diff --git a/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
--- a/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -191,6 +191,10 @@
     "lsr-term-fold", cl::Hidden, cl::init(false),
     cl::desc("Attempt to replace primary IV with other IV."));
 
+static cl::opt<bool> AllowDropSolutionIfLessProfitable(
+    "lsr-drop-sol", cl::Hidden, cl::init(false),
+    cl::desc("Attempt to drop solution if it is less profitable"));
+
 STATISTIC(NumTermFold,
           "Number of terminating condition fold recognized and performed");
 
@@ -1975,6 +1979,10 @@
   /// SmallDenseSet.
   SetVector<int64_t, SmallVector<int64_t, 8>, SmallSet<int64_t, 8>> Factors;
 
+  /// The cost of the current SCEV, the best solution by LSR will be dropped if
+  /// the solution is not profitable.
+  Cost BaselineCost;
+
   /// Interesting use types, to facilitate truncation reuse.
   SmallSetVector<Type *, 4> Types;
 
@@ -3294,6 +3302,11 @@
   BranchInst *ExitBranch = nullptr;
   bool SaveCmp = TTI.canSaveCmp(L, &ExitBranch, &SE, &LI, &DT, &AC, &TLI);
 
+  // For calculating baseline cost
+  SmallPtrSet<const SCEV *, 16> Regs;
+  DenseSet<const SCEV *> VisitedRegs;
+  DenseSet<int> VisitedLSRUse;
+
   for (const IVStrideUse &U : IU) {
     Instruction *UserInst = U.getUser();
     // Skip IV users that are part of profitable IV Chains.
@@ -3387,6 +3400,14 @@
     LF.Offset = Offset;
     LU.AllFixupsOutsideLoop &= LF.isUseFullyOutsideLoop(L);
 
+    // Create SCEV as Formula for calculating baseline cost
+    if (!VisitedLSRUse.count(LUIdx) && !LF.isUseFullyOutsideLoop(L)) {
+      Formula F;
+      F.initialMatch(S, L, SE);
+      BaselineCost.RateFormula(F, Regs, VisitedRegs, LU);
+      VisitedLSRUse.insert(LUIdx);
+    }
+
     if (!LU.WidestFixupType ||
         SE.getTypeSizeInBits(LU.WidestFixupType) <
         SE.getTypeSizeInBits(LF.OperandValToReplace->getType()))
@@ -5162,6 +5183,19 @@
              });
 
   assert(Solution.size() == Uses.size() && "Malformed solution!");
+
+  if (BaselineCost.isLess(SolutionCost)) {
+    LLVM_DEBUG(dbgs() << "The baseline solution requires ";
+               BaselineCost.print(dbgs()); dbgs() << "\n");
+    if (!AllowDropSolutionIfLessProfitable)
+      LLVM_DEBUG(dbgs() << "Baseline is more profitable than chosen solution, "
+                           "add option 'lsr-drop-sol' to drop LSR solution.");
+    else {
+      LLVM_DEBUG(dbgs() << "Baseline is more profitable than chosen "
+                           "solution, dropping LSR solution.\n";);
+      Solution.clear();
+    }
+  }
 }
 
 /// Helper for AdjustInsertPositionForExpand. Climb up the dominator tree far as
@@ -5706,7 +5740,8 @@
       MSSAU(MSSAU), AMK(PreferredAddresingMode.getNumOccurrences() > 0
                             ? PreferredAddresingMode
                             : TTI.getPreferredAddressingMode(L, &SE)),
-      Rewriter(SE, L->getHeader()->getModule()->getDataLayout(), "lsr", false) {
+      Rewriter(SE, L->getHeader()->getModule()->getDataLayout(), "lsr", false),
+      BaselineCost(L, SE, TTI, AMK) {
   // If LoopSimplify form is not available, stay out of trouble.
   if (!L->isLoopSimplifyForm())
     return;
diff --git a/llvm/test/Transforms/LoopStrengthReduce/lsr-drop-solution.ll b/llvm/test/Transforms/LoopStrengthReduce/lsr-drop-solution.ll
--- a/llvm/test/Transforms/LoopStrengthReduce/lsr-drop-solution.ll
+++ b/llvm/test/Transforms/LoopStrengthReduce/lsr-drop-solution.ll
@@ -1,5 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -O3 -mattr=+v | FileCheck %s
+; REQUIRES: asserts
+; RUN: llc < %s -O3 -mattr=+v -lsr-drop-sol | FileCheck --check-prefix=CHECK %s
+; RUN: llc < %s -O3 -mattr=+v -debug -lsr-drop-sol 2>&1 | FileCheck --check-prefix=DEBUG %s
+; RUN: llc < %s -O3 -mattr=+v -debug 2>&1 | FileCheck --check-prefix=DEBUG2 %s
 
 target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n64-S128"
 target triple = "riscv64-unknown-linux-gnu"
@@ -16,27 +19,31 @@
 ; CHECK-NEXT:    vse8.v v8, (a3)
 ; CHECK-NEXT:    ret
 ; CHECK-NEXT:  .LBB0_2: # %if.then
-; CHECK-NEXT:    li a5, 0
-; CHECK-NEXT:    add a3, a0, a2
-; CHECK-NEXT:    sub a6, a3, a4
+; CHECK-NEXT:    add a2, a0, a2
+; CHECK-NEXT:    sub a5, a2, a4
+; CHECK-NEXT:    mv a3, a0
 ; CHECK-NEXT:  .LBB0_3: # %do.body
 ; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    add a3, a1, a5
 ; CHECK-NEXT:    vsetvli zero, a4, e8, m8, ta, mu
-; CHECK-NEXT:    vle8.v v8, (a3)
-; CHECK-NEXT:    add a7, a0, a5
-; CHECK-NEXT:    add a5, a5, a4
-; CHECK-NEXT:    add a3, a0, a5
-; CHECK-NEXT:    vse8.v v8, (a7)
-; CHECK-NEXT:    bltu a3, a6, .LBB0_3
+; CHECK-NEXT:    vle8.v v8, (a1)
+; CHECK-NEXT:    vse8.v v8, (a3)
+; CHECK-NEXT:    add a3, a3, a4
+; CHECK-NEXT:    add a1, a1, a4
+; CHECK-NEXT:    bltu a3, a5, .LBB0_3
 ; CHECK-NEXT:  # %bb.4: # %do.end
-; CHECK-NEXT:    sub a2, a2, a5
+; CHECK-NEXT:    sub a2, a2, a3
 ; CHECK-NEXT:    vsetvli a2, a2, e8, m8, ta, mu
-; CHECK-NEXT:    add a1, a1, a5
 ; CHECK-NEXT:    vsetvli zero, a2, e8, m8, ta, mu
 ; CHECK-NEXT:    vle8.v v8, (a1)
 ; CHECK-NEXT:    vse8.v v8, (a3)
 ; CHECK-NEXT:    ret
+
+;DEBUG: The chosen solution requires 3 instructions 6 regs, with addrec cost 1, plus 2 base adds, plus 5 setup cost
+;DEBUG: The baseline solution requires 2 instructions 4 regs, with addrec cost 2, plus 3 setup cost
+;DEBUG: Baseline is more profitable than chosen solution, dropping LSR solution.
+
+;DEBUG2: Baseline is more profitable than chosen solution, add option 'lsr-drop-sol' to drop LSR solution.
+
 entry:
   %0 = ptrtoint ptr %a0 to i64
   %1 = tail call i64 @llvm.riscv.vsetvli.i64(i64 %a2, i64 0, i64 3)