[X86FixupSetCC] Remember the preceding eflags defining instruction while we're scanning the basic block instead of looking back for it.

[craig.topper]

Summary

We're already scanning forward through the basic block. Might as
well just remember eflags defs instead of doing a bounded search
backwards later.

Based on a comment in D71841.

[craig.topper]

Remove some now dead code.

[uweigand]

Otherwise, this LGTM.

llvm/lib/Target/X86/X86FixupSetCC.cpp

49

This is now also dead.

llvm/lib/Target/X86/X86FixupSetCC.cpp

 public:
   X86FixupSetCCPass() : MachineFunctionPass(ID) {}
 
   StringRef getPassName() const override { return "X86 Fixup SetCC"; }
 
   bool runOnMachineFunction(MachineFunction &MF) override;
 
 private:
-  // Find the preceding instruction that imp-defs eflags.
-  MachineInstr *findFlagsImpDef(MachineBasicBlock *MBB,
-                                MachineBasicBlock::reverse_iterator MI);
-
-  // Return true if this is the opcode of a SetCC instruction with a register
-  // output.
-  bool isSetCCr(unsigned Opode);
-
   MachineRegisterInfo *MRI = nullptr;
   const X86InstrInfo *TII = nullptr;
 
   enum { SearchBound = 16 };
 
   static char ID;
 };
 
 char X86FixupSetCCPass::ID = 0;
 }
 
 FunctionPass *llvm::createX86FixupSetCC() { return new X86FixupSetCCPass(); }
 
-// We expect the instruction *immediately* before the setcc to imp-def
-// EFLAGS (because of scheduling glue). To make this less brittle w.r.t
-// scheduling, look backwards until we hit the beginning of the
-// basic-block, or a small bound (to avoid quadratic behavior).
-MachineInstr *
-X86FixupSetCCPass::findFlagsImpDef(MachineBasicBlock *MBB,
-                                   MachineBasicBlock::reverse_iterator MI) {
-  // FIXME: Should this be instr_rend(), and MI be reverse_instr_iterator?
-  auto MBBStart = MBB->rend();
-  for (int i = 0; (i < SearchBound) && (MI != MBBStart); ++i, ++MI)
-    if (MI->definesRegister(X86::EFLAGS))
-      return &*MI;
-
-  return nullptr;
-}
-
 bool X86FixupSetCCPass::runOnMachineFunction(MachineFunction &MF) {
   bool Changed = false;
   MRI = &MF.getRegInfo();
   TII = MF.getSubtarget<X86Subtarget>().getInstrInfo();
 
   SmallVector<MachineInstr*, 4> ToErase;
 
   for (auto &MBB : MF) {
+    MachineInstr *FlagsDefMI = nullptr;
     for (auto &MI : MBB) {
+      // Remember the most recent preceding eflags defining instruction.
+      if (MI.definesRegister(X86::EFLAGS))
+        FlagsDefMI = &MI;
+
       // Find a setcc that is used by a zext.
       // This doesn't have to be the only use, the transformation is safe
       // regardless.
       if (MI.getOpcode() != X86::SETCCr)
         continue;
 
       MachineInstr *ZExt = nullptr;
       for (auto &Use : MRI->use_instructions(MI.getOperand(0).getReg()))
         if (Use.getOpcode() == X86::MOVZX32rr8)
           ZExt = &Use;
 
       if (!ZExt)
         continue;
 
-      // Find the preceding instruction that imp-defs eflags.
-      MachineInstr *FlagsDefMI = findFlagsImpDef(
-          MI.getParent(), MachineBasicBlock::reverse_iterator(&MI));
       if (!FlagsDefMI)
         continue;
 
       // We'd like to put something that clobbers eflags directly before
       // FlagsDefMI. This can't hurt anything after FlagsDefMI, because
       // it, itself, by definition, clobbers eflags. But it may happen that
       // FlagsDefMI also *uses* eflags, in which case the transformation is
       // invalid.

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

 ; AVX512VL-32-NEXT:    vsubsd %xmm4, %xmm3, %xmm3
 ; AVX512VL-32-NEXT:    vmovsd %xmm3, {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fldl {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fisttpll {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    movl $0, %eax
 ; AVX512VL-32-NEXT:    setae %al
 ; AVX512VL-32-NEXT:    shll $31, %eax
 ; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %eax
-; AVX512VL-32-NEXT:    movl %eax, %edi
+; AVX512VL-32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; AVX512VL-32-NEXT:    vextractf32x4 $2, %zmm0, %xmm3
 ; AVX512VL-32-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm3[1,0]
 ; AVX512VL-32-NEXT:    vcomisd %xmm1, %xmm4
 ; AVX512VL-32-NEXT:    setb %al
 ; AVX512VL-32-NEXT:    kmovw %eax, %k1
 ; AVX512VL-32-NEXT:    vmovapd %xmm1, %xmm5
 ; AVX512VL-32-NEXT:    vmovsd %xmm2, %xmm5, %xmm5 {%k1}
 ; AVX512VL-32-NEXT:    vsubsd %xmm5, %xmm4, %xmm4
 ; AVX512VL-32-NEXT:    vmovsd %xmm4, {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fldl {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fisttpll {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    movl $0, %eax
 ; AVX512VL-32-NEXT:    setae %al
 ; AVX512VL-32-NEXT:    shll $31, %eax
 ; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %eax
-; AVX512VL-32-NEXT:    movl %eax, %esi
-; AVX512VL-32-NEXT:    xorl %edx, %edx
+; AVX512VL-32-NEXT:    movl %eax, %edi
 ; AVX512VL-32-NEXT:    vcomisd %xmm1, %xmm3
 ; AVX512VL-32-NEXT:    setb %al
 ; AVX512VL-32-NEXT:    kmovw %eax, %k1
 ; AVX512VL-32-NEXT:    vmovapd %xmm1, %xmm4
 ; AVX512VL-32-NEXT:    vmovsd %xmm2, %xmm4, %xmm4 {%k1}
 ; AVX512VL-32-NEXT:    vsubsd %xmm4, %xmm3, %xmm3
 ; AVX512VL-32-NEXT:    vmovsd %xmm3, {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fldl {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fisttpll {{[0-9]+}}(%esp)
-; AVX512VL-32-NEXT:    setae %dl
-; AVX512VL-32-NEXT:    shll $31, %edx
-; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %edx
+; AVX512VL-32-NEXT:    movl $0, %eax
+; AVX512VL-32-NEXT:    setae %al
+; AVX512VL-32-NEXT:    shll $31, %eax
+; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %eax
+; AVX512VL-32-NEXT:    movl %eax, %esi
 ; AVX512VL-32-NEXT:    vextractf32x4 $3, %zmm0, %xmm3
 ; AVX512VL-32-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm3[1,0]
-; AVX512VL-32-NEXT:    xorl %ecx, %ecx
+; AVX512VL-32-NEXT:    xorl %edx, %edx
 ; AVX512VL-32-NEXT:    vcomisd %xmm1, %xmm4
 ; AVX512VL-32-NEXT:    setb %al
 ; AVX512VL-32-NEXT:    kmovw %eax, %k1
 ; AVX512VL-32-NEXT:    vmovapd %xmm1, %xmm5
 ; AVX512VL-32-NEXT:    vmovsd %xmm2, %xmm5, %xmm5 {%k1}
 ; AVX512VL-32-NEXT:    vsubsd %xmm5, %xmm4, %xmm4
 ; AVX512VL-32-NEXT:    vmovsd %xmm4, (%esp)
 ; AVX512VL-32-NEXT:    fldl (%esp)
 ; AVX512VL-32-NEXT:    fisttpll (%esp)
-; AVX512VL-32-NEXT:    setae %cl
-; AVX512VL-32-NEXT:    shll $31, %ecx
-; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %ecx
+; AVX512VL-32-NEXT:    setae %dl
+; AVX512VL-32-NEXT:    shll $31, %edx
+; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %edx
 ; AVX512VL-32-NEXT:    xorl %eax, %eax
 ; AVX512VL-32-NEXT:    vcomisd %xmm1, %xmm3
-; AVX512VL-32-NEXT:    setb %bl
-; AVX512VL-32-NEXT:    kmovw %ebx, %k1
+; AVX512VL-32-NEXT:    setb %cl
+; AVX512VL-32-NEXT:    kmovw %ecx, %k1
 ; AVX512VL-32-NEXT:    vmovapd %xmm1, %xmm4
 ; AVX512VL-32-NEXT:    vmovsd %xmm2, %xmm4, %xmm4 {%k1}
 ; AVX512VL-32-NEXT:    vsubsd %xmm4, %xmm3, %xmm3
 ; AVX512VL-32-NEXT:    vmovsd %xmm3, {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fldl {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fisttpll {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    setae %al
 ; AVX512VL-32-NEXT:    shll $31, %eax
 ; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %eax
+; AVX512VL-32-NEXT:    xorl %ecx, %ecx
 ; AVX512VL-32-NEXT:    vcomisd %xmm1, %xmm0
 ; AVX512VL-32-NEXT:    setb %bl
 ; AVX512VL-32-NEXT:    kmovw %ebx, %k1
 ; AVX512VL-32-NEXT:    vmovsd %xmm2, %xmm1, %xmm1 {%k1}
 ; AVX512VL-32-NEXT:    vsubsd %xmm1, %xmm0, %xmm0
 ; AVX512VL-32-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fldl {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fisttpll {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; AVX512VL-32-NEXT:    vpinsrd $1, %eax, %xmm0, %xmm0
 ; AVX512VL-32-NEXT:    vpinsrd $2, (%esp), %xmm0, %xmm0
-; AVX512VL-32-NEXT:    vpinsrd $3, %ecx, %xmm0, %xmm0
+; AVX512VL-32-NEXT:    vpinsrd $3, %edx, %xmm0, %xmm0
 ; AVX512VL-32-NEXT:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; AVX512VL-32-NEXT:    vpinsrd $1, %edx, %xmm1, %xmm1
+; AVX512VL-32-NEXT:    vpinsrd $1, %esi, %xmm1, %xmm1
 ; AVX512VL-32-NEXT:    vpinsrd $2, {{[0-9]+}}(%esp), %xmm1, %xmm1
-; AVX512VL-32-NEXT:    vpinsrd $3, %esi, %xmm1, %xmm1
+; AVX512VL-32-NEXT:    vpinsrd $3, %edi, %xmm1, %xmm1
 ; AVX512VL-32-NEXT:    vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero
-; AVX512VL-32-NEXT:    vpinsrd $1, %edi, %xmm2, %xmm2
+; AVX512VL-32-NEXT:    vpinsrd $1, {{[-0-9]+}}(%e{{[sb]}}p), %xmm2, %xmm2 # 4-byte Folded Reload
 ; AVX512VL-32-NEXT:    vpinsrd $2, {{[0-9]+}}(%esp), %xmm2, %xmm2
 ; AVX512VL-32-NEXT:    vpinsrd $3, {{[-0-9]+}}(%e{{[sb]}}p), %xmm2, %xmm2 # 4-byte Folded Reload
-; AVX512VL-32-NEXT:    setae %al
-; AVX512VL-32-NEXT:    movzbl %al, %eax
-; AVX512VL-32-NEXT:    shll $31, %eax
-; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %eax
+; AVX512VL-32-NEXT:    setae %cl
+; AVX512VL-32-NEXT:    shll $31, %ecx
+; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %ecx
 ; AVX512VL-32-NEXT:    vmovd {{.*#+}} xmm3 = mem[0],zero,zero,zero
-; AVX512VL-32-NEXT:    vpinsrd $1, %eax, %xmm3, %xmm3
+; AVX512VL-32-NEXT:    vpinsrd $1, %ecx, %xmm3, %xmm3
 ; AVX512VL-32-NEXT:    vpinsrd $2, {{[0-9]+}}(%esp), %xmm3, %xmm3
 ; AVX512VL-32-NEXT:    vpinsrd $3, {{[-0-9]+}}(%e{{[sb]}}p), %xmm3, %xmm3 # 4-byte Folded Reload
 ; AVX512VL-32-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
 ; AVX512VL-32-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm1
 ; AVX512VL-32-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
 ; AVX512VL-32-NEXT:    leal -12(%ebp), %esp
 ; AVX512VL-32-NEXT:    popl %esi
 ; AVX512VL-32-NEXT:    popl %edi

 ; AVX512VL-32-NEXT:    vsubss %xmm4, %xmm3, %xmm3
 ; AVX512VL-32-NEXT:    vmovss %xmm3, {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    flds {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fisttpll {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    movl $0, %eax
 ; AVX512VL-32-NEXT:    setae %al
 ; AVX512VL-32-NEXT:    shll $31, %eax
 ; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %eax
-; AVX512VL-32-NEXT:    movl %eax, %edi
+; AVX512VL-32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; AVX512VL-32-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX512VL-32-NEXT:    vmovshdup {{.*#+}} xmm4 = xmm3[1,1,3,3]
 ; AVX512VL-32-NEXT:    vcomiss %xmm1, %xmm4
 ; AVX512VL-32-NEXT:    setb %al
 ; AVX512VL-32-NEXT:    kmovw %eax, %k1
 ; AVX512VL-32-NEXT:    vmovaps %xmm1, %xmm5
 ; AVX512VL-32-NEXT:    vmovss %xmm2, %xmm5, %xmm5 {%k1}
 ; AVX512VL-32-NEXT:    vsubss %xmm5, %xmm4, %xmm4
 ; AVX512VL-32-NEXT:    vmovss %xmm4, {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    flds {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fisttpll {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    movl $0, %eax
 ; AVX512VL-32-NEXT:    setae %al
 ; AVX512VL-32-NEXT:    shll $31, %eax
 ; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %eax
-; AVX512VL-32-NEXT:    movl %eax, %esi
-; AVX512VL-32-NEXT:    xorl %edx, %edx
+; AVX512VL-32-NEXT:    movl %eax, %edi
 ; AVX512VL-32-NEXT:    vcomiss %xmm1, %xmm3
 ; AVX512VL-32-NEXT:    setb %al
 ; AVX512VL-32-NEXT:    kmovw %eax, %k1
 ; AVX512VL-32-NEXT:    vmovaps %xmm1, %xmm4
 ; AVX512VL-32-NEXT:    vmovss %xmm2, %xmm4, %xmm4 {%k1}
 ; AVX512VL-32-NEXT:    vsubss %xmm4, %xmm3, %xmm4
 ; AVX512VL-32-NEXT:    vmovss %xmm4, {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    flds {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fisttpll {{[0-9]+}}(%esp)
-; AVX512VL-32-NEXT:    setae %dl
-; AVX512VL-32-NEXT:    shll $31, %edx
-; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %edx
+; AVX512VL-32-NEXT:    movl $0, %eax
+; AVX512VL-32-NEXT:    setae %al
+; AVX512VL-32-NEXT:    shll $31, %eax
+; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %eax
+; AVX512VL-32-NEXT:    movl %eax, %esi
 ; AVX512VL-32-NEXT:    vpermilps {{.*#+}} xmm4 = xmm3[3,1,2,3]
-; AVX512VL-32-NEXT:    xorl %ecx, %ecx
+; AVX512VL-32-NEXT:    xorl %edx, %edx
 ; AVX512VL-32-NEXT:    vcomiss %xmm1, %xmm4
 ; AVX512VL-32-NEXT:    setb %al
 ; AVX512VL-32-NEXT:    kmovw %eax, %k1
 ; AVX512VL-32-NEXT:    vmovaps %xmm1, %xmm5
 ; AVX512VL-32-NEXT:    vmovss %xmm2, %xmm5, %xmm5 {%k1}
 ; AVX512VL-32-NEXT:    vsubss %xmm5, %xmm4, %xmm4
 ; AVX512VL-32-NEXT:    vmovss %xmm4, (%esp)
 ; AVX512VL-32-NEXT:    flds (%esp)
 ; AVX512VL-32-NEXT:    fisttpll (%esp)
-; AVX512VL-32-NEXT:    setae %cl
-; AVX512VL-32-NEXT:    shll $31, %ecx
-; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %ecx
+; AVX512VL-32-NEXT:    setae %dl
+; AVX512VL-32-NEXT:    shll $31, %edx
+; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %edx
 ; AVX512VL-32-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm3[1,0]
 ; AVX512VL-32-NEXT:    xorl %eax, %eax
 ; AVX512VL-32-NEXT:    vcomiss %xmm1, %xmm3
-; AVX512VL-32-NEXT:    setb %bl
-; AVX512VL-32-NEXT:    kmovw %ebx, %k1
+; AVX512VL-32-NEXT:    setb %cl
+; AVX512VL-32-NEXT:    kmovw %ecx, %k1
 ; AVX512VL-32-NEXT:    vmovaps %xmm1, %xmm4
 ; AVX512VL-32-NEXT:    vmovss %xmm2, %xmm4, %xmm4 {%k1}
 ; AVX512VL-32-NEXT:    vsubss %xmm4, %xmm3, %xmm3
 ; AVX512VL-32-NEXT:    vmovss %xmm3, {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    flds {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fisttpll {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    setae %al
 ; AVX512VL-32-NEXT:    shll $31, %eax
 ; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %eax
+; AVX512VL-32-NEXT:    xorl %ecx, %ecx
 ; AVX512VL-32-NEXT:    vcomiss %xmm1, %xmm0
 ; AVX512VL-32-NEXT:    setb %bl
 ; AVX512VL-32-NEXT:    kmovw %ebx, %k1
 ; AVX512VL-32-NEXT:    vmovss %xmm2, %xmm1, %xmm1 {%k1}
 ; AVX512VL-32-NEXT:    vsubss %xmm1, %xmm0, %xmm0
 ; AVX512VL-32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    flds {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    fisttpll {{[0-9]+}}(%esp)
 ; AVX512VL-32-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; AVX512VL-32-NEXT:    vpinsrd $1, %eax, %xmm0, %xmm0
 ; AVX512VL-32-NEXT:    vpinsrd $2, (%esp), %xmm0, %xmm0
-; AVX512VL-32-NEXT:    vpinsrd $3, %ecx, %xmm0, %xmm0
+; AVX512VL-32-NEXT:    vpinsrd $3, %edx, %xmm0, %xmm0
 ; AVX512VL-32-NEXT:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; AVX512VL-32-NEXT:    vpinsrd $1, %edx, %xmm1, %xmm1
+; AVX512VL-32-NEXT:    vpinsrd $1, %esi, %xmm1, %xmm1
 ; AVX512VL-32-NEXT:    vpinsrd $2, {{[0-9]+}}(%esp), %xmm1, %xmm1
-; AVX512VL-32-NEXT:    vpinsrd $3, %esi, %xmm1, %xmm1
+; AVX512VL-32-NEXT:    vpinsrd $3, %edi, %xmm1, %xmm1
 ; AVX512VL-32-NEXT:    vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero
-; AVX512VL-32-NEXT:    vpinsrd $1, %edi, %xmm2, %xmm2
+; AVX512VL-32-NEXT:    vpinsrd $1, {{[-0-9]+}}(%e{{[sb]}}p), %xmm2, %xmm2 # 4-byte Folded Reload
 ; AVX512VL-32-NEXT:    vpinsrd $2, {{[0-9]+}}(%esp), %xmm2, %xmm2
 ; AVX512VL-32-NEXT:    vpinsrd $3, {{[-0-9]+}}(%e{{[sb]}}p), %xmm2, %xmm2 # 4-byte Folded Reload
-; AVX512VL-32-NEXT:    setae %al
-; AVX512VL-32-NEXT:    movzbl %al, %eax
-; AVX512VL-32-NEXT:    shll $31, %eax
-; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %eax
+; AVX512VL-32-NEXT:    setae %cl
+; AVX512VL-32-NEXT:    shll $31, %ecx
+; AVX512VL-32-NEXT:    xorl {{[0-9]+}}(%esp), %ecx
 ; AVX512VL-32-NEXT:    vmovd {{.*#+}} xmm3 = mem[0],zero,zero,zero
-; AVX512VL-32-NEXT:    vpinsrd $1, %eax, %xmm3, %xmm3
+; AVX512VL-32-NEXT:    vpinsrd $1, %ecx, %xmm3, %xmm3
 ; AVX512VL-32-NEXT:    vpinsrd $2, {{[0-9]+}}(%esp), %xmm3, %xmm3
 ; AVX512VL-32-NEXT:    vpinsrd $3, {{[-0-9]+}}(%e{{[sb]}}p), %xmm3, %xmm3 # 4-byte Folded Reload
 ; AVX512VL-32-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
 ; AVX512VL-32-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm1
 ; AVX512VL-32-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
 ; AVX512VL-32-NEXT:    leal -12(%ebp), %esp
 ; AVX512VL-32-NEXT:    popl %esi
 ; AVX512VL-32-NEXT:    popl %edi