SCEV fix. In general, Add/Mul expressions should not inherit NSW/NUW.

This reverts r139450, fixes r139453, and adds much needed comments and a
unit test.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145367 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 33 insertions, 11 deletions

diff --git a/lib/Analysis/ScalarEvolution.cpp b/lib/Analysis/ScalarEvolution.cpp
index a054801..ff85906 100644
--- a/lib/Analysis/ScalarEvolution.cpp
+++ b/lib/Analysis/ScalarEvolution.cpp
@@ -3592,6 +3592,12 @@ const SCEV *ScalarEvolution::createSCEV(Value *V) {
     // because it leads to N-1 getAddExpr calls for N ultimate operands.
     // Instead, gather up all the operands and make a single getAddExpr call.
     // LLVM IR canonical form means we need only traverse the left operands.
+    //
+    // Don't apply this instruction's NSW or NUW flags to the new
+    // expression. The instruction may be guarded by control flow that the
+    // no-wrap behavior depends on. Non-control-equivalent instructions can be
+    // mapped to the same SCEV expression, and it would be incorrect to transfer
+    // NSW/NUW semantics to those operations.
     SmallVector<const SCEV *, 4> AddOps;
     AddOps.push_back(getSCEV(U->getOperand(1)));
     for (Value *Op = U->getOperand(0); ; Op = U->getOperand(0)) {
@@ -3606,16 +3612,10 @@ const SCEV *ScalarEvolution::createSCEV(Value *V) {
         AddOps.push_back(Op1);
     }
     AddOps.push_back(getSCEV(U->getOperand(0)));
-    SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap;
-    OverflowingBinaryOperator *OBO = cast<OverflowingBinaryOperator>(V);
-    if (OBO->hasNoSignedWrap())
-      Flags = setFlags(Flags, SCEV::FlagNSW);
-    if (OBO->hasNoUnsignedWrap())
-      Flags = setFlags(Flags, SCEV::FlagNUW);
-    return getAddExpr(AddOps, Flags);
+    return getAddExpr(AddOps);
   }
   case Instruction::Mul: {
-    // See the Add code above.
+    // Don't transfer NSW/NUW for the same reason as AddExpr.
     SmallVector<const SCEV *, 4> MulOps;
     MulOps.push_back(getSCEV(U->getOperand(1)));
     for (Value *Op = U->getOperand(0);
diff --git a/lib/Transforms/Scalar/IndVarSimplify.cpp b/lib/Transforms/Scalar/IndVarSimplify.cpp
index 1f5ba51..ea083e3 100644
--- a/lib/Transforms/Scalar/IndVarSimplify.cpp
+++ b/lib/Transforms/Scalar/IndVarSimplify.cpp
@@ -944,9 +944,13 @@ const SCEVAddRecExpr* WidenIV::GetExtendedOperandRecurrence(NarrowIVDefUse DU) {
   else
     return 0;
 
+  // When creating this AddExpr, don't apply the current operations NSW or NUW
+  // flags. This instruction may be guarded by control flow that the no-wrap
+  // behavior depends on. Non-control-equivalent instructions can be mapped to
+  // the same SCEV expression, and it would be incorrect to transfer NSW/NUW
+  // semantics to those operations.
   const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(
-    SE->getAddExpr(SE->getSCEV(DU.WideDef), ExtendOperExpr,
-                   IsSigned ? SCEV::FlagNSW : SCEV::FlagNUW));
+    SE->getAddExpr(SE->getSCEV(DU.WideDef), ExtendOperExpr));
 
   if (!AddRec || AddRec->getLoop() != L)
     return 0;
diff --git a/test/Analysis/ScalarEvolution/nsw.ll b/test/Analysis/ScalarEvolution/nsw.ll
index da35a6c..659cf4f 100644
--- a/test/Analysis/ScalarEvolution/nsw.ll
+++ b/test/Analysis/ScalarEvolution/nsw.ll
@@ -103,4 +103,22 @@ for.body.i.i:                                     ; preds = %entry, %for.body.i.
 ; CHECK: Loop %for.body.i.i: max backedge-taken count is ((-4 + (-1 * %begin) + %end) /u 4)
 _ZSt4fillIPiiEvT_S1_RKT0_.exit:                   ; preds = %for.body.i.i, %entry
   ret void
-}
\ No newline at end of file
+}
+
+; A single AddExpr exists for (%a + %b), which is not always <nsw>.
+; CHECK: @addnsw
+; CHECK-NOT: --> (%a + %b)<nsw>
+define i32 @addnsw(i32 %a, i32 %b) nounwind ssp {
+entry:
+  %tmp = add i32 %a, %b
+  %cmp = icmp sgt i32 %tmp, 0
+  br i1 %cmp, label %greater, label %exit
+
+greater:
+  %tmp2 = add nsw i32 %a, %b
+  br label %exit
+
+exit:
+  %result = phi i32 [ %a, %entry ], [ %tmp2, %greater ]
+  ret i32 %result
+}