[DAG] Create fptoui.sat from clamped fptosi

[dmgreen]

Summary

As an extension to D111976, this converts clamp fptosi, clamped between 0 and (2^n)-1 to a fptoui.sat. This can greatly help on targets with conversions that naturally saturate, such as Arm.

X86 disables the transform as some of the test cases increases in size. A fptoui.sat necessitates a fp clamp without native support, so there is little use in converting if the instruction is just going to be expanded.

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

4859–4860

Can these be const APInt &? They don't seem to be modified.

llvm/lib/Target/X86/X86ISelLowering.h

1104

Formatting looks wrong here

llvm/test/CodeGen/RISCV/fpclamptosat.ll

187

Something weird is happening here. We should be able to fold the slti and beqz together. I'll investigate.

[craig.topper]

llvm/test/CodeGen/RISCV/fpclamptosat.ll

187

This an artifact of DAGTypeLegalizer::IntegerExpandSetCCOperands creating a select of setccs. So we end up with a select_cc as part of the condition of another select_cc. Both of the selects get expanded to control flow and tail duplication leaves us with the silly looking code.

[dmgreen]

Thanks - updated as per comments.

[SjoerdMeijer]

Looks like a good change to me. Please wait a day or two with committing this in case @craig.topper has more comments.

[RKSimon]

I'm happy to take a look at x86 codegen later on - I'm hoping to look at https://llvm.org/PR52581 soon

[dmgreen]

Thanks - that sounds good. I wasn't sure how the X86 converts saturated but it would be nice if they could use this same code. They were just a bit bigger in places at the moment for the unsigned cases.

This is just a rebase to the latest main, now the other parts are in tree.

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

if (!Overflow)
       return CombineTo(N, DAG.getNode(ISD::MUL, DL, VT, N0, N1),
                        DAG.getConstant(0, DL, CarryVT));
   }
 
   return SDValue();
 }
 
 // Function to calculate whether the Min/Max pair of SDNodes (potentially
-// swapped around) make a signed saturate pattern, clamping to between -2^(BW-1)
-// and 2^(BW-1)-1. Returns the node being clamped and the bitwidth of the clamp
-// in BW. Should work with both SMIN/SMAX nodes and setcc/select combo. The
-// operands are the same as SimplifySelectCC. N0<N1 ? N2 : N3
+// swapped around) make a signed saturate pattern, clamping to between a signed
+// saturate of -2^(BW-1) and 2^(BW-1)-1, or an unsigned saturate of 0 and 2^BW.
+// Returns the node being clamped and the bitwidth of the clamp in BW. Should
+// work with both SMIN/SMAX nodes and setcc/select combo. The operands are the
+// same as SimplifySelectCC. N0<N1 ? N2 : N3.
 static SDValue isSaturatingMinMax(SDValue N0, SDValue N1, SDValue N2,
-                                  SDValue N3, ISD::CondCode CC, unsigned &BW) {
+                                  SDValue N3, ISD::CondCode CC, unsigned &BW,
+                                  bool &Unsigned) {
   auto isSignedMinMax = [&](SDValue N0, SDValue N1, SDValue N2, SDValue N3,
                             ISD::CondCode CC) {
     // The compare and select operand should be the same or the select operands
     // should be truncated versions of the comparison.
     if (N0 != N2 && (N2.getOpcode() != ISD::TRUNCATE || N0 != N2.getOperand(0)))
       return 0;
     // The constants need to be the same or a truncated version of each other.
     ConstantSDNode *N1C = isConstOrConstSplat(N1);
static SDValue isSaturatingMinMax(SDValue N0, SDValue N1, SDValue N2,
   unsigned Opcode1 = isSignedMinMax(N00, N01, N02, N03, N0CC);
   if (!Opcode1 || Opcode0 == Opcode1)
     return SDValue();
 
   ConstantSDNode *MinCOp = isConstOrConstSplat(Opcode0 == ISD::SMIN ? N1 : N01);
   ConstantSDNode *MaxCOp = isConstOrConstSplat(Opcode0 == ISD::SMIN ? N01 : N1);
   if (!MinCOp || !MaxCOp || MinCOp->getValueType(0) != MaxCOp->getValueType(0))
     return SDValue();
 
   const APInt &MinC = MinCOp->getAPIntValue();
   const APInt &MaxC = MaxCOp->getAPIntValue();
   APInt MinCPlus1 = MinC + 1;
-  if (-MaxC != MinCPlus1 || !MinCPlus1.isPowerOf2())
-    return SDValue();
+  if (-MaxC == MinCPlus1 && MinCPlus1.isPowerOf2()) {
     BW = MinCPlus1.exactLogBase2() + 1;
+    Unsigned = false;
+    return N02;
+  }
+
+  if (MaxC == 0 && MinCPlus1.isPowerOf2()) {
+    BW = MinCPlus1.exactLogBase2();
+    Unsigned = true;
     return N02;
   }
 
+  return SDValue();
+}
+
 static SDValue PerformMinMaxFpToSatCombine(SDValue N0, SDValue N1, SDValue N2,
                                            SDValue N3, ISD::CondCode CC,
                                            SelectionDAG &DAG) {
   unsigned BW;
-  SDValue Fp = isSaturatingMinMax(N0, N1, N2, N3, CC, BW);
+  bool Unsigned;
+  SDValue Fp = isSaturatingMinMax(N0, N1, N2, N3, CC, BW, Unsigned);
   if (!Fp || Fp.getOpcode() != ISD::FP_TO_SINT)
     return SDValue();
   EVT FPVT = Fp.getOperand(0).getValueType();
   EVT NewVT = EVT::getIntegerVT(*DAG.getContext(), BW);
   if (FPVT.isVector())
     NewVT = EVT::getVectorVT(*DAG.getContext(), NewVT,
                              FPVT.getVectorElementCount());
-  if (!DAG.getTargetLoweringInfo().shouldConvertFpToSat(
-          ISD::FP_TO_SINT_SAT, Fp.getOperand(0).getValueType(), NewVT))
+  unsigned NewOpc = Unsigned ? ISD::FP_TO_UINT_SAT : ISD::FP_TO_SINT_SAT;
+  if (!DAG.getTargetLoweringInfo().shouldConvertFpToSat(NewOpc, FPVT, NewVT))
     return SDValue();
   SDLoc DL(Fp);
-  SDValue Sat = DAG.getNode(ISD::FP_TO_SINT_SAT, DL, NewVT, Fp.getOperand(0),
+  SDValue Sat = DAG.getNode(NewOpc, DL, NewVT, Fp.getOperand(0),
                             DAG.getValueType(NewVT.getScalarType()));
-  return DAG.getSExtOrTrunc(Sat, DL, N2->getValueType(0));
+  return Unsigned ? DAG.getZExtOrTrunc(Sat, DL, N2->getValueType(0))
+                  : DAG.getSExtOrTrunc(Sat, DL, N2->getValueType(0));
 }
 
 SDValue DAGCombiner::visitIMINMAX(SDNode *N) {
   SDValue N0 = N->getOperand(0);
   SDValue N1 = N->getOperand(1);
   EVT VT = N0.getValueType();
   unsigned Opcode = N->getOpcode();
   SDLoc DL(N);

llvm/lib/Target/X86/X86ISelLowering.h

shouldTransformSignedTruncationCheck(EVT XVT,
       MVT KeptBitsVT = MVT::getIntegerVT(KeptBits);
       return VTIsOk(XVT) && VTIsOk(KeptBitsVT);
     }
 
     bool shouldExpandShift(SelectionDAG &DAG, SDNode *N) const override;
 
     bool shouldSplatInsEltVarIndex(EVT VT) const override;
 
+    bool shouldConvertFpToSat(unsigned Op, EVT FPVT, EVT VT) const override {
+      // Converting to sat variants holds little benefit on X86 as we will just
+      // need to saturate the value back using fp arithmatic.
+      return Op != ISD::FP_TO_UINT_SAT && isOperationLegalOrCustom(Op, VT);
+    }
+
     bool convertSetCCLogicToBitwiseLogic(EVT VT) const override {
       return VT.isScalarInteger();
     }
 
     /// Vector-sized comparisons are fast using PCMPEQ + PMOVMSK or PTEST.
     MVT hasFastEqualityCompare(unsigned NumBits) const override;
 
     /// Return the value type to use for ISD::SETCC.

llvm/test/CodeGen/AArch64/fpclamptosat.ll

entry:
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f64i32(double %x) {
 ; CHECK-LABEL: ustest_f64i32:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fcvtzs x8, d0
-; CHECK-NEXT:    mov w9, #-1
-; CHECK-NEXT:    cmp x8, x9
-; CHECK-NEXT:    csel x8, x8, x9, lt
-; CHECK-NEXT:    cmp x8, #0
-; CHECK-NEXT:    csel w0, w8, wzr, gt
+; CHECK-NEXT:    fcvtzu w0, d0
 ; CHECK-NEXT:    ret
 entry:
   %conv = fptosi double %x to i64
   %0 = icmp slt i64 %conv, 4294967295
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %1 = icmp sgt i64 %spec.store.select, 0
   %spec.store.select7 = select i1 %1, i64 %spec.store.select, i64 0
   %conv6 = trunc i64 %spec.store.select7 to i32
entry:
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f32i32(float %x) {
 ; CHECK-LABEL: ustest_f32i32:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fcvtzs x8, s0
-; CHECK-NEXT:    mov w9, #-1
-; CHECK-NEXT:    cmp x8, x9
-; CHECK-NEXT:    csel x8, x8, x9, lt
-; CHECK-NEXT:    cmp x8, #0
-; CHECK-NEXT:    csel w0, w8, wzr, gt
+; CHECK-NEXT:    fcvtzu w0, s0
 ; CHECK-NEXT:    ret
 entry:
   %conv = fptosi float %x to i64
   %0 = icmp slt i64 %conv, 4294967295
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %1 = icmp sgt i64 %spec.store.select, 0
   %spec.store.select7 = select i1 %1, i64 %spec.store.select, i64 0
   %conv6 = trunc i64 %spec.store.select7 to i32
entry:
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f16i32(half %x) {
 ; CHECK-CVT-LABEL: ustest_f16i32:
 ; CHECK-CVT:       // %bb.0: // %entry
 ; CHECK-CVT-NEXT:    fcvt s0, h0
-; CHECK-CVT-NEXT:    mov w9, #-1
-; CHECK-CVT-NEXT:    fcvtzs x8, s0
-; CHECK-CVT-NEXT:    cmp x8, x9
-; CHECK-CVT-NEXT:    csel x8, x8, x9, lt
-; CHECK-CVT-NEXT:    cmp x8, #0
-; CHECK-CVT-NEXT:    csel w0, w8, wzr, gt
+; CHECK-CVT-NEXT:    fcvtzu w0, s0
 ; CHECK-CVT-NEXT:    ret
 ;
 ; CHECK-FP16-LABEL: ustest_f16i32:
 ; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fcvtzs x8, h0
-; CHECK-FP16-NEXT:    mov w9, #-1
-; CHECK-FP16-NEXT:    cmp x8, x9
-; CHECK-FP16-NEXT:    csel x8, x8, x9, lt
-; CHECK-FP16-NEXT:    cmp x8, #0
-; CHECK-FP16-NEXT:    csel w0, w8, wzr, gt
+; CHECK-FP16-NEXT:    fcvtzu w0, h0
 ; CHECK-FP16-NEXT:    ret
 entry:
   %conv = fptosi half %x to i64
   %0 = icmp slt i64 %conv, 4294967295
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %1 = icmp sgt i64 %spec.store.select, 0
   %spec.store.select7 = select i1 %1, i64 %spec.store.select, i64 0
   %conv6 = trunc i64 %spec.store.select7 to i32
entry:
   %spec.store.select = call i64 @llvm.umin.i64(i64 %conv, i64 4294967295)
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f64i32_mm(double %x) {
 ; CHECK-LABEL: ustest_f64i32_mm:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fcvtzs x8, d0
-; CHECK-NEXT:    mov w9, #-1
-; CHECK-NEXT:    cmp x8, x9
-; CHECK-NEXT:    csel x8, x8, x9, lt
-; CHECK-NEXT:    cmp x8, #0
-; CHECK-NEXT:    csel x0, x8, xzr, gt
-; CHECK-NEXT:    // kill: def $w0 killed $w0 killed $x0
+; CHECK-NEXT:    fcvtzu w0, d0
 ; CHECK-NEXT:    ret
 entry:
   %conv = fptosi double %x to i64
   %spec.store.select = call i64 @llvm.smin.i64(i64 %conv, i64 4294967295)
   %spec.store.select7 = call i64 @llvm.smax.i64(i64 %spec.store.select, i64 0)
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6
 }
entry:
   %spec.store.select = call i64 @llvm.umin.i64(i64 %conv, i64 4294967295)
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f32i32_mm(float %x) {
 ; CHECK-LABEL: ustest_f32i32_mm:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fcvtzs x8, s0
-; CHECK-NEXT:    mov w9, #-1
-; CHECK-NEXT:    cmp x8, x9
-; CHECK-NEXT:    csel x8, x8, x9, lt
-; CHECK-NEXT:    cmp x8, #0
-; CHECK-NEXT:    csel x0, x8, xzr, gt
-; CHECK-NEXT:    // kill: def $w0 killed $w0 killed $x0
+; CHECK-NEXT:    fcvtzu w0, s0
 ; CHECK-NEXT:    ret
 entry:
   %conv = fptosi float %x to i64
   %spec.store.select = call i64 @llvm.smin.i64(i64 %conv, i64 4294967295)
   %spec.store.select7 = call i64 @llvm.smax.i64(i64 %spec.store.select, i64 0)
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6
 }
entry:
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f16i32_mm(half %x) {
 ; CHECK-CVT-LABEL: ustest_f16i32_mm:
 ; CHECK-CVT:       // %bb.0: // %entry
 ; CHECK-CVT-NEXT:    fcvt s0, h0
-; CHECK-CVT-NEXT:    mov w9, #-1
-; CHECK-CVT-NEXT:    fcvtzs x8, s0
-; CHECK-CVT-NEXT:    cmp x8, x9
-; CHECK-CVT-NEXT:    csel x8, x8, x9, lt
-; CHECK-CVT-NEXT:    cmp x8, #0
-; CHECK-CVT-NEXT:    csel x0, x8, xzr, gt
-; CHECK-CVT-NEXT:    // kill: def $w0 killed $w0 killed $x0
+; CHECK-CVT-NEXT:    fcvtzu w0, s0
 ; CHECK-CVT-NEXT:    ret
 ;
 ; CHECK-FP16-LABEL: ustest_f16i32_mm:
 ; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fcvtzs x8, h0
-; CHECK-FP16-NEXT:    mov w9, #-1
-; CHECK-FP16-NEXT:    cmp x8, x9
-; CHECK-FP16-NEXT:    csel x8, x8, x9, lt
-; CHECK-FP16-NEXT:    cmp x8, #0
-; CHECK-FP16-NEXT:    csel x0, x8, xzr, gt
-; CHECK-FP16-NEXT:    // kill: def $w0 killed $w0 killed $x0
+; CHECK-FP16-NEXT:    fcvtzu w0, h0
 ; CHECK-FP16-NEXT:    ret
 entry:
   %conv = fptosi half %x to i64
   %spec.store.select = call i64 @llvm.smin.i64(i64 %conv, i64 4294967295)
   %spec.store.select7 = call i64 @llvm.smax.i64(i64 %spec.store.select, i64 0)
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6
 }

llvm/test/CodeGen/AArch64/fpclamptosat_vec.ll

entry:
   %spec.store.select = select <2 x i1> %0, <2 x i64> %conv, <2 x i64> <i64 4294967295, i64 4294967295>
   %conv6 = trunc <2 x i64> %spec.store.select to <2 x i32>
   ret <2 x i32> %conv6
 }
 
 define <2 x i32> @ustest_f64i32(<2 x double> %x) {
 ; CHECK-LABEL: ustest_f64i32:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    movi v1.2d, #0x000000ffffffff
-; CHECK-NEXT:    fcvtzs v0.2d, v0.2d
-; CHECK-NEXT:    cmgt v2.2d, v1.2d, v0.2d
-; CHECK-NEXT:    bif v0.16b, v1.16b, v2.16b
-; CHECK-NEXT:    cmgt v1.2d, v0.2d, #0
-; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
-; CHECK-NEXT:    xtn v0.2s, v0.2d
+; CHECK-NEXT:    mov d1, v0.d[1]
+; CHECK-NEXT:    fcvtzu w8, d0
+; CHECK-NEXT:    fmov s0, w8
+; CHECK-NEXT:    fcvtzu w8, d1
+; CHECK-NEXT:    mov v0.s[1], w8
+; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
 ; CHECK-NEXT:    ret
 entry:
   %conv = fptosi <2 x double> %x to <2 x i64>
   %0 = icmp slt <2 x i64> %conv, <i64 4294967295, i64 4294967295>
   %spec.store.select = select <2 x i1> %0, <2 x i64> %conv, <2 x i64> <i64 4294967295, i64 4294967295>
   %1 = icmp sgt <2 x i64> %spec.store.select, zeroinitializer
   %spec.store.select7 = select <2 x i1> %1, <2 x i64> %spec.store.select, <2 x i64> zeroinitializer
   %conv6 = trunc <2 x i64> %spec.store.select7 to <2 x i32>
entry:
   %spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
   ret <4 x i32> %conv6
 }
 
 define <4 x i32> @ustest_f32i32(<4 x float> %x) {
 ; CHECK-LABEL: ustest_f32i32:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fcvtl v2.2d, v0.2s
-; CHECK-NEXT:    fcvtl2 v0.2d, v0.4s
-; CHECK-NEXT:    movi v1.2d, #0x000000ffffffff
-; CHECK-NEXT:    fcvtzs v2.2d, v2.2d
-; CHECK-NEXT:    fcvtzs v0.2d, v0.2d
-; CHECK-NEXT:    cmgt v3.2d, v1.2d, v2.2d
-; CHECK-NEXT:    cmgt v4.2d, v1.2d, v0.2d
-; CHECK-NEXT:    bif v2.16b, v1.16b, v3.16b
-; CHECK-NEXT:    bif v0.16b, v1.16b, v4.16b
-; CHECK-NEXT:    cmgt v1.2d, v2.2d, #0
-; CHECK-NEXT:    cmgt v3.2d, v0.2d, #0
-; CHECK-NEXT:    and v1.16b, v2.16b, v1.16b
-; CHECK-NEXT:    and v2.16b, v0.16b, v3.16b
-; CHECK-NEXT:    xtn v0.2s, v1.2d
-; CHECK-NEXT:    xtn2 v0.4s, v2.2d
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
 ; CHECK-NEXT:    ret
 entry:
   %conv = fptosi <4 x float> %x to <4 x i64>
   %0 = icmp slt <4 x i64> %conv, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %1 = icmp sgt <4 x i64> %spec.store.select, zeroinitializer
   %spec.store.select7 = select <4 x i1> %1, <4 x i64> %spec.store.select, <4 x i64> zeroinitializer
   %conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
entry:
   %conv = fptoui <4 x half> %x to <4 x i64>
   %0 = icmp ult <4 x i64> %conv, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
   ret <4 x i32> %conv6
 }
 
 define <4 x i32> @ustest_f16i32(<4 x half> %x) {
-; CHECK-CVT-LABEL: ustest_f16i32:
-; CHECK-CVT:       // %bb.0: // %entry
-; CHECK-CVT-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-CVT-NEXT:    mov h2, v0.h[2]
-; CHECK-CVT-NEXT:    mov h3, v0.h[1]
-; CHECK-CVT-NEXT:    fcvt s4, h0
-; CHECK-CVT-NEXT:    mov h0, v0.h[3]
-; CHECK-CVT-NEXT:    movi v1.2d, #0x000000ffffffff
-; CHECK-CVT-NEXT:    fcvt s2, h2
-; CHECK-CVT-NEXT:    fcvt s3, h3
-; CHECK-CVT-NEXT:    fcvtzs x8, s4
-; CHECK-CVT-NEXT:    fcvt s0, h0
-; CHECK-CVT-NEXT:    fcvtzs x9, s2
-; CHECK-CVT-NEXT:    fmov d2, x8
-; CHECK-CVT-NEXT:    fcvtzs x8, s3
-; CHECK-CVT-NEXT:    fmov d3, x9
-; CHECK-CVT-NEXT:    fcvtzs x9, s0
-; CHECK-CVT-NEXT:    mov v2.d[1], x8
-; CHECK-CVT-NEXT:    mov v3.d[1], x9
-; CHECK-CVT-NEXT:    cmgt v0.2d, v1.2d, v2.2d
-; CHECK-CVT-NEXT:    cmgt v4.2d, v1.2d, v3.2d
-; CHECK-CVT-NEXT:    bsl v0.16b, v2.16b, v1.16b
-; CHECK-CVT-NEXT:    bit v1.16b, v3.16b, v4.16b
-; CHECK-CVT-NEXT:    cmgt v2.2d, v0.2d, #0
-; CHECK-CVT-NEXT:    cmgt v3.2d, v1.2d, #0
-; CHECK-CVT-NEXT:    and v0.16b, v0.16b, v2.16b
-; CHECK-CVT-NEXT:    and v1.16b, v1.16b, v3.16b
-; CHECK-CVT-NEXT:    xtn v0.2s, v0.2d
-; CHECK-CVT-NEXT:    xtn2 v0.4s, v1.2d
-; CHECK-CVT-NEXT:    ret
-;
-; CHECK-FP16-LABEL: ustest_f16i32:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-FP16-NEXT:    mov h2, v0.h[2]
-; CHECK-FP16-NEXT:    mov h3, v0.h[1]
-; CHECK-FP16-NEXT:    fcvtzs x8, h0
-; CHECK-FP16-NEXT:    mov h0, v0.h[3]
-; CHECK-FP16-NEXT:    movi v1.2d, #0x000000ffffffff
-; CHECK-FP16-NEXT:    fcvtzs x9, h2
-; CHECK-FP16-NEXT:    fmov d2, x8
-; CHECK-FP16-NEXT:    fcvtzs x8, h3
-; CHECK-FP16-NEXT:    fmov d3, x9
-; CHECK-FP16-NEXT:    fcvtzs x9, h0
-; CHECK-FP16-NEXT:    mov v2.d[1], x8
-; CHECK-FP16-NEXT:    mov v3.d[1], x9
-; CHECK-FP16-NEXT:    cmgt v0.2d, v1.2d, v2.2d
-; CHECK-FP16-NEXT:    cmgt v4.2d, v1.2d, v3.2d
-; CHECK-FP16-NEXT:    bsl v0.16b, v2.16b, v1.16b
-; CHECK-FP16-NEXT:    bit v1.16b, v3.16b, v4.16b
-; CHECK-FP16-NEXT:    cmgt v2.2d, v0.2d, #0
-; CHECK-FP16-NEXT:    cmgt v3.2d, v1.2d, #0
-; CHECK-FP16-NEXT:    and v0.16b, v0.16b, v2.16b
-; CHECK-FP16-NEXT:    and v1.16b, v1.16b, v3.16b
-; CHECK-FP16-NEXT:    xtn v0.2s, v0.2d
-; CHECK-FP16-NEXT:    xtn2 v0.4s, v1.2d
-; CHECK-FP16-NEXT:    ret
+; CHECK-LABEL: ustest_f16i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-NEXT:    ret
 entry:
   %conv = fptosi <4 x half> %x to <4 x i64>
   %0 = icmp slt <4 x i64> %conv, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %1 = icmp sgt <4 x i64> %spec.store.select, zeroinitializer
   %spec.store.select7 = select <4 x i1> %1, <4 x i64> %spec.store.select, <4 x i64> zeroinitializer
   %conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
   ret <4 x i32> %conv6
entry:
   %spec.store.select = select <4 x i1> %0, <4 x i32> %conv, <4 x i32> <i32 65535, i32 65535, i32 65535, i32 65535>
   %conv6 = trunc <4 x i32> %spec.store.select to <4 x i16>
   ret <4 x i16> %conv6
 }
 
 define <4 x i16> @ustest_f32i16(<4 x float> %x) {
 ; CHECK-LABEL: ustest_f32i16:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    movi v1.2d, #0x00ffff0000ffff
-; CHECK-NEXT:    fcvtzs v0.4s, v0.4s
-; CHECK-NEXT:    smin v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    movi v1.2d, #0000000000000000
-; CHECK-NEXT:    smax v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    xtn v0.4h, v0.4s
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-NEXT:    uqxtn v0.4h, v0.4s
 ; CHECK-NEXT:    ret
 entry:
   %conv = fptosi <4 x float> %x to <4 x i32>
   %0 = icmp slt <4 x i32> %conv, <i32 65535, i32 65535, i32 65535, i32 65535>
   %spec.store.select = select <4 x i1> %0, <4 x i32> %conv, <4 x i32> <i32 65535, i32 65535, i32 65535, i32 65535>
   %1 = icmp sgt <4 x i32> %spec.store.select, zeroinitializer
   %spec.store.select7 = select <4 x i1> %1, <4 x i32> %spec.store.select, <4 x i32> zeroinitializer
   %conv6 = trunc <4 x i32> %spec.store.select7 to <4 x i16>
entry:
   %conv = fptoui <8 x half> %x to <8 x i32>
   %0 = icmp ult <8 x i32> %conv, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
   %spec.store.select = select <8 x i1> %0, <8 x i32> %conv, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
   %conv6 = trunc <8 x i32> %spec.store.select to <8 x i16>
   ret <8 x i16> %conv6
 }
 
 define <8 x i16> @ustest_f16i16(<8 x half> %x) {
-; CHECK-LABEL: ustest_f16i16:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fcvtl v2.4s, v0.4h
-; CHECK-NEXT:    fcvtl2 v0.4s, v0.8h
-; CHECK-NEXT:    movi v1.2d, #0x00ffff0000ffff
-; CHECK-NEXT:    movi v3.2d, #0000000000000000
-; CHECK-NEXT:    fcvtzs v2.4s, v2.4s
-; CHECK-NEXT:    fcvtzs v0.4s, v0.4s
-; CHECK-NEXT:    smin v2.4s, v2.4s, v1.4s
-; CHECK-NEXT:    smin v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    smax v1.4s, v2.4s, v3.4s
-; CHECK-NEXT:    smax v2.4s, v0.4s, v3.4s
-; CHECK-NEXT:    xtn v0.4h, v1.4s
-; CHECK-NEXT:    xtn2 v0.8h, v2.4s
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: ustest_f16i16:
+; CHECK-CVT:       // %bb.0: // %entry
+; CHECK-CVT-NEXT:    fcvtl2 v2.4s, v0.8h
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    movi v1.2d, #0x00ffff0000ffff
+; CHECK-CVT-NEXT:    fcvtzu v2.4s, v2.4s
+; CHECK-CVT-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-CVT-NEXT:    umin v1.4s, v2.4s, v1.4s
+; CHECK-CVT-NEXT:    uqxtn v0.4h, v0.4s
+; CHECK-CVT-NEXT:    xtn2 v0.8h, v1.4s
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: ustest_f16i16:
+; CHECK-FP16:       // %bb.0: // %entry
+; CHECK-FP16-NEXT:    fcvtzu v0.8h, v0.8h
+; CHECK-FP16-NEXT:    ret
 entry:
   %conv = fptosi <8 x half> %x to <8 x i32>
   %0 = icmp slt <8 x i32> %conv, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
   %spec.store.select = select <8 x i1> %0, <8 x i32> %conv, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
   %1 = icmp sgt <8 x i32> %spec.store.select, zeroinitializer
   %spec.store.select7 = select <8 x i1> %1, <8 x i32> %spec.store.select, <8 x i32> zeroinitializer
   %conv6 = trunc <8 x i32> %spec.store.select7 to <8 x i16>
   ret <8 x i16> %conv6
entry:
   %spec.store.select = call <2 x i64> @llvm.umin.v2i64(<2 x i64> %conv, <2 x i64> <i64 4294967295, i64 4294967295>)
   %conv6 = trunc <2 x i64> %spec.store.select to <2 x i32>
   ret <2 x i32> %conv6
 }
 
 define <2 x i32> @ustest_f64i32_mm(<2 x double> %x) {
 ; CHECK-LABEL: ustest_f64i32_mm:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    movi v1.2d, #0x000000ffffffff
-; CHECK-NEXT:    fcvtzs v0.2d, v0.2d
-; CHECK-NEXT:    cmgt v2.2d, v1.2d, v0.2d
-; CHECK-NEXT:    bif v0.16b, v1.16b, v2.16b
-; CHECK-NEXT:    cmgt v1.2d, v0.2d, #0
-; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
-; CHECK-NEXT:    xtn v0.2s, v0.2d
+; CHECK-NEXT:    mov d1, v0.d[1]
+; CHECK-NEXT:    fcvtzu w8, d0
+; CHECK-NEXT:    fmov s0, w8
+; CHECK-NEXT:    fcvtzu w8, d1
+; CHECK-NEXT:    mov v0.s[1], w8
+; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
 ; CHECK-NEXT:    ret
 entry:
   %conv = fptosi <2 x double> %x to <2 x i64>
   %spec.store.select = call <2 x i64> @llvm.smin.v2i64(<2 x i64> %conv, <2 x i64> <i64 4294967295, i64 4294967295>)
   %spec.store.select7 = call <2 x i64> @llvm.smax.v2i64(<2 x i64> %spec.store.select, <2 x i64> zeroinitializer)
   %conv6 = trunc <2 x i64> %spec.store.select7 to <2 x i32>
   ret <2 x i32> %conv6
 }
entry:
   %spec.store.select = call <4 x i64> @llvm.umin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
   %conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
   ret <4 x i32> %conv6
 }
 
 define <4 x i32> @ustest_f32i32_mm(<4 x float> %x) {
 ; CHECK-LABEL: ustest_f32i32_mm:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fcvtl v2.2d, v0.2s
-; CHECK-NEXT:    fcvtl2 v0.2d, v0.4s
-; CHECK-NEXT:    movi v1.2d, #0x000000ffffffff
-; CHECK-NEXT:    fcvtzs v2.2d, v2.2d
-; CHECK-NEXT:    fcvtzs v0.2d, v0.2d
-; CHECK-NEXT:    cmgt v3.2d, v1.2d, v2.2d
-; CHECK-NEXT:    cmgt v4.2d, v1.2d, v0.2d
-; CHECK-NEXT:    bif v2.16b, v1.16b, v3.16b
-; CHECK-NEXT:    bif v0.16b, v1.16b, v4.16b
-; CHECK-NEXT:    cmgt v1.2d, v2.2d, #0
-; CHECK-NEXT:    cmgt v3.2d, v0.2d, #0
-; CHECK-NEXT:    and v1.16b, v2.16b, v1.16b
-; CHECK-NEXT:    and v2.16b, v0.16b, v3.16b
-; CHECK-NEXT:    xtn v0.2s, v1.2d
-; CHECK-NEXT:    xtn2 v0.4s, v2.2d
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
 ; CHECK-NEXT:    ret
 entry:
   %conv = fptosi <4 x float> %x to <4 x i64>
   %spec.store.select = call <4 x i64> @llvm.smin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
   %spec.store.select7 = call <4 x i64> @llvm.smax.v4i64(<4 x i64> %spec.store.select, <4 x i64> zeroinitializer)
   %conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
   ret <4 x i32> %conv6
 }

 entry:
   %conv = fptoui <4 x half> %x to <4 x i64>
   %spec.store.select = call <4 x i64> @llvm.umin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
   %conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
   ret <4 x i32> %conv6
 }
 
 define <4 x i32> @ustest_f16i32_mm(<4 x half> %x) {
-; CHECK-CVT-LABEL: ustest_f16i32_mm:
-; CHECK-CVT:       // %bb.0: // %entry
-; CHECK-CVT-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-CVT-NEXT:    mov h2, v0.h[2]
-; CHECK-CVT-NEXT:    mov h3, v0.h[1]
-; CHECK-CVT-NEXT:    fcvt s4, h0
-; CHECK-CVT-NEXT:    mov h0, v0.h[3]
-; CHECK-CVT-NEXT:    movi v1.2d, #0x000000ffffffff
-; CHECK-CVT-NEXT:    fcvt s2, h2
-; CHECK-CVT-NEXT:    fcvt s3, h3
-; CHECK-CVT-NEXT:    fcvtzs x8, s4
-; CHECK-CVT-NEXT:    fcvt s0, h0
-; CHECK-CVT-NEXT:    fcvtzs x9, s2
-; CHECK-CVT-NEXT:    fmov d2, x8
-; CHECK-CVT-NEXT:    fcvtzs x8, s3
-; CHECK-CVT-NEXT:    fmov d3, x9
-; CHECK-CVT-NEXT:    fcvtzs x9, s0
-; CHECK-CVT-NEXT:    mov v2.d[1], x8
-; CHECK-CVT-NEXT:    mov v3.d[1], x9
-; CHECK-CVT-NEXT:    cmgt v0.2d, v1.2d, v2.2d
-; CHECK-CVT-NEXT:    cmgt v4.2d, v1.2d, v3.2d
-; CHECK-CVT-NEXT:    bsl v0.16b, v2.16b, v1.16b
-; CHECK-CVT-NEXT:    bit v1.16b, v3.16b, v4.16b
-; CHECK-CVT-NEXT:    cmgt v2.2d, v0.2d, #0
-; CHECK-CVT-NEXT:    cmgt v3.2d, v1.2d, #0
-; CHECK-CVT-NEXT:    and v0.16b, v0.16b, v2.16b
-; CHECK-CVT-NEXT:    and v1.16b, v1.16b, v3.16b
-; CHECK-CVT-NEXT:    xtn v0.2s, v0.2d
-; CHECK-CVT-NEXT:    xtn2 v0.4s, v1.2d
-; CHECK-CVT-NEXT:    ret
-;
-; CHECK-FP16-LABEL: ustest_f16i32_mm:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-FP16-NEXT:    mov h2, v0.h[2]
-; CHECK-FP16-NEXT:    mov h3, v0.h[1]
-; CHECK-FP16-NEXT:    fcvtzs x8, h0
-; CHECK-FP16-NEXT:    mov h0, v0.h[3]
-; CHECK-FP16-NEXT:    movi v1.2d, #0x000000ffffffff
-; CHECK-FP16-NEXT:    fcvtzs x9, h2
-; CHECK-FP16-NEXT:    fmov d2, x8
-; CHECK-FP16-NEXT:    fcvtzs x8, h3
-; CHECK-FP16-NEXT:    fmov d3, x9
-; CHECK-FP16-NEXT:    fcvtzs x9, h0
-; CHECK-FP16-NEXT:    mov v2.d[1], x8
-; CHECK-FP16-NEXT:    mov v3.d[1], x9
-; CHECK-FP16-NEXT:    cmgt v0.2d, v1.2d, v2.2d
-; CHECK-FP16-NEXT:    cmgt v4.2d, v1.2d, v3.2d
-; CHECK-FP16-NEXT:    bsl v0.16b, v2.16b, v1.16b
-; CHECK-FP16-NEXT:    bit v1.16b, v3.16b, v4.16b
-; CHECK-FP16-NEXT:    cmgt v2.2d, v0.2d, #0
-; CHECK-FP16-NEXT:    cmgt v3.2d, v1.2d, #0
-; CHECK-FP16-NEXT:    and v0.16b, v0.16b, v2.16b
-; CHECK-FP16-NEXT:    and v1.16b, v1.16b, v3.16b
-; CHECK-FP16-NEXT:    xtn v0.2s, v0.2d
-; CHECK-FP16-NEXT:    xtn2 v0.4s, v1.2d
-; CHECK-FP16-NEXT:    ret
+; CHECK-LABEL: ustest_f16i32_mm:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-NEXT:    ret
 entry:
   %conv = fptosi <4 x half> %x to <4 x i64>
   %spec.store.select = call <4 x i64> @llvm.smin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
   %spec.store.select7 = call <4 x i64> @llvm.smax.v4i64(<4 x i64> %spec.store.select, <4 x i64> zeroinitializer)
   %conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
   ret <4 x i32> %conv6
 }
 
entry:
   %spec.store.select = call <4 x i32> @llvm.umin.v4i32(<4 x i32> %conv, <4 x i32> <i32 65535, i32 65535, i32 65535, i32 65535>)
   %conv6 = trunc <4 x i32> %spec.store.select to <4 x i16>
   ret <4 x i16> %conv6
 }
 
 define <4 x i16> @ustest_f32i16_mm(<4 x float> %x) {
 ; CHECK-LABEL: ustest_f32i16_mm:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    movi v1.2d, #0x00ffff0000ffff
-; CHECK-NEXT:    fcvtzs v0.4s, v0.4s
-; CHECK-NEXT:    smin v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    movi v1.2d, #0000000000000000
-; CHECK-NEXT:    smax v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    xtn v0.4h, v0.4s
+; CHECK-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-NEXT:    uqxtn v0.4h, v0.4s
 ; CHECK-NEXT:    ret
 entry:
   %conv = fptosi <4 x float> %x to <4 x i32>
   %spec.store.select = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %conv, <4 x i32> <i32 65535, i32 65535, i32 65535, i32 65535>)
   %spec.store.select7 = call <4 x i32> @llvm.smax.v4i32(<4 x i32> %spec.store.select, <4 x i32> zeroinitializer)
   %conv6 = trunc <4 x i32> %spec.store.select7 to <4 x i16>
   ret <4 x i16> %conv6
 }

 entry:
   %conv = fptoui <8 x half> %x to <8 x i32>
   %spec.store.select = call <8 x i32> @llvm.umin.v8i32(<8 x i32> %conv, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>)
   %conv6 = trunc <8 x i32> %spec.store.select to <8 x i16>
   ret <8 x i16> %conv6
 }
 
 define <8 x i16> @ustest_f16i16_mm(<8 x half> %x) {
-; CHECK-LABEL: ustest_f16i16_mm:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fcvtl v2.4s, v0.4h
-; CHECK-NEXT:    fcvtl2 v0.4s, v0.8h
-; CHECK-NEXT:    movi v1.2d, #0x00ffff0000ffff
-; CHECK-NEXT:    movi v3.2d, #0000000000000000
-; CHECK-NEXT:    fcvtzs v2.4s, v2.4s
-; CHECK-NEXT:    fcvtzs v0.4s, v0.4s
-; CHECK-NEXT:    smin v2.4s, v2.4s, v1.4s
-; CHECK-NEXT:    smin v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    smax v1.4s, v2.4s, v3.4s
-; CHECK-NEXT:    smax v2.4s, v0.4s, v3.4s
-; CHECK-NEXT:    xtn v0.4h, v1.4s
-; CHECK-NEXT:    xtn2 v0.8h, v2.4s
-; CHECK-NEXT:    ret
+; CHECK-CVT-LABEL: ustest_f16i16_mm:
+; CHECK-CVT:       // %bb.0: // %entry
+; CHECK-CVT-NEXT:    fcvtl2 v2.4s, v0.8h
+; CHECK-CVT-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-CVT-NEXT:    movi v1.2d, #0x00ffff0000ffff
+; CHECK-CVT-NEXT:    fcvtzu v2.4s, v2.4s
+; CHECK-CVT-NEXT:    fcvtzu v0.4s, v0.4s
+; CHECK-CVT-NEXT:    umin v1.4s, v2.4s, v1.4s
+; CHECK-CVT-NEXT:    uqxtn v0.4h, v0.4s
+; CHECK-CVT-NEXT:    xtn2 v0.8h, v1.4s
+; CHECK-CVT-NEXT:    ret
+;
+; CHECK-FP16-LABEL: ustest_f16i16_mm:
+; CHECK-FP16:       // %bb.0: // %entry
+; CHECK-FP16-NEXT:    fcvtzu v0.8h, v0.8h
+; CHECK-FP16-NEXT:    ret
 entry:
   %conv = fptosi <8 x half> %x to <8 x i32>
   %spec.store.select = call <8 x i32> @llvm.smin.v8i32(<8 x i32> %conv, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>)
   %spec.store.select7 = call <8 x i32> @llvm.smax.v8i32(<8 x i32> %spec.store.select, <8 x i32> zeroinitializer)
   %conv6 = trunc <8 x i32> %spec.store.select7 to <8 x i16>
   ret <8 x i16> %conv6
 }
 

llvm/test/CodeGen/ARM/fpclamptosat.ll

 ; VFP2-NEXT:    movlt r2, #1
 ; VFP2-NEXT:    cmp r2, #0
 ; VFP2-NEXT:    it eq
 ; VFP2-NEXT:    moveq r0, r2
 ; VFP2-NEXT:    pop {r7, pc}
 ;
 ; FULL-LABEL: ustest_f64i32:
 ; FULL:       @ %bb.0: @ %entry
-; FULL-NEXT:    .save {r7, lr}
-; FULL-NEXT:    push {r7, lr}
-; FULL-NEXT:    vmov r0, r1, d0
-; FULL-NEXT:    bl __aeabi_d2lz
-; FULL-NEXT:    subs.w r2, r0, #-1
-; FULL-NEXT:    sbcs r2, r1, #0
-; FULL-NEXT:    cset r2, lt
-; FULL-NEXT:    cmp r2, #0
-; FULL-NEXT:    it eq
-; FULL-NEXT:    moveq.w r0, #-1
-; FULL-NEXT:    csel r1, r1, r2, ne
-; FULL-NEXT:    rsbs r3, r0, #0
-; FULL-NEXT:    mov.w r2, #0
-; FULL-NEXT:    sbcs.w r1, r2, r1
-; FULL-NEXT:    cset r1, lt
-; FULL-NEXT:    cmp r1, #0
-; FULL-NEXT:    csel r0, r0, r1, ne
-; FULL-NEXT:    pop {r7, pc}
+; FULL-NEXT:    vcvt.u32.f64 s0, d0
+; FULL-NEXT:    vmov r0, s0
+; FULL-NEXT:    bx lr
 entry:
   %conv = fptosi double %x to i64
   %0 = icmp slt i64 %conv, 4294967295
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %1 = icmp sgt i64 %spec.store.select, 0
   %spec.store.select7 = select i1 %1, i64 %spec.store.select, i64 0
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6

 ; SOFT-NEXT:  .LBB5_9: @ %entry
 ; SOFT-NEXT:    mov r2, r3
 ; SOFT-NEXT:    cmp r2, #0
 ; SOFT-NEXT:    bne .LBB5_6
 ; SOFT-NEXT:  .LBB5_10: @ %entry
 ; SOFT-NEXT:    mov r0, r2
 ; SOFT-NEXT:    pop {r4, pc}
 ;
-; VFP2-LABEL: ustest_f32i32:
-; VFP2:       @ %bb.0: @ %entry
-; VFP2-NEXT:    .save {r7, lr}
-; VFP2-NEXT:    push {r7, lr}
-; VFP2-NEXT:    vmov r0, s0
-; VFP2-NEXT:    bl __aeabi_f2lz
-; VFP2-NEXT:    subs.w r3, r0, #-1
-; VFP2-NEXT:    mov.w r2, #0
-; VFP2-NEXT:    sbcs r3, r1, #0
-; VFP2-NEXT:    mov.w r3, #0
-; VFP2-NEXT:    it lt
-; VFP2-NEXT:    movlt r3, #1
-; VFP2-NEXT:    cmp r3, #0
-; VFP2-NEXT:    ite ne
-; VFP2-NEXT:    movne r3, r1
-; VFP2-NEXT:    moveq.w r0, #-1
-; VFP2-NEXT:    rsbs r1, r0, #0
-; VFP2-NEXT:    sbcs.w r1, r2, r3
-; VFP2-NEXT:    it lt
-; VFP2-NEXT:    movlt r2, #1
-; VFP2-NEXT:    cmp r2, #0
-; VFP2-NEXT:    it eq
-; VFP2-NEXT:    moveq r0, r2
-; VFP2-NEXT:    pop {r7, pc}
-;
-; FULL-LABEL: ustest_f32i32:
-; FULL:       @ %bb.0: @ %entry
-; FULL-NEXT:    .save {r7, lr}
-; FULL-NEXT:    push {r7, lr}
-; FULL-NEXT:    vmov r0, s0
-; FULL-NEXT:    bl __aeabi_f2lz
-; FULL-NEXT:    subs.w r2, r0, #-1
-; FULL-NEXT:    sbcs r2, r1, #0
-; FULL-NEXT:    cset r2, lt
-; FULL-NEXT:    cmp r2, #0
-; FULL-NEXT:    it eq
-; FULL-NEXT:    moveq.w r0, #-1
-; FULL-NEXT:    csel r1, r1, r2, ne
-; FULL-NEXT:    rsbs r3, r0, #0
-; FULL-NEXT:    mov.w r2, #0
-; FULL-NEXT:    sbcs.w r1, r2, r1
-; FULL-NEXT:    cset r1, lt
-; FULL-NEXT:    cmp r1, #0
-; FULL-NEXT:    csel r0, r0, r1, ne
-; FULL-NEXT:    pop {r7, pc}
+; VFP-LABEL: ustest_f32i32:
+; VFP:       @ %bb.0: @ %entry
+; VFP-NEXT:    vcvt.u32.f32 s0, s0
+; VFP-NEXT:    vmov r0, s0
+; VFP-NEXT:    bx lr
 entry:
   %conv = fptosi float %x to i64
   %0 = icmp slt i64 %conv, 4294967295
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %1 = icmp sgt i64 %spec.store.select, 0
   %spec.store.select7 = select i1 %1, i64 %spec.store.select, i64 0
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6

 ; SOFT-NEXT:    pop {r4, pc}
 ;
 ; VFP2-LABEL: ustest_f16i32:
 ; VFP2:       @ %bb.0: @ %entry
 ; VFP2-NEXT:    .save {r7, lr}
 ; VFP2-NEXT:    push {r7, lr}
 ; VFP2-NEXT:    vmov r0, s0
 ; VFP2-NEXT:    bl __aeabi_h2f
-; VFP2-NEXT:    bl __aeabi_f2lz
-; VFP2-NEXT:    subs.w r3, r0, #-1
-; VFP2-NEXT:    mov.w r2, #0
-; VFP2-NEXT:    sbcs r3, r1, #0
-; VFP2-NEXT:    mov.w r3, #0
-; VFP2-NEXT:    it lt
-; VFP2-NEXT:    movlt r3, #1
-; VFP2-NEXT:    cmp r3, #0
-; VFP2-NEXT:    ite ne
-; VFP2-NEXT:    movne r3, r1
-; VFP2-NEXT:    moveq.w r0, #-1
-; VFP2-NEXT:    rsbs r1, r0, #0
-; VFP2-NEXT:    sbcs.w r1, r2, r3
-; VFP2-NEXT:    it lt
-; VFP2-NEXT:    movlt r2, #1
-; VFP2-NEXT:    cmp r2, #0
-; VFP2-NEXT:    it eq
-; VFP2-NEXT:    moveq r0, r2
+; VFP2-NEXT:    vmov s0, r0
+; VFP2-NEXT:    vcvt.u32.f32 s0, s0
+; VFP2-NEXT:    vmov r0, s0
 ; VFP2-NEXT:    pop {r7, pc}
 ;
 ; FULL-LABEL: ustest_f16i32:
 ; FULL:       @ %bb.0: @ %entry
-; FULL-NEXT:    .save {r7, lr}
-; FULL-NEXT:    push {r7, lr}
-; FULL-NEXT:    vmov.f16 r0, s0
-; FULL-NEXT:    vmov s0, r0
-; FULL-NEXT:    bl __fixhfdi
-; FULL-NEXT:    subs.w r2, r0, #-1
-; FULL-NEXT:    sbcs r2, r1, #0
-; FULL-NEXT:    cset r2, lt
-; FULL-NEXT:    cmp r2, #0
-; FULL-NEXT:    it eq
-; FULL-NEXT:    moveq.w r0, #-1
-; FULL-NEXT:    csel r1, r1, r2, ne
-; FULL-NEXT:    rsbs r3, r0, #0
-; FULL-NEXT:    mov.w r2, #0
-; FULL-NEXT:    sbcs.w r1, r2, r1
-; FULL-NEXT:    cset r1, lt
-; FULL-NEXT:    cmp r1, #0
-; FULL-NEXT:    csel r0, r0, r1, ne
-; FULL-NEXT:    pop {r7, pc}
+; FULL-NEXT:    vcvt.u32.f16 s0, s0
+; FULL-NEXT:    vmov r0, s0
+; FULL-NEXT:    bx lr
 entry:
   %conv = fptosi half %x to i64
   %0 = icmp slt i64 %conv, 4294967295
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %1 = icmp sgt i64 %spec.store.select, 0
   %spec.store.select7 = select i1 %1, i64 %spec.store.select, i64 0
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6

 ; VFP2-NEXT:    it gt
 ; VFP2-NEXT:    movgt r2, r0
 ; VFP2-NEXT:    it ne
 ; VFP2-NEXT:    movne r0, r2
 ; VFP2-NEXT:    pop {r7, pc}
 ;
 ; FULL-LABEL: ustest_f64i32_mm:
 ; FULL:       @ %bb.0: @ %entry
-; FULL-NEXT:    .save {r7, lr}
-; FULL-NEXT:    push {r7, lr}
-; FULL-NEXT:    vmov r0, r1, d0
-; FULL-NEXT:    bl __aeabi_d2lz
-; FULL-NEXT:    mov r2, r0
-; FULL-NEXT:    cmp r1, #0
-; FULL-NEXT:    it pl
-; FULL-NEXT:    movpl.w r2, #-1
-; FULL-NEXT:    csel r0, r0, r2, eq
-; FULL-NEXT:    mov.w r2, #0
-; FULL-NEXT:    csel r1, r1, r2, mi
-; FULL-NEXT:    cmp r1, #0
-; FULL-NEXT:    csel r1, r0, r2, gt
-; FULL-NEXT:    csel r0, r0, r1, eq
-; FULL-NEXT:    pop {r7, pc}
+; FULL-NEXT:    vcvt.u32.f64 s0, d0
+; FULL-NEXT:    vmov r0, s0
+; FULL-NEXT:    bx lr
 entry:
   %conv = fptosi double %x to i64
   %spec.store.select = call i64 @llvm.smin.i64(i64 %conv, i64 4294967295)
   %spec.store.select7 = call i64 @llvm.smax.i64(i64 %spec.store.select, i64 0)
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6
 }
 

 ; SOFT-NEXT:  .LBB32_9: @ %entry
 ; SOFT-NEXT:    mov r1, r2
 ; SOFT-NEXT:    cmp r3, #0
 ; SOFT-NEXT:    beq .LBB32_8
 ; SOFT-NEXT:  .LBB32_10: @ %entry
 ; SOFT-NEXT:    mov r0, r1
 ; SOFT-NEXT:    pop {r4, pc}
 ;
-; VFP2-LABEL: ustest_f32i32_mm:
-; VFP2:       @ %bb.0: @ %entry
-; VFP2-NEXT:    .save {r7, lr}
-; VFP2-NEXT:    push {r7, lr}
-; VFP2-NEXT:    vmov r0, s0
-; VFP2-NEXT:    bl __aeabi_f2lz
-; VFP2-NEXT:    mov r2, r0
-; VFP2-NEXT:    cmp r1, #0
-; VFP2-NEXT:    it pl
-; VFP2-NEXT:    movpl.w r2, #-1
-; VFP2-NEXT:    it ne
-; VFP2-NEXT:    movne r0, r2
-; VFP2-NEXT:    mov.w r2, #0
-; VFP2-NEXT:    it pl
-; VFP2-NEXT:    movpl r1, r2
-; VFP2-NEXT:    cmp r1, #0
-; VFP2-NEXT:    it gt
-; VFP2-NEXT:    movgt r2, r0
-; VFP2-NEXT:    it ne
-; VFP2-NEXT:    movne r0, r2
-; VFP2-NEXT:    pop {r7, pc}
-;
-; FULL-LABEL: ustest_f32i32_mm:
-; FULL:       @ %bb.0: @ %entry
-; FULL-NEXT:    .save {r7, lr}
-; FULL-NEXT:    push {r7, lr}
-; FULL-NEXT:    vmov r0, s0
-; FULL-NEXT:    bl __aeabi_f2lz
-; FULL-NEXT:    mov r2, r0
-; FULL-NEXT:    cmp r1, #0
-; FULL-NEXT:    it pl
-; FULL-NEXT:    movpl.w r2, #-1
-; FULL-NEXT:    csel r0, r0, r2, eq
-; FULL-NEXT:    mov.w r2, #0
-; FULL-NEXT:    csel r1, r1, r2, mi
-; FULL-NEXT:    cmp r1, #0
-; FULL-NEXT:    csel r1, r0, r2, gt
-; FULL-NEXT:    csel r0, r0, r1, eq
-; FULL-NEXT:    pop {r7, pc}
+; VFP-LABEL: ustest_f32i32_mm:
+; VFP:       @ %bb.0: @ %entry
+; VFP-NEXT:    vcvt.u32.f32 s0, s0
+; VFP-NEXT:    vmov r0, s0
+; VFP-NEXT:    bx lr
 entry:
   %conv = fptosi float %x to i64
   %spec.store.select = call i64 @llvm.smin.i64(i64 %conv, i64 4294967295)
   %spec.store.select7 = call i64 @llvm.smax.i64(i64 %spec.store.select, i64 0)
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6
 }
 

 ; SOFT-NEXT:    pop {r4, pc}
 ;
 ; VFP2-LABEL: ustest_f16i32_mm:
 ; VFP2:       @ %bb.0: @ %entry
 ; VFP2-NEXT:    .save {r7, lr}
 ; VFP2-NEXT:    push {r7, lr}
 ; VFP2-NEXT:    vmov r0, s0
 ; VFP2-NEXT:    bl __aeabi_h2f
-; VFP2-NEXT:    bl __aeabi_f2lz
-; VFP2-NEXT:    mov r2, r0
-; VFP2-NEXT:    cmp r1, #0
-; VFP2-NEXT:    it pl
-; VFP2-NEXT:    movpl.w r2, #-1
-; VFP2-NEXT:    it ne
-; VFP2-NEXT:    movne r0, r2
-; VFP2-NEXT:    mov.w r2, #0
-; VFP2-NEXT:    it pl
-; VFP2-NEXT:    movpl r1, r2
-; VFP2-NEXT:    cmp r1, #0
-; VFP2-NEXT:    it gt
-; VFP2-NEXT:    movgt r2, r0
-; VFP2-NEXT:    it ne
-; VFP2-NEXT:    movne r0, r2
+; VFP2-NEXT:    vmov s0, r0
+; VFP2-NEXT:    vcvt.u32.f32 s0, s0
+; VFP2-NEXT:    vmov r0, s0
 ; VFP2-NEXT:    pop {r7, pc}
 ;
 ; FULL-LABEL: ustest_f16i32_mm:
 ; FULL:       @ %bb.0: @ %entry
-; FULL-NEXT:    .save {r7, lr}
-; FULL-NEXT:    push {r7, lr}
-; FULL-NEXT:    vmov.f16 r0, s0
-; FULL-NEXT:    vmov s0, r0
-; FULL-NEXT:    bl __fixhfdi
-; FULL-NEXT:    mov r2, r0
-; FULL-NEXT:    cmp r1, #0
-; FULL-NEXT:    it pl
-; FULL-NEXT:    movpl.w r2, #-1
-; FULL-NEXT:    csel r0, r0, r2, eq
-; FULL-NEXT:    mov.w r2, #0
-; FULL-NEXT:    csel r1, r1, r2, mi
-; FULL-NEXT:    cmp r1, #0
-; FULL-NEXT:    csel r1, r0, r2, gt
-; FULL-NEXT:    csel r0, r0, r1, eq
-; FULL-NEXT:    pop {r7, pc}
+; FULL-NEXT:    vcvt.u32.f16 s0, s0
+; FULL-NEXT:    vmov r0, s0
+; FULL-NEXT:    bx lr
 entry:
   %conv = fptosi half %x to i64
   %spec.store.select = call i64 @llvm.smin.i64(i64 %conv, i64 4294967295)
   %spec.store.select7 = call i64 @llvm.smax.i64(i64 %spec.store.select, i64 0)
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6
 }
 

llvm/test/CodeGen/RISCV/fpclamptosat.ll

entry:
   %conv = fptoui double %x to i64
   %0 = icmp ult i64 %conv, 4294967295
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f64i32(double %x) {
-; RV32-LABEL: ustest_f64i32:
-; RV32:       # %bb.0: # %entry
-; RV32-NEXT:    addi sp, sp, -16
-; RV32-NEXT:    .cfi_def_cfa_offset 16
-; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32-NEXT:    .cfi_offset ra, -4
-; RV32-NEXT:    call __fixdfdi@plt
-; RV32-NEXT:    beqz a1, .LBB2_2
-; RV32-NEXT:  # %bb.1: # %entry
-; RV32-NEXT:    slti a2, a1, 0
-; RV32-NEXT:    beqz a2, .LBB2_3
-; RV32-NEXT:    j .LBB2_4
-; RV32-NEXT:  .LBB2_2:
-; RV32-NEXT:    addi a2, a0, 1
-; RV32-NEXT:    snez a2, a2
-; RV32-NEXT:    bnez a2, .LBB2_4
-; RV32-NEXT:  .LBB2_3: # %entry
-; RV32-NEXT:    li a1, 0
-; RV32-NEXT:    li a0, -1
-; RV32-NEXT:  .LBB2_4: # %entry
-; RV32-NEXT:    beqz a1, .LBB2_6
-; RV32-NEXT:  # %bb.5: # %entry
-; RV32-NEXT:    sgtz a1, a1
-; RV32-NEXT:    beqz a1, .LBB2_7
-; RV32-NEXT:    j .LBB2_8
-; RV32-NEXT:  .LBB2_6:
-; RV32-NEXT:    snez a1, a0
-; RV32-NEXT:    bnez a1, .LBB2_8
-; RV32-NEXT:  .LBB2_7: # %entry
-; RV32-NEXT:    li a0, 0
-; RV32-NEXT:  .LBB2_8: # %entry
-; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32-NEXT:    addi sp, sp, 16
-; RV32-NEXT:    ret
+; RV32IF-LABEL: ustest_f64i32:
+; RV32IF:       # %bb.0: # %entry
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    .cfi_def_cfa_offset 16
+; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IF-NEXT:    .cfi_offset ra, -4
+; RV32IF-NEXT:    call __fixdfdi@plt
+; RV32IF-NEXT:    beqz a1, .LBB2_2
+; RV32IF-NEXT:  # %bb.1: # %entry
+; RV32IF-NEXT:    slti a2, a1, 0
+; RV32IF-NEXT:    beqz a2, .LBB2_3
+; RV32IF-NEXT:    j .LBB2_4
+; RV32IF-NEXT:  .LBB2_2:
+; RV32IF-NEXT:    addi a2, a0, 1
+; RV32IF-NEXT:    snez a2, a2
+; RV32IF-NEXT:    bnez a2, .LBB2_4
+; RV32IF-NEXT:  .LBB2_3: # %entry
+; RV32IF-NEXT:    li a1, 0
+; RV32IF-NEXT:    li a0, -1
+; RV32IF-NEXT:  .LBB2_4: # %entry
+; RV32IF-NEXT:    beqz a1, .LBB2_6
+; RV32IF-NEXT:  # %bb.5: # %entry
+; RV32IF-NEXT:    sgtz a1, a1
+; RV32IF-NEXT:    beqz a1, .LBB2_7
+; RV32IF-NEXT:    j .LBB2_8
+; RV32IF-NEXT:  .LBB2_6:
+; RV32IF-NEXT:    snez a1, a0
+; RV32IF-NEXT:    bnez a1, .LBB2_8
+; RV32IF-NEXT:  .LBB2_7: # %entry
+; RV32IF-NEXT:    li a0, 0
+; RV32IF-NEXT:  .LBB2_8: # %entry
+; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: ustest_f64i32:
 ; RV64IF:       # %bb.0: # %entry
 ; RV64IF-NEXT:    addi sp, sp, -16
 ; RV64IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV64IF-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IF-NEXT:    .cfi_offset ra, -8
 ; RV64IF-NEXT:    call __fixdfdi@plt
 ; RV64IF-NEXT:    li a1, -1
 ; RV64IF-NEXT:    srli a1, a1, 32
 ; RV64IF-NEXT:    blt a0, a1, .LBB2_2
 ; RV64IF-NEXT:  # %bb.1: # %entry
 ; RV64IF-NEXT:    mv a0, a1
 ; RV64IF-NEXT:  .LBB2_2: # %entry
 ; RV64IF-NEXT:    bgtz a0, .LBB2_4
 ; RV64IF-NEXT:  # %bb.3: # %entry
 ; RV64IF-NEXT:    li a0, 0
 ; RV64IF-NEXT:  .LBB2_4: # %entry
 ; RV64IF-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IF-NEXT:    addi sp, sp, 16
 ; RV64IF-NEXT:    ret
 ;
+; RV32IFD-LABEL: ustest_f64i32:
+; RV32IFD:       # %bb.0: # %entry
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    .cfi_def_cfa_offset 16
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    feq.d a0, ft0, ft0
+; RV32IFD-NEXT:    bnez a0, .LBB2_2
+; RV32IFD-NEXT:  # %bb.1: # %entry
+; RV32IFD-NEXT:    li a0, 0
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+; RV32IFD-NEXT:  .LBB2_2:
+; RV32IFD-NEXT:    fcvt.wu.d a0, ft0, rtz
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
 ; RV64IFD-LABEL: ustest_f64i32:
 ; RV64IFD:       # %bb.0: # %entry
 ; RV64IFD-NEXT:    fmv.d.x ft0, a0
 ; RV64IFD-NEXT:    fcvt.l.d a0, ft0, rtz
 ; RV64IFD-NEXT:    li a1, -1
 ; RV64IFD-NEXT:    srli a1, a1, 32
 ; RV64IFD-NEXT:    bge a0, a1, .LBB2_3
 ; RV64IFD-NEXT:  # %bb.1: # %entry
entry:
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f32i32(float %x) {
 ; RV32-LABEL: ustest_f32i32:
 ; RV32:       # %bb.0: # %entry
-; RV32-NEXT:    addi sp, sp, -16
-; RV32-NEXT:    .cfi_def_cfa_offset 16
-; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32-NEXT:    .cfi_offset ra, -4
-; RV32-NEXT:    call __fixsfdi@plt
-; RV32-NEXT:    beqz a1, .LBB5_2
+; RV32-NEXT:    fmv.w.x ft0, a0
+; RV32-NEXT:    feq.s a0, ft0, ft0
+; RV32-NEXT:    bnez a0, .LBB5_2
 ; RV32-NEXT:  # %bb.1: # %entry
-; RV32-NEXT:    slti a2, a1, 0
-; RV32-NEXT:    beqz a2, .LBB5_3
-; RV32-NEXT:    j .LBB5_4
-; RV32-NEXT:  .LBB5_2:
-; RV32-NEXT:    addi a2, a0, 1
-; RV32-NEXT:    snez a2, a2
-; RV32-NEXT:    bnez a2, .LBB5_4
-; RV32-NEXT:  .LBB5_3: # %entry
-; RV32-NEXT:    li a1, 0
-; RV32-NEXT:    li a0, -1
-; RV32-NEXT:  .LBB5_4: # %entry
-; RV32-NEXT:    beqz a1, .LBB5_6
-; RV32-NEXT:  # %bb.5: # %entry
-; RV32-NEXT:    sgtz a1, a1
-; RV32-NEXT:    beqz a1, .LBB5_7
-; RV32-NEXT:    j .LBB5_8
-; RV32-NEXT:  .LBB5_6:
-; RV32-NEXT:    snez a1, a0
-; RV32-NEXT:    bnez a1, .LBB5_8
-; RV32-NEXT:  .LBB5_7: # %entry
 ; RV32-NEXT:    li a0, 0
-; RV32-NEXT:  .LBB5_8: # %entry
-; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    ret
+; RV32-NEXT:  .LBB5_2:
+; RV32-NEXT:    fcvt.wu.s a0, ft0, rtz
 ; RV32-NEXT:    ret
 ;
 ; RV64-LABEL: ustest_f32i32:
 ; RV64:       # %bb.0: # %entry
 ; RV64-NEXT:    fmv.w.x ft0, a0
 ; RV64-NEXT:    fcvt.l.s a0, ft0, rtz
 ; RV64-NEXT:    li a1, -1
 ; RV64-NEXT:    srli a1, a1, 32

 entry:
   %conv = fptoui double %x to i64
   %spec.store.select = call i64 @llvm.umin.i64(i64 %conv, i64 4294967295)
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f64i32_mm(double %x) {
-; RV32-LABEL: ustest_f64i32_mm:
-; RV32:       # %bb.0: # %entry
-; RV32-NEXT:    addi sp, sp, -16
-; RV32-NEXT:    .cfi_def_cfa_offset 16
-; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32-NEXT:    .cfi_offset ra, -4
-; RV32-NEXT:    call __fixdfdi@plt
-; RV32-NEXT:    mv a2, a0
-; RV32-NEXT:    bgez a1, .LBB29_7
-; RV32-NEXT:  # %bb.1: # %entry
-; RV32-NEXT:    bnez a1, .LBB29_8
-; RV32-NEXT:  .LBB29_2: # %entry
-; RV32-NEXT:    bgez a1, .LBB29_9
-; RV32-NEXT:  .LBB29_3: # %entry
-; RV32-NEXT:    mv a2, a0
-; RV32-NEXT:    blez a1, .LBB29_10
-; RV32-NEXT:  .LBB29_4: # %entry
-; RV32-NEXT:    beqz a1, .LBB29_6
-; RV32-NEXT:  .LBB29_5: # %entry
-; RV32-NEXT:    mv a0, a2
-; RV32-NEXT:  .LBB29_6: # %entry
-; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32-NEXT:    addi sp, sp, 16
-; RV32-NEXT:    ret
-; RV32-NEXT:  .LBB29_7: # %entry
-; RV32-NEXT:    li a2, -1
-; RV32-NEXT:    beqz a1, .LBB29_2
-; RV32-NEXT:  .LBB29_8: # %entry
-; RV32-NEXT:    mv a0, a2
-; RV32-NEXT:    bltz a1, .LBB29_3
-; RV32-NEXT:  .LBB29_9: # %entry
-; RV32-NEXT:    li a1, 0
-; RV32-NEXT:    mv a2, a0
-; RV32-NEXT:    bgtz a1, .LBB29_4
-; RV32-NEXT:  .LBB29_10: # %entry
-; RV32-NEXT:    li a2, 0
-; RV32-NEXT:    bnez a1, .LBB29_5
-; RV32-NEXT:    j .LBB29_6
+; RV32IF-LABEL: ustest_f64i32_mm:
+; RV32IF:       # %bb.0: # %entry
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    .cfi_def_cfa_offset 16
+; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IF-NEXT:    .cfi_offset ra, -4
+; RV32IF-NEXT:    call __fixdfdi@plt
+; RV32IF-NEXT:    mv a2, a0
+; RV32IF-NEXT:    bgez a1, .LBB29_7
+; RV32IF-NEXT:  # %bb.1: # %entry
+; RV32IF-NEXT:    bnez a1, .LBB29_8
+; RV32IF-NEXT:  .LBB29_2: # %entry
+; RV32IF-NEXT:    bgez a1, .LBB29_9
+; RV32IF-NEXT:  .LBB29_3: # %entry
+; RV32IF-NEXT:    mv a2, a0
+; RV32IF-NEXT:    blez a1, .LBB29_10
+; RV32IF-NEXT:  .LBB29_4: # %entry
+; RV32IF-NEXT:    beqz a1, .LBB29_6
+; RV32IF-NEXT:  .LBB29_5: # %entry
+; RV32IF-NEXT:    mv a0, a2
+; RV32IF-NEXT:  .LBB29_6: # %entry
+; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+; RV32IF-NEXT:  .LBB29_7: # %entry
+; RV32IF-NEXT:    li a2, -1
+; RV32IF-NEXT:    beqz a1, .LBB29_2
+; RV32IF-NEXT:  .LBB29_8: # %entry
+; RV32IF-NEXT:    mv a0, a2
+; RV32IF-NEXT:    bltz a1, .LBB29_3
+; RV32IF-NEXT:  .LBB29_9: # %entry
+; RV32IF-NEXT:    li a1, 0
+; RV32IF-NEXT:    mv a2, a0
+; RV32IF-NEXT:    bgtz a1, .LBB29_4
+; RV32IF-NEXT:  .LBB29_10: # %entry
+; RV32IF-NEXT:    li a2, 0
+; RV32IF-NEXT:    bnez a1, .LBB29_5
+; RV32IF-NEXT:    j .LBB29_6
 ;
 ; RV64IF-LABEL: ustest_f64i32_mm:
 ; RV64IF:       # %bb.0: # %entry
 ; RV64IF-NEXT:    addi sp, sp, -16
 ; RV64IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV64IF-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IF-NEXT:    .cfi_offset ra, -8
 ; RV64IF-NEXT:    call __fixdfdi@plt
 ; RV64IF-NEXT:    li a1, -1
 ; RV64IF-NEXT:    srli a1, a1, 32
 ; RV64IF-NEXT:    blt a0, a1, .LBB29_2
 ; RV64IF-NEXT:  # %bb.1: # %entry
 ; RV64IF-NEXT:    mv a0, a1
 ; RV64IF-NEXT:  .LBB29_2: # %entry
 ; RV64IF-NEXT:    bgtz a0, .LBB29_4
 ; RV64IF-NEXT:  # %bb.3: # %entry
 ; RV64IF-NEXT:    li a0, 0
 ; RV64IF-NEXT:  .LBB29_4: # %entry
 ; RV64IF-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IF-NEXT:    addi sp, sp, 16
 ; RV64IF-NEXT:    ret
 ;
+; RV32IFD-LABEL: ustest_f64i32_mm:
+; RV32IFD:       # %bb.0: # %entry
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    .cfi_def_cfa_offset 16
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    feq.d a0, ft0, ft0
+; RV32IFD-NEXT:    bnez a0, .LBB29_2
+; RV32IFD-NEXT:  # %bb.1: # %entry
+; RV32IFD-NEXT:    li a0, 0
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+; RV32IFD-NEXT:  .LBB29_2:
+; RV32IFD-NEXT:    fcvt.wu.d a0, ft0, rtz
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
 ; RV64IFD-LABEL: ustest_f64i32_mm:
 ; RV64IFD:       # %bb.0: # %entry
 ; RV64IFD-NEXT:    fmv.d.x ft0, a0
 ; RV64IFD-NEXT:    fcvt.l.d a0, ft0, rtz
 ; RV64IFD-NEXT:    li a1, -1
 ; RV64IFD-NEXT:    srli a1, a1, 32
 ; RV64IFD-NEXT:    bge a0, a1, .LBB29_3
 ; RV64IFD-NEXT:  # %bb.1: # %entry
entry:
   %spec.store.select = call i64 @llvm.umin.i64(i64 %conv, i64 4294967295)
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f32i32_mm(float %x) {
 ; RV32-LABEL: ustest_f32i32_mm:
 ; RV32:       # %bb.0: # %entry
-; RV32-NEXT:    addi sp, sp, -16
-; RV32-NEXT:    .cfi_def_cfa_offset 16
-; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32-NEXT:    .cfi_offset ra, -4
-; RV32-NEXT:    call __fixsfdi@plt
-; RV32-NEXT:    mv a2, a0
-; RV32-NEXT:    bgez a1, .LBB32_7
+; RV32-NEXT:    fmv.w.x ft0, a0
+; RV32-NEXT:    feq.s a0, ft0, ft0
+; RV32-NEXT:    bnez a0, .LBB32_2
 ; RV32-NEXT:  # %bb.1: # %entry
-; RV32-NEXT:    bnez a1, .LBB32_8
-; RV32-NEXT:  .LBB32_2: # %entry
-; RV32-NEXT:    bgez a1, .LBB32_9
-; RV32-NEXT:  .LBB32_3: # %entry
-; RV32-NEXT:    mv a2, a0
-; RV32-NEXT:    blez a1, .LBB32_10
-; RV32-NEXT:  .LBB32_4: # %entry
-; RV32-NEXT:    beqz a1, .LBB32_6
-; RV32-NEXT:  .LBB32_5: # %entry
-; RV32-NEXT:    mv a0, a2
-; RV32-NEXT:  .LBB32_6: # %entry
-; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    li a0, 0
+; RV32-NEXT:    ret
+; RV32-NEXT:  .LBB32_2:
+; RV32-NEXT:    fcvt.wu.s a0, ft0, rtz
 ; RV32-NEXT:    ret
-; RV32-NEXT:  .LBB32_7: # %entry
-; RV32-NEXT:    li a2, -1
-; RV32-NEXT:    beqz a1, .LBB32_2
-; RV32-NEXT:  .LBB32_8: # %entry
-; RV32-NEXT:    mv a0, a2
-; RV32-NEXT:    bltz a1, .LBB32_3
-; RV32-NEXT:  .LBB32_9: # %entry
-; RV32-NEXT:    li a1, 0
-; RV32-NEXT:    mv a2, a0
-; RV32-NEXT:    bgtz a1, .LBB32_4
-; RV32-NEXT:  .LBB32_10: # %entry
-; RV32-NEXT:    li a2, 0
-; RV32-NEXT:    bnez a1, .LBB32_5
-; RV32-NEXT:    j .LBB32_6
 ;
 ; RV64-LABEL: ustest_f32i32_mm:
 ; RV64:       # %bb.0: # %entry
 ; RV64-NEXT:    fmv.w.x ft0, a0
 ; RV64-NEXT:    fcvt.l.s a0, ft0, rtz
 ; RV64-NEXT:    li a1, -1
 ; RV64-NEXT:    srli a1, a1, 32
 ; RV64-NEXT:    bge a0, a1, .LBB32_3

llvm/test/CodeGen/Thumb2/mve-fpclamptosat_vec.ll

entry:
   %spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
   ret <4 x i32> %conv6
 }
 
 define arm_aapcs_vfpcc <4 x i32> @ustest_f32i32(<4 x float> %x) {
 ; CHECK-LABEL: ustest_f32i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r6, r7, lr}
-; CHECK-NEXT:    push {r4, r5, r6, r7, lr}
-; CHECK-NEXT:    .pad #4
-; CHECK-NEXT:    sub sp, #4
-; CHECK-NEXT:    .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    vmov q4, q0
-; CHECK-NEXT:    vmov r0, r4, d9
-; CHECK-NEXT:    bl __aeabi_f2lz
-; CHECK-NEXT:    mov r5, r0
-; CHECK-NEXT:    mov r0, r4
-; CHECK-NEXT:    mov r6, r1
-; CHECK-NEXT:    bl __aeabi_f2lz
-; CHECK-NEXT:    subs.w r2, r5, #-1
-; CHECK-NEXT:    vmov q0[2], q0[0], r5, r0
-; CHECK-NEXT:    sbcs r2, r6, #0
-; CHECK-NEXT:    mov.w r3, #0
-; CHECK-NEXT:    cset r2, lt
-; CHECK-NEXT:    vmov.i64 q5, #0xffffffff
-; CHECK-NEXT:    cmp r2, #0
-; CHECK-NEXT:    vmov q0[3], q0[1], r6, r1
-; CHECK-NEXT:    csetm r2, ne
-; CHECK-NEXT:    subs.w r0, r0, #-1
-; CHECK-NEXT:    sbcs r0, r1, #0
-; CHECK-NEXT:    bfi r3, r2, #0, #8
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    movs r7, #0
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    vmov.i32 q6, #0x0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    bfi r3, r0, #8, #8
-; CHECK-NEXT:    vmsr p0, r3
-; CHECK-NEXT:    vpsel q0, q0, q5
-; CHECK-NEXT:    vmov r0, r1, d0
-; CHECK-NEXT:    vmov r2, r3, d1
-; CHECK-NEXT:    rsbs r0, r0, #0
-; CHECK-NEXT:    sbcs.w r0, r7, r1
-; CHECK-NEXT:    mov.w r1, #0
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    bfi r1, r0, #0, #8
-; CHECK-NEXT:    rsbs r0, r2, #0
-; CHECK-NEXT:    sbcs.w r0, r7, r3
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    bfi r1, r0, #8, #8
-; CHECK-NEXT:    vmov r0, r4, d8
-; CHECK-NEXT:    vmsr p0, r1
-; CHECK-NEXT:    vpsel q7, q0, q6
-; CHECK-NEXT:    bl __aeabi_f2lz
-; CHECK-NEXT:    mov r5, r0
-; CHECK-NEXT:    mov r0, r4
-; CHECK-NEXT:    mov r6, r1
-; CHECK-NEXT:    bl __aeabi_f2lz
-; CHECK-NEXT:    subs.w r2, r5, #-1
-; CHECK-NEXT:    vmov q0[2], q0[0], r5, r0
-; CHECK-NEXT:    sbcs r2, r6, #0
-; CHECK-NEXT:    mov.w r3, #0
-; CHECK-NEXT:    cset r2, lt
-; CHECK-NEXT:    vmov q0[3], q0[1], r6, r1
-; CHECK-NEXT:    cmp r2, #0
-; CHECK-NEXT:    csetm r2, ne
-; CHECK-NEXT:    subs.w r0, r0, #-1
-; CHECK-NEXT:    sbcs r0, r1, #0
-; CHECK-NEXT:    bfi r3, r2, #0, #8
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    bfi r3, r0, #8, #8
-; CHECK-NEXT:    vmsr p0, r3
-; CHECK-NEXT:    vpsel q0, q0, q5
-; CHECK-NEXT:    vmov r0, r1, d0
-; CHECK-NEXT:    vmov r2, r3, d1
-; CHECK-NEXT:    rsbs r0, r0, #0
-; CHECK-NEXT:    sbcs.w r0, r7, r1
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    rsbs r1, r2, #0
-; CHECK-NEXT:    sbcs.w r1, r7, r3
-; CHECK-NEXT:    bfi r7, r0, #0, #8
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    bfi r7, r0, #8, #8
-; CHECK-NEXT:    vmsr p0, r7
-; CHECK-NEXT:    vpsel q0, q0, q6
-; CHECK-NEXT:    vmov.f32 s1, s2
-; CHECK-NEXT:    vmov.f32 s2, s28
-; CHECK-NEXT:    vmov.f32 s3, s30
-; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    add sp, #4
-; CHECK-NEXT:    pop {r4, r5, r6, r7, pc}
+; CHECK-NEXT:    vcvt.u32.f32 q0, q0
+; CHECK-NEXT:    bx lr
 entry:
   %conv = fptosi <4 x float> %x to <4 x i64>
   %0 = icmp slt <4 x i64> %conv, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %1 = icmp sgt <4 x i64> %spec.store.select, zeroinitializer
   %spec.store.select7 = select <4 x i1> %1, <4 x i64> %spec.store.select, <4 x i64> zeroinitializer
   %conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
   ret <4 x i32> %conv6
entry:
   %spec.store.select = call <4 x i64> @llvm.umin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
   %conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
   ret <4 x i32> %conv6
 }
 
 define arm_aapcs_vfpcc <4 x i32> @ustest_f32i32_mm(<4 x float> %x) {
 ; CHECK-LABEL: ustest_f32i32_mm:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r6, r7, lr}
-; CHECK-NEXT:    push {r4, r5, r6, r7, lr}
-; CHECK-NEXT:    .pad #4
-; CHECK-NEXT:    sub sp, #4
-; CHECK-NEXT:    .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    vmov q4, q0
-; CHECK-NEXT:    vmov r0, r4, d9
-; CHECK-NEXT:    bl __aeabi_f2lz
-; CHECK-NEXT:    mov r5, r0
-; CHECK-NEXT:    mov r0, r4
-; CHECK-NEXT:    mov r6, r1
-; CHECK-NEXT:    bl __aeabi_f2lz
-; CHECK-NEXT:    subs.w r2, r5, #-1
-; CHECK-NEXT:    vmov q0[2], q0[0], r5, r0
-; CHECK-NEXT:    sbcs r2, r6, #0
-; CHECK-NEXT:    mov.w r3, #0
-; CHECK-NEXT:    cset r2, lt
-; CHECK-NEXT:    vmov.i64 q5, #0xffffffff
-; CHECK-NEXT:    cmp r2, #0
-; CHECK-NEXT:    vmov q0[3], q0[1], r6, r1
-; CHECK-NEXT:    csetm r2, ne
-; CHECK-NEXT:    subs.w r0, r0, #-1
-; CHECK-NEXT:    sbcs r0, r1, #0
-; CHECK-NEXT:    bfi r3, r2, #0, #8
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    movs r7, #0
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    vmov.i32 q6, #0x0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    bfi r3, r0, #8, #8
-; CHECK-NEXT:    vmsr p0, r3
-; CHECK-NEXT:    vpsel q0, q0, q5
-; CHECK-NEXT:    vmov r0, r1, d0
-; CHECK-NEXT:    vmov r2, r3, d1
-; CHECK-NEXT:    rsbs r0, r0, #0
-; CHECK-NEXT:    sbcs.w r0, r7, r1
-; CHECK-NEXT:    mov.w r1, #0
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    bfi r1, r0, #0, #8
-; CHECK-NEXT:    rsbs r0, r2, #0
-; CHECK-NEXT:    sbcs.w r0, r7, r3
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    bfi r1, r0, #8, #8
-; CHECK-NEXT:    vmov r0, r4, d8
-; CHECK-NEXT:    vmsr p0, r1
-; CHECK-NEXT:    vpsel q7, q0, q6
-; CHECK-NEXT:    bl __aeabi_f2lz
-; CHECK-NEXT:    mov r5, r0
-; CHECK-NEXT:    mov r0, r4
-; CHECK-NEXT:    mov r6, r1
-; CHECK-NEXT:    bl __aeabi_f2lz
-; CHECK-NEXT:    subs.w r2, r5, #-1
-; CHECK-NEXT:    vmov q0[2], q0[0], r5, r0
-; CHECK-NEXT:    sbcs r2, r6, #0
-; CHECK-NEXT:    mov.w r3, #0
-; CHECK-NEXT:    cset r2, lt
-; CHECK-NEXT:    vmov q0[3], q0[1], r6, r1
-; CHECK-NEXT:    cmp r2, #0
-; CHECK-NEXT:    csetm r2, ne
-; CHECK-NEXT:    subs.w r0, r0, #-1
-; CHECK-NEXT:    sbcs r0, r1, #0
-; CHECK-NEXT:    bfi r3, r2, #0, #8
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    bfi r3, r0, #8, #8
-; CHECK-NEXT:    vmsr p0, r3
-; CHECK-NEXT:    vpsel q0, q0, q5
-; CHECK-NEXT:    vmov r0, r1, d0
-; CHECK-NEXT:    vmov r2, r3, d1
-; CHECK-NEXT:    rsbs r0, r0, #0
-; CHECK-NEXT:    sbcs.w r0, r7, r1
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    rsbs r1, r2, #0
-; CHECK-NEXT:    sbcs.w r1, r7, r3
-; CHECK-NEXT:    bfi r7, r0, #0, #8
-; CHECK-NEXT:    cset r0, lt
-; CHECK-NEXT:    cmp r0, #0
-; CHECK-NEXT:    csetm r0, ne
-; CHECK-NEXT:    bfi r7, r0, #8, #8
-; CHECK-NEXT:    vmsr p0, r7
-; CHECK-NEXT:    vpsel q0, q0, q6
-; CHECK-NEXT:    vmov.f32 s1, s2
-; CHECK-NEXT:    vmov.f32 s2, s28
-; CHECK-NEXT:    vmov.f32 s3, s30
-; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    add sp, #4
-; CHECK-NEXT:    pop {r4, r5, r6, r7, pc}
+; CHECK-NEXT:    vcvt.u32.f32 q0, q0
+; CHECK-NEXT:    bx lr
 entry:
   %conv = fptosi <4 x float> %x to <4 x i64>
   %spec.store.select = call <4 x i64> @llvm.smin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
   %spec.store.select7 = call <4 x i64> @llvm.smax.v4i64(<4 x i64> %spec.store.select, <4 x i64> zeroinitializer)
   %conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
   ret <4 x i32> %conv6
 }
 

llvm/test/CodeGen/WebAssembly/fpclamptosat.ll

entry:
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f64i32(double %x) {
 ; CHECK-LABEL: ustest_f64i32:
 ; CHECK:         .functype ustest_f64i32 (f64) -> (i32)
-; CHECK-NEXT:    .local i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    i64.trunc_sat_f64_s
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    i64.lt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    i64.gt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    i32.wrap_i64
+; CHECK-NEXT:    i32.trunc_sat_f64_u
 ; CHECK-NEXT:    # fallthrough-return
 entry:
   %conv = fptosi double %x to i64
   %0 = icmp slt i64 %conv, 4294967295
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %1 = icmp sgt i64 %spec.store.select, 0
   %spec.store.select7 = select i1 %1, i64 %spec.store.select, i64 0
   %conv6 = trunc i64 %spec.store.select7 to i32
entry:
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f32i32(float %x) {
 ; CHECK-LABEL: ustest_f32i32:
 ; CHECK:         .functype ustest_f32i32 (f32) -> (i32)
-; CHECK-NEXT:    .local i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    i64.lt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    i64.gt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    i32.wrap_i64
+; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
 entry:
   %conv = fptosi float %x to i64
   %0 = icmp slt i64 %conv, 4294967295
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %1 = icmp sgt i64 %spec.store.select, 0
   %spec.store.select7 = select i1 %1, i64 %spec.store.select, i64 0
   %conv6 = trunc i64 %spec.store.select7 to i32
entry:
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f16i32(half %x) {
 ; CHECK-LABEL: ustest_f16i32:
 ; CHECK:         .functype ustest_f16i32 (f32) -> (i32)
-; CHECK-NEXT:    .local i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    i64.lt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    i64.gt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    i32.wrap_i64
+; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
 entry:
   %conv = fptosi half %x to i64
   %0 = icmp slt i64 %conv, 4294967295
   %spec.store.select = select i1 %0, i64 %conv, i64 4294967295
   %1 = icmp sgt i64 %spec.store.select, 0
   %spec.store.select7 = select i1 %1, i64 %spec.store.select, i64 0
   %conv6 = trunc i64 %spec.store.select7 to i32
entry:
   %spec.store.select = call i64 @llvm.umin.i64(i64 %conv, i64 4294967295)
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f64i32_mm(double %x) {
 ; CHECK-LABEL: ustest_f64i32_mm:
 ; CHECK:         .functype ustest_f64i32_mm (f64) -> (i32)
-; CHECK-NEXT:    .local i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    i64.trunc_sat_f64_s
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    i64.lt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    i64.gt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    i32.wrap_i64
+; CHECK-NEXT:    i32.trunc_sat_f64_u
 ; CHECK-NEXT:    # fallthrough-return
 entry:
   %conv = fptosi double %x to i64
   %spec.store.select = call i64 @llvm.smin.i64(i64 %conv, i64 4294967295)
   %spec.store.select7 = call i64 @llvm.smax.i64(i64 %spec.store.select, i64 0)
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6
 }
entry:
   %spec.store.select = call i64 @llvm.umin.i64(i64 %conv, i64 4294967295)
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f32i32_mm(float %x) {
 ; CHECK-LABEL: ustest_f32i32_mm:
 ; CHECK:         .functype ustest_f32i32_mm (f32) -> (i32)
-; CHECK-NEXT:    .local i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    i64.lt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    i64.gt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    i32.wrap_i64
+; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
 entry:
   %conv = fptosi float %x to i64
   %spec.store.select = call i64 @llvm.smin.i64(i64 %conv, i64 4294967295)
   %spec.store.select7 = call i64 @llvm.smax.i64(i64 %spec.store.select, i64 0)
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6
 }
entry:
   %spec.store.select = call i64 @llvm.umin.i64(i64 %conv, i64 4294967295)
   %conv6 = trunc i64 %spec.store.select to i32
   ret i32 %conv6
 }
 
 define i32 @ustest_f16i32_mm(half %x) {
 ; CHECK-LABEL: ustest_f16i32_mm:
 ; CHECK:         .functype ustest_f16i32_mm (f32) -> (i32)
-; CHECK-NEXT:    .local i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 4294967295
-; CHECK-NEXT:    i64.lt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.const 0
-; CHECK-NEXT:    i64.gt_s
-; CHECK-NEXT:    i64.select
-; CHECK-NEXT:    i32.wrap_i64
+; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
 entry:
   %conv = fptosi half %x to i64
   %spec.store.select = call i64 @llvm.smin.i64(i64 %conv, i64 4294967295)
   %spec.store.select7 = call i64 @llvm.smax.i64(i64 %spec.store.select, i64 0)
   %conv6 = trunc i64 %spec.store.select7 to i32
   ret i32 %conv6
 }

llvm/test/CodeGen/WebAssembly/fpclamptosat_vec.ll

entry:
   %spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
   ret <4 x i32> %conv6
 }
 
 define <4 x i32> @ustest_f32i32(<4 x float> %x) {
 ; CHECK-LABEL: ustest_f32i32:
 ; CHECK:         .functype ustest_f32i32 (v128) -> (v128)
-; CHECK-NEXT:    .local v128, v128, v128
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    f32x4.extract_lane 0
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.splat
-; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    f32x4.extract_lane 1
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.replace_lane 1
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    v128.const 4294967295, 4294967295
-; CHECK-NEXT:    local.tee 2
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    i64x2.lt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    local.tee 3
-; CHECK-NEXT:    v128.const 0, 0
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64x2.gt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    f32x4.extract_lane 2
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.splat
-; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    f32x4.extract_lane 3
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.replace_lane 1
-; CHECK-NEXT:    local.tee 0
-; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    i64x2.lt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    local.tee 0
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64x2.gt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    i8x16.shuffle 0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27
+; CHECK-NEXT:    i32x4.trunc_sat_f32x4_u
 ; CHECK-NEXT:    # fallthrough-return
 entry:
   %conv = fptosi <4 x float> %x to <4 x i64>
   %0 = icmp slt <4 x i64> %conv, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %1 = icmp sgt <4 x i64> %spec.store.select, zeroinitializer
   %spec.store.select7 = select <4 x i1> %1, <4 x i64> %spec.store.select, <4 x i64> zeroinitializer
   %conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
entry:
   %spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
   ret <4 x i32> %conv6
 }
 
 define <4 x i32> @ustest_f16i32(<4 x half> %x) {
 ; CHECK-LABEL: ustest_f16i32:
 ; CHECK:         .functype ustest_f16i32 (f32, f32, f32, f32) -> (v128)
-; CHECK-NEXT:    .local v128, v128, v128
 ; CHECK-NEXT:  # %bb.0: # %entry
-; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
-; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 3
-; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
-; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 2
 ; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.splat
+; CHECK-NEXT:    i32.trunc_sat_f32_u
+; CHECK-NEXT:    i32x4.splat
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.replace_lane 1
-; CHECK-NEXT:    local.tee 4
-; CHECK-NEXT:    v128.const 4294967295, 4294967295
-; CHECK-NEXT:    local.tee 5
-; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    i64x2.lt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    local.tee 6
-; CHECK-NEXT:    v128.const 0, 0
-; CHECK-NEXT:    local.tee 4
-; CHECK-NEXT:    local.get 6
-; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    i64x2.gt_s
-; CHECK-NEXT:    v128.bitselect
+; CHECK-NEXT:    i32.trunc_sat_f32_u
+; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.splat
+; CHECK-NEXT:    call __truncsfhf2
+; CHECK-NEXT:    call __extendhfsf2
+; CHECK-NEXT:    i32.trunc_sat_f32_u
+; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.replace_lane 1
-; CHECK-NEXT:    local.tee 6
-; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    local.get 6
-; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    i64x2.lt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    local.tee 5
-; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    i64x2.gt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    i8x16.shuffle 0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27
+; CHECK-NEXT:    call __truncsfhf2
+; CHECK-NEXT:    call __extendhfsf2
+; CHECK-NEXT:    i32.trunc_sat_f32_u
+; CHECK-NEXT:    i32x4.replace_lane 3
 ; CHECK-NEXT:    # fallthrough-return
 entry:
   %conv = fptosi <4 x half> %x to <4 x i64>
   %0 = icmp slt <4 x i64> %conv, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
   %1 = icmp sgt <4 x i64> %spec.store.select, zeroinitializer
   %spec.store.select7 = select <4 x i1> %1, <4 x i64> %spec.store.select, <4 x i64> zeroinitializer
   %conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
entry:
   %spec.store.select = call <4 x i64> @llvm.umin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
   %conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
   ret <4 x i32> %conv6
 }
 
 define <4 x i32> @ustest_f32i32_mm(<4 x float> %x) {
 ; CHECK-LABEL: ustest_f32i32_mm:
 ; CHECK:         .functype ustest_f32i32_mm (v128) -> (v128)
-; CHECK-NEXT:    .local v128, v128, v128
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    f32x4.extract_lane 0
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.splat
-; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    f32x4.extract_lane 1
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.replace_lane 1
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    v128.const 4294967295, 4294967295
-; CHECK-NEXT:    local.tee 2
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    i64x2.lt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    local.tee 3
-; CHECK-NEXT:    v128.const 0, 0
-; CHECK-NEXT:    local.tee 1
-; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64x2.gt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    f32x4.extract_lane 2
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.splat
-; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    f32x4.extract_lane 3
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.replace_lane 1
-; CHECK-NEXT:    local.tee 0
-; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    i64x2.lt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    local.tee 0
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64x2.gt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    i8x16.shuffle 0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27
+; CHECK-NEXT:    i32x4.trunc_sat_f32x4_u
 ; CHECK-NEXT:    # fallthrough-return
 entry:
   %conv = fptosi <4 x float> %x to <4 x i64>
   %spec.store.select = call <4 x i64> @llvm.smin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
   %spec.store.select7 = call <4 x i64> @llvm.smax.v4i64(<4 x i64> %spec.store.select, <4 x i64> zeroinitializer)
   %conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
   ret <4 x i32> %conv6
 }
entry:
   %spec.store.select = call <4 x i64> @llvm.umin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
   %conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
   ret <4 x i32> %conv6
 }
 
 define <4 x i32> @ustest_f16i32_mm(<4 x half> %x) {
 ; CHECK-LABEL: ustest_f16i32_mm:
 ; CHECK:         .functype ustest_f16i32_mm (f32, f32, f32, f32) -> (v128)
-; CHECK-NEXT:    .local v128, v128, v128
 ; CHECK-NEXT:  # %bb.0: # %entry
-; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
-; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 3
-; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
-; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 2
 ; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    local.set 1
 ; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.splat
+; CHECK-NEXT:    i32.trunc_sat_f32_u
+; CHECK-NEXT:    i32x4.splat
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.replace_lane 1
-; CHECK-NEXT:    local.tee 4
-; CHECK-NEXT:    v128.const 4294967295, 4294967295
-; CHECK-NEXT:    local.tee 5
-; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    i64x2.lt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    local.tee 6
-; CHECK-NEXT:    v128.const 0, 0
-; CHECK-NEXT:    local.tee 4
-; CHECK-NEXT:    local.get 6
-; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    i64x2.gt_s
-; CHECK-NEXT:    v128.bitselect
+; CHECK-NEXT:    i32.trunc_sat_f32_u
+; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.splat
+; CHECK-NEXT:    call __truncsfhf2
+; CHECK-NEXT:    call __extendhfsf2
+; CHECK-NEXT:    i32.trunc_sat_f32_u
+; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    i64.trunc_sat_f32_s
-; CHECK-NEXT:    i64x2.replace_lane 1
-; CHECK-NEXT:    local.tee 6
-; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    local.get 6
-; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    i64x2.lt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    local.tee 5
-; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    i64x2.gt_s
-; CHECK-NEXT:    v128.bitselect
-; CHECK-NEXT:    i8x16.shuffle 0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27
+; CHECK-NEXT:    call __truncsfhf2
+; CHECK-NEXT:    call __extendhfsf2
+; CHECK-NEXT:    i32.trunc_sat_f32_u
+; CHECK-NEXT:    i32x4.replace_lane 3
 ; CHECK-NEXT:    # fallthrough-return
 entry:
   %conv = fptosi <4 x half> %x to <4 x i64>
   %spec.store.select = call <4 x i64> @llvm.smin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
   %spec.store.select7 = call <4 x i64> @llvm.smax.v4i64(<4 x i64> %spec.store.select, <4 x i64> zeroinitializer)
   %conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
   ret <4 x i32> %conv6
 }