Index: llvm/lib/Target/ARM/ARMISelLowering.cpp
===================================================================
--- llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -14064,7 +14064,8 @@
 }
 
 static SDValue PerformExtractEltCombine(SDNode *N,
-                                        TargetLowering::DAGCombinerInfo &DCI) {
+                                        TargetLowering::DAGCombinerInfo &DCI,
+                                        const ARMSubtarget *ST) {
   SDValue Op0 = N->getOperand(0);
   EVT VT = N->getValueType(0);
   SDLoc dl(N);
@@ -14083,6 +14084,19 @@
       return X;
   }
 
+  // extract(bitcast(BUILD_VECTOR(VMOVDRR(a, b), ..))) -> a or b
+  if (Op0.getValueType() == MVT::v4i32 &&
+      isa<ConstantSDNode>(N->getOperand(1)) &&
+      Op0.getOpcode() == ISD::BITCAST &&
+      Op0.getOperand(0).getOpcode() == ISD::BUILD_VECTOR &&
+      Op0.getOperand(0).getValueType() == MVT::v2f64) {
+    SDValue BV = Op0.getOperand(0);
+    unsigned Offset = N->getConstantOperandVal(1);
+    SDValue MOV = BV.getOperand(Offset < 2 ? 0 : 1);
+    if (MOV.getOpcode() == ARMISD::VMOVDRR)
+      return MOV.getOperand(ST->isLittle() ? Offset % 2 : 1 - Offset % 2);
+  }
+
   return SDValue();
 }
 
@@ -16502,7 +16516,8 @@
   case ISD::STORE:      return PerformSTORECombine(N, DCI, Subtarget);
   case ISD::BUILD_VECTOR: return PerformBUILD_VECTORCombine(N, DCI, Subtarget);
   case ISD::INSERT_VECTOR_ELT: return PerformInsertEltCombine(N, DCI);
-  case ISD::EXTRACT_VECTOR_ELT: return PerformExtractEltCombine(N, DCI);
+  case ISD::EXTRACT_VECTOR_ELT:
+    return PerformExtractEltCombine(N, DCI, Subtarget);
   case ISD::VECTOR_SHUFFLE: return PerformVECTOR_SHUFFLECombine(N, DCI.DAG);
   case ARMISD::VDUPLANE: return PerformVDUPLANECombine(N, DCI, Subtarget);
   case ARMISD::VDUP: return PerformVDUPCombine(N, DCI, Subtarget);
Index: llvm/test/CodeGen/Thumb2/mve-soft-float-abi.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/mve-soft-float-abi.ll
+++ llvm/test/CodeGen/Thumb2/mve-soft-float-abi.ll
@@ -51,27 +51,21 @@
 }
 
 define <2 x i64> @vector_add_i64(<2 x i64> %lhs, <2 x i64> %rhs) {
-; CHECK-FP-LABEL: vector_add_i64:
-; CHECK-FP:       @ %bb.0: @ %entry
-; CHECK-FP-NEXT:    .save {r7, lr}
-; CHECK-FP-NEXT:    push {r7, lr}
-; CHECK-FP-NEXT:    vmov d1, r2, r3
-; CHECK-FP-NEXT:    vmov d0, r0, r1
-; CHECK-FP-NEXT:    add r0, sp, #8
-; CHECK-FP-NEXT:    vldrw.u32 q1, [r0]
-; CHECK-FP-NEXT:    vmov lr, s2
-; CHECK-FP-NEXT:    vmov r0, s0
-; CHECK-FP-NEXT:    vmov r3, s4
-; CHECK-FP-NEXT:    vmov r1, s1
-; CHECK-FP-NEXT:    vmov r2, s5
-; CHECK-FP-NEXT:    vmov r12, s3
-; CHECK-FP-NEXT:    adds r0, r0, r3
-; CHECK-FP-NEXT:    vmov r3, s7
-; CHECK-FP-NEXT:    adcs r1, r2
-; CHECK-FP-NEXT:    vmov r2, s6
-; CHECK-FP-NEXT:    adds.w r2, r2, lr
-; CHECK-FP-NEXT:    adc.w r3, r3, r12
-; CHECK-FP-NEXT:    pop {r7, pc}
+; CHECK-LABEL: vector_add_i64:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    .save {r7, lr}
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    add.w r12, sp, #8
+; CHECK-NEXT:    vldrw.u32 q0, [r12]
+; CHECK-NEXT:    vmov lr, s0
+; CHECK-NEXT:    vmov r12, s1
+; CHECK-NEXT:    adds.w r0, r0, lr
+; CHECK-NEXT:    vmov lr, s2
+; CHECK-NEXT:    adc.w r1, r1, r12
+; CHECK-NEXT:    vmov r12, s3
+; CHECK-NEXT:    adds.w r2, r2, lr
+; CHECK-NEXT:    adc.w r3, r3, r12
+; CHECK-NEXT:    pop {r7, pc}
 entry:
   %sum = add <2 x i64> %lhs, %rhs
   ret <2 x i64> %sum