Index: llvm/lib/Target/ARM/ARMISelLowering.h
===================================================================
--- llvm/lib/Target/ARM/ARMISelLowering.h
+++ llvm/lib/Target/ARM/ARMISelLowering.h
@@ -350,6 +350,13 @@
     SDValue PerformCMOVToBFICombine(SDNode *N, SelectionDAG &DAG) const;
     SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
 
+    bool SimplifyDemandedBitsForTargetNode(SDValue Op,
+                                           const APInt &OriginalDemandedBits,
+                                           const APInt &OriginalDemandedElts,
+                                           KnownBits &Known,
+                                           TargetLoweringOpt &TLO,
+                                           unsigned Depth) const override;
+
     bool isDesirableToTransformToIntegerOp(unsigned Opc, EVT VT) const override;
 
     /// allowsMisalignedMemoryAccesses - Returns true if the target allows
Index: llvm/lib/Target/ARM/ARMISelLowering.cpp
===================================================================
--- llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -16167,6 +16167,35 @@
   return false;
 }
 
+bool ARMTargetLowering::SimplifyDemandedBitsForTargetNode(
+    SDValue Op, const APInt &OriginalDemandedBits,
+    const APInt &OriginalDemandedElts, KnownBits &Known, TargetLoweringOpt &TLO,
+    unsigned Depth) const {
+  unsigned Opc = Op.getOpcode();
+
+  switch (Opc) {
+  case ARMISD::ASRL:
+  case ARMISD::LSRL: {
+    // If this is result 0 and the other result is unused, see if the demand
+    // bits allow us to shrink this long shift into a standard small shift in
+    // the opposite direction.
+    if (Op.getResNo() == 0 && !Op->hasAnyUseOfValue(1) &&
+        isa<ConstantSDNode>(Op->getOperand(2))) {
+      unsigned ShAmt = Op->getConstantOperandVal(2);
+      if (ShAmt < 32 && OriginalDemandedBits.isSubsetOf(
+                             APInt::getAllOnesValue(32) << (32 - ShAmt)))
+        return TLO.CombineTo(
+            Op, TLO.DAG.getNode(
+                    ISD::SHL, SDLoc(Op), MVT::i32, Op.getOperand(1),
+                    TLO.DAG.getConstant(32 - ShAmt, SDLoc(Op), MVT::i32)));
+    }
+    break;
+  }
+  }
+
+  return TargetLowering::SimplifyDemandedBitsForTargetNode(
+      Op, OriginalDemandedBits, OriginalDemandedElts, Known, TLO, Depth);
+}
 
 //===----------------------------------------------------------------------===//
 //                           ARM Inline Assembly Support
Index: llvm/test/CodeGen/Thumb2/fir.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/fir.ll
+++ llvm/test/CodeGen/Thumb2/fir.ll
@@ -3,27 +3,15 @@
 ; RUN: llc --verify-machineinstrs -mtriple=thumbv8.1m.main-none-eabi -mattr=+dsp %s -o - | FileCheck %s -check-prefix=CHECK --check-prefix=CHECK-NOMVE
 
 define void @test1(i32* %p0, i32 *%p1, i32 *%p2, i32 *%pDst) {
-; CHECK-MVE-LABEL: test1:
-; CHECK-MVE:       @ %bb.0: @ %entry
-; CHECK-MVE-NEXT:    ldr r1, [r1]
-; CHECK-MVE-NEXT:    ldr r2, [r2]
-; CHECK-MVE-NEXT:    ldr r0, [r0]
-; CHECK-MVE-NEXT:    smull r2, r1, r2, r1
-; CHECK-MVE-NEXT:    lsrl r2, r1, #31
-; CHECK-MVE-NEXT:    bic r1, r2, #1
-; CHECK-MVE-NEXT:    add r0, r1
-; CHECK-MVE-NEXT:    str r0, [r3]
-; CHECK-MVE-NEXT:    bx lr
-;
-; CHECK-NOMVE-LABEL: test1:
-; CHECK-NOMVE:       @ %bb.0: @ %entry
-; CHECK-NOMVE-NEXT:    ldr r1, [r1]
-; CHECK-NOMVE-NEXT:    ldr r2, [r2]
-; CHECK-NOMVE-NEXT:    ldr r0, [r0]
-; CHECK-NOMVE-NEXT:    smmul r1, r2, r1
-; CHECK-NOMVE-NEXT:    add.w r0, r0, r1, lsl #1
-; CHECK-NOMVE-NEXT:    str r0, [r3]
-; CHECK-NOMVE-NEXT:    bx lr
+; CHECK-LABEL: test1:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    ldr r1, [r1]
+; CHECK-NEXT:    ldr r2, [r2]
+; CHECK-NEXT:    ldr r0, [r0]
+; CHECK-NEXT:    smmul r1, r2, r1
+; CHECK-NEXT:    add.w r0, r0, r1, lsl #1
+; CHECK-NEXT:    str r0, [r3]
+; CHECK-NEXT:    bx lr
 entry:
   %l3 = load i32, i32* %p0, align 4
   %l4 = load i32, i32* %p1, align 4
Index: llvm/test/CodeGen/Thumb2/shift_parts.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/shift_parts.ll
+++ llvm/test/CodeGen/Thumb2/shift_parts.ll
@@ -422,16 +422,10 @@
 
 
 define i32 @ashr_demand_bottommask(i64 %x) {
-; CHECK-MVE-LABEL: ashr_demand_bottommask:
-; CHECK-MVE:       @ %bb.0: @ %entry
-; CHECK-MVE-NEXT:    lsrl r0, r1, #31
-; CHECK-MVE-NEXT:    bic r0, r0, #1
-; CHECK-MVE-NEXT:    bx lr
-;
-; CHECK-NON-MVE-LABEL: ashr_demand_bottommask:
-; CHECK-NON-MVE:       @ %bb.0: @ %entry
-; CHECK-NON-MVE-NEXT:    lsls r0, r1, #1
-; CHECK-NON-MVE-NEXT:    bx lr
+; CHECK-LABEL: ashr_demand_bottommask:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    lsls r0, r1, #1
+; CHECK-NEXT:    bx lr
 entry:
   %shr = ashr i64 %x, 31
   %t = trunc i64 %shr to i32
@@ -440,16 +434,10 @@
 }
 
 define i32 @lshr_demand_bottommask(i64 %x) {
-; CHECK-MVE-LABEL: lshr_demand_bottommask:
-; CHECK-MVE:       @ %bb.0: @ %entry
-; CHECK-MVE-NEXT:    lsrl r0, r1, #31
-; CHECK-MVE-NEXT:    bic r0, r0, #1
-; CHECK-MVE-NEXT:    bx lr
-;
-; CHECK-NON-MVE-LABEL: lshr_demand_bottommask:
-; CHECK-NON-MVE:       @ %bb.0: @ %entry
-; CHECK-NON-MVE-NEXT:    lsls r0, r1, #1
-; CHECK-NON-MVE-NEXT:    bx lr
+; CHECK-LABEL: lshr_demand_bottommask:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    lsls r0, r1, #1
+; CHECK-NEXT:    bx lr
 entry:
   %shr = lshr i64 %x, 31
   %t = trunc i64 %shr to i32
@@ -470,17 +458,11 @@
 }
 
 define i32 @ashr_demand_bottommask2(i64 %x) {
-; CHECK-MVE-LABEL: ashr_demand_bottommask2:
-; CHECK-MVE:       @ %bb.0: @ %entry
-; CHECK-MVE-NEXT:    lsrl r0, r1, #31
-; CHECK-MVE-NEXT:    bic r0, r0, #3
-; CHECK-MVE-NEXT:    bx lr
-;
-; CHECK-NON-MVE-LABEL: ashr_demand_bottommask2:
-; CHECK-NON-MVE:       @ %bb.0: @ %entry
-; CHECK-NON-MVE-NEXT:    mvn r0, #2
-; CHECK-NON-MVE-NEXT:    and.w r0, r0, r1, lsl #1
-; CHECK-NON-MVE-NEXT:    bx lr
+; CHECK-LABEL: ashr_demand_bottommask2:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    mvn r0, #2
+; CHECK-NEXT:    and.w r0, r0, r1, lsl #1
+; CHECK-NEXT:    bx lr
 entry:
   %shr = ashr i64 %x, 31
   %t = trunc i64 %shr to i32
@@ -489,17 +471,11 @@
 }
 
 define i32 @lshr_demand_bottommask2(i64 %x) {
-; CHECK-MVE-LABEL: lshr_demand_bottommask2:
-; CHECK-MVE:       @ %bb.0: @ %entry
-; CHECK-MVE-NEXT:    lsrl r0, r1, #31
-; CHECK-MVE-NEXT:    bic r0, r0, #3
-; CHECK-MVE-NEXT:    bx lr
-;
-; CHECK-NON-MVE-LABEL: lshr_demand_bottommask2:
-; CHECK-NON-MVE:       @ %bb.0: @ %entry
-; CHECK-NON-MVE-NEXT:    mvn r0, #2
-; CHECK-NON-MVE-NEXT:    and.w r0, r0, r1, lsl #1
-; CHECK-NON-MVE-NEXT:    bx lr
+; CHECK-LABEL: lshr_demand_bottommask2:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    mvn r0, #2
+; CHECK-NEXT:    and.w r0, r0, r1, lsl #1
+; CHECK-NEXT:    bx lr
 entry:
   %shr = lshr i64 %x, 31
   %t = trunc i64 %shr to i32