InstrProf: Don't keep a large sparse list around just to zero it

The Terms vector here represented a polynomial of of all possible
counters, and is used to simplify expressions when generating coverage
mapping. There are a few problems with this:

1. Keeping the vector as a member is wasteful, since we clear it every
   time we use it.
2. Most expressions refer to a subset of the counters, so we end up
   iterating over a large number of zeros doing nothing a lot of the
   time.

This updates the user of the vector to store the terms locally, and
uses a sort and combine approach so that we only operate on counters
that are actually used in a given expression. For small cases this
makes very little difference, but in cases with a very large number of
counted regions this is a significant performance fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218879 91177308-0d34-0410-b5e6-96231b3b80d8

include/llvm/ProfileData/CoverageMapping.h

@@ -103,8 +103,6 @@ struct CounterExpression {
 class CounterExpressionBuilder {
   /// \brief A list of all the counter expressions
   llvm::SmallVector<CounterExpression, 16> Expressions;
-  /// \brief An array of terms used in expression simplification.
-  llvm::SmallVector<int, 16> Terms;
 
   /// \brief Return the counter which corresponds to the given expression.
   ///
@@ -113,18 +111,19 @@ class CounterExpressionBuilder {
   /// expression is added to the builder's collection of expressions.
   Counter get(const CounterExpression &E);
 
-  /// \brief Convert the expression tree represented by a counter
+  /// \brief Gather the terms of the expression tree for processing.
-  /// into a polynomial in the form of K1Counter1 + .. + KNCounterN
+  ///
-  /// where K1 .. KN are integer constants that are stored in the Terms array.
+  /// This collects each addition and subtraction referenced by the counter into
-  void extractTerms(Counter C, int Sign = 1);
+  /// a sequence that can be sorted and combined to build a simplified counter
+  /// expression.
+  void extractTerms(Counter C, int Sign,
+                    SmallVectorImpl<std::pair<unsigned, int>> &Terms);
 
   /// \brief Simplifies the given expression tree
   /// by getting rid of algebraically redundant operations.
   Counter simplify(Counter ExpressionTree);
 
 public:
-  CounterExpressionBuilder(unsigned NumCounterValues);
-
   ArrayRef<CounterExpression> getExpressions() const { return Expressions; }
 
   /// \brief Return a counter that represents the expression

lib/ProfileData/CoverageMapping.cpp

@@ -27,10 +27,6 @@ using namespace coverage;
 
 #define DEBUG_TYPE "coverage-mapping"
 
-CounterExpressionBuilder::CounterExpressionBuilder(unsigned NumCounterValues) {
-  Terms.resize(NumCounterValues);
-}
-
 Counter CounterExpressionBuilder::get(const CounterExpression &E) {
   for (unsigned I = 0, S = Expressions.size(); I < S; ++I) {
     if (Expressions[I] == E)
@@ -40,50 +36,68 @@ Counter CounterExpressionBuilder::get(const CounterExpression &E) {
   return Counter::getExpression(Expressions.size() - 1);
 }
 
-void CounterExpressionBuilder::extractTerms(Counter C, int Sign) {
+void CounterExpressionBuilder::extractTerms(
+    Counter C, int Sign, SmallVectorImpl<std::pair<unsigned, int>> &Terms) {
   switch (C.getKind()) {
   case Counter::Zero:
     break;
   case Counter::CounterValueReference:
-    Terms[C.getCounterID()] += Sign;
+    Terms.push_back(std::make_pair(C.getCounterID(), Sign));
     break;
   case Counter::Expression:
     const auto &E = Expressions[C.getExpressionID()];
-    extractTerms(E.LHS, Sign);
+    extractTerms(E.LHS, Sign, Terms);
-    extractTerms(E.RHS, E.Kind == CounterExpression::Subtract ? -Sign : Sign);
+    extractTerms(E.RHS, E.Kind == CounterExpression::Subtract ? -Sign : Sign,
+                 Terms);
     break;
   }
 }
 
 Counter CounterExpressionBuilder::simplify(Counter ExpressionTree) {
   // Gather constant terms.
-  for (auto &I : Terms)
+  llvm::SmallVector<std::pair<unsigned, int>, 32> Terms;
-    I = 0;
+  extractTerms(ExpressionTree, +1, Terms);
-  extractTerms(ExpressionTree);
+
+  // Group the terms by counter ID.
+  std::sort(Terms.begin(), Terms.end(),
+            [](const std::pair<unsigned, int> &LHS,
+               const std::pair<unsigned, int> &RHS) {
+    return LHS.first < RHS.first;
+  });
+
+  // Combine terms by counter ID to eliminate counters that sum to zero.
+  auto Prev = Terms.begin();
+  for (auto I = Prev + 1, E = Terms.end(); I != E; ++I) {
+    if (I->first == Prev->first) {
+      Prev->second += I->second;
+      continue;
+    }
+    ++Prev;
+    *Prev = *I;
+  }
+  Terms.erase(++Prev, Terms.end());
 
   Counter C;
-  // Create additions.
+  // Create additions. We do this before subtractions to avoid constructs like
-  // Note: the additions are created first
+  // ((0 - X) + Y), as opposed to (Y - X).
-  // to avoid creation of a tree like ((0 - X) + Y) instead of (Y - X).
+  for (auto Term : Terms) {
-  for (unsigned I = 0, S = Terms.size(); I < S; ++I) {
+    if (Term.second <= 0)
-    if (Terms[I] <= 0)
       continue;
-    for (int J = 0; J < Terms[I]; ++J) {
+    for (int I = 0; I < Term.second; ++I)
       if (C.isZero())
-        C = Counter::getCounter(I);
+        C = Counter::getCounter(Term.first);
       else
         C = get(CounterExpression(CounterExpression::Add, C,
-                                  Counter::getCounter(I)));
+                                  Counter::getCounter(Term.first)));
-    }
   }
 
   // Create subtractions.
-  for (unsigned I = 0, S = Terms.size(); I < S; ++I) {
+  for (auto Term : Terms) {
-    if (Terms[I] >= 0)
+    if (Term.second >= 0)
       continue;
-    for (int J = 0; J < (-Terms[I]); ++J)
+    for (int I = 0; I < -Term.second; ++I)
       C = get(CounterExpression(CounterExpression::Subtract, C,
-                                Counter::getCounter(I)));
+                                Counter::getCounter(Term.first)));
   }
   return C;
 }