Ensure -mcpu=xscale works for arm targets, after rL252903 and rL252904
[oota-llvm.git] / include / llvm / Support / BranchProbability.h
index 3620d4d5d7722b96cdfd35ee5496687db2fef43a..26bc888d1cabfb29c7a2b62cec086397925bd3ad 100644 (file)
@@ -53,6 +53,9 @@ public:
   // Create a BranchProbability object with the given numerator and 1<<31
   // as denominator.
   static BranchProbability getRaw(uint32_t N) { return BranchProbability(N); }
+  // Create a BranchProbability object from 64-bit integers.
+  static BranchProbability getBranchProbability(uint64_t Numerator,
+                                                uint64_t Denominator);
 
   // Normalize given probabilties so that the sum of them becomes approximate
   // one.
@@ -60,11 +63,6 @@ public:
   static void normalizeProbabilities(ProbabilityIter Begin,
                                      ProbabilityIter End);
 
-  // Normalize a list of weights by scaling them down so that the sum of them
-  // doesn't exceed UINT32_MAX.
-  template <class WeightListIter>
-  static void normalizeEdgeWeights(WeightListIter Begin, WeightListIter End);
-
   uint32_t getNumerator() const { return N; }
   static uint32_t getDenominator() { return D; }
 
@@ -114,6 +112,14 @@ public:
     return *this;
   }
 
+  BranchProbability &operator/=(uint32_t RHS) {
+    assert(N != UnknownN &&
+           "Unknown probability cannot participate in arithmetics.");
+    assert(RHS > 0 && "The divider cannot be zero.");
+    N /= RHS;
+    return *this;
+  }
+
   BranchProbability operator+(BranchProbability RHS) const {
     BranchProbability Prob(*this);
     return Prob += RHS;
@@ -129,41 +135,75 @@ public:
     return Prob *= RHS;
   }
 
+  BranchProbability operator/(uint32_t RHS) const {
+    BranchProbability Prob(*this);
+    return Prob /= RHS;
+  }
+
   bool operator==(BranchProbability RHS) const { return N == RHS.N; }
   bool operator!=(BranchProbability RHS) const { return !(*this == RHS); }
-  bool operator<(BranchProbability RHS) const { return N < RHS.N; }
-  bool operator>(BranchProbability RHS) const { return RHS < *this; }
-  bool operator<=(BranchProbability RHS) const { return !(RHS < *this); }
-  bool operator>=(BranchProbability RHS) const { return !(*this < RHS); }
+
+  bool operator<(BranchProbability RHS) const {
+    assert(N != UnknownN && RHS.N != UnknownN &&
+           "Unknown probability cannot participate in comparisons.");
+    return N < RHS.N;
+  }
+
+  bool operator>(BranchProbability RHS) const {
+    assert(N != UnknownN && RHS.N != UnknownN &&
+           "Unknown probability cannot participate in comparisons.");
+    return RHS < *this;
+  }
+
+  bool operator<=(BranchProbability RHS) const {
+    assert(N != UnknownN && RHS.N != UnknownN &&
+           "Unknown probability cannot participate in comparisons.");
+    return !(RHS < *this);
+  }
+
+  bool operator>=(BranchProbability RHS) const {
+    assert(N != UnknownN && RHS.N != UnknownN &&
+           "Unknown probability cannot participate in comparisons.");
+    return !(*this < RHS);
+  }
 };
 
 inline raw_ostream &operator<<(raw_ostream &OS, BranchProbability Prob) {
   return Prob.print(OS);
 }
 
-inline BranchProbability operator/(BranchProbability LHS, uint32_t RHS) {
-  assert(LHS != BranchProbability::getUnknown() &&
-         "Unknown probability cannot participate in arithmetics.");
-  return BranchProbability::getRaw(LHS.getNumerator() / RHS);
-}
-
 template <class ProbabilityIter>
 void BranchProbability::normalizeProbabilities(ProbabilityIter Begin,
                                                ProbabilityIter End) {
   if (Begin == End)
     return;
 
-  auto UnknownProbCount =
-      std::count(Begin, End, BranchProbability::getUnknown());
-  assert((UnknownProbCount == 0 ||
-          UnknownProbCount == std::distance(Begin, End)) &&
-         "Cannot normalize probabilities with known and unknown ones.");
-  (void)UnknownProbCount;
-
-  uint64_t Sum = std::accumulate(
-      Begin, End, uint64_t(0),
-      [](uint64_t S, const BranchProbability &BP) { return S + BP.N; });
-
+  unsigned UnknownProbCount = 0;
+  uint64_t Sum = std::accumulate(Begin, End, uint64_t(0),
+                                 [&](uint64_t S, const BranchProbability &BP) {
+                                   if (!BP.isUnknown())
+                                     return S + BP.N;
+                                   UnknownProbCount++;
+                                   return S;
+                                 });
+
+  if (UnknownProbCount > 0) {
+    BranchProbability ProbForUnknown = BranchProbability::getZero();
+    // If the sum of all known probabilities is less than one, evenly distribute
+    // the complement of sum to unknown probabilities. Otherwise, set unknown
+    // probabilities to zeros and continue to normalize known probabilities.
+    if (Sum < BranchProbability::getDenominator())
+      ProbForUnknown = BranchProbability::getRaw(
+          (BranchProbability::getDenominator() - Sum) / UnknownProbCount);
+
+    std::replace_if(Begin, End,
+                    [](const BranchProbability &BP) { return BP.isUnknown(); },
+                    ProbForUnknown);
+
+    if (Sum <= BranchProbability::getDenominator())
+      return;
+  }
   if (Sum == 0) {
     BranchProbability BP(1, std::distance(Begin, End));
     std::fill(Begin, End, BP);
@@ -174,49 +214,6 @@ void BranchProbability::normalizeProbabilities(ProbabilityIter Begin,
     I->N = (I->N * uint64_t(D) + Sum / 2) / Sum;
 }
 
-template <class WeightListIter>
-void BranchProbability::normalizeEdgeWeights(WeightListIter Begin,
-                                             WeightListIter End) {
-  // First we compute the sum with 64-bits of precision.
-  uint64_t Sum = std::accumulate(Begin, End, uint64_t(0));
-
-  if (Sum > UINT32_MAX) {
-    // Compute the scale necessary to cause the weights to fit, and re-sum with
-    // that scale applied.
-    assert(Sum / UINT32_MAX < UINT32_MAX &&
-           "The sum of weights exceeds UINT32_MAX^2!");
-    uint32_t Scale = Sum / UINT32_MAX + 1;
-    for (auto I = Begin; I != End; ++I)
-      *I /= Scale;
-    Sum = std::accumulate(Begin, End, uint64_t(0));
-  }
-
-  // Eliminate zero weights.
-  auto ZeroWeightNum = std::count(Begin, End, 0u);
-  if (ZeroWeightNum > 0) {
-    // If all weights are zeros, replace them by 1.
-    if (Sum == 0)
-      std::fill(Begin, End, 1u);
-    else {
-      // We are converting zeros into ones, and here we need to make sure that
-      // after this the sum won't exceed UINT32_MAX.
-      if (Sum + ZeroWeightNum > UINT32_MAX) {
-        for (auto I = Begin; I != End; ++I)
-          *I /= 2;
-        ZeroWeightNum = std::count(Begin, End, 0u);
-        Sum = std::accumulate(Begin, End, uint64_t(0));
-      }
-      // Scale up non-zero weights and turn zero weights into ones.
-      uint64_t ScalingFactor = (UINT32_MAX - ZeroWeightNum) / Sum;
-      assert(ScalingFactor >= 1);
-      if (ScalingFactor > 1)
-        for (auto I = Begin; I != End; ++I)
-          *I *= ScalingFactor;
-      std::replace(Begin, End, 0u, 1u);
-    }
-  }
-}
-
 }
 
 #endif