; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i686-linux-gnu -mattr=avx512bf16,avx512fp16,avx512vl | FileCheck %s --check-prefixes=X86
; RUN: llc < %s -mtriple=x86_64-linux-gnu | FileCheck %s --check-prefixes=CHECK,SSE2
; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=avx512bf16,avx512vl | FileCheck %s --check-prefixes=CHECK,AVX,F16,BF16
; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=avx512bf16,avx512fp16,avx512vl | FileCheck %s --check-prefixes=CHECK,AVX,F16,FP16
; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=avxneconvert,f16c | FileCheck %s --check-prefixes=CHECK,AVX,BF16,AVXNC

define void @add(ptr %pa, ptr %pb, ptr %pc) nounwind {
; X86-LABEL: add:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NEXT:    movzwl (%edx), %edx
; X86-NEXT:    shll $16, %edx
; X86-NEXT:    vmovd %edx, %xmm0
; X86-NEXT:    movzwl (%ecx), %ecx
; X86-NEXT:    shll $16, %ecx
; X86-NEXT:    vmovd %ecx, %xmm1
; X86-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; X86-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; X86-NEXT:    vpextrw $0, %xmm0, (%eax)
; X86-NEXT:    retl
;
; SSE2-LABEL: add:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rbx
; SSE2-NEXT:    movq %rdx, %rbx
; SSE2-NEXT:    movzwl (%rsi), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm1
; SSE2-NEXT:    movzwl (%rdi), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movw %ax, (%rbx)
; SSE2-NEXT:    popq %rbx
; SSE2-NEXT:    retq
;
; F16-LABEL: add:
; F16:       # %bb.0:
; F16-NEXT:    movzwl (%rsi), %eax
; F16-NEXT:    shll $16, %eax
; F16-NEXT:    vmovd %eax, %xmm0
; F16-NEXT:    movzwl (%rdi), %eax
; F16-NEXT:    shll $16, %eax
; F16-NEXT:    vmovd %eax, %xmm1
; F16-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; F16-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; F16-NEXT:    vpextrw $0, %xmm0, (%rdx)
; F16-NEXT:    retq
;
; AVXNC-LABEL: add:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    movzwl (%rsi), %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm0
; AVXNC-NEXT:    movzwl (%rdi), %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm1
; AVXNC-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %xmm0, %xmm0
; AVXNC-NEXT:    vpextrw $0, %xmm0, (%rdx)
; AVXNC-NEXT:    retq
  %a = load bfloat, ptr %pa
  %b = load bfloat, ptr %pb
  %add = fadd bfloat %a, %b
  store bfloat %add, ptr %pc
  ret void
}

define bfloat @add2(bfloat %a, bfloat %b) nounwind {
; X86-LABEL: add2:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shll $16, %eax
; X86-NEXT:    vmovd %eax, %xmm0
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shll $16, %eax
; X86-NEXT:    vmovd %eax, %xmm1
; X86-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; X86-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; X86-NEXT:    vmovw %xmm0, %eax
; X86-NEXT:    vmovw %eax, %xmm0
; X86-NEXT:    retl
;
; SSE2-LABEL: add2:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rax
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    pextrw $0, %xmm1, %ecx
; SSE2-NEXT:    shll $16, %ecx
; SSE2-NEXT:    movd %ecx, %xmm1
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    popq %rax
; SSE2-NEXT:    retq
;
; FP16-LABEL: add2:
; FP16:       # %bb.0:
; FP16-NEXT:    vmovw %xmm0, %eax
; FP16-NEXT:    vmovw %xmm1, %ecx
; FP16-NEXT:    shll $16, %ecx
; FP16-NEXT:    vmovd %ecx, %xmm0
; FP16-NEXT:    shll $16, %eax
; FP16-NEXT:    vmovd %eax, %xmm1
; FP16-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; FP16-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; FP16-NEXT:    vmovw %xmm0, %eax
; FP16-NEXT:    vmovw %eax, %xmm0
; FP16-NEXT:    retq
;
; AVXNC-LABEL: add2:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    vpextrw $0, %xmm0, %eax
; AVXNC-NEXT:    vpextrw $0, %xmm1, %ecx
; AVXNC-NEXT:    shll $16, %ecx
; AVXNC-NEXT:    vmovd %ecx, %xmm0
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm1
; AVXNC-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %xmm0, %xmm0
; AVXNC-NEXT:    vmovd %xmm0, %eax
; AVXNC-NEXT:    vpinsrw $0, %eax, %xmm0, %xmm0
; AVXNC-NEXT:    retq
  %add = fadd bfloat %a, %b
  ret bfloat %add
}

define void @add_double(ptr %pa, ptr %pb, ptr %pc) nounwind {
; X86-LABEL: add_double:
; X86:       # %bb.0:
; X86-NEXT:    pushl %ebx
; X86-NEXT:    pushl %edi
; X86-NEXT:    pushl %esi
; X86-NEXT:    subl $16, %esp
; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; X86-NEXT:    vmovsd %xmm0, (%esp)
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovw %xmm0, %edi
; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; X86-NEXT:    vmovsd %xmm0, (%esp)
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovw %xmm0, %eax
; X86-NEXT:    shll $16, %eax
; X86-NEXT:    vmovd %eax, %xmm0
; X86-NEXT:    shll $16, %edi
; X86-NEXT:    vmovd %edi, %xmm1
; X86-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; X86-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; X86-NEXT:    vmovw %xmm0, %eax
; X86-NEXT:    shll $16, %eax
; X86-NEXT:    vmovd %eax, %xmm0
; X86-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
; X86-NEXT:    vmovsd %xmm0, (%esi)
; X86-NEXT:    addl $16, %esp
; X86-NEXT:    popl %esi
; X86-NEXT:    popl %edi
; X86-NEXT:    popl %ebx
; X86-NEXT:    retl
;
; SSE2-LABEL: add_double:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rbp
; SSE2-NEXT:    pushq %r14
; SSE2-NEXT:    pushq %rbx
; SSE2-NEXT:    movq %rdx, %rbx
; SSE2-NEXT:    movq %rsi, %r14
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm1
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movd %ebp, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    cvtss2sd %xmm0, %xmm0
; SSE2-NEXT:    movsd %xmm0, (%rbx)
; SSE2-NEXT:    popq %rbx
; SSE2-NEXT:    popq %r14
; SSE2-NEXT:    popq %rbp
; SSE2-NEXT:    retq
;
; FP16-LABEL: add_double:
; FP16:       # %bb.0:
; FP16-NEXT:    pushq %rbp
; FP16-NEXT:    pushq %r14
; FP16-NEXT:    pushq %rbx
; FP16-NEXT:    movq %rdx, %rbx
; FP16-NEXT:    movq %rsi, %r14
; FP16-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovw %xmm0, %ebp
; FP16-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovw %xmm0, %eax
; FP16-NEXT:    shll $16, %eax
; FP16-NEXT:    vmovd %eax, %xmm0
; FP16-NEXT:    shll $16, %ebp
; FP16-NEXT:    vmovd %ebp, %xmm1
; FP16-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; FP16-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; FP16-NEXT:    vmovw %xmm0, %eax
; FP16-NEXT:    shll $16, %eax
; FP16-NEXT:    vmovd %eax, %xmm0
; FP16-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
; FP16-NEXT:    vmovsd %xmm0, (%rbx)
; FP16-NEXT:    popq %rbx
; FP16-NEXT:    popq %r14
; FP16-NEXT:    popq %rbp
; FP16-NEXT:    retq
;
; AVXNC-LABEL: add_double:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    pushq %rbp
; AVXNC-NEXT:    pushq %r14
; AVXNC-NEXT:    pushq %rbx
; AVXNC-NEXT:    movq %rdx, %rbx
; AVXNC-NEXT:    movq %rsi, %r14
; AVXNC-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vpextrw $0, %xmm0, %ebp
; AVXNC-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vpextrw $0, %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm0
; AVXNC-NEXT:    shll $16, %ebp
; AVXNC-NEXT:    vmovd %ebp, %xmm1
; AVXNC-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %xmm0, %xmm0
; AVXNC-NEXT:    vmovd %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm0
; AVXNC-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
; AVXNC-NEXT:    vmovsd %xmm0, (%rbx)
; AVXNC-NEXT:    popq %rbx
; AVXNC-NEXT:    popq %r14
; AVXNC-NEXT:    popq %rbp
; AVXNC-NEXT:    retq
  %la = load double, ptr %pa
  %a = fptrunc double %la to bfloat
  %lb = load double, ptr %pb
  %b = fptrunc double %lb to bfloat
  %add = fadd bfloat %a, %b
  %dadd = fpext bfloat %add to double
  store double %dadd, ptr %pc
  ret void
}

define double @add_double2(double %da, double %db) nounwind {
; X86-LABEL: add_double2:
; X86:       # %bb.0:
; X86-NEXT:    pushl %esi
; X86-NEXT:    subl $24, %esp
; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; X86-NEXT:    vmovsd %xmm0, (%esp)
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovw %xmm0, %esi
; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; X86-NEXT:    vmovsd %xmm0, (%esp)
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovw %xmm0, %eax
; X86-NEXT:    shll $16, %eax
; X86-NEXT:    vmovd %eax, %xmm0
; X86-NEXT:    shll $16, %esi
; X86-NEXT:    vmovd %esi, %xmm1
; X86-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; X86-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; X86-NEXT:    vmovw %xmm0, %eax
; X86-NEXT:    shll $16, %eax
; X86-NEXT:    vmovd %eax, %xmm0
; X86-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
; X86-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%esp)
; X86-NEXT:    fldl {{[0-9]+}}(%esp)
; X86-NEXT:    addl $24, %esp
; X86-NEXT:    popl %esi
; X86-NEXT:    retl
;
; SSE2-LABEL: add_double2:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rbx
; SSE2-NEXT:    subq $16, %rsp
; SSE2-NEXT:    movsd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 8-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm1
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd %ebx, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    cvtss2sd %xmm0, %xmm0
; SSE2-NEXT:    addq $16, %rsp
; SSE2-NEXT:    popq %rbx
; SSE2-NEXT:    retq
;
; FP16-LABEL: add_double2:
; FP16:       # %bb.0:
; FP16-NEXT:    pushq %rbx
; FP16-NEXT:    subq $16, %rsp
; FP16-NEXT:    vmovsd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovw %xmm0, %ebx
; FP16-NEXT:    vmovsd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 8-byte Reload
; FP16-NEXT:    # xmm0 = mem[0],zero
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovw %xmm0, %eax
; FP16-NEXT:    shll $16, %eax
; FP16-NEXT:    vmovd %eax, %xmm0
; FP16-NEXT:    shll $16, %ebx
; FP16-NEXT:    vmovd %ebx, %xmm1
; FP16-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; FP16-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; FP16-NEXT:    vmovw %xmm0, %eax
; FP16-NEXT:    shll $16, %eax
; FP16-NEXT:    vmovd %eax, %xmm0
; FP16-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
; FP16-NEXT:    addq $16, %rsp
; FP16-NEXT:    popq %rbx
; FP16-NEXT:    retq
;
; AVXNC-LABEL: add_double2:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    pushq %rbx
; AVXNC-NEXT:    subq $16, %rsp
; AVXNC-NEXT:    vmovsd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vpextrw $0, %xmm0, %ebx
; AVXNC-NEXT:    vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 8-byte Folded Reload
; AVXNC-NEXT:    # xmm0 = mem[0],zero
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vpextrw $0, %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm0
; AVXNC-NEXT:    shll $16, %ebx
; AVXNC-NEXT:    vmovd %ebx, %xmm1
; AVXNC-NEXT:    vaddss %xmm0, %xmm1, %xmm0
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %xmm0, %xmm0
; AVXNC-NEXT:    vmovd %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm0
; AVXNC-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
; AVXNC-NEXT:    addq $16, %rsp
; AVXNC-NEXT:    popq %rbx
; AVXNC-NEXT:    retq
  %a = fptrunc double %da to bfloat
  %b = fptrunc double %db to bfloat
  %add = fadd bfloat %a, %b
  %dadd = fpext bfloat %add to double
  ret double %dadd
}

define void @add_constant(ptr %pa, ptr %pc) nounwind {
; X86-LABEL: add_constant:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movzwl (%ecx), %ecx
; X86-NEXT:    shll $16, %ecx
; X86-NEXT:    vmovd %ecx, %xmm0
; X86-NEXT:    vaddss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
; X86-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; X86-NEXT:    vpextrw $0, %xmm0, (%eax)
; X86-NEXT:    retl
;
; SSE2-LABEL: add_constant:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rbx
; SSE2-NEXT:    movq %rsi, %rbx
; SSE2-NEXT:    movzwl (%rdi), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    addss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movw %ax, (%rbx)
; SSE2-NEXT:    popq %rbx
; SSE2-NEXT:    retq
;
; F16-LABEL: add_constant:
; F16:       # %bb.0:
; F16-NEXT:    movzwl (%rdi), %eax
; F16-NEXT:    shll $16, %eax
; F16-NEXT:    vmovd %eax, %xmm0
; F16-NEXT:    vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; F16-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; F16-NEXT:    vpextrw $0, %xmm0, (%rsi)
; F16-NEXT:    retq
;
; AVXNC-LABEL: add_constant:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    movzwl (%rdi), %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm0
; AVXNC-NEXT:    vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %xmm0, %xmm0
; AVXNC-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVXNC-NEXT:    retq
  %a = load bfloat, ptr %pa
  %add = fadd bfloat %a, 1.0
  store bfloat %add, ptr %pc
  ret void
}

define bfloat @add_constant2(bfloat %a) nounwind {
; X86-LABEL: add_constant2:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shll $16, %eax
; X86-NEXT:    vmovd %eax, %xmm0
; X86-NEXT:    vaddss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
; X86-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; X86-NEXT:    vmovw %xmm0, %eax
; X86-NEXT:    vmovw %eax, %xmm0
; X86-NEXT:    retl
;
; SSE2-LABEL: add_constant2:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rax
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    addss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    popq %rax
; SSE2-NEXT:    retq
;
; FP16-LABEL: add_constant2:
; FP16:       # %bb.0:
; FP16-NEXT:    vmovw %xmm0, %eax
; FP16-NEXT:    shll $16, %eax
; FP16-NEXT:    vmovd %eax, %xmm0
; FP16-NEXT:    vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; FP16-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; FP16-NEXT:    vmovw %xmm0, %eax
; FP16-NEXT:    vmovw %eax, %xmm0
; FP16-NEXT:    retq
;
; AVXNC-LABEL: add_constant2:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    vpextrw $0, %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm0
; AVXNC-NEXT:    vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %xmm0, %xmm0
; AVXNC-NEXT:    vmovd %xmm0, %eax
; AVXNC-NEXT:    vpinsrw $0, %eax, %xmm0, %xmm0
; AVXNC-NEXT:    retq
  %add = fadd bfloat %a, 1.0
  ret bfloat %add
}

define void @store_constant(ptr %pc) nounwind {
; X86-LABEL: store_constant:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movw $16256, (%eax) # imm = 0x3F80
; X86-NEXT:    retl
;
; CHECK-LABEL: store_constant:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movw $16256, (%rdi) # imm = 0x3F80
; CHECK-NEXT:    retq
  store bfloat 1.0, ptr %pc
  ret void
}

define void @fold_ext_trunc(ptr %pa, ptr %pc) nounwind {
; X86-LABEL: fold_ext_trunc:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movzwl (%ecx), %ecx
; X86-NEXT:    movw %cx, (%eax)
; X86-NEXT:    retl
;
; CHECK-LABEL: fold_ext_trunc:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movzwl (%rdi), %eax
; CHECK-NEXT:    movw %ax, (%rsi)
; CHECK-NEXT:    retq
  %a = load bfloat, ptr %pa
  %ext = fpext bfloat %a to float
  %trunc = fptrunc float %ext to bfloat
  store bfloat %trunc, ptr %pc
  ret void
}

define bfloat @fold_ext_trunc2(bfloat %a) nounwind {
; X86-LABEL: fold_ext_trunc2:
; X86:       # %bb.0:
; X86-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT:    retl
;
; CHECK-LABEL: fold_ext_trunc2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    retq
  %ext = fpext bfloat %a to float
  %trunc = fptrunc float %ext to bfloat
  ret bfloat %trunc
}

define <8 x bfloat> @addv(<8 x bfloat> %a, <8 x bfloat> %b) nounwind {
; X86-LABEL: addv:
; X86:       # %bb.0:
; X86-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; X86-NEXT:    vpslld $16, %ymm1, %ymm1
; X86-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; X86-NEXT:    vpslld $16, %ymm0, %ymm0
; X86-NEXT:    vaddps %ymm1, %ymm0, %ymm0
; X86-NEXT:    vcvtneps2bf16 %ymm0, %xmm0
; X86-NEXT:    vzeroupper
; X86-NEXT:    retl
;
; SSE2-LABEL: addv:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rbp
; SSE2-NEXT:    pushq %r15
; SSE2-NEXT:    pushq %r14
; SSE2-NEXT:    pushq %r13
; SSE2-NEXT:    pushq %r12
; SSE2-NEXT:    pushq %rbx
; SSE2-NEXT:    subq $56, %rsp
; SSE2-NEXT:    movq %xmm0, %rcx
; SSE2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; SSE2-NEXT:    movq %rcx, %rax
; SSE2-NEXT:    shrq $48, %rax
; SSE2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; SSE2-NEXT:    movq %xmm1, %rdx
; SSE2-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; SSE2-NEXT:    movq %rdx, %rax
; SSE2-NEXT:    shrq $48, %rax
; SSE2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; SSE2-NEXT:    movq %rcx, %rax
; SSE2-NEXT:    shrq $32, %rax
; SSE2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; SSE2-NEXT:    movq %rdx, %rax
; SSE2-NEXT:    shrq $32, %rax
; SSE2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    movq %xmm0, %r15
; SSE2-NEXT:    movq %r15, %rbx
; SSE2-NEXT:    shrq $48, %rbx
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1,1]
; SSE2-NEXT:    movq %xmm1, %r14
; SSE2-NEXT:    movq %r14, %rbp
; SSE2-NEXT:    shrq $48, %rbp
; SSE2-NEXT:    movq %r15, %r12
; SSE2-NEXT:    shrq $32, %r12
; SSE2-NEXT:    movq %r14, %r13
; SSE2-NEXT:    shrq $32, %r13
; SSE2-NEXT:    movl %r14d, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm1
; SSE2-NEXT:    movl %r15d, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:    andl $-65536, %r14d # imm = 0xFFFF0000
; SSE2-NEXT:    movd %r14d, %xmm1
; SSE2-NEXT:    andl $-65536, %r15d # imm = 0xFFFF0000
; SSE2-NEXT:    movd %r15d, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %r15d
; SSE2-NEXT:    shll $16, %r15d
; SSE2-NEXT:    addl {{[-0-9]+}}(%r{{[sb]}}p), %r15d # 4-byte Folded Reload
; SSE2-NEXT:    shll $16, %r13d
; SSE2-NEXT:    movd %r13d, %xmm1
; SSE2-NEXT:    shll $16, %r12d
; SSE2-NEXT:    movd %r12d, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movd %ebp, %xmm1
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd %ebx, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    orl %r14d, %ebx
; SSE2-NEXT:    shlq $32, %rbx
; SSE2-NEXT:    orq %r15, %rbx
; SSE2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
; SSE2-NEXT:    movl %r15d, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm1
; SSE2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload
; SSE2-NEXT:    movl %r14d, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %ebp
; SSE2-NEXT:    movq %r15, %rax
; SSE2-NEXT:    andl $-65536, %eax # imm = 0xFFFF0000
; SSE2-NEXT:    movd %eax, %xmm1
; SSE2-NEXT:    movq %r14, %rax
; SSE2-NEXT:    andl $-65536, %eax # imm = 0xFFFF0000
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %r14d
; SSE2-NEXT:    shll $16, %r14d
; SSE2-NEXT:    orl %ebp, %r14d
; SSE2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm1
; SSE2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %ebp
; SSE2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm1
; SSE2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    addss %xmm1, %xmm0
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    orl %ebp, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm0
; SSE2-NEXT:    movq %rbx, %xmm1
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-NEXT:    addq $56, %rsp
; SSE2-NEXT:    popq %rbx
; SSE2-NEXT:    popq %r12
; SSE2-NEXT:    popq %r13
; SSE2-NEXT:    popq %r14
; SSE2-NEXT:    popq %r15
; SSE2-NEXT:    popq %rbp
; SSE2-NEXT:    retq
;
; F16-LABEL: addv:
; F16:       # %bb.0:
; F16-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; F16-NEXT:    vpslld $16, %ymm1, %ymm1
; F16-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; F16-NEXT:    vpslld $16, %ymm0, %ymm0
; F16-NEXT:    vaddps %ymm1, %ymm0, %ymm0
; F16-NEXT:    vcvtneps2bf16 %ymm0, %xmm0
; F16-NEXT:    vzeroupper
; F16-NEXT:    retq
;
; AVXNC-LABEL: addv:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVXNC-NEXT:    vpslld $16, %ymm1, %ymm1
; AVXNC-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVXNC-NEXT:    vpslld $16, %ymm0, %ymm0
; AVXNC-NEXT:    vaddps %ymm1, %ymm0, %ymm0
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %ymm0, %xmm0
; AVXNC-NEXT:    vzeroupper
; AVXNC-NEXT:    retq
  %add = fadd <8 x bfloat> %a, %b
  ret <8 x bfloat> %add
}

define <2 x bfloat> @pr62997(bfloat %a, bfloat %b) {
; X86-LABEL: pr62997:
; X86:       # %bb.0:
; X86-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT:    vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; X86-NEXT:    retl
;
; SSE2-LABEL: pr62997:
; SSE2:       # %bb.0:
; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; SSE2-NEXT:    retq
;
; BF16-LABEL: pr62997:
; BF16:       # %bb.0:
; BF16-NEXT:    vpextrw $0, %xmm0, %eax
; BF16-NEXT:    vpextrw $0, %xmm1, %ecx
; BF16-NEXT:    vmovd %eax, %xmm0
; BF16-NEXT:    vpinsrw $1, %ecx, %xmm0, %xmm0
; BF16-NEXT:    retq
;
; FP16-LABEL: pr62997:
; FP16:       # %bb.0:
; FP16-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; FP16-NEXT:    retq
  %1 = insertelement <2 x bfloat> undef, bfloat %a, i64 0
  %2 = insertelement <2 x bfloat> %1, bfloat %b, i64 1
  ret <2 x bfloat> %2
}

define <32 x bfloat> @pr63017() {
; X86-LABEL: pr63017:
; X86:       # %bb.0:
; X86-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; X86-NEXT:    retl
;
; SSE2-LABEL: pr63017:
; SSE2:       # %bb.0:
; SSE2-NEXT:    xorps %xmm0, %xmm0
; SSE2-NEXT:    xorps %xmm1, %xmm1
; SSE2-NEXT:    xorps %xmm2, %xmm2
; SSE2-NEXT:    xorps %xmm3, %xmm3
; SSE2-NEXT:    retq
;
; F16-LABEL: pr63017:
; F16:       # %bb.0:
; F16-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; F16-NEXT:    retq
;
; AVXNC-LABEL: pr63017:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVXNC-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; AVXNC-NEXT:    retq
  ret <32 x bfloat> zeroinitializer
}

define <32 x bfloat> @pr63017_2() nounwind {
; X86-LABEL: pr63017_2:
; X86:       # %bb.0:
; X86-NEXT:    vpbroadcastw {{.*#+}} zmm0 = [-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0]
; X86-NEXT:    vmovdqu16 (%eax), %zmm0 {%k1}
; X86-NEXT:    retl
;
; SSE2-LABEL: pr63017_2:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %r14
; SSE2-NEXT:    pushq %rbx
; SSE2-NEXT:    subq $200, %rsp
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_1
; SSE2-NEXT:  # %bb.2: # %cond.load
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:    movd {{.*#+}} xmm0 = [-1.0E+0,0.0E+0,0.0E+0,0.0E+0]
; SSE2-NEXT:    movdqa %xmm0, %xmm1
; SSE2-NEXT:    jmp .LBB12_3
; SSE2-NEXT:  .LBB12_1:
; SSE2-NEXT:    movd {{.*#+}} xmm1 = [-1.0E+0,0.0E+0,0.0E+0,0.0E+0]
; SSE2-NEXT:    movdqa %xmm1, %xmm0
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:  .LBB12_3: # %else
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_5
; SSE2-NEXT:  # %bb.4: # %cond.load1
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:  .LBB12_5: # %else2
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_6
; SSE2-NEXT:  # %bb.7: # %cond.load4
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movdqa %xmm1, %xmm14
; SSE2-NEXT:    movdqa %xmm1, %xmm15
; SSE2-NEXT:    movdqa %xmm1, %xmm12
; SSE2-NEXT:    movdqa %xmm1, %xmm13
; SSE2-NEXT:    movdqa %xmm1, %xmm10
; SSE2-NEXT:    movdqa %xmm1, %xmm11
; SSE2-NEXT:    movdqa %xmm1, %xmm8
; SSE2-NEXT:    movdqa %xmm1, %xmm9
; SSE2-NEXT:    movdqa %xmm1, %xmm6
; SSE2-NEXT:    movdqa %xmm1, %xmm7
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    movdqa %xmm1, %xmm5
; SSE2-NEXT:    movdqa %xmm1, %xmm2
; SSE2-NEXT:    movdqa %xmm1, %xmm3
; SSE2-NEXT:    movd %eax, %xmm1
; SSE2-NEXT:    jmp .LBB12_8
; SSE2-NEXT:  .LBB12_6:
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movdqa %xmm1, %xmm14
; SSE2-NEXT:    movdqa %xmm1, %xmm15
; SSE2-NEXT:    movdqa %xmm1, %xmm12
; SSE2-NEXT:    movdqa %xmm1, %xmm13
; SSE2-NEXT:    movdqa %xmm1, %xmm10
; SSE2-NEXT:    movdqa %xmm1, %xmm11
; SSE2-NEXT:    movdqa %xmm1, %xmm8
; SSE2-NEXT:    movdqa %xmm1, %xmm9
; SSE2-NEXT:    movdqa %xmm1, %xmm6
; SSE2-NEXT:    movdqa %xmm1, %xmm7
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    movdqa %xmm1, %xmm5
; SSE2-NEXT:    movdqa %xmm1, %xmm2
; SSE2-NEXT:    movdqa %xmm1, %xmm3
; SSE2-NEXT:  .LBB12_8: # %else5
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_10
; SSE2-NEXT:  # %bb.9: # %cond.load7
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_10: # %else8
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_12
; SSE2-NEXT:  # %bb.11: # %cond.load10
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_12: # %else11
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_14
; SSE2-NEXT:  # %bb.13: # %cond.load13
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_14: # %else14
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_16
; SSE2-NEXT:  # %bb.15: # %cond.load16
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_16: # %else17
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_18
; SSE2-NEXT:  # %bb.17: # %cond.load19
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_18: # %else20
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_20
; SSE2-NEXT:  # %bb.19: # %cond.load22
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_20: # %else23
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_22
; SSE2-NEXT:  # %bb.21: # %cond.load25
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_22: # %else26
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_24
; SSE2-NEXT:  # %bb.23: # %cond.load28
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_24: # %else29
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_26
; SSE2-NEXT:  # %bb.25: # %cond.load31
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_26: # %else32
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_28
; SSE2-NEXT:  # %bb.27: # %cond.load34
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_28: # %else35
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_30
; SSE2-NEXT:  # %bb.29: # %cond.load37
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_30: # %else38
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_32
; SSE2-NEXT:  # %bb.31: # %cond.load40
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_32: # %else41
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_34
; SSE2-NEXT:  # %bb.33: # %cond.load43
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_34: # %else44
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_36
; SSE2-NEXT:  # %bb.35: # %cond.load46
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_36: # %else47
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_38
; SSE2-NEXT:  # %bb.37: # %cond.load49
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:  .LBB12_38: # %else50
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_40
; SSE2-NEXT:  # %bb.39: # %cond.load52
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm14
; SSE2-NEXT:  .LBB12_40: # %else53
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_42
; SSE2-NEXT:  # %bb.41: # %cond.load55
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm15
; SSE2-NEXT:  .LBB12_42: # %else56
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_44
; SSE2-NEXT:  # %bb.43: # %cond.load58
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm12
; SSE2-NEXT:  .LBB12_44: # %else59
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_46
; SSE2-NEXT:  # %bb.45: # %cond.load61
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm13
; SSE2-NEXT:  .LBB12_46: # %else62
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_48
; SSE2-NEXT:  # %bb.47: # %cond.load64
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm10
; SSE2-NEXT:  .LBB12_48: # %else65
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_50
; SSE2-NEXT:  # %bb.49: # %cond.load67
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm11
; SSE2-NEXT:  .LBB12_50: # %else68
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_52
; SSE2-NEXT:  # %bb.51: # %cond.load70
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm8
; SSE2-NEXT:  .LBB12_52: # %else71
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_54
; SSE2-NEXT:  # %bb.53: # %cond.load73
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm9
; SSE2-NEXT:  .LBB12_54: # %else74
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_56
; SSE2-NEXT:  # %bb.55: # %cond.load76
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm6
; SSE2-NEXT:  .LBB12_56: # %else77
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_58
; SSE2-NEXT:  # %bb.57: # %cond.load79
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm7
; SSE2-NEXT:  .LBB12_58: # %else80
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_60
; SSE2-NEXT:  # %bb.59: # %cond.load82
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm4
; SSE2-NEXT:  .LBB12_60: # %else83
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_62
; SSE2-NEXT:  # %bb.61: # %cond.load85
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm5
; SSE2-NEXT:  .LBB12_62: # %else86
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    jne .LBB12_64
; SSE2-NEXT:  # %bb.63: # %cond.load88
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm2
; SSE2-NEXT:  .LBB12_64: # %else89
; SSE2-NEXT:    xorl %eax, %eax
; SSE2-NEXT:    testb %al, %al
; SSE2-NEXT:    movd %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    movd %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:    jne .LBB12_65
; SSE2-NEXT:  # %bb.66: # %cond.load91
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SSE2-NEXT:    jmp .LBB12_67
; SSE2-NEXT:  .LBB12_65:
; SSE2-NEXT:    movd %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; SSE2-NEXT:  .LBB12_67: # %else92
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebx, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm0
; SSE2-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebx, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm0
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE2-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebx, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm0
; SSE2-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebx, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm0
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE2-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebx, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm0
; SSE2-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebx, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm0
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE2-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebx, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm0
; SSE2-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
; SSE2-NEXT:    # xmm0 = mem[0],zero,zero,zero
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebx, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm0
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm0[0]
; SSE2-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE2-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE2-NEXT:    addq $200, %rsp
; SSE2-NEXT:    popq %rbx
; SSE2-NEXT:    popq %r14
; SSE2-NEXT:    retq
;
; FP16-LABEL: pr63017_2:
; FP16:       # %bb.0:
; FP16-NEXT:    vpbroadcastw {{.*#+}} zmm0 = [-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0,-1.875E+0]
; FP16-NEXT:    vmovdqu16 (%rax), %zmm0 {%k1}
; FP16-NEXT:    retq
;
; AVXNC-LABEL: pr63017_2:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    vpbroadcastw {{.*#+}} ymm0 = [49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024]
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    vmovdqa %ymm0, %ymm1
; AVXNC-NEXT:    jne .LBB12_2
; AVXNC-NEXT:  # %bb.1: # %cond.load
; AVXNC-NEXT:    vpbroadcastw {{.*#+}} ymm1 = [49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024]
; AVXNC-NEXT:    vpbroadcastw {{.*#+}} ymm0 = [49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024]
; AVXNC-NEXT:    vpinsrw $0, (%rax), %xmm0, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVXNC-NEXT:  .LBB12_2: # %else
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_4
; AVXNC-NEXT:  # %bb.3: # %cond.load1
; AVXNC-NEXT:    vpinsrw $1, (%rax), %xmm0, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVXNC-NEXT:  .LBB12_4: # %else2
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_6
; AVXNC-NEXT:  # %bb.5: # %cond.load4
; AVXNC-NEXT:    vpinsrw $2, (%rax), %xmm0, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVXNC-NEXT:  .LBB12_6: # %else5
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_8
; AVXNC-NEXT:  # %bb.7: # %cond.load7
; AVXNC-NEXT:    vpinsrw $3, (%rax), %xmm0, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVXNC-NEXT:  .LBB12_8: # %else8
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_10
; AVXNC-NEXT:  # %bb.9: # %cond.load10
; AVXNC-NEXT:    vpinsrw $4, (%rax), %xmm0, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVXNC-NEXT:  .LBB12_10: # %else11
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_12
; AVXNC-NEXT:  # %bb.11: # %cond.load13
; AVXNC-NEXT:    vpinsrw $5, (%rax), %xmm0, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVXNC-NEXT:  .LBB12_12: # %else14
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_14
; AVXNC-NEXT:  # %bb.13: # %cond.load16
; AVXNC-NEXT:    vpinsrw $6, (%rax), %xmm0, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVXNC-NEXT:  .LBB12_14: # %else17
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_16
; AVXNC-NEXT:  # %bb.15: # %cond.load19
; AVXNC-NEXT:    vpinsrw $7, (%rax), %xmm0, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVXNC-NEXT:  .LBB12_16: # %else20
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_18
; AVXNC-NEXT:  # %bb.17: # %cond.load22
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_18: # %else23
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_20
; AVXNC-NEXT:  # %bb.19: # %cond.load25
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm0[0],ymm2[1],ymm0[2,3,4,5,6,7,8],ymm2[9],ymm0[10,11,12,13,14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_20: # %else26
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_22
; AVXNC-NEXT:  # %bb.21: # %cond.load28
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm0[0,1],ymm2[2],ymm0[3,4,5,6,7,8,9],ymm2[10],ymm0[11,12,13,14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_22: # %else29
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_24
; AVXNC-NEXT:  # %bb.23: # %cond.load31
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm0[0,1,2],ymm2[3],ymm0[4,5,6,7,8,9,10],ymm2[11],ymm0[12,13,14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_24: # %else32
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_26
; AVXNC-NEXT:  # %bb.25: # %cond.load34
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm0[0,1,2,3],ymm2[4],ymm0[5,6,7,8,9,10,11],ymm2[12],ymm0[13,14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_26: # %else35
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_28
; AVXNC-NEXT:  # %bb.27: # %cond.load37
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm0[0,1,2,3,4],ymm2[5],ymm0[6,7,8,9,10,11,12],ymm2[13],ymm0[14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_28: # %else38
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_30
; AVXNC-NEXT:  # %bb.29: # %cond.load40
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm0[0,1,2,3,4,5],ymm2[6],ymm0[7,8,9,10,11,12,13],ymm2[14],ymm0[15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_30: # %else41
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_32
; AVXNC-NEXT:  # %bb.31: # %cond.load43
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm0[0,1,2,3,4,5,6],ymm2[7],ymm0[8,9,10,11,12,13,14],ymm2[15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_32: # %else44
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_34
; AVXNC-NEXT:  # %bb.33: # %cond.load46
; AVXNC-NEXT:    vpinsrw $0, (%rax), %xmm1, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVXNC-NEXT:  .LBB12_34: # %else47
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_36
; AVXNC-NEXT:  # %bb.35: # %cond.load49
; AVXNC-NEXT:    vpinsrw $1, (%rax), %xmm1, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVXNC-NEXT:  .LBB12_36: # %else50
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_38
; AVXNC-NEXT:  # %bb.37: # %cond.load52
; AVXNC-NEXT:    vpinsrw $2, (%rax), %xmm1, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVXNC-NEXT:  .LBB12_38: # %else53
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_40
; AVXNC-NEXT:  # %bb.39: # %cond.load55
; AVXNC-NEXT:    vpinsrw $3, (%rax), %xmm1, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVXNC-NEXT:  .LBB12_40: # %else56
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_42
; AVXNC-NEXT:  # %bb.41: # %cond.load58
; AVXNC-NEXT:    vpinsrw $4, (%rax), %xmm1, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVXNC-NEXT:  .LBB12_42: # %else59
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_44
; AVXNC-NEXT:  # %bb.43: # %cond.load61
; AVXNC-NEXT:    vpinsrw $5, (%rax), %xmm1, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVXNC-NEXT:  .LBB12_44: # %else62
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_46
; AVXNC-NEXT:  # %bb.45: # %cond.load64
; AVXNC-NEXT:    vpinsrw $6, (%rax), %xmm1, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVXNC-NEXT:  .LBB12_46: # %else65
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_48
; AVXNC-NEXT:  # %bb.47: # %cond.load67
; AVXNC-NEXT:    vpinsrw $7, (%rax), %xmm1, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVXNC-NEXT:  .LBB12_48: # %else68
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_50
; AVXNC-NEXT:  # %bb.49: # %cond.load70
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_50: # %else71
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_52
; AVXNC-NEXT:  # %bb.51: # %cond.load73
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm1[0],ymm2[1],ymm1[2,3,4,5,6,7,8],ymm2[9],ymm1[10,11,12,13,14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_52: # %else74
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_54
; AVXNC-NEXT:  # %bb.53: # %cond.load76
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm1[0,1],ymm2[2],ymm1[3,4,5,6,7,8,9],ymm2[10],ymm1[11,12,13,14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_54: # %else77
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_56
; AVXNC-NEXT:  # %bb.55: # %cond.load79
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm1[0,1,2],ymm2[3],ymm1[4,5,6,7,8,9,10],ymm2[11],ymm1[12,13,14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_56: # %else80
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_58
; AVXNC-NEXT:  # %bb.57: # %cond.load82
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4],ymm1[5,6,7,8,9,10,11],ymm2[12],ymm1[13,14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_58: # %else83
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_60
; AVXNC-NEXT:  # %bb.59: # %cond.load85
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6,7,8,9,10,11,12],ymm2[13],ymm1[14,15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_60: # %else86
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_62
; AVXNC-NEXT:  # %bb.61: # %cond.load88
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7,8,9,10,11,12,13],ymm2[14],ymm1[15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_62: # %else89
; AVXNC-NEXT:    xorl %eax, %eax
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB12_64
; AVXNC-NEXT:  # %bb.63: # %cond.load91
; AVXNC-NEXT:    vpbroadcastw (%rax), %ymm2
; AVXNC-NEXT:    vpblendw {{.*#+}} ymm2 = ymm1[0,1,2,3,4,5,6],ymm2[7],ymm1[8,9,10,11,12,13,14],ymm2[15]
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
; AVXNC-NEXT:  .LBB12_64: # %else92
; AVXNC-NEXT:    retq
  %1 = call <32 x bfloat> @llvm.masked.load.v32bf16.p0(ptr poison, i32 2, <32 x i1> poison, <32 x bfloat> <bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80>)
  ret <32 x bfloat> %1
}

define <32 x bfloat> @pr62997_3(<32 x bfloat> %0, bfloat %1) {
; X86-LABEL: pr62997_3:
; X86:       # %bb.0:
; X86-NEXT:    vpinsrw $1, {{[0-9]+}}(%esp), %xmm0, %xmm1
; X86-NEXT:    vinserti32x4 $0, %xmm1, %zmm0, %zmm0
; X86-NEXT:    retl
;
; SSE2-LABEL: pr62997_3:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq %xmm0, %rax
; SSE2-NEXT:    movabsq $-4294967296, %rcx # imm = 0xFFFFFFFF00000000
; SSE2-NEXT:    andq %rax, %rcx
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    pextrw $0, %xmm4, %edx
; SSE2-NEXT:    shll $16, %edx
; SSE2-NEXT:    orl %eax, %edx
; SSE2-NEXT:    orq %rcx, %rdx
; SSE2-NEXT:    movq %rdx, %xmm4
; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
; SSE2-NEXT:    retq
;
; FP16-LABEL: pr62997_3:
; FP16:       # %bb.0:
; FP16-NEXT:    vmovw %xmm1, %eax
; FP16-NEXT:    vpinsrw $1, %eax, %xmm0, %xmm1
; FP16-NEXT:    vinserti32x4 $0, %xmm1, %zmm0, %zmm0
; FP16-NEXT:    retq
;
; AVXNC-LABEL: pr62997_3:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    vpextrw $0, %xmm2, %eax
; AVXNC-NEXT:    vpinsrw $1, %eax, %xmm0, %xmm2
; AVXNC-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVXNC-NEXT:    retq
  %3 = insertelement <32 x bfloat> %0, bfloat %1, i64 1
  ret <32 x bfloat> %3
}

declare <32 x bfloat> @llvm.masked.load.v32bf16.p0(ptr, i32, <32 x i1>, <32 x bfloat>)

define <4 x float> @pr64460_1(<4 x bfloat> %a) {
; X86-LABEL: pr64460_1:
; X86:       # %bb.0:
; X86-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; X86-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; X86-NEXT:    retl
;
; SSE2-LABEL: pr64460_1:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pextrw $1, %xmm0, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm2
; SSE2-NEXT:    movd %xmm0, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm1
; SSE2-NEXT:    pextrw $3, %xmm0, %eax
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm2
; SSE2-NEXT:    movd %xmm0, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE2-NEXT:    movdqa %xmm1, %xmm0
; SSE2-NEXT:    retq
;
; AVX-LABEL: pr64460_1:
; AVX:       # %bb.0:
; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; AVX-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; AVX-NEXT:    retq
  %b = fpext <4 x bfloat> %a to <4 x float>
  ret <4 x float> %b
}

define <8 x float> @pr64460_2(<8 x bfloat> %a) {
; X86-LABEL: pr64460_2:
; X86:       # %bb.0:
; X86-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; X86-NEXT:    vpslld $16, %ymm0, %ymm0
; X86-NEXT:    retl
;
; SSE2-LABEL: pr64460_2:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq %xmm0, %rdx
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    movq %xmm0, %rcx
; SSE2-NEXT:    movq %rcx, %rax
; SSE2-NEXT:    shrq $32, %rax
; SSE2-NEXT:    movq %rdx, %rsi
; SSE2-NEXT:    shrq $32, %rsi
; SSE2-NEXT:    movl %edx, %edi
; SSE2-NEXT:    andl $-65536, %edi # imm = 0xFFFF0000
; SSE2-NEXT:    movd %edi, %xmm1
; SSE2-NEXT:    movl %edx, %edi
; SSE2-NEXT:    shll $16, %edi
; SSE2-NEXT:    movd %edi, %xmm0
; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE2-NEXT:    shrq $48, %rdx
; SSE2-NEXT:    shll $16, %edx
; SSE2-NEXT:    movd %edx, %xmm1
; SSE2-NEXT:    shll $16, %esi
; SSE2-NEXT:    movd %esi, %xmm2
; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
; SSE2-NEXT:    movl %ecx, %edx
; SSE2-NEXT:    andl $-65536, %edx # imm = 0xFFFF0000
; SSE2-NEXT:    movd %edx, %xmm2
; SSE2-NEXT:    movl %ecx, %edx
; SSE2-NEXT:    shll $16, %edx
; SSE2-NEXT:    movd %edx, %xmm1
; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; SSE2-NEXT:    shrq $48, %rcx
; SSE2-NEXT:    shll $16, %ecx
; SSE2-NEXT:    movd %ecx, %xmm2
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm3
; SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
; SSE2-NEXT:    retq
;
; AVX-LABEL: pr64460_2:
; AVX:       # %bb.0:
; AVX-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVX-NEXT:    vpslld $16, %ymm0, %ymm0
; AVX-NEXT:    retq
  %b = fpext <8 x bfloat> %a to <8 x float>
  ret <8 x float> %b
}

define <16 x float> @pr64460_3(<16 x bfloat> %a) {
; X86-LABEL: pr64460_3:
; X86:       # %bb.0:
; X86-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; X86-NEXT:    vpslld $16, %zmm0, %zmm0
; X86-NEXT:    retl
;
; SSE2-LABEL: pr64460_3:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq %xmm1, %rdi
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1,1]
; SSE2-NEXT:    movq %xmm1, %rcx
; SSE2-NEXT:    movq %rcx, %rax
; SSE2-NEXT:    shrq $32, %rax
; SSE2-NEXT:    movq %xmm0, %r9
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    movq %xmm0, %rsi
; SSE2-NEXT:    movq %rsi, %rdx
; SSE2-NEXT:    shrq $32, %rdx
; SSE2-NEXT:    movq %rdi, %r8
; SSE2-NEXT:    shrq $32, %r8
; SSE2-NEXT:    movq %r9, %r10
; SSE2-NEXT:    shrq $32, %r10
; SSE2-NEXT:    movl %r9d, %r11d
; SSE2-NEXT:    andl $-65536, %r11d # imm = 0xFFFF0000
; SSE2-NEXT:    movd %r11d, %xmm1
; SSE2-NEXT:    movl %r9d, %r11d
; SSE2-NEXT:    shll $16, %r11d
; SSE2-NEXT:    movd %r11d, %xmm0
; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE2-NEXT:    shrq $48, %r9
; SSE2-NEXT:    shll $16, %r9d
; SSE2-NEXT:    movd %r9d, %xmm1
; SSE2-NEXT:    shll $16, %r10d
; SSE2-NEXT:    movd %r10d, %xmm2
; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
; SSE2-NEXT:    movl %edi, %r9d
; SSE2-NEXT:    andl $-65536, %r9d # imm = 0xFFFF0000
; SSE2-NEXT:    movd %r9d, %xmm1
; SSE2-NEXT:    movl %edi, %r9d
; SSE2-NEXT:    shll $16, %r9d
; SSE2-NEXT:    movd %r9d, %xmm2
; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE2-NEXT:    shrq $48, %rdi
; SSE2-NEXT:    shll $16, %edi
; SSE2-NEXT:    movd %edi, %xmm1
; SSE2-NEXT:    shll $16, %r8d
; SSE2-NEXT:    movd %r8d, %xmm3
; SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
; SSE2-NEXT:    movl %esi, %edi
; SSE2-NEXT:    andl $-65536, %edi # imm = 0xFFFF0000
; SSE2-NEXT:    movd %edi, %xmm3
; SSE2-NEXT:    movl %esi, %edi
; SSE2-NEXT:    shll $16, %edi
; SSE2-NEXT:    movd %edi, %xmm1
; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
; SSE2-NEXT:    shrq $48, %rsi
; SSE2-NEXT:    shll $16, %esi
; SSE2-NEXT:    movd %esi, %xmm3
; SSE2-NEXT:    shll $16, %edx
; SSE2-NEXT:    movd %edx, %xmm4
; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm4[0]
; SSE2-NEXT:    movl %ecx, %edx
; SSE2-NEXT:    andl $-65536, %edx # imm = 0xFFFF0000
; SSE2-NEXT:    movd %edx, %xmm4
; SSE2-NEXT:    movl %ecx, %edx
; SSE2-NEXT:    shll $16, %edx
; SSE2-NEXT:    movd %edx, %xmm3
; SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
; SSE2-NEXT:    shrq $48, %rcx
; SSE2-NEXT:    shll $16, %ecx
; SSE2-NEXT:    movd %ecx, %xmm4
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm5
; SSE2-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1]
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm5[0]
; SSE2-NEXT:    retq
;
; F16-LABEL: pr64460_3:
; F16:       # %bb.0:
; F16-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; F16-NEXT:    vpslld $16, %zmm0, %zmm0
; F16-NEXT:    retq
;
; AVXNC-LABEL: pr64460_3:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVXNC-NEXT:    vpslld $16, %ymm1, %ymm2
; AVXNC-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVXNC-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVXNC-NEXT:    vpslld $16, %ymm0, %ymm1
; AVXNC-NEXT:    vmovdqa %ymm2, %ymm0
; AVXNC-NEXT:    retq
  %b = fpext <16 x bfloat> %a to <16 x float>
  ret <16 x float> %b
}

define <8 x double> @pr64460_4(<8 x bfloat> %a) {
; X86-LABEL: pr64460_4:
; X86:       # %bb.0:
; X86-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; X86-NEXT:    vpslld $16, %ymm0, %ymm0
; X86-NEXT:    vcvtps2pd %ymm0, %zmm0
; X86-NEXT:    retl
;
; SSE2-LABEL: pr64460_4:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq %xmm0, %rsi
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    movq %xmm0, %rdx
; SSE2-NEXT:    movq %rdx, %rax
; SSE2-NEXT:    shrq $32, %rax
; SSE2-NEXT:    movq %rdx, %rcx
; SSE2-NEXT:    shrq $48, %rcx
; SSE2-NEXT:    movq %rsi, %rdi
; SSE2-NEXT:    shrq $32, %rdi
; SSE2-NEXT:    movq %rsi, %r8
; SSE2-NEXT:    shrq $48, %r8
; SSE2-NEXT:    movl %esi, %r9d
; SSE2-NEXT:    andl $-65536, %r9d # imm = 0xFFFF0000
; SSE2-NEXT:    movd %r9d, %xmm0
; SSE2-NEXT:    cvtss2sd %xmm0, %xmm1
; SSE2-NEXT:    shll $16, %esi
; SSE2-NEXT:    movd %esi, %xmm0
; SSE2-NEXT:    cvtss2sd %xmm0, %xmm0
; SSE2-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-NEXT:    shll $16, %r8d
; SSE2-NEXT:    movd %r8d, %xmm1
; SSE2-NEXT:    cvtss2sd %xmm1, %xmm2
; SSE2-NEXT:    shll $16, %edi
; SSE2-NEXT:    movd %edi, %xmm1
; SSE2-NEXT:    cvtss2sd %xmm1, %xmm1
; SSE2-NEXT:    movlhps {{.*#+}} xmm1 = xmm1[0],xmm2[0]
; SSE2-NEXT:    movl %edx, %esi
; SSE2-NEXT:    andl $-65536, %esi # imm = 0xFFFF0000
; SSE2-NEXT:    movd %esi, %xmm2
; SSE2-NEXT:    cvtss2sd %xmm2, %xmm3
; SSE2-NEXT:    shll $16, %edx
; SSE2-NEXT:    movd %edx, %xmm2
; SSE2-NEXT:    cvtss2sd %xmm2, %xmm2
; SSE2-NEXT:    movlhps {{.*#+}} xmm2 = xmm2[0],xmm3[0]
; SSE2-NEXT:    shll $16, %ecx
; SSE2-NEXT:    movd %ecx, %xmm3
; SSE2-NEXT:    cvtss2sd %xmm3, %xmm4
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm3
; SSE2-NEXT:    cvtss2sd %xmm3, %xmm3
; SSE2-NEXT:    movlhps {{.*#+}} xmm3 = xmm3[0],xmm4[0]
; SSE2-NEXT:    retq
;
; F16-LABEL: pr64460_4:
; F16:       # %bb.0:
; F16-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; F16-NEXT:    vpslld $16, %ymm0, %ymm0
; F16-NEXT:    vcvtps2pd %ymm0, %zmm0
; F16-NEXT:    retq
;
; AVXNC-LABEL: pr64460_4:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    vpextrw $3, %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm1
; AVXNC-NEXT:    vcvtss2sd %xmm1, %xmm1, %xmm1
; AVXNC-NEXT:    vpextrw $2, %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm2
; AVXNC-NEXT:    vcvtss2sd %xmm2, %xmm2, %xmm2
; AVXNC-NEXT:    vmovlhps {{.*#+}} xmm1 = xmm2[0],xmm1[0]
; AVXNC-NEXT:    vpextrw $1, %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm2
; AVXNC-NEXT:    vcvtss2sd %xmm2, %xmm2, %xmm2
; AVXNC-NEXT:    vmovd %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm3
; AVXNC-NEXT:    vcvtss2sd %xmm3, %xmm3, %xmm3
; AVXNC-NEXT:    vmovlhps {{.*#+}} xmm2 = xmm3[0],xmm2[0]
; AVXNC-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm2
; AVXNC-NEXT:    vpextrw $7, %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm1
; AVXNC-NEXT:    vcvtss2sd %xmm1, %xmm1, %xmm1
; AVXNC-NEXT:    vpextrw $6, %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm3
; AVXNC-NEXT:    vcvtss2sd %xmm3, %xmm3, %xmm3
; AVXNC-NEXT:    vmovlhps {{.*#+}} xmm1 = xmm3[0],xmm1[0]
; AVXNC-NEXT:    vpextrw $5, %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm3
; AVXNC-NEXT:    vcvtss2sd %xmm3, %xmm3, %xmm3
; AVXNC-NEXT:    vpextrw $4, %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm0
; AVXNC-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
; AVXNC-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm3[0]
; AVXNC-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1
; AVXNC-NEXT:    vmovaps %ymm2, %ymm0
; AVXNC-NEXT:    retq
  %b = fpext <8 x bfloat> %a to <8 x double>
  ret <8 x double> %b
}

define <4 x bfloat> @fptrunc_v4f32(<4 x float> %a) nounwind {
; X86-LABEL: fptrunc_v4f32:
; X86:       # %bb.0:
; X86-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
; X86-NEXT:    vcvtneps2bf16 %ymm0, %xmm0
; X86-NEXT:    vzeroupper
; X86-NEXT:    retl
;
; SSE2-LABEL: fptrunc_v4f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    subq $72, %rsp
; SSE2-NEXT:    movaps %xmm0, (%rsp) # 16-byte Spill
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE2-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE2-NEXT:    addq $72, %rsp
; SSE2-NEXT:    retq
;
; F16-LABEL: fptrunc_v4f32:
; F16:       # %bb.0:
; F16-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
; F16-NEXT:    vcvtneps2bf16 %ymm0, %xmm0
; F16-NEXT:    vzeroupper
; F16-NEXT:    retq
;
; AVXNC-LABEL: fptrunc_v4f32:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %ymm0, %xmm0
; AVXNC-NEXT:    vzeroupper
; AVXNC-NEXT:    retq
  %b = fptrunc <4 x float> %a to <4 x bfloat>
  ret <4 x bfloat> %b
}

define <8 x bfloat> @fptrunc_v8f32(<8 x float> %a) nounwind {
; X86-LABEL: fptrunc_v8f32:
; X86:       # %bb.0:
; X86-NEXT:    vcvtneps2bf16 %ymm0, %xmm0
; X86-NEXT:    vzeroupper
; X86-NEXT:    retl
;
; SSE2-LABEL: fptrunc_v8f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rbp
; SSE2-NEXT:    pushq %r14
; SSE2-NEXT:    pushq %rbx
; SSE2-NEXT:    subq $32, %rsp
; SSE2-NEXT:    movaps %xmm1, (%rsp) # 16-byte Spill
; SSE2-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %ebx
; SSE2-NEXT:    orl %ebp, %ebx
; SSE2-NEXT:    shlq $32, %rbx
; SSE2-NEXT:    orq %r14, %rbx
; SSE2-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebp, %r14d
; SSE2-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebp, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm1
; SSE2-NEXT:    movq %rbx, %xmm0
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-NEXT:    addq $32, %rsp
; SSE2-NEXT:    popq %rbx
; SSE2-NEXT:    popq %r14
; SSE2-NEXT:    popq %rbp
; SSE2-NEXT:    retq
;
; F16-LABEL: fptrunc_v8f32:
; F16:       # %bb.0:
; F16-NEXT:    vcvtneps2bf16 %ymm0, %xmm0
; F16-NEXT:    vzeroupper
; F16-NEXT:    retq
;
; AVXNC-LABEL: fptrunc_v8f32:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %ymm0, %xmm0
; AVXNC-NEXT:    vzeroupper
; AVXNC-NEXT:    retq
  %b = fptrunc <8 x float> %a to <8 x bfloat>
  ret <8 x bfloat> %b
}

define <16 x bfloat> @fptrunc_v16f32(<16 x float> %a) nounwind {
; X86-LABEL: fptrunc_v16f32:
; X86:       # %bb.0:
; X86-NEXT:    vcvtneps2bf16 %zmm0, %ymm0
; X86-NEXT:    retl
;
; SSE2-LABEL: fptrunc_v16f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rbp
; SSE2-NEXT:    pushq %r15
; SSE2-NEXT:    pushq %r14
; SSE2-NEXT:    pushq %r12
; SSE2-NEXT:    pushq %rbx
; SSE2-NEXT:    subq $64, %rsp
; SSE2-NEXT:    movaps %xmm3, (%rsp) # 16-byte Spill
; SSE2-NEXT:    movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movaps %xmm2, %xmm0
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1],xmm2[1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %ebx
; SSE2-NEXT:    orl %ebp, %ebx
; SSE2-NEXT:    shlq $32, %rbx
; SSE2-NEXT:    orq %r14, %rbx
; SSE2-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r15d
; SSE2-NEXT:    orl %ebp, %r15d
; SSE2-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebp, %r14d
; SSE2-NEXT:    shlq $32, %r14
; SSE2-NEXT:    orq %r15, %r14
; SSE2-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r12d
; SSE2-NEXT:    orl %ebp, %r12d
; SSE2-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r15d
; SSE2-NEXT:    orl %ebp, %r15d
; SSE2-NEXT:    shlq $32, %r15
; SSE2-NEXT:    orq %r12, %r15
; SSE2-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r12d
; SSE2-NEXT:    orl %ebp, %r12d
; SSE2-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebp, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r12, %rax
; SSE2-NEXT:    movq %rax, %xmm1
; SSE2-NEXT:    movq %r15, %xmm0
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-NEXT:    movq %r14, %xmm2
; SSE2-NEXT:    movq %rbx, %xmm1
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
; SSE2-NEXT:    addq $64, %rsp
; SSE2-NEXT:    popq %rbx
; SSE2-NEXT:    popq %r12
; SSE2-NEXT:    popq %r14
; SSE2-NEXT:    popq %r15
; SSE2-NEXT:    popq %rbp
; SSE2-NEXT:    retq
;
; F16-LABEL: fptrunc_v16f32:
; F16:       # %bb.0:
; F16-NEXT:    vcvtneps2bf16 %zmm0, %ymm0
; F16-NEXT:    retq
;
; AVXNC-LABEL: fptrunc_v16f32:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %ymm0, %xmm0
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %ymm1, %xmm1
; AVXNC-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVXNC-NEXT:    retq
  %b = fptrunc <16 x float> %a to <16 x bfloat>
  ret <16 x bfloat> %b
}

define <8 x bfloat> @fptrunc_v8f64(<8 x double> %a) nounwind {
; X86-LABEL: fptrunc_v8f64:
; X86:       # %bb.0:
; X86-NEXT:    subl $204, %esp
; X86-NEXT:    vmovups %zmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 64-byte Spill
; X86-NEXT:    vextractf128 $1, %ymm0, %xmm0
; X86-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
; X86-NEXT:    vmovlps %xmm0, (%esp)
; X86-NEXT:    vzeroupper
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
; X86-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
; X86-NEXT:    vmovhps %xmm0, (%esp)
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
; X86-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %zmm0 # 64-byte Reload
; X86-NEXT:    vmovlps %xmm0, (%esp)
; X86-NEXT:    vzeroupper
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
; X86-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %zmm0 # 64-byte Reload
; X86-NEXT:    vmovhps %xmm0, (%esp)
; X86-NEXT:    vzeroupper
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
; X86-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %zmm0 # 64-byte Reload
; X86-NEXT:    vextractf32x4 $2, %zmm0, %xmm0
; X86-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
; X86-NEXT:    vmovlps %xmm0, (%esp)
; X86-NEXT:    vzeroupper
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
; X86-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
; X86-NEXT:    vmovhps %xmm0, (%esp)
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
; X86-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %zmm0 # 64-byte Reload
; X86-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
; X86-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
; X86-NEXT:    vmovlps %xmm0, (%esp)
; X86-NEXT:    vzeroupper
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
; X86-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
; X86-NEXT:    vmovhps %xmm0, (%esp)
; X86-NEXT:    calll __truncdfbf2
; X86-NEXT:    vmovdqa {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
; X86-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; X86-NEXT:    vmovdqa {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
; X86-NEXT:    vpunpcklwd {{[-0-9]+}}(%e{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
; X86-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3]
; X86-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; X86-NEXT:    vmovdqa {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
; X86-NEXT:    vpunpcklwd {{[-0-9]+}}(%e{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
; X86-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3]
; X86-NEXT:    vmovdqa {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
; X86-NEXT:    vpunpcklwd {{[-0-9]+}}(%e{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
; X86-NEXT:    # xmm2 = xmm2[0],mem[0],xmm2[1],mem[1],xmm2[2],mem[2],xmm2[3],mem[3]
; X86-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; X86-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; X86-NEXT:    addl $204, %esp
; X86-NEXT:    retl
;
; SSE2-LABEL: fptrunc_v8f64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rbp
; SSE2-NEXT:    pushq %r14
; SSE2-NEXT:    pushq %rbx
; SSE2-NEXT:    subq $64, %rsp
; SSE2-NEXT:    movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    movaps %xmm1, (%rsp) # 16-byte Spill
; SSE2-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebx
; SSE2-NEXT:    shll $16, %ebx
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebx, %r14d
; SSE2-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %ebx
; SSE2-NEXT:    orl %ebp, %ebx
; SSE2-NEXT:    shlq $32, %rbx
; SSE2-NEXT:    orq %r14, %rbx
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %r14d
; SSE2-NEXT:    orl %ebp, %r14d
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %ebp
; SSE2-NEXT:    shll $16, %ebp
; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE2-NEXT:    callq __truncdfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    movzwl %ax, %eax
; SSE2-NEXT:    orl %ebp, %eax
; SSE2-NEXT:    shlq $32, %rax
; SSE2-NEXT:    orq %r14, %rax
; SSE2-NEXT:    movq %rax, %xmm1
; SSE2-NEXT:    movq %rbx, %xmm0
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-NEXT:    addq $64, %rsp
; SSE2-NEXT:    popq %rbx
; SSE2-NEXT:    popq %r14
; SSE2-NEXT:    popq %rbp
; SSE2-NEXT:    retq
;
; FP16-LABEL: fptrunc_v8f64:
; FP16:       # %bb.0:
; FP16-NEXT:    subq $184, %rsp
; FP16-NEXT:    vmovupd %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
; FP16-NEXT:    vextractf128 $1, %ymm0, %xmm0
; FP16-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; FP16-NEXT:    vshufpd {{.*#+}} xmm0 = xmm0[1,0]
; FP16-NEXT:    vzeroupper
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; FP16-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; FP16-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; FP16-NEXT:    # xmm0 = mem[1,0]
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; FP16-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
; FP16-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
; FP16-NEXT:    vzeroupper
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; FP16-NEXT:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
; FP16-NEXT:    vextractf32x4 $2, %zmm0, %xmm0
; FP16-NEXT:    vmovapd %xmm0, (%rsp) # 16-byte Spill
; FP16-NEXT:    vshufpd {{.*#+}} xmm0 = xmm0[1,0]
; FP16-NEXT:    vzeroupper
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; FP16-NEXT:    vmovaps (%rsp), %xmm0 # 16-byte Reload
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
; FP16-NEXT:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
; FP16-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
; FP16-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; FP16-NEXT:    vshufpd {{.*#+}} xmm0 = xmm0[1,0]
; FP16-NEXT:    vzeroupper
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; FP16-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; FP16-NEXT:    callq __truncdfbf2@PLT
; FP16-NEXT:    vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
; FP16-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
; FP16-NEXT:    vmovdqa (%rsp), %xmm1 # 16-byte Reload
; FP16-NEXT:    vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
; FP16-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3]
; FP16-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; FP16-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; FP16-NEXT:    vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
; FP16-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3]
; FP16-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; FP16-NEXT:    vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
; FP16-NEXT:    # xmm2 = xmm2[0],mem[0],xmm2[1],mem[1],xmm2[2],mem[2],xmm2[3],mem[3]
; FP16-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; FP16-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; FP16-NEXT:    addq $184, %rsp
; FP16-NEXT:    retq
;
; AVXNC-LABEL: fptrunc_v8f64:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    pushq %rbp
; AVXNC-NEXT:    pushq %r15
; AVXNC-NEXT:    pushq %r14
; AVXNC-NEXT:    pushq %r13
; AVXNC-NEXT:    pushq %r12
; AVXNC-NEXT:    pushq %rbx
; AVXNC-NEXT:    subq $168, %rsp
; AVXNC-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVXNC-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVXNC-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVXNC-NEXT:    vzeroupper
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVXNC-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVXNC-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVXNC-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
; AVXNC-NEXT:    vzeroupper
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVXNC-NEXT:    vpermilpd $1, (%rsp), %xmm0 # 16-byte Folded Reload
; AVXNC-NEXT:    # xmm0 = mem[1,0]
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vmovapd %xmm0, (%rsp) # 16-byte Spill
; AVXNC-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVXNC-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVXNC-NEXT:    vzeroupper
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVXNC-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; AVXNC-NEXT:    # xmm0 = mem[1,0]
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVXNC-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVXNC-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVXNC-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVXNC-NEXT:    vzeroupper
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVXNC-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; AVXNC-NEXT:    # xmm0 = mem[1,0]
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vpextrw $0, %xmm0, %eax
; AVXNC-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVXNC-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVXNC-NEXT:    vpextrw $0, %xmm0, %ebp
; AVXNC-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVXNC-NEXT:    vpextrw $0, %xmm0, %r14d
; AVXNC-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVXNC-NEXT:    vpextrw $0, %xmm0, %r15d
; AVXNC-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
; AVXNC-NEXT:    vpextrw $0, %xmm0, %r12d
; AVXNC-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVXNC-NEXT:    vpextrw $0, %xmm0, %r13d
; AVXNC-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVXNC-NEXT:    vpextrw $0, %xmm0, %ebx
; AVXNC-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; AVXNC-NEXT:    # xmm0 = mem[1,0]
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vpextrw $0, %xmm0, %eax
; AVXNC-NEXT:    vmovd %ebx, %xmm0
; AVXNC-NEXT:    vpinsrw $1, %eax, %xmm0, %xmm0
; AVXNC-NEXT:    vpinsrw $2, %r13d, %xmm0, %xmm0
; AVXNC-NEXT:    vpinsrw $3, %r12d, %xmm0, %xmm0
; AVXNC-NEXT:    vpinsrw $4, %r15d, %xmm0, %xmm0
; AVXNC-NEXT:    vpinsrw $5, %r14d, %xmm0, %xmm0
; AVXNC-NEXT:    vpinsrw $6, %ebp, %xmm0, %xmm0
; AVXNC-NEXT:    vpinsrw $7, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 4-byte Folded Reload
; AVXNC-NEXT:    addq $168, %rsp
; AVXNC-NEXT:    popq %rbx
; AVXNC-NEXT:    popq %r12
; AVXNC-NEXT:    popq %r13
; AVXNC-NEXT:    popq %r14
; AVXNC-NEXT:    popq %r15
; AVXNC-NEXT:    popq %rbp
; AVXNC-NEXT:    retq
  %b = fptrunc <8 x double> %a to <8 x bfloat>
  ret <8 x bfloat> %b
}

define <32 x bfloat> @test_v8bf16_v32bf16(ptr %0) {
; X86-LABEL: test_v8bf16_v32bf16:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    vbroadcastf32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X86-NEXT:    retl
;
; SSE2-LABEL: test_v8bf16_v32bf16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movaps (%rdi), %xmm0
; SSE2-NEXT:    movaps %xmm0, %xmm1
; SSE2-NEXT:    movaps %xmm0, %xmm2
; SSE2-NEXT:    movaps %xmm0, %xmm3
; SSE2-NEXT:    retq
;
; F16-LABEL: test_v8bf16_v32bf16:
; F16:       # %bb.0:
; F16-NEXT:    vbroadcastf32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; F16-NEXT:    retq
;
; AVXNC-LABEL: test_v8bf16_v32bf16:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1]
; AVXNC-NEXT:    vmovaps %ymm0, %ymm1
; AVXNC-NEXT:    retq
  %2 = load <8 x bfloat>, ptr %0, align 16
  %3 = shufflevector <8 x bfloat> %2, <8 x bfloat> %2, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  ret <32 x bfloat> %3
}

define <16 x bfloat> @concat_v8bf16(<8 x bfloat> %x, <8 x bfloat> %y) {
; X86-LABEL: concat_v8bf16:
; X86:       # %bb.0:
; X86-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
; X86-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; X86-NEXT:    retl
;
; SSE2-LABEL: concat_v8bf16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    retq
;
; AVX-LABEL: concat_v8bf16:
; AVX:       # %bb.0:
; AVX-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
; AVX-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX-NEXT:    retq
  %a = shufflevector <8 x bfloat> %x, <8 x bfloat> %y, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  ret <16 x bfloat> %a
}

define <8 x bfloat> @extract_v32bf16_v8bf16(<32 x bfloat> %x) {
; X86-LABEL: extract_v32bf16_v8bf16:
; X86:       # %bb.0:
; X86-NEXT:    vextractf128 $1, %ymm0, %xmm0
; X86-NEXT:    vzeroupper
; X86-NEXT:    retl
;
; SSE2-LABEL: extract_v32bf16_v8bf16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pextrw $0, %xmm1, %eax
; SSE2-NEXT:    pextrw $1, %xmm1, %ecx
; SSE2-NEXT:    shll $16, %ecx
; SSE2-NEXT:    orl %eax, %ecx
; SSE2-NEXT:    pextrw $2, %xmm1, %eax
; SSE2-NEXT:    pextrw $3, %xmm1, %edx
; SSE2-NEXT:    shll $16, %edx
; SSE2-NEXT:    orl %eax, %edx
; SSE2-NEXT:    shlq $32, %rdx
; SSE2-NEXT:    orq %rcx, %rdx
; SSE2-NEXT:    pextrw $4, %xmm1, %eax
; SSE2-NEXT:    pextrw $5, %xmm1, %ecx
; SSE2-NEXT:    shll $16, %ecx
; SSE2-NEXT:    orl %eax, %ecx
; SSE2-NEXT:    pextrw $6, %xmm1, %eax
; SSE2-NEXT:    pextrw $7, %xmm1, %esi
; SSE2-NEXT:    shll $16, %esi
; SSE2-NEXT:    orl %eax, %esi
; SSE2-NEXT:    shlq $32, %rsi
; SSE2-NEXT:    orq %rcx, %rsi
; SSE2-NEXT:    movq %rsi, %xmm1
; SSE2-NEXT:    movq %rdx, %xmm0
; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-NEXT:    retq
;
; AVX-LABEL: extract_v32bf16_v8bf16:
; AVX:       # %bb.0:
; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX-NEXT:    vzeroupper
; AVX-NEXT:    retq
  %a = shufflevector <32 x bfloat> %x, <32 x bfloat> undef, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  ret <8 x bfloat> %a
}

define <16 x bfloat> @concat_zero_v8bf16(<8 x bfloat> %x, <8 x bfloat> %y) {
; X86-LABEL: concat_zero_v8bf16:
; X86:       # %bb.0:
; X86-NEXT:    vmovaps %xmm0, %xmm0
; X86-NEXT:    retl
;
; SSE2-LABEL: concat_zero_v8bf16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    xorps %xmm1, %xmm1
; SSE2-NEXT:    retq
;
; AVX-LABEL: concat_zero_v8bf16:
; AVX:       # %bb.0:
; AVX-NEXT:    vmovaps %xmm0, %xmm0
; AVX-NEXT:    retq
  %a = shufflevector <8 x bfloat> %x, <8 x bfloat> zeroinitializer, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  ret <16 x bfloat> %a
}

define <16 x bfloat> @concat_dup_v8bf16(<8 x bfloat> %x, <8 x bfloat> %y) {
; X86-LABEL: concat_dup_v8bf16:
; X86:       # %bb.0:
; X86-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; X86-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; X86-NEXT:    retl
;
; SSE2-LABEL: concat_dup_v8bf16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0,0]
; SSE2-NEXT:    retq
;
; AVX-LABEL: concat_dup_v8bf16:
; AVX:       # %bb.0:
; AVX-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; AVX-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX-NEXT:    retq
  %a = shufflevector <8 x bfloat> %x, <8 x bfloat> %y, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  ret <16 x bfloat> %a
}