From 05f3219b38dad2806beda523fe83d43ffae3927c Mon Sep 17 00:00:00 2001
From: Jacob Bramley <jacob.bramley@arm.com>
Date: Thu, 5 Aug 2021 13:36:05 +0100
Subject: [PATCH] [AArch64] Lower fpto*i.sat intrinsics for NEON.

Following on from D102353, extend the fpto*i.sat intrinsics to use NEON
fcvt* instructions.

Differential Revision: https://reviews.llvm.org/D108460
---
 llvm/lib/Target/AArch64/AArch64ISelLowering.cpp |  67 ++-
 llvm/lib/Target/AArch64/AArch64ISelLowering.h   |   1 +
 llvm/lib/Target/AArch64/AArch64InstrInfo.td     |  16 +
 llvm/test/CodeGen/AArch64/fptosi-sat-vector.ll  | 702 ++++++++++++------------
 llvm/test/CodeGen/AArch64/fptoui-sat-vector.ll  | 684 ++++++++++++-----------
 5 files changed, 789 insertions(+), 681 deletions(-)

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 338145b..1a6e68d 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -998,6 +998,9 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
     setOperationAction(ISD::UINT_TO_FP, MVT::v1i64, Expand);
     setOperationAction(ISD::FP_ROUND, MVT::v1f64, Expand);
 
+    setOperationAction(ISD::FP_TO_SINT_SAT, MVT::v1i64, Expand);
+    setOperationAction(ISD::FP_TO_UINT_SAT, MVT::v1i64, Expand);
+
     setOperationAction(ISD::MUL, MVT::v1i64, Expand);
 
     // AArch64 doesn't have a direct vector ->f32 conversion instructions for
@@ -1461,6 +1464,8 @@ void AArch64TargetLowering::addTypeForNEON(MVT VT) {
 
   setOperationAction(ISD::FP_TO_SINT, VT, Custom);
   setOperationAction(ISD::FP_TO_UINT, VT, Custom);
+  setOperationAction(ISD::FP_TO_SINT_SAT, VT, Custom);
+  setOperationAction(ISD::FP_TO_UINT_SAT, VT, Custom);
 
   if (!VT.isFloatingPoint())
     setOperationAction(ISD::ABS, VT, Legal);
@@ -3382,31 +3387,77 @@ SDValue AArch64TargetLowering::LowerFP_TO_INT(SDValue Op,
   return SDValue();
 }
 
+SDValue
+AArch64TargetLowering::LowerVectorFP_TO_INT_SAT(SDValue Op,
+                                                SelectionDAG &DAG) const {
+  // AArch64 FP-to-int conversions saturate to the destination element size, so
+  // we can lower common saturating conversions to simple instructions.
+  SDValue SrcVal = Op.getOperand(0);
+  EVT SrcVT = SrcVal.getValueType();
+  EVT DstVT = Op.getValueType();
+  EVT SatVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
+
+  uint64_t SrcElementWidth = SrcVT.getScalarSizeInBits();
+  uint64_t DstElementWidth = DstVT.getScalarSizeInBits();
+  uint64_t SatWidth = SatVT.getScalarSizeInBits();
+  assert(SatWidth <= DstElementWidth &&
+         "Saturation width cannot exceed result width");
+
+  // TODO: Consider lowering to SVE operations, as in LowerVectorFP_TO_INT.
+  // Currently, the `llvm.fpto[su]i.sat.*` instrinsics don't accept scalable
+  // types, so this is hard to reach.
+  if (DstVT.isScalableVector())
+    return SDValue();
+
+  // TODO: Saturate to SatWidth explicitly.
+  if (SatWidth != DstElementWidth)
+    return SDValue();
+
+  EVT SrcElementVT = SrcVT.getVectorElementType();
+
+  // In the absence of FP16 support, promote f16 to f32, like
+  // LowerVectorFP_TO_INT().
+  if (SrcElementVT == MVT::f16 && !Subtarget->hasFullFP16()) {
+    MVT F32VT = MVT::getVectorVT(MVT::f32, SrcVT.getVectorNumElements());
+    return DAG.getNode(Op.getOpcode(), SDLoc(Op), DstVT,
+                       DAG.getNode(ISD::FP_EXTEND, SDLoc(Op), F32VT, SrcVal),
+                       Op.getOperand(1));
+  }
+
+  // Cases that we can emit directly.
+  if ((SrcElementWidth == DstElementWidth) &&
+      (SrcElementVT == MVT::f64 || SrcElementVT == MVT::f32 ||
+       (SrcElementVT == MVT::f16 && Subtarget->hasFullFP16()))) {
+    return Op;
+  }
+
+  // For all other cases, fall back on the expanded form.
+  return SDValue();
+}
+
 SDValue AArch64TargetLowering::LowerFP_TO_INT_SAT(SDValue Op,
                                                   SelectionDAG &DAG) const {
   // AArch64 FP-to-int conversions saturate to the destination register size, so
   // we can lower common saturating conversions to simple instructions.
   SDValue SrcVal = Op.getOperand(0);
-
   EVT SrcVT = SrcVal.getValueType();
-  EVT DstVT = Op.getValueType();
 
+  if (SrcVT.isVector())
+    return LowerVectorFP_TO_INT_SAT(Op, DAG);
+
+  EVT DstVT = Op.getValueType();
   EVT SatVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
   uint64_t SatWidth = SatVT.getScalarSizeInBits();
   uint64_t DstWidth = DstVT.getScalarSizeInBits();
   assert(SatWidth <= DstWidth && "Saturation width cannot exceed result width");
 
-  // TODO: Support lowering of NEON and SVE conversions.
-  if (SrcVT.isVector())
-    return SDValue();
-
   // TODO: Saturate to SatWidth explicitly.
   if (SatWidth != DstWidth)
     return SDValue();
 
-  // In the absence of FP16 support, promote f32 to f16, like LowerFP_TO_INT().
+  // In the absence of FP16 support, promote f16 to f32, like LowerFP_TO_INT().
   if (SrcVT == MVT::f16 && !Subtarget->hasFullFP16())
-    return DAG.getNode(Op.getOpcode(), SDLoc(Op), Op.getValueType(),
+    return DAG.getNode(Op.getOpcode(), SDLoc(Op), DstVT,
                        DAG.getNode(ISD::FP_EXTEND, SDLoc(Op), MVT::f32, SrcVal),
                        Op.getOperand(1));
 
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.h b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
index 0a4c48b..ac805c2 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.h
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
@@ -970,6 +970,7 @@ private:
   SDValue LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerVectorFP_TO_INT(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerVectorFP_TO_INT_SAT(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFP_TO_INT_SAT(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 990ae3e..d5798b0 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -4148,6 +4148,22 @@ defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
 
+// AArch64's FCVT instructions saturate when out of range.
+multiclass SIMDTwoVectorFPToIntSatPats<SDNode to_int_sat, string INST> {
+  def : Pat<(v4i16 (to_int_sat v4f16:$Rn, i16)),
+            (!cast<Instruction>(INST # v4f16) v4f16:$Rn)>;
+  def : Pat<(v8i16 (to_int_sat v8f16:$Rn, i16)),
+            (!cast<Instruction>(INST # v8f16) v8f16:$Rn)>;
+  def : Pat<(v2i32 (to_int_sat v2f32:$Rn, i32)),
+            (!cast<Instruction>(INST # v2f32) v2f32:$Rn)>;
+  def : Pat<(v4i32 (to_int_sat v4f32:$Rn, i32)),
+            (!cast<Instruction>(INST # v4f32) v4f32:$Rn)>;
+  def : Pat<(v2i64 (to_int_sat v2f64:$Rn, i64)),
+            (!cast<Instruction>(INST # v2f64) v2f64:$Rn)>;
+}
+defm : SIMDTwoVectorFPToIntSatPats<fp_to_sint_sat, "FCVTZS">;
+defm : SIMDTwoVectorFPToIntSatPats<fp_to_uint_sat, "FCVTZU">;
+
 def : Pat<(v4i16 (int_aarch64_neon_fcvtzs v4f16:$Rn)), (FCVTZSv4f16 $Rn)>;
 def : Pat<(v8i16 (int_aarch64_neon_fcvtzs v8f16:$Rn)), (FCVTZSv8f16 $Rn)>;
 def : Pat<(v2i32 (int_aarch64_neon_fcvtzs v2f32:$Rn)), (FCVTZSv2f32 $Rn)>;
diff --git a/llvm/test/CodeGen/AArch64/fptosi-sat-vector.ll b/llvm/test/CodeGen/AArch64/fptosi-sat-vector.ll
index 8cdf873..6319d70 100644
--- a/llvm/test/CodeGen/AArch64/fptosi-sat-vector.ll
+++ b/llvm/test/CodeGen/AArch64/fptosi-sat-vector.ll
@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64 < %s | FileCheck %s
+; RUN: llc < %s -mtriple=aarch64 | FileCheck %s --check-prefixes=CHECK,CHECK-CVT
+; RUN: llc < %s -mtriple=aarch64 -mattr=+fullfp16 | FileCheck %s --check-prefixes=CHECK,CHECK-FP16
 
 ;
 ; Float to signed 32-bit -- Vector size variation
@@ -17,13 +18,7 @@ declare <8 x i32> @llvm.fptosi.sat.v8f32.v8i32 (<8 x float>)
 define <1 x i32> @test_signed_v1f32_v1i32(<1 x float> %f) {
 ; CHECK-LABEL: test_signed_v1f32_v1i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    fcvtzs v0.2s, v0.2s
 ; CHECK-NEXT:    ret
     %x = call <1 x i32> @llvm.fptosi.sat.v1f32.v1i32(<1 x float> %f)
     ret <1 x i32> %x
@@ -32,13 +27,7 @@ define <1 x i32> @test_signed_v1f32_v1i32(<1 x float> %f) {
 define <2 x i32> @test_signed_v2f32_v2i32(<2 x float> %f) {
 ; CHECK-LABEL: test_signed_v2f32_v2i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    fcvtzs v0.2s, v0.2s
 ; CHECK-NEXT:    ret
     %x = call <2 x i32> @llvm.fptosi.sat.v2f32.v2i32(<2 x float> %f)
     ret <2 x i32> %x
@@ -47,17 +36,7 @@ define <2 x i32> @test_signed_v2f32_v2i32(<2 x float> %f) {
 define <3 x i32> @test_signed_v3f32_v3i32(<3 x float> %f) {
 ; CHECK-LABEL: test_signed_v3f32_v3i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    mov s2, v0.s[2]
-; CHECK-NEXT:    mov s3, v0.s[3]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    fcvtzs w9, s2
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    mov v0.s[2], w9
-; CHECK-NEXT:    fcvtzs w8, s3
-; CHECK-NEXT:    mov v0.s[3], w8
+; CHECK-NEXT:    fcvtzs v0.4s, v0.4s
 ; CHECK-NEXT:    ret
     %x = call <3 x i32> @llvm.fptosi.sat.v3f32.v3i32(<3 x float> %f)
     ret <3 x i32> %x
@@ -66,17 +45,7 @@ define <3 x i32> @test_signed_v3f32_v3i32(<3 x float> %f) {
 define <4 x i32> @test_signed_v4f32_v4i32(<4 x float> %f) {
 ; CHECK-LABEL: test_signed_v4f32_v4i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    mov s2, v0.s[2]
-; CHECK-NEXT:    mov s3, v0.s[3]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    fcvtzs w9, s2
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    mov v0.s[2], w9
-; CHECK-NEXT:    fcvtzs w8, s3
-; CHECK-NEXT:    mov v0.s[3], w8
+; CHECK-NEXT:    fcvtzs v0.4s, v0.4s
 ; CHECK-NEXT:    ret
     %x = call <4 x i32> @llvm.fptosi.sat.v4f32.v4i32(<4 x float> %f)
     ret <4 x i32> %x
@@ -85,11 +54,21 @@ define <4 x i32> @test_signed_v4f32_v4i32(<4 x float> %f) {
 define <5 x i32> @test_signed_v5f32_v5i32(<5 x float> %f) {
 ; CHECK-LABEL: test_signed_v5f32_v5i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzs w0, s0
-; CHECK-NEXT:    fcvtzs w1, s1
-; CHECK-NEXT:    fcvtzs w2, s2
-; CHECK-NEXT:    fcvtzs w3, s3
-; CHECK-NEXT:    fcvtzs w4, s4
+; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
+; CHECK-NEXT:    // kill: def $s1 killed $s1 def $q1
+; CHECK-NEXT:    // kill: def $s2 killed $s2 def $q2
+; CHECK-NEXT:    // kill: def $s4 killed $s4 def $q4
+; CHECK-NEXT:    // kill: def $s3 killed $s3 def $q3
+; CHECK-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-NEXT:    mov v0.s[3], v3.s[0]
+; CHECK-NEXT:    fcvtzs v4.4s, v4.4s
+; CHECK-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-NEXT:    mov w1, v0.s[1]
+; CHECK-NEXT:    mov w2, v0.s[2]
+; CHECK-NEXT:    mov w3, v0.s[3]
+; CHECK-NEXT:    fmov w0, s0
+; CHECK-NEXT:    fmov w4, s4
 ; CHECK-NEXT:    ret
     %x = call <5 x i32> @llvm.fptosi.sat.v5f32.v5i32(<5 x float> %f)
     ret <5 x i32> %x
@@ -98,15 +77,24 @@ define <5 x i32> @test_signed_v5f32_v5i32(<5 x float> %f) {
 define <6 x i32> @test_signed_v6f32_v6i32(<6 x float> %f) {
 ; CHECK-LABEL: test_signed_v6f32_v6i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzs w8, s4
-; CHECK-NEXT:    fcvtzs w5, s5
-; CHECK-NEXT:    fcvtzs w0, s0
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    mov v0.s[1], w5
-; CHECK-NEXT:    fcvtzs w1, s1
-; CHECK-NEXT:    fcvtzs w2, s2
-; CHECK-NEXT:    fcvtzs w3, s3
-; CHECK-NEXT:    fmov w4, s0
+; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
+; CHECK-NEXT:    // kill: def $s1 killed $s1 def $q1
+; CHECK-NEXT:    // kill: def $s4 killed $s4 def $q4
+; CHECK-NEXT:    // kill: def $s2 killed $s2 def $q2
+; CHECK-NEXT:    // kill: def $s5 killed $s5 def $q5
+; CHECK-NEXT:    // kill: def $s3 killed $s3 def $q3
+; CHECK-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-NEXT:    mov v4.s[1], v5.s[0]
+; CHECK-NEXT:    mov v0.s[3], v3.s[0]
+; CHECK-NEXT:    fcvtzs v1.4s, v4.4s
+; CHECK-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-NEXT:    mov w5, v1.s[1]
+; CHECK-NEXT:    mov w1, v0.s[1]
+; CHECK-NEXT:    mov w2, v0.s[2]
+; CHECK-NEXT:    mov w3, v0.s[3]
+; CHECK-NEXT:    fmov w4, s1
+; CHECK-NEXT:    fmov w0, s0
 ; CHECK-NEXT:    ret
     %x = call <6 x i32> @llvm.fptosi.sat.v6f32.v6i32(<6 x float> %f)
     ret <6 x i32> %x
@@ -115,18 +103,27 @@ define <6 x i32> @test_signed_v6f32_v6i32(<6 x float> %f) {
 define <7 x i32> @test_signed_v7f32_v7i32(<7 x float> %f) {
 ; CHECK-LABEL: test_signed_v7f32_v7i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzs w8, s4
-; CHECK-NEXT:    fcvtzs w9, s5
-; CHECK-NEXT:    fcvtzs w0, s0
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzs w6, s6
-; CHECK-NEXT:    mov v0.s[1], w9
-; CHECK-NEXT:    mov v0.s[2], w6
-; CHECK-NEXT:    fcvtzs w1, s1
-; CHECK-NEXT:    fcvtzs w2, s2
-; CHECK-NEXT:    fcvtzs w3, s3
-; CHECK-NEXT:    mov w5, v0.s[1]
-; CHECK-NEXT:    fmov w4, s0
+; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
+; CHECK-NEXT:    // kill: def $s4 killed $s4 def $q4
+; CHECK-NEXT:    // kill: def $s1 killed $s1 def $q1
+; CHECK-NEXT:    // kill: def $s5 killed $s5 def $q5
+; CHECK-NEXT:    // kill: def $s2 killed $s2 def $q2
+; CHECK-NEXT:    // kill: def $s6 killed $s6 def $q6
+; CHECK-NEXT:    // kill: def $s3 killed $s3 def $q3
+; CHECK-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-NEXT:    mov v4.s[1], v5.s[0]
+; CHECK-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-NEXT:    mov v4.s[2], v6.s[0]
+; CHECK-NEXT:    mov v0.s[3], v3.s[0]
+; CHECK-NEXT:    fcvtzs v1.4s, v4.4s
+; CHECK-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-NEXT:    mov w5, v1.s[1]
+; CHECK-NEXT:    mov w6, v1.s[2]
+; CHECK-NEXT:    mov w1, v0.s[1]
+; CHECK-NEXT:    mov w2, v0.s[2]
+; CHECK-NEXT:    mov w3, v0.s[3]
+; CHECK-NEXT:    fmov w4, s1
+; CHECK-NEXT:    fmov w0, s0
 ; CHECK-NEXT:    ret
     %x = call <7 x i32> @llvm.fptosi.sat.v7f32.v7i32(<7 x float> %f)
     ret <7 x i32> %x
@@ -135,30 +132,8 @@ define <7 x i32> @test_signed_v7f32_v7i32(<7 x float> %f) {
 define <8 x i32> @test_signed_v8f32_v8i32(<8 x float> %f) {
 ; CHECK-LABEL: test_signed_v8f32_v8i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov s2, v0.s[1]
-; CHECK-NEXT:    mov s3, v0.s[2]
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    fcvtzs w9, s1
-; CHECK-NEXT:    fcvtzs w10, s2
-; CHECK-NEXT:    mov s2, v1.s[1]
-; CHECK-NEXT:    fcvtzs w11, s3
-; CHECK-NEXT:    mov s3, v1.s[2]
-; CHECK-NEXT:    fcvtzs w12, s2
-; CHECK-NEXT:    fmov s2, w8
-; CHECK-NEXT:    fcvtzs w8, s3
-; CHECK-NEXT:    fmov s3, w9
-; CHECK-NEXT:    mov v2.s[1], w10
-; CHECK-NEXT:    mov v3.s[1], w12
-; CHECK-NEXT:    mov s0, v0.s[3]
-; CHECK-NEXT:    mov v2.s[2], w11
-; CHECK-NEXT:    mov s1, v1.s[3]
-; CHECK-NEXT:    mov v3.s[2], w8
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    mov v2.s[3], w8
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    mov v3.s[3], w8
-; CHECK-NEXT:    mov v0.16b, v2.16b
-; CHECK-NEXT:    mov v1.16b, v3.16b
+; CHECK-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-NEXT:    fcvtzs v1.4s, v1.4s
 ; CHECK-NEXT:    ret
     %x = call <8 x i32> @llvm.fptosi.sat.v8f32.v8i32(<8 x float> %f)
     ret <8 x i32> %x
@@ -590,202 +565,239 @@ declare <7 x i32> @llvm.fptosi.sat.v7f16.v7i32 (<7 x half>)
 declare <8 x i32> @llvm.fptosi.sat.v8f16.v8i32 (<8 x half>)
 
 define <1 x i32> @test_signed_v1f16_v1i32(<1 x half> %f) {
-; CHECK-LABEL: test_signed_v1f16_v1i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v1f16_v1i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvt s0, h0
+; CHECK-CVT-NEXT:    fcvtzs w8, s0
+; CHECK-CVT-NEXT:    fmov s0, w8
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v1f16_v1i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    fcvtzs w8, h0
+; CHECK-FP16-NEXT:    fmov s0, w8
+; CHECK-FP16-NEXT:    ret
     %x = call <1 x i32> @llvm.fptosi.sat.v1f16.v1i32(<1 x half> %f)
     ret <1 x i32> %x
 }
 
 define <2 x i32> @test_signed_v2f16_v2i32(<2 x half> %f) {
-; CHECK-LABEL: test_signed_v2f16_v2i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov h0, v0.h[1]
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v2f16_v2i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-CVT-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v2f16_v2i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-FP16-NEXT:    fcvtzs w8, h0
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    fmov s0, w8
+; CHECK-FP16-NEXT:    fcvtzs w8, h1
+; CHECK-FP16-NEXT:    mov v0.s[1], w8
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-FP16-NEXT:    ret
     %x = call <2 x i32> @llvm.fptosi.sat.v2f16.v2i32(<2 x half> %f)
     ret <2 x i32> %x
 }
 
 define <3 x i32> @test_signed_v3f16_v3i32(<3 x half> %f) {
-; CHECK-LABEL: test_signed_v3f16_v3i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fmov s1, w8
-; CHECK-NEXT:    fcvtzs w8, s2
-; CHECK-NEXT:    mov h2, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v1.s[1], w8
-; CHECK-NEXT:    fcvtzs w8, s2
-; CHECK-NEXT:    mov v1.s[2], w8
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    mov v1.s[3], w8
-; CHECK-NEXT:    mov v0.16b, v1.16b
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v3f16_v3i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v3f16_v3i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-FP16-NEXT:    fcvtzs w8, h0
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    mov h3, v0.h[3]
+; CHECK-FP16-NEXT:    fmov s0, w8
+; CHECK-FP16-NEXT:    fcvtzs w8, h1
+; CHECK-FP16-NEXT:    fcvtzs w9, h2
+; CHECK-FP16-NEXT:    mov v0.s[1], w8
+; CHECK-FP16-NEXT:    mov v0.s[2], w9
+; CHECK-FP16-NEXT:    fcvtzs w8, h3
+; CHECK-FP16-NEXT:    mov v0.s[3], w8
+; CHECK-FP16-NEXT:    ret
     %x = call <3 x i32> @llvm.fptosi.sat.v3f16.v3i32(<3 x half> %f)
     ret <3 x i32> %x
 }
 
 define <4 x i32> @test_signed_v4f16_v4i32(<4 x half> %f) {
-; CHECK-LABEL: test_signed_v4f16_v4i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fmov s1, w8
-; CHECK-NEXT:    fcvtzs w8, s2
-; CHECK-NEXT:    mov h2, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v1.s[1], w8
-; CHECK-NEXT:    fcvtzs w8, s2
-; CHECK-NEXT:    mov v1.s[2], w8
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    mov v1.s[3], w8
-; CHECK-NEXT:    mov v0.16b, v1.16b
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v4f16_v4i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v4f16_v4i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-FP16-NEXT:    fcvtzs w8, h0
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    mov h3, v0.h[3]
+; CHECK-FP16-NEXT:    fmov s0, w8
+; CHECK-FP16-NEXT:    fcvtzs w8, h1
+; CHECK-FP16-NEXT:    fcvtzs w9, h2
+; CHECK-FP16-NEXT:    mov v0.s[1], w8
+; CHECK-FP16-NEXT:    mov v0.s[2], w9
+; CHECK-FP16-NEXT:    fcvtzs w8, h3
+; CHECK-FP16-NEXT:    mov v0.s[3], w8
+; CHECK-FP16-NEXT:    ret
     %x = call <4 x i32> @llvm.fptosi.sat.v4f16.v4i32(<4 x half> %f)
     ret <4 x i32> %x
 }
 
 define <5 x i32> @test_signed_v5f16_v5i32(<5 x half> %f) {
-; CHECK-LABEL: test_signed_v5f16_v5i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    fcvtzs w0, s1
-; CHECK-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvtzs w4, s1
-; CHECK-NEXT:    mov h1, v0.h[1]
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvtzs w1, s1
-; CHECK-NEXT:    mov h1, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    fcvtzs w2, s1
-; CHECK-NEXT:    fcvtzs w3, s0
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v5f16_v5i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v1.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtl2 v0.4s, v0.8h
+; CHECK-CVT-NEXT:    fcvtzs v1.4s, v1.4s
+; CHECK-CVT-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-CVT-NEXT:    mov w1, v1.s[1]
+; CHECK-CVT-NEXT:    mov w2, v1.s[2]
+; CHECK-CVT-NEXT:    mov w3, v1.s[3]
+; CHECK-CVT-NEXT:    fmov w4, s0
+; CHECK-CVT-NEXT:    fmov w0, s1
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v5f16_v5i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-FP16-NEXT:    fcvtzs w0, h0
+; CHECK-FP16-NEXT:    mov h2, v0.h[1]
+; CHECK-FP16-NEXT:    mov h3, v0.h[2]
+; CHECK-FP16-NEXT:    mov h0, v0.h[3]
+; CHECK-FP16-NEXT:    fcvtzs w4, h1
+; CHECK-FP16-NEXT:    fcvtzs w1, h2
+; CHECK-FP16-NEXT:    fcvtzs w2, h3
+; CHECK-FP16-NEXT:    fcvtzs w3, h0
+; CHECK-FP16-NEXT:    ret
     %x = call <5 x i32> @llvm.fptosi.sat.v5f16.v5i32(<5 x half> %f)
     ret <5 x i32> %x
 }
 
 define <6 x i32> @test_signed_v6f16_v6i32(<6 x half> %f) {
-; CHECK-LABEL: test_signed_v6f16_v6i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    fcvt s2, h0
-; CHECK-NEXT:    fcvtzs w0, s2
-; CHECK-NEXT:    fcvt s2, h1
-; CHECK-NEXT:    fcvtzs w8, s2
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvtzs w1, s2
-; CHECK-NEXT:    mov h2, v0.h[2]
-; CHECK-NEXT:    mov h1, v1.h[1]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvtzs w2, s2
-; CHECK-NEXT:    fmov s2, w8
-; CHECK-NEXT:    fcvtzs w5, s1
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v2.s[1], w5
-; CHECK-NEXT:    fcvtzs w3, s0
-; CHECK-NEXT:    fmov w4, s2
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v6f16_v6i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v1.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtl2 v0.4s, v0.8h
+; CHECK-CVT-NEXT:    fcvtzs v1.4s, v1.4s
+; CHECK-CVT-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-CVT-NEXT:    mov w5, v0.s[1]
+; CHECK-CVT-NEXT:    mov w1, v1.s[1]
+; CHECK-CVT-NEXT:    mov w2, v1.s[2]
+; CHECK-CVT-NEXT:    mov w3, v1.s[3]
+; CHECK-CVT-NEXT:    fmov w4, s0
+; CHECK-CVT-NEXT:    fmov w0, s1
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v6f16_v6i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-FP16-NEXT:    mov h2, v0.h[1]
+; CHECK-FP16-NEXT:    fcvtzs w1, h2
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    fcvtzs w8, h1
+; CHECK-FP16-NEXT:    mov h1, v1.h[1]
+; CHECK-FP16-NEXT:    fcvtzs w2, h2
+; CHECK-FP16-NEXT:    fmov s2, w8
+; CHECK-FP16-NEXT:    fcvtzs w5, h1
+; CHECK-FP16-NEXT:    fcvtzs w0, h0
+; CHECK-FP16-NEXT:    mov h0, v0.h[3]
+; CHECK-FP16-NEXT:    mov v2.s[1], w5
+; CHECK-FP16-NEXT:    fcvtzs w3, h0
+; CHECK-FP16-NEXT:    fmov w4, s2
+; CHECK-FP16-NEXT:    ret
     %x = call <6 x i32> @llvm.fptosi.sat.v6f16.v6i32(<6 x half> %f)
     ret <6 x i32> %x
 }
 
 define <7 x i32> @test_signed_v7f16_v7i32(<7 x half> %f) {
-; CHECK-LABEL: test_signed_v7f16_v7i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    fcvt s2, h0
-; CHECK-NEXT:    mov h3, v0.h[1]
-; CHECK-NEXT:    fcvtzs w0, s2
-; CHECK-NEXT:    fcvt s2, h1
-; CHECK-NEXT:    fcvt s3, h3
-; CHECK-NEXT:    fcvtzs w8, s2
-; CHECK-NEXT:    mov h2, v0.h[2]
-; CHECK-NEXT:    fcvtzs w1, s3
-; CHECK-NEXT:    mov h3, v1.h[1]
-; CHECK-NEXT:    mov h1, v1.h[2]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s3, h3
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvtzs w2, s2
-; CHECK-NEXT:    fmov s2, w8
-; CHECK-NEXT:    fcvtzs w8, s3
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvtzs w6, s1
-; CHECK-NEXT:    mov v2.s[1], w8
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v2.s[2], w6
-; CHECK-NEXT:    fcvtzs w3, s0
-; CHECK-NEXT:    mov w5, v2.s[1]
-; CHECK-NEXT:    fmov w4, s2
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v7f16_v7i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v1.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtl2 v0.4s, v0.8h
+; CHECK-CVT-NEXT:    fcvtzs v1.4s, v1.4s
+; CHECK-CVT-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-CVT-NEXT:    mov w5, v0.s[1]
+; CHECK-CVT-NEXT:    mov w6, v0.s[2]
+; CHECK-CVT-NEXT:    mov w1, v1.s[1]
+; CHECK-CVT-NEXT:    mov w2, v1.s[2]
+; CHECK-CVT-NEXT:    mov w3, v1.s[3]
+; CHECK-CVT-NEXT:    fmov w4, s0
+; CHECK-CVT-NEXT:    fmov w0, s1
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v7f16_v7i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-FP16-NEXT:    mov h2, v0.h[1]
+; CHECK-FP16-NEXT:    mov h3, v0.h[2]
+; CHECK-FP16-NEXT:    fcvtzs w8, h1
+; CHECK-FP16-NEXT:    fcvtzs w1, h2
+; CHECK-FP16-NEXT:    mov h2, v1.h[1]
+; CHECK-FP16-NEXT:    mov h1, v1.h[2]
+; CHECK-FP16-NEXT:    fcvtzs w2, h3
+; CHECK-FP16-NEXT:    fmov s3, w8
+; CHECK-FP16-NEXT:    fcvtzs w8, h2
+; CHECK-FP16-NEXT:    fcvtzs w6, h1
+; CHECK-FP16-NEXT:    mov v3.s[1], w8
+; CHECK-FP16-NEXT:    fcvtzs w0, h0
+; CHECK-FP16-NEXT:    mov h0, v0.h[3]
+; CHECK-FP16-NEXT:    mov v3.s[2], w6
+; CHECK-FP16-NEXT:    fcvtzs w3, h0
+; CHECK-FP16-NEXT:    mov w5, v3.s[1]
+; CHECK-FP16-NEXT:    fmov w4, s3
+; CHECK-FP16-NEXT:    ret
     %x = call <7 x i32> @llvm.fptosi.sat.v7f16.v7i32(<7 x half> %f)
     ret <7 x i32> %x
 }
 
 define <8 x i32> @test_signed_v8f16_v8i32(<8 x half> %f) {
-; CHECK-LABEL: test_signed_v8f16_v8i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ext v3.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    fcvt s1, h3
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvtzs w9, s1
-; CHECK-NEXT:    mov h1, v0.h[2]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvtzs w10, s2
-; CHECK-NEXT:    mov h2, v3.h[1]
-; CHECK-NEXT:    fcvtzs w11, s1
-; CHECK-NEXT:    mov h1, v3.h[2]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvtzs w12, s2
-; CHECK-NEXT:    fmov s2, w8
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    fmov s1, w9
-; CHECK-NEXT:    mov h3, v3.h[3]
-; CHECK-NEXT:    mov v1.s[1], w12
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s3, h3
-; CHECK-NEXT:    mov v2.s[1], w10
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v1.s[2], w8
-; CHECK-NEXT:    fcvtzs w8, s3
-; CHECK-NEXT:    mov v2.s[2], w11
-; CHECK-NEXT:    mov v1.s[3], w8
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    mov v2.s[3], w8
-; CHECK-NEXT:    mov v0.16b, v2.16b
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v8f16_v8i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl2 v1.4s, v0.8h
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtzs v1.4s, v1.4s
+; CHECK-CVT-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v8f16_v8i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    ext v3.16b, v0.16b, v0.16b, #8
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    fcvtzs w9, h3
+; CHECK-FP16-NEXT:    fcvtzs w10, h1
+; CHECK-FP16-NEXT:    mov h1, v3.h[1]
+; CHECK-FP16-NEXT:    fcvtzs w8, h0
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    mov h4, v3.h[2]
+; CHECK-FP16-NEXT:    fcvtzs w12, h1
+; CHECK-FP16-NEXT:    fmov s1, w9
+; CHECK-FP16-NEXT:    fcvtzs w11, h2
+; CHECK-FP16-NEXT:    fmov s2, w8
+; CHECK-FP16-NEXT:    fcvtzs w8, h4
+; CHECK-FP16-NEXT:    mov v1.s[1], w12
+; CHECK-FP16-NEXT:    mov h3, v3.h[3]
+; CHECK-FP16-NEXT:    mov v2.s[1], w10
+; CHECK-FP16-NEXT:    mov h0, v0.h[3]
+; CHECK-FP16-NEXT:    mov v1.s[2], w8
+; CHECK-FP16-NEXT:    fcvtzs w8, h3
+; CHECK-FP16-NEXT:    mov v2.s[2], w11
+; CHECK-FP16-NEXT:    mov v1.s[3], w8
+; CHECK-FP16-NEXT:    fcvtzs w8, h0
+; CHECK-FP16-NEXT:    mov v2.s[3], w8
+; CHECK-FP16-NEXT:    mov v0.16b, v2.16b
+; CHECK-FP16-NEXT:    ret
     %x = call <8 x i32> @llvm.fptosi.sat.v8f16.v8i32(<8 x half> %f)
     ret <8 x i32> %x
 }
@@ -943,13 +955,7 @@ define <2 x i19> @test_signed_v2f32_v2i19(<2 x float> %f) {
 define <2 x i32> @test_signed_v2f32_v2i32_duplicate(<2 x float> %f) {
 ; CHECK-LABEL: test_signed_v2f32_v2i32_duplicate:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    fcvtzs v0.2s, v0.2s
 ; CHECK-NEXT:    ret
     %x = call <2 x i32> @llvm.fptosi.sat.v2f32.v2i32(<2 x float> %f)
     ret <2 x i32> %x
@@ -1324,11 +1330,7 @@ define <2 x i50> @test_signed_v2f64_v2i50(<2 x double> %f) {
 define <2 x i64> @test_signed_v2f64_v2i64(<2 x double> %f) {
 ; CHECK-LABEL: test_signed_v2f64_v2i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzs x8, d0
-; CHECK-NEXT:    mov d1, v0.d[1]
-; CHECK-NEXT:    fmov d0, x8
-; CHECK-NEXT:    fcvtzs x8, d1
-; CHECK-NEXT:    mov v0.d[1], x8
+; CHECK-NEXT:    fcvtzs v0.2d, v0.2d
 ; CHECK-NEXT:    ret
     %x = call <2 x i64> @llvm.fptosi.sat.v2f64.v2i64(<2 x double> %f)
     ret <2 x i64> %x
@@ -1615,47 +1617,48 @@ define <4 x i13> @test_signed_v4f16_v4i13(<4 x half> %f) {
 }
 
 define <4 x i16> @test_signed_v4f16_v4i16(<4 x half> %f) {
-; CHECK-LABEL: test_signed_v4f16_v4i16:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #-956301312
-; CHECK-NEXT:    mov w9, #65024
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    movk w9, #18175, lsl #16
-; CHECK-NEXT:    fcvt s2, h0
-; CHECK-NEXT:    fmov s3, w8
-; CHECK-NEXT:    fmov s4, w9
-; CHECK-NEXT:    fmaxnm s5, s2, s3
-; CHECK-NEXT:    mov h1, v0.h[1]
-; CHECK-NEXT:    fminnm s5, s5, s4
-; CHECK-NEXT:    fcvtzs w8, s5
-; CHECK-NEXT:    mov h5, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvt s6, h0
-; CHECK-NEXT:    fmaxnm s0, s1, s3
-; CHECK-NEXT:    fminnm s0, s0, s4
-; CHECK-NEXT:    fcvt s5, h5
-; CHECK-NEXT:    fcmp s1, s1
-; CHECK-NEXT:    fcvtzs w9, s0
-; CHECK-NEXT:    fmaxnm s0, s5, s3
-; CHECK-NEXT:    csel w9, wzr, w9, vs
-; CHECK-NEXT:    fcmp s2, s2
-; CHECK-NEXT:    fmaxnm s1, s6, s3
-; CHECK-NEXT:    fminnm s3, s0, s4
-; CHECK-NEXT:    csel w8, wzr, w8, vs
-; CHECK-NEXT:    fminnm s1, s1, s4
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzs w8, s3
-; CHECK-NEXT:    fcmp s5, s5
-; CHECK-NEXT:    mov v0.h[1], w9
-; CHECK-NEXT:    fcvtzs w9, s1
-; CHECK-NEXT:    csel w8, wzr, w8, vs
-; CHECK-NEXT:    fcmp s6, s6
-; CHECK-NEXT:    mov v0.h[2], w8
-; CHECK-NEXT:    csel w8, wzr, w9, vs
-; CHECK-NEXT:    mov v0.h[3], w8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v4f16_v4i16:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    mov w8, #-956301312
+; CHECK-CVT-NEXT:    mov w9, #65024
+; CHECK-CVT-NEXT:    movk w9, #18175, lsl #16
+; CHECK-CVT-NEXT:    mov s1, v0.s[1]
+; CHECK-CVT-NEXT:    fmov s2, w8
+; CHECK-CVT-NEXT:    fmov s3, w9
+; CHECK-CVT-NEXT:    fcmp s1, s1
+; CHECK-CVT-NEXT:    fmaxnm s1, s1, s2
+; CHECK-CVT-NEXT:    fmaxnm s4, s0, s2
+; CHECK-CVT-NEXT:    fminnm s1, s1, s3
+; CHECK-CVT-NEXT:    mov s5, v0.s[2]
+; CHECK-CVT-NEXT:    fminnm s4, s4, s3
+; CHECK-CVT-NEXT:    fcvtzs w9, s1
+; CHECK-CVT-NEXT:    fcvtzs w8, s4
+; CHECK-CVT-NEXT:    mov s4, v0.s[3]
+; CHECK-CVT-NEXT:    fmaxnm s1, s5, s2
+; CHECK-CVT-NEXT:    csel w9, wzr, w9, vs
+; CHECK-CVT-NEXT:    fcmp s0, s0
+; CHECK-CVT-NEXT:    fmaxnm s2, s4, s2
+; CHECK-CVT-NEXT:    fminnm s1, s1, s3
+; CHECK-CVT-NEXT:    csel w8, wzr, w8, vs
+; CHECK-CVT-NEXT:    fminnm s2, s2, s3
+; CHECK-CVT-NEXT:    fmov s0, w8
+; CHECK-CVT-NEXT:    fcvtzs w8, s1
+; CHECK-CVT-NEXT:    fcmp s5, s5
+; CHECK-CVT-NEXT:    mov v0.h[1], w9
+; CHECK-CVT-NEXT:    fcvtzs w9, s2
+; CHECK-CVT-NEXT:    csel w8, wzr, w8, vs
+; CHECK-CVT-NEXT:    fcmp s4, s4
+; CHECK-CVT-NEXT:    mov v0.h[2], w8
+; CHECK-CVT-NEXT:    csel w8, wzr, w9, vs
+; CHECK-CVT-NEXT:    mov v0.h[3], w8
+; CHECK-CVT-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v4f16_v4i16:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    fcvtzs v0.4h, v0.4h
+; CHECK-FP16-NEXT:    ret
     %x = call <4 x i16> @llvm.fptosi.sat.v4f16.v4i16(<4 x half> %f)
     ret <4 x i16> %x
 }
@@ -1706,26 +1709,27 @@ define <4 x i19> @test_signed_v4f16_v4i19(<4 x half> %f) {
 }
 
 define <4 x i32> @test_signed_v4f16_v4i32_duplicate(<4 x half> %f) {
-; CHECK-LABEL: test_signed_v4f16_v4i32_duplicate:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvtzs w8, s1
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fmov s1, w8
-; CHECK-NEXT:    fcvtzs w8, s2
-; CHECK-NEXT:    mov h2, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v1.s[1], w8
-; CHECK-NEXT:    fcvtzs w8, s2
-; CHECK-NEXT:    mov v1.s[2], w8
-; CHECK-NEXT:    fcvtzs w8, s0
-; CHECK-NEXT:    mov v1.s[3], w8
-; CHECK-NEXT:    mov v0.16b, v1.16b
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v4f16_v4i32_duplicate:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtzs v0.4s, v0.4s
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v4f16_v4i32_duplicate:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-FP16-NEXT:    fcvtzs w8, h0
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    mov h3, v0.h[3]
+; CHECK-FP16-NEXT:    fmov s0, w8
+; CHECK-FP16-NEXT:    fcvtzs w8, h1
+; CHECK-FP16-NEXT:    fcvtzs w9, h2
+; CHECK-FP16-NEXT:    mov v0.s[1], w8
+; CHECK-FP16-NEXT:    mov v0.s[2], w9
+; CHECK-FP16-NEXT:    fcvtzs w8, h3
+; CHECK-FP16-NEXT:    mov v0.s[3], w8
+; CHECK-FP16-NEXT:    ret
     %x = call <4 x i32> @llvm.fptosi.sat.v4f16.v4i32(<4 x half> %f)
     ret <4 x i32> %x
 }
@@ -1781,26 +1785,42 @@ define <4 x i50> @test_signed_v4f16_v4i50(<4 x half> %f) {
 }
 
 define <4 x i64> @test_signed_v4f16_v4i64(<4 x half> %f) {
-; CHECK-LABEL: test_signed_v4f16_v4i64:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvtzs x8, s1
-; CHECK-NEXT:    fcvt s1, h2
-; CHECK-NEXT:    fmov d2, x8
-; CHECK-NEXT:    fcvtzs x8, s1
-; CHECK-NEXT:    mov h1, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    mov v2.d[1], x8
-; CHECK-NEXT:    fcvtzs x8, s1
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    fmov d1, x8
-; CHECK-NEXT:    fcvtzs x8, s0
-; CHECK-NEXT:    mov v1.d[1], x8
-; CHECK-NEXT:    mov v0.16b, v2.16b
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_signed_v4f16_v4i64:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-CVT-NEXT:    fcvt s1, h0
+; CHECK-CVT-NEXT:    mov h2, v0.h[1]
+; CHECK-CVT-NEXT:    fcvtzs x8, s1
+; CHECK-CVT-NEXT:    fcvt s1, h2
+; CHECK-CVT-NEXT:    fmov d2, x8
+; CHECK-CVT-NEXT:    fcvtzs x8, s1
+; CHECK-CVT-NEXT:    mov h1, v0.h[2]
+; CHECK-CVT-NEXT:    mov h0, v0.h[3]
+; CHECK-CVT-NEXT:    fcvt s1, h1
+; CHECK-CVT-NEXT:    mov v2.d[1], x8
+; CHECK-CVT-NEXT:    fcvtzs x8, s1
+; CHECK-CVT-NEXT:    fcvt s0, h0
+; CHECK-CVT-NEXT:    fmov d1, x8
+; CHECK-CVT-NEXT:    fcvtzs x8, s0
+; CHECK-CVT-NEXT:    mov v1.d[1], x8
+; CHECK-CVT-NEXT:    mov v0.16b, v2.16b
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_signed_v4f16_v4i64:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-FP16-NEXT:    fcvtzs x8, h0
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    mov h3, v0.h[3]
+; CHECK-FP16-NEXT:    fmov d0, x8
+; CHECK-FP16-NEXT:    fcvtzs x8, h1
+; CHECK-FP16-NEXT:    fcvtzs x9, h2
+; CHECK-FP16-NEXT:    mov v0.d[1], x8
+; CHECK-FP16-NEXT:    fmov d1, x9
+; CHECK-FP16-NEXT:    fcvtzs x8, h3
+; CHECK-FP16-NEXT:    mov v1.d[1], x8
+; CHECK-FP16-NEXT:    ret
     %x = call <4 x i64> @llvm.fptosi.sat.v4f16.v4i64(<4 x half> %f)
     ret <4 x i64> %x
 }
diff --git a/llvm/test/CodeGen/AArch64/fptoui-sat-vector.ll b/llvm/test/CodeGen/AArch64/fptoui-sat-vector.ll
index 22bdbbb..11b02dd 100644
--- a/llvm/test/CodeGen/AArch64/fptoui-sat-vector.ll
+++ b/llvm/test/CodeGen/AArch64/fptoui-sat-vector.ll
@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64 < %s | FileCheck %s
+; RUN: llc < %s -mtriple=aarch64 | FileCheck %s --check-prefixes=CHECK,CHECK-CVT
+; RUN: llc < %s -mtriple=aarch64 -mattr=+fullfp16 | FileCheck %s --check-prefixes=CHECK,CHECK-FP16
 
 ;
 ; Float to unsigned 32-bit -- Vector size variation
@@ -17,13 +18,7 @@ declare <8 x i32> @llvm.fptoui.sat.v8f32.v8i32 (<8 x float>)
 define <1 x i32> @test_unsigned_v1f32_v1i32(<1 x float> %f) {
 ; CHECK-LABEL: test_unsigned_v1f32_v1i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    fcvtzu v0.2s, v0.2s
 ; CHECK-NEXT:    ret
     %x = call <1 x i32> @llvm.fptoui.sat.v1f32.v1i32(<1 x float> %f)
     ret <1 x i32> %x
@@ -32,13 +27,7 @@ define <1 x i32> @test_unsigned_v1f32_v1i32(<1 x float> %f) {
 define <2 x i32> @test_unsigned_v2f32_v2i32(<2 x float> %f) {
 ; CHECK-LABEL: test_unsigned_v2f32_v2i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    fcvtzu v0.2s, v0.2s
 ; CHECK-NEXT:    ret
     %x = call <2 x i32> @llvm.fptoui.sat.v2f32.v2i32(<2 x float> %f)
     ret <2 x i32> %x
@@ -47,17 +36,7 @@ define <2 x i32> @test_unsigned_v2f32_v2i32(<2 x float> %f) {
 define <3 x i32> @test_unsigned_v3f32_v3i32(<3 x float> %f) {
 ; CHECK-LABEL: test_unsigned_v3f32_v3i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    mov s2, v0.s[2]
-; CHECK-NEXT:    mov s3, v0.s[3]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    fcvtzu w9, s2
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    mov v0.s[2], w9
-; CHECK-NEXT:    fcvtzu w8, s3
-; CHECK-NEXT:    mov v0.s[3], w8
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
 ; CHECK-NEXT:    ret
     %x = call <3 x i32> @llvm.fptoui.sat.v3f32.v3i32(<3 x float> %f)
     ret <3 x i32> %x
@@ -66,17 +45,7 @@ define <3 x i32> @test_unsigned_v3f32_v3i32(<3 x float> %f) {
 define <4 x i32> @test_unsigned_v4f32_v4i32(<4 x float> %f) {
 ; CHECK-LABEL: test_unsigned_v4f32_v4i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    mov s2, v0.s[2]
-; CHECK-NEXT:    mov s3, v0.s[3]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    fcvtzu w9, s2
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    mov v0.s[2], w9
-; CHECK-NEXT:    fcvtzu w8, s3
-; CHECK-NEXT:    mov v0.s[3], w8
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
 ; CHECK-NEXT:    ret
     %x = call <4 x i32> @llvm.fptoui.sat.v4f32.v4i32(<4 x float> %f)
     ret <4 x i32> %x
@@ -85,11 +54,21 @@ define <4 x i32> @test_unsigned_v4f32_v4i32(<4 x float> %f) {
 define <5 x i32> @test_unsigned_v5f32_v5i32(<5 x float> %f) {
 ; CHECK-LABEL: test_unsigned_v5f32_v5i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzu w0, s0
-; CHECK-NEXT:    fcvtzu w1, s1
-; CHECK-NEXT:    fcvtzu w2, s2
-; CHECK-NEXT:    fcvtzu w3, s3
-; CHECK-NEXT:    fcvtzu w4, s4
+; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
+; CHECK-NEXT:    // kill: def $s1 killed $s1 def $q1
+; CHECK-NEXT:    // kill: def $s2 killed $s2 def $q2
+; CHECK-NEXT:    // kill: def $s4 killed $s4 def $q4
+; CHECK-NEXT:    // kill: def $s3 killed $s3 def $q3
+; CHECK-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-NEXT:    mov v0.s[3], v3.s[0]
+; CHECK-NEXT:    fcvtzu v4.4s, v4.4s
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-NEXT:    mov w1, v0.s[1]
+; CHECK-NEXT:    mov w2, v0.s[2]
+; CHECK-NEXT:    mov w3, v0.s[3]
+; CHECK-NEXT:    fmov w0, s0
+; CHECK-NEXT:    fmov w4, s4
 ; CHECK-NEXT:    ret
     %x = call <5 x i32> @llvm.fptoui.sat.v5f32.v5i32(<5 x float> %f)
     ret <5 x i32> %x
@@ -98,15 +77,24 @@ define <5 x i32> @test_unsigned_v5f32_v5i32(<5 x float> %f) {
 define <6 x i32> @test_unsigned_v6f32_v6i32(<6 x float> %f) {
 ; CHECK-LABEL: test_unsigned_v6f32_v6i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzu w8, s4
-; CHECK-NEXT:    fcvtzu w5, s5
-; CHECK-NEXT:    fcvtzu w0, s0
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    mov v0.s[1], w5
-; CHECK-NEXT:    fcvtzu w1, s1
-; CHECK-NEXT:    fcvtzu w2, s2
-; CHECK-NEXT:    fcvtzu w3, s3
-; CHECK-NEXT:    fmov w4, s0
+; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
+; CHECK-NEXT:    // kill: def $s1 killed $s1 def $q1
+; CHECK-NEXT:    // kill: def $s4 killed $s4 def $q4
+; CHECK-NEXT:    // kill: def $s2 killed $s2 def $q2
+; CHECK-NEXT:    // kill: def $s5 killed $s5 def $q5
+; CHECK-NEXT:    // kill: def $s3 killed $s3 def $q3
+; CHECK-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-NEXT:    mov v4.s[1], v5.s[0]
+; CHECK-NEXT:    mov v0.s[3], v3.s[0]
+; CHECK-NEXT:    fcvtzu v1.4s, v4.4s
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-NEXT:    mov w5, v1.s[1]
+; CHECK-NEXT:    mov w1, v0.s[1]
+; CHECK-NEXT:    mov w2, v0.s[2]
+; CHECK-NEXT:    mov w3, v0.s[3]
+; CHECK-NEXT:    fmov w4, s1
+; CHECK-NEXT:    fmov w0, s0
 ; CHECK-NEXT:    ret
     %x = call <6 x i32> @llvm.fptoui.sat.v6f32.v6i32(<6 x float> %f)
     ret <6 x i32> %x
@@ -115,18 +103,27 @@ define <6 x i32> @test_unsigned_v6f32_v6i32(<6 x float> %f) {
 define <7 x i32> @test_unsigned_v7f32_v7i32(<7 x float> %f) {
 ; CHECK-LABEL: test_unsigned_v7f32_v7i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzu w8, s4
-; CHECK-NEXT:    fcvtzu w9, s5
-; CHECK-NEXT:    fcvtzu w0, s0
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzu w6, s6
-; CHECK-NEXT:    mov v0.s[1], w9
-; CHECK-NEXT:    mov v0.s[2], w6
-; CHECK-NEXT:    fcvtzu w1, s1
-; CHECK-NEXT:    fcvtzu w2, s2
-; CHECK-NEXT:    fcvtzu w3, s3
-; CHECK-NEXT:    mov w5, v0.s[1]
-; CHECK-NEXT:    fmov w4, s0
+; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
+; CHECK-NEXT:    // kill: def $s4 killed $s4 def $q4
+; CHECK-NEXT:    // kill: def $s1 killed $s1 def $q1
+; CHECK-NEXT:    // kill: def $s5 killed $s5 def $q5
+; CHECK-NEXT:    // kill: def $s2 killed $s2 def $q2
+; CHECK-NEXT:    // kill: def $s6 killed $s6 def $q6
+; CHECK-NEXT:    // kill: def $s3 killed $s3 def $q3
+; CHECK-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-NEXT:    mov v4.s[1], v5.s[0]
+; CHECK-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-NEXT:    mov v4.s[2], v6.s[0]
+; CHECK-NEXT:    mov v0.s[3], v3.s[0]
+; CHECK-NEXT:    fcvtzu v1.4s, v4.4s
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-NEXT:    mov w5, v1.s[1]
+; CHECK-NEXT:    mov w6, v1.s[2]
+; CHECK-NEXT:    mov w1, v0.s[1]
+; CHECK-NEXT:    mov w2, v0.s[2]
+; CHECK-NEXT:    mov w3, v0.s[3]
+; CHECK-NEXT:    fmov w4, s1
+; CHECK-NEXT:    fmov w0, s0
 ; CHECK-NEXT:    ret
     %x = call <7 x i32> @llvm.fptoui.sat.v7f32.v7i32(<7 x float> %f)
     ret <7 x i32> %x
@@ -135,30 +132,8 @@ define <7 x i32> @test_unsigned_v7f32_v7i32(<7 x float> %f) {
 define <8 x i32> @test_unsigned_v8f32_v8i32(<8 x float> %f) {
 ; CHECK-LABEL: test_unsigned_v8f32_v8i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov s2, v0.s[1]
-; CHECK-NEXT:    mov s3, v0.s[2]
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    fcvtzu w9, s1
-; CHECK-NEXT:    fcvtzu w10, s2
-; CHECK-NEXT:    mov s2, v1.s[1]
-; CHECK-NEXT:    fcvtzu w11, s3
-; CHECK-NEXT:    mov s3, v1.s[2]
-; CHECK-NEXT:    fcvtzu w12, s2
-; CHECK-NEXT:    fmov s2, w8
-; CHECK-NEXT:    fcvtzu w8, s3
-; CHECK-NEXT:    fmov s3, w9
-; CHECK-NEXT:    mov v2.s[1], w10
-; CHECK-NEXT:    mov v3.s[1], w12
-; CHECK-NEXT:    mov s0, v0.s[3]
-; CHECK-NEXT:    mov v2.s[2], w11
-; CHECK-NEXT:    mov s1, v1.s[3]
-; CHECK-NEXT:    mov v3.s[2], w8
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    mov v2.s[3], w8
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    mov v3.s[3], w8
-; CHECK-NEXT:    mov v0.16b, v2.16b
-; CHECK-NEXT:    mov v1.16b, v3.16b
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-NEXT:    fcvtzu v1.4s, v1.4s
 ; CHECK-NEXT:    ret
     %x = call <8 x i32> @llvm.fptoui.sat.v8f32.v8i32(<8 x float> %f)
     ret <8 x i32> %x
@@ -519,202 +494,239 @@ declare <7 x i32> @llvm.fptoui.sat.v7f16.v7i32 (<7 x half>)
 declare <8 x i32> @llvm.fptoui.sat.v8f16.v8i32 (<8 x half>)
 
 define <1 x i32> @test_unsigned_v1f16_v1i32(<1 x half> %f) {
-; CHECK-LABEL: test_unsigned_v1f16_v1i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v1f16_v1i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvt s0, h0
+; CHECK-CVT-NEXT:    fcvtzu w8, s0
+; CHECK-CVT-NEXT:    fmov s0, w8
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v1f16_v1i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    fcvtzu w8, h0
+; CHECK-FP16-NEXT:    fmov s0, w8
+; CHECK-FP16-NEXT:    ret
     %x = call <1 x i32> @llvm.fptoui.sat.v1f16.v1i32(<1 x half> %f)
     ret <1 x i32> %x
 }
 
 define <2 x i32> @test_unsigned_v2f16_v2i32(<2 x half> %f) {
-; CHECK-LABEL: test_unsigned_v2f16_v2i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov h0, v0.h[1]
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v2f16_v2i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-CVT-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v2f16_v2i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-FP16-NEXT:    fcvtzu w8, h0
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    fmov s0, w8
+; CHECK-FP16-NEXT:    fcvtzu w8, h1
+; CHECK-FP16-NEXT:    mov v0.s[1], w8
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-FP16-NEXT:    ret
     %x = call <2 x i32> @llvm.fptoui.sat.v2f16.v2i32(<2 x half> %f)
     ret <2 x i32> %x
 }
 
 define <3 x i32> @test_unsigned_v3f16_v3i32(<3 x half> %f) {
-; CHECK-LABEL: test_unsigned_v3f16_v3i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fmov s1, w8
-; CHECK-NEXT:    fcvtzu w8, s2
-; CHECK-NEXT:    mov h2, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v1.s[1], w8
-; CHECK-NEXT:    fcvtzu w8, s2
-; CHECK-NEXT:    mov v1.s[2], w8
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    mov v1.s[3], w8
-; CHECK-NEXT:    mov v0.16b, v1.16b
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v3f16_v3i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v3f16_v3i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-FP16-NEXT:    fcvtzu w8, h0
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    mov h3, v0.h[3]
+; CHECK-FP16-NEXT:    fmov s0, w8
+; CHECK-FP16-NEXT:    fcvtzu w8, h1
+; CHECK-FP16-NEXT:    fcvtzu w9, h2
+; CHECK-FP16-NEXT:    mov v0.s[1], w8
+; CHECK-FP16-NEXT:    mov v0.s[2], w9
+; CHECK-FP16-NEXT:    fcvtzu w8, h3
+; CHECK-FP16-NEXT:    mov v0.s[3], w8
+; CHECK-FP16-NEXT:    ret
     %x = call <3 x i32> @llvm.fptoui.sat.v3f16.v3i32(<3 x half> %f)
     ret <3 x i32> %x
 }
 
 define <4 x i32> @test_unsigned_v4f16_v4i32(<4 x half> %f) {
-; CHECK-LABEL: test_unsigned_v4f16_v4i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fmov s1, w8
-; CHECK-NEXT:    fcvtzu w8, s2
-; CHECK-NEXT:    mov h2, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v1.s[1], w8
-; CHECK-NEXT:    fcvtzu w8, s2
-; CHECK-NEXT:    mov v1.s[2], w8
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    mov v1.s[3], w8
-; CHECK-NEXT:    mov v0.16b, v1.16b
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v4f16_v4i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v4f16_v4i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-FP16-NEXT:    fcvtzu w8, h0
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    mov h3, v0.h[3]
+; CHECK-FP16-NEXT:    fmov s0, w8
+; CHECK-FP16-NEXT:    fcvtzu w8, h1
+; CHECK-FP16-NEXT:    fcvtzu w9, h2
+; CHECK-FP16-NEXT:    mov v0.s[1], w8
+; CHECK-FP16-NEXT:    mov v0.s[2], w9
+; CHECK-FP16-NEXT:    fcvtzu w8, h3
+; CHECK-FP16-NEXT:    mov v0.s[3], w8
+; CHECK-FP16-NEXT:    ret
     %x = call <4 x i32> @llvm.fptoui.sat.v4f16.v4i32(<4 x half> %f)
     ret <4 x i32> %x
 }
 
 define <5 x i32> @test_unsigned_v5f16_v5i32(<5 x half> %f) {
-; CHECK-LABEL: test_unsigned_v5f16_v5i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    fcvtzu w0, s1
-; CHECK-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvtzu w4, s1
-; CHECK-NEXT:    mov h1, v0.h[1]
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvtzu w1, s1
-; CHECK-NEXT:    mov h1, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    fcvtzu w2, s1
-; CHECK-NEXT:    fcvtzu w3, s0
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v5f16_v5i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v1.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtl2 v0.4s, v0.8h
+; CHECK-CVT-NEXT:    fcvtzu v1.4s, v1.4s
+; CHECK-CVT-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-CVT-NEXT:    mov w1, v1.s[1]
+; CHECK-CVT-NEXT:    mov w2, v1.s[2]
+; CHECK-CVT-NEXT:    mov w3, v1.s[3]
+; CHECK-CVT-NEXT:    fmov w4, s0
+; CHECK-CVT-NEXT:    fmov w0, s1
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v5f16_v5i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-FP16-NEXT:    fcvtzu w0, h0
+; CHECK-FP16-NEXT:    mov h2, v0.h[1]
+; CHECK-FP16-NEXT:    mov h3, v0.h[2]
+; CHECK-FP16-NEXT:    mov h0, v0.h[3]
+; CHECK-FP16-NEXT:    fcvtzu w4, h1
+; CHECK-FP16-NEXT:    fcvtzu w1, h2
+; CHECK-FP16-NEXT:    fcvtzu w2, h3
+; CHECK-FP16-NEXT:    fcvtzu w3, h0
+; CHECK-FP16-NEXT:    ret
     %x = call <5 x i32> @llvm.fptoui.sat.v5f16.v5i32(<5 x half> %f)
     ret <5 x i32> %x
 }
 
 define <6 x i32> @test_unsigned_v6f16_v6i32(<6 x half> %f) {
-; CHECK-LABEL: test_unsigned_v6f16_v6i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    fcvt s2, h0
-; CHECK-NEXT:    fcvtzu w0, s2
-; CHECK-NEXT:    fcvt s2, h1
-; CHECK-NEXT:    fcvtzu w8, s2
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvtzu w1, s2
-; CHECK-NEXT:    mov h2, v0.h[2]
-; CHECK-NEXT:    mov h1, v1.h[1]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvtzu w2, s2
-; CHECK-NEXT:    fmov s2, w8
-; CHECK-NEXT:    fcvtzu w5, s1
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v2.s[1], w5
-; CHECK-NEXT:    fcvtzu w3, s0
-; CHECK-NEXT:    fmov w4, s2
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v6f16_v6i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v1.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtl2 v0.4s, v0.8h
+; CHECK-CVT-NEXT:    fcvtzu v1.4s, v1.4s
+; CHECK-CVT-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-CVT-NEXT:    mov w5, v0.s[1]
+; CHECK-CVT-NEXT:    mov w1, v1.s[1]
+; CHECK-CVT-NEXT:    mov w2, v1.s[2]
+; CHECK-CVT-NEXT:    mov w3, v1.s[3]
+; CHECK-CVT-NEXT:    fmov w4, s0
+; CHECK-CVT-NEXT:    fmov w0, s1
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v6f16_v6i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-FP16-NEXT:    mov h2, v0.h[1]
+; CHECK-FP16-NEXT:    fcvtzu w1, h2
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    fcvtzu w8, h1
+; CHECK-FP16-NEXT:    mov h1, v1.h[1]
+; CHECK-FP16-NEXT:    fcvtzu w2, h2
+; CHECK-FP16-NEXT:    fmov s2, w8
+; CHECK-FP16-NEXT:    fcvtzu w5, h1
+; CHECK-FP16-NEXT:    fcvtzu w0, h0
+; CHECK-FP16-NEXT:    mov h0, v0.h[3]
+; CHECK-FP16-NEXT:    mov v2.s[1], w5
+; CHECK-FP16-NEXT:    fcvtzu w3, h0
+; CHECK-FP16-NEXT:    fmov w4, s2
+; CHECK-FP16-NEXT:    ret
     %x = call <6 x i32> @llvm.fptoui.sat.v6f16.v6i32(<6 x half> %f)
     ret <6 x i32> %x
 }
 
 define <7 x i32> @test_unsigned_v7f16_v7i32(<7 x half> %f) {
-; CHECK-LABEL: test_unsigned_v7f16_v7i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    fcvt s2, h0
-; CHECK-NEXT:    mov h3, v0.h[1]
-; CHECK-NEXT:    fcvtzu w0, s2
-; CHECK-NEXT:    fcvt s2, h1
-; CHECK-NEXT:    fcvt s3, h3
-; CHECK-NEXT:    fcvtzu w8, s2
-; CHECK-NEXT:    mov h2, v0.h[2]
-; CHECK-NEXT:    fcvtzu w1, s3
-; CHECK-NEXT:    mov h3, v1.h[1]
-; CHECK-NEXT:    mov h1, v1.h[2]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s3, h3
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvtzu w2, s2
-; CHECK-NEXT:    fmov s2, w8
-; CHECK-NEXT:    fcvtzu w8, s3
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvtzu w6, s1
-; CHECK-NEXT:    mov v2.s[1], w8
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v2.s[2], w6
-; CHECK-NEXT:    fcvtzu w3, s0
-; CHECK-NEXT:    mov w5, v2.s[1]
-; CHECK-NEXT:    fmov w4, s2
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v7f16_v7i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v1.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtl2 v0.4s, v0.8h
+; CHECK-CVT-NEXT:    fcvtzu v1.4s, v1.4s
+; CHECK-CVT-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-CVT-NEXT:    mov w5, v0.s[1]
+; CHECK-CVT-NEXT:    mov w6, v0.s[2]
+; CHECK-CVT-NEXT:    mov w1, v1.s[1]
+; CHECK-CVT-NEXT:    mov w2, v1.s[2]
+; CHECK-CVT-NEXT:    mov w3, v1.s[3]
+; CHECK-CVT-NEXT:    fmov w4, s0
+; CHECK-CVT-NEXT:    fmov w0, s1
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v7f16_v7i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-FP16-NEXT:    mov h2, v0.h[1]
+; CHECK-FP16-NEXT:    mov h3, v0.h[2]
+; CHECK-FP16-NEXT:    fcvtzu w8, h1
+; CHECK-FP16-NEXT:    fcvtzu w1, h2
+; CHECK-FP16-NEXT:    mov h2, v1.h[1]
+; CHECK-FP16-NEXT:    mov h1, v1.h[2]
+; CHECK-FP16-NEXT:    fcvtzu w2, h3
+; CHECK-FP16-NEXT:    fmov s3, w8
+; CHECK-FP16-NEXT:    fcvtzu w8, h2
+; CHECK-FP16-NEXT:    fcvtzu w6, h1
+; CHECK-FP16-NEXT:    mov v3.s[1], w8
+; CHECK-FP16-NEXT:    fcvtzu w0, h0
+; CHECK-FP16-NEXT:    mov h0, v0.h[3]
+; CHECK-FP16-NEXT:    mov v3.s[2], w6
+; CHECK-FP16-NEXT:    fcvtzu w3, h0
+; CHECK-FP16-NEXT:    mov w5, v3.s[1]
+; CHECK-FP16-NEXT:    fmov w4, s3
+; CHECK-FP16-NEXT:    ret
     %x = call <7 x i32> @llvm.fptoui.sat.v7f16.v7i32(<7 x half> %f)
     ret <7 x i32> %x
 }
 
 define <8 x i32> @test_unsigned_v8f16_v8i32(<8 x half> %f) {
-; CHECK-LABEL: test_unsigned_v8f16_v8i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ext v3.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    fcvt s1, h3
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvtzu w9, s1
-; CHECK-NEXT:    mov h1, v0.h[2]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvtzu w10, s2
-; CHECK-NEXT:    mov h2, v3.h[1]
-; CHECK-NEXT:    fcvtzu w11, s1
-; CHECK-NEXT:    mov h1, v3.h[2]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    fcvtzu w12, s2
-; CHECK-NEXT:    fmov s2, w8
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    fmov s1, w9
-; CHECK-NEXT:    mov h3, v3.h[3]
-; CHECK-NEXT:    mov v1.s[1], w12
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s3, h3
-; CHECK-NEXT:    mov v2.s[1], w10
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v1.s[2], w8
-; CHECK-NEXT:    fcvtzu w8, s3
-; CHECK-NEXT:    mov v2.s[2], w11
-; CHECK-NEXT:    mov v1.s[3], w8
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    mov v2.s[3], w8
-; CHECK-NEXT:    mov v0.16b, v2.16b
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v8f16_v8i32:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl2 v1.4s, v0.8h
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtzu v1.4s, v1.4s
+; CHECK-CVT-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v8f16_v8i32:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    ext v3.16b, v0.16b, v0.16b, #8
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    fcvtzu w9, h3
+; CHECK-FP16-NEXT:    fcvtzu w10, h1
+; CHECK-FP16-NEXT:    mov h1, v3.h[1]
+; CHECK-FP16-NEXT:    fcvtzu w8, h0
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    mov h4, v3.h[2]
+; CHECK-FP16-NEXT:    fcvtzu w12, h1
+; CHECK-FP16-NEXT:    fmov s1, w9
+; CHECK-FP16-NEXT:    fcvtzu w11, h2
+; CHECK-FP16-NEXT:    fmov s2, w8
+; CHECK-FP16-NEXT:    fcvtzu w8, h4
+; CHECK-FP16-NEXT:    mov v1.s[1], w12
+; CHECK-FP16-NEXT:    mov h3, v3.h[3]
+; CHECK-FP16-NEXT:    mov v2.s[1], w10
+; CHECK-FP16-NEXT:    mov h0, v0.h[3]
+; CHECK-FP16-NEXT:    mov v1.s[2], w8
+; CHECK-FP16-NEXT:    fcvtzu w8, h3
+; CHECK-FP16-NEXT:    mov v2.s[2], w11
+; CHECK-FP16-NEXT:    mov v1.s[3], w8
+; CHECK-FP16-NEXT:    fcvtzu w8, h0
+; CHECK-FP16-NEXT:    mov v2.s[3], w8
+; CHECK-FP16-NEXT:    mov v0.16b, v2.16b
+; CHECK-FP16-NEXT:    ret
     %x = call <8 x i32> @llvm.fptoui.sat.v8f16.v8i32(<8 x half> %f)
     ret <8 x i32> %x
 }
@@ -848,13 +860,7 @@ define <2 x i19> @test_unsigned_v2f32_v2i19(<2 x float> %f) {
 define <2 x i32> @test_unsigned_v2f32_v2i32_duplicate(<2 x float> %f) {
 ; CHECK-LABEL: test_unsigned_v2f32_v2i32_duplicate:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    fcvtzu v0.2s, v0.2s
 ; CHECK-NEXT:    ret
     %x = call <2 x i32> @llvm.fptoui.sat.v2f32.v2i32(<2 x float> %f)
     ret <2 x i32> %x
@@ -1161,11 +1167,7 @@ define <2 x i50> @test_unsigned_v2f64_v2i50(<2 x double> %f) {
 define <2 x i64> @test_unsigned_v2f64_v2i64(<2 x double> %f) {
 ; CHECK-LABEL: test_unsigned_v2f64_v2i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzu x8, d0
-; CHECK-NEXT:    mov d1, v0.d[1]
-; CHECK-NEXT:    fmov d0, x8
-; CHECK-NEXT:    fcvtzu x8, d1
-; CHECK-NEXT:    mov v0.d[1], x8
+; CHECK-NEXT:    fcvtzu v0.2d, v0.2d
 ; CHECK-NEXT:    ret
     %x = call <2 x i64> @llvm.fptoui.sat.v2f64.v2i64(<2 x double> %f)
     ret <2 x i64> %x
@@ -1394,38 +1396,39 @@ define <4 x i13> @test_unsigned_v4f16_v4i13(<4 x half> %f) {
 }
 
 define <4 x i16> @test_unsigned_v4f16_v4i16(<4 x half> %f) {
-; CHECK-LABEL: test_unsigned_v4f16_v4i16:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov w8, #65280
-; CHECK-NEXT:    mov h3, v0.h[1]
-; CHECK-NEXT:    mov h4, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    movi d2, #0000000000000000
-; CHECK-NEXT:    movk w8, #18303, lsl #16
-; CHECK-NEXT:    fcvt s3, h3
-; CHECK-NEXT:    fcvt s4, h4
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    fmaxnm s1, s1, s2
-; CHECK-NEXT:    fmaxnm s3, s3, s2
-; CHECK-NEXT:    fmaxnm s4, s4, s2
-; CHECK-NEXT:    fmaxnm s0, s0, s2
-; CHECK-NEXT:    fmov s2, w8
-; CHECK-NEXT:    fminnm s1, s1, s2
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    fminnm s1, s3, s2
-; CHECK-NEXT:    fminnm s3, s4, s2
-; CHECK-NEXT:    fminnm s2, s0, s2
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    mov v0.h[1], w8
-; CHECK-NEXT:    fcvtzu w8, s3
-; CHECK-NEXT:    mov v0.h[2], w8
-; CHECK-NEXT:    fcvtzu w8, s2
-; CHECK-NEXT:    mov v0.h[3], w8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v4f16_v4i16:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    movi d1, #0000000000000000
+; CHECK-CVT-NEXT:    mov w8, #65280
+; CHECK-CVT-NEXT:    movk w8, #18303, lsl #16
+; CHECK-CVT-NEXT:    fmaxnm s2, s0, s1
+; CHECK-CVT-NEXT:    mov s3, v0.s[1]
+; CHECK-CVT-NEXT:    mov s4, v0.s[2]
+; CHECK-CVT-NEXT:    mov s0, v0.s[3]
+; CHECK-CVT-NEXT:    fmaxnm s3, s3, s1
+; CHECK-CVT-NEXT:    fmaxnm s4, s4, s1
+; CHECK-CVT-NEXT:    fmaxnm s0, s0, s1
+; CHECK-CVT-NEXT:    fmov s1, w8
+; CHECK-CVT-NEXT:    fminnm s2, s2, s1
+; CHECK-CVT-NEXT:    fcvtzu w8, s2
+; CHECK-CVT-NEXT:    fminnm s2, s3, s1
+; CHECK-CVT-NEXT:    fminnm s3, s4, s1
+; CHECK-CVT-NEXT:    fminnm s1, s0, s1
+; CHECK-CVT-NEXT:    fmov s0, w8
+; CHECK-CVT-NEXT:    fcvtzu w8, s2
+; CHECK-CVT-NEXT:    mov v0.h[1], w8
+; CHECK-CVT-NEXT:    fcvtzu w8, s3
+; CHECK-CVT-NEXT:    mov v0.h[2], w8
+; CHECK-CVT-NEXT:    fcvtzu w8, s1
+; CHECK-CVT-NEXT:    mov v0.h[3], w8
+; CHECK-CVT-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v4f16_v4i16:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    fcvtzu v0.4h, v0.4h
+; CHECK-FP16-NEXT:    ret
     %x = call <4 x i16> @llvm.fptoui.sat.v4f16.v4i16(<4 x half> %f)
     ret <4 x i16> %x
 }
@@ -1467,26 +1470,27 @@ define <4 x i19> @test_unsigned_v4f16_v4i19(<4 x half> %f) {
 }
 
 define <4 x i32> @test_unsigned_v4f16_v4i32_duplicate(<4 x half> %f) {
-; CHECK-LABEL: test_unsigned_v4f16_v4i32_duplicate:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvtzu w8, s1
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fmov s1, w8
-; CHECK-NEXT:    fcvtzu w8, s2
-; CHECK-NEXT:    mov h2, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s2, h2
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    mov v1.s[1], w8
-; CHECK-NEXT:    fcvtzu w8, s2
-; CHECK-NEXT:    mov v1.s[2], w8
-; CHECK-NEXT:    fcvtzu w8, s0
-; CHECK-NEXT:    mov v1.s[3], w8
-; CHECK-NEXT:    mov v0.16b, v1.16b
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v4f16_v4i32_duplicate:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v4f16_v4i32_duplicate:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-FP16-NEXT:    fcvtzu w8, h0
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    mov h3, v0.h[3]
+; CHECK-FP16-NEXT:    fmov s0, w8
+; CHECK-FP16-NEXT:    fcvtzu w8, h1
+; CHECK-FP16-NEXT:    fcvtzu w9, h2
+; CHECK-FP16-NEXT:    mov v0.s[1], w8
+; CHECK-FP16-NEXT:    mov v0.s[2], w9
+; CHECK-FP16-NEXT:    fcvtzu w8, h3
+; CHECK-FP16-NEXT:    mov v0.s[3], w8
+; CHECK-FP16-NEXT:    ret
     %x = call <4 x i32> @llvm.fptoui.sat.v4f16.v4i32(<4 x half> %f)
     ret <4 x i32> %x
 }
@@ -1531,26 +1535,42 @@ define <4 x i50> @test_unsigned_v4f16_v4i50(<4 x half> %f) {
 }
 
 define <4 x i64> @test_unsigned_v4f16_v4i64(<4 x half> %f) {
-; CHECK-LABEL: test_unsigned_v4f16_v4i64:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    fcvt s1, h0
-; CHECK-NEXT:    mov h2, v0.h[1]
-; CHECK-NEXT:    fcvtzu x8, s1
-; CHECK-NEXT:    fcvt s1, h2
-; CHECK-NEXT:    fmov d2, x8
-; CHECK-NEXT:    fcvtzu x8, s1
-; CHECK-NEXT:    mov h1, v0.h[2]
-; CHECK-NEXT:    mov h0, v0.h[3]
-; CHECK-NEXT:    fcvt s1, h1
-; CHECK-NEXT:    mov v2.d[1], x8
-; CHECK-NEXT:    fcvtzu x8, s1
-; CHECK-NEXT:    fcvt s0, h0
-; CHECK-NEXT:    fmov d1, x8
-; CHECK-NEXT:    fcvtzu x8, s0
-; CHECK-NEXT:    mov v1.d[1], x8
-; CHECK-NEXT:    mov v0.16b, v2.16b
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: test_unsigned_v4f16_v4i64:
+; CHECK-CVT:       // %bb.0:
+; CHECK-CVT-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-CVT-NEXT:    fcvt s1, h0
+; CHECK-CVT-NEXT:    mov h2, v0.h[1]
+; CHECK-CVT-NEXT:    fcvtzu x8, s1
+; CHECK-CVT-NEXT:    fcvt s1, h2
+; CHECK-CVT-NEXT:    fmov d2, x8
+; CHECK-CVT-NEXT:    fcvtzu x8, s1
+; CHECK-CVT-NEXT:    mov h1, v0.h[2]
+; CHECK-CVT-NEXT:    mov h0, v0.h[3]
+; CHECK-CVT-NEXT:    fcvt s1, h1
+; CHECK-CVT-NEXT:    mov v2.d[1], x8
+; CHECK-CVT-NEXT:    fcvtzu x8, s1
+; CHECK-CVT-NEXT:    fcvt s0, h0
+; CHECK-CVT-NEXT:    fmov d1, x8
+; CHECK-CVT-NEXT:    fcvtzu x8, s0
+; CHECK-CVT-NEXT:    mov v1.d[1], x8
+; CHECK-CVT-NEXT:    mov v0.16b, v2.16b
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: test_unsigned_v4f16_v4i64:
+; CHECK-FP16:       // %bb.0:
+; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-FP16-NEXT:    fcvtzu x8, h0
+; CHECK-FP16-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-NEXT:    mov h3, v0.h[3]
+; CHECK-FP16-NEXT:    fmov d0, x8
+; CHECK-FP16-NEXT:    fcvtzu x8, h1
+; CHECK-FP16-NEXT:    fcvtzu x9, h2
+; CHECK-FP16-NEXT:    mov v0.d[1], x8
+; CHECK-FP16-NEXT:    fmov d1, x9
+; CHECK-FP16-NEXT:    fcvtzu x8, h3
+; CHECK-FP16-NEXT:    mov v1.d[1], x8
+; CHECK-FP16-NEXT:    ret
     %x = call <4 x i64> @llvm.fptoui.sat.v4f16.v4i64(<4 x half> %f)
     ret <4 x i64> %x
 }
-- 
2.7.4