[FPEnv][X86][SystemZ] Use different algorithms for i64->double uint_to_fp under strictfp to avoid producing -0.0 when rounding toward negative infinity

[craig.topper]

Summary

Some of our conversion algorithms produce -0.0 when converting unsigned i64 to double when the rounding mode is round toward negative. This switches them to other algorithms that don't have this problem. Since it is undefined behavior to change rounding mode with the non-strict nodes, this patch only changes the behavior for strict nodes.

There are still problems with unsigned i32 conversions too which I'll try to fix in another patch.

Fixes part of PR47393

[efriedma]

llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp

2440

What types is this correct for? I guess the integer type has to be larger than the mantissa by some number of bits, so the divide by 2 doesn't affect the rounding?

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6473

Should we drop the isStrictFPOpcode check later in the function?

[uweigand]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6472

Why is this testing isStrictFPOpcode? The only difference between strict and non-strict operations is that the former respect exceptions (and current rounding mode, if applicable). The behavior of signed zeros is not affected by those; even a non-strict UINT_TO_FP should never produce a negative zero (unless the NoSignedZero flags is set, I guess).

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6472

The issue only occurs if the rounding mode has been changed to FE_DOWNWARD. Since it’s undefined behavior to change rounding mode and use non-strict unit_to_fp, I didn’t change those.

[uweigand]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6472

Ah, I see. Thanks!

6495–6498

This comment is not accurate, then?

[craig.topper]

-Remove strict FP code that became dead with the new early outs
-Improve comments

[efriedma]

LGTM with one minor comment.

llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp

2440

I'd like to see a comment briefly describing the widths where this works.

[craig.topper]

Add the requested comment. @efriedma can you check my logic

[efriedma]

llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp

2440

Comment looks right.

llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp

if (Node->isStrictFPOpcode()) {
       Result = DAG.getFPExtendOrRound(Sub, dl, DestVT);
     }
     return Result;
   }
   // Code below here assumes !isSigned without checking again.
   assert(!isSigned && "Legalize cannot Expand SINT_TO_FP for i64 yet");
 
   // TODO: Generalize this for use with other types.
-  if ((SrcVT == MVT::i32 || SrcVT == MVT::i64) && DestVT == MVT::f32) {
-    LLVM_DEBUG(dbgs() << "Converting unsigned i32/i64 to f32\n");
+  if (((SrcVT == MVT::i32 || SrcVT == MVT::i64) && DestVT == MVT::f32) ||
+      (SrcVT == MVT::i64 && DestVT == MVT::f64)) {
+    LLVM_DEBUG(dbgs() << "Converting unsigned i32/i64 to f32/f64\n");
     // For unsigned conversions, convert them to signed conversions using the
     // algorithm from the x86_64 __floatundisf in compiler_rt. That method
     // should be valid for i32->f32 as well.
 
     // TODO: This really should be implemented using a branch rather than a
     // select.  We happen to get lucky and machinesink does the right
     // thing most of the time.  This would be a good candidate for a
     // pseudo-op, or, even better, for whole-function isel.

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

if (Strict) {
     Result = DAG.getSelect(dl, DstVT, Sel, True, False);
   }
   return true;
 }
 
 bool TargetLowering::expandUINT_TO_FP(SDNode *Node, SDValue &Result,
                                       SDValue &Chain,
                                       SelectionDAG &DAG) const {
-  unsigned OpNo = Node->isStrictFPOpcode() ? 1 : 0;
-  SDValue Src = Node->getOperand(OpNo);
+  // This transform is not correct for converting 0 when rounding mode is set
+  // to round toward negative infinity which will produce -0.0. So disable under
+  // strictfp.
+  if (Node->isStrictFPOpcode())
+    return false;
+
+  SDValue Src = Node->getOperand(0);
   EVT SrcVT = Src.getValueType();
   EVT DstVT = Node->getValueType(0);
 
   if (SrcVT.getScalarType() != MVT::i64 || DstVT.getScalarType() != MVT::f64)
     return false;
 
   // Only expand vector types if we have the appropriate vector bit operations.
   if (SrcVT.isVector() && (!isOperationLegalOrCustom(ISD::SRL, SrcVT) ||
                            !isOperationLegalOrCustom(ISD::FADD, DstVT) ||
                            !isOperationLegalOrCustom(ISD::FSUB, DstVT) ||
                            !isOperationLegalOrCustomOrPromote(ISD::OR, SrcVT) ||
                            !isOperationLegalOrCustomOrPromote(ISD::AND, SrcVT)))
     return false;
 
   SDLoc dl(SDValue(Node, 0));
   EVT ShiftVT = getShiftAmountTy(SrcVT, DAG.getDataLayout());
 
   // Implementation of unsigned i64 to f64 following the algorithm in
-  // __floatundidf in compiler_rt. This implementation has the advantage
-  // of performing rounding correctly, both in the default rounding mode
-  // and in all alternate rounding modes.
+  // __floatundidf in compiler_rt.  This implementation performs rounding
+  // correctly in all rounding modes with the exception of converting 0
+  // when rounding toward negative infinity. In that case the fsub will produce
+  // -0.0. This will be added to +0.0 and produce -0.0 which is incorrect.
   SDValue TwoP52 = DAG.getConstant(UINT64_C(0x4330000000000000), dl, SrcVT);
   SDValue TwoP84PlusTwoP52 = DAG.getConstantFP(
       BitsToDouble(UINT64_C(0x4530000000100000)), dl, DstVT);
   SDValue TwoP84 = DAG.getConstant(UINT64_C(0x4530000000000000), dl, SrcVT);
   SDValue LoMask = DAG.getConstant(UINT64_C(0x00000000FFFFFFFF), dl, SrcVT);
   SDValue HiShift = DAG.getConstant(32, dl, ShiftVT);
 
   SDValue Lo = DAG.getNode(ISD::AND, dl, SrcVT, Src, LoMask);
   SDValue Hi = DAG.getNode(ISD::SRL, dl, SrcVT, Src, HiShift);
   SDValue LoOr = DAG.getNode(ISD::OR, dl, SrcVT, Lo, TwoP52);
   SDValue HiOr = DAG.getNode(ISD::OR, dl, SrcVT, Hi, TwoP84);
   SDValue LoFlt = DAG.getBitcast(DstVT, LoOr);
   SDValue HiFlt = DAG.getBitcast(DstVT, HiOr);
-  if (Node->isStrictFPOpcode()) {
-    SDValue HiSub =
-        DAG.getNode(ISD::STRICT_FSUB, dl, {DstVT, MVT::Other},
-                    {Node->getOperand(0), HiFlt, TwoP84PlusTwoP52});
-    Result = DAG.getNode(ISD::STRICT_FADD, dl, {DstVT, MVT::Other},
-                         {HiSub.getValue(1), LoFlt, HiSub});
-    Chain = Result.getValue(1);
-  } else {
-    SDValue HiSub =
-        DAG.getNode(ISD::FSUB, dl, DstVT, HiFlt, TwoP84PlusTwoP52);
-    Result = DAG.getNode(ISD::FADD, dl, DstVT, LoFlt, HiSub);
-  }
   return true;
 }
 
 SDValue TargetLowering::expandFMINNUM_FMAXNUM(SDNode *Node,
                                               SelectionDAG &DAG) const {
   SDLoc dl(Node);
   unsigned NewOp = Node->getOpcode() == ISD::FMINNUM ?
     ISD::FMINNUM_IEEE : ISD::FMAXNUM_IEEE;

llvm/lib/Target/X86/X86ISelLowering.cpp

static bool shouldUseHorizontalOp(bool IsSingleSource, SelectionDAG &DAG,
   bool IsOptimizingSize = DAG.shouldOptForSize();
   bool HasFastHOps = Subtarget.hasFastHorizontalOps();
   return !IsSingleSource || IsOptimizingSize || HasFastHOps;
 }
 
 /// 64-bit unsigned integer to double expansion.
 static SDValue LowerUINT_TO_FP_i64(SDValue Op, SelectionDAG &DAG,
                                    const X86Subtarget &Subtarget) {
+  // We can't use this algorithm for strict fp. It produces -0.0 instead of +0.0
+  // when converting 0 when rounding toward negative infinity. Caller will
+  // fall back to Expand for when i64 or is legal or use FILD in 32-bit mode.
+  assert(!Op->isStrictFPOpcode() && "Expected non-strict uint_to_fp!");
   // This algorithm is not obvious. Here it is what we're trying to output:
   /*
      movq       %rax,  %xmm0
      punpckldq  (c0),  %xmm0  // c0: (uint4){ 0x43300000U, 0x45300000U, 0U, 0U }
      subpd      (c1),  %xmm0  // c1: (double2){ 0x1.0p52, 0x1.0p52 * 0x1.0p32 }
      #ifdef __SSE3__
        haddpd   %xmm0, %xmm0
      #else
        pshufd   $0x4e, %xmm0, %xmm1
        addpd    %xmm1, %xmm0
      #endif
   */
 
-  bool IsStrict = Op->isStrictFPOpcode();
-  unsigned OpNo = IsStrict ? 1 : 0;
   SDLoc dl(Op);
   LLVMContext *Context = DAG.getContext();
 
   // Build some magic constants.
   static const uint32_t CV0[] = { 0x43300000, 0x45300000, 0, 0 };
   Constant *C0 = ConstantDataVector::get(*Context, CV0);
   auto PtrVT = DAG.getTargetLoweringInfo().getPointerTy(DAG.getDataLayout());
   SDValue CPIdx0 = DAG.getConstantPool(C0, PtrVT, Align(16));
 
   SmallVector<Constant*,2> CV1;
   CV1.push_back(
     ConstantFP::get(*Context, APFloat(APFloat::IEEEdouble(),
                                       APInt(64, 0x4330000000000000ULL))));
   CV1.push_back(
     ConstantFP::get(*Context, APFloat(APFloat::IEEEdouble(),
                                       APInt(64, 0x4530000000000000ULL))));
   Constant *C1 = ConstantVector::get(CV1);
   SDValue CPIdx1 = DAG.getConstantPool(C1, PtrVT, Align(16));
 
   // Load the 64-bit value into an XMM register.
   SDValue XR1 =
-      DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64, Op.getOperand(OpNo));
+      DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64, Op.getOperand(0));
   SDValue CLod0 =
       DAG.getLoad(MVT::v4i32, dl, DAG.getEntryNode(), CPIdx0,
                   MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
                   /* Alignment = */ 16);
   SDValue Unpck1 =
       getUnpackl(DAG, dl, MVT::v4i32, DAG.getBitcast(MVT::v4i32, XR1), CLod0);
 
   SDValue CLod1 =
       DAG.getLoad(MVT::v2f64, dl, CLod0.getValue(1), CPIdx1,
                   MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
                   /* Alignment = */ 16);
   SDValue XR2F = DAG.getBitcast(MVT::v2f64, Unpck1);
   SDValue Sub;
   SDValue Chain;
   // TODO: Are there any fast-math-flags to propagate here?
-  if (IsStrict) {
-    Sub = DAG.getNode(ISD::STRICT_FSUB, dl, {MVT::v2f64, MVT::Other},
-                      {Op.getOperand(0), XR2F, CLod1});
-    Chain = Sub.getValue(1);
-  } else
-    Sub = DAG.getNode(ISD::FSUB, dl, MVT::v2f64, XR2F, CLod1);
   SDValue Result;
 
-  if (!IsStrict && Subtarget.hasSSE3() &&
+  if (Subtarget.hasSSE3() &&
       shouldUseHorizontalOp(true, DAG, Subtarget)) {
-    // FIXME: Do we need a STRICT version of FHADD?
     Result = DAG.getNode(X86ISD::FHADD, dl, MVT::v2f64, Sub, Sub);
   } else {
     SDValue Shuffle = DAG.getVectorShuffle(MVT::v2f64, dl, Sub, Sub, {1,-1});
-    if (IsStrict) {
-      Result = DAG.getNode(ISD::STRICT_FADD, dl, {MVT::v2f64, MVT::Other},
-                           {Chain, Shuffle, Sub});
-      Chain = Result.getValue(1);
-    } else
-      Result = DAG.getNode(ISD::FADD, dl, MVT::v2f64, Shuffle, Sub);
   }
   Result = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Result,
                        DAG.getIntPtrConstant(0, dl));
-  if (IsStrict)
-    return DAG.getMergeValues({Result, Chain}, dl);
-
   return Result;
 }
 
 /// 32-bit unsigned integer to float expansion.
 static SDValue LowerUINT_TO_FP_i32(SDValue Op, SelectionDAG &DAG,
                                    const X86Subtarget &Subtarget) {
   unsigned OpNo = Op.getNode()->isStrictFPOpcode() ? 1 : 0;
   SDLoc dl(Op);
if (IsStrict)
       return DAG.getNode(ISD::STRICT_SINT_TO_FP, dl, {DstVT, MVT::Other},
                          {Chain, Src});
     return DAG.getNode(ISD::SINT_TO_FP, dl, DstVT, Src);
   }
 
   if (SDValue V = LowerI64IntToFP_AVX512DQ(Op, DAG, Subtarget))
     return V;
 
-  if (SrcVT == MVT::i64 && DstVT == MVT::f64 && X86ScalarSSEf64)
+  // The transform for i64->f64 isn't correct for 0 when rounding to negative
+  // infinity. It produces -0.0, so disable under strictfp.
+  if (SrcVT == MVT::i64 && DstVT == MVT::f64 && X86ScalarSSEf64 && !IsStrict)
     return LowerUINT_TO_FP_i64(Op, DAG, Subtarget);
   if (SrcVT == MVT::i32 && X86ScalarSSEf64 && DstVT != MVT::f80)
     return LowerUINT_TO_FP_i32(Op, DAG, Subtarget);
-  if (Subtarget.is64Bit() && SrcVT == MVT::i64 && DstVT == MVT::f32)
+  if (Subtarget.is64Bit() && SrcVT == MVT::i64 &&
+      (DstVT == MVT::f32 || DstVT == MVT::f64))
     return SDValue();
 
   // Make a 64-bit buffer, and use it to build an FILD.
   SDValue StackSlot = DAG.CreateStackTemporary(MVT::i64, 8);
   int SSFI = cast<FrameIndexSDNode>(StackSlot)->getIndex();
   MachinePointerInfo MPI =
     MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SSFI);
   if (SrcVT == MVT::i32) {

llvm/test/CodeGen/SystemZ/fp-strict-conv-08.ll

%conv = call float @llvm.experimental.constrained.uitofp.f32.i64(i64 %i,
                                                metadata !"round.dynamic",
                                                metadata !"fpexcept.strict") #0
   ret float %conv
 }
 
 ; Test i64->f64.
 define double @f2(i64 %i) #0 {
 ; CHECK-LABEL: f2:
-; CHECK: ldgr
+; CHECK: cdgbr
 ; CHECK: adbr
 ; CHECK: br %r14
   %conv = call double @llvm.experimental.constrained.uitofp.f64.i64(i64 %i,
                                                metadata !"round.dynamic",
                                                metadata !"fpexcept.strict") #0
   ret double %conv
 }
 

llvm/test/CodeGen/X86/fp-intrinsics.ll

 ; X87-NEXT:    fldl {{[0-9]+}}(%esp)
 ; X87-NEXT:    wait
 ; X87-NEXT:    addl $20, %esp
 ; X87-NEXT:    .cfi_def_cfa_offset 4
 ; X87-NEXT:    retl
 ;
 ; X86-SSE-LABEL: uifdl:
 ; X86-SSE:       # %bb.0: # %entry
-; X86-SSE-NEXT:    subl $12, %esp
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 16
+; X86-SSE-NEXT:    subl $28, %esp
+; X86-SSE-NEXT:    .cfi_def_cfa_offset 32
+; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
-; X86-SSE-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; X86-SSE-NEXT:    subpd {{\.LCPI.*}}, %xmm0
-; X86-SSE-NEXT:    movapd %xmm0, %xmm1
-; X86-SSE-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
-; X86-SSE-NEXT:    addpd %xmm0, %xmm1
-; X86-SSE-NEXT:    movlpd %xmm1, (%esp)
+; X86-SSE-NEXT:    movlps %xmm0, {{[0-9]+}}(%esp)
+; X86-SSE-NEXT:    shrl $31, %eax
+; X86-SSE-NEXT:    fildll {{[0-9]+}}(%esp)
+; X86-SSE-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; X86-SSE-NEXT:    fstpl {{[0-9]+}}(%esp)
+; X86-SSE-NEXT:    wait
+; X86-SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; X86-SSE-NEXT:    movsd %xmm0, (%esp)
 ; X86-SSE-NEXT:    fldl (%esp)
 ; X86-SSE-NEXT:    wait
-; X86-SSE-NEXT:    addl $12, %esp
+; X86-SSE-NEXT:    addl $28, %esp
 ; X86-SSE-NEXT:    .cfi_def_cfa_offset 4
 ; X86-SSE-NEXT:    retl
 ;
 ; SSE-LABEL: uifdl:
 ; SSE:       # %bb.0: # %entry
-; SSE-NEXT:    movq %rdi, %xmm1
-; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; SSE-NEXT:    subpd {{.*}}(%rip), %xmm1
-; SSE-NEXT:    movapd %xmm1, %xmm0
-; SSE-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
-; SSE-NEXT:    addpd %xmm1, %xmm0
+; SSE-NEXT:    movq %rdi, %rax
+; SSE-NEXT:    shrq %rax
+; SSE-NEXT:    movl %edi, %ecx
+; SSE-NEXT:    andl $1, %ecx
+; SSE-NEXT:    orq %rax, %rcx
+; SSE-NEXT:    testq %rdi, %rdi
+; SSE-NEXT:    cmovnsq %rdi, %rcx
+; SSE-NEXT:    cvtsi2sd %rcx, %xmm0
+; SSE-NEXT:    jns .LBB48_2
+; SSE-NEXT:  # %bb.1:
+; SSE-NEXT:    addsd %xmm0, %xmm0
+; SSE-NEXT:  .LBB48_2: # %entry
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: uifdl:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vmovq %rdi, %xmm0
-; AVX1-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; AVX1-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
-; AVX1-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX1-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    movq %rdi, %rax
+; AVX1-NEXT:    shrq %rax
+; AVX1-NEXT:    movl %edi, %ecx
+; AVX1-NEXT:    andl $1, %ecx
+; AVX1-NEXT:    orq %rax, %rcx
+; AVX1-NEXT:    testq %rdi, %rdi
+; AVX1-NEXT:    cmovnsq %rdi, %rcx
+; AVX1-NEXT:    vcvtsi2sd %rcx, %xmm0, %xmm0
+; AVX1-NEXT:    jns .LBB48_2
+; AVX1-NEXT:  # %bb.1:
+; AVX1-NEXT:    vaddsd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:  .LBB48_2: # %entry
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: uifdl:
 ; AVX512:       # %bb.0: # %entry
 ; AVX512-NEXT:    vcvtusi2sd %rdi, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 entry:
   %result = call double @llvm.experimental.constrained.uitofp.f64.i64(i64 %x,

llvm/test/CodeGen/X86/fp-strict-scalar-inttofp.ll

 ; SSE-X86-LABEL: uitofp_i64tof64:
 ; SSE-X86:       # %bb.0:
 ; SSE-X86-NEXT:    pushl %ebp
 ; SSE-X86-NEXT:    .cfi_def_cfa_offset 8
 ; SSE-X86-NEXT:    .cfi_offset %ebp, -8
 ; SSE-X86-NEXT:    movl %esp, %ebp
 ; SSE-X86-NEXT:    .cfi_def_cfa_register %ebp
 ; SSE-X86-NEXT:    andl $-8, %esp
-; SSE-X86-NEXT:    subl $8, %esp
+; SSE-X86-NEXT:    subl $24, %esp
+; SSE-X86-NEXT:    movl 12(%ebp), %eax
 ; SSE-X86-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
-; SSE-X86-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; SSE-X86-NEXT:    subpd {{\.LCPI.*}}, %xmm0
-; SSE-X86-NEXT:    movapd %xmm0, %xmm1
-; SSE-X86-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
-; SSE-X86-NEXT:    addpd %xmm0, %xmm1
-; SSE-X86-NEXT:    movlpd %xmm1, (%esp)
+; SSE-X86-NEXT:    movlps %xmm0, {{[0-9]+}}(%esp)
+; SSE-X86-NEXT:    shrl $31, %eax
+; SSE-X86-NEXT:    fildll {{[0-9]+}}(%esp)
+; SSE-X86-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; SSE-X86-NEXT:    fstpl {{[0-9]+}}(%esp)
+; SSE-X86-NEXT:    wait
+; SSE-X86-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE-X86-NEXT:    movsd %xmm0, (%esp)
 ; SSE-X86-NEXT:    fldl (%esp)
 ; SSE-X86-NEXT:    wait
 ; SSE-X86-NEXT:    movl %ebp, %esp
 ; SSE-X86-NEXT:    popl %ebp
 ; SSE-X86-NEXT:    .cfi_def_cfa %esp, 4
 ; SSE-X86-NEXT:    retl
 ;
 ; SSE-X64-LABEL: uitofp_i64tof64:
 ; SSE-X64:       # %bb.0:
-; SSE-X64-NEXT:    movq %rdi, %xmm1
-; SSE-X64-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; SSE-X64-NEXT:    subpd {{.*}}(%rip), %xmm1
-; SSE-X64-NEXT:    movapd %xmm1, %xmm0
-; SSE-X64-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
-; SSE-X64-NEXT:    addpd %xmm1, %xmm0
+; SSE-X64-NEXT:    movq %rdi, %rax
+; SSE-X64-NEXT:    shrq %rax
+; SSE-X64-NEXT:    movl %edi, %ecx
+; SSE-X64-NEXT:    andl $1, %ecx
+; SSE-X64-NEXT:    orq %rax, %rcx
+; SSE-X64-NEXT:    testq %rdi, %rdi
+; SSE-X64-NEXT:    cmovnsq %rdi, %rcx
+; SSE-X64-NEXT:    cvtsi2sd %rcx, %xmm0
+; SSE-X64-NEXT:    jns .LBB18_2
+; SSE-X64-NEXT:  # %bb.1:
+; SSE-X64-NEXT:    addsd %xmm0, %xmm0
+; SSE-X64-NEXT:  .LBB18_2:
 ; SSE-X64-NEXT:    retq
 ;
 ; AVX-X86-LABEL: uitofp_i64tof64:
 ; AVX-X86:       # %bb.0:
 ; AVX-X86-NEXT:    pushl %ebp
 ; AVX-X86-NEXT:    .cfi_def_cfa_offset 8
 ; AVX-X86-NEXT:    .cfi_offset %ebp, -8
 ; AVX-X86-NEXT:    movl %esp, %ebp
 ; AVX-X86-NEXT:    .cfi_def_cfa_register %ebp
 ; AVX-X86-NEXT:    andl $-8, %esp
-; AVX-X86-NEXT:    subl $8, %esp
+; AVX-X86-NEXT:    subl $24, %esp
+; AVX-X86-NEXT:    movl 12(%ebp), %eax
 ; AVX-X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; AVX-X86-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; AVX-X86-NEXT:    vsubpd {{\.LCPI.*}}, %xmm0, %xmm0
-; AVX-X86-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX-X86-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
-; AVX-X86-NEXT:    vmovlpd %xmm0, (%esp)
+; AVX-X86-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp)
+; AVX-X86-NEXT:    shrl $31, %eax
+; AVX-X86-NEXT:    fildll {{[0-9]+}}(%esp)
+; AVX-X86-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; AVX-X86-NEXT:    fstpl {{[0-9]+}}(%esp)
+; AVX-X86-NEXT:    wait
+; AVX-X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-X86-NEXT:    vmovsd %xmm0, (%esp)
 ; AVX-X86-NEXT:    fldl (%esp)
 ; AVX-X86-NEXT:    wait
 ; AVX-X86-NEXT:    movl %ebp, %esp
 ; AVX-X86-NEXT:    popl %ebp
 ; AVX-X86-NEXT:    .cfi_def_cfa %esp, 4
 ; AVX-X86-NEXT:    retl
 ;
 ; AVX1-X64-LABEL: uitofp_i64tof64:
 ; AVX1-X64:       # %bb.0:
-; AVX1-X64-NEXT:    vmovq %rdi, %xmm0
-; AVX1-X64-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; AVX1-X64-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
-; AVX1-X64-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX1-X64-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
+; AVX1-X64-NEXT:    movq %rdi, %rax
+; AVX1-X64-NEXT:    shrq %rax
+; AVX1-X64-NEXT:    movl %edi, %ecx
+; AVX1-X64-NEXT:    andl $1, %ecx
+; AVX1-X64-NEXT:    orq %rax, %rcx
+; AVX1-X64-NEXT:    testq %rdi, %rdi
+; AVX1-X64-NEXT:    cmovnsq %rdi, %rcx
+; AVX1-X64-NEXT:    vcvtsi2sd %rcx, %xmm0, %xmm0
+; AVX1-X64-NEXT:    jns .LBB18_2
+; AVX1-X64-NEXT:  # %bb.1:
+; AVX1-X64-NEXT:    vaddsd %xmm0, %xmm0, %xmm0
+; AVX1-X64-NEXT:  .LBB18_2:
 ; AVX1-X64-NEXT:    retq
 ;
 ; AVX512-X64-LABEL: uitofp_i64tof64:
 ; AVX512-X64:       # %bb.0:
 ; AVX512-X64-NEXT:    vcvtusi2sd %rdi, %xmm0, %xmm0
 ; AVX512-X64-NEXT:    retq
 ;
 ; X87-LABEL: uitofp_i64tof64:

llvm/test/CodeGen/X86/vec-strict-inttofp-128.ll

%result = call <2 x double> @llvm.experimental.constrained.sitofp.v2f64.v2i64(<2 x i64> %x,
                                                               metadata !"round.dynamic",
                                                               metadata !"fpexcept.strict") #0
   ret <2 x double> %result
 }
 
 define <2 x double> @uitofp_v2i64_v2f64(<2 x i64> %x) #0 {
 ; SSE-32-LABEL: uitofp_v2i64_v2f64:
 ; SSE-32:       # %bb.0:
-; SSE-32-NEXT:    movdqa {{.*#+}} xmm1 = [4294967295,0,4294967295,0]
-; SSE-32-NEXT:    pand %xmm0, %xmm1
-; SSE-32-NEXT:    por {{\.LCPI.*}}, %xmm1
-; SSE-32-NEXT:    psrlq $32, %xmm0
-; SSE-32-NEXT:    por {{\.LCPI.*}}, %xmm0
-; SSE-32-NEXT:    subpd {{\.LCPI.*}}, %xmm0
-; SSE-32-NEXT:    addpd %xmm1, %xmm0
+; SSE-32-NEXT:    pushl %ebp
+; SSE-32-NEXT:    .cfi_def_cfa_offset 8
+; SSE-32-NEXT:    .cfi_offset %ebp, -8
+; SSE-32-NEXT:    movl %esp, %ebp
+; SSE-32-NEXT:    .cfi_def_cfa_register %ebp
+; SSE-32-NEXT:    andl $-8, %esp
+; SSE-32-NEXT:    subl $32, %esp
+; SSE-32-NEXT:    movq %xmm0, {{[0-9]+}}(%esp)
+; SSE-32-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; SSE-32-NEXT:    movq %xmm1, {{[0-9]+}}(%esp)
+; SSE-32-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; SSE-32-NEXT:    movd %xmm1, %eax
+; SSE-32-NEXT:    shrl $31, %eax
+; SSE-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; SSE-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; SSE-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; SSE-32-NEXT:    wait
+; SSE-32-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
+; SSE-32-NEXT:    movd %xmm0, %eax
+; SSE-32-NEXT:    shrl $31, %eax
+; SSE-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; SSE-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; SSE-32-NEXT:    fstpl (%esp)
+; SSE-32-NEXT:    wait
+; SSE-32-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE-32-NEXT:    movhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
+; SSE-32-NEXT:    movl %ebp, %esp
+; SSE-32-NEXT:    popl %ebp
+; SSE-32-NEXT:    .cfi_def_cfa %esp, 4
 ; SSE-32-NEXT:    retl
 ;
 ; SSE-64-LABEL: uitofp_v2i64_v2f64:
 ; SSE-64:       # %bb.0:
-; SSE-64-NEXT:    movdqa {{.*#+}} xmm1 = [4294967295,4294967295]
-; SSE-64-NEXT:    pand %xmm0, %xmm1
-; SSE-64-NEXT:    por {{.*}}(%rip), %xmm1
-; SSE-64-NEXT:    psrlq $32, %xmm0
-; SSE-64-NEXT:    por {{.*}}(%rip), %xmm0
-; SSE-64-NEXT:    subpd {{.*}}(%rip), %xmm0
-; SSE-64-NEXT:    addpd %xmm1, %xmm0
+; SSE-64-NEXT:    movdqa %xmm0, %xmm1
+; SSE-64-NEXT:    movq %xmm0, %rax
+; SSE-64-NEXT:    movq %rax, %rcx
+; SSE-64-NEXT:    shrq %rcx
+; SSE-64-NEXT:    movl %eax, %edx
+; SSE-64-NEXT:    andl $1, %edx
+; SSE-64-NEXT:    orq %rcx, %rdx
+; SSE-64-NEXT:    testq %rax, %rax
+; SSE-64-NEXT:    cmovnsq %rax, %rdx
+; SSE-64-NEXT:    xorps %xmm0, %xmm0
+; SSE-64-NEXT:    cvtsi2sd %rdx, %xmm0
+; SSE-64-NEXT:    jns .LBB21_2
+; SSE-64-NEXT:  # %bb.1:
+; SSE-64-NEXT:    addsd %xmm0, %xmm0
+; SSE-64-NEXT:  .LBB21_2:
+; SSE-64-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
+; SSE-64-NEXT:    movq %xmm1, %rax
+; SSE-64-NEXT:    movq %rax, %rcx
+; SSE-64-NEXT:    shrq %rcx
+; SSE-64-NEXT:    movl %eax, %edx
+; SSE-64-NEXT:    andl $1, %edx
+; SSE-64-NEXT:    orq %rcx, %rdx
+; SSE-64-NEXT:    testq %rax, %rax
+; SSE-64-NEXT:    cmovnsq %rax, %rdx
+; SSE-64-NEXT:    xorps %xmm1, %xmm1
+; SSE-64-NEXT:    cvtsi2sd %rdx, %xmm1
+; SSE-64-NEXT:    jns .LBB21_4
+; SSE-64-NEXT:  # %bb.3:
+; SSE-64-NEXT:    addsd %xmm1, %xmm1
+; SSE-64-NEXT:  .LBB21_4:
+; SSE-64-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; SSE-64-NEXT:    retq
 ;
 ; SSE41-32-LABEL: uitofp_v2i64_v2f64:
 ; SSE41-32:       # %bb.0:
-; SSE41-32-NEXT:    movdqa {{.*#+}} xmm1 = [4294967295,0,4294967295,0]
-; SSE41-32-NEXT:    pand %xmm0, %xmm1
-; SSE41-32-NEXT:    por {{\.LCPI.*}}, %xmm1
-; SSE41-32-NEXT:    psrlq $32, %xmm0
-; SSE41-32-NEXT:    por {{\.LCPI.*}}, %xmm0
-; SSE41-32-NEXT:    subpd {{\.LCPI.*}}, %xmm0
-; SSE41-32-NEXT:    addpd %xmm1, %xmm0
+; SSE41-32-NEXT:    pushl %ebp
+; SSE41-32-NEXT:    .cfi_def_cfa_offset 8
+; SSE41-32-NEXT:    .cfi_offset %ebp, -8
+; SSE41-32-NEXT:    movl %esp, %ebp
+; SSE41-32-NEXT:    .cfi_def_cfa_register %ebp
+; SSE41-32-NEXT:    andl $-8, %esp
+; SSE41-32-NEXT:    subl $32, %esp
+; SSE41-32-NEXT:    movq %xmm0, {{[0-9]+}}(%esp)
+; SSE41-32-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; SSE41-32-NEXT:    movq %xmm1, {{[0-9]+}}(%esp)
+; SSE41-32-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; SSE41-32-NEXT:    movd %xmm1, %eax
+; SSE41-32-NEXT:    shrl $31, %eax
+; SSE41-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; SSE41-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; SSE41-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; SSE41-32-NEXT:    wait
+; SSE41-32-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
+; SSE41-32-NEXT:    movd %xmm0, %eax
+; SSE41-32-NEXT:    shrl $31, %eax
+; SSE41-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; SSE41-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; SSE41-32-NEXT:    fstpl (%esp)
+; SSE41-32-NEXT:    wait
+; SSE41-32-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE41-32-NEXT:    movhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
+; SSE41-32-NEXT:    movl %ebp, %esp
+; SSE41-32-NEXT:    popl %ebp
+; SSE41-32-NEXT:    .cfi_def_cfa %esp, 4
 ; SSE41-32-NEXT:    retl
 ;
 ; SSE41-64-LABEL: uitofp_v2i64_v2f64:
 ; SSE41-64:       # %bb.0:
-; SSE41-64-NEXT:    movdqa {{.*#+}} xmm1 = [4294967295,4294967295]
-; SSE41-64-NEXT:    pand %xmm0, %xmm1
-; SSE41-64-NEXT:    por {{.*}}(%rip), %xmm1
-; SSE41-64-NEXT:    psrlq $32, %xmm0
-; SSE41-64-NEXT:    por {{.*}}(%rip), %xmm0
-; SSE41-64-NEXT:    subpd {{.*}}(%rip), %xmm0
-; SSE41-64-NEXT:    addpd %xmm1, %xmm0
+; SSE41-64-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-64-NEXT:    movq %xmm0, %rax
+; SSE41-64-NEXT:    movq %rax, %rcx
+; SSE41-64-NEXT:    shrq %rcx
+; SSE41-64-NEXT:    movl %eax, %edx
+; SSE41-64-NEXT:    andl $1, %edx
+; SSE41-64-NEXT:    orq %rcx, %rdx
+; SSE41-64-NEXT:    testq %rax, %rax
+; SSE41-64-NEXT:    cmovnsq %rax, %rdx
+; SSE41-64-NEXT:    xorps %xmm0, %xmm0
+; SSE41-64-NEXT:    cvtsi2sd %rdx, %xmm0
+; SSE41-64-NEXT:    jns .LBB21_2
+; SSE41-64-NEXT:  # %bb.1:
+; SSE41-64-NEXT:    addsd %xmm0, %xmm0
+; SSE41-64-NEXT:  .LBB21_2:
+; SSE41-64-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
+; SSE41-64-NEXT:    movq %xmm1, %rax
+; SSE41-64-NEXT:    movq %rax, %rcx
+; SSE41-64-NEXT:    shrq %rcx
+; SSE41-64-NEXT:    movl %eax, %edx
+; SSE41-64-NEXT:    andl $1, %edx
+; SSE41-64-NEXT:    orq %rcx, %rdx
+; SSE41-64-NEXT:    testq %rax, %rax
+; SSE41-64-NEXT:    cmovnsq %rax, %rdx
+; SSE41-64-NEXT:    xorps %xmm1, %xmm1
+; SSE41-64-NEXT:    cvtsi2sd %rdx, %xmm1
+; SSE41-64-NEXT:    jns .LBB21_4
+; SSE41-64-NEXT:  # %bb.3:
+; SSE41-64-NEXT:    addsd %xmm1, %xmm1
+; SSE41-64-NEXT:  .LBB21_4:
+; SSE41-64-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; SSE41-64-NEXT:    retq
 ;
-; AVX1-32-LABEL: uitofp_v2i64_v2f64:
-; AVX1-32:       # %bb.0:
-; AVX1-32-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-32-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
-; AVX1-32-NEXT:    vpor {{\.LCPI.*}}, %xmm1, %xmm1
-; AVX1-32-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX1-32-NEXT:    vpor {{\.LCPI.*}}, %xmm0, %xmm0
-; AVX1-32-NEXT:    vsubpd {{\.LCPI.*}}, %xmm0, %xmm0
-; AVX1-32-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
-; AVX1-32-NEXT:    retl
+; AVX-32-LABEL: uitofp_v2i64_v2f64:
+; AVX-32:       # %bb.0:
+; AVX-32-NEXT:    pushl %ebp
+; AVX-32-NEXT:    .cfi_def_cfa_offset 8
+; AVX-32-NEXT:    .cfi_offset %ebp, -8
+; AVX-32-NEXT:    movl %esp, %ebp
+; AVX-32-NEXT:    .cfi_def_cfa_register %ebp
+; AVX-32-NEXT:    andl $-8, %esp
+; AVX-32-NEXT:    subl $32, %esp
+; AVX-32-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX-32-NEXT:    vmovlps %xmm1, {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    vextractps $1, %xmm0, %eax
+; AVX-32-NEXT:    shrl $31, %eax
+; AVX-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; AVX-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    wait
+; AVX-32-NEXT:    vextractps $3, %xmm0, %eax
+; AVX-32-NEXT:    shrl $31, %eax
+; AVX-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; AVX-32-NEXT:    fstpl (%esp)
+; AVX-32-NEXT:    wait
+; AVX-32-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-32-NEXT:    vmovhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
+; AVX-32-NEXT:    movl %ebp, %esp
+; AVX-32-NEXT:    popl %ebp
+; AVX-32-NEXT:    .cfi_def_cfa %esp, 4
+; AVX-32-NEXT:    retl
 ;
 ; AVX1-64-LABEL: uitofp_v2i64_v2f64:
 ; AVX1-64:       # %bb.0:
-; AVX1-64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-64-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
-; AVX1-64-NEXT:    vpor {{.*}}(%rip), %xmm1, %xmm1
-; AVX1-64-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX1-64-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
-; AVX1-64-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
-; AVX1-64-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
+; AVX1-64-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX1-64-NEXT:    movq %rax, %rcx
+; AVX1-64-NEXT:    shrq %rcx
+; AVX1-64-NEXT:    movl %eax, %edx
+; AVX1-64-NEXT:    andl $1, %edx
+; AVX1-64-NEXT:    orq %rcx, %rdx
+; AVX1-64-NEXT:    testq %rax, %rax
+; AVX1-64-NEXT:    cmovnsq %rax, %rdx
+; AVX1-64-NEXT:    vcvtsi2sd %rdx, %xmm1, %xmm1
+; AVX1-64-NEXT:    jns .LBB21_2
+; AVX1-64-NEXT:  # %bb.1:
+; AVX1-64-NEXT:    vaddsd %xmm1, %xmm1, %xmm1
+; AVX1-64-NEXT:  .LBB21_2:
+; AVX1-64-NEXT:    vmovq %xmm0, %rax
+; AVX1-64-NEXT:    movq %rax, %rcx
+; AVX1-64-NEXT:    shrq %rcx
+; AVX1-64-NEXT:    movl %eax, %edx
+; AVX1-64-NEXT:    andl $1, %edx
+; AVX1-64-NEXT:    orq %rcx, %rdx
+; AVX1-64-NEXT:    testq %rax, %rax
+; AVX1-64-NEXT:    cmovnsq %rax, %rdx
+; AVX1-64-NEXT:    vcvtsi2sd %rdx, %xmm2, %xmm0
+; AVX1-64-NEXT:    jns .LBB21_4
+; AVX1-64-NEXT:  # %bb.3:
+; AVX1-64-NEXT:    vaddsd %xmm0, %xmm0, %xmm0
+; AVX1-64-NEXT:  .LBB21_4:
+; AVX1-64-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX1-64-NEXT:    retq
 ;
-; AVX512F-32-LABEL: uitofp_v2i64_v2f64:
-; AVX512F-32:       # %bb.0:
-; AVX512F-32-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-32-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX512F-32-NEXT:    vpor {{\.LCPI.*}}, %xmm1, %xmm1
-; AVX512F-32-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX512F-32-NEXT:    vpor {{\.LCPI.*}}, %xmm0, %xmm0
-; AVX512F-32-NEXT:    vsubpd {{\.LCPI.*}}, %xmm0, %xmm0
-; AVX512F-32-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
-; AVX512F-32-NEXT:    retl
-;
 ; AVX512F-64-LABEL: uitofp_v2i64_v2f64:
 ; AVX512F-64:       # %bb.0:
-; AVX512F-64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-64-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX512F-64-NEXT:    vpor {{.*}}(%rip), %xmm1, %xmm1
-; AVX512F-64-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX512F-64-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
-; AVX512F-64-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
-; AVX512F-64-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
+; AVX512F-64-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX512F-64-NEXT:    vcvtusi2sd %rax, %xmm1, %xmm1
+; AVX512F-64-NEXT:    vmovq %xmm0, %rax
+; AVX512F-64-NEXT:    vcvtusi2sd %rax, %xmm2, %xmm0
+; AVX512F-64-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX512F-64-NEXT:    retq
 ;
-; AVX512VL-32-LABEL: uitofp_v2i64_v2f64:
-; AVX512VL-32:       # %bb.0:
-; AVX512VL-32-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-32-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX512VL-32-NEXT:    vpor {{\.LCPI.*}}, %xmm1, %xmm1
-; AVX512VL-32-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX512VL-32-NEXT:    vpor {{\.LCPI.*}}, %xmm0, %xmm0
-; AVX512VL-32-NEXT:    vsubpd {{\.LCPI.*}}, %xmm0, %xmm0
-; AVX512VL-32-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
-; AVX512VL-32-NEXT:    retl
-;
 ; AVX512VL-64-LABEL: uitofp_v2i64_v2f64:
 ; AVX512VL-64:       # %bb.0:
-; AVX512VL-64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-64-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX512VL-64-NEXT:    vpor {{.*}}(%rip), %xmm1, %xmm1
-; AVX512VL-64-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX512VL-64-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
-; AVX512VL-64-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
-; AVX512VL-64-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
+; AVX512VL-64-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX512VL-64-NEXT:    vcvtusi2sd %rax, %xmm1, %xmm1
+; AVX512VL-64-NEXT:    vmovq %xmm0, %rax
+; AVX512VL-64-NEXT:    vcvtusi2sd %rax, %xmm2, %xmm0
+; AVX512VL-64-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX512VL-64-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_v2i64_v2f64:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vmovaps %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vcvtuqq2pd %zmm0, %zmm0
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512DQ-NEXT:    vzeroupper

llvm/test/CodeGen/X86/vec-strict-inttofp-256.ll

 ; AVX512DQVL-NEXT:    ret{{[l|q]}}
  %result = call <4 x double> @llvm.experimental.constrained.sitofp.v4f64.v4i64(<4 x i64> %x,
                                                               metadata !"round.dynamic",
                                                               metadata !"fpexcept.strict") #0
   ret <4 x double> %result
 }
 
 define <4 x double> @uitofp_v4i64_v4f64(<4 x i64> %x) #0 {
-; AVX1-32-LABEL: uitofp_v4i64_v4f64:
-; AVX1-32:       # %bb.0:
-; AVX1-32-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; AVX1-32-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX1-32-NEXT:    vorps {{\.LCPI.*}}, %ymm1, %ymm1
-; AVX1-32-NEXT:    vpsrlq $32, %xmm0, %xmm2
-; AVX1-32-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-32-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX1-32-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
-; AVX1-32-NEXT:    vorpd {{\.LCPI.*}}, %ymm0, %ymm0
-; AVX1-32-NEXT:    vsubpd {{\.LCPI.*}}, %ymm0, %ymm0
-; AVX1-32-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
-; AVX1-32-NEXT:    retl
+; AVX-32-LABEL: uitofp_v4i64_v4f64:
+; AVX-32:       # %bb.0:
+; AVX-32-NEXT:    pushl %ebp
+; AVX-32-NEXT:    .cfi_def_cfa_offset 8
+; AVX-32-NEXT:    .cfi_offset %ebp, -8
+; AVX-32-NEXT:    movl %esp, %ebp
+; AVX-32-NEXT:    .cfi_def_cfa_register %ebp
+; AVX-32-NEXT:    andl $-8, %esp
+; AVX-32-NEXT:    subl $64, %esp
+; AVX-32-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX-32-NEXT:    vmovlps %xmm1, {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX-32-NEXT:    vmovlps %xmm1, {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    vpermilps {{.*#+}} xmm2 = xmm1[2,3,2,3]
+; AVX-32-NEXT:    vmovlps %xmm2, {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    vextractps $1, %xmm0, %eax
+; AVX-32-NEXT:    shrl $31, %eax
+; AVX-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; AVX-32-NEXT:    fstpl (%esp)
+; AVX-32-NEXT:    wait
+; AVX-32-NEXT:    vextractps $3, %xmm0, %eax
+; AVX-32-NEXT:    shrl $31, %eax
+; AVX-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; AVX-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    wait
+; AVX-32-NEXT:    vextractps $1, %xmm1, %eax
+; AVX-32-NEXT:    shrl $31, %eax
+; AVX-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; AVX-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    wait
+; AVX-32-NEXT:    vextractps $3, %xmm1, %eax
+; AVX-32-NEXT:    shrl $31, %eax
+; AVX-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; AVX-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; AVX-32-NEXT:    wait
+; AVX-32-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-32-NEXT:    vmovhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
+; AVX-32-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX-32-NEXT:    vmovhps {{.*#+}} xmm1 = xmm1[0,1],mem[0,1]
+; AVX-32-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX-32-NEXT:    movl %ebp, %esp
+; AVX-32-NEXT:    popl %ebp
+; AVX-32-NEXT:    .cfi_def_cfa %esp, 4
+; AVX-32-NEXT:    retl
 ;
 ; AVX1-64-LABEL: uitofp_v4i64_v4f64:
 ; AVX1-64:       # %bb.0:
 ; AVX1-64-NEXT:    vxorps %xmm1, %xmm1, %xmm1
 ; AVX1-64-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX1-64-NEXT:    vorps {{.*}}(%rip), %ymm1, %ymm1
-; AVX1-64-NEXT:    vpsrlq $32, %xmm0, %xmm2
-; AVX1-64-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-64-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-64-NEXT:    vpextrq $1, %xmm2, %rax
+; AVX1-64-NEXT:    vcvtsi2sd %rax, %xmm3, %xmm3
+; AVX1-64-NEXT:    vmovq %xmm2, %rax
+; AVX1-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm2
+; AVX1-64-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX1-64-NEXT:    vpextrq $1, %xmm1, %rax
+; AVX1-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm3
+; AVX1-64-NEXT:    vmovq %xmm1, %rax
+; AVX1-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm1
+; AVX1-64-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm3[0]
+; AVX1-64-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX1-64-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-64-NEXT:    vpsrlq $32, %xmm2, %xmm2
+; AVX1-64-NEXT:    vpextrq $1, %xmm2, %rax
+; AVX1-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm3
+; AVX1-64-NEXT:    vmovq %xmm2, %rax
+; AVX1-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm2
+; AVX1-64-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
 ; AVX1-64-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX1-64-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
-; AVX1-64-NEXT:    vorpd {{.*}}(%rip), %ymm0, %ymm0
-; AVX1-64-NEXT:    vsubpd {{.*}}(%rip), %ymm0, %ymm0
-; AVX1-64-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
+; AVX1-64-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX1-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm3
+; AVX1-64-NEXT:    vmovq %xmm0, %rax
+; AVX1-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm0
+; AVX1-64-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm3[0]
+; AVX1-64-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-64-NEXT:    vmulpd {{.*}}(%rip), %ymm0, %ymm0
+; AVX1-64-NEXT:    vaddpd %ymm1, %ymm0, %ymm0
 ; AVX1-64-NEXT:    retq
 ;
-; AVX2-32-LABEL: uitofp_v4i64_v4f64:
-; AVX2-32:       # %bb.0:
-; AVX2-32-NEXT:    vpsrlq $32, %ymm0, %ymm1
-; AVX2-32-NEXT:    vpor {{\.LCPI.*}}, %ymm1, %ymm1
-; AVX2-32-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25]
-; AVX2-32-NEXT:    vsubpd %ymm2, %ymm1, %ymm1
-; AVX2-32-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
-; AVX2-32-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
-; AVX2-32-NEXT:    vpor {{\.LCPI.*}}, %ymm0, %ymm0
-; AVX2-32-NEXT:    vaddpd %ymm1, %ymm0, %ymm0
-; AVX2-32-NEXT:    retl
-;
 ; AVX2-64-LABEL: uitofp_v4i64_v4f64:
 ; AVX2-64:       # %bb.0:
-; AVX2-64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-64-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX2-64-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200]
-; AVX2-64-NEXT:    vpor %ymm2, %ymm1, %ymm1
-; AVX2-64-NEXT:    vpsrlq $32, %ymm0, %ymm0
-; AVX2-64-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4985484787499139072,4985484787499139072,4985484787499139072,4985484787499139072]
-; AVX2-64-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; AVX2-64-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25]
-; AVX2-64-NEXT:    vsubpd %ymm2, %ymm0, %ymm0
+; AVX2-64-NEXT:    vpsrlq $32, %ymm0, %ymm1
+; AVX2-64-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX2-64-NEXT:    vpextrq $1, %xmm2, %rax
+; AVX2-64-NEXT:    vcvtsi2sd %rax, %xmm3, %xmm3
+; AVX2-64-NEXT:    vmovq %xmm2, %rax
+; AVX2-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm2
+; AVX2-64-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX2-64-NEXT:    vpextrq $1, %xmm1, %rax
+; AVX2-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm3
+; AVX2-64-NEXT:    vmovq %xmm1, %rax
+; AVX2-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm1
+; AVX2-64-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm3[0]
+; AVX2-64-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX2-64-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4.294967296E+9,4.294967296E+9,4.294967296E+9,4.294967296E+9]
+; AVX2-64-NEXT:    vmulpd %ymm2, %ymm1, %ymm1
+; AVX2-64-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
+; AVX2-64-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
+; AVX2-64-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX2-64-NEXT:    vpextrq $1, %xmm2, %rax
+; AVX2-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm3
+; AVX2-64-NEXT:    vmovq %xmm2, %rax
+; AVX2-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm2
+; AVX2-64-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX2-64-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX2-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm3
+; AVX2-64-NEXT:    vmovq %xmm0, %rax
+; AVX2-64-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm0
+; AVX2-64-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm3[0]
+; AVX2-64-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX2-64-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
 ; AVX2-64-NEXT:    retq
 ;
-; AVX512F-32-LABEL: uitofp_v4i64_v4f64:
-; AVX512F-32:       # %bb.0:
-; AVX512F-32-NEXT:    vpsrlq $32, %ymm0, %ymm1
-; AVX512F-32-NEXT:    vpor {{\.LCPI.*}}, %ymm1, %ymm1
-; AVX512F-32-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25]
-; AVX512F-32-NEXT:    vsubpd %ymm2, %ymm1, %ymm1
-; AVX512F-32-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
-; AVX512F-32-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
-; AVX512F-32-NEXT:    vpor {{\.LCPI.*}}, %ymm0, %ymm0
-; AVX512F-32-NEXT:    vaddpd %ymm1, %ymm0, %ymm0
-; AVX512F-32-NEXT:    retl
-;
 ; AVX512F-64-LABEL: uitofp_v4i64_v4f64:
 ; AVX512F-64:       # %bb.0:
-; AVX512F-64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-64-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX512F-64-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200]
-; AVX512F-64-NEXT:    vpor %ymm2, %ymm1, %ymm1
-; AVX512F-64-NEXT:    vpsrlq $32, %ymm0, %ymm0
-; AVX512F-64-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4985484787499139072,4985484787499139072,4985484787499139072,4985484787499139072]
-; AVX512F-64-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; AVX512F-64-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25]
-; AVX512F-64-NEXT:    vsubpd %ymm2, %ymm0, %ymm0
-; AVX512F-64-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
+; AVX512F-64-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX512F-64-NEXT:    vpextrq $1, %xmm1, %rax
+; AVX512F-64-NEXT:    vcvtusi2sd %rax, %xmm2, %xmm2
+; AVX512F-64-NEXT:    vmovq %xmm1, %rax
+; AVX512F-64-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm1
+; AVX512F-64-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; AVX512F-64-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX512F-64-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm2
+; AVX512F-64-NEXT:    vmovq %xmm0, %rax
+; AVX512F-64-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm0
+; AVX512F-64-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0]
+; AVX512F-64-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX512F-64-NEXT:    retq
 ;
-; AVX512VL-32-LABEL: uitofp_v4i64_v4f64:
-; AVX512VL-32:       # %bb.0:
-; AVX512VL-32-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-32-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX512VL-32-NEXT:    vpor {{\.LCPI.*}}, %ymm1, %ymm1
-; AVX512VL-32-NEXT:    vpsrlq $32, %ymm0, %ymm0
-; AVX512VL-32-NEXT:    vpor {{\.LCPI.*}}, %ymm0, %ymm0
-; AVX512VL-32-NEXT:    vsubpd {{\.LCPI.*}}{1to4}, %ymm0, %ymm0
-; AVX512VL-32-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
-; AVX512VL-32-NEXT:    retl
-;
 ; AVX512VL-64-LABEL: uitofp_v4i64_v4f64:
 ; AVX512VL-64:       # %bb.0:
-; AVX512VL-64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-64-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX512VL-64-NEXT:    vporq {{.*}}(%rip){1to4}, %ymm1, %ymm1
-; AVX512VL-64-NEXT:    vpsrlq $32, %ymm0, %ymm0
-; AVX512VL-64-NEXT:    vporq {{.*}}(%rip){1to4}, %ymm0, %ymm0
-; AVX512VL-64-NEXT:    vsubpd {{.*}}(%rip){1to4}, %ymm0, %ymm0
-; AVX512VL-64-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
+; AVX512VL-64-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX512VL-64-NEXT:    vpextrq $1, %xmm1, %rax
+; AVX512VL-64-NEXT:    vcvtusi2sd %rax, %xmm2, %xmm2
+; AVX512VL-64-NEXT:    vmovq %xmm1, %rax
+; AVX512VL-64-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm1
+; AVX512VL-64-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; AVX512VL-64-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX512VL-64-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm2
+; AVX512VL-64-NEXT:    vmovq %xmm0, %rax
+; AVX512VL-64-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm0
+; AVX512VL-64-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0]
+; AVX512VL-64-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX512VL-64-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_v4i64_v4f64:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vmovaps %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vcvtuqq2pd %zmm0, %zmm0
 ; AVX512DQ-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512DQ-NEXT:    ret{{[l|q]}}

llvm/test/CodeGen/X86/vec-strict-inttofp-512.ll

%result = call <8 x double> @llvm.experimental.constrained.sitofp.v8f64.v8i64(<8 x i64> %x,
                                                               metadata !"round.dynamic",
                                                               metadata !"fpexcept.strict") #0
   ret <8 x double> %result
 }
 
 define <8 x double> @uitofp_v8i64_v8f64(<8 x i64> %x) #0 {
 ; NODQ-32-LABEL: uitofp_v8i64_v8f64:
 ; NODQ-32:       # %bb.0:
-; NODQ-32-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [0,1127219200,0,1127219200,0,1127219200,0,1127219200,0,1127219200,0,1127219200,0,1127219200,0,1127219200]
-; NODQ-32-NEXT:    vpternlogq $248, {{\.LCPI.*}}, %zmm0, %zmm1
-; NODQ-32-NEXT:    vpsrlq $32, %zmm0, %zmm0
-; NODQ-32-NEXT:    vporq {{\.LCPI.*}}, %zmm0, %zmm0
-; NODQ-32-NEXT:    vsubpd {{\.LCPI.*}}{1to8}, %zmm0, %zmm0
-; NODQ-32-NEXT:    vaddpd %zmm0, %zmm1, %zmm0
+; NODQ-32-NEXT:    pushl %ebp
+; NODQ-32-NEXT:    .cfi_def_cfa_offset 8
+; NODQ-32-NEXT:    .cfi_offset %ebp, -8
+; NODQ-32-NEXT:    movl %esp, %ebp
+; NODQ-32-NEXT:    .cfi_def_cfa_register %ebp
+; NODQ-32-NEXT:    andl $-8, %esp
+; NODQ-32-NEXT:    subl $128, %esp
+; NODQ-32-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
+; NODQ-32-NEXT:    vmovlps %xmm2, {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    vpermilps {{.*#+}} xmm1 = xmm2[2,3,2,3]
+; NODQ-32-NEXT:    vmovlps %xmm1, {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    vextractf32x4 $3, %zmm0, %xmm3
+; NODQ-32-NEXT:    vmovlps %xmm3, {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    vpermilps {{.*#+}} xmm1 = xmm3[2,3,2,3]
+; NODQ-32-NEXT:    vmovlps %xmm1, {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; NODQ-32-NEXT:    vmovlps %xmm1, {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; NODQ-32-NEXT:    vmovlps %xmm1, {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    vpermilps {{.*#+}} xmm4 = xmm1[2,3,2,3]
+; NODQ-32-NEXT:    vmovlps %xmm4, {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    vextractps $1, %xmm2, %eax
+; NODQ-32-NEXT:    shrl $31, %eax
+; NODQ-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; NODQ-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    wait
+; NODQ-32-NEXT:    vextractps $3, %xmm2, %eax
+; NODQ-32-NEXT:    shrl $31, %eax
+; NODQ-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; NODQ-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    wait
+; NODQ-32-NEXT:    vextractps $1, %xmm3, %eax
+; NODQ-32-NEXT:    shrl $31, %eax
+; NODQ-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; NODQ-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    wait
+; NODQ-32-NEXT:    vextractps $3, %xmm3, %eax
+; NODQ-32-NEXT:    shrl $31, %eax
+; NODQ-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; NODQ-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    wait
+; NODQ-32-NEXT:    vextractps $1, %xmm0, %eax
+; NODQ-32-NEXT:    shrl $31, %eax
+; NODQ-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; NODQ-32-NEXT:    fstpl (%esp)
+; NODQ-32-NEXT:    wait
+; NODQ-32-NEXT:    vextractps $3, %xmm0, %eax
+; NODQ-32-NEXT:    shrl $31, %eax
+; NODQ-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; NODQ-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    wait
+; NODQ-32-NEXT:    vextractps $1, %xmm1, %eax
+; NODQ-32-NEXT:    shrl $31, %eax
+; NODQ-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; NODQ-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    wait
+; NODQ-32-NEXT:    vextractps $3, %xmm1, %eax
+; NODQ-32-NEXT:    shrl $31, %eax
+; NODQ-32-NEXT:    fildll {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    fadds {{\.LCPI.*}}(,%eax,4)
+; NODQ-32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; NODQ-32-NEXT:    wait
+; NODQ-32-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; NODQ-32-NEXT:    vmovhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
+; NODQ-32-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; NODQ-32-NEXT:    vmovhps {{.*#+}} xmm1 = xmm1[0,1],mem[0,1]
+; NODQ-32-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; NODQ-32-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; NODQ-32-NEXT:    vmovhps {{.*#+}} xmm1 = xmm1[0,1],mem[0,1]
+; NODQ-32-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
+; NODQ-32-NEXT:    vmovhps {{.*#+}} xmm2 = xmm2[0,1],mem[0,1]
+; NODQ-32-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; NODQ-32-NEXT:    vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
+; NODQ-32-NEXT:    movl %ebp, %esp
+; NODQ-32-NEXT:    popl %ebp
+; NODQ-32-NEXT:    .cfi_def_cfa %esp, 4
 ; NODQ-32-NEXT:    retl
 ;
 ; NODQ-64-LABEL: uitofp_v8i64_v8f64:
 ; NODQ-64:       # %bb.0:
-; NODQ-64-NEXT:    vpbroadcastq {{.*#+}} zmm1 = [4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200]
-; NODQ-64-NEXT:    vpternlogq $248, {{.*}}(%rip){1to8}, %zmm0, %zmm1
-; NODQ-64-NEXT:    vpsrlq $32, %zmm0, %zmm0
-; NODQ-64-NEXT:    vporq {{.*}}(%rip){1to8}, %zmm0, %zmm0
-; NODQ-64-NEXT:    vsubpd {{.*}}(%rip){1to8}, %zmm0, %zmm0
-; NODQ-64-NEXT:    vaddpd %zmm0, %zmm1, %zmm0
+; NODQ-64-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
+; NODQ-64-NEXT:    vpextrq $1, %xmm1, %rax
+; NODQ-64-NEXT:    vcvtusi2sd %rax, %xmm2, %xmm2
+; NODQ-64-NEXT:    vmovq %xmm1, %rax
+; NODQ-64-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm1
+; NODQ-64-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; NODQ-64-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
+; NODQ-64-NEXT:    vpextrq $1, %xmm2, %rax
+; NODQ-64-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm3
+; NODQ-64-NEXT:    vmovq %xmm2, %rax
+; NODQ-64-NEXT:    vcvtusi2sd %rax, %xmm4, %xmm2
+; NODQ-64-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; NODQ-64-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
+; NODQ-64-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; NODQ-64-NEXT:    vpextrq $1, %xmm2, %rax
+; NODQ-64-NEXT:    vcvtusi2sd %rax, %xmm4, %xmm3
+; NODQ-64-NEXT:    vmovq %xmm2, %rax
+; NODQ-64-NEXT:    vcvtusi2sd %rax, %xmm4, %xmm2
+; NODQ-64-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; NODQ-64-NEXT:    vpextrq $1, %xmm0, %rax
+; NODQ-64-NEXT:    vcvtusi2sd %rax, %xmm4, %xmm3
+; NODQ-64-NEXT:    vmovq %xmm0, %rax
+; NODQ-64-NEXT:    vcvtusi2sd %rax, %xmm4, %xmm0
+; NODQ-64-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm3[0]
+; NODQ-64-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; NODQ-64-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
 ; NODQ-64-NEXT:    retq
 ;
 ; DQ-LABEL: uitofp_v8i64_v8f64:
 ; DQ:       # %bb.0:
 ; DQ-NEXT:    vcvtuqq2pd %zmm0, %zmm0
 ; DQ-NEXT:    ret{{[l|q]}}
  %result = call <8 x double> @llvm.experimental.constrained.uitofp.v8f64.v8i64(<8 x i64> %x,
                                                               metadata !"round.dynamic",

llvm/test/CodeGen/X86/vector-constrained-fp-intrinsics.ll

%result = call <1 x float>
                                                metadata !"round.dynamic",
                                                metadata !"fpexcept.strict") #0
   ret <1 x float> %result
 }
 
 define <1 x double> @constrained_vector_uitofp_v1f64_v1i64(<1 x i64> %x) #0 {
 ; CHECK-LABEL: constrained_vector_uitofp_v1f64_v1i64:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    movq %rdi, %xmm1
-; CHECK-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; CHECK-NEXT:    subpd {{.*}}(%rip), %xmm1
-; CHECK-NEXT:    movapd %xmm1, %xmm0
-; CHECK-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
-; CHECK-NEXT:    addpd %xmm1, %xmm0
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    shrq %rax
+; CHECK-NEXT:    movl %edi, %ecx
+; CHECK-NEXT:    andl $1, %ecx
+; CHECK-NEXT:    orq %rax, %rcx
+; CHECK-NEXT:    testq %rdi, %rdi
+; CHECK-NEXT:    cmovnsq %rdi, %rcx
+; CHECK-NEXT:    cvtsi2sd %rcx, %xmm0
+; CHECK-NEXT:    jns .LBB169_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    addsd %xmm0, %xmm0
+; CHECK-NEXT:  .LBB169_2: # %entry
 ; CHECK-NEXT:    retq
 ;
 ; AVX1-LABEL: constrained_vector_uitofp_v1f64_v1i64:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vmovq %rdi, %xmm0
-; AVX1-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; AVX1-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
-; AVX1-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX1-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    movq %rdi, %rax
+; AVX1-NEXT:    shrq %rax
+; AVX1-NEXT:    movl %edi, %ecx
+; AVX1-NEXT:    andl $1, %ecx
+; AVX1-NEXT:    orq %rax, %rcx
+; AVX1-NEXT:    testq %rdi, %rdi
+; AVX1-NEXT:    cmovnsq %rdi, %rcx
+; AVX1-NEXT:    vcvtsi2sd %rcx, %xmm0, %xmm0
+; AVX1-NEXT:    jns .LBB169_2
+; AVX1-NEXT:  # %bb.1:
+; AVX1-NEXT:    vaddsd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:  .LBB169_2: # %entry
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: constrained_vector_uitofp_v1f64_v1i64:
 ; AVX512:       # %bb.0: # %entry
 ; AVX512-NEXT:    vcvtusi2sd %rdi, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 entry:
   %result = call <1 x double>
%result = call <2 x float>
                                                metadata !"round.dynamic",
                                                metadata !"fpexcept.strict") #0
   ret <2 x float> %result
 }
 
 define <2 x double> @constrained_vector_uitofp_v2f64_v2i64(<2 x i64> %x) #0 {
 ; CHECK-LABEL: constrained_vector_uitofp_v2f64_v2i64:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [4294967295,4294967295]
-; CHECK-NEXT:    pand %xmm0, %xmm1
-; CHECK-NEXT:    por {{.*}}(%rip), %xmm1
-; CHECK-NEXT:    psrlq $32, %xmm0
-; CHECK-NEXT:    por {{.*}}(%rip), %xmm0
-; CHECK-NEXT:    subpd {{.*}}(%rip), %xmm0
-; CHECK-NEXT:    addpd %xmm1, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    movq %xmm0, %rax
+; CHECK-NEXT:    movq %rax, %rcx
+; CHECK-NEXT:    shrq %rcx
+; CHECK-NEXT:    movl %eax, %edx
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    orq %rcx, %rdx
+; CHECK-NEXT:    testq %rax, %rax
+; CHECK-NEXT:    cmovnsq %rax, %rdx
+; CHECK-NEXT:    xorps %xmm0, %xmm0
+; CHECK-NEXT:    cvtsi2sd %rdx, %xmm0
+; CHECK-NEXT:    jns .LBB173_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    addsd %xmm0, %xmm0
+; CHECK-NEXT:  .LBB173_2: # %entry
+; CHECK-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
+; CHECK-NEXT:    movq %xmm1, %rax
+; CHECK-NEXT:    movq %rax, %rcx
+; CHECK-NEXT:    shrq %rcx
+; CHECK-NEXT:    movl %eax, %edx
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    orq %rcx, %rdx
+; CHECK-NEXT:    testq %rax, %rax
+; CHECK-NEXT:    cmovnsq %rax, %rdx
+; CHECK-NEXT:    xorps %xmm1, %xmm1
+; CHECK-NEXT:    cvtsi2sd %rdx, %xmm1
+; CHECK-NEXT:    jns .LBB173_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    addsd %xmm1, %xmm1
+; CHECK-NEXT:  .LBB173_4: # %entry
+; CHECK-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; CHECK-NEXT:    retq
 ;
 ; AVX1-LABEL: constrained_vector_uitofp_v2f64_v2i64:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
-; AVX1-NEXT:    vpor {{.*}}(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX1-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
-; AVX1-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
-; AVX1-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX1-NEXT:    movq %rax, %rcx
+; AVX1-NEXT:    shrq %rcx
+; AVX1-NEXT:    movl %eax, %edx
+; AVX1-NEXT:    andl $1, %edx
+; AVX1-NEXT:    orq %rcx, %rdx
+; AVX1-NEXT:    testq %rax, %rax
+; AVX1-NEXT:    cmovnsq %rax, %rdx
+; AVX1-NEXT:    vcvtsi2sd %rdx, %xmm1, %xmm1
+; AVX1-NEXT:    jns .LBB173_2
+; AVX1-NEXT:  # %bb.1:
+; AVX1-NEXT:    vaddsd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:  .LBB173_2: # %entry
+; AVX1-NEXT:    vmovq %xmm0, %rax
+; AVX1-NEXT:    movq %rax, %rcx
+; AVX1-NEXT:    shrq %rcx
+; AVX1-NEXT:    movl %eax, %edx
+; AVX1-NEXT:    andl $1, %edx
+; AVX1-NEXT:    orq %rcx, %rdx
+; AVX1-NEXT:    testq %rax, %rax
+; AVX1-NEXT:    cmovnsq %rax, %rdx
+; AVX1-NEXT:    vcvtsi2sd %rdx, %xmm2, %xmm0
+; AVX1-NEXT:    jns .LBB173_4
+; AVX1-NEXT:  # %bb.3:
+; AVX1-NEXT:    vaddsd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:  .LBB173_4: # %entry
+; AVX1-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX1-NEXT:    retq
 ;
 ; AVX512F-LABEL: constrained_vector_uitofp_v2f64_v2i64:
 ; AVX512F:       # %bb.0: # %entry
-; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX512F-NEXT:    vpor {{.*}}(%rip), %xmm1, %xmm1
-; AVX512F-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX512F-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
-; AVX512F-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
-; AVX512F-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
+; AVX512F-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX512F-NEXT:    vcvtusi2sd %rax, %xmm1, %xmm1
+; AVX512F-NEXT:    vmovq %xmm0, %rax
+; AVX512F-NEXT:    vcvtusi2sd %rax, %xmm2, %xmm0
+; AVX512F-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: constrained_vector_uitofp_v2f64_v2i64:
 ; AVX512DQ:       # %bb.0: # %entry
 ; AVX512DQ-NEXT:    vmovaps %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vcvtuqq2pd %zmm0, %zmm0
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512DQ-NEXT:    vzeroupper
%result = call <3 x float>
                                                metadata !"round.dynamic",
                                                metadata !"fpexcept.strict") #0
   ret <3 x float> %result
 }
 
 define <3 x double> @constrained_vector_uitofp_v3f64_v3i64(<3 x i64> %x) #0 {
 ; CHECK-LABEL: constrained_vector_uitofp_v3f64_v3i64:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    movq %rdi, %xmm1
-; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = [1127219200,1160773632,0,0]
-; CHECK-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; CHECK-NEXT:    movapd {{.*#+}} xmm3 = [4.503599627370496E+15,1.9342813113834067E+25]
-; CHECK-NEXT:    subpd %xmm3, %xmm1
-; CHECK-NEXT:    movapd %xmm1, %xmm0
-; CHECK-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
-; CHECK-NEXT:    addpd %xmm1, %xmm0
-; CHECK-NEXT:    movq %rsi, %xmm4
-; CHECK-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; CHECK-NEXT:    subpd %xmm3, %xmm4
-; CHECK-NEXT:    movapd %xmm4, %xmm1
-; CHECK-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm4[1]
-; CHECK-NEXT:    addpd %xmm4, %xmm1
-; CHECK-NEXT:    movq %rdx, %xmm4
-; CHECK-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; CHECK-NEXT:    subpd %xmm3, %xmm4
-; CHECK-NEXT:    movapd %xmm4, %xmm2
-; CHECK-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm4[1]
-; CHECK-NEXT:    addpd %xmm4, %xmm2
-; CHECK-NEXT:    movlpd %xmm2, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    shrq %rax
+; CHECK-NEXT:    movl %edi, %ecx
+; CHECK-NEXT:    andl $1, %ecx
+; CHECK-NEXT:    orq %rax, %rcx
+; CHECK-NEXT:    testq %rdi, %rdi
+; CHECK-NEXT:    cmovnsq %rdi, %rcx
+; CHECK-NEXT:    cvtsi2sd %rcx, %xmm0
+; CHECK-NEXT:    jns .LBB177_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    addsd %xmm0, %xmm0
+; CHECK-NEXT:  .LBB177_2: # %entry
+; CHECK-NEXT:    movq %rsi, %rax
+; CHECK-NEXT:    shrq %rax
+; CHECK-NEXT:    movl %esi, %ecx
+; CHECK-NEXT:    andl $1, %ecx
+; CHECK-NEXT:    orq %rax, %rcx
+; CHECK-NEXT:    testq %rsi, %rsi
+; CHECK-NEXT:    cmovnsq %rsi, %rcx
+; CHECK-NEXT:    cvtsi2sd %rcx, %xmm1
+; CHECK-NEXT:    jns .LBB177_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    addsd %xmm1, %xmm1
+; CHECK-NEXT:  .LBB177_4: # %entry
+; CHECK-NEXT:    movq %rdx, %rax
+; CHECK-NEXT:    shrq %rax
+; CHECK-NEXT:    movl %edx, %ecx
+; CHECK-NEXT:    andl $1, %ecx
+; CHECK-NEXT:    orq %rax, %rcx
+; CHECK-NEXT:    testq %rdx, %rdx
+; CHECK-NEXT:    cmovnsq %rdx, %rcx
+; CHECK-NEXT:    cvtsi2sd %rcx, %xmm2
+; CHECK-NEXT:    jns .LBB177_6
+; CHECK-NEXT:  # %bb.5:
+; CHECK-NEXT:    addsd %xmm2, %xmm2
+; CHECK-NEXT:  .LBB177_6: # %entry
+; CHECK-NEXT:    movsd %xmm2, -{{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    fldl -{{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    wait
 ; CHECK-NEXT:    retq
 ;
 ; AVX1-LABEL: constrained_vector_uitofp_v3f64_v3i64:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vmovapd {{.*#+}} xmm1 = [1127219200,1160773632,0,0]
-; AVX1-NEXT:    vunpcklps {{.*#+}} xmm2 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX1-NEXT:    vmovapd {{.*#+}} xmm3 = [4.503599627370496E+15,1.9342813113834067E+25]
-; AVX1-NEXT:    vsubpd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm2[1,0]
-; AVX1-NEXT:    vaddpd %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm4 = xmm0[2,3,2,3]
-; AVX1-NEXT:    vunpcklps {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
-; AVX1-NEXT:    vsubpd %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm4[1,0]
-; AVX1-NEXT:    vaddpd %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm4[0]
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX1-NEXT:    movq %rax, %rcx
+; AVX1-NEXT:    shrq %rcx
+; AVX1-NEXT:    movl %eax, %edx
+; AVX1-NEXT:    andl $1, %edx
+; AVX1-NEXT:    orq %rcx, %rdx
+; AVX1-NEXT:    testq %rax, %rax
+; AVX1-NEXT:    cmovnsq %rax, %rdx
+; AVX1-NEXT:    vcvtsi2sd %rdx, %xmm1, %xmm1
+; AVX1-NEXT:    jns .LBB177_2
+; AVX1-NEXT:  # %bb.1:
+; AVX1-NEXT:    vaddsd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:  .LBB177_2: # %entry
+; AVX1-NEXT:    vmovq %xmm0, %rax
+; AVX1-NEXT:    movq %rax, %rcx
+; AVX1-NEXT:    shrq %rcx
+; AVX1-NEXT:    movl %eax, %edx
+; AVX1-NEXT:    andl $1, %edx
+; AVX1-NEXT:    orq %rcx, %rdx
+; AVX1-NEXT:    testq %rax, %rax
+; AVX1-NEXT:    cmovnsq %rax, %rdx
+; AVX1-NEXT:    vcvtsi2sd %rdx, %xmm2, %xmm2
+; AVX1-NEXT:    jns .LBB177_4
+; AVX1-NEXT:  # %bb.3:
+; AVX1-NEXT:    vaddsd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:  .LBB177_4: # %entry
+; AVX1-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm2[0],xmm1[0]
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX1-NEXT:    vsubpd %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX1-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
+; AVX1-NEXT:    vmovq %xmm0, %rax
+; AVX1-NEXT:    movq %rax, %rcx
+; AVX1-NEXT:    shrq %rcx
+; AVX1-NEXT:    movl %eax, %edx
+; AVX1-NEXT:    andl $1, %edx
+; AVX1-NEXT:    orq %rcx, %rdx
+; AVX1-NEXT:    testq %rax, %rax
+; AVX1-NEXT:    cmovnsq %rax, %rdx
+; AVX1-NEXT:    vcvtsi2sd %rdx, %xmm3, %xmm0
+; AVX1-NEXT:    jns .LBB177_6
+; AVX1-NEXT:  # %bb.5:
+; AVX1-NEXT:    vaddsd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:  .LBB177_6: # %entry
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: constrained_vector_uitofp_v3f64_v3i64:
 ; AVX512:       # %bb.0: # %entry
 ; AVX512-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX512-NEXT:    vcvtusi2sd %rax, %xmm1, %xmm1
 ; AVX512-NEXT:    vmovq %xmm0, %rax
 ; AVX512-NEXT:    vcvtusi2sd %rax, %xmm2, %xmm2
%result = call <4 x float>
                                                metadata !"round.dynamic",
                                                metadata !"fpexcept.strict") #0
   ret <4 x float> %result
 }
 
 define <4 x double> @constrained_vector_uitofp_v4f64_v4i64(<4 x i64> %x) #0 {
 ; CHECK-LABEL: constrained_vector_uitofp_v4f64_v4i64:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = [4294967295,4294967295]
-; CHECK-NEXT:    movdqa %xmm1, %xmm3
-; CHECK-NEXT:    pand %xmm2, %xmm3
-; CHECK-NEXT:    movdqa {{.*#+}} xmm4 = [4841369599423283200,4841369599423283200]
-; CHECK-NEXT:    por %xmm4, %xmm3
-; CHECK-NEXT:    psrlq $32, %xmm1
-; CHECK-NEXT:    movdqa {{.*#+}} xmm5 = [4985484787499139072,4985484787499139072]
-; CHECK-NEXT:    por %xmm5, %xmm1
-; CHECK-NEXT:    movapd {{.*#+}} xmm6 = [1.9342813118337666E+25,1.9342813118337666E+25]
-; CHECK-NEXT:    subpd %xmm6, %xmm1
-; CHECK-NEXT:    addpd %xmm3, %xmm1
-; CHECK-NEXT:    pand %xmm0, %xmm2
-; CHECK-NEXT:    por %xmm4, %xmm2
-; CHECK-NEXT:    psrlq $32, %xmm0
-; CHECK-NEXT:    por %xmm5, %xmm0
-; CHECK-NEXT:    subpd %xmm6, %xmm0
-; CHECK-NEXT:    addpd %xmm2, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-NEXT:    movq %xmm0, %rax
+; CHECK-NEXT:    movq %rax, %rcx
+; CHECK-NEXT:    shrq %rcx
+; CHECK-NEXT:    movl %eax, %edx
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    orq %rcx, %rdx
+; CHECK-NEXT:    testq %rax, %rax
+; CHECK-NEXT:    cmovnsq %rax, %rdx
+; CHECK-NEXT:    xorps %xmm0, %xmm0
+; CHECK-NEXT:    cvtsi2sd %rdx, %xmm0
+; CHECK-NEXT:    jns .LBB181_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    addsd %xmm0, %xmm0
+; CHECK-NEXT:  .LBB181_2: # %entry
+; CHECK-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,2,3]
+; CHECK-NEXT:    movq %xmm2, %rax
+; CHECK-NEXT:    movq %rax, %rcx
+; CHECK-NEXT:    shrq %rcx
+; CHECK-NEXT:    movl %eax, %edx
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    orq %rcx, %rdx
+; CHECK-NEXT:    testq %rax, %rax
+; CHECK-NEXT:    cmovnsq %rax, %rdx
+; CHECK-NEXT:    cvtsi2sd %rdx, %xmm3
+; CHECK-NEXT:    jns .LBB181_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    addsd %xmm3, %xmm3
+; CHECK-NEXT:  .LBB181_4: # %entry
+; CHECK-NEXT:    movq %xmm1, %rax
+; CHECK-NEXT:    movq %rax, %rcx
+; CHECK-NEXT:    shrq %rcx
+; CHECK-NEXT:    movl %eax, %edx
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    orq %rcx, %rdx
+; CHECK-NEXT:    testq %rax, %rax
+; CHECK-NEXT:    cmovnsq %rax, %rdx
+; CHECK-NEXT:    xorps %xmm2, %xmm2
+; CHECK-NEXT:    cvtsi2sd %rdx, %xmm2
+; CHECK-NEXT:    jns .LBB181_6
+; CHECK-NEXT:  # %bb.5:
+; CHECK-NEXT:    addsd %xmm2, %xmm2
+; CHECK-NEXT:  .LBB181_6: # %entry
+; CHECK-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm3[0]
+; CHECK-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
+; CHECK-NEXT:    movq %xmm1, %rax
+; CHECK-NEXT:    movq %rax, %rcx
+; CHECK-NEXT:    shrq %rcx
+; CHECK-NEXT:    movl %eax, %edx
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    orq %rcx, %rdx
+; CHECK-NEXT:    testq %rax, %rax
+; CHECK-NEXT:    cmovnsq %rax, %rdx
+; CHECK-NEXT:    xorps %xmm1, %xmm1
+; CHECK-NEXT:    cvtsi2sd %rdx, %xmm1
+; CHECK-NEXT:    jns .LBB181_8
+; CHECK-NEXT:  # %bb.7:
+; CHECK-NEXT:    addsd %xmm1, %xmm1
+; CHECK-NEXT:  .LBB181_8: # %entry
+; CHECK-NEXT:    unpcklpd {{.*#+}} xmm2 = xmm2[0],xmm1[0]
+; CHECK-NEXT:    movapd %xmm2, %xmm1
 ; CHECK-NEXT:    retq
 ;
 ; AVX1-LABEL: constrained_vector_uitofp_v4f64_v4i64:
 ; AVX1:       # %bb.0: # %entry
 ; AVX1-NEXT:    vxorps %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX1-NEXT:    vorps {{.*}}(%rip), %ymm1, %ymm1
-; AVX1-NEXT:    vpsrlq $32, %xmm0, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpextrq $1, %xmm2, %rax
+; AVX1-NEXT:    vcvtsi2sd %rax, %xmm3, %xmm3
+; AVX1-NEXT:    vmovq %xmm2, %rax
+; AVX1-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm2
+; AVX1-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX1-NEXT:    vpextrq $1, %xmm1, %rax
+; AVX1-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm3
+; AVX1-NEXT:    vmovq %xmm1, %rax
+; AVX1-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm1
+; AVX1-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm3[0]
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpsrlq $32, %xmm2, %xmm2
+; AVX1-NEXT:    vpextrq $1, %xmm2, %rax
+; AVX1-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm3
+; AVX1-NEXT:    vmovq %xmm2, %rax
+; AVX1-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm2
+; AVX1-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
 ; AVX1-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
-; AVX1-NEXT:    vorpd {{.*}}(%rip), %ymm0, %ymm0
-; AVX1-NEXT:    vsubpd {{.*}}(%rip), %ymm0, %ymm0
-; AVX1-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX1-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm3
+; AVX1-NEXT:    vmovq %xmm0, %rax
+; AVX1-NEXT:    vcvtsi2sd %rax, %xmm4, %xmm0
+; AVX1-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm3[0]
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vmulpd {{.*}}(%rip), %ymm0, %ymm0
+; AVX1-NEXT:    vaddpd %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX512F-LABEL: constrained_vector_uitofp_v4f64_v4i64:
 ; AVX512F:       # %bb.0: # %entry
-; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200]
-; AVX512F-NEXT:    vpor %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsrlq $32, %ymm0, %ymm0
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4985484787499139072,4985484787499139072,4985484787499139072,4985484787499139072]
-; AVX512F-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25]
-; AVX512F-NEXT:    vsubpd %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX512F-NEXT:    vpextrq $1, %xmm1, %rax
+; AVX512F-NEXT:    vcvtusi2sd %rax, %xmm2, %xmm2
+; AVX512F-NEXT:    vmovq %xmm1, %rax
+; AVX512F-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm1
+; AVX512F-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; AVX512F-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX512F-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm2
+; AVX512F-NEXT:    vmovq %xmm0, %rax
+; AVX512F-NEXT:    vcvtusi2sd %rax, %xmm3, %xmm0
+; AVX512F-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0]
+; AVX512F-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: constrained_vector_uitofp_v4f64_v4i64:
 ; AVX512DQ:       # %bb.0: # %entry
 ; AVX512DQ-NEXT:    vmovaps %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vcvtuqq2pd %zmm0, %zmm0
 ; AVX512DQ-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512DQ-NEXT:    retq