[SCEV] Cache results during GetMinTrailingZeros query

Differential Revision: https://reviews.llvm.org/D29759

llvm-svn: 295060

include/llvm/Analysis/ScalarEvolution.h

@@ -543,6 +543,12 @@ private:
   /// predicate by splitting it into a set of independent predicates.
   bool ProvingSplitPredicate;
 
+  /// Memoized values for the GetMinTrailingZeros
+  DenseMap<const SCEV *, uint32_t> MinTrailingZerosCache;
+
+  /// Private helper method for the GetMinTrailingZeros method
+  uint32_t GetMinTrailingZerosImpl(const SCEV *S);
+
   /// Information about the number of loop iterations for which a loop exit's
   /// branch condition evaluates to the not-taken path.  This is a temporary
   /// pair of exact and max expressions that are eventually summarized in

lib/Analysis/ScalarEvolution.cpp

@@ -4437,8 +4437,7 @@ const SCEV *ScalarEvolution::createNodeForGEP(GEPOperator *GEP) {
   return getGEPExpr(GEP, IndexExprs);
 }
 
-uint32_t
-ScalarEvolution::GetMinTrailingZeros(const SCEV *S) {
+uint32_t ScalarEvolution::GetMinTrailingZerosImpl(const SCEV *S) {
   if (const SCEVConstant *C = dyn_cast<SCEVConstant>(S))
     return C->getAPInt().countTrailingZeros();
 
@@ -4448,14 +4447,16 @@ ScalarEvolution::GetMinTrailingZeros(const SCEV *S) {
 
   if (const SCEVZeroExtendExpr *E = dyn_cast<SCEVZeroExtendExpr>(S)) {
     uint32_t OpRes = GetMinTrailingZeros(E->getOperand());
-    return OpRes == getTypeSizeInBits(E->getOperand()->getType()) ?
-             getTypeSizeInBits(E->getType()) : OpRes;
+    return OpRes == getTypeSizeInBits(E->getOperand()->getType())
+               ? getTypeSizeInBits(E->getType())
+               : OpRes;
   }
 
   if (const SCEVSignExtendExpr *E = dyn_cast<SCEVSignExtendExpr>(S)) {
     uint32_t OpRes = GetMinTrailingZeros(E->getOperand());
-    return OpRes == getTypeSizeInBits(E->getOperand()->getType()) ?
-             getTypeSizeInBits(E->getType()) : OpRes;
+    return OpRes == getTypeSizeInBits(E->getOperand()->getType())
+               ? getTypeSizeInBits(E->getType())
+               : OpRes;
   }
 
   if (const SCEVAddExpr *A = dyn_cast<SCEVAddExpr>(S)) {
@@ -4472,8 +4473,8 @@ ScalarEvolution::GetMinTrailingZeros(const SCEV *S) {
     uint32_t BitWidth = getTypeSizeInBits(M->getType());
     for (unsigned i = 1, e = M->getNumOperands();
          SumOpRes != BitWidth && i != e; ++i)
-      SumOpRes = std::min(SumOpRes + GetMinTrailingZeros(M->getOperand(i)),
-                          BitWidth);
+      SumOpRes =
+          std::min(SumOpRes + GetMinTrailingZeros(M->getOperand(i)), BitWidth);
     return SumOpRes;
   }
 
@@ -4514,6 +4515,17 @@ ScalarEvolution::GetMinTrailingZeros(const SCEV *S) {
   return 0;
 }
 
+uint32_t ScalarEvolution::GetMinTrailingZeros(const SCEV *S) {
+  auto I = MinTrailingZerosCache.find(S);
+  if (I != MinTrailingZerosCache.end())
+    return I->second;
+
+  uint32_t Result = GetMinTrailingZerosImpl(S);
+  auto InsertPair = MinTrailingZerosCache.insert({S, Result});
+  assert(InsertPair.second && "Should insert a new key");
+  return InsertPair.first->second;
+}
+
 /// Helper method to assign a range to V from metadata present in the IR.
 static Optional<ConstantRange> GetRangeFromMetadata(Value *V) {
   if (Instruction *I = dyn_cast<Instruction>(V))
@@ -9510,6 +9522,7 @@ ScalarEvolution::ScalarEvolution(ScalarEvolution &&Arg)
       ValueExprMap(std::move(Arg.ValueExprMap)),
       PendingLoopPredicates(std::move(Arg.PendingLoopPredicates)),
       WalkingBEDominatingConds(false), ProvingSplitPredicate(false),
+      MinTrailingZerosCache(std::move(Arg.MinTrailingZerosCache)),
       BackedgeTakenCounts(std::move(Arg.BackedgeTakenCounts)),
       PredicatedBackedgeTakenCounts(
           std::move(Arg.PredicatedBackedgeTakenCounts)),
@@ -9915,6 +9928,7 @@ void ScalarEvolution::forgetMemoizedResults(const SCEV *S) {
   SignedRanges.erase(S);
   ExprValueMap.erase(S);
   HasRecMap.erase(S);
+  MinTrailingZerosCache.erase(S);
 
   auto RemoveSCEVFromBackedgeMap =
       [S, this](DenseMap<const Loop *, BackedgeTakenInfo> &Map) {

test/Analysis/ScalarEvolution/pr18606-min-zeros.ll

@@ -0,0 +1,63 @@
+; RUN: opt -S -indvars < %s | FileCheck %s
+
+; CHECK: @test
+; CHECK: %5 = add i32 %local_6_, %local_0_
+; CEHCK: %37 = mul i32 %36, %36
+
+define i32 @test(i32, i32) {
+bci_0:
+  br label %bci_30
+
+bci_68:                                           ; preds = %bci_45
+  %local_6_.lcssa = phi i32 [ %local_6_, %bci_45 ]
+  %.lcssa1.lcssa = phi i32 [ %37, %bci_45 ]
+  %.lcssa.lcssa = phi i32 [ 34, %bci_45 ]
+  %2 = add i32 %local_6_.lcssa, 262
+  %3 = add i32 %2, %.lcssa1.lcssa
+  %4 = add i32 %3, %.lcssa.lcssa
+  ret i32 %4
+
+bci_30:                                           ; preds = %bci_45, %bci_0
+  %local_0_ = phi i32 [ %0, %bci_0 ], [ %38, %bci_45 ]
+  %local_6_ = phi i32 [ 2, %bci_0 ], [ %39, %bci_45 ]
+  %5 = add i32 %local_6_, %local_0_
+  br label %bci_45
+
+bci_45:                                           ; preds = %bci_30
+  %6 = mul i32 %5, %5
+  %7 = mul i32 %6, %6
+  %8 = mul i32 %7, %7
+  %9 = mul i32 %8, %8
+  %10 = mul i32 %9, %9
+  %11 = mul i32 %10, %10
+  %12 = mul i32 %11, %11
+  %13 = mul i32 %12, %12
+  %14 = mul i32 %13, %13
+  %15 = mul i32 %14, %14
+  %16 = mul i32 %15, %15
+  %17 = mul i32 %16, %16
+  %18 = mul i32 %17, %17
+  %19 = mul i32 %18, %18
+  %20 = mul i32 %19, %19
+  %21 = mul i32 %20, %20
+  %22 = mul i32 %21, %21
+  %23 = mul i32 %22, %22
+  %24 = mul i32 %23, %23
+  %25 = mul i32 %24, %24
+  %26 = mul i32 %25, %25
+  %27 = mul i32 %26, %26
+  %28 = mul i32 %27, %27
+  %29 = mul i32 %28, %28
+  %30 = mul i32 %29, %29
+  %31 = mul i32 %30, %30
+  %32 = mul i32 %31, %31
+  %33 = mul i32 %32, %32
+  %34 = mul i32 %33, %33
+  %35 = mul i32 %34, %34
+  %36 = mul i32 %35, %35
+  %37 = mul i32 %36, %36
+  %38 = add i32 %37, -11
+  %39 = add i32 %local_6_, 1
+  %40 = icmp sgt i32 %39, 76
+  br i1 %40, label %bci_68, label %bci_30
+}