AMDGPU/SI: Fix read2 merging into a super register.

[oota-llvm.git] / lib / Analysis / ScalarEvolution.cpp

diff --git a/lib/Analysis/ScalarEvolution.cpp b/lib/Analysis/ScalarEvolution.cpp
index 81e07e99dca14dc91f6a7eb7cf4567e1991371a9..9c7c1754e387acbc4a2dc8a539bf626084ee1b1f 100644 (file)
--- a/lib/Analysis/ScalarEvolution.cpp
+++ b/lib/Analysis/ScalarEvolution.cpp
@@ -627,7 +627,7 @@ namespace {
       llvm_unreachable("Unknown SCEV kind!");
     }
   };
-} // namespace
+}
 
 /// GroupByComplexity - Given a list of SCEV objects, order them by their
 /// complexity, and group objects of the same complexity together by value.
@@ -689,7 +689,7 @@ struct FindSCEVSize {
     return false;
   }
 };
-} // namespace
+}
 
 // Returns the size of the SCEV S.
 static inline int sizeOfSCEV(const SCEV *S) {
@@ -937,7 +937,7 @@ private:
   const SCEV *Denominator, *Quotient, *Remainder, *Zero, *One;
 };
 
-} // namespace
+}
 
 //===----------------------------------------------------------------------===//
 //                      Simple SCEV method implementations
@@ -1248,7 +1248,7 @@ struct ExtendOpTraits<SCEVZeroExtendExpr> : public ExtendOpTraitsBase {
 
 const ExtendOpTraitsBase::GetExtendExprTy ExtendOpTraits<
     SCEVZeroExtendExpr>::GetExtendExpr = &ScalarEvolution::getZeroExtendExpr;
-} // namespace
+}
 
 // The recurrence AR has been shown to have no signed/unsigned wrap or something
 // close to it. Typically, if we can prove NSW/NUW for AR, then we can just as
@@ -3300,7 +3300,7 @@ namespace {
     }
     bool isDone() const { return FindOne; }
   };
-} // namespace
+}
 
 bool ScalarEvolution::checkValidity(const SCEV *S) const {
   FindInvalidSCEVUnknown F;
@@ -7594,7 +7594,7 @@ struct FindUndefs {
     return Found;
   }
 };
-} // namespace
+}
 
 // Return true when S contains at least an undef value.
 static inline bool
@@ -7644,14 +7644,14 @@ struct SCEVCollectTerms {
   }
   bool isDone() const { return false; }
 };
-} // namespace
+}
 
 /// Find parametric terms in this SCEVAddRecExpr.
-void SCEVAddRecExpr::collectParametricTerms(
-    ScalarEvolution &SE, SmallVectorImpl<const SCEV *> &Terms) const {
+void ScalarEvolution::collectParametricTerms(const SCEV *Expr,
+    SmallVectorImpl<const SCEV *> &Terms) {
   SmallVector<const SCEV *, 4> Strides;
-  SCEVCollectStrides StrideCollector(SE, Strides);
-  visitAll(this, StrideCollector);
+  SCEVCollectStrides StrideCollector(*this, Strides);
+  visitAll(Expr, StrideCollector);
 
   DEBUG({
       dbgs() << "Strides:\n";
@@ -7737,7 +7737,7 @@ struct FindParameter {
     return FoundParameter;
   }
 };
-} // namespace
+}
 
 // Returns true when S contains at least a SCEVUnknown parameter.
 static inline bool
@@ -7867,19 +7867,23 @@ void ScalarEvolution::findArrayDimensions(SmallVectorImpl<const SCEV *> &Terms,
 
 /// Third step of delinearization: compute the access functions for the
 /// Subscripts based on the dimensions in Sizes.
-void SCEVAddRecExpr::computeAccessFunctions(
-    ScalarEvolution &SE, SmallVectorImpl<const SCEV *> &Subscripts,
-    SmallVectorImpl<const SCEV *> &Sizes) const {
+void ScalarEvolution::computeAccessFunctions(
+    const SCEV *Expr, SmallVectorImpl<const SCEV *> &Subscripts,
+    SmallVectorImpl<const SCEV *> &Sizes) {
 
   // Early exit in case this SCEV is not an affine multivariate function.
-  if (Sizes.empty() || !this->isAffine())
+  if (Sizes.empty())
     return;
 
-  const SCEV *Res = this;
+  if (auto AR = dyn_cast<SCEVAddRecExpr>(Expr))
+    if (!AR->isAffine())
+      return;
+
+  const SCEV *Res = Expr;
   int Last = Sizes.size() - 1;
   for (int i = Last; i >= 0; i--) {
     const SCEV *Q, *R;
-    SCEVDivision::divide(SE, Res, Sizes[i], &Q, &R);
+    SCEVDivision::divide(*this, Res, Sizes[i], &Q, &R);
 
     DEBUG({
         dbgs() << "Res: " << *Res << "\n";
@@ -7971,31 +7975,31 @@ void SCEVAddRecExpr::computeAccessFunctions(
 /// asking for the SCEV of the memory access with respect to all enclosing
 /// loops, calling SCEV->delinearize on that and printing the results.
 
-void SCEVAddRecExpr::delinearize(ScalarEvolution &SE,
+void ScalarEvolution::delinearize(const SCEV *Expr,
                                  SmallVectorImpl<const SCEV *> &Subscripts,
                                  SmallVectorImpl<const SCEV *> &Sizes,
-                                 const SCEV *ElementSize) const {
+                                 const SCEV *ElementSize) {
   // First step: collect parametric terms.
   SmallVector<const SCEV *, 4> Terms;
-  collectParametricTerms(SE, Terms);
+  collectParametricTerms(Expr, Terms);
 
   if (Terms.empty())
     return;
 
   // Second step: find subscript sizes.
-  SE.findArrayDimensions(Terms, Sizes, ElementSize);
+  findArrayDimensions(Terms, Sizes, ElementSize);
 
   if (Sizes.empty())
     return;
 
   // Third step: compute the access functions for each subscript.
-  computeAccessFunctions(SE, Subscripts, Sizes);
+  computeAccessFunctions(Expr, Subscripts, Sizes);
 
   if (Subscripts.empty())
     return;
 
   DEBUG({
-      dbgs() << "succeeded to delinearize " << *this << "\n";
+      dbgs() << "succeeded to delinearize " << *Expr << "\n";
       dbgs() << "ArrayDecl[UnknownSize]";
       for (const SCEV *S : Sizes)
         dbgs() << "[" << *S << "]";
@@ -8418,7 +8422,7 @@ struct SCEVSearch {
   }
   bool isDone() const { return IsFound; }
 };
-} // namespace
+}
 
 bool ScalarEvolution::hasOperand(const SCEV *S, const SCEV *Op) const {
   SCEVSearch Search(Op);