[SCEV] Better preserve wrapping info in SimplifyICmpOperands for UGE. (#144404)

Update SimplifyICmpOperands to only try subtracting 1 from RHS first, if
RHS is an op we can fold the subtract directly into. Otherwise try
adding to LHS first, as we can preserve NUW flags.

This improves results in a few cases, including the modified test case
from berkeley-abc and new code to be added in
#128061.

Note that there are more cases where the results can be improved by
better ordering here which I'll try to investigate as follow-up.

PR: #144404

Author: fhahn

llvm/lib/Analysis/ScalarEvolution.cpp

@@ -10892,7 +10892,12 @@ bool ScalarEvolution::SimplifyICmpOperands(CmpPredicate &Pred, const SCEV *&LHS,
     }
     break;
   case ICmpInst::ICMP_UGE:
-    if (!getUnsignedRangeMin(RHS).isMinValue()) {
+    // If RHS is an op we can fold the -1, try that first.
+    // Otherwise prefer LHS to preserve the nuw flag.
+    if ((isa<SCEVConstant>(RHS) ||
+         (isa<SCEVAddExpr, SCEVAddRecExpr>(RHS) &&
+          isa<SCEVConstant>(cast<SCEVNAryExpr>(RHS)->getOperand(0)))) &&
+        !getUnsignedRangeMin(RHS).isMinValue()) {
       RHS = getAddExpr(getConstant(RHS->getType(), (uint64_t)-1, true), RHS);
       Pred = ICmpInst::ICMP_UGT;
       Changed = true;
@@ -10901,6 +10906,10 @@ bool ScalarEvolution::SimplifyICmpOperands(CmpPredicate &Pred, const SCEV *&LHS,
                        SCEV::FlagNUW);
       Pred = ICmpInst::ICMP_UGT;
       Changed = true;
+    } else if (!getUnsignedRangeMin(RHS).isMinValue()) {
+      RHS = getAddExpr(getConstant(RHS->getType(), (uint64_t)-1, true), RHS);
+      Pred = ICmpInst::ICMP_UGT;
+      Changed = true;
     }
     break;
   default:

llvm/test/Transforms/IndVarSimplify/simplify-icmp-operands-order.ll

@@ -53,15 +53,12 @@ loop.latch:
 
 define void @test_simplifycompare_rhs_not_constant1() {
 ; CHECK-LABEL: define void @test_simplifycompare_rhs_not_constant1() {
-; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:  [[ENTRY:.*:]]
 ; CHECK-NEXT:    [[P:%.*]] = alloca i64, align 8
 ; CHECK-NEXT:    br label %[[LOOP:.*]]
 ; CHECK:       [[LOOP]]:
-; CHECK-NEXT:    [[PTR_IV:%.*]] = phi ptr [ [[P]], %[[ENTRY]] ], [ [[PTR_IV_NEXT:%.*]], %[[LOOP]] ]
-; CHECK-NEXT:    [[PTR_IV_NEXT]] = getelementptr i8, ptr [[PTR_IV]], i64 -8
-; CHECK-NEXT:    call void @use(ptr [[PTR_IV]])
-; CHECK-NEXT:    [[EC:%.*]] = icmp ult ptr [[PTR_IV_NEXT]], [[P]]
-; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]]
+; CHECK-NEXT:    call void @use(ptr [[P]])
+; CHECK-NEXT:    br i1 true, label %[[EXIT:.*]], label %[[LOOP]]
 ; CHECK:       [[EXIT]]:
 ; CHECK-NEXT:    ret void
 ;