CodeGen: Handle CapturedStmt in instrumentation based profiling

CapturedStmt was being ignored by instrumentation based profiling, and
its counters attributed to the containing function. Instead, we need
to treat this as a top level entity, like we do with blocks.

llvm-svn: 206231

5 files changed, 101 insertions, 0 deletions

diff --git a/clang/lib/CodeGen/CGStmt.cpp b/clang/lib/CodeGen/CGStmt.cpp
index d50d8b959a8..9ea53b3e1fb 100644
--- a/clang/lib/CodeGen/CGStmt.cpp
+++ b/clang/lib/CodeGen/CGStmt.cpp
@@ -1962,8 +1962,11 @@ CodeGenFunction::GenerateCapturedStmtFunction(const CapturedDecl *CD,
     CXXThisValue = EmitLoadOfLValue(ThisLValue, Loc).getScalarVal();
   }
 
+  PGO.assignRegionCounters(CD, F);
   CapturedStmtInfo->EmitBody(*this, CD->getBody());
   FinishFunction(CD->getBodyRBrace());
+  PGO.emitInstrumentationData();
+  PGO.destroyRegionCounters();
 
   return F;
 }
diff --git a/clang/lib/CodeGen/CodeGenFunction.h b/clang/lib/CodeGen/CodeGenFunction.h
index 7d5a59e99ed..ec9601e0644 100644
--- a/clang/lib/CodeGen/CodeGenFunction.h
+++ b/clang/lib/CodeGen/CodeGenFunction.h
@@ -195,6 +195,8 @@ public:
 
     /// \brief Emit the captured statement body.
     virtual void EmitBody(CodeGenFunction &CGF, Stmt *S) {
+      RegionCounter Cnt = CGF.getPGORegionCounter(S);
+      Cnt.beginRegion(CGF.Builder);
       CGF.EmitStmt(S);
     }
 
diff --git a/clang/lib/CodeGen/CodeGenPGO.cpp b/clang/lib/CodeGen/CodeGenPGO.cpp
index 4dbd01a6619..80c7e87de36 100644
--- a/clang/lib/CodeGen/CodeGenPGO.cpp
+++ b/clang/lib/CodeGen/CodeGenPGO.cpp
@@ -335,6 +335,7 @@ namespace {
     // traverse them in the parent context.
     bool TraverseBlockExpr(BlockExpr *BE) { return true; }
     bool TraverseLambdaBody(LambdaExpr *LE) { return true; }
+    bool TraverseCapturedStmt(CapturedStmt *CS) { return true; }
 
     bool VisitDecl(const Decl *D) {
       switch (D->getKind()) {
@@ -347,6 +348,7 @@ namespace {
       case Decl::CXXConversion:
       case Decl::ObjCMethod:
       case Decl::Block:
+      case Decl::Captured:
         CounterMap[D->getBody()] = NextCounter++;
         break;
       }
@@ -437,6 +439,14 @@ namespace {
     // parent context.
     void VisitLambdaExpr(const LambdaExpr *LE) {}
 
+    void VisitCapturedDecl(const CapturedDecl *D) {
+      // Counter tracks entry to the capture body.
+      RegionCounter Cnt(PGO, D->getBody());
+      Cnt.beginRegion();
+      CountMap[D->getBody()] = PGO.getCurrentRegionCount();
+      Visit(D->getBody());
+    }
+
     void VisitObjCMethodDecl(const ObjCMethodDecl *D) {
       // Counter tracks entry to the method body.
       RegionCounter Cnt(PGO, D->getBody());
@@ -838,6 +848,8 @@ void CodeGenPGO::mapRegionCounters(const Decl *D) {
     Walker.TraverseDecl(const_cast<ObjCMethodDecl *>(MD));
   else if (const BlockDecl *BD = dyn_cast_or_null<BlockDecl>(D))
     Walker.TraverseDecl(const_cast<BlockDecl *>(BD));
+  else if (const CapturedDecl *CD = dyn_cast_or_null<CapturedDecl>(D))
+    Walker.TraverseDecl(const_cast<CapturedDecl *>(CD));
   NumRegionCounters = Walker.NextCounter;
   // FIXME: The number of counters isn't sufficient for the hash
   FunctionHash = NumRegionCounters;
@@ -852,6 +864,8 @@ void CodeGenPGO::computeRegionCounts(const Decl *D) {
     Walker.VisitObjCMethodDecl(MD);
   else if (const BlockDecl *BD = dyn_cast_or_null<BlockDecl>(D))
     Walker.VisitBlockDecl(BD);
+  else if (const CapturedDecl *CD = dyn_cast_or_null<CapturedDecl>(D))
+    Walker.VisitCapturedDecl(const_cast<CapturedDecl *>(CD));
 }
 
 void CodeGenPGO::applyFunctionAttributes(PGOProfileData *PGOData,
diff --git a/clang/test/Profile/Inputs/c-captured.profdata b/clang/test/Profile/Inputs/c-captured.profdata
new file mode 100644
index 00000000000..c2490d9b680
--- /dev/null
+++ b/clang/test/Profile/Inputs/c-captured.profdata
@@ -0,0 +1,25 @@
+c-captured.c:__captured_stmt
+2
+2
+1
+1
+
+c-captured.c:__captured_stmt1
+3
+3
+1
+10
+1
+
+main
+1
+1
+1
+
+debug_captured
+3
+3
+1
+1
+1
+
diff --git a/clang/test/Profile/c-captured.c b/clang/test/Profile/c-captured.c
new file mode 100644
index 00000000000..a2678ec6283
--- /dev/null
+++ b/clang/test/Profile/c-captured.c
@@ -0,0 +1,57 @@
+// RUN: %clang_cc1 -triple x86_64-apple-macosx10.9 -main-file-name c-captured.c %s -o - -emit-llvm -fprofile-instr-generate | FileCheck -check-prefix=PGOGEN -check-prefix=PGOALL %s
+// RUN: %clang_cc1 -triple x86_64-apple-macosx10.9 -main-file-name c-captured.c %s -o - -emit-llvm -fprofile-instr-use=%S/Inputs/c-captured.profdata | FileCheck -check-prefix=PGOUSE -check-prefix=PGOALL %s
+
+// PGOGEN: @[[DCC:__llvm_profile_counters_debug_captured]] = global [3 x i64] zeroinitializer
+// PGOGEN: @[[CSC:__llvm_profile_counters___captured_stmt]] = internal global [2 x i64] zeroinitializer
+// PGOGEN: @[[C1C:__llvm_profile_counters___captured_stmt1]] = internal global [3 x i64] zeroinitializer
+
+// PGOALL-LABEL: define void @debug_captured()
+// PGOGEN: store {{.*}} @[[DCC]], i64 0, i64 0
+void debug_captured() {
+  int x = 10;
+
+  // Check both debug_captured counters, so we can do this all in one pass
+  // PGOGEN: store {{.*}} @[[DCC]], i64 0, i64 1
+  // PGOUSE: br {{.*}} !prof ![[DC1:[0-9]+]]
+  // PGOGEN: store {{.*}} @[[DCC]], i64 0, i64 2
+  // PGOUSE: br {{.*}} !prof ![[DC2:[0-9]+]]
+  // PGOALL: ret
+
+  // PGOALL-LABEL: define internal void @__captured_stmt(
+  // PGOGEN: store {{.*}} @[[CSC]], i64 0, i64 0
+  #pragma clang __debug captured
+  {
+    // PGOGEN: store {{.*}} @[[CSC]], i64 0, i64 1
+    // PGOUSE: br {{.*}} !prof ![[CS1:[0-9]+]]
+    if (x) {}
+    // PGOALL: ret
+  }
+
+  if (x) {} // This is DC1. Checked above.
+
+  // PGOALL-LABEL: define internal void @__captured_stmt1(
+  // PGOGEN: store {{.*}} @[[C1C]], i64 0, i64 0
+  #pragma clang __debug captured
+  {
+    // PGOGEN: store {{.*}} @[[C1C]], i64 0, i64 1
+    // PGOUSE: br {{.*}} !prof ![[C11:[0-9]+]]
+    for (int i = 0; i < x; ++i) {}
+    // PGOGEN: store {{.*}} @[[C1C]], i64 0, i64 2
+    // PGOUSE: br {{.*}} !prof ![[C12:[0-9]+]]
+    if (x) {}
+    // PGOALL: ret
+  }
+
+  if (x) {} // This is DC2. Checked above.
+}
+
+// PGOUSE-DAG: ![[DC1]] = metadata !{metadata !"branch_weights", i32 2, i32 1}
+// PGOUSE-DAG: ![[DC2]] = metadata !{metadata !"branch_weights", i32 2, i32 1}
+// PGOUSE-DAG: ![[CS1]] = metadata !{metadata !"branch_weights", i32 2, i32 1}
+// PGOUSE-DAG: ![[C11]] = metadata !{metadata !"branch_weights", i32 11, i32 2}
+// PGOUSE-DAG: ![[C12]] = metadata !{metadata !"branch_weights", i32 2, i32 1}
+
+int main(int argc, const char *argv[]) {
+  debug_captured();
+  return 0;
+}