Vendor import of llvm trunk r308421:vendor/llvm/llvm-trunk-r308421

https://llvm.org/svn/llvm-project/llvm/trunk@308421

28 files changed, 446 insertions, 219 deletions

diff --git a/lib/Target/AMDGPU/AMDGPU.h b/lib/Target/AMDGPU/AMDGPU.h
index 5a799b2d88d0..568682899be5 100644
--- a/lib/Target/AMDGPU/AMDGPU.h
+++ b/lib/Target/AMDGPU/AMDGPU.h
@@ -56,7 +56,7 @@ extern char &AMDGPUMachineCFGStructurizerID;
 
 void initializeAMDGPUAlwaysInlinePass(PassRegistry&);
 
-ModulePass *createAMDGPUAnnotateKernelFeaturesPass();
+Pass *createAMDGPUAnnotateKernelFeaturesPass();
 void initializeAMDGPUAnnotateKernelFeaturesPass(PassRegistry &);
 extern char &AMDGPUAnnotateKernelFeaturesID;
 
diff --git a/lib/Target/AMDGPU/AMDGPUAnnotateKernelFeatures.cpp b/lib/Target/AMDGPU/AMDGPUAnnotateKernelFeatures.cpp
index 7235d8fae332..c68e5861ff25 100644
--- a/lib/Target/AMDGPU/AMDGPUAnnotateKernelFeatures.cpp
+++ b/lib/Target/AMDGPU/AMDGPUAnnotateKernelFeatures.cpp
@@ -15,8 +15,10 @@
 #include "AMDGPU.h"
 #include "AMDGPUSubtarget.h"
 #include "llvm/ADT/Triple.h"
+#include "llvm/Analysis/CallGraphSCCPass.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
 #include "llvm/IR/Constants.h"
+#include "llvm/IR/InstIterator.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/Module.h"
 
@@ -26,26 +28,27 @@ using namespace llvm;
 
 namespace {
 
-class AMDGPUAnnotateKernelFeatures : public ModulePass {
+class AMDGPUAnnotateKernelFeatures : public CallGraphSCCPass {
 private:
+  const TargetMachine *TM = nullptr;
   AMDGPUAS AS;
-  static bool hasAddrSpaceCast(const Function &F, AMDGPUAS AS);
 
-  void addAttrToCallers(Function *Intrin, StringRef AttrName);
-  bool addAttrsForIntrinsics(Module &M, ArrayRef<StringRef[2]>);
+  bool addFeatureAttributes(Function &F);
 
 public:
   static char ID;
 
-  AMDGPUAnnotateKernelFeatures() : ModulePass(ID) {}
-  bool runOnModule(Module &M) override;
+  AMDGPUAnnotateKernelFeatures() : CallGraphSCCPass(ID) {}
+
+  bool doInitialization(CallGraph &CG) override;
+  bool runOnSCC(CallGraphSCC &SCC) override;
   StringRef getPassName() const override {
     return "AMDGPU Annotate Kernel Features";
   }
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.setPreservesAll();
-    ModulePass::getAnalysisUsage(AU);
+    CallGraphSCCPass::getAnalysisUsage(AU);
   }
 
   static bool visitConstantExpr(const ConstantExpr *CE, AMDGPUAS AS);
@@ -121,16 +124,130 @@ bool AMDGPUAnnotateKernelFeatures::visitConstantExprsRecursively(
   return false;
 }
 
-// Return true if an addrspacecast is used that requires the queue ptr.
-bool AMDGPUAnnotateKernelFeatures::hasAddrSpaceCast(const Function &F,
-    AMDGPUAS AS) {
+// We do not need to note the x workitem or workgroup id because they are always
+// initialized.
+//
+// TODO: We should not add the attributes if the known compile time workgroup
+// size is 1 for y/z.
+static StringRef intrinsicToAttrName(Intrinsic::ID ID,
+                                     bool &NonKernelOnly,
+                                     bool &IsQueuePtr) {
+  switch (ID) {
+  case Intrinsic::amdgcn_workitem_id_x:
+    NonKernelOnly = true;
+    return "amdgpu-work-item-id-x";
+  case Intrinsic::amdgcn_workgroup_id_x:
+    NonKernelOnly = true;
+    return "amdgpu-work-group-id-x";
+  case Intrinsic::amdgcn_workitem_id_y:
+  case Intrinsic::r600_read_tidig_y:
+    return "amdgpu-work-item-id-y";
+  case Intrinsic::amdgcn_workitem_id_z:
+  case Intrinsic::r600_read_tidig_z:
+    return "amdgpu-work-item-id-z";
+  case Intrinsic::amdgcn_workgroup_id_y:
+  case Intrinsic::r600_read_tgid_y:
+    return "amdgpu-work-group-id-y";
+  case Intrinsic::amdgcn_workgroup_id_z:
+  case Intrinsic::r600_read_tgid_z:
+    return "amdgpu-work-group-id-z";
+  case Intrinsic::amdgcn_dispatch_ptr:
+    return "amdgpu-dispatch-ptr";
+  case Intrinsic::amdgcn_dispatch_id:
+    return "amdgpu-dispatch-id";
+  case Intrinsic::amdgcn_kernarg_segment_ptr:
+  case Intrinsic::amdgcn_implicitarg_ptr:
+    return "amdgpu-kernarg-segment-ptr";
+  case Intrinsic::amdgcn_queue_ptr:
+  case Intrinsic::trap:
+  case Intrinsic::debugtrap:
+    IsQueuePtr = true;
+    return "amdgpu-queue-ptr";
+  default:
+    return "";
+  }
+}
+
+static bool handleAttr(Function &Parent, const Function &Callee,
+                       StringRef Name) {
+  if (Callee.hasFnAttribute(Name)) {
+    Parent.addFnAttr(Name);
+    return true;
+  }
+
+  return false;
+}
+
+static void copyFeaturesToFunction(Function &Parent, const Function &Callee,
+                                   bool &NeedQueuePtr) {
+  // X ids unnecessarily propagated to kernels.
+  static const StringRef AttrNames[] = {
+    { "amdgpu-work-item-id-x" },
+    { "amdgpu-work-item-id-y" },
+    { "amdgpu-work-item-id-z" },
+    { "amdgpu-work-group-id-x" },
+    { "amdgpu-work-group-id-y" },
+    { "amdgpu-work-group-id-z" },
+    { "amdgpu-dispatch-ptr" },
+    { "amdgpu-dispatch-id" },
+    { "amdgpu-kernarg-segment-ptr" }
+  };
+
+  if (handleAttr(Parent, Callee, "amdgpu-queue-ptr"))
+    NeedQueuePtr = true;
+
+  for (StringRef AttrName : AttrNames)
+    handleAttr(Parent, Callee, AttrName);
+}
+
+bool AMDGPUAnnotateKernelFeatures::addFeatureAttributes(Function &F) {
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>(F);
+  bool HasFlat = ST.hasFlatAddressSpace();
+  bool HasApertureRegs = ST.hasApertureRegs();
   SmallPtrSet<const Constant *, 8> ConstantExprVisited;
 
-  for (const BasicBlock &BB : F) {
-    for (const Instruction &I : BB) {
+  bool Changed = false;
+  bool NeedQueuePtr = false;
+  bool HaveCall = false;
+  bool IsFunc = !AMDGPU::isEntryFunctionCC(F.getCallingConv());
+
+  for (BasicBlock &BB : F) {
+    for (Instruction &I : BB) {
+      CallSite CS(&I);
+      if (CS) {
+        Function *Callee = CS.getCalledFunction();
+
+        // TODO: Do something with indirect calls.
+        if (!Callee) {
+          if (!CS.isInlineAsm())
+            HaveCall = true;
+          continue;
+        }
+
+        Intrinsic::ID IID = Callee->getIntrinsicID();
+        if (IID == Intrinsic::not_intrinsic) {
+          HaveCall = true;
+          copyFeaturesToFunction(F, *Callee, NeedQueuePtr);
+          Changed = true;
+        } else {
+          bool NonKernelOnly = false;
+          StringRef AttrName = intrinsicToAttrName(IID,
+                                                   NonKernelOnly, NeedQueuePtr);
+          if (!AttrName.empty() && (IsFunc || !NonKernelOnly)) {
+            F.addFnAttr(AttrName);
+            Changed = true;
+          }
+        }
+      }
+
+      if (NeedQueuePtr || HasApertureRegs)
+        continue;
+
       if (const AddrSpaceCastInst *ASC = dyn_cast<AddrSpaceCastInst>(&I)) {
-        if (castRequiresQueuePtr(ASC, AS))
-          return true;
+        if (castRequiresQueuePtr(ASC, AS)) {
+          NeedQueuePtr = true;
+          continue;
+        }
       }
 
       for (const Use &U : I.operands()) {
@@ -138,100 +255,57 @@ bool AMDGPUAnnotateKernelFeatures::hasAddrSpaceCast(const Function &F,
         if (!OpC)
           continue;
 
-        if (visitConstantExprsRecursively(OpC, ConstantExprVisited, AS))
-          return true;
+        if (visitConstantExprsRecursively(OpC, ConstantExprVisited, AS)) {
+          NeedQueuePtr = true;
+          break;
+        }
       }
     }
   }
 
-  return false;
-}
-
-void AMDGPUAnnotateKernelFeatures::addAttrToCallers(Function *Intrin,
-                                                    StringRef AttrName) {
-  SmallPtrSet<Function *, 4> SeenFuncs;
-
-  for (User *U : Intrin->users()) {
-    // CallInst is the only valid user for an intrinsic.
-    CallInst *CI = cast<CallInst>(U);
-
-    Function *CallingFunction = CI->getParent()->getParent();
-    if (SeenFuncs.insert(CallingFunction).second)
-      CallingFunction->addFnAttr(AttrName);
+  if (NeedQueuePtr) {
+    F.addFnAttr("amdgpu-queue-ptr");
+    Changed = true;
   }
-}
-
-bool AMDGPUAnnotateKernelFeatures::addAttrsForIntrinsics(
-  Module &M,
-  ArrayRef<StringRef[2]> IntrinsicToAttr) {
-  bool Changed = false;
 
-  for (const StringRef *Arr  : IntrinsicToAttr) {
-    if (Function *Fn = M.getFunction(Arr[0])) {
-      addAttrToCallers(Fn, Arr[1]);
-      Changed = true;
-    }
+  // TODO: We could refine this to captured pointers that could possibly be
+  // accessed by flat instructions. For now this is mostly a poor way of
+  // estimating whether there are calls before argument lowering.
+  if (HasFlat && !IsFunc && HaveCall) {
+    F.addFnAttr("amdgpu-flat-scratch");
+    Changed = true;
   }
 
   return Changed;
 }
 
-bool AMDGPUAnnotateKernelFeatures::runOnModule(Module &M) {
+bool AMDGPUAnnotateKernelFeatures::runOnSCC(CallGraphSCC &SCC) {
+  Module &M = SCC.getCallGraph().getModule();
   Triple TT(M.getTargetTriple());
-  AS = AMDGPU::getAMDGPUAS(M);
-
-  static const StringRef IntrinsicToAttr[][2] = {
-    // .x omitted
-    { "llvm.amdgcn.workitem.id.y", "amdgpu-work-item-id-y" },
-    { "llvm.amdgcn.workitem.id.z", "amdgpu-work-item-id-z" },
-
-    { "llvm.amdgcn.workgroup.id.y", "amdgpu-work-group-id-y" },
-    { "llvm.amdgcn.workgroup.id.z", "amdgpu-work-group-id-z" },
-
-    { "llvm.r600.read.tgid.y", "amdgpu-work-group-id-y" },
-    { "llvm.r600.read.tgid.z", "amdgpu-work-group-id-z" },
-
-    // .x omitted
-    { "llvm.r600.read.tidig.y", "amdgpu-work-item-id-y" },
-    { "llvm.r600.read.tidig.z", "amdgpu-work-item-id-z" }
-  };
 
-  static const StringRef HSAIntrinsicToAttr[][2] = {
-    { "llvm.amdgcn.dispatch.ptr", "amdgpu-dispatch-ptr" },
-    { "llvm.amdgcn.queue.ptr", "amdgpu-queue-ptr" },
-    { "llvm.amdgcn.dispatch.id", "amdgpu-dispatch-id" },
-    { "llvm.trap", "amdgpu-queue-ptr" },
-    { "llvm.debugtrap", "amdgpu-queue-ptr" }
-  };
-
-  // TODO: We should not add the attributes if the known compile time workgroup
-  // size is 1 for y/z.
-
-  // TODO: Intrinsics that require queue ptr.
+  bool Changed = false;
+  for (CallGraphNode *I : SCC) {
+    Function *F = I->getFunction();
+    if (!F || F->isDeclaration())
+      continue;
 
-  // We do not need to note the x workitem or workgroup id because they are
-  // always initialized.
+    Changed |= addFeatureAttributes(*F);
+  }
 
-  bool Changed = addAttrsForIntrinsics(M, IntrinsicToAttr);
-  if (TT.getOS() == Triple::AMDHSA || TT.getOS() == Triple::Mesa3D) {
-    Changed |= addAttrsForIntrinsics(M, HSAIntrinsicToAttr);
 
-    for (Function &F : M) {
-      if (F.hasFnAttribute("amdgpu-queue-ptr"))
-        continue;
+  return Changed;
+}
 
-      auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
-      bool HasApertureRegs = TPC && TPC->getTM<TargetMachine>()
-                                        .getSubtarget<AMDGPUSubtarget>(F)
-                                        .hasApertureRegs();
-      if (!HasApertureRegs && hasAddrSpaceCast(F, AS))
-        F.addFnAttr("amdgpu-queue-ptr");
-    }
-  }
+bool AMDGPUAnnotateKernelFeatures::doInitialization(CallGraph &CG) {
+  auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
+  if (!TPC)
+    report_fatal_error("TargetMachine is required");
 
-  return Changed;
+  AS = AMDGPU::getAMDGPUAS(CG.getModule());
+  TM = &TPC->getTM<TargetMachine>();
+  return false;
 }
 
-ModulePass *llvm::createAMDGPUAnnotateKernelFeaturesPass() {
+Pass *llvm::createAMDGPUAnnotateKernelFeaturesPass() {
   return new AMDGPUAnnotateKernelFeatures();
 }
diff --git a/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp b/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
index 83ad1a5c6ee3..2247814cfe55 100644
--- a/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ b/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -268,20 +268,11 @@ bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
                                  CurrentProgramInfo.ScratchSize,
                                  getFunctionCodeSize(MF));
 
-      OutStreamer->emitRawComment(" codeLenInByte = " +
-                                  Twine(getFunctionCodeSize(MF)), false);
-      OutStreamer->emitRawComment(
-        " NumSgprs: " + Twine(CurrentProgramInfo.NumSGPR), false);
-      OutStreamer->emitRawComment(
-        " NumVgprs: " + Twine(CurrentProgramInfo.NumVGPR), false);
-
       OutStreamer->emitRawComment(
         " FloatMode: " + Twine(CurrentProgramInfo.FloatMode), false);
       OutStreamer->emitRawComment(
         " IeeeMode: " + Twine(CurrentProgramInfo.IEEEMode), false);
       OutStreamer->emitRawComment(
-        " ScratchSize: " + Twine(CurrentProgramInfo.ScratchSize), false);
-      OutStreamer->emitRawComment(
         " LDSByteSize: " + Twine(CurrentProgramInfo.LDSSize) +
         " bytes/workgroup (compile time only)", false);
 
diff --git a/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index 2553cf4da0fe..258b1737deb3 100644
--- a/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -573,6 +573,8 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(const TargetMachine &TM,
   setTargetDAGCombine(ISD::FSUB);
   setTargetDAGCombine(ISD::FNEG);
   setTargetDAGCombine(ISD::FABS);
+  setTargetDAGCombine(ISD::AssertZext);
+  setTargetDAGCombine(ISD::AssertSext);
 }
 
 //===----------------------------------------------------------------------===//
@@ -883,7 +885,7 @@ CCAssignFn *AMDGPUCallLowering::CCAssignFnForReturn(CallingConv::ID CC,
 
 /// When the SelectionDAGBuilder computes the Ins, it takes care of splitting
 /// input values across multiple registers.  Each item in the Ins array
-/// represents a single value that will be stored in regsters.  Ins[x].VT is
+/// represents a single value that will be stored in registers.  Ins[x].VT is
 /// the value type of the value that will be stored in the register, so
 /// whatever SDNode we lower the argument to needs to be this type.
 ///
@@ -2591,6 +2593,31 @@ SDValue AMDGPUTargetLowering::performClampCombine(SDNode *N,
   return SDValue(CSrc, 0);
 }
 
+// FIXME: This should go in generic DAG combiner with an isTruncateFree check,
+// but isTruncateFree is inaccurate for i16 now because of SALU vs. VALU
+// issues.
+SDValue AMDGPUTargetLowering::performAssertSZExtCombine(SDNode *N,
+                                                        DAGCombinerInfo &DCI) const {
+  SelectionDAG &DAG = DCI.DAG;
+  SDValue N0 = N->getOperand(0);
+
+  // (vt2 (assertzext (truncate vt0:x), vt1)) ->
+  //     (vt2 (truncate (assertzext vt0:x, vt1)))
+  if (N0.getOpcode() == ISD::TRUNCATE) {
+    SDValue N1 = N->getOperand(1);
+    EVT ExtVT = cast<VTSDNode>(N1)->getVT();
+    SDLoc SL(N);
+
+    SDValue Src = N0.getOperand(0);
+    EVT SrcVT = Src.getValueType();
+    if (SrcVT.bitsGE(ExtVT)) {
+      SDValue NewInReg = DAG.getNode(N->getOpcode(), SL, SrcVT, Src, N1);
+      return DAG.getNode(ISD::TRUNCATE, SL, N->getValueType(0), NewInReg);
+    }
+  }
+
+  return SDValue();
+}
 /// Split the 64-bit value \p LHS into two 32-bit components, and perform the
 /// binary operation \p Opc to it with the corresponding constant operands.
 SDValue AMDGPUTargetLowering::splitBinaryBitConstantOpImpl(
@@ -3521,6 +3548,9 @@ SDValue AMDGPUTargetLowering::PerformDAGCombine(SDNode *N,
 
     break;
   }
+  case ISD::AssertZext:
+  case ISD::AssertSext:
+    return performAssertSZExtCombine(N, DCI);
   }
   return SDValue();
 }
diff --git a/lib/Target/AMDGPU/AMDGPUISelLowering.h b/lib/Target/AMDGPU/AMDGPUISelLowering.h
index a45234e2b39f..d85aada6053a 100644
--- a/lib/Target/AMDGPU/AMDGPUISelLowering.h
+++ b/lib/Target/AMDGPU/AMDGPUISelLowering.h
@@ -76,6 +76,7 @@ protected:
   SDValue performLoadCombine(SDNode *N, DAGCombinerInfo &DCI) const;
   SDValue performStoreCombine(SDNode *N, DAGCombinerInfo &DCI) const;
   SDValue performClampCombine(SDNode *N, DAGCombinerInfo &DCI) const;
+  SDValue performAssertSZExtCombine(SDNode *N, DAGCombinerInfo &DCI) const;
 
   SDValue splitBinaryBitConstantOpImpl(DAGCombinerInfo &DCI, const SDLoc &SL,
                                        unsigned Opc, SDValue LHS,
diff --git a/lib/Target/AMDGPU/AMDGPUSubtarget.cpp b/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
index 1bc5a52053ec..779617629010 100644
--- a/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
+++ b/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
@@ -277,7 +277,7 @@ std::pair<unsigned, unsigned> AMDGPUSubtarget::getWavesPerEU(
   // Make sure requested values are compatible with values implied by requested
   // minimum/maximum flat work group sizes.
   if (RequestedFlatWorkGroupSize &&
-      Requested.first > MinImpliedByFlatWorkGroupSize)
+      Requested.first < MinImpliedByFlatWorkGroupSize)
     return Default;
 
   return Requested;
diff --git a/lib/Target/AMDGPU/AMDGPUSubtarget.h b/lib/Target/AMDGPU/AMDGPUSubtarget.h
index 22cede59086a..d4b6a5fe8020 100644
--- a/lib/Target/AMDGPU/AMDGPUSubtarget.h
+++ b/lib/Target/AMDGPU/AMDGPUSubtarget.h
@@ -359,6 +359,10 @@ public:
     return FP64FP16Denormals;
   }
 
+  bool supportsMinMaxDenormModes() const {
+    return getGeneration() >= AMDGPUSubtarget::GFX9;
+  }
+
   bool hasFPExceptions() const {
     return FPExceptions;
   }
diff --git a/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp b/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
index e3c90f250600..b37c274102bc 100644
--- a/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
+++ b/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
@@ -1208,7 +1208,7 @@ bool AMDGPUOperand::isInlinableImm(MVT type) const {
 }
 
 bool AMDGPUOperand::isLiteralImm(MVT type) const {
-  // Check that this imediate can be added as literal
+  // Check that this immediate can be added as literal
   if (!isImmTy(ImmTyNone)) {
     return false;
   }
diff --git a/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp b/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
index f26e49295e69..966c6fec20c6 100644
--- a/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
+++ b/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
@@ -87,6 +87,7 @@ DECODE_OPERAND(Decode##RegClass##RegisterClass, decodeOperand_##RegClass)
 DECODE_OPERAND_REG(VGPR_32)
 DECODE_OPERAND_REG(VS_32)
 DECODE_OPERAND_REG(VS_64)
+DECODE_OPERAND_REG(VS_128)
 
 DECODE_OPERAND_REG(VReg_64)
 DECODE_OPERAND_REG(VReg_96)
@@ -318,6 +319,10 @@ MCOperand AMDGPUDisassembler::decodeOperand_VS_64(unsigned Val) const {
   return decodeSrcOp(OPW64, Val);
 }
 
+MCOperand AMDGPUDisassembler::decodeOperand_VS_128(unsigned Val) const {
+  return decodeSrcOp(OPW128, Val);
+}
+
 MCOperand AMDGPUDisassembler::decodeOperand_VSrc16(unsigned Val) const {
   return decodeSrcOp(OPW16, Val);
 }
diff --git a/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h b/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h
index 3d71db909e20..4c755be09999 100644
--- a/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h
+++ b/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h
@@ -70,6 +70,7 @@ public:
   MCOperand decodeOperand_VGPR_32(unsigned Val) const;
   MCOperand decodeOperand_VS_32(unsigned Val) const;
   MCOperand decodeOperand_VS_64(unsigned Val) const;
+  MCOperand decodeOperand_VS_128(unsigned Val) const;
   MCOperand decodeOperand_VSrc16(unsigned Val) const;
   MCOperand decodeOperand_VSrcV216(unsigned Val) const;
 
diff --git a/lib/Target/AMDGPU/SIFoldOperands.cpp b/lib/Target/AMDGPU/SIFoldOperands.cpp
index 3af242d9ea66..0aad8f0843d6 100644
--- a/lib/Target/AMDGPU/SIFoldOperands.cpp
+++ b/lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -653,6 +653,7 @@ void SIFoldOperands::foldInstOperand(MachineInstr &MI,
         // again. The same constant folded instruction could also have a second
         // use operand.
         NextUse = MRI->use_begin(Dst.getReg());
+        FoldList.clear();
         continue;
       }
 
diff --git a/lib/Target/AMDGPU/SIFrameLowering.cpp b/lib/Target/AMDGPU/SIFrameLowering.cpp
index 08a64de38501..7334781916d8 100644
--- a/lib/Target/AMDGPU/SIFrameLowering.cpp
+++ b/lib/Target/AMDGPU/SIFrameLowering.cpp
@@ -158,7 +158,7 @@ SIFrameLowering::getReservedPrivateSegmentWaveByteOffsetReg(
   // No replacement necessary.
   if (ScratchWaveOffsetReg == AMDGPU::NoRegister ||
       !MRI.isPhysRegUsed(ScratchWaveOffsetReg)) {
-    assert(MFI->getStackPtrOffsetReg() == AMDGPU::NoRegister);
+    assert(MFI->getStackPtrOffsetReg() == AMDGPU::SP_REG);
     return std::make_pair(AMDGPU::NoRegister, AMDGPU::NoRegister);
   }
 
@@ -246,13 +246,16 @@ void SIFrameLowering::emitEntryFunctionPrologue(MachineFunction &MF,
   // this point it appears we need the setup. This part of the prolog should be
   // emitted after frame indices are eliminated.
 
-  if (MF.getFrameInfo().hasStackObjects() && MFI->hasFlatScratchInit())
+  if (MFI->hasFlatScratchInit())
     emitFlatScratchInit(ST, MF, MBB);
 
   unsigned SPReg = MFI->getStackPtrOffsetReg();
-  if (SPReg != AMDGPU::NoRegister) {
+  if (SPReg != AMDGPU::SP_REG) {
+    assert(MRI.isReserved(SPReg) && "SPReg used but not reserved");
+
     DebugLoc DL;
-    int64_t StackSize = MF.getFrameInfo().getStackSize();
+    const MachineFrameInfo &FrameInfo = MF.getFrameInfo();
+    int64_t StackSize = FrameInfo.getStackSize();
 
     if (StackSize == 0) {
       BuildMI(MBB, MBB.begin(), DL, TII->get(AMDGPU::COPY), SPReg)
diff --git a/lib/Target/AMDGPU/SIISelLowering.cpp b/lib/Target/AMDGPU/SIISelLowering.cpp
index 2ba570b9ebbb..2356405f0919 100644
--- a/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -1171,8 +1171,7 @@ static void allocateSystemSGPRs(CCState &CCInfo,
 static void reservePrivateMemoryRegs(const TargetMachine &TM,
                                      MachineFunction &MF,
                                      const SIRegisterInfo &TRI,
-                                     SIMachineFunctionInfo &Info,
-                                     bool NeedSP) {
+                                     SIMachineFunctionInfo &Info) {
   // Now that we've figured out where the scratch register inputs are, see if
   // should reserve the arguments and use them directly.
   MachineFrameInfo &MFI = MF.getFrameInfo();
@@ -1234,15 +1233,6 @@ static void reservePrivateMemoryRegs(const TargetMachine &TM,
       Info.setScratchWaveOffsetReg(ReservedOffsetReg);
     }
   }
-
-  if (NeedSP) {
-    unsigned ReservedStackPtrOffsetReg = TRI.reservedStackPtrOffsetReg(MF);
-    Info.setStackPtrOffsetReg(ReservedStackPtrOffsetReg);
-
-    assert(Info.getStackPtrOffsetReg() != Info.getFrameOffsetReg());
-    assert(!TRI.isSubRegister(Info.getScratchRSrcReg(),
-                              Info.getStackPtrOffsetReg()));
-  }
 }
 
 SDValue SITargetLowering::LowerFormalArguments(
@@ -1380,10 +1370,37 @@ SDValue SITargetLowering::LowerFormalArguments(
 
     unsigned Reg = VA.getLocReg();
     const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg, VT);
+    EVT ValVT = VA.getValVT();
 
     Reg = MF.addLiveIn(Reg, RC);
     SDValue Val = DAG.getCopyFromReg(Chain, DL, Reg, VT);
 
+    // If this is an 8 or 16-bit value, it is really passed promoted
+    // to 32 bits. Insert an assert[sz]ext to capture this, then
+    // truncate to the right size.
+    switch (VA.getLocInfo()) {
+    case CCValAssign::Full:
+      break;
+    case CCValAssign::BCvt:
+      Val = DAG.getNode(ISD::BITCAST, DL, ValVT, Val);
+      break;
+    case CCValAssign::SExt:
+      Val = DAG.getNode(ISD::AssertSext, DL, VT, Val,
+                        DAG.getValueType(ValVT));
+      Val = DAG.getNode(ISD::TRUNCATE, DL, ValVT, Val);
+      break;
+    case CCValAssign::ZExt:
+      Val = DAG.getNode(ISD::AssertZext, DL, VT, Val,
+                        DAG.getValueType(ValVT));
+      Val = DAG.getNode(ISD::TRUNCATE, DL, ValVT, Val);
+      break;
+    case CCValAssign::AExt:
+      Val = DAG.getNode(ISD::TRUNCATE, DL, ValVT, Val);
+      break;
+    default:
+      llvm_unreachable("Unknown loc info!");
+    }
+
     if (IsShader && Arg.VT.isVector()) {
       // Build a vector from the registers
       Type *ParamType = FType->getParamType(Arg.getOrigArgIndex());
@@ -1410,25 +1427,13 @@ SDValue SITargetLowering::LowerFormalArguments(
     InVals.push_back(Val);
   }
 
-  const MachineFrameInfo &FrameInfo = MF.getFrameInfo();
-
-  // TODO: Could maybe omit SP if only tail calls?
-  bool NeedSP = FrameInfo.hasCalls() || FrameInfo.hasVarSizedObjects();
-
   // Start adding system SGPRs.
   if (IsEntryFunc) {
     allocateSystemSGPRs(CCInfo, MF, *Info, CallConv, IsShader);
-    reservePrivateMemoryRegs(getTargetMachine(), MF, *TRI, *Info, NeedSP);
   } else {
     CCInfo.AllocateReg(Info->getScratchRSrcReg());
     CCInfo.AllocateReg(Info->getScratchWaveOffsetReg());
     CCInfo.AllocateReg(Info->getFrameOffsetReg());
-
-    if (NeedSP) {
-      unsigned StackPtrReg = findFirstFreeSGPR(CCInfo);
-      CCInfo.AllocateReg(StackPtrReg);
-      Info->setStackPtrOffsetReg(StackPtrReg);
-    }
   }
 
   return Chains.empty() ? Chain :
@@ -4624,8 +4629,8 @@ static bool isKnownNeverSNan(SelectionDAG &DAG, SDValue Op) {
   return DAG.isKnownNeverNaN(Op);
 }
 
-static bool isCanonicalized(SDValue Op, const SISubtarget *ST,
-                            unsigned MaxDepth=5) {
+static bool isCanonicalized(SelectionDAG &DAG, SDValue Op,
+                            const SISubtarget *ST, unsigned MaxDepth=5) {
   // If source is a result of another standard FP operation it is already in
   // canonical form.
 
@@ -4663,7 +4668,7 @@ static bool isCanonicalized(SDValue Op, const SISubtarget *ST,
   case ISD::FNEG:
   case ISD::FABS:
     return (MaxDepth > 0) &&
-           isCanonicalized(Op.getOperand(0), ST, MaxDepth - 1);
+           isCanonicalized(DAG, Op.getOperand(0), ST, MaxDepth - 1);
 
   case ISD::FSIN:
   case ISD::FCOS:
@@ -4672,16 +4677,19 @@ static bool isCanonicalized(SDValue Op, const SISubtarget *ST,
 
   // In pre-GFX9 targets V_MIN_F32 and others do not flush denorms.
   // For such targets need to check their input recursively.
-  // TODO: on GFX9+ we could return true without checking provided no-nan
-  // mode, since canonicalization is also used to quiet sNaNs.
   case ISD::FMINNUM:
   case ISD::FMAXNUM:
   case ISD::FMINNAN:
   case ISD::FMAXNAN:
 
+    if (ST->supportsMinMaxDenormModes() &&
+        DAG.isKnownNeverNaN(Op.getOperand(0)) &&
+        DAG.isKnownNeverNaN(Op.getOperand(1)))
+      return true;
+
     return (MaxDepth > 0) &&
-           isCanonicalized(Op.getOperand(0), ST, MaxDepth - 1) &&
-           isCanonicalized(Op.getOperand(1), ST, MaxDepth - 1);
+           isCanonicalized(DAG, Op.getOperand(0), ST, MaxDepth - 1) &&
+           isCanonicalized(DAG, Op.getOperand(1), ST, MaxDepth - 1);
 
   case ISD::ConstantFP: {
     auto F = cast<ConstantFPSDNode>(Op)->getValueAPF();
@@ -4700,11 +4708,19 @@ SDValue SITargetLowering::performFCanonicalizeCombine(
 
   if (!CFP) {
     SDValue N0 = N->getOperand(0);
+    EVT VT = N0.getValueType().getScalarType();
+    auto ST = getSubtarget();
+
+    if (((VT == MVT::f32 && ST->hasFP32Denormals()) ||
+         (VT == MVT::f64 && ST->hasFP64Denormals()) ||
+         (VT == MVT::f16 && ST->hasFP16Denormals())) &&
+        DAG.isKnownNeverNaN(N0))
+      return N0;
 
     bool IsIEEEMode = Subtarget->enableIEEEBit(DAG.getMachineFunction());
 
     if ((IsIEEEMode || isKnownNeverSNan(DAG, N0)) &&
-        isCanonicalized(N0, getSubtarget()))
+        isCanonicalized(DAG, N0, ST))
       return N0;
 
     return SDValue();
@@ -5813,3 +5829,44 @@ SITargetLowering::getConstraintType(StringRef Constraint) const {
   }
   return TargetLowering::getConstraintType(Constraint);
 }
+
+// Figure out which registers should be reserved for stack access. Only after
+// the function is legalized do we know all of the non-spill stack objects or if
+// calls are present.
+void SITargetLowering::finalizeLowering(MachineFunction &MF) const {
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+  SIMachineFunctionInfo *Info = MF.getInfo<SIMachineFunctionInfo>();
+  const MachineFrameInfo &MFI = MF.getFrameInfo();
+  const SISubtarget &ST = MF.getSubtarget<SISubtarget>();
+  const SIRegisterInfo *TRI = ST.getRegisterInfo();
+
+  if (Info->isEntryFunction()) {
+    // Callable functions have fixed registers used for stack access.
+    reservePrivateMemoryRegs(getTargetMachine(), MF, *TRI, *Info);
+  }
+
+  // We have to assume the SP is needed in case there are calls in the function
+  // during lowering. Calls are only detected after the function is
+  // lowered. We're about to reserve registers, so don't bother using it if we
+  // aren't really going to use it.
+  bool NeedSP = !Info->isEntryFunction() ||
+    MFI.hasVarSizedObjects() ||
+    MFI.hasCalls();
+
+  if (NeedSP) {
+    unsigned ReservedStackPtrOffsetReg = TRI->reservedStackPtrOffsetReg(MF);
+    Info->setStackPtrOffsetReg(ReservedStackPtrOffsetReg);
+
+    assert(Info->getStackPtrOffsetReg() != Info->getFrameOffsetReg());
+    assert(!TRI->isSubRegister(Info->getScratchRSrcReg(),
+                               Info->getStackPtrOffsetReg()));
+    MRI.replaceRegWith(AMDGPU::SP_REG, Info->getStackPtrOffsetReg());
+  }
+
+  MRI.replaceRegWith(AMDGPU::PRIVATE_RSRC_REG, Info->getScratchRSrcReg());
+  MRI.replaceRegWith(AMDGPU::FP_REG, Info->getFrameOffsetReg());
+  MRI.replaceRegWith(AMDGPU::SCRATCH_WAVE_OFFSET_REG,
+                     Info->getScratchWaveOffsetReg());
+
+  TargetLoweringBase::finalizeLowering(MF);
+}
diff --git a/lib/Target/AMDGPU/SIISelLowering.h b/lib/Target/AMDGPU/SIISelLowering.h
index 83392a7ab1b2..e6bb3d6cd419 100644
--- a/lib/Target/AMDGPU/SIISelLowering.h
+++ b/lib/Target/AMDGPU/SIISelLowering.h
@@ -232,6 +232,8 @@ public:
   ConstraintType getConstraintType(StringRef Constraint) const override;
   SDValue copyToM0(SelectionDAG &DAG, SDValue Chain, const SDLoc &DL,
                    SDValue V) const;
+
+  void finalizeLowering(MachineFunction &MF) const override;
 };
 
 } // End namespace llvm
diff --git a/lib/Target/AMDGPU/SIInstrInfo.cpp b/lib/Target/AMDGPU/SIInstrInfo.cpp
index 160f8837d49c..a7e0feb10b9f 100644
--- a/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -3408,8 +3408,8 @@ void SIInstrInfo::legalizeOperands(MachineInstr &MI) const {
 }
 
 void SIInstrInfo::moveToVALU(MachineInstr &TopInst) const {
-  SmallVector<MachineInstr *, 128> Worklist;
-  Worklist.push_back(&TopInst);
+  SetVectorType Worklist;
+  Worklist.insert(&TopInst);
 
   while (!Worklist.empty()) {
     MachineInstr &Inst = *Worklist.pop_back_val();
@@ -3610,7 +3610,7 @@ void SIInstrInfo::moveToVALU(MachineInstr &TopInst) const {
   }
 }
 
-void SIInstrInfo::lowerScalarAbs(SmallVectorImpl<MachineInstr *> &Worklist,
+void SIInstrInfo::lowerScalarAbs(SetVectorType &Worklist,
                                  MachineInstr &Inst) const {
   MachineBasicBlock &MBB = *Inst.getParent();
   MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
@@ -3635,7 +3635,7 @@ void SIInstrInfo::lowerScalarAbs(SmallVectorImpl<MachineInstr *> &Worklist,
 }
 
 void SIInstrInfo::splitScalar64BitUnaryOp(
-    SmallVectorImpl<MachineInstr *> &Worklist, MachineInstr &Inst,
+    SetVectorType &Worklist, MachineInstr &Inst,
     unsigned Opcode) const {
   MachineBasicBlock &MBB = *Inst.getParent();
   MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
@@ -3686,7 +3686,7 @@ void SIInstrInfo::splitScalar64BitUnaryOp(
 }
 
 void SIInstrInfo::splitScalar64BitBinaryOp(
-    SmallVectorImpl<MachineInstr *> &Worklist, MachineInstr &Inst,
+    SetVectorType &Worklist, MachineInstr &Inst,
     unsigned Opcode) const {
   MachineBasicBlock &MBB = *Inst.getParent();
   MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
@@ -3753,7 +3753,7 @@ void SIInstrInfo::splitScalar64BitBinaryOp(
 }
 
 void SIInstrInfo::splitScalar64BitBCNT(
-    SmallVectorImpl<MachineInstr *> &Worklist, MachineInstr &Inst) const {
+    SetVectorType &Worklist, MachineInstr &Inst) const {
   MachineBasicBlock &MBB = *Inst.getParent();
   MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
 
@@ -3789,7 +3789,7 @@ void SIInstrInfo::splitScalar64BitBCNT(
   addUsersToMoveToVALUWorklist(ResultReg, MRI, Worklist);
 }
 
-void SIInstrInfo::splitScalar64BitBFE(SmallVectorImpl<MachineInstr *> &Worklist,
+void SIInstrInfo::splitScalar64BitBFE(SetVectorType &Worklist,
                                       MachineInstr &Inst) const {
   MachineBasicBlock &MBB = *Inst.getParent();
   MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
@@ -3853,12 +3853,12 @@ void SIInstrInfo::splitScalar64BitBFE(SmallVectorImpl<MachineInstr *> &Worklist,
 void SIInstrInfo::addUsersToMoveToVALUWorklist(
   unsigned DstReg,
   MachineRegisterInfo &MRI,
-  SmallVectorImpl<MachineInstr *> &Worklist) const {
+  SetVectorType &Worklist) const {
   for (MachineRegisterInfo::use_iterator I = MRI.use_begin(DstReg),
          E = MRI.use_end(); I != E;) {
     MachineInstr &UseMI = *I->getParent();
     if (!canReadVGPR(UseMI, I.getOperandNo())) {
-      Worklist.push_back(&UseMI);
+      Worklist.insert(&UseMI);
 
       do {
         ++I;
@@ -3869,7 +3869,7 @@ void SIInstrInfo::addUsersToMoveToVALUWorklist(
   }
 }
 
-void SIInstrInfo::movePackToVALU(SmallVectorImpl<MachineInstr *> &Worklist,
+void SIInstrInfo::movePackToVALU(SetVectorType &Worklist,
                                  MachineRegisterInfo &MRI,
                                  MachineInstr &Inst) const {
   unsigned ResultReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
@@ -3932,7 +3932,7 @@ void SIInstrInfo::movePackToVALU(SmallVectorImpl<MachineInstr *> &Worklist,
 }
 
 void SIInstrInfo::addSCCDefUsersToVALUWorklist(
-    MachineInstr &SCCDefInst, SmallVectorImpl<MachineInstr *> &Worklist) const {
+    MachineInstr &SCCDefInst, SetVectorType &Worklist) const {
   // This assumes that all the users of SCC are in the same block
   // as the SCC def.
   for (MachineInstr &MI :
@@ -3943,7 +3943,7 @@ void SIInstrInfo::addSCCDefUsersToVALUWorklist(
       return;
 
     if (MI.findRegisterUseOperandIdx(AMDGPU::SCC) != -1)
-      Worklist.push_back(&MI);
+      Worklist.insert(&MI);
   }
 }
 
diff --git a/lib/Target/AMDGPU/SIInstrInfo.h b/lib/Target/AMDGPU/SIInstrInfo.h
index d00c0d4a7f4e..3dd5bc89e6c7 100644
--- a/lib/Target/AMDGPU/SIInstrInfo.h
+++ b/lib/Target/AMDGPU/SIInstrInfo.h
@@ -19,6 +19,7 @@
 #include "AMDGPUInstrInfo.h"
 #include "SIDefines.h"
 #include "SIRegisterInfo.h"
+#include "llvm/ADT/SetVector.h"
 
 namespace llvm {
 
@@ -38,6 +39,8 @@ private:
     EXECZ = 3
   };
 
+  typedef SmallSetVector<MachineInstr *, 32> SetVectorType;
+
   static unsigned getBranchOpcode(BranchPredicate Cond);
   static BranchPredicate getBranchPredicate(unsigned Opcode);
 
@@ -56,30 +59,30 @@ private:
 
   void swapOperands(MachineInstr &Inst) const;
 
-  void lowerScalarAbs(SmallVectorImpl<MachineInstr *> &Worklist,
+  void lowerScalarAbs(SetVectorType &Worklist,
                       MachineInstr &Inst) const;
 
-  void splitScalar64BitUnaryOp(SmallVectorImpl<MachineInstr *> &Worklist,
+  void splitScalar64BitUnaryOp(SetVectorType &Worklist,
                                MachineInstr &Inst, unsigned Opcode) const;
 
-  void splitScalar64BitBinaryOp(SmallVectorImpl<MachineInstr *> &Worklist,
+  void splitScalar64BitBinaryOp(SetVectorType &Worklist,
                                 MachineInstr &Inst, unsigned Opcode) const;
 
-  void splitScalar64BitBCNT(SmallVectorImpl<MachineInstr *> &Worklist,
+  void splitScalar64BitBCNT(SetVectorType &Worklist,
                             MachineInstr &Inst) const;
-  void splitScalar64BitBFE(SmallVectorImpl<MachineInstr *> &Worklist,
+  void splitScalar64BitBFE(SetVectorType &Worklist,
                            MachineInstr &Inst) const;
-  void movePackToVALU(SmallVectorImpl<MachineInstr *> &Worklist,
+  void movePackToVALU(SetVectorType &Worklist,
                       MachineRegisterInfo &MRI,
                       MachineInstr &Inst) const;
 
   void addUsersToMoveToVALUWorklist(
     unsigned Reg, MachineRegisterInfo &MRI,
-    SmallVectorImpl<MachineInstr *> &Worklist) const;
+    SetVectorType &Worklist) const;
 
   void
   addSCCDefUsersToVALUWorklist(MachineInstr &SCCDefInst,
-                               SmallVectorImpl<MachineInstr *> &Worklist) const;
+                               SetVectorType &Worklist) const;
 
   const TargetRegisterClass *
   getDestEquivalentVGPRClass(const MachineInstr &Inst) const;
diff --git a/lib/Target/AMDGPU/SIInstrInfo.td b/lib/Target/AMDGPU/SIInstrInfo.td
index ffb01363e131..088173680fa8 100644
--- a/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/lib/Target/AMDGPU/SIInstrInfo.td
@@ -1436,7 +1436,7 @@ class VOPProfile <list<ValueType> _ArgVT> {
 
   field bit IsPacked = isPackedType<Src0VT>.ret;
   field bit HasOpSel = IsPacked;
-  field bit HasOMod = !if(HasOpSel, 0, HasModifiers);
+  field bit HasOMod = !if(HasOpSel, 0, isFloatType<DstVT>.ret);
   field bit HasSDWAOMod = isFloatType<DstVT>.ret;
 
   field bit HasExt = getHasExt<NumSrcArgs, DstVT, Src0VT, Src1VT>.ret;
diff --git a/lib/Target/AMDGPU/SIInstructions.td b/lib/Target/AMDGPU/SIInstructions.td
index bcc685015cf5..ba69e42d9125 100644
--- a/lib/Target/AMDGPU/SIInstructions.td
+++ b/lib/Target/AMDGPU/SIInstructions.td
@@ -1060,7 +1060,7 @@ def : Pat <
 
 class FPToI1Pat<Instruction Inst, int KOne, ValueType kone_type, ValueType vt, SDPatternOperator fp_to_int> : Pat <
   (i1 (fp_to_int (vt (VOP3Mods vt:$src0, i32:$src0_modifiers)))),
-  (i1 (Inst 0, (kone_type KOne), $src0_modifiers, $src0, DSTCLAMP.NONE, DSTOMOD.NONE))
+  (i1 (Inst 0, (kone_type KOne), $src0_modifiers, $src0, DSTCLAMP.NONE))
 >;
 
 def : FPToI1Pat<V_CMP_EQ_F32_e64, CONST.FP32_ONE, i32, f32, fp_to_uint>;
diff --git a/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp b/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
index 3203c38dae34..a7c8166ff6d2 100644
--- a/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
+++ b/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
@@ -23,10 +23,10 @@ using namespace llvm;
 SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
   : AMDGPUMachineFunction(MF),
     TIDReg(AMDGPU::NoRegister),
-    ScratchRSrcReg(AMDGPU::NoRegister),
-    ScratchWaveOffsetReg(AMDGPU::NoRegister),
-    FrameOffsetReg(AMDGPU::NoRegister),
-    StackPtrOffsetReg(AMDGPU::NoRegister),
+    ScratchRSrcReg(AMDGPU::PRIVATE_RSRC_REG),
+    ScratchWaveOffsetReg(AMDGPU::SCRATCH_WAVE_OFFSET_REG),
+    FrameOffsetReg(AMDGPU::FP_REG),
+    StackPtrOffsetReg(AMDGPU::SP_REG),
     PrivateSegmentBufferUserSGPR(AMDGPU::NoRegister),
     DispatchPtrUserSGPR(AMDGPU::NoRegister),
     QueuePtrUserSGPR(AMDGPU::NoRegister),
@@ -42,6 +42,9 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
     WorkGroupIDZSystemSGPR(AMDGPU::NoRegister),
     WorkGroupInfoSystemSGPR(AMDGPU::NoRegister),
     PrivateSegmentWaveByteOffsetSystemSGPR(AMDGPU::NoRegister),
+    WorkItemIDXVGPR(AMDGPU::NoRegister),
+    WorkItemIDYVGPR(AMDGPU::NoRegister),
+    WorkItemIDZVGPR(AMDGPU::NoRegister),
     PSInputAddr(0),
     PSInputEnable(0),
     ReturnsVoid(true),
@@ -87,12 +90,14 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
     ScratchWaveOffsetReg = AMDGPU::SGPR4;
     FrameOffsetReg = AMDGPU::SGPR5;
     StackPtrOffsetReg = AMDGPU::SGPR32;
-    return;
+
+    // FIXME: Not really a system SGPR.
+    PrivateSegmentWaveByteOffsetSystemSGPR = ScratchWaveOffsetReg;
   }
 
   CallingConv::ID CC = F->getCallingConv();
   if (CC == CallingConv::AMDGPU_KERNEL || CC == CallingConv::SPIR_KERNEL) {
-    KernargSegmentPtr = true;
+    KernargSegmentPtr = !F->arg_empty();
     WorkGroupIDX = true;
     WorkItemIDX = true;
   } else if (CC == CallingConv::AMDGPU_PS) {
@@ -101,17 +106,25 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
 
   if (ST.debuggerEmitPrologue()) {
     // Enable everything.
+    WorkGroupIDX = true;
     WorkGroupIDY = true;
     WorkGroupIDZ = true;
+    WorkItemIDX = true;
     WorkItemIDY = true;
     WorkItemIDZ = true;
   } else {
+    if (F->hasFnAttribute("amdgpu-work-group-id-x"))
+      WorkGroupIDX = true;
+
     if (F->hasFnAttribute("amdgpu-work-group-id-y"))
       WorkGroupIDY = true;
 
     if (F->hasFnAttribute("amdgpu-work-group-id-z"))
       WorkGroupIDZ = true;
 
+    if (F->hasFnAttribute("amdgpu-work-item-id-x"))
+      WorkItemIDX = true;
+
     if (F->hasFnAttribute("amdgpu-work-item-id-y"))
       WorkItemIDY = true;
 
@@ -119,25 +132,28 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
       WorkItemIDZ = true;
   }
 
-  // X, XY, and XYZ are the only supported combinations, so make sure Y is
-  // enabled if Z is.
-  if (WorkItemIDZ)
-    WorkItemIDY = true;
-
   const MachineFrameInfo &FrameInfo = MF.getFrameInfo();
   bool MaySpill = ST.isVGPRSpillingEnabled(*F);
-  bool HasStackObjects = FrameInfo.hasStackObjects() || FrameInfo.hasCalls();
+  bool HasStackObjects = FrameInfo.hasStackObjects();
+
+  if (isEntryFunction()) {
+    // X, XY, and XYZ are the only supported combinations, so make sure Y is
+    // enabled if Z is.
+    if (WorkItemIDZ)
+      WorkItemIDY = true;
 
-  if (HasStackObjects || MaySpill) {
-    PrivateSegmentWaveByteOffset = true;
+    if (HasStackObjects || MaySpill) {
+      PrivateSegmentWaveByteOffset = true;
 
-    // HS and GS always have the scratch wave offset in SGPR5 on GFX9.
-    if (ST.getGeneration() >= AMDGPUSubtarget::GFX9 &&
-        (CC == CallingConv::AMDGPU_HS || CC == CallingConv::AMDGPU_GS))
-      PrivateSegmentWaveByteOffsetSystemSGPR = AMDGPU::SGPR5;
+      // HS and GS always have the scratch wave offset in SGPR5 on GFX9.
+      if (ST.getGeneration() >= AMDGPUSubtarget::GFX9 &&
+          (CC == CallingConv::AMDGPU_HS || CC == CallingConv::AMDGPU_GS))
+        PrivateSegmentWaveByteOffsetSystemSGPR = AMDGPU::SGPR5;
+    }
   }
 
-  if (ST.isAmdCodeObjectV2(MF)) {
+  bool IsCOV2 = ST.isAmdCodeObjectV2(MF);
+  if (IsCOV2) {
     if (HasStackObjects || MaySpill)
       PrivateSegmentBuffer = true;
 
@@ -154,11 +170,15 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
       ImplicitBufferPtr = true;
   }
 
-  // We don't need to worry about accessing spills with flat instructions.
-  // TODO: On VI where we must use flat for global, we should be able to omit
-  // this if it is never used for generic access.
-  if (HasStackObjects && ST.hasFlatAddressSpace() && ST.isAmdHsaOS())
-    FlatScratchInit = true;
+  if (F->hasFnAttribute("amdgpu-kernarg-segment-ptr"))
+    KernargSegmentPtr = true;
+
+  if (ST.hasFlatAddressSpace() && isEntryFunction() && IsCOV2) {
+    // TODO: This could be refined a lot. The attribute is a poor way of
+    // detecting calls that may require it before argument lowering.
+    if (HasStackObjects || F->hasFnAttribute("amdgpu-flat-scratch"))
+      FlatScratchInit = true;
+  }
 }
 
 unsigned SIMachineFunctionInfo::addPrivateSegmentBuffer(
diff --git a/lib/Target/AMDGPU/SIMachineFunctionInfo.h b/lib/Target/AMDGPU/SIMachineFunctionInfo.h
index 05aa249584bf..4c7f38a09a48 100644
--- a/lib/Target/AMDGPU/SIMachineFunctionInfo.h
+++ b/lib/Target/AMDGPU/SIMachineFunctionInfo.h
@@ -119,6 +119,11 @@ class SIMachineFunctionInfo final : public AMDGPUMachineFunction {
   unsigned WorkGroupInfoSystemSGPR;
   unsigned PrivateSegmentWaveByteOffsetSystemSGPR;
 
+  // VGPR inputs. These are always v0, v1 and v2 for entry functions.
+  unsigned WorkItemIDXVGPR;
+  unsigned WorkItemIDYVGPR;
+  unsigned WorkItemIDZVGPR;
+
   // Graphics info.
   unsigned PSInputAddr;
   unsigned PSInputEnable;
@@ -377,10 +382,13 @@ public:
   }
 
   void setStackPtrOffsetReg(unsigned Reg) {
-    assert(Reg != AMDGPU::NoRegister && "Should never be unset");
     StackPtrOffsetReg = Reg;
   }
 
+  // Note the unset value for this is AMDGPU::SP_REG rather than
+  // NoRegister. This is mostly a workaround for MIR tests where state that
+  // can't be directly computed from the function is not preserved in serialized
+  // MIR.
   unsigned getStackPtrOffsetReg() const {
     return StackPtrOffsetReg;
   }
diff --git a/lib/Target/AMDGPU/SIRegisterInfo.cpp b/lib/Target/AMDGPU/SIRegisterInfo.cpp
index ef6ad4ad0c8f..4a3fbb4593bb 100644
--- a/lib/Target/AMDGPU/SIRegisterInfo.cpp
+++ b/lib/Target/AMDGPU/SIRegisterInfo.cpp
@@ -207,7 +207,11 @@ BitVector SIRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
     assert(!isSubRegister(ScratchRSrcReg, ScratchWaveOffsetReg));
   }
 
+  // We have to assume the SP is needed in case there are calls in the function,
+  // which is detected after the function is lowered. If we aren't really going
+  // to need SP, don't bother reserving it.
   unsigned StackPtrReg = MFI->getStackPtrOffsetReg();
+
   if (StackPtrReg != AMDGPU::NoRegister) {
     reserveRegisterTuples(Reserved, StackPtrReg);
     assert(!isSubRegister(ScratchRSrcReg, StackPtrReg));
diff --git a/lib/Target/AMDGPU/SIRegisterInfo.td b/lib/Target/AMDGPU/SIRegisterInfo.td
index fc808011cd88..54ea7805e18d 100644
--- a/lib/Target/AMDGPU/SIRegisterInfo.td
+++ b/lib/Target/AMDGPU/SIRegisterInfo.td
@@ -23,6 +23,13 @@ class SIReg <string n, bits<16> regIdx = 0> : Register<n>,
 def VCC_LO : SIReg<"vcc_lo", 106>;
 def VCC_HI : SIReg<"vcc_hi", 107>;
 
+// Pseudo-registers: Used as placeholders during isel and immediately
+// replaced, never seeing the verifier.
+def PRIVATE_RSRC_REG : SIReg<"", 0>;
+def FP_REG : SIReg<"", 0>;
+def SP_REG : SIReg<"", 0>;
+def SCRATCH_WAVE_OFFSET_REG : SIReg<"", 0>;
+
 // VCC for 64-bit instructions
 def VCC : RegisterWithSubRegs<"vcc", [VCC_LO, VCC_HI]>,
           DwarfRegAlias<VCC_LO> {
@@ -267,7 +274,8 @@ def VGPR_512 : RegisterTuples<[sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7,
 def SReg_32_XM0_XEXEC : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16], 32,
   (add SGPR_32, VCC_LO, VCC_HI, FLAT_SCR_LO, FLAT_SCR_HI,
    TTMP_32, TMA_LO, TMA_HI, TBA_LO, TBA_HI, SRC_SHARED_BASE, SRC_SHARED_LIMIT,
-   SRC_PRIVATE_BASE, SRC_PRIVATE_LIMIT)> {
+   SRC_PRIVATE_BASE, SRC_PRIVATE_LIMIT,
+   FP_REG, SP_REG, SCRATCH_WAVE_OFFSET_REG)> {
   let AllocationPriority = 7;
 }
 
@@ -314,7 +322,8 @@ def TTMP_128 : RegisterClass<"AMDGPU", [v4i32, v16i8, v2i64], 32, (add TTMP_128R
   let isAllocatable = 0;
 }
 
-def SReg_128 : RegisterClass<"AMDGPU", [v4i32, v16i8, v2i64], 32, (add SGPR_128, TTMP_128)> {
+def SReg_128 : RegisterClass<"AMDGPU", [v4i32, v16i8, v2i64], 32,
+  (add SGPR_128, TTMP_128)> {
   let AllocationPriority = 10;
 }
 
@@ -464,7 +473,9 @@ defm SCSrc : RegInlineOperand<"SReg", "SCSrc"> ;
 
 defm VSrc : RegImmOperand<"VS", "VSrc">;
 
-def VSrc_128 : RegisterOperand<VReg_128>;
+def VSrc_128 : RegisterOperand<VReg_128> {
+  let DecoderMethod = "DecodeVS_128RegisterClass";
+}
 
 //===----------------------------------------------------------------------===//
 //  VSrc_* Operands with an VGPR
diff --git a/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp b/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
index 26515b27bb77..67ad904ca972 100644
--- a/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
+++ b/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
@@ -539,23 +539,9 @@ bool isSGPR(unsigned Reg, const MCRegisterInfo* TRI) {
 }
 
 bool isRegIntersect(unsigned Reg0, unsigned Reg1, const MCRegisterInfo* TRI) {
-
-  if (Reg0 == Reg1) {
-    return true;
+  for (MCRegAliasIterator R(Reg0, TRI, true); R.isValid(); ++R) {
+    if (*R == Reg1) return true;
   }
-
-  unsigned SubReg0 = TRI->getSubReg(Reg0, 1);
-  if (SubReg0 == 0) {
-    return TRI->getSubRegIndex(Reg1, Reg0) > 0;
-  }
-
-  for (unsigned Idx = 2; SubReg0 > 0; ++Idx) {
-    if (isRegIntersect(Reg1, SubReg0, TRI)) {
-      return true;
-    }
-    SubReg0 = TRI->getSubReg(Reg0, Idx);
-  }
-
   return false;
 }
 
diff --git a/lib/Target/AMDGPU/VOP2Instructions.td b/lib/Target/AMDGPU/VOP2Instructions.td
index 7b9bc71ad4c7..d5acb49b4f39 100644
--- a/lib/Target/AMDGPU/VOP2Instructions.td
+++ b/lib/Target/AMDGPU/VOP2Instructions.td
@@ -117,7 +117,10 @@ class VOP2_SDWA_Pseudo <string OpName, VOPProfile P, list<dag> pattern=[]> :
 class getVOP2Pat64 <SDPatternOperator node, VOPProfile P> : LetDummies {
   list<dag> ret = !if(P.HasModifiers,
     [(set P.DstVT:$vdst,
-      (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)),
+      (node (P.Src0VT
+              !if(P.HasOMod,
+                  (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod),
+                  (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp))),
             (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers))))],
     [(set P.DstVT:$vdst, (node P.Src0VT:$src0, P.Src1VT:$src1))]);
 }
@@ -813,9 +816,11 @@ let SubtargetPredicate = isVI in {
 
 // Aliases to simplify matching of floating-point instructions that
 // are VOP2 on SI and VOP3 on VI.
-class SI2_VI3Alias <string name, Instruction inst> : InstAlias <
+class SI2_VI3Alias <string name, VOP3_Real inst> : InstAlias <
   name#" $dst, $src0, $src1",
-  (inst VGPR_32:$dst, 0, VCSrc_f32:$src0, 0, VCSrc_f32:$src1, 0, 0)
+  !if(inst.Pfl.HasOMod,
+      (inst VGPR_32:$dst, 0, VCSrc_f32:$src0, 0, VCSrc_f32:$src1, 0, 0),
+      (inst VGPR_32:$dst, 0, VCSrc_f32:$src0, 0, VCSrc_f32:$src1, 0))
 >, PredicateControl {
   let UseInstAsmMatchConverter = 0;
   let AsmVariantName = AMDGPUAsmVariants.VOP3;
diff --git a/lib/Target/AMDGPU/VOP3Instructions.td b/lib/Target/AMDGPU/VOP3Instructions.td
index a8ca593f14ed..92ed0706dc01 100644
--- a/lib/Target/AMDGPU/VOP3Instructions.td
+++ b/lib/Target/AMDGPU/VOP3Instructions.td
@@ -12,17 +12,21 @@
 //===----------------------------------------------------------------------===//
 
 class getVOP3ModPat<VOPProfile P, SDPatternOperator node> {
+  dag src0 = !if(P.HasOMod,
+    (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod),
+    (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp));
+
   list<dag> ret3 = [(set P.DstVT:$vdst,
-    (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)),
+    (node (P.Src0VT src0),
           (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers)),
           (P.Src2VT (VOP3Mods P.Src2VT:$src2, i32:$src2_modifiers))))];
 
   list<dag> ret2 = [(set P.DstVT:$vdst,
-    (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)),
+    (node (P.Src0VT src0),
           (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers))))];
 
   list<dag> ret1 = [(set P.DstVT:$vdst,
-    (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod))))];
+    (node (P.Src0VT src0)))];
 
   list<dag> ret = !if(!eq(P.NumSrcArgs, 3), ret3,
                   !if(!eq(P.NumSrcArgs, 2), ret2,
@@ -92,6 +96,7 @@ class VOP3_Profile<VOPProfile P> : VOPProfile<P.ArgVT> {
 class VOP3b_Profile<ValueType vt> : VOPProfile<[vt, vt, vt, vt]> {
   // v_div_scale_{f32|f64} do not support input modifiers.
   let HasModifiers = 0;
+  let HasOMod = 0;
   let Outs64 = (outs DstRC:$vdst, SReg_64:$sdst);
   let Asm64 = " $vdst, $sdst, $src0, $src1, $src2";
 }
diff --git a/lib/Target/AMDGPU/VOP3PInstructions.td b/lib/Target/AMDGPU/VOP3PInstructions.td
index f2de1f995726..3becf758aaa3 100644
--- a/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -34,6 +34,9 @@ class VOP3_VOP3PInst<string OpName, VOPProfile P, SDPatternOperator node = null_
 
 let isCommutable = 1 in {
 def V_PK_FMA_F16 : VOP3PInst<"v_pk_fma_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16_V2F16>, fma>;
+def V_PK_MAD_I16 : VOP3PInst<"v_pk_mad_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16_V2I16>>;
+def V_PK_MAD_U16 : VOP3PInst<"v_pk_mad_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16_V2I16>>;
+
 def V_PK_ADD_F16 : VOP3PInst<"v_pk_add_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16>, fadd>;
 def V_PK_MUL_F16 : VOP3PInst<"v_pk_mul_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16>, fmul>;
 def V_PK_MAX_F16 : VOP3PInst<"v_pk_max_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16>, fmaxnum>;
@@ -41,7 +44,6 @@ def V_PK_MIN_F16 : VOP3PInst<"v_pk_min_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16>
 
 def V_PK_ADD_U16 : VOP3PInst<"v_pk_add_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, add>;
 def V_PK_ADD_I16 : VOP3PInst<"v_pk_add_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>>;
-def V_PK_SUB_I16 : VOP3PInst<"v_pk_sub_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, sub>;
 def V_PK_MUL_LO_U16 : VOP3PInst<"v_pk_mul_lo_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, mul>;
 
 def V_PK_MIN_I16 : VOP3PInst<"v_pk_min_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, smin>;
@@ -50,6 +52,9 @@ def V_PK_MAX_I16 : VOP3PInst<"v_pk_max_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>
 def V_PK_MAX_U16 : VOP3PInst<"v_pk_max_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, umax>;
 }
 
+def V_PK_SUB_U16 : VOP3PInst<"v_pk_sub_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>>;
+def V_PK_SUB_I16 : VOP3PInst<"v_pk_sub_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, sub>;
+
 def V_PK_LSHLREV_B16 : VOP3PInst<"v_pk_lshlrev_b16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, lshl_rev>;
 def V_PK_ASHRREV_I16 : VOP3PInst<"v_pk_ashrrev_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, ashr_rev>;
 def V_PK_LSHRREV_B16 : VOP3PInst<"v_pk_lshrrev_b16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, lshr_rev>;
@@ -71,6 +76,7 @@ multiclass VOP3P_Real_vi<bits<10> op> {
   }
 }
 
+defm V_PK_MAD_I16 : VOP3P_Real_vi <0x380>;
 defm V_PK_MUL_LO_U16 : VOP3P_Real_vi <0x381>;
 defm V_PK_ADD_I16 : VOP3P_Real_vi <0x382>;
 defm V_PK_SUB_I16 : VOP3P_Real_vi <0x383>;
@@ -79,8 +85,10 @@ defm V_PK_LSHRREV_B16 : VOP3P_Real_vi <0x385>;
 defm V_PK_ASHRREV_I16 : VOP3P_Real_vi <0x386>;
 defm V_PK_MAX_I16 : VOP3P_Real_vi <0x387>;
 defm V_PK_MIN_I16 : VOP3P_Real_vi <0x388>;
+defm V_PK_MAD_U16 : VOP3P_Real_vi <0x389>;
 
 defm V_PK_ADD_U16 : VOP3P_Real_vi <0x38a>;
+defm V_PK_SUB_U16 : VOP3P_Real_vi <0x38b>;
 defm V_PK_MAX_U16 : VOP3P_Real_vi <0x38c>;
 defm V_PK_MIN_U16 : VOP3P_Real_vi <0x38d>;
 defm V_PK_FMA_F16 : VOP3P_Real_vi <0x38e>;
diff --git a/lib/Target/AMDGPU/VOPCInstructions.td b/lib/Target/AMDGPU/VOPCInstructions.td
index f3482a22d5dc..b636fc9be431 100644
--- a/lib/Target/AMDGPU/VOPCInstructions.td
+++ b/lib/Target/AMDGPU/VOPCInstructions.td
@@ -148,6 +148,19 @@ class VOPCInstAlias <VOP3_Pseudo ps, Instruction inst, VOPProfile p = ps.Pfl> :
   let SubtargetPredicate = AssemblerPredicate;
 }
 
+class getVOPCPat64 <PatLeaf cond, VOPProfile P> : LetDummies {
+  list<dag> ret = !if(P.HasModifiers,
+      [(set i1:$sdst,
+        (setcc (P.Src0VT
+                  !if(P.HasOMod,
+                    (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod),
+                    (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp))),
+               (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers)),
+               cond))],
+      [(set i1:$sdst, (setcc P.Src0VT:$src0, P.Src1VT:$src1, cond))]);
+}
+
+
 multiclass VOPC_Pseudos <string opName,
                          VOPC_Profile P,
                          PatLeaf cond = COND_NULL,
@@ -163,14 +176,7 @@ multiclass VOPC_Pseudos <string opName,
     let isCommutable = 1;
   }
 
-  def _e64 : VOP3_Pseudo<opName, P,
-    !if(P.HasModifiers,
-      [(set i1:$sdst,
-          (setcc (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers,
-                                      i1:$clamp, i32:$omod)),
-                 (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers)),
-                 cond))],
-      [(set i1:$sdst, (setcc P.Src0VT:$src0, P.Src1VT:$src1, cond))])>,
+  def _e64 : VOP3_Pseudo<opName, P, getVOPCPat64<cond, P>.ret>,
     Commutable_REV<revOp#"_e64", !eq(revOp, opName)> {
     let Defs = !if(DefExec, [EXEC], []);
     let SchedRW = P.Schedule;
@@ -634,7 +640,7 @@ class FCMP_Pattern <PatLeaf cond, Instruction inst, ValueType vt> : Pat <
   (i64 (AMDGPUsetcc (vt (VOP3Mods vt:$src0, i32:$src0_modifiers)),
                    (vt (VOP3Mods vt:$src1, i32:$src1_modifiers)), cond)),
   (inst $src0_modifiers, $src0, $src1_modifiers, $src1,
-        DSTCLAMP.NONE, DSTOMOD.NONE)
+        DSTCLAMP.NONE)
 >;
 
 def : FCMP_Pattern <COND_OEQ, V_CMP_EQ_F32_e64, f32>;
diff --git a/lib/Target/AMDGPU/VOPInstructions.td b/lib/Target/AMDGPU/VOPInstructions.td
index 77b7952b22a8..b47538ba0349 100644
--- a/lib/Target/AMDGPU/VOPInstructions.td
+++ b/lib/Target/AMDGPU/VOPInstructions.td
@@ -136,6 +136,8 @@ class VOP3_Real <VOP3_Pseudo ps, int EncodingFamily> :
   let TSFlags            = ps.TSFlags;
   let UseNamedOperandTable = ps.UseNamedOperandTable;
   let Uses                 = ps.Uses;
+
+  VOPProfile Pfl = ps.Pfl;
 }
 
 // XXX - Is there any reason to distingusih this from regular VOP3