[CodeGen] Move CodeGenPGO behind unique_ptr (NFC) (llvm#142155)

The InstrProf headers are very expensive. Avoid including them in all of
CodeGen/ by moving the CodeGenPGO member behind a unqiue_ptr.

This reduces clang build time by 0.8%.

Author: nikic

clang/lib/CodeGen/CGBlocks.cpp

@@ -17,6 +17,7 @@
 #include "CGOpenCLRuntime.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
+#include "CodeGenPGO.h"
 #include "ConstantEmitter.h"
 #include "TargetInfo.h"
 #include "clang/AST/Attr.h"
@@ -1522,7 +1523,7 @@ llvm::Function *CodeGenFunction::GenerateBlockFunction(
   if (IsLambdaConversionToBlock)
     EmitLambdaBlockInvokeBody();
   else {
-    PGO.assignRegionCounters(GlobalDecl(blockDecl), fn);
+    PGO->assignRegionCounters(GlobalDecl(blockDecl), fn);
     incrementProfileCounter(blockDecl->getBody());
     EmitStmt(blockDecl->getBody());
   }

clang/lib/CodeGen/CGCall.cpp

@@ -21,6 +21,7 @@
 #include "CGRecordLayout.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
+#include "CodeGenPGO.h"
 #include "TargetInfo.h"
 #include "clang/AST/Attr.h"
 #include "clang/AST/Decl.h"
@@ -5900,7 +5901,7 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
   // For more details, see the comment before the definition of
   // IPVK_IndirectCallTarget in InstrProfData.inc.
   if (!CI->getCalledFunction())
-    PGO.valueProfile(Builder, llvm::IPVK_IndirectCallTarget, CI, CalleePtr);
+    PGO->valueProfile(Builder, llvm::IPVK_IndirectCallTarget, CI, CalleePtr);
 
   // In ObjC ARC mode with no ObjC ARC exception safety, tell the ARC
   // optimizer it can aggressively ignore unwind edges.

clang/lib/CodeGen/CGDecl.cpp

@@ -18,6 +18,7 @@
 #include "CGOpenMPRuntime.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
+#include "CodeGenPGO.h"
 #include "ConstantEmitter.h"
 #include "EHScopeStack.h"
 #include "PatternInit.h"
@@ -368,7 +369,7 @@ CodeGenFunction::AddInitializerToStaticVarDecl(const VarDecl &D,
     return GV;
   }
 
-  PGO.markStmtMaybeUsed(D.getInit()); // FIXME: Too lazy
+  PGO->markStmtMaybeUsed(D.getInit()); // FIXME: Too lazy
 
 #ifndef NDEBUG
   CharUnits VarSize = CGM.getContext().getTypeSizeInChars(D.getType()) +
@@ -1944,7 +1945,7 @@ void CodeGenFunction::EmitAutoVarInit(const AutoVarEmission &emission) {
   // unless it contains a label.
   if (!HaveInsertPoint()) {
     if (!Init || !ContainsLabel(Init)) {
-      PGO.markStmtMaybeUsed(Init);
+      PGO->markStmtMaybeUsed(Init);
       return;
     }
     EnsureInsertPoint();
@@ -2057,7 +2058,7 @@ void CodeGenFunction::EmitAutoVarInit(const AutoVarEmission &emission) {
     return EmitExprAsInit(Init, &D, lv, capturedByInit);
   }
 
-  PGO.markStmtMaybeUsed(Init);
+  PGO->markStmtMaybeUsed(Init);
 
   if (!emission.IsConstantAggregate) {
     // For simple scalar/complex initialization, store the value directly.

clang/lib/CodeGen/CGExpr.cpp

@@ -21,6 +21,7 @@
 #include "CGRecordLayout.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
+#include "CodeGenPGO.h"
 #include "ConstantEmitter.h"
 #include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
@@ -207,7 +208,7 @@ RawAddress CodeGenFunction::CreateMemTempWithoutCast(QualType Ty,
 /// EvaluateExprAsBool - Perform the usual unary conversions on the specified
 /// expression and compare the result against zero, returning an Int1Ty value.
 llvm::Value *CodeGenFunction::EvaluateExprAsBool(const Expr *E) {
-  PGO.setCurrentStmt(E);
+  PGO->setCurrentStmt(E);
   if (const MemberPointerType *MPT = E->getType()->getAs<MemberPointerType>()) {
     llvm::Value *MemPtr = EmitScalarExpr(E);
     return CGM.getCXXABI().EmitMemberPointerIsNotNull(*this, MemPtr, MPT);

clang/lib/CodeGen/CGObjC.cpp

@@ -14,6 +14,7 @@
 #include "CGObjCRuntime.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
+#include "CodeGenPGO.h"
 #include "ConstantEmitter.h"
 #include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
@@ -805,7 +806,7 @@ static llvm::Value *emitARCRetainLoadOfScalar(CodeGenFunction &CGF,
 /// its pointer, name, and types registered in the class structure.
 void CodeGenFunction::GenerateObjCMethod(const ObjCMethodDecl *OMD) {
   StartObjCMethod(OMD, OMD->getClassInterface());
-  PGO.assignRegionCounters(GlobalDecl(OMD), CurFn);
+  PGO->assignRegionCounters(GlobalDecl(OMD), CurFn);
   assert(isa<CompoundStmt>(OMD->getBody()));
   incrementProfileCounter(OMD->getBody());
   EmitCompoundStmtWithoutScope(*cast<CompoundStmt>(OMD->getBody()));

clang/lib/CodeGen/CGStmt.cpp

@@ -14,6 +14,7 @@
 #include "CGOpenMPRuntime.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
+#include "CodeGenPGO.h"
 #include "TargetInfo.h"
 #include "clang/AST/Attr.h"
 #include "clang/AST/Expr.h"
@@ -59,7 +60,7 @@ void CodeGenFunction::EmitStopPoint(const Stmt *S) {
 
 void CodeGenFunction::EmitStmt(const Stmt *S, ArrayRef<const Attr *> Attrs) {
   assert(S && "Null statement?");
-  PGO.setCurrentStmt(S);
+  PGO->setCurrentStmt(S);
 
   // These statements have their own debug info handling.
   if (EmitSimpleStmt(S, Attrs))
@@ -76,7 +77,7 @@ void CodeGenFunction::EmitStmt(const Stmt *S, ArrayRef<const Attr *> Attrs) {
       // Verify that any decl statements were handled as simple, they may be in
       // scope of subsequent reachable statements.
       assert(!isa<DeclStmt>(*S) && "Unexpected DeclStmt!");
-      PGO.markStmtMaybeUsed(S);
+      PGO->markStmtMaybeUsed(S);
       return;
     }
 
@@ -917,7 +918,7 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
         RunCleanupsScope ExecutedScope(*this);
         EmitStmt(Executed);
       }
-      PGO.markStmtMaybeUsed(Skipped);
+      PGO->markStmtMaybeUsed(Skipped);
       return;
     }
   }
@@ -2265,7 +2266,7 @@ void CodeGenFunction::EmitSwitchStmt(const SwitchStmt &S) {
       for (unsigned i = 0, e = CaseStmts.size(); i != e; ++i)
         EmitStmt(CaseStmts[i]);
       incrementProfileCounter(&S);
-      PGO.markStmtMaybeUsed(S.getBody());
+      PGO->markStmtMaybeUsed(S.getBody());
 
       // Now we want to restore the saved switch instance so that nested
       // switches continue to function properly
@@ -2308,7 +2309,7 @@ void CodeGenFunction::EmitSwitchStmt(const SwitchStmt &S) {
                             llvm::MDNode::get(CGM.getLLVMContext(), Vals));
   }
 
-  if (PGO.haveRegionCounts()) {
+  if (PGO->haveRegionCounts()) {
     // Walk the SwitchCase list to find how many there are.
     uint64_t DefaultCount = 0;
     unsigned NumCases = 0;
@@ -3329,7 +3330,7 @@ CodeGenFunction::GenerateCapturedStmtFunction(const CapturedStmt &S) {
     CXXThisValue = EmitLoadOfLValue(ThisLValue, Loc).getScalarVal();
   }
 
-  PGO.assignRegionCounters(GlobalDecl(CD), F);
+  PGO->assignRegionCounters(GlobalDecl(CD), F);
   CapturedStmtInfo->EmitBody(*this, CD->getBody());
   FinishFunction(CD->getBodyRBrace());
 

clang/lib/CodeGen/CGStmtOpenMP.cpp

@@ -15,6 +15,7 @@
 #include "CGOpenMPRuntime.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
+#include "CodeGenPGO.h"
 #include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/Attr.h"
@@ -710,7 +711,7 @@ CodeGenFunction::GenerateOpenMPCapturedStmtFunction(const CapturedStmt &S,
   (void)LocalScope.Privatize();
   for (const auto &VLASizePair : WrapperVLASizes)
     VLASizeMap[VLASizePair.second.first] = VLASizePair.second.second;
-  PGO.assignRegionCounters(GlobalDecl(CD), F);
+  PGO->assignRegionCounters(GlobalDecl(CD), F);
   CapturedStmtInfo->EmitBody(*this, CD->getBody());
   (void)LocalScope.ForceCleanup();
   FinishFunction(CD->getBodyRBrace());

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -80,7 +80,8 @@ CodeGenFunction::CodeGenFunction(CodeGenModule &cgm, bool suppressNewContext)
       Builder(cgm, cgm.getModule().getContext(), llvm::ConstantFolder(),
               CGBuilderInserterTy(this)),
       SanOpts(CGM.getLangOpts().Sanitize), CurFPFeatures(CGM.getLangOpts()),
-      DebugInfo(CGM.getModuleDebugInfo()), PGO(cgm),
+      DebugInfo(CGM.getModuleDebugInfo()),
+      PGO(std::make_unique<CodeGenPGO>(cgm)),
       ShouldEmitLifetimeMarkers(
           shouldEmitLifetimeMarkers(CGM.getCodeGenOpts(), CGM.getLangOpts())) {
   if (!suppressNewContext)
@@ -1564,7 +1565,7 @@ void CodeGenFunction::GenerateCode(GlobalDecl GD, llvm::Function *Fn,
     CurFn->addFnAttr(llvm::Attribute::MustProgress);
 
   // Generate the body of the function.
-  PGO.assignRegionCounters(GD, CurFn);
+  PGO->assignRegionCounters(GD, CurFn);
   if (isa<CXXDestructorDecl>(FD))
     EmitDestructorBody(Args);
   else if (isa<CXXConstructorDecl>(FD))
@@ -1650,7 +1651,7 @@ void CodeGenFunction::GenerateCode(GlobalDecl GD, llvm::Function *Fn,
   // Emit the standard function epilogue.
   FinishFunction(BodyRange.getEnd());
 
-  PGO.verifyCounterMap();
+  PGO->verifyCounterMap();
 
   // If we haven't marked the function nothrow through other means, do
   // a quick pass now to see if we can.
@@ -1774,7 +1775,7 @@ bool CodeGenFunction::ConstantFoldsToSimpleInteger(const Expr *Cond,
   if (!AllowLabels && CodeGenFunction::ContainsLabel(Cond))
     return false;  // Contains a label.
 
-  PGO.markStmtMaybeUsed(Cond);
+  PGO->markStmtMaybeUsed(Cond);
   ResultInt = Int;
   return true;
 }
@@ -3355,4 +3356,4 @@ void CodeGenFunction::addInstToNewSourceAtom(llvm::Instruction *KeyInstruction,
     ApplyAtomGroup Grp(getDebugInfo());
     DI->addInstToCurrentSourceAtom(KeyInstruction, Backup);
   }
-}
+}

clang/lib/CodeGen/CodeGenFunction.h

@@ -17,7 +17,6 @@
 #include "CGLoopInfo.h"
 #include "CGValue.h"
 #include "CodeGenModule.h"
-#include "CodeGenPGO.h"
 #include "EHScopeStack.h"
 #include "VarBypassDetector.h"
 #include "clang/AST/CharUnits.h"
@@ -90,6 +89,7 @@ class OSLogBufferLayout;
 
 namespace CodeGen {
 class CodeGenTypes;
+class CodeGenPGO;
 class CGCallee;
 class CGFunctionInfo;
 class CGBlockInfo;
@@ -1670,7 +1670,7 @@ class CodeGenFunction : public CodeGenTypeCache {
   llvm::Value *emitCondLikelihoodViaExpectIntrinsic(llvm::Value *Cond,
                                                     Stmt::Likelihood LH);
 
-  CodeGenPGO PGO;
+  std::unique_ptr<CodeGenPGO> PGO;
 
   /// Bitmap used by MC/DC to track condition outcomes of a boolean expression.
   Address MCDCCondBitmapAddr = Address::invalid();
@@ -1683,12 +1683,9 @@ class CodeGenFunction : public CodeGenTypeCache {
                                             uint64_t LoopCount) const;
 
 public:
-  auto getIsCounterPair(const Stmt *S) const { return PGO.getIsCounterPair(S); }
-
-  void markStmtAsUsed(bool Skipped, const Stmt *S) {
-    PGO.markStmtAsUsed(Skipped, S);
-  }
-  void markStmtMaybeUsed(const Stmt *S) { PGO.markStmtMaybeUsed(S); }
+  std::pair<bool, bool> getIsCounterPair(const Stmt *S) const;
+  void markStmtAsUsed(bool Skipped, const Stmt *S);
+  void markStmtMaybeUsed(const Stmt *S);
 
   /// Increment the profiler's counter for the given statement by \p StepV.
   /// If \p StepV is null, the default increment is 1.
@@ -1702,57 +1699,32 @@ class CodeGenFunction : public CodeGenTypeCache {
 
   /// Allocate a temp value on the stack that MCDC can use to track condition
   /// results.
-  void maybeCreateMCDCCondBitmap() {
-    if (isMCDCCoverageEnabled()) {
-      PGO.emitMCDCParameters(Builder);
-      MCDCCondBitmapAddr =
-          CreateIRTemp(getContext().UnsignedIntTy, "mcdc.addr");
-    }
-  }
+  void maybeCreateMCDCCondBitmap();
 
   bool isBinaryLogicalOp(const Expr *E) const {
     const BinaryOperator *BOp = dyn_cast<BinaryOperator>(E->IgnoreParens());
     return (BOp && BOp->isLogicalOp());
   }
 
   /// Zero-init the MCDC temp value.
-  void maybeResetMCDCCondBitmap(const Expr *E) {
-    if (isMCDCCoverageEnabled() && isBinaryLogicalOp(E)) {
-      PGO.emitMCDCCondBitmapReset(Builder, E, MCDCCondBitmapAddr);
-      PGO.setCurrentStmt(E);
-    }
-  }
+  void maybeResetMCDCCondBitmap(const Expr *E);
 
   /// Increment the profiler's counter for the given expression by \p StepV.
   /// If \p StepV is null, the default increment is 1.
-  void maybeUpdateMCDCTestVectorBitmap(const Expr *E) {
-    if (isMCDCCoverageEnabled() && isBinaryLogicalOp(E)) {
-      PGO.emitMCDCTestVectorBitmapUpdate(Builder, E, MCDCCondBitmapAddr, *this);
-      PGO.setCurrentStmt(E);
-    }
-  }
+  void maybeUpdateMCDCTestVectorBitmap(const Expr *E);
 
   /// Update the MCDC temp value with the condition's evaluated result.
-  void maybeUpdateMCDCCondBitmap(const Expr *E, llvm::Value *Val) {
-    if (isMCDCCoverageEnabled()) {
-      PGO.emitMCDCCondBitmapUpdate(Builder, E, MCDCCondBitmapAddr, Val, *this);
-      PGO.setCurrentStmt(E);
-    }
-  }
+  void maybeUpdateMCDCCondBitmap(const Expr *E, llvm::Value *Val);
 
   /// Get the profiler's count for the given statement.
-  uint64_t getProfileCount(const Stmt *S) {
-    return PGO.getStmtCount(S).value_or(0);
-  }
+  uint64_t getProfileCount(const Stmt *S);
 
   /// Set the profiler's current count.
-  void setCurrentProfileCount(uint64_t Count) {
-    PGO.setCurrentRegionCount(Count);
-  }
+  void setCurrentProfileCount(uint64_t Count);
 
   /// Get the profiler's current count. This is generally the count for the most
   /// recently incremented counter.
-  uint64_t getCurrentProfileCount() { return PGO.getCurrentRegionCount(); }
+  uint64_t getCurrentProfileCount();
 
   /// See CGDebugInfo::addInstToCurrentSourceAtom.
   void addInstToCurrentSourceAtom(llvm::Instruction *KeyInstruction,