llvm/test/CodeGen/X86/bfloat.ll

; 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:    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:    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:    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:    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:    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:    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 bfloat @fold_from_half(half %a) nounwind {
; X86-LABEL: fold_from_half:
; X86:       # %bb.0:
; X86-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
; X86-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; X86-NEXT:    retl
;
; SSE2-LABEL: fold_from_half:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rax
; SSE2-NEXT:    callq __extendhfsf2@PLT
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    popq %rax
; SSE2-NEXT:    retq
;
; FP16-LABEL: fold_from_half:
; FP16:       # %bb.0:
; FP16-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
; FP16-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; FP16-NEXT:    retq
;
; AVXNC-LABEL: fold_from_half:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    vcvtph2ps %xmm0, %xmm0
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %xmm0, %xmm0
; AVXNC-NEXT:    retq
  %ext = fpext half %a to float
  %trunc = fptrunc float %ext to bfloat
  ret bfloat %trunc
}

define half @fold_to_half(bfloat %a) nounwind {
; X86-LABEL: fold_to_half:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shll $16, %eax
; X86-NEXT:    vmovd %eax, %xmm0
; X86-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
; X86-NEXT:    retl
;
; SSE2-LABEL: fold_to_half:
; 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:    callq __truncsfhf2@PLT
; SSE2-NEXT:    popq %rax
; SSE2-NEXT:    retq
;
; BF16-LABEL: fold_to_half:
; BF16:       # %bb.0:
; BF16-NEXT:    vpextrw $0, %xmm0, %eax
; BF16-NEXT:    shll $16, %eax
; BF16-NEXT:    vmovd %eax, %xmm0
; BF16-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
; BF16-NEXT:    retq
;
; FP16-LABEL: fold_to_half:
; FP16:       # %bb.0:
; FP16-NEXT:    vmovw %xmm0, %eax
; FP16-NEXT:    shll $16, %eax
; FP16-NEXT:    vmovd %eax, %xmm0
; FP16-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
; FP16-NEXT:    retq
  %ext = fpext bfloat %a to float
  %trunc = fptrunc float %ext to half
  ret half %trunc
}

define bfloat @bitcast_from_half(half %a) nounwind {
; X86-LABEL: bitcast_from_half:
; X86:       # %bb.0:
; X86-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT:    retl
;
; CHECK-LABEL: bitcast_from_half:
; CHECK:       # %bb.0:
; CHECK-NEXT:    retq
  %bc = bitcast half %a to bfloat
  ret bfloat %bc
}

define half @bitcast_to_half(bfloat %a) nounwind {
; X86-LABEL: bitcast_to_half:
; X86:       # %bb.0:
; X86-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT:    retl
;
; CHECK-LABEL: bitcast_to_half:
; CHECK:       # %bb.0:
; CHECK-NEXT:    retq
  %bc = bitcast bfloat %a to half
  ret half %bc
}

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, %xmm1, %eax
; BF16-NEXT:    vpinsrw $1, %eax, %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:    testb %al, %al
; SSE2-NEXT:    jne .LBB16_1
; SSE2-NEXT:  # %bb.2: # %cond.load
; SSE2-NEXT:    movzwl (%rax), %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    jmp .LBB16_3
; SSE2-NEXT:  .LBB16_1:
; SSE2-NEXT:    movd {{.*#+}} xmm0 = [-1.0E+0,0.0E+0,0.0E+0,0.0E+0]
; SSE2-NEXT:  .LBB16_3:
; SSE2-NEXT:    pushq %r14
; SSE2-NEXT:    pushq %rbx
; SSE2-NEXT:    subq $88, %rsp
; SSE2-NEXT:    movd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
; 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 $88, %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:    vbroadcastss {{.*#+}} ymm0 = [49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024]
; AVXNC-NEXT:    testb %al, %al
; AVXNC-NEXT:    jne .LBB16_2
; AVXNC-NEXT:  # %bb.1: # %cond.load
; AVXNC-NEXT:    vmovups (%rax), %ymm0
; AVXNC-NEXT:  .LBB16_2:
; AVXNC-NEXT:    vmovaps %ymm0, %ymm1
; 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:    pxor %xmm1, %xmm1
; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; 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:    pxor %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm2, %xmm2
; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-NEXT:    movdqa %xmm2, %xmm0
; 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:    pxor %xmm3, %xmm3
; SSE2-NEXT:    pxor %xmm5, %xmm5
; SSE2-NEXT:    punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3]
; SSE2-NEXT:    pxor %xmm4, %xmm4
; SSE2-NEXT:    punpckhwd {{.*#+}} xmm4 = xmm4[4],xmm0[4],xmm4[5],xmm0[5],xmm4[6],xmm0[6],xmm4[7],xmm0[7]
; SSE2-NEXT:    pxor %xmm2, %xmm2
; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm1[4],xmm3[5],xmm1[5],xmm3[6],xmm1[6],xmm3[7],xmm1[7]
; SSE2-NEXT:    movdqa %xmm5, %xmm0
; SSE2-NEXT:    movdqa %xmm4, %xmm1
; 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:    pxor %xmm3, %xmm3
; SSE2-NEXT:    pxor %xmm1, %xmm1
; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; SSE2-NEXT:    cvtps2pd %xmm1, %xmm4
; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
; SSE2-NEXT:    cvtps2pd %xmm3, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,2,3]
; SSE2-NEXT:    cvtps2pd %xmm0, %xmm1
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[2,3,2,3]
; SSE2-NEXT:    cvtps2pd %xmm0, %xmm3
; SSE2-NEXT:    movaps %xmm4, %xmm0
; 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:    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:    vcvtps2pd %xmm1, %ymm0
; AVXNC-NEXT:    vextracti128 $1, %ymm1, %xmm1
; AVXNC-NEXT:    vcvtps2pd %xmm1, %ymm1
; 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, (%rsp) # 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:    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:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVXNC-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVXNC-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVXNC-NEXT:    vshufpd {{.*#+}} xmm0 = xmm0[1,0]
; AVXNC-NEXT:    vzeroupper
; AVXNC-NEXT:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVXNC-NEXT:    vmovups (%rsp), %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, (%rsp), %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 (%rsp), %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, (%rsp) # 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, %ebx
; AVXNC-NEXT:    vmovdqa (%rsp), %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 {{[-0-9]+}}(%r{{[sb]}}p), %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:    callq __truncdfbf2@PLT
; AVXNC-NEXT:    vpextrw $0, %xmm0, %eax
; AVXNC-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVXNC-NEXT:    vpinsrw $1, %r13d, %xmm0, %xmm0
; AVXNC-NEXT:    vpinsrw $2, %eax, %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, %ebx, %xmm0, %xmm0
; 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
}

define float @trunc_ext(float %a) nounwind {
; X86-LABEL: trunc_ext:
; X86:       # %bb.0:
; X86-NEXT:    pushl %eax
; X86-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; X86-NEXT:    vmovw %xmm0, %eax
; X86-NEXT:    shll $16, %eax
; X86-NEXT:    vmovd %eax, %xmm0
; X86-NEXT:    vmovd %xmm0, (%esp)
; X86-NEXT:    flds (%esp)
; X86-NEXT:    popl %eax
; X86-NEXT:    retl
;
; SSE2-LABEL: trunc_ext:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pushq %rax
; SSE2-NEXT:    callq __truncsfbf2@PLT
; SSE2-NEXT:    pextrw $0, %xmm0, %eax
; SSE2-NEXT:    shll $16, %eax
; SSE2-NEXT:    movd %eax, %xmm0
; SSE2-NEXT:    popq %rax
; SSE2-NEXT:    retq
;
; FP16-LABEL: trunc_ext:
; FP16:       # %bb.0:
; FP16-NEXT:    vcvtneps2bf16 %xmm0, %xmm0
; FP16-NEXT:    vmovw %xmm0, %eax
; FP16-NEXT:    shll $16, %eax
; FP16-NEXT:    vmovd %eax, %xmm0
; FP16-NEXT:    retq
;
; AVXNC-LABEL: trunc_ext:
; AVXNC:       # %bb.0:
; AVXNC-NEXT:    {vex} vcvtneps2bf16 %xmm0, %xmm0
; AVXNC-NEXT:    vmovd %xmm0, %eax
; AVXNC-NEXT:    shll $16, %eax
; AVXNC-NEXT:    vmovd %eax, %xmm0
; AVXNC-NEXT:    retq
  %b = fptrunc float %a to bfloat
  %c = fpext bfloat %b to float
  ret float %c
}

define void @PR92471(ptr %0, ptr %1) nounwind {
; X86-LABEL: PR92471:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-NEXT:    vpinsrd $1, 4(%ecx), %xmm0, %xmm0
; X86-NEXT:    vpinsrd $2, 8(%ecx), %xmm0, %xmm0
; X86-NEXT:    vpinsrw $6, 12(%ecx), %xmm0, %xmm0
; 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:    vextracti128 $1, %ymm0, %xmm1
; X86-NEXT:    vpextrd $2, %xmm1, 24(%eax)
; X86-NEXT:    vpextrd $1, %xmm1, 20(%eax)
; X86-NEXT:    vmovd %xmm1, 16(%eax)
; X86-NEXT:    vmovdqu %xmm0, (%eax)
; X86-NEXT:    vzeroupper
; X86-NEXT:    retl
;
; SSE2-LABEL: PR92471:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE2-NEXT:    pinsrw $2, 12(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm2, %xmm2
; SSE2-NEXT:    pxor %xmm3, %xmm3
; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
; SSE2-NEXT:    movdqu %xmm2, (%rsi)
; SSE2-NEXT:    movq %xmm3, 16(%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[2,3,2,3]
; SSE2-NEXT:    movd %xmm0, 24(%rsi)
; SSE2-NEXT:    retq
;
; AVX-LABEL: PR92471:
; AVX:       # %bb.0:
; AVX-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX-NEXT:    vpinsrd $2, 8(%rdi), %xmm0, %xmm0
; AVX-NEXT:    vpinsrw $6, 12(%rdi), %xmm0, %xmm0
; 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:    vextracti128 $1, %ymm0, %xmm1
; AVX-NEXT:    vpextrd $2, %xmm1, 24(%rsi)
; AVX-NEXT:    vmovq %xmm1, 16(%rsi)
; AVX-NEXT:    vmovdqu %xmm0, (%rsi)
; AVX-NEXT:    vzeroupper
; AVX-NEXT:    retq
  %3 = load <7 x bfloat>, ptr %0, align 2
  %4 = fpext <7 x bfloat> %3 to <7 x float>
  store <7 x float> %4, ptr %1, align 4
  ret void
}

define bfloat @PR108936(x86_fp80 %0) nounwind {
; X86-LABEL: PR108936:
; X86:       # %bb.0:
; X86-NEXT:    subl $12, %esp
; X86-NEXT:    fldt {{[0-9]+}}(%esp)
; X86-NEXT:    fstpt (%esp)
; X86-NEXT:    calll __truncxfbf2
; X86-NEXT:    addl $12, %esp
; X86-NEXT:    retl
;
; CHECK-LABEL: PR108936:
; CHECK:       # %bb.0:
; CHECK-NEXT:    subq $24, %rsp
; CHECK-NEXT:    fldt {{[0-9]+}}(%rsp)
; CHECK-NEXT:    fstpt (%rsp)
; CHECK-NEXT:    callq __truncxfbf2@PLT
; CHECK-NEXT:    addq $24, %rsp
; CHECK-NEXT:    retq
  %2 = fptrunc x86_fp80 %0 to bfloat
  ret bfloat %2
}

define bfloat @PR115710(fp128 %0) nounwind {
; X86-LABEL: PR115710:
; X86:       # %bb.0:
; X86-NEXT:    subl $28, %esp
; X86-NEXT:    vmovaps {{[0-9]+}}(%esp), %xmm0
; X86-NEXT:    vmovups %xmm0, (%esp)
; X86-NEXT:    calll __trunctfbf2
; X86-NEXT:    addl $28, %esp
; X86-NEXT:    retl
;
; CHECK-LABEL: PR115710:
; CHECK:       # %bb.0:
; CHECK-NEXT:    pushq %rax
; CHECK-NEXT:    callq __trunctfbf2@PLT
; CHECK-NEXT:    popq %rax
; CHECK-NEXT:    retq
  %2 = fptrunc fp128 %0 to bfloat
  ret bfloat %2
}