diff --git a/llvm/lib/Target/ARM/ARMISelLowering.h b/llvm/lib/Target/ARM/ARMISelLowering.h
--- a/llvm/lib/Target/ARM/ARMISelLowering.h
+++ b/llvm/lib/Target/ARM/ARMISelLowering.h
@@ -208,6 +208,9 @@
       VQMOVNs,      // Vector (V) Saturating (Q) Move and Narrow (N), signed (s)
       VQMOVNu,      // Vector (V) Saturating (Q) Move and Narrow (N), unsigned (u)
 
+      // MVE float <> half converts
+      VCVTN,        // MVE vcvt f32 -> f16, truncating into either the bottom or top lanes
+
       // Vector multiply long:
       VMULLs,       // ...signed
       VMULLu,       // ...unsigned
diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -1692,6 +1692,7 @@
   case ARMISD::VMOVN:         return "ARMISD::VMOVN";
   case ARMISD::VQMOVNs:       return "ARMISD::VQMOVNs";
   case ARMISD::VQMOVNu:       return "ARMISD::VQMOVNu";
+  case ARMISD::VCVTN:         return "ARMISD::VCVTN";
   case ARMISD::VMULLs:        return "ARMISD::VMULLs";
   case ARMISD::VMULLu:        return "ARMISD::VMULLu";
   case ARMISD::VADDVs:        return "ARMISD::VADDVs";
@@ -7243,6 +7244,60 @@
   return true;
 }
 
+// Reconstruct an MVE VCVT from a BuildVector of scalar fptrunc, all extract
+// from a pair of inputs. For example:
+// BUILDVECTOR(FP_ROUND(EXTRACT_ELT(X, 0),
+//             FP_ROUND(EXTRACT_ELT(Y, 0),
+//             FP_ROUND(EXTRACT_ELT(X, 1),
+//             FP_ROUND(EXTRACT_ELT(Y, 1), ...)
+static SDValue LowerBuildVectorOfFPTrunc(SDValue BV, SelectionDAG &DAG,
+                                         const ARMSubtarget *ST) {
+  assert(BV.getOpcode() == ISD::BUILD_VECTOR && "Unknown opcode!");
+  if (!ST->hasMVEFloatOps())
+    return SDValue();
+
+  SDLoc dl(BV);
+  EVT VT = BV.getValueType();
+  if (VT != MVT::v8f16)
+    return SDValue();
+
+  // We are looking for a buildvector of fptrunc elements, where all the
+  // elements are interleavingly extracted from two sources. Check the first two
+  // items are valid enough and extract some info from them (they are checked
+  // properly in the loop below).
+  if (BV.getOperand(0).getOpcode() != ISD::FP_ROUND ||
+      BV.getOperand(0).getOperand(0).getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
+      BV.getOperand(0).getOperand(0).getConstantOperandVal(1) != 0)
+    return SDValue();
+  if (BV.getOperand(1).getOpcode() != ISD::FP_ROUND ||
+      BV.getOperand(1).getOperand(0).getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
+      BV.getOperand(1).getOperand(0).getConstantOperandVal(1) != 0)
+    return SDValue();
+  SDValue Op0 = BV.getOperand(0).getOperand(0).getOperand(0);
+  SDValue Op1 = BV.getOperand(1).getOperand(0).getOperand(0);
+  if (Op0.getValueType() != MVT::v4f32 || Op1.getValueType() != MVT::v4f32)
+    return SDValue();
+
+  // Check all the values in the BuildVector line up with our expectations.
+  for (int i = 1; i < 4; i++) {
+    auto Check = [](SDValue Trunc, SDValue Op, int Idx) {
+      return Trunc.getOpcode() == ISD::FP_ROUND &&
+             Trunc.getOperand(0).getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
+             Trunc.getOperand(0).getOperand(0) == Op &&
+             Trunc.getOperand(0).getConstantOperandVal(1) == Idx;
+    };
+    if (!Check(BV.getOperand(i * 2 + 0), Op0, i))
+      return SDValue();
+    if (!Check(BV.getOperand(i * 2 + 1), Op1, i))
+      return SDValue();
+  }
+
+  SDValue N1 = DAG.getNode(ARMISD::VCVTN, dl, VT, DAG.getUNDEF(VT), Op0,
+                           DAG.getConstant(0, dl, MVT::i32));
+  return DAG.getNode(ARMISD::VCVTN, dl, VT, N1, Op1,
+                     DAG.getConstant(1, dl, MVT::i32));
+}
+
 // If N is an integer constant that can be moved into a register in one
 // instruction, return an SDValue of such a constant (will become a MOV
 // instruction).  Otherwise return null.
@@ -7498,12 +7553,16 @@
   if (isConstant)
     return SDValue();
 
-  // Empirical tests suggest this is rarely worth it for vectors of length <= 2.
-  if (NumElts >= 4) {
-    SDValue shuffle = ReconstructShuffle(Op, DAG);
-    if (shuffle != SDValue())
+  // Reconstruct the BUILDVECTOR to one of the legal shuffles (such as vext and
+  // vmovn). Empirical tests suggest this is rarely worth it for vectors of
+  // length <= 2.
+  if (NumElts >= 4)
+    if (SDValue shuffle = ReconstructShuffle(Op, DAG))
       return shuffle;
-  }
+
+  // Attempt to turn a buildvector of scalar fptrunc's back into VCVT's
+  if (SDValue VCVT = LowerBuildVectorOfFPTrunc(Op, DAG, Subtarget))
+    return VCVT;
 
   if (ST->hasNEON() && VT.is128BitVector() && VT != MVT::v2f64 && VT != MVT::v4f32) {
     // If we haven't found an efficient lowering, try splitting a 128-bit vector
diff --git a/llvm/lib/Target/ARM/ARMInstrMVE.td b/llvm/lib/Target/ARM/ARMInstrMVE.td
--- a/llvm/lib/Target/ARM/ARMInstrMVE.td
+++ b/llvm/lib/Target/ARM/ARMInstrMVE.td
@@ -4782,6 +4782,8 @@
   let retainsPreviousHalfElement = 1;
 }
 
+def MVEvcvtn       : SDNode<"ARMISD::VCVTN", SDTARMVMOVNQ>;
+
 multiclass MVE_VCVT_f2h_m<string iname, int half> {
   def "": MVE_VCVT_ff<iname, "f16.f32", 0b0, half,
                       (ins MQPR:$Qd_src), vpred_n, "$Qd = $Qd_src">;
@@ -4796,6 +4798,9 @@
                          (v4i1 VCCR:$mask))),
               (v8f16 (Inst (v8f16 MQPR:$Qd_src), (v4f32 MQPR:$Qm),
                            ARMVCCThen, (v4i1 VCCR:$mask)))>;
+
+    def : Pat<(v8f16 (MVEvcvtn (v8f16 MQPR:$Qd_src), (v4f32 MQPR:$Qm), (i32 half))),
+              (v8f16 (Inst (v8f16 MQPR:$Qd_src), (v4f32 MQPR:$Qm)))>;
   }
 }
 
diff --git a/llvm/test/CodeGen/Thumb2/mve-fp16convertloops.ll b/llvm/test/CodeGen/Thumb2/mve-fp16convertloops.ll
--- a/llvm/test/CodeGen/Thumb2/mve-fp16convertloops.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-fp16convertloops.ll
@@ -726,46 +726,24 @@
 ; CHECK-NEXT:    dls lr, lr
 ; CHECK-NEXT:  .LBB9_1: @ %vector.body
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    vldrh.u16 q2, [r0], #16
-; CHECK-NEXT:    vcvtb.f32.f16 s7, s11
-; CHECK-NEXT:    vmovx.f16 s13, s11
-; CHECK-NEXT:    vcvtb.f32.f16 s6, s10
-; CHECK-NEXT:    vmovx.f16 s14, s8
-; CHECK-NEXT:    vcvtb.f32.f16 s5, s9
-; CHECK-NEXT:    vcvtb.f32.f16 s4, s8
-; CHECK-NEXT:    vmovx.f16 s8, s10
-; CHECK-NEXT:    vmul.f32 q1, q1, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s4
-; CHECK-NEXT:    vcvtb.f32.f16 s19, s13
-; CHECK-NEXT:    vmov r2, s12
-; CHECK-NEXT:    vmovx.f16 s12, s9
-; CHECK-NEXT:    vcvtb.f32.f16 s18, s8
-; CHECK-NEXT:    vcvtb.f32.f16 s17, s12
-; CHECK-NEXT:    vcvtb.f32.f16 s16, s14
-; CHECK-NEXT:    vmul.f32 q2, q4, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s5
-; CHECK-NEXT:    vmov r3, s12
-; CHECK-NEXT:    vmov.16 q3[0], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vmov.16 q3[1], r3
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s9
-; CHECK-NEXT:    vmov.16 q3[2], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s6
-; CHECK-NEXT:    vmov.16 q3[3], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s10
-; CHECK-NEXT:    vmov.16 q3[4], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q3[5], r2
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s11
-; CHECK-NEXT:    vmov.16 q3[6], r2
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vmov.16 q3[7], r2
-; CHECK-NEXT:    vstrb.8 q3, [r1], #16
+; CHECK-NEXT:    vldrh.u16 q1, [r0], #16
+; CHECK-NEXT:    vmovx.f16 s8, s7
+; CHECK-NEXT:    vmovx.f16 s14, s6
+; CHECK-NEXT:    vcvtb.f32.f16 s11, s8
+; CHECK-NEXT:    vmovx.f16 s13, s5
+; CHECK-NEXT:    vcvtb.f32.f16 s10, s14
+; CHECK-NEXT:    vmovx.f16 s12, s4
+; CHECK-NEXT:    vcvtb.f32.f16 s9, s13
+; CHECK-NEXT:    vcvtb.f32.f16 s19, s7
+; CHECK-NEXT:    vcvtb.f32.f16 s18, s6
+; CHECK-NEXT:    vcvtb.f32.f16 s17, s5
+; CHECK-NEXT:    vcvtb.f32.f16 s16, s4
+; CHECK-NEXT:    vcvtb.f32.f16 s8, s12
+; CHECK-NEXT:    vmul.f32 q1, q4, q0
+; CHECK-NEXT:    vmul.f32 q2, q2, q0
+; CHECK-NEXT:    vcvtb.f16.f32 q1, q1
+; CHECK-NEXT:    vcvtt.f16.f32 q1, q2
+; CHECK-NEXT:    vstrb.8 q1, [r1], #16
 ; CHECK-NEXT:    le lr, .LBB9_1
 ; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
 ; CHECK-NEXT:    vpop {d8, d9}
@@ -817,85 +795,41 @@
 ; CHECK-NEXT:    dls lr, lr
 ; CHECK-NEXT:  .LBB10_1: @ %vector.body
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    vldrh.u16 q2, [r0]
-; CHECK-NEXT:    vcvtb.f32.f16 s7, s11
-; CHECK-NEXT:    vmovx.f16 s13, s11
-; CHECK-NEXT:    vcvtb.f32.f16 s6, s10
-; CHECK-NEXT:    vmovx.f16 s14, s8
-; CHECK-NEXT:    vcvtb.f32.f16 s5, s9
-; CHECK-NEXT:    vcvtb.f32.f16 s4, s8
-; CHECK-NEXT:    vmovx.f16 s8, s10
-; CHECK-NEXT:    vmul.f32 q1, q1, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s4
-; CHECK-NEXT:    vcvtb.f32.f16 s19, s13
-; CHECK-NEXT:    vmov r2, s12
-; CHECK-NEXT:    vmovx.f16 s12, s9
-; CHECK-NEXT:    vcvtb.f32.f16 s18, s8
-; CHECK-NEXT:    vcvtb.f32.f16 s17, s12
-; CHECK-NEXT:    vcvtb.f32.f16 s16, s14
-; CHECK-NEXT:    vmul.f32 q2, q4, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s5
-; CHECK-NEXT:    vmov r3, s12
-; CHECK-NEXT:    vmov.16 q3[0], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vmov.16 q3[1], r3
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s9
-; CHECK-NEXT:    vmov.16 q3[2], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s6
-; CHECK-NEXT:    vmov.16 q3[3], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s10
-; CHECK-NEXT:    vmov.16 q3[4], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q3[5], r2
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s11
-; CHECK-NEXT:    vmov.16 q3[6], r2
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vldrh.u16 q2, [r0, #16]!
-; CHECK-NEXT:    vmov.16 q3[7], r2
-; CHECK-NEXT:    vstrh.16 q3, [r1]
-; CHECK-NEXT:    vmovx.f16 s12, s11
-; CHECK-NEXT:    vmovx.f16 s14, s10
+; CHECK-NEXT:    vldrh.u16 q1, [r0]
+; CHECK-NEXT:    vmovx.f16 s8, s7
+; CHECK-NEXT:    vmovx.f16 s14, s6
+; CHECK-NEXT:    vcvtb.f32.f16 s11, s8
+; CHECK-NEXT:    vmovx.f16 s13, s5
+; CHECK-NEXT:    vcvtb.f32.f16 s10, s14
+; CHECK-NEXT:    vmovx.f16 s12, s4
+; CHECK-NEXT:    vcvtb.f32.f16 s9, s13
+; CHECK-NEXT:    vcvtb.f32.f16 s19, s7
+; CHECK-NEXT:    vcvtb.f32.f16 s18, s6
+; CHECK-NEXT:    vcvtb.f32.f16 s17, s5
+; CHECK-NEXT:    vcvtb.f32.f16 s16, s4
+; CHECK-NEXT:    vcvtb.f32.f16 s8, s12
+; CHECK-NEXT:    vmul.f32 q1, q4, q0
+; CHECK-NEXT:    vmul.f32 q2, q2, q0
+; CHECK-NEXT:    vcvtb.f16.f32 q1, q1
+; CHECK-NEXT:    vcvtt.f16.f32 q1, q2
+; CHECK-NEXT:    vstrh.16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q1, [r0, #16]!
+; CHECK-NEXT:    vmovx.f16 s12, s7
+; CHECK-NEXT:    vmovx.f16 s14, s6
 ; CHECK-NEXT:    vcvtb.f32.f16 s19, s12
-; CHECK-NEXT:    vmovx.f16 s4, s9
+; CHECK-NEXT:    vmovx.f16 s8, s5
 ; CHECK-NEXT:    vcvtb.f32.f16 s18, s14
-; CHECK-NEXT:    vmovx.f16 s6, s8
-; CHECK-NEXT:    vcvtb.f32.f16 s17, s4
-; CHECK-NEXT:    vcvtb.f32.f16 s16, s6
+; CHECK-NEXT:    vmovx.f16 s10, s4
+; CHECK-NEXT:    vcvtb.f32.f16 s17, s8
+; CHECK-NEXT:    vcvtb.f32.f16 s16, s10
+; CHECK-NEXT:    vcvtb.f32.f16 s11, s7
+; CHECK-NEXT:    vcvtb.f32.f16 s10, s6
+; CHECK-NEXT:    vcvtb.f32.f16 s9, s5
+; CHECK-NEXT:    vcvtb.f32.f16 s8, s4
 ; CHECK-NEXT:    vmul.f32 q1, q4, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s4
-; CHECK-NEXT:    vmov r2, s12
-; CHECK-NEXT:    vcvtb.f32.f16 s15, s11
-; CHECK-NEXT:    vcvtb.f32.f16 s14, s10
-; CHECK-NEXT:    vcvtb.f32.f16 s13, s9
-; CHECK-NEXT:    vcvtb.f32.f16 s12, s8
-; CHECK-NEXT:    vmul.f32 q3, q3, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s13
-; CHECK-NEXT:    vmov r3, s8
-; CHECK-NEXT:    vmov.16 q2[0], r3
-; CHECK-NEXT:    vmov.16 q2[1], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s5
-; CHECK-NEXT:    vmov.16 q2[2], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s14
-; CHECK-NEXT:    vmov.16 q2[3], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s6
-; CHECK-NEXT:    vmov.16 q2[4], r2
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s15
-; CHECK-NEXT:    vmov.16 q2[5], r2
-; CHECK-NEXT:    vmov r2, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q2[6], r2
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vmov.16 q2[7], r2
+; CHECK-NEXT:    vmul.f32 q2, q2, q0
+; CHECK-NEXT:    vcvtb.f16.f32 q2, q2
+; CHECK-NEXT:    vcvtt.f16.f32 q2, q1
 ; CHECK-NEXT:    vstrb.8 q2, [r1, #16]!
 ; CHECK-NEXT:    le lr, .LBB10_1
 ; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
diff --git a/llvm/test/CodeGen/Thumb2/mve-vcvt16.ll b/llvm/test/CodeGen/Thumb2/mve-vcvt16.ll
--- a/llvm/test/CodeGen/Thumb2/mve-vcvt16.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-vcvt16.ll
@@ -100,31 +100,8 @@
 define arm_aapcs_vfpcc <8 x half> @shuffle_trunc1(<4 x float> %src1, <4 x float> %src2) {
 ; CHECK-LABEL: shuffle_trunc1:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmov q2, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s8
-; CHECK-NEXT:    vmov r0, s0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s4
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov.16 q0[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s9
-; CHECK-NEXT:    vmov.16 q0[1], r1
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s5
-; CHECK-NEXT:    vmov.16 q0[2], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s10
-; CHECK-NEXT:    vmov.16 q0[3], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s6
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s11
-; CHECK-NEXT:    vmov.16 q0[5], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q0[6], r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.16 q0[7], r0
+; CHECK-NEXT:    vcvtb.f16.f32 q0, q0
+; CHECK-NEXT:    vcvtt.f16.f32 q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %strided.vec = shufflevector <4 x float> %src1, <4 x float> %src2, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>
@@ -135,31 +112,9 @@
 define arm_aapcs_vfpcc <8 x half> @shuffle_trunc2(<4 x float> %src1, <4 x float> %src2) {
 ; CHECK-LABEL: shuffle_trunc2:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s4
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s0
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vmov.16 q2[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s5
-; CHECK-NEXT:    vmov.16 q2[1], r1
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s1
-; CHECK-NEXT:    vmov.16 q2[2], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s6
-; CHECK-NEXT:    vmov.16 q2[3], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s2
-; CHECK-NEXT:    vmov.16 q2[4], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q2[5], r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s3
-; CHECK-NEXT:    vmov.16 q2[6], r0
-; CHECK-NEXT:    vmov r0, s0
-; CHECK-NEXT:    vmov.16 q2[7], r0
-; CHECK-NEXT:    vmov q0, q2
+; CHECK-NEXT:    vcvtb.f16.f32 q1, q1
+; CHECK-NEXT:    vcvtt.f16.f32 q1, q0
+; CHECK-NEXT:    vmov q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %strided.vec = shufflevector <4 x float> %src1, <4 x float> %src2, <8 x i32> <i32 4, i32 0, i32 5, i32 1, i32 6, i32 2, i32 7, i32 3>
@@ -170,59 +125,10 @@
 define arm_aapcs_vfpcc <16 x half> @shuffle_trunc3(<8 x float> %src1, <8 x float> %src2) {
 ; CHECK-LABEL: shuffle_trunc3:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9, d10}
-; CHECK-NEXT:    vpush {d8, d9, d10}
-; CHECK-NEXT:    vmov q4, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s16
-; CHECK-NEXT:    vmov r0, s0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s8
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov.16 q0[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s17
-; CHECK-NEXT:    vmov.16 q0[1], r1
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s9
-; CHECK-NEXT:    vmov.16 q0[2], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s18
-; CHECK-NEXT:    vmov.16 q0[3], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s10
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s19
-; CHECK-NEXT:    vmov.16 q0[5], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s11
-; CHECK-NEXT:    vmov.16 q0[6], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s4
-; CHECK-NEXT:    vmov.16 q0[7], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s12
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vmov.16 q2[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s5
-; CHECK-NEXT:    vmov.16 q2[1], r1
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s13
-; CHECK-NEXT:    vmov.16 q2[2], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s6
-; CHECK-NEXT:    vmov.16 q2[3], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s14
-; CHECK-NEXT:    vmov.16 q2[4], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q2[5], r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s15
-; CHECK-NEXT:    vmov.16 q2[6], r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.16 q2[7], r0
-; CHECK-NEXT:    vmov q1, q2
-; CHECK-NEXT:    vpop {d8, d9, d10}
+; CHECK-NEXT:    vcvtb.f16.f32 q0, q0
+; CHECK-NEXT:    vcvtb.f16.f32 q1, q1
+; CHECK-NEXT:    vcvtt.f16.f32 q0, q2
+; CHECK-NEXT:    vcvtt.f16.f32 q1, q3
 ; CHECK-NEXT:    bx lr
 entry:
   %strided.vec = shufflevector <8 x float> %src1, <8 x float> %src2, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>
@@ -233,59 +139,12 @@
 define arm_aapcs_vfpcc <16 x half> @shuffle_trunc4(<8 x float> %src1, <8 x float> %src2) {
 ; CHECK-LABEL: shuffle_trunc4:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9, d10}
-; CHECK-NEXT:    vpush {d8, d9, d10}
-; CHECK-NEXT:    vmov q4, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s8
-; CHECK-NEXT:    vmov r0, s0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s16
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov.16 q0[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s9
-; CHECK-NEXT:    vmov.16 q0[1], r1
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s17
-; CHECK-NEXT:    vmov.16 q0[2], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s10
-; CHECK-NEXT:    vmov.16 q0[3], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s18
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s11
-; CHECK-NEXT:    vmov.16 q0[5], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s19
-; CHECK-NEXT:    vmov.16 q0[6], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s12
-; CHECK-NEXT:    vmov.16 q0[7], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s4
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vmov.16 q2[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s13
-; CHECK-NEXT:    vmov.16 q2[1], r1
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s5
-; CHECK-NEXT:    vmov.16 q2[2], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s14
-; CHECK-NEXT:    vmov.16 q2[3], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s6
-; CHECK-NEXT:    vmov.16 q2[4], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s15
-; CHECK-NEXT:    vmov.16 q2[5], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q2[6], r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.16 q2[7], r0
-; CHECK-NEXT:    vmov q1, q2
-; CHECK-NEXT:    vpop {d8, d9, d10}
+; CHECK-NEXT:    vcvtb.f16.f32 q2, q2
+; CHECK-NEXT:    vcvtb.f16.f32 q3, q3
+; CHECK-NEXT:    vcvtt.f16.f32 q2, q0
+; CHECK-NEXT:    vcvtt.f16.f32 q3, q1
+; CHECK-NEXT:    vmov q0, q2
+; CHECK-NEXT:    vmov q1, q3
 ; CHECK-NEXT:    bx lr
 entry:
   %strided.vec = shufflevector <8 x float> %src1, <8 x float> %src2, <16 x i32> <i32 8, i32 0, i32 9, i32 1, i32 10, i32 2, i32 11, i32 3, i32 12, i32 4, i32 13, i32 5, i32 14, i32 6, i32 15, i32 7>
@@ -296,31 +155,8 @@
 define arm_aapcs_vfpcc <8 x half> @shuffle_trunc5(<4 x float> %src1, <4 x float> %src2) {
 ; CHECK-LABEL: shuffle_trunc5:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmov q2, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s8
-; CHECK-NEXT:    vmov r0, s0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s4
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov.16 q0[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s9
-; CHECK-NEXT:    vmov.16 q0[1], r1
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s5
-; CHECK-NEXT:    vmov.16 q0[2], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s10
-; CHECK-NEXT:    vmov.16 q0[3], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s6
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s11
-; CHECK-NEXT:    vmov.16 q0[5], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q0[6], r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.16 q0[7], r0
+; CHECK-NEXT:    vcvtb.f16.f32 q0, q0
+; CHECK-NEXT:    vcvtt.f16.f32 q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %out1 = fptrunc <4 x float> %src1 to <4 x half>
@@ -332,31 +168,9 @@
 define arm_aapcs_vfpcc <8 x half> @shuffle_trunc6(<4 x float> %src1, <4 x float> %src2) {
 ; CHECK-LABEL: shuffle_trunc6:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s4
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s0
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vmov.16 q2[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s5
-; CHECK-NEXT:    vmov.16 q2[1], r1
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s1
-; CHECK-NEXT:    vmov.16 q2[2], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s6
-; CHECK-NEXT:    vmov.16 q2[3], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s2
-; CHECK-NEXT:    vmov.16 q2[4], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q2[5], r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s3
-; CHECK-NEXT:    vmov.16 q2[6], r0
-; CHECK-NEXT:    vmov r0, s0
-; CHECK-NEXT:    vmov.16 q2[7], r0
-; CHECK-NEXT:    vmov q0, q2
+; CHECK-NEXT:    vcvtb.f16.f32 q1, q1
+; CHECK-NEXT:    vcvtt.f16.f32 q1, q0
+; CHECK-NEXT:    vmov q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %out1 = fptrunc <4 x float> %src1 to <4 x half>
@@ -368,59 +182,10 @@
 define arm_aapcs_vfpcc <16 x half> @shuffle_trunc7(<8 x float> %src1, <8 x float> %src2) {
 ; CHECK-LABEL: shuffle_trunc7:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9, d10}
-; CHECK-NEXT:    vpush {d8, d9, d10}
-; CHECK-NEXT:    vmov q4, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s16
-; CHECK-NEXT:    vmov r0, s0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s8
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov.16 q0[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s17
-; CHECK-NEXT:    vmov.16 q0[1], r1
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s9
-; CHECK-NEXT:    vmov.16 q0[2], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s18
-; CHECK-NEXT:    vmov.16 q0[3], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s10
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s19
-; CHECK-NEXT:    vmov.16 q0[5], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s11
-; CHECK-NEXT:    vmov.16 q0[6], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s4
-; CHECK-NEXT:    vmov.16 q0[7], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s12
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vmov.16 q2[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s5
-; CHECK-NEXT:    vmov.16 q2[1], r1
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s13
-; CHECK-NEXT:    vmov.16 q2[2], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s6
-; CHECK-NEXT:    vmov.16 q2[3], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s14
-; CHECK-NEXT:    vmov.16 q2[4], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q2[5], r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s15
-; CHECK-NEXT:    vmov.16 q2[6], r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.16 q2[7], r0
-; CHECK-NEXT:    vmov q1, q2
-; CHECK-NEXT:    vpop {d8, d9, d10}
+; CHECK-NEXT:    vcvtb.f16.f32 q0, q0
+; CHECK-NEXT:    vcvtb.f16.f32 q1, q1
+; CHECK-NEXT:    vcvtt.f16.f32 q0, q2
+; CHECK-NEXT:    vcvtt.f16.f32 q1, q3
 ; CHECK-NEXT:    bx lr
 entry:
   %out1 = fptrunc <8 x float> %src1 to <8 x half>
@@ -432,59 +197,12 @@
 define arm_aapcs_vfpcc <16 x half> @shuffle_trunc8(<8 x float> %src1, <8 x float> %src2) {
 ; CHECK-LABEL: shuffle_trunc8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9, d10}
-; CHECK-NEXT:    vpush {d8, d9, d10}
-; CHECK-NEXT:    vmov q4, q0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s8
-; CHECK-NEXT:    vmov r0, s0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s16
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov.16 q0[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s9
-; CHECK-NEXT:    vmov.16 q0[1], r1
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s17
-; CHECK-NEXT:    vmov.16 q0[2], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s10
-; CHECK-NEXT:    vmov.16 q0[3], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s18
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s11
-; CHECK-NEXT:    vmov.16 q0[5], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s19
-; CHECK-NEXT:    vmov.16 q0[6], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s12
-; CHECK-NEXT:    vmov.16 q0[7], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s4
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vmov.16 q2[0], r0
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s13
-; CHECK-NEXT:    vmov.16 q2[1], r1
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s5
-; CHECK-NEXT:    vmov.16 q2[2], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s14
-; CHECK-NEXT:    vmov.16 q2[3], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s6
-; CHECK-NEXT:    vmov.16 q2[4], r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s15
-; CHECK-NEXT:    vmov.16 q2[5], r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q2[6], r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.16 q2[7], r0
-; CHECK-NEXT:    vmov q1, q2
-; CHECK-NEXT:    vpop {d8, d9, d10}
+; CHECK-NEXT:    vcvtb.f16.f32 q2, q2
+; CHECK-NEXT:    vcvtb.f16.f32 q3, q3
+; CHECK-NEXT:    vcvtt.f16.f32 q2, q0
+; CHECK-NEXT:    vcvtt.f16.f32 q3, q1
+; CHECK-NEXT:    vmov q0, q2
+; CHECK-NEXT:    vmov q1, q3
 ; CHECK-NEXT:    bx lr
 entry:
   %out1 = fptrunc <8 x float> %src1 to <8 x half>
@@ -750,31 +468,9 @@
 define arm_aapcs_vfpcc void @store_shuffletrunc_8(<8 x half>* %src, <4 x float> %val1, <4 x float> %val2) {
 ; CHECK-LABEL: store_shuffletrunc_8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s0
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vcvtb.f16.f32 s8, s4
-; CHECK-NEXT:    vmov r2, s8
-; CHECK-NEXT:    vmov.16 q2[0], r1
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s1
-; CHECK-NEXT:    vmov.16 q2[1], r2
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s5
-; CHECK-NEXT:    vmov.16 q2[2], r1
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s2
-; CHECK-NEXT:    vmov.16 q2[3], r1
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s6
-; CHECK-NEXT:    vmov.16 q2[4], r1
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s3
-; CHECK-NEXT:    vmov.16 q2[5], r1
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s7
-; CHECK-NEXT:    vmov.16 q2[6], r1
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov.16 q2[7], r1
-; CHECK-NEXT:    vstrw.32 q2, [r0]
+; CHECK-NEXT:    vcvtb.f16.f32 q0, q0
+; CHECK-NEXT:    vcvtt.f16.f32 q0, q1
+; CHECK-NEXT:    vstrw.32 q0, [r0]
 ; CHECK-NEXT:    bx lr
 entry:
   %strided.vec = shufflevector <4 x float> %val1, <4 x float> %val2, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>
@@ -786,59 +482,12 @@
 define arm_aapcs_vfpcc void @store_shuffletrunc_16(<16 x half>* %src, <8 x float> %val1, <8 x float> %val2) {
 ; CHECK-LABEL: store_shuffletrunc_16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9, d10}
-; CHECK-NEXT:    vpush {d8, d9, d10}
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s4
-; CHECK-NEXT:    vmov r1, s16
-; CHECK-NEXT:    vcvtb.f16.f32 s16, s12
-; CHECK-NEXT:    vmov r2, s16
-; CHECK-NEXT:    vmov.16 q4[0], r1
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s5
-; CHECK-NEXT:    vmov.16 q4[1], r2
-; CHECK-NEXT:    vmov r1, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s13
-; CHECK-NEXT:    vmov.16 q4[2], r1
-; CHECK-NEXT:    vmov r1, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s6
-; CHECK-NEXT:    vmov.16 q4[3], r1
-; CHECK-NEXT:    vmov r1, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s20, s14
-; CHECK-NEXT:    vmov.16 q4[4], r1
-; CHECK-NEXT:    vmov r1, s20
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s7
-; CHECK-NEXT:    vmov.16 q4[5], r1
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s15
-; CHECK-NEXT:    vmov.16 q4[6], r1
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    vmov.16 q4[7], r1
-; CHECK-NEXT:    vstrw.32 q4, [r0, #16]
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s8
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    vcvtb.f16.f32 s4, s0
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s1
-; CHECK-NEXT:    vmov.16 q1[0], r2
-; CHECK-NEXT:    vmov.16 q1[1], r1
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s9
-; CHECK-NEXT:    vmov.16 q1[2], r1
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s2
-; CHECK-NEXT:    vmov.16 q1[3], r1
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s12, s10
-; CHECK-NEXT:    vmov.16 q1[4], r1
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s3
-; CHECK-NEXT:    vmov.16 q1[5], r1
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vcvtb.f16.f32 s0, s11
-; CHECK-NEXT:    vmov.16 q1[6], r1
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov.16 q1[7], r1
-; CHECK-NEXT:    vstrw.32 q1, [r0]
-; CHECK-NEXT:    vpop {d8, d9, d10}
+; CHECK-NEXT:    vcvtb.f16.f32 q1, q1
+; CHECK-NEXT:    vcvtb.f16.f32 q0, q0
+; CHECK-NEXT:    vcvtt.f16.f32 q1, q3
+; CHECK-NEXT:    vcvtt.f16.f32 q0, q2
+; CHECK-NEXT:    vstrw.32 q1, [r0, #16]
+; CHECK-NEXT:    vstrw.32 q0, [r0]
 ; CHECK-NEXT:    bx lr
 entry:
   %strided.vec = shufflevector <8 x float> %val1, <8 x float> %val2, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>