[llvm-exegesis] Improve Register Setup.

[gchatelet]

Summary

Added function to set a register to a particular value + tests.
Add EFLAGS test, use new setRegTo instead of setRegToConstant.

[courbet]

tools/llvm-exegesis/lib/Assembler.cpp

173–174

let's keep this code factored out in generateSnippetSetupCode().

tools/llvm-exegesis/lib/BenchmarkRunner.h

46 ↗

(On Diff #164660)

you can kill that now.

47 ↗

(On Diff #164660)

where are you setting these ?

tools/llvm-exegesis/lib/RegisterValue.h

1 ↗

(On Diff #164660)

Can you please split this to a different patch ?

tools/llvm-exegesis/lib/Target.h

38–39

Please comment what happens when the register bit width is not the same as that of the value.

52

As discussed, let's not change this API (+fillMemoryOperands, getMaxMemoryAccessSize). It already handles not supporting memory operands, which is valid for a target.

tools/llvm-exegesis/lib/X86/Target.cpp

104

Conceptually, the "load immediate" opcode should be a function of the size of the register AND value, not just value (e.g. MOV64ri vs MOV64ri32). See my comment below.

121

Is there any case when the size of the value is not the same as that of the register ? If not, let's make this a requirement and crash if the value bit width is not exactly that of the register.

323

let's fail gracefully instead of crashing here. That's not the users' fault.

[gchatelet]

• Addressing courbet's comments.

[gchatelet]

• Addressing courbet's comments.

[gchatelet]

• Addressing courbet's comments.

[courbet]

tools/llvm-exegesis/lib/Target.cpp

95

this will break exegesis on AARch64. I think we should implement it first.

unittests/tools/llvm-exegesis/AArch64/TargetTest.cpp

44

let's reinstate that.

[gchatelet]

• Addressing courbet's comments.
• Merge conflicts, reinstated AArch64 tests

[gchatelet]

• Added 'Set' functions for AArch64 GPRs.

[courbet]

tools/llvm-exegesis/lib/AArch64/Target.cpp

32

getLoadImmediateOpcode

tools/llvm-exegesis/lib/Assembler.cpp

40

Please keep the comment.

183

You need to keep this.

[gchatelet]

• Restore tracking of snipped completion.

tools/llvm-exegesis/lib/AArch64/Target.cpp

 //===-- Target.cpp ----------------------------------------------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 #include "../Target.h"
 #include "../Latency.h"
 #include "AArch64.h"
+#include "AArch64RegisterInfo.h"
 
 namespace exegesis {
 
 namespace {
 
 class AArch64LatencyBenchmarkRunner : public LatencyBenchmarkRunner {
 public:
   AArch64LatencyBenchmarkRunner(const LLVMState &State)
       : LatencyBenchmarkRunner(State) {}
 
 private:
   const char *getCounterName() const override {
     // All AArch64 subtargets have CPU_CYCLES as the cycle counter name
     return "CPU_CYCLES";
   }
 };
 
+namespace {
+
+static unsigned getLoadImmediateOpcode(unsigned RegBitWidth) {
+  switch (RegBitWidth) {
+  case 32:
+    return llvm::AArch64::MOVi32imm;
+  case 64:
+    return llvm::AArch64::MOVi64imm;
+  }
+  llvm_unreachable("Invalid Value Width");
+}
+
+// Generates instruction to load an immediate value into a register.
+static llvm::MCInst loadImmediate(unsigned Reg, unsigned RegBitWidth,
+                                  const llvm::APInt &Value) {
+  if (Value.getBitWidth() > RegBitWidth)
+    llvm_unreachable("Value must fit in the Register");
+  return llvm::MCInstBuilder(getLoadImmediateOpcode(RegBitWidth))
+      .addReg(Reg)
+      .addImm(Value.getZExtValue());
+}
+
+} // namespace
+
 class ExegesisAArch64Target : public ExegesisTarget {
+  std::vector<llvm::MCInst> setRegTo(const llvm::MCSubtargetInfo &STI,
+                                     unsigned Reg,
+                                     const llvm::APInt &Value) const override {
+    if (llvm::AArch64::GPR32RegClass.contains(Reg))
+      return {loadImmediate(Reg, 32, Value)};
+    if (llvm::AArch64::GPR64RegClass.contains(Reg))
+      return {loadImmediate(Reg, 64, Value)};
+    llvm::errs() << "setRegTo is not implemented, results will be unreliable\n";
+    return {};
+  }
+
   bool matchesArch(llvm::Triple::ArchType Arch) const override {
     return Arch == llvm::Triple::aarch64 || Arch == llvm::Triple::aarch64_be;
   }
+
   void addTargetSpecificPasses(llvm::PassManagerBase &PM) const override {
     // Function return is a pseudo-instruction that needs to be expanded
     PM.add(llvm::createAArch64ExpandPseudoPass());
   }
+
   std::unique_ptr<BenchmarkRunner>
   createLatencyBenchmarkRunner(const LLVMState &State) const override {
     return llvm::make_unique<AArch64LatencyBenchmarkRunner>(State);
   }
 };
 
 } // namespace
 

tools/llvm-exegesis/lib/Assembler.h

 namespace exegesis {
 
 class ExegesisTarget;
 
 // Gather the set of reserved registers (depends on function's calling
 // convention and target machine).
 llvm::BitVector getFunctionReservedRegs(const llvm::TargetMachine &TM);
 
+// A simple object storing the value for a particular register.
+struct RegisterValue {
+  unsigned Register;
+  llvm::APInt Value;
+};
+
 // Creates a temporary `void foo(char*)` function containing the provided
 // Instructions. Runs a set of llvm Passes to provide correct prologue and
 // epilogue. Once the MachineFunction is ready, it is assembled for TM to
 // AsmStream, the temporary function is eventually discarded.
 void assembleToStream(const ExegesisTarget &ET,
                       std::unique_ptr<llvm::LLVMTargetMachine> TM,
                       llvm::ArrayRef<unsigned> LiveIns,
-                      llvm::ArrayRef<unsigned> RegsToDef,
+                      llvm::ArrayRef<RegisterValue> RegisterInitialValues,
                       llvm::ArrayRef<llvm::MCInst> Instructions,
                       llvm::raw_pwrite_stream &AsmStream);
 
 // Creates an ObjectFile in the format understood by the host.
 // Note: the resulting object keeps a copy of Buffer so it can be discarded once
 // this function returns.
 llvm::object::OwningBinary<llvm::object::ObjectFile>
 getObjectFromBuffer(llvm::StringRef Buffer);

tools/llvm-exegesis/lib/Assembler.cpp

 #include "llvm/Support/MemoryBuffer.h"
 
 namespace exegesis {
 
 static constexpr const char ModuleID[] = "ExegesisInfoTest";
 static constexpr const char FunctionID[] = "foo";
 
 static std::vector<llvm::MCInst>
-generateSnippetSetupCode(const llvm::ArrayRef<unsigned> RegsToDef,
-                         const ExegesisTarget &ET,
-                         const llvm::LLVMTargetMachine &TM, bool &IsComplete) {
-  IsComplete = true;
+generateSnippetSetupCode(const ExegesisTarget &ET,
+                         const llvm::MCSubtargetInfo *const MSI,
+                         llvm::ArrayRef<RegisterValue> RegisterInitialValues,
+                         bool &IsSnippetSetupComplete) {
   std::vector<llvm::MCInst> Result;
-  for (const unsigned Reg : RegsToDef) {
+  for (const RegisterValue &RV : RegisterInitialValues) {
     // Load a constant in the register.
-    const auto Code = ET.setRegToConstant(*TM.getMCSubtargetInfo(), Reg);
-    if (Code.empty())
-      IsComplete = false;
-    Result.insert(Result.end(), Code.begin(), Code.end());
+    const auto SetRegisterCode = ET.setRegTo(*MSI, RV.Register, RV.Value);
+    if (SetRegisterCode.empty())
+      IsSnippetSetupComplete = false;
+    Result.insert(Result.end(), SetRegisterCode.begin(), SetRegisterCode.end());
   }
   return Result;
 }
 
 // Small utility function to add named passes.
 static bool addPass(llvm::PassManagerBase &PM, llvm::StringRef PassName,
                     llvm::TargetPassConfig &TPC) {
   const llvm::PassRegistry *PR = llvm::PassRegistry::getPassRegistry();
llvm::MachineFunction &MF =
       createVoidVoidPtrMachineFunction(FunctionID, Module.get(), MMI.get());
   // Saving reserved registers for client.
   return MF.getSubtarget().getRegisterInfo()->getReservedRegs(MF);
 }
 
 void assembleToStream(const ExegesisTarget &ET,
                       std::unique_ptr<llvm::LLVMTargetMachine> TM,
                       llvm::ArrayRef<unsigned> LiveIns,
-                      llvm::ArrayRef<unsigned> RegsToDef,
+                      llvm::ArrayRef<RegisterValue> RegisterInitialValues,
                       llvm::ArrayRef<llvm::MCInst> Instructions,
                       llvm::raw_pwrite_stream &AsmStream) {
   std::unique_ptr<llvm::LLVMContext> Context =
       llvm::make_unique<llvm::LLVMContext>();
   std::unique_ptr<llvm::Module> Module =
       createModule(Context, TM->createDataLayout());
   std::unique_ptr<llvm::MachineModuleInfo> MMI =
       llvm::make_unique<llvm::MachineModuleInfo>(TM.get());
   llvm::MachineFunction &MF =
       createVoidVoidPtrMachineFunction(FunctionID, Module.get(), MMI.get());
 
   // We need to instruct the passes that we're done with SSA and virtual
   // registers.
   auto &Properties = MF.getProperties();
   Properties.set(llvm::MachineFunctionProperties::Property::NoVRegs);
   Properties.reset(llvm::MachineFunctionProperties::Property::IsSSA);
 
   for (const unsigned Reg : LiveIns)
     MF.getRegInfo().addLiveIn(Reg);
 
   bool IsSnippetSetupComplete = false;
-  std::vector<llvm::MCInst> SnippetWithSetup =
-      generateSnippetSetupCode(RegsToDef, ET, *TM, IsSnippetSetupComplete);
-  if (!SnippetWithSetup.empty()) {
-    SnippetWithSetup.insert(SnippetWithSetup.end(), Instructions.begin(),
-                            Instructions.end());
-    Instructions = SnippetWithSetup;
-  }
+  std::vector<llvm::MCInst> Code =
+      generateSnippetSetupCode(ET, TM->getMCSubtargetInfo(),
+                               RegisterInitialValues, IsSnippetSetupComplete);
+
+  Code.insert(Code.end(), Instructions.begin(), Instructions.end());
+
   // If the snippet setup is not complete, we disable liveliness tracking. This
   // means that we won't know what values are in the registers.
   if (!IsSnippetSetupComplete)
     Properties.reset(llvm::MachineFunctionProperties::Property::TracksLiveness);
 
   // prologue/epilogue pass needs the reserved registers to be frozen, this
   // is usually done by the SelectionDAGISel pass.
   MF.getRegInfo().freezeReservedRegs(MF);
 
   // Fill the MachineFunction from the instructions.
-  fillMachineFunction(MF, LiveIns, Instructions);
+  fillMachineFunction(MF, LiveIns, Code);
 
   // We create the pass manager, run the passes to populate AsmBuffer.
   llvm::MCContext &MCContext = MMI->getContext();
   llvm::legacy::PassManager PM;
 
   llvm::TargetLibraryInfoImpl TLII(llvm::Triple(Module->getTargetTriple()));
   PM.add(new llvm::TargetLibraryInfoWrapperPass(TLII));
 

tools/llvm-exegesis/lib/BenchmarkCode.h

 
 // A collection of instructions that are to be assembled, executed and measured.
 struct BenchmarkCode {
   // The sequence of instructions that are to be repeated.
   std::vector<llvm::MCInst> Instructions;
 
   // Before the code is executed some instructions are added to setup the
   // registers initial values.
-  std::vector<unsigned> RegsToDef;
+  std::vector<RegisterValue> RegisterInitialValues;
 
   // We also need to provide the registers that are live on entry for the
   // assembler to generate proper prologue/epilogue.
   std::vector<unsigned> LiveIns;
 
   // Informations about how this configuration was built.
   std::string Info;
 };
 
 } // namespace exegesis
 
 #endif // LLVM_TOOLS_LLVM_EXEGESIS_BENCHMARKCODE_H

tools/llvm-exegesis/lib/BenchmarkRunner.cpp

BenchmarkRunner::writeObjectFile(const BenchmarkCode &BC,
                                  llvm::ArrayRef<llvm::MCInst> Code) const {
   int ResultFD = 0;
   llvm::SmallString<256> ResultPath;
   if (llvm::Error E = llvm::errorCodeToError(llvm::sys::fs::createTemporaryFile(
           "snippet", "o", ResultFD, ResultPath)))
     return std::move(E);
   llvm::raw_fd_ostream OFS(ResultFD, true /*ShouldClose*/);
   assembleToStream(State.getExegesisTarget(), State.createTargetMachine(),
-                   BC.LiveIns, BC.RegsToDef, Code, OFS);
+                   BC.LiveIns, BC.RegisterInitialValues, Code, OFS);
   return ResultPath.str();
 }
 
 } // namespace exegesis

tools/llvm-exegesis/lib/SnippetGenerator.h

public:
 
   virtual ~SnippetGenerator();
 
   // Calls generateCodeTemplate and expands it into one or more BenchmarkCode.
   llvm::Expected<std::vector<BenchmarkCode>>
   generateConfigurations(unsigned Opcode) const;
 
   // Given a snippet, computes which registers the setup code needs to define.
-  std::vector<unsigned>
-  computeRegsToDef(const std::vector<InstructionBuilder> &Snippet) const;
+  std::vector<RegisterValue> computeRegisterInitialValues(
+      const std::vector<InstructionBuilder> &Snippet) const;
 
 protected:
   const LLVMState &State;
   const RegisterAliasingTrackerCache RATC;
 
   // Generates a single code template that has a self-dependency.
   llvm::Expected<CodeTemplate>
   generateSelfAliasingCodeTemplate(const Instruction &Instr) const;

tools/llvm-exegesis/lib/SnippetGenerator.cpp

// TODO: Generate as many BenchmarkCode as needed.
         IB.randomizeUnsetVariables(
             CT.ScratchSpacePointerInReg
                 ? RATC.getRegister(CT.ScratchSpacePointerInReg).aliasedBits()
                 : RATC.emptyRegisters());
         BC.Instructions.push_back(IB.build());
       }
       if (CT.ScratchSpacePointerInReg)
         BC.LiveIns.push_back(CT.ScratchSpacePointerInReg);
-      BC.RegsToDef = computeRegsToDef(CT.Instructions);
+      BC.RegisterInitialValues = computeRegisterInitialValues(CT.Instructions);
       Output.push_back(std::move(BC));
     }
     return Output;
   } else
     return E.takeError();
 }
 
-std::vector<unsigned> SnippetGenerator::computeRegsToDef(
+std::vector<RegisterValue> SnippetGenerator::computeRegisterInitialValues(
     const std::vector<InstructionBuilder> &Instructions) const {
   // Collect all register uses and create an assignment for each of them.
   // Ignore memory operands which are handled separately.
   // Loop invariant: DefinedRegs[i] is true iif it has been set at least once
   // before the current instruction.
   llvm::BitVector DefinedRegs = RATC.emptyRegisters();
-  std::vector<unsigned> RegsToDef;
+  std::vector<RegisterValue> RIV;
   for (const InstructionBuilder &IB : Instructions) {
     // Returns the register that this Operand sets or uses, or 0 if this is not
     // a register.
     const auto GetOpReg = [&IB](const Operand &Op) -> unsigned {
       if (Op.IsMem)
         return 0;
       if (Op.ImplicitReg)
         return *Op.ImplicitReg;
       if (Op.IsExplicit && IB.getValueFor(Op).isReg())
         return IB.getValueFor(Op).getReg();
       return 0;
     };
     // Collect used registers that have never been def'ed.
     for (const Operand &Op : IB.Instr.Operands) {
       if (!Op.IsDef) {
         const unsigned Reg = GetOpReg(Op);
         if (Reg > 0 && !DefinedRegs.test(Reg)) {
-          RegsToDef.push_back(Reg);
+          RIV.push_back(RegisterValue{Reg, llvm::APInt()});
           DefinedRegs.set(Reg);
         }
       }
     }
     // Mark defs as having been def'ed.
     for (const Operand &Op : IB.Instr.Operands) {
       if (Op.IsDef) {
         const unsigned Reg = GetOpReg(Op);
         if (Reg > 0)
           DefinedRegs.set(Reg);
       }
     }
   }
-  return RegsToDef;
+  return RIV;
 }
 
 llvm::Expected<CodeTemplate> SnippetGenerator::generateSelfAliasingCodeTemplate(
     const Instruction &Instr) const {
   const AliasingConfigurations SelfAliasing(Instr, Instr);
   if (SelfAliasing.empty()) {
     return llvm::make_error<SnippetGeneratorFailure>("empty self aliasing");
   }

tools/llvm-exegesis/lib/Target.h

 
 namespace exegesis {
 
 class ExegesisTarget {
 public:
   // Targets can use this to add target-specific passes in assembleToStream();
   virtual void addTargetSpecificPasses(llvm::PassManagerBase &PM) const {}
 
   // Generates code to move a constant into a the given register.
+  // Precondition: Value must fit into Reg.
   virtual std::vector<llvm::MCInst>
-  setRegToConstant(const llvm::MCSubtargetInfo &STI, unsigned Reg) const {
-    return {};
-  }
+  setRegTo(const llvm::MCSubtargetInfo &STI, unsigned Reg,
+           const llvm::APInt &Value) const = 0;
 
-  // Returns the register pointing to scratch memory, or 0 if this target does
-  // not support memory operands. The benchmark function uses the default
-  // calling convention.
+  // Returns the register pointing to scratch memory, or 0 if this target
+  // does not support memory operands. The benchmark function uses the
+  // default calling convention.
   virtual unsigned getScratchMemoryRegister(const llvm::Triple &) const {
     return 0;
   }
 
   // Fills memory operands with references to the address at [Reg] + Offset.
   virtual void fillMemoryOperands(InstructionBuilder &IB, unsigned Reg,
                                   unsigned Offset) const {
     llvm_unreachable(
         "fillMemoryOperands() requires getScratchMemoryRegister() > 0");
   }
 
   // Returns the maximum number of bytes a load/store instruction can access at
   // once. This is typically the size of the largest register available on the
   // processor. Note that this only used as a hint to generate independant

tools/llvm-exegesis/lib/Target.cpp

ExegesisTarget::createUopsBenchmarkRunner(const LLVMState &State) const {
   return llvm::make_unique<UopsBenchmarkRunner>(State);
 }
 
 namespace {
 
 // Default implementation.
 class ExegesisDefaultTarget : public ExegesisTarget {
 private:
+  std::vector<llvm::MCInst> setRegTo(const llvm::MCSubtargetInfo &STI,
+                                     unsigned Reg,
+                                     const llvm::APInt &Value) const override {
+    llvm_unreachable("Not yet implemented");
+  }
+
   bool matchesArch(llvm::Triple::ArchType Arch) const override {
     llvm_unreachable("never called");
     return false;
   }
 };
 
 } // namespace
 
 const ExegesisTarget &ExegesisTarget::getDefault() {
   static ExegesisDefaultTarget Target;
   return Target;
 }
 
 } // namespace exegesis

tools/llvm-exegesis/lib/X86/Target.cpp

return generateUnconstrainedCodeTemplate(
         Instr, "instruction does not grow/shrink the FP stack");
   }
   llvm::Expected<CodeTemplate> handleCondMovFP(const Instruction &Instr) const {
     return generateUnconstrainedCodeTemplate(
         Instr, "instruction does not grow/shrink the FP stack");
   }
 };
 
+static unsigned GetLoadImmediateOpcode(unsigned RegBitWidth) {
+  switch (RegBitWidth) {
+  case 8:
+    return llvm::X86::MOV8ri;
+  case 16:
+    return llvm::X86::MOV16ri;
+  case 32:
+    return llvm::X86::MOV32ri;
+  case 64:
+    return llvm::X86::MOV64ri;
+  }
+  llvm_unreachable("Invalid Value Width");
+}
+
+// Generates instruction to load an immediate value into a register.
+static llvm::MCInst loadImmediate(unsigned Reg, unsigned RegBitWidth,
+                                  const llvm::APInt &Value) {
+  if (Value.getBitWidth() > RegBitWidth)
+    llvm_unreachable("Value must fit in the Register");
+  return llvm::MCInstBuilder(GetLoadImmediateOpcode(RegBitWidth))
+      .addReg(Reg)
+      .addImm(Value.getZExtValue());
+}
+
+// Allocates scratch memory on the stack.
+static llvm::MCInst allocateStackSpace(unsigned Bytes) {
+  return llvm::MCInstBuilder(llvm::X86::SUB64ri8)
+      .addReg(llvm::X86::RSP)
+      .addReg(llvm::X86::RSP)
+      .addImm(Bytes);
+}
+
+// Fills scratch memory at offset `OffsetBytes` with value `Imm`.
+static llvm::MCInst fillStackSpace(unsigned MovOpcode, unsigned OffsetBytes,
+                                   uint64_t Imm) {
+  return llvm::MCInstBuilder(MovOpcode)
+      // Address = ESP
+      .addReg(llvm::X86::RSP) // BaseReg
+      .addImm(1)              // ScaleAmt
+      .addReg(0)              // IndexReg
+      .addImm(OffsetBytes)    // Disp
+      .addReg(0)              // Segment
+      // Immediate.
+      .addImm(Imm);
+}
+
+// Loads scratch memory into register `Reg` using opcode `RMOpcode`.
+static llvm::MCInst loadToReg(unsigned Reg, unsigned RMOpcode) {
+  return llvm::MCInstBuilder(RMOpcode)
+      .addReg(Reg)
+      // Address = ESP
+      .addReg(llvm::X86::RSP) // BaseReg
+      .addImm(1)              // ScaleAmt
+      .addReg(0)              // IndexReg
+      .addImm(0)              // Disp
+      .addReg(0);             // Segment
+}
+
+// Releases scratch memory.
+static llvm::MCInst releaseStackSpace(unsigned Bytes) {
+  return llvm::MCInstBuilder(llvm::X86::ADD64ri8)
+      .addReg(llvm::X86::RSP)
+      .addReg(llvm::X86::RSP)
+      .addImm(Bytes);
+}
+
+// Reserves some space on the stack, fills it with the content of the provided
+// constant and provide methods to load the stack value into a register.
+struct ConstantInliner {
+  explicit ConstantInliner(const llvm::APInt &Constant)
+      : StackSize(Constant.getBitWidth() / 8) {
+    assert(Constant.getBitWidth() % 8 == 0 && "Must be a multiple of 8");
+    add(allocateStackSpace(StackSize));
+    size_t ByteOffset = 0;
+    for (; StackSize - ByteOffset >= 4; ByteOffset += 4)
+      add(fillStackSpace(
+          llvm::X86::MOV32mi, ByteOffset,
+          Constant.extractBits(32, ByteOffset * 8).getZExtValue()));
+    if (StackSize - ByteOffset >= 2) {
+      add(fillStackSpace(
+          llvm::X86::MOV16mi, ByteOffset,
+          Constant.extractBits(16, ByteOffset * 8).getZExtValue()));
+      ByteOffset += 2;
+    }
+    if (StackSize - ByteOffset >= 1)
+      add(fillStackSpace(
+          llvm::X86::MOV8mi, ByteOffset,
+          Constant.extractBits(8, ByteOffset * 8).getZExtValue()));
+  }
+
+  std::vector<llvm::MCInst> loadAndFinalize(unsigned Reg, unsigned RegBitWidth,
+                                            unsigned Opcode) {
+    assert(StackSize * 8 == RegBitWidth &&
+           "Value does not have the correct size");
+    add(loadToReg(Reg, Opcode));
+    add(releaseStackSpace(StackSize));
+    return std::move(Instructions);
+  }
+
+  std::vector<llvm::MCInst>
+  loadX87AndFinalize(unsigned Reg, unsigned RegBitWidth, unsigned Opcode) {
+    assert(StackSize * 8 == RegBitWidth &&
+           "Value does not have the correct size");
+    add(llvm::MCInstBuilder(Opcode)
+            .addReg(llvm::X86::RSP) // BaseReg
+            .addImm(1)              // ScaleAmt
+            .addReg(0)              // IndexReg
+            .addImm(0)              // Disp
+            .addReg(0));            // Segment
+    if (Reg != llvm::X86::ST0)
+      add(llvm::MCInstBuilder(llvm::X86::ST_Frr).addReg(Reg));
+    add(releaseStackSpace(StackSize));
+    return std::move(Instructions);
+  }
+
+  std::vector<llvm::MCInst> popFlagAndFinalize() {
+    assert(StackSize * 8 == 64 && "Value does not have the correct size");
+    add(llvm::MCInstBuilder(llvm::X86::POPF64));
+    return std::move(Instructions);
+  }
+
+private:
+  ConstantInliner &add(const llvm::MCInst &Inst) {
+    Instructions.push_back(Inst);
+    return *this;
+  }
+
+  const size_t StackSize;
+  std::vector<llvm::MCInst> Instructions;
+};
+
 class ExegesisX86Target : public ExegesisTarget {
   void addTargetSpecificPasses(llvm::PassManagerBase &PM) const override {
     // Lowers FP pseudo-instructions, e.g. ABS_Fp32 -> ABS_F.
     PM.add(llvm::createX86FloatingPointStackifierPass());
   }
 
   unsigned getScratchMemoryRegister(const llvm::Triple &TT) const override {
     if (!TT.isArch64Bit()) {
for (size_t I = 0, E = IB.Instr.Operands.size(); I < E; ++I) {
         assert(Op->IsMem);
         assert(Op->IsExplicit);
         IB.getValueFor(*Op) = llvm::MCOperand::createReg(0); // Segment
         // Case2: segment:index addressing. We assume that ES is 0.
       }
     }
   }
 
-  std::vector<llvm::MCInst> setRegToConstant(const llvm::MCSubtargetInfo &STI,
-                                             unsigned Reg) const override {
-    // GPR.
+  std::vector<llvm::MCInst> setRegTo(const llvm::MCSubtargetInfo &STI,
+                                     unsigned Reg,
+                                     const llvm::APInt &Value) const override {
     if (llvm::X86::GR8RegClass.contains(Reg))
-      return {llvm::MCInstBuilder(llvm::X86::MOV8ri).addReg(Reg).addImm(1)};
+      return {loadImmediate(Reg, 8, Value)};
     if (llvm::X86::GR16RegClass.contains(Reg))
-      return {llvm::MCInstBuilder(llvm::X86::MOV16ri).addReg(Reg).addImm(1)};
+      return {loadImmediate(Reg, 16, Value)};
     if (llvm::X86::GR32RegClass.contains(Reg))
-      return {llvm::MCInstBuilder(llvm::X86::MOV32ri).addReg(Reg).addImm(1)};
+      return {loadImmediate(Reg, 32, Value)};
     if (llvm::X86::GR64RegClass.contains(Reg))
-      return {llvm::MCInstBuilder(llvm::X86::MOV64ri32).addReg(Reg).addImm(1)};
-    // MMX.
+      return {loadImmediate(Reg, 64, Value)};
+    ConstantInliner CI(Value);
     if (llvm::X86::VR64RegClass.contains(Reg))
-      return setVectorRegToConstant(Reg, 8, llvm::X86::MMX_MOVQ64rm);
-    // {X,Y,Z}MM.
+      return CI.loadAndFinalize(Reg, 64, llvm::X86::MMX_MOVQ64rm);
     if (llvm::X86::VR128XRegClass.contains(Reg)) {
       if (STI.getFeatureBits()[llvm::X86::FeatureAVX512])
-        return setVectorRegToConstant(Reg, 16, llvm::X86::VMOVDQU32Z128rm);
+        return CI.loadAndFinalize(Reg, 128, llvm::X86::VMOVDQU32Z128rm);
       if (STI.getFeatureBits()[llvm::X86::FeatureAVX])
-        return setVectorRegToConstant(Reg, 16, llvm::X86::VMOVDQUrm);
-      return setVectorRegToConstant(Reg, 16, llvm::X86::MOVDQUrm);
+        return CI.loadAndFinalize(Reg, 128, llvm::X86::VMOVDQUrm);
+      return CI.loadAndFinalize(Reg, 128, llvm::X86::MOVDQUrm);
     }
     if (llvm::X86::VR256XRegClass.contains(Reg)) {
       if (STI.getFeatureBits()[llvm::X86::FeatureAVX512])
-        return setVectorRegToConstant(Reg, 32, llvm::X86::VMOVDQU32Z256rm);
-      return setVectorRegToConstant(Reg, 32, llvm::X86::VMOVDQUYrm);
+        return CI.loadAndFinalize(Reg, 256, llvm::X86::VMOVDQU32Z256rm);
+      if (STI.getFeatureBits()[llvm::X86::FeatureAVX])
+        return CI.loadAndFinalize(Reg, 256, llvm::X86::VMOVDQUYrm);
     }
     if (llvm::X86::VR512RegClass.contains(Reg))
-      return setVectorRegToConstant(Reg, 64, llvm::X86::VMOVDQU32Zrm);
-    // X87.
-    if (llvm::X86::RFP32RegClass.contains(Reg) ||
-        llvm::X86::RFP64RegClass.contains(Reg) ||
-        llvm::X86::RFP80RegClass.contains(Reg))
-      return setVectorRegToConstant(Reg, 8, llvm::X86::LD_Fp64m);
-    if (Reg == llvm::X86::EFLAGS) {
-      // Set all flags to 0 but the bits that are "reserved and set to 1".
-      constexpr const uint32_t kImmValue = 0x00007002u;
-      std::vector<llvm::MCInst> Result;
-      Result.push_back(allocateStackSpace(8));
-      Result.push_back(fillStackSpace(llvm::X86::MOV64mi32, 0, kImmValue));
-      Result.push_back(llvm::MCInstBuilder(llvm::X86::POPF64)); // Also pops.
-      return Result;
-    }
-    return {};
+      if (STI.getFeatureBits()[llvm::X86::FeatureAVX512])
+        return CI.loadAndFinalize(Reg, 512, llvm::X86::VMOVDQU32Zrm);
+    if (llvm::X86::RSTRegClass.contains(Reg)) {
+      if (Value.getBitWidth() == 32)
+        return CI.loadX87AndFinalize(Reg, 32, llvm::X86::LD_F32m);
+      if (Value.getBitWidth() == 64)
+        return CI.loadX87AndFinalize(Reg, 64, llvm::X86::LD_F64m);
+      if (Value.getBitWidth() == 80)
+        return CI.loadX87AndFinalize(Reg, 80, llvm::X86::LD_F80m);
+    }
+    if (Reg == llvm::X86::EFLAGS)
+      return CI.popFlagAndFinalize();
+    return {}; // Not yet implemented.
   }
 
   std::unique_ptr<SnippetGenerator>
   createLatencySnippetGenerator(const LLVMState &State) const override {
     return llvm::make_unique<X86SnippetGenerator<X86LatencyImpl>>(State);
   }
 
   std::unique_ptr<SnippetGenerator>
   createUopsSnippetGenerator(const LLVMState &State) const override {
     return llvm::make_unique<X86SnippetGenerator<X86UopsImpl>>(State);
   }
 
   bool matchesArch(llvm::Triple::ArchType Arch) const override {
     return Arch == llvm::Triple::x86_64 || Arch == llvm::Triple::x86;
   }
-
-private:
-  // setRegToConstant() specialized for a vector register of size
-  // `RegSizeBytes`. `RMOpcode` is the opcode used to do a memory -> vector
-  // register load.
-  static std::vector<llvm::MCInst>
-  setVectorRegToConstant(const unsigned Reg, const unsigned RegSizeBytes,
-                         const unsigned RMOpcode) {
-    // There is no instruction to directly set XMM, go through memory.
-    // Since vector values can be interpreted as integers of various sizes (8
-    // to 64 bits) as well as floats and double, so we chose an immediate
-    // value that has set bits for all byte values and is a normal float/
-    // double. 0x40404040 is ~32.5 when interpreted as a double and ~3.0f when
-    // interpreted as a float.
-    constexpr const uint32_t kImmValue = 0x40404040u;
-    std::vector<llvm::MCInst> Result;
-    Result.push_back(allocateStackSpace(RegSizeBytes));
-    constexpr const unsigned kMov32NumBytes = 4;
-    for (unsigned Disp = 0; Disp < RegSizeBytes; Disp += kMov32NumBytes) {
-      Result.push_back(fillStackSpace(llvm::X86::MOV32mi, Disp, kImmValue));
-    }
-    Result.push_back(loadToReg(Reg, RMOpcode));
-    Result.push_back(releaseStackSpace(RegSizeBytes));
-    return Result;
-  }
-
-  // Allocates scratch memory on the stack.
-  static llvm::MCInst allocateStackSpace(unsigned Bytes) {
-    return llvm::MCInstBuilder(llvm::X86::SUB64ri8)
-        .addReg(llvm::X86::RSP)
-        .addReg(llvm::X86::RSP)
-        .addImm(Bytes);
-  }
-
-  // Fills scratch memory at offset `OffsetBytes` with value `Imm`.
-  static llvm::MCInst fillStackSpace(unsigned MovOpcode, unsigned OffsetBytes,
-                                     uint64_t Imm) {
-    return llvm::MCInstBuilder(MovOpcode)
-        // Address = ESP
-        .addReg(llvm::X86::RSP) // BaseReg
-        .addImm(1)              // ScaleAmt
-        .addReg(0)              // IndexReg
-        .addImm(OffsetBytes)    // Disp
-        .addReg(0)              // Segment
-        // Immediate.
-        .addImm(Imm);
-  }
-
-  // Loads scratch memory into register `Reg` using opcode `RMOpcode`.
-  static llvm::MCInst loadToReg(unsigned Reg, unsigned RMOpcode) {
-    return llvm::MCInstBuilder(RMOpcode)
-        .addReg(Reg)
-        // Address = ESP
-        .addReg(llvm::X86::RSP) // BaseReg
-        .addImm(1)              // ScaleAmt
-        .addReg(0)              // IndexReg
-        .addImm(0)              // Disp
-        .addReg(0);             // Segment
-  }
-
-  // Releases scratch memory.
-  static llvm::MCInst releaseStackSpace(unsigned Bytes) {
-    return llvm::MCInstBuilder(llvm::X86::ADD64ri8)
-        .addReg(llvm::X86::RSP)
-        .addReg(llvm::X86::RSP)
-        .addImm(Bytes);
-  }
 };
 
 } // namespace
 
 static ExegesisTarget *getTheExegesisX86Target() {
   static ExegesisX86Target Target;
   return &Target;
 }
 
 void InitializeX86ExegesisTarget() {
   ExegesisTarget::registerTarget(getTheExegesisX86Target());
 }
 
 } // namespace exegesis

unittests/tools/llvm-exegesis/AArch64/TargetTest.cpp

 #include "gtest/gtest.h"
 
 namespace exegesis {
 
 void InitializeAArch64ExegesisTarget();
 
 namespace {
 
+using llvm::APInt;
+using llvm::MCInst;
 using testing::Gt;
+using testing::IsEmpty;
+using testing::Not;
 using testing::NotNull;
-using testing::SizeIs;
 
 constexpr const char kTriple[] = "aarch64-unknown-linux";
+constexpr const char kGenericCpu[] = "generic";
+constexpr const char kNoFeatures[] = "";
 
 class AArch64TargetTest : public ::testing::Test {
 protected:
   AArch64TargetTest()
       : ExegesisTarget_(ExegesisTarget::lookup(llvm::Triple(kTriple))) {
     EXPECT_THAT(ExegesisTarget_, NotNull());
     std::string error;
     Target_ = llvm::TargetRegistry::lookupTarget(kTriple, error);
     EXPECT_THAT(Target_, NotNull());
+    STI_.reset(
+        Target_->createMCSubtargetInfo(kTriple, kGenericCpu, kNoFeatures));
   }
+
   static void SetUpTestCase() {
     LLVMInitializeAArch64TargetInfo();
     LLVMInitializeAArch64Target();
     LLVMInitializeAArch64TargetMC();
     InitializeAArch64ExegesisTarget();
   }
 
+  std::vector<MCInst> setRegTo(unsigned Reg, const APInt &Value) {
+    return ExegesisTarget_->setRegTo(*STI_, Reg, Value);
+  }
+
   const llvm::Target *Target_;
   const ExegesisTarget *const ExegesisTarget_;
+  std::unique_ptr<llvm::MCSubtargetInfo> STI_;
 };
 
 TEST_F(AArch64TargetTest, SetRegToConstant) {
-  const std::unique_ptr<llvm::MCSubtargetInfo> STI(
-      Target_->createMCSubtargetInfo(kTriple, "generic", ""));
-  // The AArch64 target currently doesn't know how to set register values
-  const auto Insts = ExegesisTarget_->setRegToConstant(*STI, llvm::AArch64::X0);
-  EXPECT_THAT(Insts, SizeIs(0));
+  // The AArch64 target currently doesn't know how to set register values.
+  const auto Insts = setRegTo(llvm::AArch64::X0, llvm::APInt());
+  EXPECT_THAT(Insts, Not(IsEmpty()));
 }
 
 } // namespace
 } // namespace exegesis

unittests/tools/llvm-exegesis/ARM/AssemblerTest.cpp

static void SetUpTestCase() {
     LLVMInitializeARMTargetInfo();
     LLVMInitializeARMTargetMC();
     LLVMInitializeARMTarget();
     LLVMInitializeARMAsmPrinter();
   }
 };
 
 TEST_F(ARMMachineFunctionGeneratorTest, DISABLED_JitFunction) {
-  Check(ExegesisTarget::getDefault(), {}, llvm::MCInst(), 0x1e, 0xff, 0x2f,
-        0xe1);
+  Check({}, llvm::MCInst(), 0x1e, 0xff, 0x2f, 0xe1);
 }
 
 TEST_F(ARMMachineFunctionGeneratorTest, DISABLED_JitFunctionADDrr) {
-  Check(ExegesisTarget::getDefault(), {llvm::ARM::R0},
+  Check({{llvm::ARM::R0, llvm::APInt()}},
         MCInstBuilder(llvm::ARM::ADDrr)
             .addReg(llvm::ARM::R0)
             .addReg(llvm::ARM::R0)
             .addReg(llvm::ARM::R0)
             .addImm(llvm::ARMCC::AL)
             .addReg(0)
             .addReg(0),
         0x00, 0x00, 0x80, 0xe0, 0x1e, 0xff, 0x2f, 0xe1);
 }
 
 } // namespace
 } // namespace exegesis

unittests/tools/llvm-exegesis/Common/AssemblerUtils.h

 namespace exegesis {
 
 class MachineFunctionGeneratorBaseTest : public ::testing::Test {
 protected:
   MachineFunctionGeneratorBaseTest(const std::string &TT,
                                    const std::string &CpuName)
       : TT(TT), CpuName(CpuName),
         CanExecute(llvm::Triple(TT).getArch() ==
-                   llvm::Triple(llvm::sys::getProcessTriple()).getArch()) {
+                   llvm::Triple(llvm::sys::getProcessTriple()).getArch()),
+        ET(ExegesisTarget::lookup(llvm::Triple(TT))) {
+    assert(ET);
     if (!CanExecute) {
       llvm::outs() << "Skipping execution, host:"
                    << llvm::sys::getProcessTriple() << ", target:" << TT
                    << "\n";
     }
   }
 
   template <class... Bs>
-  inline void Check(const ExegesisTarget &ET,
-                    llvm::ArrayRef<unsigned> RegsToDef, llvm::MCInst MCInst,
-                    Bs... Bytes) {
+  inline void Check(llvm::ArrayRef<RegisterValue> RegisterInitialValues,
+                    llvm::MCInst MCInst, Bs... Bytes) {
     ExecutableFunction Function =
-        (MCInst.getOpcode() == 0) ? assembleToFunction(ET, RegsToDef, {})
-                                  : assembleToFunction(ET, RegsToDef, {MCInst});
+        (MCInst.getOpcode() == 0)
+            ? assembleToFunction(RegisterInitialValues, {})
+            : assembleToFunction(RegisterInitialValues, {MCInst});
     ASSERT_THAT(Function.getFunctionBytes().str(),
                 testing::ElementsAre(Bytes...));
     if (CanExecute) {
       BenchmarkRunner::ScratchSpace Scratch;
       Function(Scratch.ptr());
     }
   }
 
 private:
   std::unique_ptr<llvm::LLVMTargetMachine> createTargetMachine() {
     std::string Error;
     const llvm::Target *TheTarget =
         llvm::TargetRegistry::lookupTarget(TT, Error);
     EXPECT_TRUE(TheTarget) << Error << " " << TT;
     const llvm::TargetOptions Options;
     llvm::TargetMachine *TM = TheTarget->createTargetMachine(
         TT, CpuName, "", Options, llvm::Reloc::Model::Static);
     EXPECT_TRUE(TM) << TT << " " << CpuName;
     return std::unique_ptr<llvm::LLVMTargetMachine>(
         static_cast<llvm::LLVMTargetMachine *>(TM));
   }
 
   ExecutableFunction
-  assembleToFunction(const ExegesisTarget &ET,
-                     llvm::ArrayRef<unsigned> RegsToDef,
+  assembleToFunction(llvm::ArrayRef<RegisterValue> RegisterInitialValues,
                      llvm::ArrayRef<llvm::MCInst> Instructions) {
     llvm::SmallString<256> Buffer;
     llvm::raw_svector_ostream AsmStream(Buffer);
-    assembleToStream(ET, createTargetMachine(), /*LiveIns=*/{},
-                     RegsToDef, Instructions,
-                     AsmStream);
+    assembleToStream(*ET, createTargetMachine(), /*LiveIns=*/{},
+                     RegisterInitialValues, Instructions, AsmStream);
     return ExecutableFunction(createTargetMachine(),
                               getObjectFromBuffer(AsmStream.str()));
   }
 
   const std::string TT;
   const std::string CpuName;
   const bool CanExecute;
+  const ExegesisTarget *const ET;
 };
 
 } // namespace exegesis
 
 #endif

unittests/tools/llvm-exegesis/X86/AssemblerTest.cpp

static void SetUpTestCase() {
     LLVMInitializeX86TargetMC();
     LLVMInitializeX86Target();
     LLVMInitializeX86AsmPrinter();
     InitializeX86ExegesisTarget();
   }
 };
 
 TEST_F(X86MachineFunctionGeneratorTest, DISABLED_JitFunction) {
-  Check(ExegesisTarget::getDefault(), {}, llvm::MCInst(), 0xc3);
-}
-
-TEST_F(X86MachineFunctionGeneratorTest, DISABLED_JitFunctionXOR32rr_Default) {
-  Check(ExegesisTarget::getDefault(), {EAX},
-        MCInstBuilder(XOR32rr).addReg(EAX).addReg(EAX).addReg(EAX), 0x31, 0xc0,
-        0xc3);
+  Check({}, llvm::MCInst(), 0xc3);
 }
 
 TEST_F(X86MachineFunctionGeneratorTest, DISABLED_JitFunctionXOR32rr_X86) {
-  const auto *ET = ExegesisTarget::lookup(llvm::Triple("x86_64-unknown-linux"));
-  ASSERT_NE(ET, nullptr);
-  Check(*ET, {EAX}, MCInstBuilder(XOR32rr).addReg(EAX).addReg(EAX).addReg(EAX),
+  Check({{EAX, llvm::APInt(32, 1)}},
+        MCInstBuilder(XOR32rr).addReg(EAX).addReg(EAX).addReg(EAX),
         // mov eax, 1
         0xb8, 0x01, 0x00, 0x00, 0x00,
         // xor eax, eax
         0x31, 0xc0, 0xc3);
 }
 
 TEST_F(X86MachineFunctionGeneratorTest, DISABLED_JitFunctionMOV64ri) {
-  Check(ExegesisTarget::getDefault(), {},
-        MCInstBuilder(MOV64ri32).addReg(RAX).addImm(42), 0x48, 0xc7, 0xc0, 0x2a,
-        0x00, 0x00, 0x00, 0xc3);
+  Check({}, MCInstBuilder(MOV64ri32).addReg(RAX).addImm(42), 0x48, 0xc7, 0xc0,
+        0x2a, 0x00, 0x00, 0x00, 0xc3);
 }
 
 TEST_F(X86MachineFunctionGeneratorTest, DISABLED_JitFunctionMOV32ri) {
-  Check(ExegesisTarget::getDefault(), {},
-        MCInstBuilder(MOV32ri).addReg(EAX).addImm(42), 0xb8, 0x2a, 0x00, 0x00,
-        0x00, 0xc3);
+  Check({}, MCInstBuilder(MOV32ri).addReg(EAX).addImm(42), 0xb8, 0x2a, 0x00,
+        0x00, 0x00, 0xc3);
 }
 
 } // namespace
 } // namespace exegesis

unittests/tools/llvm-exegesis/X86/SnippetGeneratorTest.cpp

private:
   generateCodeTemplate(unsigned Opcode) const override {
     return llvm::make_error<llvm::StringError>("not implemented",
                                                llvm::inconvertibleErrorCode());
   }
 };
 
 using FakeSnippetGeneratorTest = SnippetGeneratorTest<FakeSnippetGenerator>;
 
-TEST_F(FakeSnippetGeneratorTest, ComputeRegsToDefAdd16ri) {
+testing::Matcher<const RegisterValue &> IsRegisterValue(unsigned Reg,
+                                                        llvm::APInt Value) {
+  return testing::AllOf(testing::Field(&RegisterValue::Register, Reg),
+                        testing::Field(&RegisterValue::Value, Value));
+}
+
+TEST_F(FakeSnippetGeneratorTest, ComputeRegisterInitialValuesAdd16ri) {
   // ADD16ri:
   // explicit def 0       : reg RegClass=GR16
   // explicit use 1       : reg RegClass=GR16 | TIED_TO:0
   // explicit use 2       : imm
   // implicit def         : EFLAGS
   InstructionBuilder IB(Generator.createInstruction(llvm::X86::ADD16ri));
   IB.getValueFor(IB.Instr.Variables[0]) =
       llvm::MCOperand::createReg(llvm::X86::AX);
   std::vector<InstructionBuilder> Snippet;
   Snippet.push_back(std::move(IB));
-  const auto RegsToDef = Generator.computeRegsToDef(Snippet);
-  EXPECT_THAT(RegsToDef, UnorderedElementsAre(llvm::X86::AX));
+  const auto RIV = Generator.computeRegisterInitialValues(Snippet);
+  EXPECT_THAT(RIV, ElementsAre(IsRegisterValue(llvm::X86::AX, llvm::APInt())));
 }
 
-TEST_F(FakeSnippetGeneratorTest, ComputeRegsToDefAdd64rr) {
+TEST_F(FakeSnippetGeneratorTest, ComputeRegisterInitialValuesAdd64rr) {
   // ADD64rr:
   //  mov64ri rax, 42
   //  add64rr rax, rax, rbx
   // -> only rbx needs defining.
   std::vector<InstructionBuilder> Snippet;
   {
     InstructionBuilder Mov(Generator.createInstruction(llvm::X86::MOV64ri));
     Mov.getValueFor(Mov.Instr.Variables[0]) =
         llvm::MCOperand::createReg(llvm::X86::RAX);
     Mov.getValueFor(Mov.Instr.Variables[1]) = llvm::MCOperand::createImm(42);
     Snippet.push_back(std::move(Mov));
   }
   {
     InstructionBuilder Add(Generator.createInstruction(llvm::X86::ADD64rr));
     Add.getValueFor(Add.Instr.Variables[0]) =
         llvm::MCOperand::createReg(llvm::X86::RAX);
     Add.getValueFor(Add.Instr.Variables[1]) =
         llvm::MCOperand::createReg(llvm::X86::RBX);
     Snippet.push_back(std::move(Add));
   }
 
-  const auto RegsToDef = Generator.computeRegsToDef(Snippet);
-  EXPECT_THAT(RegsToDef, UnorderedElementsAre(llvm::X86::RBX));
+  const auto RIV = Generator.computeRegisterInitialValues(Snippet);
+  EXPECT_THAT(RIV, ElementsAre(IsRegisterValue(llvm::X86::RBX, llvm::APInt())));
 }
 
 } // namespace
 } // namespace exegesis

unittests/tools/llvm-exegesis/X86/TargetTest.cpp

 #include "Target.h"
 
 #include <cassert>
 #include <memory>
 
 #include "MCTargetDesc/X86MCTargetDesc.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/TargetSelect.h"
 #include "gmock/gmock.h"
 #include "gtest/gtest.h"
 
+#include "llvm/MC/MCInstPrinter.h"
+
+namespace llvm {
+
+bool operator==(const MCOperand &a, const MCOperand &b) {
+  if (a.isImm() && b.isImm())
+    return a.getImm() == b.getImm();
+  if (a.isReg() && b.isReg())
+    return a.getReg() == b.getReg();
+  return false;
+}
+
+bool operator==(const MCInst &a, const MCInst &b) {
+  if (a.getOpcode() != b.getOpcode())
+    return false;
+  if (a.getNumOperands() != b.getNumOperands())
+    return false;
+  for (unsigned I = 0; I < a.getNumOperands(); ++I) {
+    if (!(a.getOperand(I) == b.getOperand(I)))
+      return false;
+  }
+  return true;
+}
+
+} // namespace llvm
+
 namespace exegesis {
 
 void InitializeX86ExegesisTarget();
 
 namespace {
 
+using testing::AllOf;
+using testing::ElementsAre;
+using testing::ElementsAreArray;
+using testing::Eq;
 using testing::Gt;
+using testing::Matcher;
 using testing::NotNull;
+using testing::Property;
 using testing::SizeIs;
 
+using llvm::APInt;
+using llvm::MCInst;
+using llvm::MCInstBuilder;
+using llvm::MCOperand;
+
+Matcher<MCOperand> IsImm(int64_t Value) {
+  return AllOf(Property(&MCOperand::isImm, Eq(true)),
+               Property(&MCOperand::getImm, Eq(Value)));
+}
+
+Matcher<MCOperand> IsReg(unsigned Reg) {
+  return AllOf(Property(&MCOperand::isReg, Eq(true)),
+               Property(&MCOperand::getReg, Eq(Reg)));
+}
+
+Matcher<MCInst> OpcodeIs(unsigned Opcode) {
+  return Property(&MCInst::getOpcode, Eq(Opcode));
+}
+
+Matcher<MCInst> IsMovImmediate(unsigned Opcode, int64_t Reg, int64_t Value) {
+  return AllOf(OpcodeIs(Opcode), ElementsAre(IsReg(Reg), IsImm(Value)));
+}
+
+Matcher<MCInst> IsMovValueToStack(unsigned Opcode, int64_t Value,
+                                  size_t Offset) {
+  return AllOf(OpcodeIs(Opcode),
+               ElementsAre(IsReg(llvm::X86::RSP), IsImm(1), IsReg(0),
+                           IsImm(Offset), IsReg(0), IsImm(Value)));
+}
+
+Matcher<MCInst> IsMovValueFromStack(unsigned Opcode, unsigned Reg) {
+  return AllOf(OpcodeIs(Opcode),
+               ElementsAre(IsReg(Reg), IsReg(llvm::X86::RSP), IsImm(1),
+                           IsReg(0), IsImm(0), IsReg(0)));
+}
+
+Matcher<MCInst> IsStackAllocate(unsigned Size) {
+  return AllOf(
+      OpcodeIs(llvm::X86::SUB64ri8),
+      ElementsAre(IsReg(llvm::X86::RSP), IsReg(llvm::X86::RSP), IsImm(Size)));
+}
+
+Matcher<MCInst> IsStackDeallocate(unsigned Size) {
+  return AllOf(
+      OpcodeIs(llvm::X86::ADD64ri8),
+      ElementsAre(IsReg(llvm::X86::RSP), IsReg(llvm::X86::RSP), IsImm(Size)));
+}
+
 constexpr const char kTriple[] = "x86_64-unknown-linux";
+constexpr const char kFeaturesEmpty[] = "";
+constexpr const char kFeaturesAvx[] = "+avx";
+constexpr const char kFeaturesAvx512VL[] = "+avx512vl";
+constexpr const char kCpuCore2[] = "core2";
 
+template <const char *CPU, const char *Features>
 class X86TargetTest : public ::testing::Test {
 protected:
   X86TargetTest()
       : ExegesisTarget_(ExegesisTarget::lookup(llvm::Triple(kTriple))) {
     EXPECT_THAT(ExegesisTarget_, NotNull());
     std::string error;
     Target_ = llvm::TargetRegistry::lookupTarget(kTriple, error);
     EXPECT_THAT(Target_, NotNull());
+    STI_.reset(Target_->createMCSubtargetInfo(kTriple, kCpuCore2, Features));
   }
+
   static void SetUpTestCase() {
     LLVMInitializeX86TargetInfo();
     LLVMInitializeX86Target();
     LLVMInitializeX86TargetMC();
     InitializeX86ExegesisTarget();
   }
 
+  std::vector<MCInst> setRegTo(unsigned Reg, const APInt &Value) {
+    return ExegesisTarget_->setRegTo(*STI_, Reg, Value);
+  }
+
   const llvm::Target *Target_;
   const ExegesisTarget *const ExegesisTarget_;
+  std::unique_ptr<llvm::MCSubtargetInfo> STI_;
 };
 
-TEST_F(X86TargetTest, SetRegToConstantGPR) {
-  const std::unique_ptr<llvm::MCSubtargetInfo> STI(
-      Target_->createMCSubtargetInfo(kTriple, "core2", ""));
-  const auto Insts = ExegesisTarget_->setRegToConstant(*STI, llvm::X86::EAX);
-  EXPECT_THAT(Insts, SizeIs(1));
-  EXPECT_EQ(Insts[0].getOpcode(), llvm::X86::MOV32ri);
-  EXPECT_EQ(Insts[0].getOperand(0).getReg(), llvm::X86::EAX);
-}
-
-TEST_F(X86TargetTest, SetRegToConstantXMM_SSE2) {
-  const std::unique_ptr<llvm::MCSubtargetInfo> STI(
-      Target_->createMCSubtargetInfo(kTriple, "core2", ""));
-  const auto Insts = ExegesisTarget_->setRegToConstant(*STI, llvm::X86::XMM1);
-  EXPECT_THAT(Insts, SizeIs(7U));
-  EXPECT_EQ(Insts[0].getOpcode(), llvm::X86::SUB64ri8);
-  EXPECT_EQ(Insts[1].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[2].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[3].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[4].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[5].getOpcode(), llvm::X86::MOVDQUrm);
-  EXPECT_EQ(Insts[6].getOpcode(), llvm::X86::ADD64ri8);
-}
-
-TEST_F(X86TargetTest, SetRegToConstantXMM_AVX) {
-  const std::unique_ptr<llvm::MCSubtargetInfo> STI(
-      Target_->createMCSubtargetInfo(kTriple, "core2", "+avx"));
-  const auto Insts = ExegesisTarget_->setRegToConstant(*STI, llvm::X86::XMM1);
-  EXPECT_THAT(Insts, SizeIs(7U));
-  EXPECT_EQ(Insts[0].getOpcode(), llvm::X86::SUB64ri8);
-  EXPECT_EQ(Insts[1].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[2].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[3].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[4].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[5].getOpcode(), llvm::X86::VMOVDQUrm);
-  EXPECT_EQ(Insts[6].getOpcode(), llvm::X86::ADD64ri8);
-}
-
-TEST_F(X86TargetTest, SetRegToConstantXMM_AVX512) {
-  const std::unique_ptr<llvm::MCSubtargetInfo> STI(
-      Target_->createMCSubtargetInfo(kTriple, "core2", "+avx512vl"));
-  const auto Insts = ExegesisTarget_->setRegToConstant(*STI, llvm::X86::XMM1);
-  EXPECT_THAT(Insts, SizeIs(7U));
-  EXPECT_EQ(Insts[0].getOpcode(), llvm::X86::SUB64ri8);
-  EXPECT_EQ(Insts[1].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[2].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[3].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[4].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[5].getOpcode(), llvm::X86::VMOVDQU32Z128rm);
-  EXPECT_EQ(Insts[6].getOpcode(), llvm::X86::ADD64ri8);
-}
-
-TEST_F(X86TargetTest, SetRegToConstantMMX) {
-  const std::unique_ptr<llvm::MCSubtargetInfo> STI(
-      Target_->createMCSubtargetInfo(kTriple, "core2", ""));
-  const auto Insts = ExegesisTarget_->setRegToConstant(*STI, llvm::X86::MM1);
-  EXPECT_THAT(Insts, SizeIs(5U));
-  EXPECT_EQ(Insts[0].getOpcode(), llvm::X86::SUB64ri8);
-  EXPECT_EQ(Insts[1].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[2].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[3].getOpcode(), llvm::X86::MMX_MOVQ64rm);
-  EXPECT_EQ(Insts[4].getOpcode(), llvm::X86::ADD64ri8);
-}
-
-TEST_F(X86TargetTest, SetRegToConstantYMM_AVX) {
-  const std::unique_ptr<llvm::MCSubtargetInfo> STI(
-      Target_->createMCSubtargetInfo(kTriple, "core2", "+avx"));
-  const auto Insts = ExegesisTarget_->setRegToConstant(*STI, llvm::X86::YMM1);
-  EXPECT_THAT(Insts, SizeIs(11U));
-  EXPECT_EQ(Insts[0].getOpcode(), llvm::X86::SUB64ri8);
-  EXPECT_EQ(Insts[1].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[2].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[3].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[4].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[5].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[6].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[7].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[8].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[9].getOpcode(), llvm::X86::VMOVDQUYrm);
-  EXPECT_EQ(Insts[10].getOpcode(), llvm::X86::ADD64ri8);
-}
-
-TEST_F(X86TargetTest, SetRegToConstantYMM_AVX512) {
-  const std::unique_ptr<llvm::MCSubtargetInfo> STI(
-      Target_->createMCSubtargetInfo(kTriple, "core2", "+avx512vl"));
-  const auto Insts = ExegesisTarget_->setRegToConstant(*STI, llvm::X86::YMM1);
-  EXPECT_THAT(Insts, SizeIs(11U));
-  EXPECT_EQ(Insts[0].getOpcode(), llvm::X86::SUB64ri8);
-  EXPECT_EQ(Insts[1].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[2].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[3].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[4].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[5].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[6].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[7].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[8].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[9].getOpcode(), llvm::X86::VMOVDQU32Z256rm);
-  EXPECT_EQ(Insts[10].getOpcode(), llvm::X86::ADD64ri8);
-}
-
-TEST_F(X86TargetTest, SetRegToConstantZMM_AVX512) {
-  const std::unique_ptr<llvm::MCSubtargetInfo> STI(
-      Target_->createMCSubtargetInfo(kTriple, "core2", "+avx512vl"));
-  const auto Insts = ExegesisTarget_->setRegToConstant(*STI, llvm::X86::ZMM1);
-  EXPECT_THAT(Insts, SizeIs(19U));
-  EXPECT_EQ(Insts[0].getOpcode(), llvm::X86::SUB64ri8);
-  EXPECT_EQ(Insts[1].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[2].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[3].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[4].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[5].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[6].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[7].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[8].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[9].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[10].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[11].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[12].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[13].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[14].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[15].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[16].getOpcode(), llvm::X86::MOV32mi);
-  EXPECT_EQ(Insts[17].getOpcode(), llvm::X86::VMOVDQU32Zrm);
-  EXPECT_EQ(Insts[18].getOpcode(), llvm::X86::ADD64ri8);
+using Core2TargetTest = X86TargetTest<kCpuCore2, kFeaturesEmpty>;
+using Core2AvxTargetTest = X86TargetTest<kCpuCore2, kFeaturesAvx>;
+using Core2Avx512TargetTest = X86TargetTest<kCpuCore2, kFeaturesAvx512VL>;
+
+TEST_F(Core2TargetTest, SetFlags) {
+  const unsigned Reg = llvm::X86::EFLAGS;
+  EXPECT_THAT(
+      setRegTo(Reg, APInt(64, 0x1111222233334444ULL)),
+      ElementsAre(IsStackAllocate(8),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x33334444UL, 0),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x11112222UL, 4),
+                  OpcodeIs(llvm::X86::POPF64)));
+}
+
+TEST_F(Core2TargetTest, SetRegToGR8Value) {
+  const uint8_t Value = 0xFFU;
+  const unsigned Reg = llvm::X86::AL;
+  EXPECT_THAT(setRegTo(Reg, APInt(8, Value)),
+              ElementsAre(IsMovImmediate(llvm::X86::MOV8ri, Reg, Value)));
+}
+
+TEST_F(Core2TargetTest, SetRegToGR16Value) {
+  const uint16_t Value = 0xFFFFU;
+  const unsigned Reg = llvm::X86::BX;
+  EXPECT_THAT(setRegTo(Reg, APInt(16, Value)),
+              ElementsAre(IsMovImmediate(llvm::X86::MOV16ri, Reg, Value)));
+}
+
+TEST_F(Core2TargetTest, SetRegToGR32Value) {
+  const uint32_t Value = 0x7FFFFU;
+  const unsigned Reg = llvm::X86::ECX;
+  EXPECT_THAT(setRegTo(Reg, APInt(32, Value)),
+              ElementsAre(IsMovImmediate(llvm::X86::MOV32ri, Reg, Value)));
+}
+
+TEST_F(Core2TargetTest, SetRegToGR64Value) {
+  const uint64_t Value = 0x7FFFFFFFFFFFFFFFULL;
+  const unsigned Reg = llvm::X86::RDX;
+  EXPECT_THAT(setRegTo(Reg, APInt(64, Value)),
+              ElementsAre(IsMovImmediate(llvm::X86::MOV64ri, Reg, Value)));
+}
+
+TEST_F(Core2TargetTest, SetRegToVR64Value) {
+  EXPECT_THAT(
+      setRegTo(llvm::X86::MM0, APInt(64, 0x1111222233334444ULL)),
+      ElementsAre(IsStackAllocate(8),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x33334444UL, 0),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x11112222UL, 4),
+                  IsMovValueFromStack(llvm::X86::MMX_MOVQ64rm, llvm::X86::MM0),
+                  IsStackDeallocate(8)));
+}
+
+TEST_F(Core2TargetTest, SetRegToVR128Value_Use_MOVDQUrm) {
+  EXPECT_THAT(
+      setRegTo(llvm::X86::XMM0,
+               APInt(128, "11112222333344445555666677778888", 16)),
+      ElementsAre(IsStackAllocate(16),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x77778888UL, 0),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x55556666UL, 4),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x33334444UL, 8),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x11112222UL, 12),
+                  IsMovValueFromStack(llvm::X86::MOVDQUrm, llvm::X86::XMM0),
+                  IsStackDeallocate(16)));
+}
+
+TEST_F(Core2AvxTargetTest, SetRegToVR128Value_Use_VMOVDQUrm) {
+  EXPECT_THAT(
+      setRegTo(llvm::X86::XMM0,
+               APInt(128, "11112222333344445555666677778888", 16)),
+      ElementsAre(IsStackAllocate(16),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x77778888UL, 0),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x55556666UL, 4),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x33334444UL, 8),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x11112222UL, 12),
+                  IsMovValueFromStack(llvm::X86::VMOVDQUrm, llvm::X86::XMM0),
+                  IsStackDeallocate(16)));
+}
+
+TEST_F(Core2Avx512TargetTest, SetRegToVR128Value_Use_VMOVDQU32Z128rm) {
+  EXPECT_THAT(
+      setRegTo(llvm::X86::XMM0,
+               APInt(128, "11112222333344445555666677778888", 16)),
+      ElementsAre(
+          IsStackAllocate(16),
+          IsMovValueToStack(llvm::X86::MOV32mi, 0x77778888UL, 0),
+          IsMovValueToStack(llvm::X86::MOV32mi, 0x55556666UL, 4),
+          IsMovValueToStack(llvm::X86::MOV32mi, 0x33334444UL, 8),
+          IsMovValueToStack(llvm::X86::MOV32mi, 0x11112222UL, 12),
+          IsMovValueFromStack(llvm::X86::VMOVDQU32Z128rm, llvm::X86::XMM0),
+          IsStackDeallocate(16)));
+}
+
+TEST_F(Core2AvxTargetTest, SetRegToVR256Value_Use_VMOVDQUYrm) {
+  const char ValueStr[] =
+      "1111111122222222333333334444444455555555666666667777777788888888";
+  EXPECT_THAT(setRegTo(llvm::X86::YMM0, APInt(256, ValueStr, 16)),
+              ElementsAreArray(
+                  {IsStackAllocate(32),
+                   IsMovValueToStack(llvm::X86::MOV32mi, 0x88888888UL, 0),
+                   IsMovValueToStack(llvm::X86::MOV32mi, 0x77777777UL, 4),
+                   IsMovValueToStack(llvm::X86::MOV32mi, 0x66666666UL, 8),
+                   IsMovValueToStack(llvm::X86::MOV32mi, 0x55555555UL, 12),
+                   IsMovValueToStack(llvm::X86::MOV32mi, 0x44444444UL, 16),
+                   IsMovValueToStack(llvm::X86::MOV32mi, 0x33333333UL, 20),
+                   IsMovValueToStack(llvm::X86::MOV32mi, 0x22222222UL, 24),
+                   IsMovValueToStack(llvm::X86::MOV32mi, 0x11111111UL, 28),
+                   IsMovValueFromStack(llvm::X86::VMOVDQUYrm, llvm::X86::YMM0),
+                   IsStackDeallocate(32)}));
+}
+
+TEST_F(Core2Avx512TargetTest, SetRegToVR256Value_Use_VMOVDQU32Z256rm) {
+  const char ValueStr[] =
+      "1111111122222222333333334444444455555555666666667777777788888888";
+  EXPECT_THAT(
+      setRegTo(llvm::X86::YMM0, APInt(256, ValueStr, 16)),
+      ElementsAreArray(
+          {IsStackAllocate(32),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x88888888UL, 0),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x77777777UL, 4),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x66666666UL, 8),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x55555555UL, 12),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x44444444UL, 16),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x33333333UL, 20),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x22222222UL, 24),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x11111111UL, 28),
+           IsMovValueFromStack(llvm::X86::VMOVDQU32Z256rm, llvm::X86::YMM0),
+           IsStackDeallocate(32)}));
+}
+
+TEST_F(Core2Avx512TargetTest, SetRegToVR512Value) {
+  const char ValueStr[] =
+      "1111111122222222333333334444444455555555666666667777777788888888"
+      "99999999AAAAAAAABBBBBBBBCCCCCCCCDDDDDDDDEEEEEEEEFFFFFFFF00000000";
+  EXPECT_THAT(
+      setRegTo(llvm::X86::ZMM0, APInt(512, ValueStr, 16)),
+      ElementsAreArray(
+          {IsStackAllocate(64),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x00000000UL, 0),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0xFFFFFFFFUL, 4),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0xEEEEEEEEUL, 8),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0xDDDDDDDDUL, 12),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0xCCCCCCCCUL, 16),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0xBBBBBBBBUL, 20),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0xAAAAAAAAUL, 24),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x99999999UL, 28),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x88888888UL, 32),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x77777777UL, 36),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x66666666UL, 40),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x55555555UL, 44),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x44444444UL, 48),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x33333333UL, 52),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x22222222UL, 56),
+           IsMovValueToStack(llvm::X86::MOV32mi, 0x11111111UL, 60),
+           IsMovValueFromStack(llvm::X86::VMOVDQU32Zrm, llvm::X86::ZMM0),
+           IsStackDeallocate(64)}));
+}
+
+TEST_F(Core2TargetTest, SetRegToST0_32Bits) {
+  EXPECT_THAT(
+      setRegTo(llvm::X86::ST0, APInt(32, 0x11112222ULL)),
+      ElementsAre(IsStackAllocate(4),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x11112222UL, 0),
+                  OpcodeIs(llvm::X86::LD_F32m), IsStackDeallocate(4)));
+}
+
+TEST_F(Core2TargetTest, SetRegToST1_32Bits) {
+  const MCInst CopySt0ToSt1 =
+      llvm::MCInstBuilder(llvm::X86::ST_Frr).addReg(llvm::X86::ST1);
+  EXPECT_THAT(
+      setRegTo(llvm::X86::ST1, APInt(32, 0x11112222ULL)),
+      ElementsAre(IsStackAllocate(4),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x11112222UL, 0),
+                  OpcodeIs(llvm::X86::LD_F32m), CopySt0ToSt1,
+                  IsStackDeallocate(4)));
+}
+
+TEST_F(Core2TargetTest, SetRegToST0_64Bits) {
+  EXPECT_THAT(
+      setRegTo(llvm::X86::ST0, APInt(64, 0x1111222233334444ULL)),
+      ElementsAre(IsStackAllocate(8),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x33334444UL, 0),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x11112222UL, 4),
+                  OpcodeIs(llvm::X86::LD_F64m), IsStackDeallocate(8)));
+}
+
+TEST_F(Core2TargetTest, SetRegToST0_80Bits) {
+  EXPECT_THAT(
+      setRegTo(llvm::X86::ST0, APInt(80, "11112222333344445555", 16)),
+      ElementsAre(IsStackAllocate(10),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x44445555UL, 0),
+                  IsMovValueToStack(llvm::X86::MOV32mi, 0x22223333UL, 4),
+                  IsMovValueToStack(llvm::X86::MOV16mi, 0x1111UL, 8),
+                  OpcodeIs(llvm::X86::LD_F80m), IsStackDeallocate(10)));
 }
 
 } // namespace
 } // namespace exegesis