diff --git a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
--- a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
@@ -3996,12 +3996,24 @@
 
       // Now, create the subregister insert from SrcReg.
       Register InsertReg = MRI.createVirtualRegister(&AArch64::FPR128RegClass);
+      auto SubImm = [&]() {
+        switch (WideTy.getScalarSizeInBits() * NumElts) {
+        default:
+          llvm_unreachable("unexpected subreg size");
+        case 16:
+          return AArch64::hsub;
+        case 32:
+          return AArch64::ssub;
+        case 64:
+          return AArch64::dsub;
+        }
+      }();
       MachineInstr &InsMI =
           *BuildMI(MBB, I, I.getDebugLoc(),
                    TII.get(TargetOpcode::INSERT_SUBREG), InsertReg)
                .addUse(ImpDefReg)
                .addUse(SrcReg)
-               .addImm(AArch64::dsub);
+               .addImm(SubImm);
 
       constrainSelectedInstRegOperands(ImpDefMI, TII, TRI, RBI);
       constrainSelectedInstRegOperands(InsMI, TII, TRI, RBI);
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.h b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.h
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.h
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.h
@@ -35,6 +35,8 @@
                          MachineInstr &MI) const override;
 
 private:
+  bool legalizeBSwap(MachineInstr &MI, MachineRegisterInfo &MRI,
+                     MachineIRBuilder &MIRBuilder) const;
   bool legalizeVaArg(MachineInstr &MI, MachineRegisterInfo &MRI,
                      MachineIRBuilder &MIRBuilder) const;
   bool legalizeLoadStore(MachineInstr &MI, MachineRegisterInfo &MRI,
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -103,7 +103,8 @@
   getActionDefinitionsBuilder(G_BSWAP)
       .legalFor({s32, s64, v4s32, v2s32, v2s64})
       .clampScalar(0, s32, s64)
-      .widenScalarToNextPow2(0);
+      .widenScalarToNextPow2(0)
+      .customIf(typeIs(0, v2s16));
 
   getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL, G_AND, G_OR, G_XOR})
       .legalFor({s32, s64, v2s32, v4s32, v4s16, v8s16, v16s8, v8s8})
@@ -784,6 +785,8 @@
   case TargetOpcode::G_LOAD:
   case TargetOpcode::G_STORE:
     return legalizeLoadStore(MI, MRI, MIRBuilder, Observer);
+  case TargetOpcode::G_BSWAP:
+    return legalizeBSwap(MI, MRI, MIRBuilder);
   case TargetOpcode::G_SHL:
   case TargetOpcode::G_ASHR:
   case TargetOpcode::G_LSHR:
@@ -994,6 +997,35 @@
   return true;
 }
 
+bool AArch64LegalizerInfo::legalizeBSwap(MachineInstr &MI,
+                                         MachineRegisterInfo &MRI,
+                                         MachineIRBuilder &MIRBuilder) const {
+  assert(MI.getOpcode() == TargetOpcode::G_BSWAP);
+
+  Register ValReg = MI.getOperand(1).getReg();
+  const LLT v4s16 = LLT::fixed_vector(4, 16);
+  const LLT v2s32 = LLT::fixed_vector(2, 32);
+  const LLT v8s8 = LLT::fixed_vector(8, 8);
+  const LLT s32 = LLT::scalar(32);
+  assert(LLT::fixed_vector(2, 16) == MRI.getType(ValReg));
+
+  auto Undef = MIRBuilder.buildUndef(v2s32);
+  auto Bitcast = MIRBuilder.buildBitcast(s32, ValReg);
+  auto Zero = MIRBuilder.buildConstant(s32, 0);
+  auto Insert =
+      MIRBuilder.buildInsertVectorElement(v2s32, Undef, Bitcast, Zero);
+  auto Bitcast3 = MIRBuilder.buildBitcast(v8s8, Insert);
+  auto Rev32 = MIRBuilder.buildInstr(AArch64::G_REV32, {v8s8}, {Bitcast3});
+  auto Bitcast2 = MIRBuilder.buildBitcast(v2s32, Rev32);
+  auto Amt = MIRBuilder.buildConstant(v2s32, 16);
+  auto UShr =
+      MIRBuilder.buildInstr(TargetOpcode::G_LSHR, {v2s32}, {Bitcast2, Amt});
+  auto Extract = MIRBuilder.buildExtractVectorElement(s32, UShr, Zero);
+  MIRBuilder.buildBitcast({MI.getOperand(0).getReg()}, Extract);
+  MI.eraseFromParent();
+  return true;
+}
+
 bool AArch64LegalizerInfo::legalizeVaArg(MachineInstr &MI,
                                          MachineRegisterInfo &MRI,
                                          MachineIRBuilder &MIRBuilder) const {
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-bswap.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-bswap.mir
--- a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-bswap.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-bswap.mir
@@ -8,6 +8,8 @@
 
   define i16 @bswap_s16(i16 %a) { ret i16 0 }
 
+  define <2 x i16> @bswap_2xi16(<2 x i16> %a) { ret <2 x i16> <i16 0, i16 0> }
+
   attributes #0 = { nounwind readnone speculatable willreturn }
 
 ...
@@ -42,3 +44,41 @@
     RET_ReallyLR implicit $w0
 
 ...
+---
+name:            bswap_2xi16
+alignment:       4
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: _ }
+  - { id: 1, class: _ }
+liveins:
+  - { reg: '$s0' }
+frameInfo:
+  maxAlignment:    1
+machineFunctionInfo: {}
+body:             |
+  bb.1:
+    liveins: $s0
+
+    ; CHECK-LABEL: name: bswap_2xi16
+    ; CHECK: liveins: $s0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $s0
+    ; CHECK: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
+    ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[IVEC:%[0-9]+]]:_(<2 x s32>) = G_INSERT_VECTOR_ELT [[DEF]], [[BITCAST]](s32), [[C]](s32)
+    ; CHECK: [[BITCAST1:%[0-9]+]]:fpr64(<8 x s8>) = G_BITCAST [[IVEC]](<2 x s32>)
+    ; CHECK: [[REV32v8i8_:%[0-9]+]]:fpr64(<8 x s8>) = REV32v8i8 [[BITCAST1]](<8 x s8>)
+    ; CHECK: [[BITCAST2:%[0-9]+]]:fpr64(<2 x s32>) = G_BITCAST [[REV32v8i8_]](<8 x s8>)
+    ; CHECK: [[USHRv2i32_shift:%[0-9]+]]:fpr64(<2 x s32>) = USHRv2i32_shift [[BITCAST2]](<2 x s32>), 16
+    ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[USHRv2i32_shift]](<2 x s32>), [[C1]](s64)
+    ; CHECK: [[BITCAST3:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[EVEC]](s32)
+    ; CHECK: $s0 = COPY [[BITCAST3]](<2 x s16>)
+    ; CHECK: RET_ReallyLR
+    %0:_(<2 x s16>) = COPY $s0
+    %1:_(<2 x s16>) = G_BSWAP %0
+    $s0 = COPY %1(<2 x s16>)
+    RET_ReallyLR
+
+...