feat(CodeGen): Add PrefetchInsertion pass

Author: rlavaee

llvm/include/llvm/CodeGen/MachineBasicBlock.h

@@ -1299,7 +1299,7 @@ class MachineBasicBlock
 
   MCSymbol *getCallInstSymbol(unsigned CallInstNumber) const;
 
-  const SmallVector<MCSymbol *, 4>& getCallInstSymbols() const {
+  const SmallVector<MCSymbol *, 4> &getCallInstSymbols() const {
     return CallInstSymbols;
   }
 

llvm/include/llvm/CodeGen/MachineInstr.h

@@ -123,8 +123,9 @@ class MachineInstr
     NoUSWrap = 1 << 20,      // Instruction supports geps
                              // no unsigned signed wrap.
     SameSign = 1 << 21,      // Both operands have the same sign.
-    InBounds = 1 << 22       // Pointer arithmetic remains inbounds.
+    InBounds = 1 << 22,      // Pointer arithmetic remains inbounds.
                              // Implies NoUSWrap.
+    Prefetch = 1 << 23,      // Instruction is a prefetch.
   };
 
 private:

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

@@ -18,7 +18,6 @@
 #include "WasmException.h"
 #include "WinCFGuard.h"
 #include "WinException.h"
-#include "llvm/Support/SMLoc.h"
 #include "llvm/ADT/APFloat.h"
 #include "llvm/ADT/APInt.h"
 #include "llvm/ADT/BitmaskEnum.h"
@@ -120,6 +119,7 @@
 #include "llvm/Support/Format.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/Path.h"
+#include "llvm/Support/SMLoc.h"
 #include "llvm/Support/VCSRevision.h"
 #include "llvm/Support/VirtualFileSystem.h"
 #include "llvm/Support/raw_ostream.h"

llvm/lib/CodeGen/BasicBlockSections.cpp

@@ -106,7 +106,8 @@ class BasicBlockSections : public MachineFunctionPass {
 public:
   static char ID;
 
-  BasicBlockSectionsProfileReaderWrapperPass *BBSectionsProfileReader = nullptr;
+  // BasicBlockSectionsProfileReaderWrapperPass *BBSectionsProfileReader =
+  // nullptr;
 
   BasicBlockSections() : MachineFunctionPass(ID) {
     initializeBasicBlockSectionsPass(*PassRegistry::getPassRegistry());

llvm/lib/CodeGen/CodeGenPrepare.cpp

@@ -22,6 +22,7 @@
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/Analysis/BlockFrequencyInfo.h"
 #include "llvm/Analysis/BranchProbabilityInfo.h"
 #include "llvm/Analysis/FloatingPointPredicateUtils.h"

llvm/lib/Target/X86/PrefetchInsertion.cpp

@@ -0,0 +1,209 @@
+//===-- PrefetchInsertion.cpp ---=========-----------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// Prefetch insertion pass implementation.
+//===----------------------------------------------------------------------===//
+/// Prefetch insertion pass.
+//===----------------------------------------------------------------------===//
+
+#include "MCTargetDesc/X86MCTargetDesc.h"
+#include "X86.h"
+#include "X86InstrBuilder.h"
+#include "X86InstrInfo.h"
+#include "X86MachineFunctionInfo.h"
+#include "X86Subtarget.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/StringExtras.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/CodeGen/BasicBlockSectionUtils.h"
+#include "llvm/CodeGen/BasicBlockSectionsProfileReader.h"
+#include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineOperand.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/CodeGen/TargetInstrInfo.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/Target/TargetMachine.h"
+#include <map>
+
+using namespace llvm;
+#define DEBUG_TYPE "prefetchinsertion"
+
+static cl::opt<bool> UseCodePrefetchInstruction(
+    "use-code-prefetch-instruction",
+    cl::desc("Whether to use the new prefetchit1 instruction."), cl::init(true),
+    cl::Hidden);
+static cl::opt<bool> PrefetchNextAddress(
+    "prefetch-next-address",
+    cl::desc(
+        "Whether to prefetch the next address instead of the target address."),
+    cl::init(false), cl::Hidden);
+
+namespace {} // end anonymous namespace
+
+namespace llvm {
+class PrefetchInsertion : public MachineFunctionPass {
+public:
+  static char ID;
+
+  BasicBlockSectionsProfileReaderWrapperPass *BBSectionsProfileReader = nullptr;
+
+  PrefetchInsertion() : MachineFunctionPass(ID) {
+    initializePrefetchInsertionPass(*PassRegistry::getPassRegistry());
+  }
+
+  StringRef getPassName() const override { return "Prefetch Insertion Pass"; }
+
+  void getAnalysisUsage(AnalysisUsage &AU) const override;
+
+  /// Identify basic blocks that need separate sections and prepare to emit them
+  /// accordingly.
+  bool runOnMachineFunction(MachineFunction &MF) override;
+};
+
+} // namespace llvm
+
+char PrefetchInsertion::ID = 0;
+INITIALIZE_PASS_BEGIN(
+    PrefetchInsertion, "prefetch-insertion",
+    "Applies path clonings for the -basic-block-sections=list option", false,
+    false)
+INITIALIZE_PASS_DEPENDENCY(BasicBlockSectionsProfileReaderWrapperPass)
+INITIALIZE_PASS_END(
+    PrefetchInsertion, "prefetch-insertion",
+    "Applies path clonings for the -basic-block-sections=list option", false,
+    false)
+
+bool PrefetchInsertion::runOnMachineFunction(MachineFunction &MF) {
+  assert(MF.getTarget().getBBSectionsType() == BasicBlockSection::List &&
+         "BB Sections list not enabled!");
+  if (hasInstrProfHashMismatch(MF))
+    return false;
+  // errs() << "Running on " << MF.getName() << "\n";
+  Function &F = MF.getFunction();
+  auto PtrTy = PointerType::getUnqual(F.getParent()->getContext());
+  DenseSet<BBPosition> PrefetchTargets =
+      getAnalysis<BasicBlockSectionsProfileReaderWrapperPass>()
+          .getPrefetchTargetsForFunction(MF.getName());
+  // errs() << "Targets: Function: " << F.getName() << " "
+  //        << PrefetchTargets.size() << "\n";
+  DenseMap<UniqueBBID, SmallVector<unsigned>> PrefetchTargetsByBBID;
+  for (const auto &P : PrefetchTargets)
+    PrefetchTargetsByBBID[P.BBID].push_back(P.BBOffset);
+  for (auto &[BBID, V] : PrefetchTargetsByBBID)
+    llvm::sort(V);
+  for (auto &BB : MF)
+    BB.setPrefetchTargets(PrefetchTargetsByBBID[*BB.getBBID()]);
+
+  for (const BBPosition &P : PrefetchTargets) {
+    SmallString<128> PrefetchTargetName("__llvm_prefetch_target_");
+    PrefetchTargetName += F.getName();
+    PrefetchTargetName += "_";
+    PrefetchTargetName += utostr(P.BBID.BaseID);
+    PrefetchTargetName += "_";
+    PrefetchTargetName += utostr(P.BBOffset);
+    F.getParent()->getOrInsertGlobal(PrefetchTargetName, PtrTy);
+  }
+
+  SmallVector<PrefetchHint> PrefetchHints =
+      getAnalysis<BasicBlockSectionsProfileReaderWrapperPass>()
+          .getPrefetchHintsForFunction(MF.getName());
+  // errs() << "Hints: Function: " << F.getName() << " " << PrefetchHints.size()
+  //        << "\n";
+  for (const PrefetchHint &H : PrefetchHints) {
+    SmallString<128> PrefetchTargetName("__llvm_prefetch_target_");
+    PrefetchTargetName += H.TargetFunctionName;
+    PrefetchTargetName += "_";
+    PrefetchTargetName += utostr(H.TargetPosition.BBID.BaseID);
+    PrefetchTargetName += "_";
+    PrefetchTargetName += utostr(H.TargetPosition.BBOffset);
+    F.getParent()->getOrInsertGlobal(PrefetchTargetName, PtrTy);
+  }
+
+  DenseMap<UniqueBBID, std::map<unsigned, SmallVector<PrefetchTarget>>>
+      PrefetchHintsByBBID;
+  for (const auto &H : PrefetchHints) {
+    PrefetchHintsByBBID[H.SitePosition.BBID][H.SitePosition.BBOffset].push_back(
+        PrefetchTarget{H.TargetFunctionName, H.TargetPosition.BBID,
+                       H.TargetPosition.BBOffset});
+  }
+  const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
+  for (auto &BB : MF) {
+    auto It = PrefetchHintsByBBID.find(*BB.getBBID());
+    if (It == PrefetchHintsByBBID.end())
+      continue;
+    auto BBPrefetchHintIt = It->second.begin();
+    unsigned NumInsts = 0;
+    auto E = BB.getFirstTerminator();
+    unsigned NumCallsites = 0;
+    for (auto I = BB.instr_begin();;) {
+      auto Current = I;
+      if (NumCallsites >= BBPrefetchHintIt->first || Current == E) {
+        for (const auto &PrefetchTarget : BBPrefetchHintIt->second) {
+          SmallString<128> PrefetchTargetName("__llvm_prefetch_target_");
+          PrefetchTargetName += PrefetchTarget.TargetFunction;
+          PrefetchTargetName += "_";
+          PrefetchTargetName += utostr(PrefetchTarget.TargetBBID.BaseID);
+          PrefetchTargetName += "_";
+          PrefetchTargetName += utostr(PrefetchTarget.TargetBBOffset);
+          auto *GV =
+              MF.getFunction().getParent()->getNamedValue(PrefetchTargetName);
+          // errs() << "Inserting prefetch for " << GV->getName() << " at "
+          //        << MF.getName() << " " << BB.getName() << " " << NumInsts
+          //        << "\n";
+          MachineInstr *PFetch = MF.CreateMachineInstr(
+              UseCodePrefetchInstruction ? TII->get(X86::PREFETCHIT1)
+                                         : TII->get(X86::PREFETCHT1),
+              Current != BB.instr_end() ? Current->getDebugLoc() : DebugLoc(),
+              true);
+          PFetch->setFlag(MachineInstr::Prefetch);
+          MachineInstrBuilder MIB(MF, PFetch);
+          if (!PrefetchNextAddress) {
+            MIB.addMemOperand(MF.getMachineMemOperand(
+                MachinePointerInfo(GV), MachineMemOperand::MOLoad, /*s=*/8,
+                /*base_alignment=*/llvm::Align(1)));
+          }
+          MIB.addReg(X86::RIP).addImm(1).addReg(X86::NoRegister);
+          if (PrefetchNextAddress)
+            MIB.addImm(0);
+          else
+            MIB.addGlobalAddress(GV);
+          MIB.addReg(X86::NoRegister);
+          BB.insert(Current, PFetch);
+        }
+        ++BBPrefetchHintIt;
+        if (BBPrefetchHintIt == PrefetchHintsByBBID[*BB.getBBID()].end())
+          break;
+      }
+      if (Current != E) {
+        // Print the assembly for the instruction.
+        if (!Current->isPosition() && !Current->isImplicitDef() &&
+            !Current->isKill() && !Current->isDebugInstr()) {
+          ++NumInsts;
+        }
+        if (Current->isCall())
+          ++NumCallsites;
+        ++I;
+      }
+    }
+  }
+  return true;
+}
+
+void PrefetchInsertion::getAnalysisUsage(AnalysisUsage &AU) const {
+  AU.setPreservesAll();
+  AU.addRequired<BasicBlockSectionsProfileReaderWrapperPass>();
+  MachineFunctionPass::getAnalysisUsage(AU);
+}
+
+FunctionPass *llvm::createPrefetchInsertionPass() {
+  return new PrefetchInsertion();
+}

llvm/lib/Target/X86/X86.h

@@ -166,6 +166,13 @@ FunctionPass *createX86IndirectThunksPass();
 /// This pass replaces ret instructions with jmp's to __x86_return thunk.
 FunctionPass *createX86ReturnThunksPass();
 
+/// This pass ensures instructions featuring a memory operand
+/// have distinctive <LineNumber, Discriminator> (with respect to each other)
+FunctionPass *createX86DiscriminateMemOpsPass();
+
+/// This pass applies profiling information to insert cache prefetches.
+FunctionPass *createX86InsertPrefetchPass();
+
 /// This pass insert wait instruction after X87 instructions which could raise
 /// fp exceptions when strict-fp enabled.
 FunctionPass *createX86InsertX87waitPass();

llvm/lib/Target/X86/X86TargetMachine.cpp

@@ -624,6 +624,9 @@ void X86PassConfig::addPreEmitPass2() {
   // after all real instructions have been added to the epilog.
   if (TT.isOSWindows() && TT.isX86_64())
     addPass(createX86WinEHUnwindV2Pass());
+
+  if (TM->getBBSectionsType() == llvm::BasicBlockSection::List)
+    addPass(createPrefetchInsertionPass());
 }
 
 bool X86PassConfig::addPostFastRegAllocRewrite() {