>;
+
+defm : DSAtomicRetPat<DS_WRXCHG_RTN_B32, i32, atomic_swap_local>;
+defm : DSAtomicRetPat<DS_ADD_RTN_U32, i32, atomic_load_add_local>;
+defm : DSAtomicRetPat<DS_SUB_RTN_U32, i32, atomic_load_sub_local>;
+defm : DSAtomicRetPat<DS_AND_RTN_B32, i32, atomic_load_and_local>;
+defm : DSAtomicRetPat<DS_OR_RTN_B32, i32, atomic_load_or_local>;
+defm : DSAtomicRetPat<DS_XOR_RTN_B32, i32, atomic_load_xor_local>;
+defm : DSAtomicRetPat<DS_MIN_RTN_I32, i32, atomic_load_min_local>;
+defm : DSAtomicRetPat<DS_MAX_RTN_I32, i32, atomic_load_max_local>;
+defm : DSAtomicRetPat<DS_MIN_RTN_U32, i32, atomic_load_umin_local>;
+defm : DSAtomicRetPat<DS_MAX_RTN_U32, i32, atomic_load_umax_local>;
+
+defm : DSAtomicCmpXChg<DS_CMPST_RTN_B32, i32, atomic_cmp_swap_32_local>;
+
+// 64-bit atomics.
+defm : DSAtomicIncRetPat<DS_INC_RTN_U64, i64,
+                         S_MOV_B64, atomic_load_add_local>;
+defm : DSAtomicIncRetPat<DS_DEC_RTN_U64, i64,
+                         S_MOV_B64, atomic_load_sub_local>;
+
+defm : DSAtomicRetPat<DS_WRXCHG_RTN_B64, i64, atomic_swap_local>;
+defm : DSAtomicRetPat<DS_ADD_RTN_U64, i64, atomic_load_add_local>;
+defm : DSAtomicRetPat<DS_SUB_RTN_U64, i64, atomic_load_sub_local>;
+defm : DSAtomicRetPat<DS_AND_RTN_B64, i64, atomic_load_and_local>;
+defm : DSAtomicRetPat<DS_OR_RTN_B64, i64, atomic_load_or_local>;
+defm : DSAtomicRetPat<DS_XOR_RTN_B64, i64, atomic_load_xor_local>;
+defm : DSAtomicRetPat<DS_MIN_RTN_I64, i64, atomic_load_min_local>;
+defm : DSAtomicRetPat<DS_MAX_RTN_I64, i64, atomic_load_max_local>;
+defm : DSAtomicRetPat<DS_MIN_RTN_U64, i64, atomic_load_umin_local>;
+defm : DSAtomicRetPat<DS_MAX_RTN_U64, i64, atomic_load_umax_local>;
+
+defm : DSAtomicCmpXChg<DS_CMPST_RTN_B64, i64, atomic_cmp_swap_64_local>;
+
+
+//===----------------------------------------------------------------------===//
+// MUBUF Patterns
+//===----------------------------------------------------------------------===//
+
+multiclass MUBUFLoad_Pattern <MUBUF Instr_ADDR64, ValueType vt,
+                              PatFrag constant_ld> {
   def : Pat <
-    (st vt:$value, (add i64:$ptr, i64:$offset)),
-    (Instr $value, (SI_ADDR64_RSRC $ptr), $offset, 0)
-   >;
+     (vt (constant_ld (add i64:$ptr, i64:$offset))),
+     (Instr_ADDR64 (SI_ADDR64_RSRC $ptr), $offset, 0)
+  >;
+
 }
 
-defm : MUBUFStore_Pattern <BUFFER_STORE_BYTE, i32, truncstorei8_global>;
-defm : MUBUFStore_Pattern <BUFFER_STORE_SHORT, i32, truncstorei16_global>;
-defm : MUBUFStore_Pattern <BUFFER_STORE_DWORD, i32, global_store>;
-defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX2, i64, global_store>;
-defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX2, v2i32, global_store>;
-defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX4, v4i32, global_store>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_SBYTE_ADDR64, i32, sextloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_ADDR64, i32, az_extloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_SSHORT_ADDR64, i32, sextloadi16_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_USHORT_ADDR64, i32, az_extloadi16_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORD_ADDR64, i32, constant_load>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, v2i32, constant_load>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX4_ADDR64, v4i32, constant_load>;
+
+class MUBUFScratchLoadPat <MUBUF Instr, ValueType vt, PatFrag ld> : Pat <
+  (vt (ld (MUBUFScratch v4i32:$srsrc, i32:$vaddr,
+                        i32:$soffset, u16imm:$offset))),
+  (Instr $srsrc, $vaddr, $soffset, $offset, 0, 0, 0)
+>;
+
+def : MUBUFScratchLoadPat <BUFFER_LOAD_SBYTE_OFFEN, i32, sextloadi8_private>;
+def : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, i32, extloadi8_private>;
+def : MUBUFScratchLoadPat <BUFFER_LOAD_SSHORT_OFFEN, i32, sextloadi16_private>;
+def : MUBUFScratchLoadPat <BUFFER_LOAD_USHORT_OFFEN, i32, extloadi16_private>;
+def : MUBUFScratchLoadPat <BUFFER_LOAD_DWORD_OFFEN, i32, load_private>;
+def : MUBUFScratchLoadPat <BUFFER_LOAD_DWORDX2_OFFEN, v2i32, load_private>;
+def : MUBUFScratchLoadPat <BUFFER_LOAD_DWORDX4_OFFEN, v4i32, load_private>;
 
 // BUFFER_LOAD_DWORD*, addr64=0
 multiclass MUBUF_Load_Dword <ValueType vt, MUBUF offset, MUBUF offen, MUBUF idxen,
                              MUBUF bothen> {
 
   def : Pat <
-    (vt (int_SI_buffer_load_dword i128:$rsrc, i32:$vaddr, i32:$soffset,
+    (vt (int_SI_buffer_load_dword v4i32:$rsrc, (i32 imm), i32:$soffset,
                                   imm:$offset, 0, 0, imm:$glc, imm:$slc,
                                   imm:$tfe)),
-    (offset $rsrc, $vaddr, (as_i16imm $offset), $soffset, (as_i1imm $glc),
+    (offset $rsrc, (as_i16imm $offset), $soffset, (as_i1imm $glc),
             (as_i1imm $slc), (as_i1imm $tfe))
   >;
 
   def : Pat <
-    (vt (int_SI_buffer_load_dword i128:$rsrc, i32:$vaddr, i32:$soffset,
-                                  imm, 1, 0, imm:$glc, imm:$slc,
+    (vt (int_SI_buffer_load_dword v4i32:$rsrc, i32:$vaddr, i32:$soffset,
+                                  imm:$offset, 1, 0, imm:$glc, imm:$slc,
                                   imm:$tfe)),
-    (offen $rsrc, $vaddr, $soffset, (as_i1imm $glc), (as_i1imm $slc),
+    (offen $rsrc, $vaddr, $soffset, (as_i16imm $offset), (as_i1imm $glc), (as_i1imm $slc),
            (as_i1imm $tfe))
   >;
 
   def : Pat <
-    (vt (int_SI_buffer_load_dword i128:$rsrc, i32:$vaddr, i32:$soffset,
+    (vt (int_SI_buffer_load_dword v4i32:$rsrc, i32:$vaddr, i32:$soffset,
                                   imm:$offset, 0, 1, imm:$glc, imm:$slc,
                                   imm:$tfe)),
     (idxen $rsrc, $vaddr, (as_i16imm $offset), $soffset, (as_i1imm $glc),
@@ -2030,7 +2659,7 @@ multiclass MUBUF_Load_Dword <ValueType vt, MUBUF offset, MUBUF offen, MUBUF idxe
   >;
 
   def : Pat <
-    (vt (int_SI_buffer_load_dword i128:$rsrc, v2i32:$vaddr, i32:$soffset,
+    (vt (int_SI_buffer_load_dword v4i32:$rsrc, v2i32:$vaddr, i32:$soffset,
                                   imm, 1, 1, imm:$glc, imm:$slc,
                                   imm:$tfe)),
     (bothen $rsrc, $vaddr, $soffset, (as_i1imm $glc), (as_i1imm $slc),
@@ -2045,13 +2674,41 @@ defm : MUBUF_Load_Dword <v2i32, BUFFER_LOAD_DWORDX2_OFFSET, BUFFER_LOAD_DWORDX2_
 defm : MUBUF_Load_Dword <v4i32, BUFFER_LOAD_DWORDX4_OFFSET, BUFFER_LOAD_DWORDX4_OFFEN,
                          BUFFER_LOAD_DWORDX4_IDXEN, BUFFER_LOAD_DWORDX4_BOTHEN>;
 
+class MUBUFScratchStorePat <MUBUF Instr, ValueType vt, PatFrag st> : Pat <
+  (st vt:$value, (MUBUFAddr32 v4i32:$srsrc, i32:$vaddr, i32:$soffset,
+                              u16imm:$offset, i1imm:$offen, i1imm:$idxen,
+                              i1imm:$glc, i1imm:$slc, i1imm:$tfe)),
+  (Instr $value, $srsrc, $vaddr, $soffset, $offset, $offen, $idxen,
+         $glc, $slc, $tfe)
+>;
+
+def : MUBUFScratchStorePat <BUFFER_STORE_BYTE, i32, truncstorei8_private>;
+def : MUBUFScratchStorePat <BUFFER_STORE_SHORT, i32, truncstorei16_private>;
+def : MUBUFScratchStorePat <BUFFER_STORE_DWORD, i32, store_private>;
+def : MUBUFScratchStorePat <BUFFER_STORE_DWORDX2, v2i32, store_private>;
+def : MUBUFScratchStorePat <BUFFER_STORE_DWORDX4, v4i32, store_private>;
+
+/*
+class MUBUFStore_Pattern <MUBUF Instr, ValueType vt, PatFrag st> : Pat <
+  (st vt:$value, (MUBUFScratch v4i32:$srsrc, i64:$vaddr, u16imm:$offset)),
+  (Instr $value, $srsrc, $vaddr, $offset)
+>;
+
+def : MUBUFStore_Pattern <BUFFER_STORE_BYTE_ADDR64, i32, truncstorei8_private>;
+def : MUBUFStore_Pattern <BUFFER_STORE_SHORT_ADDR64, i32, truncstorei16_private>;
+def : MUBUFStore_Pattern <BUFFER_STORE_DWORD_ADDR64, i32, store_private>;
+def : MUBUFStore_Pattern <BUFFER_STORE_DWORDX2_ADDR64, v2i32, store_private>;
+def : MUBUFStore_Pattern <BUFFER_STORE_DWORDX4_ADDR64, v4i32, store_private>;
+
+*/
+
 //===----------------------------------------------------------------------===//
 // MTBUF Patterns
 //===----------------------------------------------------------------------===//
 
 // TBUFFER_STORE_FORMAT_*, addr64=0
 class MTBUF_StoreResource <ValueType vt, int num_channels, MTBUF opcode> : Pat<
-  (SItbuffer_store i128:$rsrc, vt:$vdata, num_channels, i32:$vaddr,
+  (SItbuffer_store v4i32:$rsrc, vt:$vdata, num_channels, i32:$vaddr,
                    i32:$soffset, imm:$inst_offset, imm:$dfmt,
                    imm:$nfmt, imm:$offen, imm:$idxen,
                    imm:$glc, imm:$slc, imm:$tfe),
@@ -2066,90 +2723,198 @@ def : MTBUF_StoreResource <v2i32, 2, TBUFFER_STORE_FORMAT_XY>;
 def : MTBUF_StoreResource <v4i32, 3, TBUFFER_STORE_FORMAT_XYZ>;
 def : MTBUF_StoreResource <v4i32, 4, TBUFFER_STORE_FORMAT_XYZW>;
 
+let SubtargetPredicate = isCI in {
+
+// Sea island new arithmetic instructinos
+let neverHasSideEffects = 1 in {
+defm V_TRUNC_F64 : VOP1_64 <0x00000017, "V_TRUNC_F64",
+  [(set f64:$dst, (ftrunc f64:$src0))]
+>;
+defm V_CEIL_F64 : VOP1_64 <0x00000018, "V_CEIL_F64",
+  [(set f64:$dst, (fceil f64:$src0))]
+>;
+defm V_FLOOR_F64 : VOP1_64 <0x0000001A, "V_FLOOR_F64",
+  [(set f64:$dst, (ffloor f64:$src0))]
+>;
+defm V_RNDNE_F64 : VOP1_64 <0x00000019, "V_RNDNE_F64",
+  [(set f64:$dst, (frint f64:$src0))]
+>;
+
+defm V_QSAD_PK_U16_U8 : VOP3_32 <0x00000173, "V_QSAD_PK_U16_U8", []>;
+defm V_MQSAD_U16_U8 : VOP3_32 <0x000000172, "V_MQSAD_U16_U8", []>;
+defm V_MQSAD_U32_U8 : VOP3_32 <0x00000175, "V_MQSAD_U32_U8", []>;
+def V_MAD_U64_U32 : VOP3_64 <0x00000176, "V_MAD_U64_U32", []>;
+
+// XXX - Does this set VCC?
+def V_MAD_I64_I32 : VOP3_64 <0x00000177, "V_MAD_I64_I32", []>;
+} // End neverHasSideEffects = 1
+
+// Remaining instructions:
+// FLAT_*
+// S_CBRANCH_CDBGUSER
+// S_CBRANCH_CDBGSYS
+// S_CBRANCH_CDBGSYS_OR_USER
+// S_CBRANCH_CDBGSYS_AND_USER
+// S_DCACHE_INV_VOL
+// V_EXP_LEGACY_F32
+// V_LOG_LEGACY_F32
+// DS_NOP
+// DS_GWS_SEMA_RELEASE_ALL
+// DS_WRAP_RTN_B32
+// DS_CNDXCHG32_RTN_B64
+// DS_WRITE_B96
+// DS_WRITE_B128
+// DS_CONDXCHG32_RTN_B128
+// DS_READ_B96
+// DS_READ_B128
+// BUFFER_LOAD_DWORDX3
+// BUFFER_STORE_DWORDX3
+
+} // End iSCI
+
+
 /********** ====================== **********/
 /**********   Indirect adressing   **********/
 /********** ====================== **********/
 
-multiclass SI_INDIRECT_Pattern <ValueType vt, SI_INDIRECT_DST IndDst> {
+multiclass SI_INDIRECT_Pattern <ValueType vt, ValueType eltvt, SI_INDIRECT_DST IndDst> {
 
   // 1. Extract with offset
   def : Pat<
     (vector_extract vt:$vec, (add i32:$idx, imm:$off)),
-    (f32 (SI_INDIRECT_SRC (IMPLICIT_DEF), $vec, $idx, imm:$off))
+    (eltvt (SI_INDIRECT_SRC (IMPLICIT_DEF), $vec, $idx, imm:$off))
   >;
 
   // 2. Extract without offset
   def : Pat<
     (vector_extract vt:$vec, i32:$idx),
-    (f32 (SI_INDIRECT_SRC (IMPLICIT_DEF), $vec, $idx, 0))
+    (eltvt (SI_INDIRECT_SRC (IMPLICIT_DEF), $vec, $idx, 0))
   >;
 
   // 3. Insert with offset
   def : Pat<
-    (vector_insert vt:$vec, f32:$val, (add i32:$idx, imm:$off)),
+    (vector_insert vt:$vec, eltvt:$val, (add i32:$idx, imm:$off)),
     (IndDst (IMPLICIT_DEF), $vec, $idx, imm:$off, $val)
   >;
 
   // 4. Insert without offset
   def : Pat<
-    (vector_insert vt:$vec, f32:$val, i32:$idx),
+    (vector_insert vt:$vec, eltvt:$val, i32:$idx),
     (IndDst (IMPLICIT_DEF), $vec, $idx, 0, $val)
   >;
 }
 
-defm : SI_INDIRECT_Pattern <v2f32, SI_INDIRECT_DST_V2>;
-defm : SI_INDIRECT_Pattern <v4f32, SI_INDIRECT_DST_V4>;
-defm : SI_INDIRECT_Pattern <v8f32, SI_INDIRECT_DST_V8>;
-defm : SI_INDIRECT_Pattern <v16f32, SI_INDIRECT_DST_V16>;
+defm : SI_INDIRECT_Pattern <v2f32, f32, SI_INDIRECT_DST_V2>;
+defm : SI_INDIRECT_Pattern <v4f32, f32, SI_INDIRECT_DST_V4>;
+defm : SI_INDIRECT_Pattern <v8f32, f32, SI_INDIRECT_DST_V8>;
+defm : SI_INDIRECT_Pattern <v16f32, f32, SI_INDIRECT_DST_V16>;
 
-/********** =============== **********/
-/**********   Conditions    **********/
-/********** =============== **********/
-
-def : Pat<
-  (i1 (setcc f32:$src0, f32:$src1, SETO)),
-  (V_CMP_O_F32_e64 $src0, $src1)
->;
-
-def : Pat<
-  (i1 (setcc f32:$src0, f32:$src1, SETUO)),
-  (V_CMP_U_F32_e64 $src0, $src1)
->;
+defm : SI_INDIRECT_Pattern <v2i32, i32, SI_INDIRECT_DST_V2>;
+defm : SI_INDIRECT_Pattern <v4i32, i32, SI_INDIRECT_DST_V4>;
+defm : SI_INDIRECT_Pattern <v8i32, i32, SI_INDIRECT_DST_V8>;
+defm : SI_INDIRECT_Pattern <v16i32, i32, SI_INDIRECT_DST_V16>;
 
 //===----------------------------------------------------------------------===//
-// Miscellaneous Patterns
+// Conversion Patterns
 //===----------------------------------------------------------------------===//
 
+def : Pat<(i32 (sext_inreg i32:$src, i1)),
+  (S_BFE_I32 i32:$src, 65536)>; // 0 | 1 << 16
+
+// TODO: Match 64-bit BFE. SI has a 64-bit BFE, but it's scalar only so it
+// might not be worth the effort, and will need to expand to shifts when
+// fixing SGPR copies.
+
+// Handle sext_inreg in i64
 def : Pat <
-  (i64 (trunc i128:$x)),
+  (i64 (sext_inreg i64:$src, i1)),
   (INSERT_SUBREG (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
-    (i32 (EXTRACT_SUBREG $x, sub0)), sub0),
-    (i32 (EXTRACT_SUBREG $x, sub1)), sub1)
+    (S_BFE_I32 (EXTRACT_SUBREG i64:$src, sub0), 65536), sub0), // 0 | 1 << 16
+    (S_MOV_B32 -1), sub1)
 >;
 
 def : Pat <
-  (i32 (trunc i64:$a)),
-  (EXTRACT_SUBREG $a, sub0)
+  (i64 (sext_inreg i64:$src, i8)),
+  (INSERT_SUBREG (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
+    (S_SEXT_I32_I8 (EXTRACT_SUBREG i64:$src, sub0)), sub0),
+    (S_MOV_B32 -1), sub1)
 >;
 
 def : Pat <
-  (i1 (trunc i32:$a)),
-  (V_CMP_EQ_I32_e64 (V_AND_B32_e32 (i32 1), $a), 1)
+  (i64 (sext_inreg i64:$src, i16)),
+  (INSERT_SUBREG (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
+    (S_SEXT_I32_I16 (EXTRACT_SUBREG i64:$src, sub0)), sub0),
+    (S_MOV_B32 -1), sub1)
 >;
 
-// V_ADD_I32_e32/S_ADD_I32 produces carry in VCC/SCC. For the vector
-// case, the sgpr-copies pass will fix this to use the vector version.
+class ZExt_i64_i32_Pat <SDNode ext> : Pat <
+  (i64 (ext i32:$src)),
+  (INSERT_SUBREG (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $src, sub0),
+    (S_MOV_B32 0), sub1)
+>;
+
+class ZExt_i64_i1_Pat <SDNode ext> : Pat <
+  (i64 (ext i1:$src)),
+  (INSERT_SUBREG
+    (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
+      (V_CNDMASK_B32_e64 (i32 0), (i32 1), $src), sub0),
+    (S_MOV_B32 0), sub1)
+>;
+
+
+def : ZExt_i64_i32_Pat<zext>;
+def : ZExt_i64_i32_Pat<anyext>;
+def : ZExt_i64_i1_Pat<zext>;
+def : ZExt_i64_i1_Pat<anyext>;
+
 def : Pat <
-  (i32 (addc i32:$src0, i32:$src1)),
-  (S_ADD_I32 $src0, $src1)
+  (i64 (sext i32:$src)),
+    (INSERT_SUBREG
+      (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $src, sub0),
+      (S_ASHR_I32 $src, 31), sub1)
 >;
 
 def : Pat <
-  (or i64:$a, i64:$b),
+  (i64 (sext i1:$src)),
   (INSERT_SUBREG
-    (INSERT_SUBREG (IMPLICIT_DEF),
-      (V_OR_B32_e32 (EXTRACT_SUBREG $a, sub0), (EXTRACT_SUBREG $b, sub0)), sub0),
-    (V_OR_B32_e32 (EXTRACT_SUBREG $a, sub1), (EXTRACT_SUBREG $b, sub1)), sub1)
+    (INSERT_SUBREG
+      (i64 (IMPLICIT_DEF)),
+      (V_CNDMASK_B32_e64 0, -1, $src), sub0),
+    (V_CNDMASK_B32_e64 0, -1, $src), sub1)
+>;
+
+def : Pat <
+  (f32 (sint_to_fp i1:$src)),
+  (V_CNDMASK_B32_e64 (i32 0), CONST.FP32_NEG_ONE, $src)
+>;
+
+def : Pat <
+  (f32 (uint_to_fp i1:$src)),
+  (V_CNDMASK_B32_e64 (i32 0), CONST.FP32_ONE, $src)
+>;
+
+def : Pat <
+  (f64 (sint_to_fp i1:$src)),
+    (V_CVT_F64_I32_e32 (V_CNDMASK_B32_e64 (i32 0), (i32 -1), $src))
+>;
+
+def : Pat <
+  (f64 (uint_to_fp i1:$src)),
+  (V_CVT_F64_U32_e32 (V_CNDMASK_B32_e64 (i32 0), (i32 1), $src))
+>;
+
+//===----------------------------------------------------------------------===//
+// Miscellaneous Patterns
+//===----------------------------------------------------------------------===//
+
+def : Pat <
+  (i32 (trunc i64:$a)),
+  (EXTRACT_SUBREG $a, sub0)
+>;
+
+def : Pat <
+  (i1 (trunc i32:$a)),
+  (V_CMP_EQ_I32_e64 (V_AND_B32_e32 (i32 1), $a), 1)
 >;
 
 //============================================================================//
diff --git a/contrib/llvm/lib/Target/R600/SIIntrinsics.td b/contrib/llvm/lib/Target/R600/SIIntrinsics.td
index 00e32c0..027a0a2 100644
--- a/contrib/llvm/lib/Target/R600/SIIntrinsics.td
+++ b/contrib/llvm/lib/Target/R600/SIIntrinsics.td
@@ -54,15 +54,132 @@ let TargetPrefix = "SI", isTarget = 1 in {
 
   def int_SI_sendmsg : Intrinsic <[], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
 
+  // Fully-flexible SAMPLE instruction.
+  class SampleRaw : Intrinsic <
+    [llvm_v4f32_ty],    // vdata(VGPR)
+    [llvm_anyint_ty,    // vaddr(VGPR)
+     llvm_v8i32_ty,     // rsrc(SGPR)
+     llvm_v4i32_ty,     // sampler(SGPR)
+     llvm_i32_ty,       // dmask(imm)
+     llvm_i32_ty,       // unorm(imm)
+     llvm_i32_ty,       // r128(imm)
+     llvm_i32_ty,       // da(imm)
+     llvm_i32_ty,       // glc(imm)
+     llvm_i32_ty,       // slc(imm)
+     llvm_i32_ty,       // tfe(imm)
+     llvm_i32_ty],      // lwe(imm)
+    [IntrNoMem]>;
+
+  // Image instruction without a sampler.
+  class Image : Intrinsic <
+    [llvm_v4f32_ty],    // vdata(VGPR)
+    [llvm_anyint_ty,    // vaddr(VGPR)
+     llvm_v8i32_ty,     // rsrc(SGPR)
+     llvm_i32_ty,       // dmask(imm)
+     llvm_i32_ty,       // unorm(imm)
+     llvm_i32_ty,       // r128(imm)
+     llvm_i32_ty,       // da(imm)
+     llvm_i32_ty,       // glc(imm)
+     llvm_i32_ty,       // slc(imm)
+     llvm_i32_ty,       // tfe(imm)
+     llvm_i32_ty],      // lwe(imm)
+    [IntrNoMem]>;
+
+  // Basic sample
+  def int_SI_image_sample : SampleRaw;
+  def int_SI_image_sample_cl : SampleRaw;
+  def int_SI_image_sample_d : SampleRaw;
+  def int_SI_image_sample_d_cl : SampleRaw;
+  def int_SI_image_sample_l : SampleRaw;
+  def int_SI_image_sample_b : SampleRaw;
+  def int_SI_image_sample_b_cl : SampleRaw;
+  def int_SI_image_sample_lz : SampleRaw;
+  def int_SI_image_sample_cd : SampleRaw;
+  def int_SI_image_sample_cd_cl : SampleRaw;
+
+  // Sample with comparison
+  def int_SI_image_sample_c : SampleRaw;
+  def int_SI_image_sample_c_cl : SampleRaw;
+  def int_SI_image_sample_c_d : SampleRaw;
+  def int_SI_image_sample_c_d_cl : SampleRaw;
+  def int_SI_image_sample_c_l : SampleRaw;
+  def int_SI_image_sample_c_b : SampleRaw;
+  def int_SI_image_sample_c_b_cl : SampleRaw;
+  def int_SI_image_sample_c_lz : SampleRaw;
+  def int_SI_image_sample_c_cd : SampleRaw;
+  def int_SI_image_sample_c_cd_cl : SampleRaw;
+
+  // Sample with offsets
+  def int_SI_image_sample_o : SampleRaw;
+  def int_SI_image_sample_cl_o : SampleRaw;
+  def int_SI_image_sample_d_o : SampleRaw;
+  def int_SI_image_sample_d_cl_o : SampleRaw;
+  def int_SI_image_sample_l_o : SampleRaw;
+  def int_SI_image_sample_b_o : SampleRaw;
+  def int_SI_image_sample_b_cl_o : SampleRaw;
+  def int_SI_image_sample_lz_o : SampleRaw;
+  def int_SI_image_sample_cd_o : SampleRaw;
+  def int_SI_image_sample_cd_cl_o : SampleRaw;
+
+  // Sample with comparison and offsets
+  def int_SI_image_sample_c_o : SampleRaw;
+  def int_SI_image_sample_c_cl_o : SampleRaw;
+  def int_SI_image_sample_c_d_o : SampleRaw;
+  def int_SI_image_sample_c_d_cl_o : SampleRaw;
+  def int_SI_image_sample_c_l_o : SampleRaw;
+  def int_SI_image_sample_c_b_o : SampleRaw;
+  def int_SI_image_sample_c_b_cl_o : SampleRaw;
+  def int_SI_image_sample_c_lz_o : SampleRaw;
+  def int_SI_image_sample_c_cd_o : SampleRaw;
+  def int_SI_image_sample_c_cd_cl_o : SampleRaw;
+
+  // Basic gather4
+  def int_SI_gather4 : SampleRaw;
+  def int_SI_gather4_cl : SampleRaw;
+  def int_SI_gather4_l : SampleRaw;
+  def int_SI_gather4_b : SampleRaw;
+  def int_SI_gather4_b_cl : SampleRaw;
+  def int_SI_gather4_lz : SampleRaw;
+
+  // Gather4 with comparison
+  def int_SI_gather4_c : SampleRaw;
+  def int_SI_gather4_c_cl : SampleRaw;
+  def int_SI_gather4_c_l : SampleRaw;
+  def int_SI_gather4_c_b : SampleRaw;
+  def int_SI_gather4_c_b_cl : SampleRaw;
+  def int_SI_gather4_c_lz : SampleRaw;
+
+  // Gather4 with offsets
+  def int_SI_gather4_o : SampleRaw;
+  def int_SI_gather4_cl_o : SampleRaw;
+  def int_SI_gather4_l_o : SampleRaw;
+  def int_SI_gather4_b_o : SampleRaw;
+  def int_SI_gather4_b_cl_o : SampleRaw;
+  def int_SI_gather4_lz_o : SampleRaw;
+
+  // Gather4 with comparison and offsets
+  def int_SI_gather4_c_o : SampleRaw;
+  def int_SI_gather4_c_cl_o : SampleRaw;
+  def int_SI_gather4_c_l_o : SampleRaw;
+  def int_SI_gather4_c_b_o : SampleRaw;
+  def int_SI_gather4_c_b_cl_o : SampleRaw;
+  def int_SI_gather4_c_lz_o : SampleRaw;
+
+  def int_SI_getlod : SampleRaw;
+
+  // Image instrinsics.
+  def int_SI_image_load : Image;
+  def int_SI_image_load_mip : Image;
+  def int_SI_getresinfo : Image;
+
+  // Deprecated image and sample intrinsics.
   class Sample : Intrinsic <[llvm_v4f32_ty], [llvm_anyvector_ty, llvm_v32i8_ty, llvm_anyint_ty, llvm_i32_ty], [IntrNoMem]>;
 
   def int_SI_sample : Sample;
   def int_SI_sampleb : Sample;
   def int_SI_sampled : Sample;
   def int_SI_samplel : Sample;
-
   def int_SI_imageload : Intrinsic <[llvm_v4i32_ty], [llvm_anyvector_ty, llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>;
-
   def int_SI_resinfo : Intrinsic <[llvm_v4i32_ty], [llvm_i32_ty, llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>;
 
   /* Interpolation Intrinsics */
diff --git a/contrib/llvm/lib/Target/R600/SILowerControlFlow.cpp b/contrib/llvm/lib/Target/R600/SILowerControlFlow.cpp
index ef867d3..75b5a5e 100644
--- a/contrib/llvm/lib/Target/R600/SILowerControlFlow.cpp
+++ b/contrib/llvm/lib/Target/R600/SILowerControlFlow.cpp
@@ -55,6 +55,7 @@
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/Constants.h"
 
 using namespace llvm;
 
@@ -66,8 +67,8 @@ private:
   static const unsigned SkipThreshold = 12;
 
   static char ID;
-  const TargetRegisterInfo *TRI;
-  const TargetInstrInfo *TII;
+  const SIRegisterInfo *TRI;
+  const SIInstrInfo *TII;
 
   bool shouldSkip(MachineBasicBlock *From, MachineBasicBlock *To);
 
@@ -85,17 +86,18 @@ private:
   void Kill(MachineInstr &MI);
   void Branch(MachineInstr &MI);
 
+  void InitM0ForLDS(MachineBasicBlock::iterator MI);
   void LoadM0(MachineInstr &MI, MachineInstr *MovRel);
   void IndirectSrc(MachineInstr &MI);
   void IndirectDst(MachineInstr &MI);
 
 public:
   SILowerControlFlowPass(TargetMachine &tm) :
-    MachineFunctionPass(ID), TRI(0), TII(0) { }
+    MachineFunctionPass(ID), TRI(nullptr), TII(nullptr) { }
 
-  virtual bool runOnMachineFunction(MachineFunction &MF);
+  bool runOnMachineFunction(MachineFunction &MF) override;
 
-  const char *getPassName() const {
+  const char *getPassName() const override {
     return "SI Lower control flow instructions";
   }
 
@@ -109,23 +111,6 @@ FunctionPass *llvm::createSILowerControlFlowPass(TargetMachine &tm) {
   return new SILowerControlFlowPass(tm);
 }
 
-static bool isDS(unsigned Opcode) {
-  switch(Opcode) {
-  default: return false;
-  case AMDGPU::DS_ADD_U32_RTN:
-  case AMDGPU::DS_SUB_U32_RTN:
-  case AMDGPU::DS_WRITE_B32:
-  case AMDGPU::DS_WRITE_B8:
-  case AMDGPU::DS_WRITE_B16:
-  case AMDGPU::DS_READ_B32:
-  case AMDGPU::DS_READ_I8:
-  case AMDGPU::DS_READ_U8:
-  case AMDGPU::DS_READ_I16:
-  case AMDGPU::DS_READ_U16:
-    return true;
-  }
-}
-
 bool SILowerControlFlowPass::shouldSkip(MachineBasicBlock *From,
                                         MachineBasicBlock *To) {
 
@@ -162,7 +147,7 @@ void SILowerControlFlowPass::SkipIfDead(MachineInstr &MI) {
   MachineBasicBlock &MBB = *MI.getParent();
   DebugLoc DL = MI.getDebugLoc();
 
-  if (MBB.getParent()->getInfo<SIMachineFunctionInfo>()->ShaderType !=
+  if (MBB.getParent()->getInfo<SIMachineFunctionInfo>()->getShaderType() !=
       ShaderType::PIXEL ||
       !shouldSkip(&MBB, &MBB.getParent()->back()))
     return;
@@ -302,33 +287,50 @@ void SILowerControlFlowPass::EndCf(MachineInstr &MI) {
 }
 
 void SILowerControlFlowPass::Branch(MachineInstr &MI) {
-  MachineBasicBlock *Next = MI.getParent()->getNextNode();
-  MachineBasicBlock *Target = MI.getOperand(0).getMBB();
-  if (Target == Next)
+  if (MI.getOperand(0).getMBB() == MI.getParent()->getNextNode())
     MI.eraseFromParent();
-  else
-    assert(0);
+
+  // If these aren't equal, this is probably an infinite loop.
 }
 
 void SILowerControlFlowPass::Kill(MachineInstr &MI) {
-
   MachineBasicBlock &MBB = *MI.getParent();
   DebugLoc DL = MI.getDebugLoc();
-
-  // Kill is only allowed in pixel / geometry shaders
-  assert(MBB.getParent()->getInfo<SIMachineFunctionInfo>()->ShaderType ==
-         ShaderType::PIXEL ||
-         MBB.getParent()->getInfo<SIMachineFunctionInfo>()->ShaderType ==
-         ShaderType::GEOMETRY);
-
-  // Clear this pixel from the exec mask if the operand is negative
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_CMPX_LE_F32_e32), AMDGPU::VCC)
-          .addImm(0)
-          .addOperand(MI.getOperand(0));
+  const MachineOperand &Op = MI.getOperand(0);
+
+#ifndef NDEBUG
+  const SIMachineFunctionInfo *MFI
+    = MBB.getParent()->getInfo<SIMachineFunctionInfo>();
+  // Kill is only allowed in pixel / geometry shaders.
+  assert(MFI->getShaderType() == ShaderType::PIXEL ||
+         MFI->getShaderType() == ShaderType::GEOMETRY);
+#endif
+
+  // Clear this thread from the exec mask if the operand is negative
+  if ((Op.isImm() || Op.isFPImm())) {
+    // Constant operand: Set exec mask to 0 or do nothing
+    if (Op.isImm() ? (Op.getImm() & 0x80000000) :
+        Op.getFPImm()->isNegative()) {
+      BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B64), AMDGPU::EXEC)
+              .addImm(0);
+    }
+  } else {
+    BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_CMPX_LE_F32_e32), AMDGPU::VCC)
+           .addImm(0)
+           .addOperand(Op);
+  }
 
   MI.eraseFromParent();
 }
 
+/// The m0 register stores the maximum allowable address for LDS reads and
+/// writes.  Its value must be at least the size in bytes of LDS allocated by
+/// the shader.  For simplicity, we set it to the maximum possible value.
+void SILowerControlFlowPass::InitM0ForLDS(MachineBasicBlock::iterator MI) {
+    BuildMI(*MI->getParent(), MI, MI->getDebugLoc(),  TII->get(AMDGPU::S_MOV_B32),
+            AMDGPU::M0).addImm(0xffffffff);
+}
+
 void SILowerControlFlowPass::LoadM0(MachineInstr &MI, MachineInstr *MovRel) {
 
   MachineBasicBlock &MBB = *MI.getParent();
@@ -342,51 +344,57 @@ void SILowerControlFlowPass::LoadM0(MachineInstr &MI, MachineInstr *MovRel) {
     BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B32), AMDGPU::M0)
             .addReg(Idx);
     MBB.insert(I, MovRel);
-    MI.eraseFromParent();
-    return;
-  }
+  } else {
 
-  assert(AMDGPU::SReg_64RegClass.contains(Save));
-  assert(AMDGPU::VReg_32RegClass.contains(Idx));
+    assert(AMDGPU::SReg_64RegClass.contains(Save));
+    assert(AMDGPU::VReg_32RegClass.contains(Idx));
 
-  // Save the EXEC mask
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B64), Save)
-          .addReg(AMDGPU::EXEC);
+    // Save the EXEC mask
+    BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B64), Save)
+            .addReg(AMDGPU::EXEC);
 
-  // Read the next variant into VCC (lower 32 bits) <- also loop target
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32_e32), AMDGPU::VCC)
-          .addReg(Idx);
+    // Read the next variant into VCC (lower 32 bits) <- also loop target
+    BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32),
+            AMDGPU::VCC_LO)
+            .addReg(Idx);
 
-  // Move index from VCC into M0
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B32), AMDGPU::M0)
-          .addReg(AMDGPU::VCC);
+    // Move index from VCC into M0
+    BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B32), AMDGPU::M0)
+            .addReg(AMDGPU::VCC_LO);
 
-  // Compare the just read M0 value to all possible Idx values
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_CMP_EQ_U32_e32), AMDGPU::VCC)
-          .addReg(AMDGPU::M0)
-          .addReg(Idx);
+    // Compare the just read M0 value to all possible Idx values
+    BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_CMP_EQ_U32_e32), AMDGPU::VCC)
+            .addReg(AMDGPU::M0)
+            .addReg(Idx);
 
-  // Update EXEC, save the original EXEC value to VCC
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_AND_SAVEEXEC_B64), AMDGPU::VCC)
-          .addReg(AMDGPU::VCC);
+    // Update EXEC, save the original EXEC value to VCC
+    BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_AND_SAVEEXEC_B64), AMDGPU::VCC)
+            .addReg(AMDGPU::VCC);
 
-  // Do the actual move
-  MBB.insert(I, MovRel);
+    // Do the actual move
+    MBB.insert(I, MovRel);
 
-  // Update EXEC, switch all done bits to 0 and all todo bits to 1
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_XOR_B64), AMDGPU::EXEC)
-          .addReg(AMDGPU::EXEC)
-          .addReg(AMDGPU::VCC);
+    // Update EXEC, switch all done bits to 0 and all todo bits to 1
+    BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_XOR_B64), AMDGPU::EXEC)
+            .addReg(AMDGPU::EXEC)
+            .addReg(AMDGPU::VCC);
 
-  // Loop back to V_READFIRSTLANE_B32 if there are still variants to cover
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_CBRANCH_EXECNZ))
-          .addImm(-7)
-          .addReg(AMDGPU::EXEC);
+    // Loop back to V_READFIRSTLANE_B32 if there are still variants to cover
+    BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_CBRANCH_EXECNZ))
+            .addImm(-7)
+            .addReg(AMDGPU::EXEC);
 
-  // Restore EXEC
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B64), AMDGPU::EXEC)
-          .addReg(Save);
+    // Restore EXEC
+    BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B64), AMDGPU::EXEC)
+            .addReg(Save);
 
+  }
+  // FIXME: Are there any values other than the LDS address clamp that need to
+  // be stored in the m0 register and may be live for more than a few
+  // instructions?  If so, we should save the m0 register at the beginning
+  // of this function and restore it here.
+  // FIXME: Add support for LDS direct loads.
+  InitM0ForLDS(&MI);
   MI.eraseFromParent();
 }
 
@@ -434,8 +442,8 @@ void SILowerControlFlowPass::IndirectDst(MachineInstr &MI) {
 }
 
 bool SILowerControlFlowPass::runOnMachineFunction(MachineFunction &MF) {
-  TII = MF.getTarget().getInstrInfo();
-  TRI = MF.getTarget().getRegisterInfo();
+  TII = static_cast<const SIInstrInfo*>(MF.getTarget().getInstrInfo());
+  TRI = static_cast<const SIRegisterInfo*>(MF.getTarget().getRegisterInfo());
   SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
 
   bool HaveKill = false;
@@ -447,12 +455,12 @@ bool SILowerControlFlowPass::runOnMachineFunction(MachineFunction &MF) {
        BI != BE; ++BI) {
 
     MachineBasicBlock &MBB = *BI;
-    for (MachineBasicBlock::iterator I = MBB.begin(), Next = llvm::next(I);
-         I != MBB.end(); I = Next) {
+    MachineBasicBlock::iterator I, Next;
+    for (I = MBB.begin(); I != MBB.end(); I = Next) {
+      Next = std::next(I);
 
-      Next = llvm::next(I);
       MachineInstr &MI = *I;
-      if (isDS(MI.getOpcode())) {
+      if (TII->isDS(MI.getOpcode())) {
         NeedM0 = true;
         NeedWQM = true;
       }
@@ -531,11 +539,10 @@ bool SILowerControlFlowPass::runOnMachineFunction(MachineFunction &MF) {
     MachineBasicBlock &MBB = MF.front();
     // Initialize M0 to a value that won't cause LDS access to be discarded
     // due to offset clamping
-    BuildMI(MBB, MBB.getFirstNonPHI(), DebugLoc(), TII->get(AMDGPU::S_MOV_B32),
-            AMDGPU::M0).addImm(0xffffffff);
+    InitM0ForLDS(MBB.getFirstNonPHI());
   }
 
-  if (NeedWQM && MFI->ShaderType == ShaderType::PIXEL) {
+  if (NeedWQM && MFI->getShaderType() == ShaderType::PIXEL) {
     MachineBasicBlock &MBB = MF.front();
     BuildMI(MBB, MBB.getFirstNonPHI(), DebugLoc(), TII->get(AMDGPU::S_WQM_B64),
             AMDGPU::EXEC).addReg(AMDGPU::EXEC);
diff --git a/contrib/llvm/lib/Target/R600/SILowerI1Copies.cpp b/contrib/llvm/lib/Target/R600/SILowerI1Copies.cpp
new file mode 100644
index 0000000..db19235
--- /dev/null
+++ b/contrib/llvm/lib/Target/R600/SILowerI1Copies.cpp
@@ -0,0 +1,154 @@
+//===-- SILowerI1Copies.cpp - Lower I1 Copies -----------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// i1 values are usually inserted by the CFG Structurize pass and they are
+/// unique in that they can be copied from VALU to SALU registers.
+/// This is not possible for any other value type.  Since there are no
+/// MOV instructions for i1, we to use V_CMP_* and V_CNDMASK to move the i1.
+///
+//===----------------------------------------------------------------------===//
+//
+
+#define DEBUG_TYPE "si-i1-copies"
+#include "AMDGPU.h"
+#include "SIInstrInfo.h"
+#include "llvm/CodeGen/LiveIntervalAnalysis.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Function.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Target/TargetMachine.h"
+
+using namespace llvm;
+
+namespace {
+
+class SILowerI1Copies : public MachineFunctionPass {
+public:
+  static char ID;
+
+public:
+  SILowerI1Copies() : MachineFunctionPass(ID) {
+    initializeSILowerI1CopiesPass(*PassRegistry::getPassRegistry());
+  }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF) override;
+
+  virtual const char *getPassName() const override {
+    return "SI Lower il Copies";
+  }
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const override {
+  AU.addRequired<MachineDominatorTree>();
+    AU.setPreservesCFG();
+    MachineFunctionPass::getAnalysisUsage(AU);
+  }
+};
+
+} // End anonymous namespace.
+
+INITIALIZE_PASS_BEGIN(SILowerI1Copies, DEBUG_TYPE,
+                      "SI Lower il Copies", false, false)
+INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
+INITIALIZE_PASS_END(SILowerI1Copies, DEBUG_TYPE,
+                    "SI Lower il Copies", false, false)
+
+char SILowerI1Copies::ID = 0;
+
+char &llvm::SILowerI1CopiesID = SILowerI1Copies::ID;
+
+FunctionPass *llvm::createSILowerI1CopiesPass() {
+  return new SILowerI1Copies();
+}
+
+bool SILowerI1Copies::runOnMachineFunction(MachineFunction &MF) {
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+  const SIInstrInfo *TII = static_cast<const SIInstrInfo *>(
+      MF.getTarget().getInstrInfo());
+  const TargetRegisterInfo *TRI = MF.getTarget().getRegisterInfo();
+  std::vector<unsigned> I1Defs;
+
+  for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
+                                                  BI != BE; ++BI) {
+
+    MachineBasicBlock &MBB = *BI;
+    MachineBasicBlock::iterator I, Next;
+    for (I = MBB.begin(); I != MBB.end(); I = Next) {
+      Next = std::next(I);
+      MachineInstr &MI = *I;
+
+      if (MI.getOpcode() == AMDGPU::V_MOV_I1) {
+        I1Defs.push_back(MI.getOperand(0).getReg());
+        MI.setDesc(TII->get(AMDGPU::V_MOV_B32_e32));
+        continue;
+      }
+
+      if (MI.getOpcode() == AMDGPU::V_AND_I1) {
+        I1Defs.push_back(MI.getOperand(0).getReg());
+        MI.setDesc(TII->get(AMDGPU::V_AND_B32_e32));
+        continue;
+      }
+
+      if (MI.getOpcode() == AMDGPU::V_OR_I1) {
+        I1Defs.push_back(MI.getOperand(0).getReg());
+        MI.setDesc(TII->get(AMDGPU::V_OR_B32_e32));
+        continue;
+      }
+
+      if (MI.getOpcode() == AMDGPU::V_XOR_I1) {
+        I1Defs.push_back(MI.getOperand(0).getReg());
+        MI.setDesc(TII->get(AMDGPU::V_XOR_B32_e32));
+        continue;
+      }
+
+      if (MI.getOpcode() != AMDGPU::COPY ||
+          !TargetRegisterInfo::isVirtualRegister(MI.getOperand(0).getReg()) ||
+          !TargetRegisterInfo::isVirtualRegister(MI.getOperand(1).getReg()))
+        continue;
+
+
+      const TargetRegisterClass *DstRC =
+          MRI.getRegClass(MI.getOperand(0).getReg());
+      const TargetRegisterClass *SrcRC =
+          MRI.getRegClass(MI.getOperand(1).getReg());
+
+      if (DstRC == &AMDGPU::VReg_1RegClass &&
+          TRI->getCommonSubClass(SrcRC, &AMDGPU::SGPR_64RegClass)) {
+        I1Defs.push_back(MI.getOperand(0).getReg());
+        BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(AMDGPU::V_CNDMASK_B32_e64))
+                .addOperand(MI.getOperand(0))
+                .addImm(0)
+                .addImm(-1)
+                .addOperand(MI.getOperand(1))
+                .addImm(0)
+                .addImm(0)
+                .addImm(0)
+                .addImm(0);
+        MI.eraseFromParent();
+      } else if (TRI->getCommonSubClass(DstRC, &AMDGPU::SGPR_64RegClass) &&
+                 SrcRC == &AMDGPU::VReg_1RegClass) {
+        BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(AMDGPU::V_CMP_NE_I32_e64))
+                .addOperand(MI.getOperand(0))
+                .addImm(0)
+                .addOperand(MI.getOperand(1))
+                .addImm(0)
+                .addImm(0)
+                .addImm(0)
+                .addImm(0);
+        MI.eraseFromParent();
+      }
+    }
+  }
+
+  for (unsigned Reg : I1Defs)
+    MRI.setRegClass(Reg, &AMDGPU::VReg_32RegClass);
+
+  return false;
+}
diff --git a/contrib/llvm/lib/Target/R600/SIMachineFunctionInfo.cpp b/contrib/llvm/lib/Target/R600/SIMachineFunctionInfo.cpp
index 071f9fa..c53a7e1 100644
--- a/contrib/llvm/lib/Target/R600/SIMachineFunctionInfo.cpp
+++ b/contrib/llvm/lib/Target/R600/SIMachineFunctionInfo.cpp
@@ -10,6 +10,13 @@
 
 
 #include "SIMachineFunctionInfo.h"
+#include "SIInstrInfo.h"
+#include "SIRegisterInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/LLVMContext.h"
+
+#define MAX_LANES 64
 
 using namespace llvm;
 
@@ -19,4 +26,72 @@ void SIMachineFunctionInfo::anchor() {}
 
 SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
   : AMDGPUMachineFunction(MF),
-    PSInputAddr(0) { }
+    PSInputAddr(0),
+    SpillTracker(),
+    NumUserSGPRs(0) { }
+
+static unsigned createLaneVGPR(MachineRegisterInfo &MRI, MachineFunction *MF) {
+  unsigned VGPR = MRI.createVirtualRegister(&AMDGPU::VReg_32RegClass);
+
+  // We need to add this register as live out for the function, in order to
+  // have the live range calculated directly.
+  //
+  // When register spilling begins, we have already calculated the live
+  // live intervals for all the registers.  Since we are spilling SGPRs to
+  // VGPRs, we need to update the Lane VGPR's live interval every time we
+  // spill or restore a register.
+  //
+  // Unfortunately, there is no good way to update the live interval as
+  // the TargetInstrInfo callbacks for spilling and restoring don't give
+  // us access to the live interval information.
+  //
+  // We are lucky, though, because the InlineSpiller calls
+  // LiveRangeEdit::calculateRegClassAndHint() which iterates through
+  // all the new register that have been created when restoring a register
+  // and calls LiveIntervals::getInterval(), which creates and computes
+  // the live interval for the newly created register.  However, once this
+  // live intervals is created, it doesn't change and since we usually reuse
+  // the Lane VGPR multiple times, this means any uses after the first aren't
+  // added to the live interval.
+  //
+  // To work around this, we add Lane VGPRs to the functions live out list,
+  // so that we can guarantee its live range will cover all of its uses.
+
+  for (MachineBasicBlock &MBB : *MF) {
+    if (MBB.back().getOpcode() == AMDGPU::S_ENDPGM) {
+      MBB.back().addOperand(*MF, MachineOperand::CreateReg(VGPR, false, true));
+      return VGPR;
+    }
+  }
+
+  LLVMContext &Ctx = MF->getFunction()->getContext();
+  Ctx.emitError("Could not find S_ENDPGM instruction.");
+
+  return VGPR;
+}
+
+unsigned SIMachineFunctionInfo::RegSpillTracker::reserveLanes(
+    MachineRegisterInfo &MRI, MachineFunction *MF, unsigned NumRegs) {
+  unsigned StartLane = CurrentLane;
+  CurrentLane += NumRegs;
+  if (!LaneVGPR) {
+    LaneVGPR = createLaneVGPR(MRI, MF);
+  } else {
+    if (CurrentLane >= MAX_LANES) {
+      StartLane = CurrentLane = 0;
+      LaneVGPR = createLaneVGPR(MRI, MF);
+    }
+  }
+  return StartLane;
+}
+
+void SIMachineFunctionInfo::RegSpillTracker::addSpilledReg(unsigned FrameIndex,
+                                                           unsigned Reg,
+                                                           int Lane) {
+  SpilledRegisters[FrameIndex] = SpilledReg(Reg, Lane);
+}
+
+const SIMachineFunctionInfo::SpilledReg&
+SIMachineFunctionInfo::RegSpillTracker::getSpilledReg(unsigned FrameIndex) {
+  return SpilledRegisters[FrameIndex];
+}
diff --git a/contrib/llvm/lib/Target/R600/SIMachineFunctionInfo.h b/contrib/llvm/lib/Target/R600/SIMachineFunctionInfo.h
index 2f1961c..9684d28 100644
--- a/contrib/llvm/lib/Target/R600/SIMachineFunctionInfo.h
+++ b/contrib/llvm/lib/Target/R600/SIMachineFunctionInfo.h
@@ -16,16 +16,50 @@
 #define SIMACHINEFUNCTIONINFO_H_
 
 #include "AMDGPUMachineFunction.h"
+#include <map>
 
 namespace llvm {
 
+class MachineRegisterInfo;
+
 /// This class keeps track of the SPI_SP_INPUT_ADDR config register, which
 /// tells the hardware which interpolation parameters to load.
 class SIMachineFunctionInfo : public AMDGPUMachineFunction {
-  virtual void anchor();
+  void anchor() override;
 public:
+
+  struct SpilledReg {
+    unsigned VGPR;
+    int Lane;
+    SpilledReg(unsigned R, int L) : VGPR (R), Lane (L) { }
+    SpilledReg() : VGPR(0), Lane(-1) { }
+    bool hasLane() { return Lane != -1;}
+  };
+
+  struct RegSpillTracker {
+  private:
+    unsigned CurrentLane;
+    std::map<unsigned, SpilledReg> SpilledRegisters;
+  public:
+    unsigned LaneVGPR;
+    RegSpillTracker() : CurrentLane(0), SpilledRegisters(), LaneVGPR(0) { }
+    /// \p NumRegs The number of consecutive registers what need to be spilled.
+    ///            This function will ensure that all registers are stored in
+    ///            the same VGPR.
+    /// \returns The lane to be used for storing the first register.
+    unsigned reserveLanes(MachineRegisterInfo &MRI, MachineFunction *MF,
+                          unsigned NumRegs = 1);
+    void addSpilledReg(unsigned FrameIndex, unsigned Reg, int Lane = -1);
+    const SpilledReg& getSpilledReg(unsigned FrameIndex);
+    bool programSpillsRegisters() { return !SpilledRegisters.empty(); }
+  };
+
+  // SIMachineFunctionInfo definition
+
   SIMachineFunctionInfo(const MachineFunction &MF);
   unsigned PSInputAddr;
+  struct RegSpillTracker SpillTracker;
+  unsigned NumUserSGPRs;
 };
 
 } // End namespace llvm
diff --git a/contrib/llvm/lib/Target/R600/SIRegisterInfo.cpp b/contrib/llvm/lib/Target/R600/SIRegisterInfo.cpp
index ed0bbaf..2a9a2ac 100644
--- a/contrib/llvm/lib/Target/R600/SIRegisterInfo.cpp
+++ b/contrib/llvm/lib/Target/R600/SIRegisterInfo.cpp
@@ -14,22 +14,23 @@
 
 
 #include "SIRegisterInfo.h"
-#include "AMDGPUTargetMachine.h"
+#include "AMDGPUSubtarget.h"
 #include "SIInstrInfo.h"
+#include "SIMachineFunctionInfo.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/RegisterScavenging.h"
 
 using namespace llvm;
 
-SIRegisterInfo::SIRegisterInfo(AMDGPUTargetMachine &tm)
-: AMDGPURegisterInfo(tm),
-  TM(tm)
+SIRegisterInfo::SIRegisterInfo(const AMDGPUSubtarget &st)
+: AMDGPURegisterInfo(st)
   { }
 
 BitVector SIRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   BitVector Reserved(getNumRegs());
   Reserved.set(AMDGPU::EXEC);
   Reserved.set(AMDGPU::INDIRECT_BASE_ADDR);
-  const SIInstrInfo *TII = static_cast<const SIInstrInfo*>(TM.getInstrInfo());
-  TII->reserveIndirectRegisters(Reserved, MF);
   return Reserved;
 }
 
@@ -38,12 +39,27 @@ unsigned SIRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC,
   return RC->getNumRegs();
 }
 
-const TargetRegisterClass *
-SIRegisterInfo::getISARegClass(const TargetRegisterClass * rc) const {
-  switch (rc->getID()) {
-  case AMDGPU::GPRF32RegClassID:
-    return &AMDGPU::VReg_32RegClass;
-  default: return rc;
+bool SIRegisterInfo::requiresRegisterScavenging(const MachineFunction &Fn) const {
+  return Fn.getFrameInfo()->hasStackObjects();
+}
+
+void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
+                                        int SPAdj, unsigned FIOperandNum,
+                                        RegScavenger *RS) const {
+  MachineFunction *MF = MI->getParent()->getParent();
+  MachineFrameInfo *FrameInfo = MF->getFrameInfo();
+  const SIInstrInfo *TII = static_cast<const SIInstrInfo*>(ST.getInstrInfo());
+  MachineOperand &FIOp = MI->getOperand(FIOperandNum);
+  int Index = MI->getOperand(FIOperandNum).getIndex();
+  int64_t Offset = FrameInfo->getObjectOffset(Index);
+
+  FIOp.ChangeToImmediate(Offset);
+  if (!TII->isImmOperandLegal(MI, FIOperandNum, FIOp)) {
+    unsigned TmpReg = RS->scavengeRegister(&AMDGPU::VReg_32RegClass, MI, SPAdj);
+    BuildMI(*MI->getParent(), MI, MI->getDebugLoc(),
+            TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
+            .addImm(Offset);
+    FIOp.ChangeToRegister(TmpReg, false);
   }
 }
 
@@ -56,7 +72,7 @@ const TargetRegisterClass * SIRegisterInfo::getCFGStructurizerRegClass(
 }
 
 unsigned SIRegisterInfo::getHWRegIndex(unsigned Reg) const {
-  return getEncodingValue(Reg);
+  return getEncodingValue(Reg) & 0xff;
 }
 
 const TargetRegisterClass *SIRegisterInfo::getPhysRegClass(unsigned Reg) const {
@@ -71,13 +87,12 @@ const TargetRegisterClass *SIRegisterInfo::getPhysRegClass(unsigned Reg) const {
     &AMDGPU::SReg_256RegClass
   };
 
-  for (unsigned i = 0, e = sizeof(BaseClasses) /
-                           sizeof(const TargetRegisterClass*); i != e; ++i) {
-    if (BaseClasses[i]->contains(Reg)) {
-      return BaseClasses[i];
+  for (const TargetRegisterClass *BaseClass : BaseClasses) {
+    if (BaseClass->contains(Reg)) {
+      return BaseClass;
     }
   }
-  return NULL;
+  return nullptr;
 }
 
 bool SIRegisterInfo::isSGPRClass(const TargetRegisterClass *RC) const {
@@ -113,7 +128,7 @@ const TargetRegisterClass *SIRegisterInfo::getEquivalentVGPRClass(
     } else if (getCommonSubClass(SRC, &AMDGPU::SReg_512RegClass)) {
       return &AMDGPU::VReg_512RegClass;
     }
-    return NULL;
+    return nullptr;
 }
 
 const TargetRegisterClass *SIRegisterInfo::getSubRegClass(
@@ -122,10 +137,52 @@ const TargetRegisterClass *SIRegisterInfo::getSubRegClass(
     return RC;
 
   // If this register has a sub-register, we can safely assume it is a 32-bit
-  // register, becuase all of SI's sub-registers are 32-bit.
+  // register, because all of SI's sub-registers are 32-bit.
   if (isSGPRClass(RC)) {
     return &AMDGPU::SGPR_32RegClass;
   } else {
     return &AMDGPU::VGPR_32RegClass;
   }
 }
+
+unsigned SIRegisterInfo::getPhysRegSubReg(unsigned Reg,
+                                          const TargetRegisterClass *SubRC,
+                                          unsigned Channel) const {
+  unsigned Index = getHWRegIndex(Reg);
+  return SubRC->getRegister(Index + Channel);
+}
+
+bool SIRegisterInfo::regClassCanUseImmediate(int RCID) const {
+  switch (RCID) {
+  default: return false;
+  case AMDGPU::SSrc_32RegClassID:
+  case AMDGPU::SSrc_64RegClassID:
+  case AMDGPU::VSrc_32RegClassID:
+  case AMDGPU::VSrc_64RegClassID:
+    return true;
+  }
+}
+
+bool SIRegisterInfo::regClassCanUseImmediate(
+                             const TargetRegisterClass *RC) const {
+  return regClassCanUseImmediate(RC->getID());
+}
+
+unsigned SIRegisterInfo::getPreloadedValue(const MachineFunction &MF,
+                                           enum PreloadedValue Value) const {
+
+  const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
+  switch (Value) {
+  case SIRegisterInfo::TGID_X:
+    return AMDGPU::SReg_32RegClass.getRegister(MFI->NumUserSGPRs + 0);
+  case SIRegisterInfo::TGID_Y:
+    return AMDGPU::SReg_32RegClass.getRegister(MFI->NumUserSGPRs + 1);
+  case SIRegisterInfo::TGID_Z:
+    return AMDGPU::SReg_32RegClass.getRegister(MFI->NumUserSGPRs + 2);
+  case SIRegisterInfo::SCRATCH_WAVE_OFFSET:
+    return AMDGPU::SReg_32RegClass.getRegister(MFI->NumUserSGPRs + 4);
+  case SIRegisterInfo::SCRATCH_PTR:
+    return AMDGPU::SGPR2_SGPR3;
+  }
+  llvm_unreachable("unexpected preloaded value type");
+}
diff --git a/contrib/llvm/lib/Target/R600/SIRegisterInfo.h b/contrib/llvm/lib/Target/R600/SIRegisterInfo.h
index 8148f7f..5d0235c 100644
--- a/contrib/llvm/lib/Target/R600/SIRegisterInfo.h
+++ b/contrib/llvm/lib/Target/R600/SIRegisterInfo.h
@@ -20,29 +20,26 @@
 
 namespace llvm {
 
-class AMDGPUTargetMachine;
-
 struct SIRegisterInfo : public AMDGPURegisterInfo {
-  AMDGPUTargetMachine &TM;
 
-  SIRegisterInfo(AMDGPUTargetMachine &tm);
+  SIRegisterInfo(const AMDGPUSubtarget &st);
+
+  BitVector getReservedRegs(const MachineFunction &MF) const override;
 
-  virtual BitVector getReservedRegs(const MachineFunction &MF) const;
+  unsigned getRegPressureLimit(const TargetRegisterClass *RC,
+                               MachineFunction &MF) const override;
 
-  virtual unsigned getRegPressureLimit(const TargetRegisterClass *RC,
-                                       MachineFunction &MF) const;
+  bool requiresRegisterScavenging(const MachineFunction &Fn) const override;
 
-  /// \param RC is an AMDIL reg class.
-  ///
-  /// \returns the SI register class that is equivalent to \p RC.
-  virtual const TargetRegisterClass *
-    getISARegClass(const TargetRegisterClass *RC) const;
+  void eliminateFrameIndex(MachineBasicBlock::iterator MI, int SPAdj,
+                           unsigned FIOperandNum,
+                           RegScavenger *RS) const override;
 
   /// \brief get the register class of the specified type to use in the
   /// CFGStructurizer
-  virtual const TargetRegisterClass * getCFGStructurizerRegClass(MVT VT) const;
+  const TargetRegisterClass * getCFGStructurizerRegClass(MVT VT) const override;
 
-  virtual unsigned getHWRegIndex(unsigned Reg) const;
+  unsigned getHWRegIndex(unsigned Reg) const override;
 
   /// \brief Return the 'base' register class for this register.
   /// e.g. SGPR0 => SReg_32, VGPR => VReg_32 SGPR0_SGPR1 -> SReg_32, etc.
@@ -63,6 +60,33 @@ struct SIRegisterInfo : public AMDGPURegisterInfo {
   /// be returned.
   const TargetRegisterClass *getSubRegClass(const TargetRegisterClass *RC,
                                             unsigned SubIdx) const;
+
+  /// \p Channel This is the register channel (e.g. a value from 0-16), not the
+  ///            SubReg index.
+  /// \returns The sub-register of Reg that is in Channel.
+  unsigned getPhysRegSubReg(unsigned Reg, const TargetRegisterClass *SubRC,
+                            unsigned Channel) const;
+
+  /// \returns True if operands defined with this register class can accept
+  /// inline immediates.
+  bool regClassCanUseImmediate(int RCID) const;
+
+  /// \returns True if operands defined with this register class can accept
+  /// inline immediates.
+  bool regClassCanUseImmediate(const TargetRegisterClass *RC) const;
+
+  enum PreloadedValue {
+    TGID_X,
+    TGID_Y,
+    TGID_Z,
+    SCRATCH_WAVE_OFFSET,
+    SCRATCH_PTR
+  };
+
+  /// \brief Returns the physical register that \p Value is stored in.
+  unsigned getPreloadedValue(const MachineFunction &MF,
+                             enum PreloadedValue Value) const;
+
 };
 
 } // End namespace llvm
diff --git a/contrib/llvm/lib/Target/R600/SIRegisterInfo.td b/contrib/llvm/lib/Target/R600/SIRegisterInfo.td
index 49bdbc9..8974b63 100644
--- a/contrib/llvm/lib/Target/R600/SIRegisterInfo.td
+++ b/contrib/llvm/lib/Target/R600/SIRegisterInfo.td
@@ -17,7 +17,16 @@ class SIReg <string n, bits<16> encoding = 0> : Register<n> {
 }
 
 // Special Registers
-def VCC : SIReg<"VCC", 106>;
+def VCC_LO : SIReg<"vcc_lo", 106>;
+def VCC_HI : SIReg<"vcc_hi", 107>;
+
+// VCC for 64-bit instructions
+def VCC : RegisterWithSubRegs<"VCC", [VCC_LO, VCC_HI]> {
+  let Namespace = "AMDGPU";
+  let SubRegIndices = [sub0, sub1];
+  let HWEncoding = 106;
+}
+
 def EXEC : SIReg<"EXEC", 126>;
 def SCC : SIReg<"SCC", 253>;
 def M0 : SIReg <"M0", 124>;
@@ -150,7 +159,7 @@ def M0Reg : RegisterClass<"AMDGPU", [i32], 32, (add M0)>;
 
 // Register class for all scalar registers (SGPRs + Special Registers)
 def SReg_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
-  (add SGPR_32, M0Reg)
+  (add SGPR_32, M0Reg, VCC_LO)
 >;
 
 def SGPR_64 : RegisterClass<"AMDGPU", [v2i32, i64], 64, (add SGPR_64Regs)>;
@@ -159,7 +168,7 @@ def SReg_64 : RegisterClass<"AMDGPU", [v2i32, i64, i1], 64,
   (add SGPR_64Regs, VCCReg, EXECReg)
 >;
 
-def SReg_128 : RegisterClass<"AMDGPU", [i128, v4i32], 128, (add SGPR_128)>;
+def SReg_128 : RegisterClass<"AMDGPU", [v4i32, v16i8], 128, (add SGPR_128)>;
 
 def SReg_256 : RegisterClass<"AMDGPU", [v32i8, v8i32, v8f32], 256, (add SGPR_256)>;
 
@@ -174,14 +183,16 @@ def VReg_96 : RegisterClass<"AMDGPU", [untyped], 96, (add VGPR_96)> {
   let Size = 96;
 }
 
-def VReg_128 : RegisterClass<"AMDGPU", [v4i32, v4f32, i128], 128, (add VGPR_128)>;
+def VReg_128 : RegisterClass<"AMDGPU", [v4i32, v4f32], 128, (add VGPR_128)>;
 
 def VReg_256 : RegisterClass<"AMDGPU", [v32i8, v8i32, v8f32], 256, (add VGPR_256)>;
 
 def VReg_512 : RegisterClass<"AMDGPU", [v16i32, v16f32], 512, (add VGPR_512)>;
 
+def VReg_1 : RegisterClass<"AMDGPU", [i1], 32, (add VGPR_32)>;
+
 //===----------------------------------------------------------------------===//
-//  [SV]Src_* register classes, can have either an immediate or an register
+//  [SV]Src_(32|64) register classes, can have either an immediate or an register
 //===----------------------------------------------------------------------===//
 
 def SSrc_32 : RegisterClass<"AMDGPU", [i32, f32], 32, (add SReg_32)>;
@@ -192,3 +203,9 @@ def VSrc_32 : RegisterClass<"AMDGPU", [i32, f32], 32, (add VReg_32, SReg_32)>;
 
 def VSrc_64 : RegisterClass<"AMDGPU", [i64, f64], 64, (add VReg_64, SReg_64)>;
 
+//===----------------------------------------------------------------------===//
+// SGPR and VGPR register classes
+//===----------------------------------------------------------------------===//
+
+def VSrc_128 : RegisterClass<"AMDGPU", [v4i32, v4f32], 128,
+                             (add VReg_128, SReg_128)>;
diff --git a/contrib/llvm/lib/Target/R600/SIShrinkInstructions.cpp b/contrib/llvm/lib/Target/R600/SIShrinkInstructions.cpp
new file mode 100644
index 0000000..745c4b6
--- /dev/null
+++ b/contrib/llvm/lib/Target/R600/SIShrinkInstructions.cpp
@@ -0,0 +1,194 @@
+//===-- SIShrinkInstructions.cpp - Shrink Instructions --------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// The pass tries to use the 32-bit encoding for instructions when possible.
+//===----------------------------------------------------------------------===//
+//
+
+#include "AMDGPU.h"
+#include "SIInstrInfo.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Function.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Target/TargetMachine.h"
+
+#define DEBUG_TYPE "si-shrink-instructions"
+
+STATISTIC(NumInstructionsShrunk,
+          "Number of 64-bit instruction reduced to 32-bit.");
+
+namespace llvm {
+  void initializeSIShrinkInstructionsPass(PassRegistry&);
+}
+
+using namespace llvm;
+
+namespace {
+
+class SIShrinkInstructions : public MachineFunctionPass {
+public:
+  static char ID;
+
+public:
+  SIShrinkInstructions() : MachineFunctionPass(ID) {
+  }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF) override;
+
+  virtual const char *getPassName() const override {
+    return "SI Shrink Instructions";
+  }
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.setPreservesCFG();
+    MachineFunctionPass::getAnalysisUsage(AU);
+  }
+};
+
+} // End anonymous namespace.
+
+INITIALIZE_PASS_BEGIN(SIShrinkInstructions, DEBUG_TYPE,
+                      "SI Lower il Copies", false, false)
+INITIALIZE_PASS_END(SIShrinkInstructions, DEBUG_TYPE,
+                    "SI Lower il Copies", false, false)
+
+char SIShrinkInstructions::ID = 0;
+
+FunctionPass *llvm::createSIShrinkInstructionsPass() {
+  return new SIShrinkInstructions();
+}
+
+static bool isVGPR(const MachineOperand *MO, const SIRegisterInfo &TRI,
+                   const MachineRegisterInfo &MRI) {
+  if (!MO->isReg())
+    return false;
+
+  if (TargetRegisterInfo::isVirtualRegister(MO->getReg()))
+    return TRI.hasVGPRs(MRI.getRegClass(MO->getReg()));
+
+  return TRI.hasVGPRs(TRI.getPhysRegClass(MO->getReg()));
+}
+
+static bool canShrink(MachineInstr &MI, const SIInstrInfo *TII,
+                      const SIRegisterInfo &TRI,
+                      const MachineRegisterInfo &MRI) {
+
+  const MachineOperand *Src2 = TII->getNamedOperand(MI, AMDGPU::OpName::src2);
+  // Can't shrink instruction with three operands.
+  if (Src2)
+    return false;
+
+  const MachineOperand *Src1 = TII->getNamedOperand(MI, AMDGPU::OpName::src1);
+  const MachineOperand *Src1Mod =
+      TII->getNamedOperand(MI, AMDGPU::OpName::src1_modifiers);
+
+  if (Src1 && (!isVGPR(Src1, TRI, MRI) || Src1Mod->getImm() != 0))
+    return false;
+
+  // We don't need to check src0, all input types are legal, so just make
+  // sure src0 isn't using any modifiers.
+  const MachineOperand *Src0Mod =
+      TII->getNamedOperand(MI, AMDGPU::OpName::src0_modifiers);
+  if (Src0Mod && Src0Mod->getImm() != 0)
+    return false;
+
+  // Check output modifiers
+  const MachineOperand *Omod = TII->getNamedOperand(MI, AMDGPU::OpName::omod);
+  if (Omod && Omod->getImm() != 0)
+    return false;
+
+  const MachineOperand *Clamp = TII->getNamedOperand(MI, AMDGPU::OpName::clamp);
+  return !Clamp || Clamp->getImm() == 0;
+}
+
+bool SIShrinkInstructions::runOnMachineFunction(MachineFunction &MF) {
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+  const SIInstrInfo *TII = static_cast<const SIInstrInfo *>(
+      MF.getTarget().getInstrInfo());
+  const SIRegisterInfo &TRI = TII->getRegisterInfo();
+  std::vector<unsigned> I1Defs;
+
+  for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
+                                                  BI != BE; ++BI) {
+
+    MachineBasicBlock &MBB = *BI;
+    MachineBasicBlock::iterator I, Next;
+    for (I = MBB.begin(); I != MBB.end(); I = Next) {
+      Next = std::next(I);
+      MachineInstr &MI = *I;
+
+      if (!TII->hasVALU32BitEncoding(MI.getOpcode()))
+        continue;
+
+      if (!canShrink(MI, TII, TRI, MRI)) {
+        // Try commtuing the instruction and see if that enables us to shrink
+        // it.
+        if (!MI.isCommutable() || !TII->commuteInstruction(&MI) ||
+            !canShrink(MI, TII, TRI, MRI))
+          continue;
+      }
+
+      int Op32 = AMDGPU::getVOPe32(MI.getOpcode());
+
+      // Op32 could be -1 here if we started with an instruction that had a
+      // a 32-bit encoding and then commuted it to an instruction that did not.
+      if (Op32 == -1)
+        continue;
+
+      if (TII->isVOPC(Op32)) {
+        unsigned DstReg = MI.getOperand(0).getReg();
+        if (TargetRegisterInfo::isVirtualRegister(DstReg)) {
+          // VOPC instructions can only write to the VCC register.  We can't
+          // force them to use VCC here, because the register allocator
+          // has trouble with sequences like this, which cause the allocator
+          // to run out of registes if vreg0 and vreg1 belong to the VCCReg
+          // register class:
+          // vreg0 = VOPC;
+          // vreg1 = VOPC;
+          // S_AND_B64 vreg0, vreg1
+          //
+          // So, instead of forcing the instruction to write to VCC, we provide a
+          // hint to the register allocator to use VCC and then we
+          // we will run this pass again after RA and shrink it if it outpus to
+          // VCC.
+          MRI.setRegAllocationHint(MI.getOperand(0).getReg(), 0, AMDGPU::VCC);
+          continue;
+        }
+        if (DstReg != AMDGPU::VCC)
+          continue;
+      }
+
+      // We can shrink this instruction
+      DEBUG(dbgs() << "Shrinking "; MI.dump(); dbgs() << "\n";);
+
+      MachineInstrBuilder MIB =
+          BuildMI(MBB, I, MI.getDebugLoc(), TII->get(Op32));
+
+      // dst
+      MIB.addOperand(MI.getOperand(0));
+
+      MIB.addOperand(*TII->getNamedOperand(MI, AMDGPU::OpName::src0));
+
+      const MachineOperand *Src1 =
+          TII->getNamedOperand(MI, AMDGPU::OpName::src1);
+      if (Src1)
+        MIB.addOperand(*Src1);
+
+      for (const MachineOperand &MO : MI.implicit_operands())
+        MIB.addOperand(MO);
+
+      DEBUG(dbgs() << "e32 MI = "; MI.dump(); dbgs() << "\n";);
+      ++NumInstructionsShrunk;
+      MI.eraseFromParent();
+    }
+  }
+  return false;
+}
diff --git a/contrib/llvm/lib/Target/R600/SITypeRewriter.cpp b/contrib/llvm/lib/Target/R600/SITypeRewriter.cpp
index f194d8b..367963a 100644
--- a/contrib/llvm/lib/Target/R600/SITypeRewriter.cpp
+++ b/contrib/llvm/lib/Target/R600/SITypeRewriter.cpp
@@ -22,9 +22,8 @@
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPU.h"
-
 #include "llvm/IR/IRBuilder.h"
-#include "llvm/InstVisitor.h"
+#include "llvm/IR/InstVisitor.h"
 
 using namespace llvm;
 
@@ -36,13 +35,13 @@ class SITypeRewriter : public FunctionPass,
   static char ID;
   Module *Mod;
   Type *v16i8;
-  Type *i128;
+  Type *v4i32;
 
 public:
   SITypeRewriter() : FunctionPass(ID) { }
-  virtual bool doInitialization(Module &M);
-  virtual bool runOnFunction(Function &F);
-  virtual const char *getPassName() const {
+  bool doInitialization(Module &M) override;
+  bool runOnFunction(Function &F) override;
+  const char *getPassName() const override {
     return "SI Type Rewriter";
   }
   void visitLoadInst(LoadInst &I);
@@ -57,7 +56,7 @@ char SITypeRewriter::ID = 0;
 bool SITypeRewriter::doInitialization(Module &M) {
   Mod = &M;
   v16i8 = VectorType::get(Type::getInt8Ty(M.getContext()), 16);
-  i128 = Type::getIntNTy(M.getContext(), 128);
+  v4i32 = VectorType::get(Type::getInt32Ty(M.getContext()), 4);
   return false;
 }
 
@@ -70,11 +69,11 @@ bool SITypeRewriter::runOnFunction(Function &F) {
     StringRef Str = A.getValueAsString();
     Str.getAsInteger(0, ShaderType);
   }
-  if (ShaderType != ShaderType::COMPUTE) {
-    visit(F);
-  }
+  if (ShaderType == ShaderType::COMPUTE)
+    return false;
 
   visit(F);
+  visit(F);
 
   return false;
 }
@@ -85,7 +84,8 @@ void SITypeRewriter::visitLoadInst(LoadInst &I) {
   Type *ElemTy = PtrTy->getPointerElementType();
   IRBuilder<> Builder(&I);
   if (ElemTy == v16i8)  {
-    Value *BitCast = Builder.CreateBitCast(Ptr, Type::getIntNPtrTy(I.getContext(), 128, 2));
+    Value *BitCast = Builder.CreateBitCast(Ptr,
+        PointerType::get(v4i32,PtrTy->getPointerAddressSpace()));
     LoadInst *Load = Builder.CreateLoad(BitCast);
     SmallVector <std::pair<unsigned, MDNode*>, 8> MD;
     I.getAllMetadataOtherThanDebugLoc(MD);
@@ -100,6 +100,7 @@ void SITypeRewriter::visitLoadInst(LoadInst &I) {
 
 void SITypeRewriter::visitCallInst(CallInst &I) {
   IRBuilder<> Builder(&I);
+
   SmallVector <Value*, 8> Args;
   SmallVector <Type*, 8> Types;
   bool NeedToReplace = false;
@@ -108,18 +109,17 @@ void SITypeRewriter::visitCallInst(CallInst &I) {
   for (unsigned i = 0, e = I.getNumArgOperands(); i != e; ++i) {
     Value *Arg = I.getArgOperand(i);
     if (Arg->getType() == v16i8) {
-      Args.push_back(Builder.CreateBitCast(Arg, i128));
-      Types.push_back(i128);
+      Args.push_back(Builder.CreateBitCast(Arg, v4i32));
+      Types.push_back(v4i32);
       NeedToReplace = true;
-      Name = Name + ".i128";
+      Name = Name + ".v4i32";
     } else if (Arg->getType()->isVectorTy() &&
                Arg->getType()->getVectorNumElements() == 1 &&
                Arg->getType()->getVectorElementType() ==
                                               Type::getInt32Ty(I.getContext())){
       Type *ElementTy = Arg->getType()->getVectorElementType();
       std::string TypeName = "i32";
-      InsertElementInst *Def = dyn_cast<InsertElementInst>(Arg);
-      assert(Def);
+      InsertElementInst *Def = cast<InsertElementInst>(Arg);
       Args.push_back(Def->getOperand(1));
       Types.push_back(ElementTy);
       std::string VecTypeName = "v1" + TypeName;
@@ -145,12 +145,12 @@ void SITypeRewriter::visitCallInst(CallInst &I) {
 
 void SITypeRewriter::visitBitCast(BitCastInst &I) {
   IRBuilder<> Builder(&I);
-  if (I.getDestTy() != i128) {
+  if (I.getDestTy() != v4i32) {
     return;
   }
 
   if (BitCastInst *Op = dyn_cast<BitCastInst>(I.getOperand(0))) {
-    if (Op->getSrcTy() == i128) {
+    if (Op->getSrcTy() == v4i32) {
       I.replaceAllUsesWith(Op->getOperand(0));
       I.eraseFromParent();
     }