Strengthen the "non-constant stride must dominate loop preheader" check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64703 91177308-0d34-0410-b5e6-96231b3b80d8

include/llvm/Analysis/ScalarEvolution.h

@@ -89,6 +89,10 @@ namespace llvm {
                                       const SCEVHandle &Conc,
                                       ScalarEvolution &SE) const = 0;
 
+    /// dominates - Return true if elements that makes up this SCEV dominates
+    /// the specified basic block.
+    virtual bool dominates(BasicBlock *BB, DominatorTree *DT) const = 0;
+
     /// print - Print out the internal representation of this scalar to the
     /// specified stream.  This should really only be used for debugging
     /// purposes.
@@ -124,6 +128,10 @@ namespace llvm {
                                       const SCEVHandle &Conc,
                                       ScalarEvolution &SE) const;
 
+    virtual bool dominates(BasicBlock *BB, DominatorTree *DT) const {
+      return true;
+    }
+
     /// Methods for support type inquiry through isa, cast, and dyn_cast:
     static inline bool classof(const SCEVCouldNotCompute *S) { return true; }
     static bool classof(const SCEV *S);

include/llvm/Analysis/ScalarEvolutionExpressions.h

@@ -20,6 +20,7 @@ namespace llvm {
   class ConstantInt;
   class ConstantRange;
   class APInt;
+  class DominatorTree;
 
   enum SCEVTypes {
     // These should be ordered in terms of increasing complexity to make the
@@ -58,6 +59,10 @@ namespace llvm {
       return this;
     }
 
+    bool dominates(BasicBlock *BB, DominatorTree *DT) const {
+      return true;
+    }
+
     virtual void print(std::ostream &OS) const;
     void print(std::ostream *OS) const { if (OS) print(*OS); }
 
@@ -100,6 +105,8 @@ namespace llvm {
       return SE.getTruncateExpr(H, Ty);
     }
 
+    virtual bool dominates(BasicBlock *BB, DominatorTree *DT) const;
+
     virtual void print(std::ostream &OS) const;
     void print(std::ostream *OS) const { if (OS) print(*OS); }
 
@@ -142,6 +149,8 @@ namespace llvm {
       return SE.getZeroExtendExpr(H, Ty);
     }
 
+    bool dominates(BasicBlock *BB, DominatorTree *DT) const;
+
     virtual void print(std::ostream &OS) const;
     void print(std::ostream *OS) const { if (OS) print(*OS); }
 
@@ -184,6 +193,8 @@ namespace llvm {
       return SE.getSignExtendExpr(H, Ty);
     }
 
+    bool dominates(BasicBlock *BB, DominatorTree *DT) const;
+
     virtual void print(std::ostream &OS) const;
     void print(std::ostream *OS) const { if (OS) print(*OS); }
 
@@ -250,6 +261,8 @@ namespace llvm {
                                                  const SCEVHandle &Conc,
                                                  ScalarEvolution &SE) const;
 
+    bool dominates(BasicBlock *BB, DominatorTree *DT) const;
+
     virtual const char *getOperationStr() const = 0;
 
     virtual const Type *getType() const { return getOperand(0)->getType(); }
@@ -343,6 +356,7 @@ namespace llvm {
         return SE.getUDivExpr(L, R);
     }
 
+    bool dominates(BasicBlock *BB, DominatorTree *DT) const;
 
     virtual const Type *getType() const;
 
@@ -437,6 +451,8 @@ namespace llvm {
                                                  const SCEVHandle &Conc,
                                                  ScalarEvolution &SE) const;
 
+    bool dominates(BasicBlock *BB, DominatorTree *DT) const;
+
     virtual void print(std::ostream &OS) const;
     void print(std::ostream *OS) const { if (OS) print(*OS); }
 
@@ -518,6 +534,8 @@ namespace llvm {
       return this;
     }
 
+    bool dominates(BasicBlock *BB, DominatorTree *DT) const;
+
     virtual const Type *getType() const;
 
     virtual void print(std::ostream &OS) const;

lib/Analysis/ScalarEvolution.cpp

@@ -66,6 +66,7 @@
 #include "llvm/GlobalVariable.h"
 #include "llvm/Instructions.h"
 #include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Analysis/Dominators.h"
 #include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Assembly/Writer.h"
 #include "llvm/Transforms/Scalar.h"
@@ -205,6 +206,10 @@ SCEVTruncateExpr::~SCEVTruncateExpr() {
   SCEVTruncates->erase(std::make_pair(Op, Ty));
 }
 
+bool SCEVTruncateExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
+  return Op->dominates(BB, DT);
+}
+
 void SCEVTruncateExpr::print(std::ostream &OS) const {
   OS << "(truncate " << *Op << " to " << *Ty << ")";
 }
@@ -227,6 +232,10 @@ SCEVZeroExtendExpr::~SCEVZeroExtendExpr() {
   SCEVZeroExtends->erase(std::make_pair(Op, Ty));
 }
 
+bool SCEVZeroExtendExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
+  return Op->dominates(BB, DT);
+}
+
 void SCEVZeroExtendExpr::print(std::ostream &OS) const {
   OS << "(zeroextend " << *Op << " to " << *Ty << ")";
 }
@@ -249,6 +258,10 @@ SCEVSignExtendExpr::~SCEVSignExtendExpr() {
   SCEVSignExtends->erase(std::make_pair(Op, Ty));
 }
 
+bool SCEVSignExtendExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
+  return Op->dominates(BB, DT);
+}
+
 void SCEVSignExtendExpr::print(std::ostream &OS) const {
   OS << "(signextend " << *Op << " to " << *Ty << ")";
 }
@@ -306,6 +319,14 @@ replaceSymbolicValuesWithConcrete(const SCEVHandle &Sym,
   return this;
 }
 
+bool SCEVCommutativeExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
+  for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
+    if (!getOperand(i)->dominates(BB, DT))
+      return false;
+  }
+  return true;
+}
+
 
 // SCEVUDivs - Only allow the creation of one SCEVUDivExpr for any particular
 // input.  Don't use a SCEVHandle here, or else the object will never be
@@ -317,6 +338,10 @@ SCEVUDivExpr::~SCEVUDivExpr() {
   SCEVUDivs->erase(std::make_pair(LHS, RHS));
 }
 
+bool SCEVUDivExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
+  return LHS->dominates(BB, DT) && RHS->dominates(BB, DT);
+}
+
 void SCEVUDivExpr::print(std::ostream &OS) const {
   OS << "(" << *LHS << " /u " << *RHS << ")";
 }
@@ -337,6 +362,15 @@ SCEVAddRecExpr::~SCEVAddRecExpr() {
                                                            Operands.end())));
 }
 
+bool SCEVAddRecExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
+  for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
+    if (!getOperand(i)->dominates(BB, DT))
+      return false;
+  }
+  return true;
+}
+
+
 SCEVHandle SCEVAddRecExpr::
 replaceSymbolicValuesWithConcrete(const SCEVHandle &Sym,
                                   const SCEVHandle &Conc,
@@ -391,6 +425,12 @@ bool SCEVUnknown::isLoopInvariant(const Loop *L) const {
   return true;
 }
 
+bool SCEVUnknown::dominates(BasicBlock *BB, DominatorTree *DT) const {
+  if (Instruction *I = dyn_cast<Instruction>(getValue()))
+    return DT->dominates(I->getParent(), BB);
+  return true;
+}
+
 const Type *SCEVUnknown::getType() const {
   return V->getType();
 }

lib/Transforms/Scalar/LoopStrengthReduce.cpp

@@ -438,16 +438,11 @@ static bool getSCEVStartAndStride(const SCEVHandle &SH, Loop *L,
   Start = SE->getAddExpr(Start, AddRec->getOperand(0));
   
   if (!isa<SCEVConstant>(AddRec->getOperand(1))) {
-    // If stride is an instruction, make sure it dominates the loop header.
+    // If stride is an instruction, make sure it dominates the loop preheader.
     // Otherwise we could end up with a use before def situation.
-    if (SCEVUnknown *SU = dyn_cast<SCEVUnknown>(AddRec->getOperand(1))) {
-      if (Instruction *I = dyn_cast<Instruction>(SU->getValue())) {
-        BasicBlock *StrideBB = I->getParent();
-        BasicBlock *Preheader = L->getLoopPreheader();
-        if (!DT->dominates(StrideBB, Preheader))
-          return false;
-      }
-    }
+    BasicBlock *Preheader = L->getLoopPreheader();
+    if (!AddRec->getOperand(1)->dominates(Preheader, DT))
+      return false;
 
     DOUT << "[" << L->getHeader()->getName()
          << "] Variable stride: " << *AddRec << "\n";

test/Transforms/LoopStrengthReduce/pr3399.ll

@@ -0,0 +1,32 @@
+; RUN: llvm-as < %s | opt -loop-reduce | llvm-dis
+; PR3399
+
+@g_53 = external global i32		; <i32*> [#uses=1]
+
+define i32 @foo() nounwind {
+bb5.thread:
+	br label %bb
+
+bb:		; preds = %bb5, %bb5.thread
+	%indvar = phi i32 [ 0, %bb5.thread ], [ %indvar.next, %bb5 ]		; <i32> [#uses=2]
+	br i1 false, label %bb5, label %bb1
+
+bb1:		; preds = %bb
+	%l_2.0.reg2mem.0 = sub i32 0, %indvar		; <i32> [#uses=1]
+	%0 = volatile load i32* @g_53, align 4		; <i32> [#uses=1]
+	%1 = trunc i32 %l_2.0.reg2mem.0 to i16		; <i16> [#uses=1]
+	%2 = trunc i32 %0 to i16		; <i16> [#uses=1]
+	%3 = mul i16 %2, %1		; <i16> [#uses=1]
+	%4 = icmp eq i16 %3, 0		; <i1> [#uses=1]
+	br i1 %4, label %bb7, label %bb2
+
+bb2:		; preds = %bb2, %bb1
+	br label %bb2
+
+bb5:		; preds = %bb
+	%indvar.next = add i32 %indvar, 1		; <i32> [#uses=1]
+	br label %bb
+
+bb7:		; preds = %bb1
+	ret i32 1
+}