[llvm-exegesis] Add a snippet generator to generate snippets to compute ROB sizes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@346394 91177308-0d34-0410-b5e6-96231b3b80d8

tools/llvm-exegesis/lib/Assembler.cpp

@@ -32,10 +32,19 @@ static constexpr const char FunctionID[] = "foo";
 static std::vector<llvm::MCInst>
 generateSnippetSetupCode(const ExegesisTarget &ET,
                          const llvm::MCSubtargetInfo *const MSI,
+                         const unsigned ScratchReg,
+                         llvm::ArrayRef<unsigned> ScratchRegisterCopies,
                          llvm::ArrayRef<RegisterValue> RegisterInitialValues,
                          bool &IsSnippetSetupComplete) {
   IsSnippetSetupComplete = true;
   std::vector<llvm::MCInst> Result;
+  // Copy registers.
+  for (const unsigned Reg : ScratchRegisterCopies) {
+    assert(ScratchReg > 0 && "scratch reg copies but no scratch reg");
+    const auto CopyRegisterCode = ET.copyReg(*MSI, Reg, ScratchReg);
+    Result.insert(Result.end(), CopyRegisterCode.begin(), CopyRegisterCode.end());
+  }
+  // Load values in registers.
   for (const RegisterValue &RV : RegisterInitialValues) {
     // Load a constant in the register.
     const auto SetRegisterCode = ET.setRegTo(*MSI, RV.Register, RV.Value);
@@ -155,6 +164,7 @@ llvm::BitVector getFunctionReservedRegs(const llvm::TargetMachine &TM) {
 void assembleToStream(const ExegesisTarget &ET,
                       std::unique_ptr<llvm::LLVMTargetMachine> TM,
                       llvm::ArrayRef<unsigned> LiveIns,
+                      llvm::ArrayRef<unsigned> ScratchRegisterCopies,
                       llvm::ArrayRef<RegisterValue> RegisterInitialValues,
                       llvm::ArrayRef<llvm::MCInst> Instructions,
                       llvm::raw_pwrite_stream &AsmStream) {
@@ -178,7 +188,7 @@ void assembleToStream(const ExegesisTarget &ET,
 
   bool IsSnippetSetupComplete;
   std::vector<llvm::MCInst> Code =
-      generateSnippetSetupCode(ET, TM->getMCSubtargetInfo(),
+      generateSnippetSetupCode(ET, TM->getMCSubtargetInfo(), ET.getScratchMemoryRegister(TM->getTargetTriple()), ScratchRegisterCopies,
                                RegisterInitialValues, IsSnippetSetupComplete);
 
   Code.insert(Code.end(), Instructions.begin(), Instructions.end());
@@ -199,7 +209,7 @@ void assembleToStream(const ExegesisTarget &ET,
   llvm::MCContext &MCContext = MMI->getContext();
   llvm::legacy::PassManager PM;
 
-  llvm::TargetLibraryInfoImpl TLII(llvm::Triple(Module->getTargetTriple()));
+  llvm::TargetLibraryInfoImpl TLII(Triple(Module->getTargetTriple()));
   PM.add(new llvm::TargetLibraryInfoWrapperPass(TLII));
 
   llvm::TargetPassConfig *TPC = TM->createPassConfig(PM);

tools/llvm-exegesis/lib/Assembler.h

@@ -48,6 +48,7 @@ llvm::BitVector getFunctionReservedRegs(const llvm::TargetMachine &TM);
 void assembleToStream(const ExegesisTarget &ET,
                       std::unique_ptr<llvm::LLVMTargetMachine> TM,
                       llvm::ArrayRef<unsigned> LiveIns,
+                      llvm::ArrayRef<unsigned> ScratchRegisterCopies,
                       llvm::ArrayRef<RegisterValue> RegisterInitialValues,
                       llvm::ArrayRef<llvm::MCInst> Instructions,
                       llvm::raw_pwrite_stream &AsmStream);

tools/llvm-exegesis/lib/BenchmarkCode.h

@@ -27,6 +27,10 @@ struct BenchmarkCode {
   // registers initial values.
   std::vector<RegisterValue> RegisterInitialValues;
 
+  // Before the code is executed some instructions are added to copy the
+  // scratch register into the specified registers.
+  std::vector<unsigned> ScratchRegisterCopies;
+
   // We also need to provide the registers that are live on entry for the
   // assembler to generate proper prologue/epilogue.
   std::vector<unsigned> LiveIns;

tools/llvm-exegesis/lib/BenchmarkResult.h

@@ -58,7 +58,7 @@ struct BenchmarkMeasure {
 // The result of an instruction benchmark.
 struct InstructionBenchmark {
   InstructionBenchmarkKey Key;
-  enum ModeE { Unknown, Latency, Uops };
+  enum ModeE { Unknown, Latency, Uops, ROBSize };
   ModeE Mode;
   std::string CpuName;
   std::string LLVMTriple;

tools/llvm-exegesis/lib/BenchmarkRunner.cpp

@@ -168,7 +168,7 @@ BenchmarkRunner::writeObjectFile(const BenchmarkCode &BC,
     return std::move(E);
   llvm::raw_fd_ostream OFS(ResultFD, true /*ShouldClose*/);
   assembleToStream(State.getExegesisTarget(), State.createTargetMachine(),
-                   BC.LiveIns, BC.RegisterInitialValues, Code, OFS);
+                   BC.LiveIns, BC.ScratchRegisterCopies, BC.RegisterInitialValues, Code, OFS);
   return ResultPath.str();
 }
 

tools/llvm-exegesis/lib/CMakeLists.txt

@@ -23,6 +23,7 @@ add_library(LLVMExegesis
   LlvmState.cpp
   MCInstrDescView.cpp
   PerfHelper.cpp
+  ROBSize.cpp
   RegisterAliasing.cpp
   SnippetGenerator.cpp
   RegisterValue.cpp

tools/llvm-exegesis/lib/CodeTemplate.h

@@ -17,6 +17,7 @@
 #define LLVM_TOOLS_LLVM_EXEGESIS_CODETEMPLATE_H
 
 #include "MCInstrDescView.h"
+#include "RegisterValue.h"
 #include "llvm/ADT/BitmaskEnum.h"
 
 namespace llvm {
@@ -120,6 +121,9 @@ struct CodeTemplate {
   std::string Info;
   // The list of the instructions for this template.
   std::vector<InstructionTemplate> Instructions;
+  // The list of registers in which to copy the scratch register as a setup
+  // step.
+  std::vector<unsigned> ScratchRegisterCopies;
   // If the template uses the provided scratch memory, the register in which
   // the pointer to this memory is passed in to the function.
   unsigned ScratchSpacePointerInReg = 0;

tools/llvm-exegesis/lib/ROBSize.cpp

@@ -0,0 +1,69 @@
+//===-- Uops.cpp ------------------------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "ROBSize.h"
+
+#include "Assembler.h"
+#include "BenchmarkRunner.h"
+#include "MCInstrDescView.h"
+#include "Target.h"
+
+namespace llvm {
+namespace exegesis {
+
+ROBSizeSnippetGenerator::~ROBSizeSnippetGenerator() = default;
+
+llvm::Expected<std::vector<CodeTemplate>>
+ROBSizeSnippetGenerator::generateCodeTemplates(const Instruction &Instr) const {
+  CodeTemplate CT;
+  // const llvm::BitVector *ScratchSpaceAliasedRegs = nullptr;
+  const auto &ET = State.getExegesisTarget();
+  const auto &TM = State.getTargetMachine();
+
+  CT.ScratchSpacePointerInReg =
+      ET.getScratchMemoryRegister(TM.getTargetTriple());
+  if (CT.ScratchSpacePointerInReg == 0)
+    return llvm::make_error<BenchmarkFailure>(
+        "Infeasible : target does not support memory instructions");
+  // ScratchSpaceAliasedRegs =
+  //      &State.getRATC().getRegister(CT.ScratchSpacePointerInReg).aliasedBits();
+
+  const unsigned ECX = 50u; // FIXME: pick any available register.
+  const unsigned EDX = 52u; // FIXME: pick any available register.
+  CT.ScratchRegisterCopies.push_back(ECX);
+  CT.ScratchRegisterCopies.push_back(EDX);
+
+  /*
+    const llvm::TargetInstrInfo *const TII =
+    State.getSubtargetInfo().getInstrInfo(); MCInst NopInst;
+    TII->getNoop(NopInst);
+    */
+  Instruction ChaseRegInst(State.getInstrInfo(), State.getRATC(), ET.getChaseRegOpcode());
+  //errs() << ChaseRegInst.Variables.size() << "\n";
+  assert(ChaseRegInst.Variables.size() >= 2 && "'mov reg, [reg]'' should have at least two variables");
+  InstructionTemplate IT1(ChaseRegInst);
+  IT1.getValueFor(ChaseRegInst.Variables[0]) = MCOperand::createReg(ECX);
+  ET.fillMemoryOperands(IT1, ECX, 0);
+  CT.Instructions.push_back(std::move(IT1));
+  InstructionTemplate IT2(ChaseRegInst);
+  IT2.getValueFor(ChaseRegInst.Variables[0]) = MCOperand::createReg(EDX);
+  ET.fillMemoryOperands(IT2, EDX, 0);
+  CT.Instructions.push_back(std::move(IT2));
+
+  // const auto &ReservedRegisters = State.getRATC().reservedRegisters();
+  // No tied variables, we pick random values for defs.
+  llvm::BitVector Defs(State.getRegInfo().getNumRegs());
+  CT.Info =
+      "instruction has no tied variables picking Uses different from defs";
+  // CT.Instructions.push_back(std::move(IT));
+  return getSingleton(std::move(CT));
+}
+
+} // namespace exegesis
+} // namespace llvm

tools/llvm-exegesis/lib/ROBSize.h

@@ -0,0 +1,36 @@
+//===-- Uops.h --------------------------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// A BenchmarkRunner implementation to measure uop decomposition.
+///
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_TOOLS_LLVM_EXEGESIS_ROBSIZE_H
+#define LLVM_TOOLS_LLVM_EXEGESIS_ROBSIZE_H
+
+#include "BenchmarkRunner.h"
+#include "SnippetGenerator.h"
+
+namespace llvm {
+namespace exegesis {
+
+class ROBSizeSnippetGenerator : public SnippetGenerator {
+public:
+  ROBSizeSnippetGenerator(const LLVMState &State) : SnippetGenerator(State) {}
+  ~ROBSizeSnippetGenerator() override;
+
+  llvm::Expected<std::vector<CodeTemplate>>
+  generateCodeTemplates(const Instruction &Instr) const override;
+};
+
+} // namespace exegesis
+} // namespace llvm
+
+#endif // LLVM_TOOLS_LLVM_EXEGESIS_ROBSIZE_H

tools/llvm-exegesis/lib/RegisterValue.h

@@ -14,6 +14,9 @@
 ///
 //===----------------------------------------------------------------------===//
 
+#ifndef LLVM_TOOLS_LLVM_EXEGESIS_REGISTERVALUE_H
+#define LLVM_TOOLS_LLVM_EXEGESIS_REGISTERVALUE_H
+
 #include <llvm/ADT/APFloat.h>
 #include <llvm/ADT/APInt.h>
 
@@ -22,6 +25,7 @@ namespace exegesis {
 
 // A simple object storing the value for a particular register.
 struct RegisterValue {
+  static RegisterValue zero(unsigned Reg) { return {Reg, llvm::APInt()}; }
   unsigned Register;
   llvm::APInt Value;
 };
@@ -45,3 +49,5 @@ llvm::APInt bitcastFloatValue(const llvm::fltSemantics &FltSemantics,
 
 } // namespace exegesis
 } // namespace llvm
+
+#endif // LLVM_TOOLS_LLVM_EXEGESIS_REGISTERVALUE_H

tools/llvm-exegesis/lib/SnippetGenerator.cpp

@@ -56,8 +56,9 @@ SnippetGenerator::generateConfigurations(const Instruction &Instr) const {
         }
         if (CT.ScratchSpacePointerInReg)
           BC.LiveIns.push_back(CT.ScratchSpacePointerInReg);
+        BC.ScratchRegisterCopies = CT.ScratchRegisterCopies;
         BC.RegisterInitialValues =
-            computeRegisterInitialValues(CT.Instructions);
+            computeRegisterInitialValues(BC.ScratchRegisterCopies, CT.Instructions);
         Output.push_back(std::move(BC));
       }
     }
@@ -67,12 +68,15 @@ SnippetGenerator::generateConfigurations(const Instruction &Instr) const {
 }
 
 std::vector<RegisterValue> SnippetGenerator::computeRegisterInitialValues(
+  const std::vector<unsigned> &ScratchRegisterCopies,
     const std::vector<InstructionTemplate> &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 = State.getRATC().emptyRegisters();
+  for (const auto& Reg : ScratchRegisterCopies)
+    DefinedRegs.set(Reg);
   std::vector<RegisterValue> RIV;
   for (const InstructionTemplate &IT : Instructions) {
     // Returns the register that this Operand sets or uses, or 0 if this is not
@@ -91,7 +95,7 @@ std::vector<RegisterValue> SnippetGenerator::computeRegisterInitialValues(
       if (Op.isUse()) {
         const unsigned Reg = GetOpReg(Op);
         if (Reg > 0 && !DefinedRegs.test(Reg)) {
-          RIV.push_back(RegisterValue{Reg, llvm::APInt()});
+          RIV.push_back(RegisterValue::zero(Reg));
           DefinedRegs.set(Reg);
         }
       }

tools/llvm-exegesis/lib/SnippetGenerator.h

@@ -62,6 +62,7 @@ public:
 
   // Given a snippet, computes which registers the setup code needs to define.
   std::vector<RegisterValue> computeRegisterInitialValues(
+    const std::vector<unsigned> &ScratchRegisterCopies,
       const std::vector<InstructionTemplate> &Snippet) const;
 
 protected:

tools/llvm-exegesis/lib/Target.cpp

@@ -9,6 +9,7 @@
 #include "Target.h"
 
 #include "Latency.h"
+#include "ROBSize.h"
 #include "Uops.h"
 
 namespace llvm {
@@ -37,6 +38,31 @@ void ExegesisTarget::registerTarget(ExegesisTarget *Target) {
   FirstTarget = Target;
 }
 
+std::unique_ptr<SnippetGenerator>
+ExegesisTarget::createLatencySnippetGenerator(const LLVMState &State) const {
+  return llvm::make_unique<LatencySnippetGenerator>(State);
+}
+
+std::unique_ptr<SnippetGenerator>
+ExegesisTarget::createUopsSnippetGenerator(const LLVMState &State) const {
+  return llvm::make_unique<UopsSnippetGenerator>(State);
+}
+
+std::unique_ptr<SnippetGenerator>
+static createROBSizeSnippetGenerator(const LLVMState &State) {
+  return llvm::make_unique<ROBSizeSnippetGenerator>(State);
+}
+
+std::unique_ptr<BenchmarkRunner>
+ExegesisTarget::createLatencyBenchmarkRunner(const LLVMState &State) const {
+  return llvm::make_unique<LatencyBenchmarkRunner>(State);
+}
+
+std::unique_ptr<BenchmarkRunner>
+ExegesisTarget::createUopsBenchmarkRunner(const LLVMState &State) const {
+  return llvm::make_unique<UopsBenchmarkRunner>(State);
+}
+
 std::unique_ptr<SnippetGenerator>
 ExegesisTarget::createSnippetGenerator(InstructionBenchmark::ModeE Mode,
                                        const LLVMState &State) const {
@@ -47,6 +73,8 @@ ExegesisTarget::createSnippetGenerator(InstructionBenchmark::ModeE Mode,
     return createLatencySnippetGenerator(State);
   case InstructionBenchmark::Uops:
     return createUopsSnippetGenerator(State);
+  case InstructionBenchmark::ROBSize:
+    return createROBSizeSnippetGenerator(State);
   }
   return nullptr;
 }
@@ -58,6 +86,7 @@ ExegesisTarget::createBenchmarkRunner(InstructionBenchmark::ModeE Mode,
   case InstructionBenchmark::Unknown:
     return nullptr;
   case InstructionBenchmark::Latency:
+  case InstructionBenchmark::ROBSize:
     return createLatencyBenchmarkRunner(State);
   case InstructionBenchmark::Uops:
     return createUopsBenchmarkRunner(State);
@@ -65,26 +94,6 @@ ExegesisTarget::createBenchmarkRunner(InstructionBenchmark::ModeE Mode,
   return nullptr;
 }
 
-std::unique_ptr<SnippetGenerator>
-ExegesisTarget::createLatencySnippetGenerator(const LLVMState &State) const {
-  return llvm::make_unique<LatencySnippetGenerator>(State);
-}
-
-std::unique_ptr<SnippetGenerator>
-ExegesisTarget::createUopsSnippetGenerator(const LLVMState &State) const {
-  return llvm::make_unique<UopsSnippetGenerator>(State);
-}
-
-std::unique_ptr<BenchmarkRunner>
-ExegesisTarget::createLatencyBenchmarkRunner(const LLVMState &State) const {
-  return llvm::make_unique<LatencyBenchmarkRunner>(State);
-}
-
-std::unique_ptr<BenchmarkRunner>
-ExegesisTarget::createUopsBenchmarkRunner(const LLVMState &State) const {
-  return llvm::make_unique<UopsBenchmarkRunner>(State);
-}
-
 static_assert(std::is_pod<PfmCountersInfo>::value,
               "We shouldn't have dynamic initialization here");
 const PfmCountersInfo PfmCountersInfo::Default = {nullptr, nullptr, nullptr, 0u};
@@ -123,6 +132,11 @@ private:
     llvm_unreachable("Not yet implemented");
   }
 
+  std::vector<llvm::MCInst> copyReg(const llvm::MCSubtargetInfo &STI,
+                                     unsigned ToReg, unsigned FromReg) const override {
+    llvm_unreachable("Not yet implemented");
+  }
+
   bool matchesArch(llvm::Triple::ArchType Arch) const override {
     llvm_unreachable("never called");
     return false;

tools/llvm-exegesis/lib/Target.h

@@ -76,6 +76,11 @@ public:
   setRegTo(const llvm::MCSubtargetInfo &STI, unsigned Reg,
            const llvm::APInt &Value) const = 0;
 
+  // Generates code to copy `FromReg` to `ToReg`.
+  // Precondition: Registers must be the same size.
+  virtual std::vector<llvm::MCInst>
+  copyReg(const llvm::MCSubtargetInfo &STI, unsigned ToReg, unsigned FromReg) 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.
@@ -83,10 +88,16 @@ public:
     return 0;
   }
 
+  // Returns the opcode to move the value at `[Reg]` into `Reg`, where `Reg` is
+  // the from the same register class as getScratchMemoryRegister().
+  virtual unsigned getChaseRegOpcode() const {
+    llvm_unreachable(
+      "fillMemoryOperands() requires getScratchMemoryRegister() > 0");
+  }
+
   // Fills memory operands with references to the address at [Reg] + Offset.
   virtual void fillMemoryOperands(InstructionTemplate &IT, unsigned Reg,
                                   unsigned Offset) const {
-
     llvm_unreachable(
         "fillMemoryOperands() requires getScratchMemoryRegister() > 0");
   }

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

@@ -484,6 +484,19 @@ private:
     return {}; // Not yet implemented.
   }
 
+  std::vector<llvm::MCInst> copyReg(const llvm::MCSubtargetInfo &STI,
+                                    unsigned ToReg,
+                                    unsigned FromReg) const override {
+    if (llvm::X86::GR64RegClass.contains(ToReg))
+      assert(llvm::X86::GR64RegClass.contains(FromReg) && "registers must be the same size");
+      return {llvm::MCInstBuilder(X86::MOV64rr).addReg(ToReg).addReg(FromReg)};
+    return {}; // Not yet implemented.
+  }
+
+  unsigned getChaseRegOpcode() const override {
+    return X86::MOV64rm;
+  }
+
   std::unique_ptr<SnippetGenerator>
   createLatencySnippetGenerator(const LLVMState &State) const override {
     return llvm::make_unique<X86LatencySnippetGenerator>(State);

tools/llvm-exegesis/llvm-exegesis.cpp

@@ -63,6 +63,8 @@ static cl::opt<exegesis::InstructionBenchmark::ModeE>
                                         "latency", "Instruction Latency"),
                              clEnumValN(exegesis::InstructionBenchmark::Uops,
                                         "uops", "Uop Decomposition"),
+                             clEnumValN(exegesis::InstructionBenchmark::ROBSize,
+                                        "rob_size", "ROB Size"),
                              // When not asking for a specific benchmark mode,
                              // we'll analyse the results.
                              clEnumValN(exegesis::InstructionBenchmark::Unknown,
@@ -201,7 +203,6 @@ public:
       return;
     if (CommentText.consume_front("DEFREG")) {
       // LLVM-EXEGESIS-DEFREF <reg> <hex_value>
-      RegisterValue RegVal;
       llvm::SmallVector<llvm::StringRef, 2> Parts;
       CommentText.split(Parts, ' ', /*unlimited splits*/ -1,
                         /*do not keep empty strings*/ false);
@@ -210,6 +211,7 @@ public:
                      << "\n";
         ++InvalidComments;
       }
+      RegisterValue RegVal;
       if (!(RegVal.Register = findRegisterByName(Parts[0].trim()))) {
         llvm::errs() << "unknown register in 'LLVM-EXEGESIS-DEFREG "
                      << CommentText << "\n";