InstrProf: Fix handling of profile counters in the body of range based for

We were assigning the counter for the body of the loop to the loop
variable initialization for some reason here, but our tests completely
lacked coverage for range-for loops. This fixes that and makes the
logic generally more similar to the logic for a regular for.

llvm-svn: 236277

diff --git a/clang/lib/CodeGen/CodeGenPGO.cpp b/clang/lib/CodeGen/CodeGenPGO.cpp
index 27e4c7f..c90b025 100644 (file)
--- a/clang/lib/CodeGen/CodeGenPGO.cpp
+++ b/clang/lib/CodeGen/CodeGenPGO.cpp
@@ -447,6 +447,7 @@ struct ComputeRegionCounts : public ConstStmtVisitor<ComputeRegionCounts> {
 
   void VisitCXXForRangeStmt(const CXXForRangeStmt *S) {
     RecordStmtCount(S);
+    Visit(S->getLoopVarStmt());
     Visit(S->getRangeStmt());
     Visit(S->getBeginEndStmt());
     // Counter tracks the body of the loop.
@@ -455,8 +456,7 @@ struct ComputeRegionCounts : public ConstStmtVisitor<ComputeRegionCounts> {
     // Visit the body region first. (This is basically the same as a while
     // loop; see further comments in VisitWhileStmt.)
     Cnt.beginRegion();
-    CountMap[S->getLoopVarStmt()] = PGO.getCurrentRegionCount();
-    Visit(S->getLoopVarStmt());
+    CountMap[S->getBody()] = PGO.getCurrentRegionCount();
     Visit(S->getBody());
     Cnt.adjustForControlFlow();
 
@@ -475,7 +475,6 @@ struct ComputeRegionCounts : public ConstStmtVisitor<ComputeRegionCounts> {
                               BC.ContinueCount);
     CountMap[S->getCond()] = PGO.getCurrentRegionCount();
     Visit(S->getCond());
-    Cnt.adjustForControlFlow();
     Cnt.applyAdjustmentsToRegion(BC.BreakCount + BC.ContinueCount);
     RecordNextStmtCount = true;
   }

diff --git a/clang/test/Profile/Inputs/cxx-rangefor.proftext b/clang/test/Profile/Inputs/cxx-rangefor.proftext
new file mode 100644 (file)
index 0000000..7d2d1ef
--- /dev/null
+++ b/clang/test/Profile/Inputs/cxx-rangefor.proftext
@@ -0,0 +1,13 @@
+_Z9range_forv
+0x000000000014a28a
+5
+1
+4
+1
+1
+1
+
+main
+0
+1
+1

diff --git a/clang/test/Profile/cxx-rangefor.cpp b/clang/test/Profile/cxx-rangefor.cpp
new file mode 100644 (file)
index 0000000..f30cdc7
--- /dev/null
+++ b/clang/test/Profile/cxx-rangefor.cpp
@@ -0,0 +1,44 @@
+// Tests for instrumentation of C++11 range-for
+
+// RUN: %clang_cc1 -x c++ %s -triple %itanium_abi_triple -main-file-name cxx-rangefor.cpp -std=c++11 -o - -emit-llvm -fprofile-instr-generate > %tgen
+// RUN: FileCheck --input-file=%tgen -check-prefix=CHECK -check-prefix=PGOGEN %s
+
+// RUN: llvm-profdata merge %S/Inputs/cxx-rangefor.proftext -o %t.profdata
+// RUN: %clang_cc1 -x c++ %s -triple %itanium_abi_triple -main-file-name cxx-rangefor.cpp -std=c++11 -o - -emit-llvm -fprofile-instr-use=%t.profdata > %tuse
+// RUN: FileCheck --input-file=%tuse -check-prefix=CHECK -check-prefix=PGOUSE %s
+
+// PGOGEN: @[[RFC:__llvm_profile_counters__Z9range_forv]] = private global [5 x i64] zeroinitializer
+
+// CHECK-LABEL: define void @_Z9range_forv()
+// PGOGEN: store {{.*}} @[[RFC]], i64 0, i64 0
+void range_for() {
+  int arr[] = {1, 2, 3, 4, 5};
+  int sum = 0;
+  // PGOGEN: store {{.*}} @[[RFC]], i64 0, i64 1
+  // PGOUSE: br {{.*}} !prof ![[RF1:[0-9]+]]
+  for (auto i : arr) {
+    // PGOGEN: store {{.*}} @[[RFC]], i64 0, i64 2
+    // PGOUSE: br {{.*}} !prof ![[RF2:[0-9]+]]
+    if (i == 3)
+      continue;
+    sum += i;
+    // PGOGEN: store {{.*}} @[[RFC]], i64 0, i64 3
+    // PGOUSE: br {{.*}} !prof ![[RF3:[0-9]+]]
+    if (sum >= 7)
+      break;
+  }
+
+  // PGOGEN: store {{.*}} @[[RFC]], i64 0, i64 4
+  // PGOUSE: br {{.*}} !prof ![[RF4:[0-9]+]]
+  if (sum) {}
+}
+
+// PGOUSE-DAG: ![[RF1]] = !{!"branch_weights", i32 5, i32 1}
+// PGOUSE-DAG: ![[RF2]] = !{!"branch_weights", i32 2, i32 4}
+// PGOUSE-DAG: ![[RF3]] = !{!"branch_weights", i32 2, i32 3}
+// PGOUSE-DAG: ![[RF4]] = !{!"branch_weights", i32 2, i32 1}
+
+int main(int argc, const char *argv[]) {
+  range_for();
+  return 0;
+}