[SCEV] Don't require NUW at first add when checking A+C1 < (A+C2)<nuw>

llvm/lib/Analysis/ScalarEvolution.cpp

@@ -11418,8 +11418,7 @@ bool ScalarEvolution::isKnownPredicateViaNoOverflow(CmpPredicate Pred,
       XNonConstOp = X;
       XFlagsPresent = ExpectedFlags;
     }
-    if (!isa<SCEVConstant>(XConstOp) ||
-        (XFlagsPresent & ExpectedFlags) != ExpectedFlags)
+    if (!isa<SCEVConstant>(XConstOp))
       return false;
 
     if (!splitBinaryAdd(Y, YConstOp, YNonConstOp, YFlagsPresent)) {
@@ -11428,12 +11427,20 @@ bool ScalarEvolution::isKnownPredicateViaNoOverflow(CmpPredicate Pred,
       YFlagsPresent = ExpectedFlags;
     }
 
-    if (!isa<SCEVConstant>(YConstOp) ||
-        (YFlagsPresent & ExpectedFlags) != ExpectedFlags)
+    if (YNonConstOp != XNonConstOp)
       return false;
 
-    if (YNonConstOp != XNonConstOp)
+    if (!isa<SCEVConstant>(YConstOp))
+      return false;
+
+    // When matching ADDs with NUW flags (and unsigned predicates), only the
+    // second ADD (with the larger constant) requires NUW.
+    if ((YFlagsPresent & ExpectedFlags) != ExpectedFlags)
+      return false;
+    if (ExpectedFlags != SCEV::FlagNUW &&
+        (XFlagsPresent & ExpectedFlags) != ExpectedFlags) {
       return false;
+    }
 
     OutC1 = cast<SCEVConstant>(XConstOp)->getAPInt();
     OutC2 = cast<SCEVConstant>(YConstOp)->getAPInt();
@@ -11472,7 +11479,7 @@ bool ScalarEvolution::isKnownPredicateViaNoOverflow(CmpPredicate Pred,
     std::swap(LHS, RHS);
     [[fallthrough]];
   case ICmpInst::ICMP_ULE:
-    // (X + C1)<nuw> u<= (X + C2)<nuw> for C1 u<= C2.
+    // (X + C1) u<= (X + C2)<nuw> for C1 u<= C2.
     if (MatchBinaryAddToConst(LHS, RHS, C1, C2, SCEV::FlagNUW) && C1.ule(C2))
       return true;
 
@@ -11482,7 +11489,7 @@ bool ScalarEvolution::isKnownPredicateViaNoOverflow(CmpPredicate Pred,
     std::swap(LHS, RHS);
     [[fallthrough]];
   case ICmpInst::ICMP_ULT:
-    // (X + C1)<nuw> u< (X + C2)<nuw> if C1 u< C2.
+    // (X + C1) u< (X + C2)<nuw> if C1 u< C2.
     if (MatchBinaryAddToConst(LHS, RHS, C1, C2, SCEV::FlagNUW) && C1.ult(C2))
       return true;
     break;

llvm/test/Analysis/LoopAccessAnalysis/different-strides-safe-dep-due-to-backedge-taken-count.ll

@@ -106,17 +106,11 @@ exit:
   ret void
 }
 
-; TOOD: The loop should be safe without dependence, as all accesses to %l are
-; completely before the first store.
 define void @backward_dep_known_safe_due_to_backedge_taken_count(ptr %A) {
 ; CHECK-LABEL: 'backward_dep_known_safe_due_to_backedge_taken_count'
 ; CHECK-NEXT:    loop:
-; CHECK-NEXT:      Memory dependences are safe with a maximum safe vector width of 8160 bits
+; CHECK-NEXT:      Memory dependences are safe
 ; CHECK-NEXT:      Dependences:
-; CHECK-NEXT:        BackwardVectorizable:
-; CHECK-NEXT:            %l = load i32, ptr %gep, align 4 ->
-; CHECK-NEXT:            store i32 %add, ptr %gep.mul.2, align 4
-; CHECK-EMPTY:
 ; CHECK-NEXT:      Run-time memory checks:
 ; CHECK-NEXT:      Grouped accesses:
 ; CHECK-EMPTY:

llvm/test/Analysis/LoopAccessAnalysis/positive-dependence-distance-different-access-sizes.ll

@@ -8,21 +8,10 @@ target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
 define void @test_distance_positive_independent_via_trip_count(ptr %A) {
 ; CHECK-LABEL: 'test_distance_positive_independent_via_trip_count'
 ; CHECK-NEXT:    loop:
-; CHECK-NEXT:      Memory dependences are safe with run-time checks
+; CHECK-NEXT:      Memory dependences are safe
 ; CHECK-NEXT:      Dependences:
 ; CHECK-NEXT:      Run-time memory checks:
-; CHECK-NEXT:      Check 0:
-; CHECK-NEXT:        Comparing group GRP0:
-; CHECK-NEXT:          %gep.A.400 = getelementptr inbounds i32, ptr %A.400, i64 %iv
-; CHECK-NEXT:        Against group GRP1:
-; CHECK-NEXT:          %gep.A = getelementptr inbounds i8, ptr %A, i64 %iv
 ; CHECK-NEXT:      Grouped accesses:
-; CHECK-NEXT:        Group GRP0:
-; CHECK-NEXT:          (Low: (400 + %A)<nuw> High: (804 + %A))
-; CHECK-NEXT:            Member: {(400 + %A)<nuw>,+,4}<nuw><%loop>
-; CHECK-NEXT:        Group GRP1:
-; CHECK-NEXT:          (Low: %A High: (101 + %A))
-; CHECK-NEXT:            Member: {%A,+,1}<nuw><%loop>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
 ; CHECK-NEXT:      SCEV assumptions: