; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2-SSSE3,SSE2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE2-SSSE3,SSSE3
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX12,AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX12,AVX2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512

;
; 128-bit vectors
;

define <2 x i64> @ext_i2_2i64(i2 %a0) {
; SSE2-SSSE3-LABEL: ext_i2_2i64:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    andb $3, %dil
; SSE2-SSSE3-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shlq $62, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movq %rcx, %xmm1
; SSE2-SSSE3-NEXT:    shlq $63, %rax
; SSE2-SSSE3-NEXT:    sarq $63, %rax
; SSE2-SSSE3-NEXT:    movq %rax, %xmm0
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-SSSE3-NEXT:    retq
;
; AVX12-LABEL: ext_i2_2i64:
; AVX12:       # BB#0:
; AVX12-NEXT:    andb $3, %dil
; AVX12-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX12-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $62, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vmovq %rcx, %xmm0
; AVX12-NEXT:    shlq $63, %rax
; AVX12-NEXT:    sarq $63, %rax
; AVX12-NEXT:    vmovq %rax, %xmm1
; AVX12-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX12-NEXT:    retq
;
; AVX512-LABEL: ext_i2_2i64:
; AVX512:       # BB#0:
; AVX512-NEXT:    andb $3, %dil
; AVX512-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX512-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; AVX512-NEXT:    kmovd %eax, %k1
; AVX512-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
; AVX512-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %1 = bitcast i2 %a0 to <2 x i1>
  %2 = sext <2 x i1> %1 to <2 x i64>
  ret <2 x i64> %2
}

define <4 x i32> @ext_i4_4i32(i4 %a0) {
; SSE2-SSSE3-LABEL: ext_i4_4i32:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    andb $15, %dil
; SSE2-SSSE3-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shlq $60, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shlq $61, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shlq $62, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    shlq $63, %rax
; SSE2-SSSE3-NEXT:    sarq $63, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm0
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-SSSE3-NEXT:    retq
;
; AVX12-LABEL: ext_i4_4i32:
; AVX12:       # BB#0:
; AVX12-NEXT:    andb $15, %dil
; AVX12-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX12-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $62, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    movq %rax, %rdx
; AVX12-NEXT:    shlq $63, %rdx
; AVX12-NEXT:    sarq $63, %rdx
; AVX12-NEXT:    vmovd %edx, %xmm0
; AVX12-NEXT:    vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $61, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrd $2, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    shlq $60, %rax
; AVX12-NEXT:    sarq $63, %rax
; AVX12-NEXT:    vpinsrd $3, %eax, %xmm0, %xmm0
; AVX12-NEXT:    retq
;
; AVX512-LABEL: ext_i4_4i32:
; AVX512:       # BB#0:
; AVX512-NEXT:    andb $15, %dil
; AVX512-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX512-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; AVX512-NEXT:    kmovd %eax, %k1
; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; AVX512-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %1 = bitcast i4 %a0 to <4 x i1>
  %2 = sext <4 x i1> %1 to <4 x i32>
  ret <4 x i32> %2
}

define <8 x i16> @ext_i8_8i16(i8 %a0) {
; SSE2-SSSE3-LABEL: ext_i8_8i16:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movsbq -{{[0-9]+}}(%rsp), %rax
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shrq $7, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shlq $57, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shlq $58, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shlq $59, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shlq $60, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shlq $61, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    shlq $62, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm3
; SSE2-SSSE3-NEXT:    shlq $63, %rax
; SSE2-SSSE3-NEXT:    sarq $63, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm0
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-SSSE3-NEXT:    retq
;
; AVX12-LABEL: ext_i8_8i16:
; AVX12:       # BB#0:
; AVX12-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX12-NEXT:    movsbq -{{[0-9]+}}(%rsp), %rax
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $62, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    movq %rax, %rdx
; AVX12-NEXT:    shlq $63, %rdx
; AVX12-NEXT:    sarq $63, %rdx
; AVX12-NEXT:    vmovd %edx, %xmm0
; AVX12-NEXT:    vpinsrw $1, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $61, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrw $2, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $60, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrw $3, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $59, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrw $4, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $58, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $57, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    shrq $7, %rax
; AVX12-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
; AVX12-NEXT:    retq
;
; AVX512-LABEL: ext_i8_8i16:
; AVX512:       # BB#0:
; AVX512-NEXT:    kmovd %edi, %k0
; AVX512-NEXT:    vpmovm2w %k0, %xmm0
; AVX512-NEXT:    retq
  %1 = bitcast i8 %a0 to <8 x i1>
  %2 = sext <8 x i1> %1 to <8 x i16>
  ret <8 x i16> %2
}

define <16 x i8> @ext_i16_16i8(i16 %a0) {
; SSE2-SSSE3-LABEL: ext_i16_16i8:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    pushq %rbp
; SSE2-SSSE3-NEXT:  .Lcfi0:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 16
; SSE2-SSSE3-NEXT:    pushq %r15
; SSE2-SSSE3-NEXT:  .Lcfi1:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 24
; SSE2-SSSE3-NEXT:    pushq %r14
; SSE2-SSSE3-NEXT:  .Lcfi2:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 32
; SSE2-SSSE3-NEXT:    pushq %r13
; SSE2-SSSE3-NEXT:  .Lcfi3:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 40
; SSE2-SSSE3-NEXT:    pushq %r12
; SSE2-SSSE3-NEXT:  .Lcfi4:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 48
; SSE2-SSSE3-NEXT:    pushq %rbx
; SSE2-SSSE3-NEXT:  .Lcfi5:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 56
; SSE2-SSSE3-NEXT:  .Lcfi6:
; SSE2-SSSE3-NEXT:    .cfi_offset %rbx, -56
; SSE2-SSSE3-NEXT:  .Lcfi7:
; SSE2-SSSE3-NEXT:    .cfi_offset %r12, -48
; SSE2-SSSE3-NEXT:  .Lcfi8:
; SSE2-SSSE3-NEXT:    .cfi_offset %r13, -40
; SSE2-SSSE3-NEXT:  .Lcfi9:
; SSE2-SSSE3-NEXT:    .cfi_offset %r14, -32
; SSE2-SSSE3-NEXT:  .Lcfi10:
; SSE2-SSSE3-NEXT:    .cfi_offset %r15, -24
; SSE2-SSSE3-NEXT:  .Lcfi11:
; SSE2-SSSE3-NEXT:    .cfi_offset %rbp, -16
; SSE2-SSSE3-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movswq -{{[0-9]+}}(%rsp), %rax
; SSE2-SSSE3-NEXT:    movq %rax, %r8
; SSE2-SSSE3-NEXT:    movq %rax, %r9
; SSE2-SSSE3-NEXT:    movq %rax, %r10
; SSE2-SSSE3-NEXT:    movq %rax, %r11
; SSE2-SSSE3-NEXT:    movq %rax, %r14
; SSE2-SSSE3-NEXT:    movq %rax, %r15
; SSE2-SSSE3-NEXT:    movq %rax, %r12
; SSE2-SSSE3-NEXT:    movq %rax, %r13
; SSE2-SSSE3-NEXT:    movq %rax, %rbx
; SSE2-SSSE3-NEXT:    movq %rax, %rcx
; SSE2-SSSE3-NEXT:    movq %rax, %rdx
; SSE2-SSSE3-NEXT:    movq %rax, %rsi
; SSE2-SSSE3-NEXT:    movq %rax, %rdi
; SSE2-SSSE3-NEXT:    movq %rax, %rbp
; SSE2-SSSE3-NEXT:    shrq $15, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm0
; SSE2-SSSE3-NEXT:    movq %rax, %rbp
; SSE2-SSSE3-NEXT:    movsbq %al, %rax
; SSE2-SSSE3-NEXT:    shlq $49, %r8
; SSE2-SSSE3-NEXT:    sarq $63, %r8
; SSE2-SSSE3-NEXT:    movd %r8d, %xmm1
; SSE2-SSSE3-NEXT:    shlq $50, %r9
; SSE2-SSSE3-NEXT:    sarq $63, %r9
; SSE2-SSSE3-NEXT:    movd %r9d, %xmm2
; SSE2-SSSE3-NEXT:    shlq $51, %r10
; SSE2-SSSE3-NEXT:    sarq $63, %r10
; SSE2-SSSE3-NEXT:    movd %r10d, %xmm3
; SSE2-SSSE3-NEXT:    shlq $52, %r11
; SSE2-SSSE3-NEXT:    sarq $63, %r11
; SSE2-SSSE3-NEXT:    movd %r11d, %xmm4
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    shlq $53, %r14
; SSE2-SSSE3-NEXT:    sarq $63, %r14
; SSE2-SSSE3-NEXT:    movd %r14d, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
; SSE2-SSSE3-NEXT:    shlq $54, %r15
; SSE2-SSSE3-NEXT:    sarq $63, %r15
; SSE2-SSSE3-NEXT:    movd %r15d, %xmm2
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
; SSE2-SSSE3-NEXT:    shlq $55, %r12
; SSE2-SSSE3-NEXT:    sarq $63, %r12
; SSE2-SSSE3-NEXT:    movd %r12d, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
; SSE2-SSSE3-NEXT:    shlq $60, %r13
; SSE2-SSSE3-NEXT:    sarq $63, %r13
; SSE2-SSSE3-NEXT:    movd %r13d, %xmm4
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
; SSE2-SSSE3-NEXT:    shlq $61, %rbx
; SSE2-SSSE3-NEXT:    sarq $63, %rbx
; SSE2-SSSE3-NEXT:    movd %ebx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; SSE2-SSSE3-NEXT:    shlq $62, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm5
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
; SSE2-SSSE3-NEXT:    shlq $63, %rdx
; SSE2-SSSE3-NEXT:    sarq $63, %rdx
; SSE2-SSSE3-NEXT:    movd %edx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3],xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7]
; SSE2-SSSE3-NEXT:    shlq $58, %rsi
; SSE2-SSSE3-NEXT:    sarq $63, %rsi
; SSE2-SSSE3-NEXT:    movd %esi, %xmm3
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3],xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
; SSE2-SSSE3-NEXT:    shlq $59, %rdi
; SSE2-SSSE3-NEXT:    sarq $63, %rdi
; SSE2-SSSE3-NEXT:    movd %edi, %xmm4
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3],xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
; SSE2-SSSE3-NEXT:    shlq $57, %rbp
; SSE2-SSSE3-NEXT:    sarq $63, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm2
; SSE2-SSSE3-NEXT:    shrq $7, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm3
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3],xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1],xmm4[2],xmm2[2],xmm4[3],xmm2[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-SSSE3-NEXT:    popq %rbx
; SSE2-SSSE3-NEXT:    popq %r12
; SSE2-SSSE3-NEXT:    popq %r13
; SSE2-SSSE3-NEXT:    popq %r14
; SSE2-SSSE3-NEXT:    popq %r15
; SSE2-SSSE3-NEXT:    popq %rbp
; SSE2-SSSE3-NEXT:    retq
;
; AVX12-LABEL: ext_i16_16i8:
; AVX12:       # BB#0:
; AVX12-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; AVX12-NEXT:    movswq -{{[0-9]+}}(%rsp), %rax
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $62, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    movq %rax, %rdx
; AVX12-NEXT:    shlq $63, %rdx
; AVX12-NEXT:    sarq $63, %rdx
; AVX12-NEXT:    vmovd %edx, %xmm0
; AVX12-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $61, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $60, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $3, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $59, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $4, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $58, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $5, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $57, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $6, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movsbq %al, %rcx
; AVX12-NEXT:    shrq $7, %rcx
; AVX12-NEXT:    vpinsrb $7, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $55, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $8, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $54, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $9, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $53, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $10, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $52, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $11, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $51, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $12, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $50, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $13, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    movq %rax, %rcx
; AVX12-NEXT:    shlq $49, %rcx
; AVX12-NEXT:    sarq $63, %rcx
; AVX12-NEXT:    vpinsrb $14, %ecx, %xmm0, %xmm0
; AVX12-NEXT:    shrq $15, %rax
; AVX12-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
; AVX12-NEXT:    retq
;
; AVX512-LABEL: ext_i16_16i8:
; AVX512:       # BB#0:
; AVX512-NEXT:    kmovd %edi, %k0
; AVX512-NEXT:    vpmovm2b %k0, %xmm0
; AVX512-NEXT:    retq
  %1 = bitcast i16 %a0 to <16 x i1>
  %2 = sext <16 x i1> %1 to <16 x i8>
  ret <16 x i8> %2
}

;
; 256-bit vectors
;

define <4 x i64> @ext_i4_4i64(i4 %a0) {
; SSE2-SSSE3-LABEL: ext_i4_4i64:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    andb $15, %dil
; SSE2-SSSE3-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $3, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $2, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; SSE2-SSSE3-NEXT:    movd %eax, %xmm2
; SSE2-SSSE3-NEXT:    shrl %eax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm0
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm1[0]
; SSE2-SSSE3-NEXT:    pand {{.*}}(%rip), %xmm2
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,1,1,3]
; SSE2-SSSE3-NEXT:    psllq $63, %xmm0
; SSE2-SSSE3-NEXT:    psrad $31, %xmm0
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,1,3,3]
; SSE2-SSSE3-NEXT:    psllq $63, %xmm1
; SSE2-SSSE3-NEXT:    psrad $31, %xmm1
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; SSE2-SSSE3-NEXT:    retq
;
; AVX1-LABEL: ext_i4_4i64:
; AVX1:       # BB#0:
; AVX1-NEXT:    andb $15, %dil
; AVX1-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX1-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; AVX1-NEXT:    movq %rax, %rcx
; AVX1-NEXT:    shlq $60, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vmovq %rcx, %xmm0
; AVX1-NEXT:    movq %rax, %rcx
; AVX1-NEXT:    shlq $61, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vmovq %rcx, %xmm1
; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX1-NEXT:    movq %rax, %rcx
; AVX1-NEXT:    shlq $62, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vmovq %rcx, %xmm1
; AVX1-NEXT:    shlq $63, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vmovq %rax, %xmm2
; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: ext_i4_4i64:
; AVX2:       # BB#0:
; AVX2-NEXT:    andb $15, %dil
; AVX2-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; AVX2-NEXT:    movq %rax, %rcx
; AVX2-NEXT:    shlq $60, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vmovq %rcx, %xmm0
; AVX2-NEXT:    movq %rax, %rcx
; AVX2-NEXT:    shlq $61, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vmovq %rcx, %xmm1
; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX2-NEXT:    movq %rax, %rcx
; AVX2-NEXT:    shlq $62, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vmovq %rcx, %xmm1
; AVX2-NEXT:    shlq $63, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vmovq %rax, %xmm2
; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
; AVX2-NEXT:    retq
;
; AVX512-LABEL: ext_i4_4i64:
; AVX512:       # BB#0:
; AVX512-NEXT:    andb $15, %dil
; AVX512-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX512-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; AVX512-NEXT:    kmovd %eax, %k1
; AVX512-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
; AVX512-NEXT:    # kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
; AVX512-NEXT:    retq
  %1 = bitcast i4 %a0 to <4 x i1>
  %2 = sext <4 x i1> %1 to <4 x i64>
  ret <4 x i64> %2
}

define <8 x i32> @ext_i8_8i32(i8 %a0) {
; SSE2-SSSE3-LABEL: ext_i8_8i32:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $3, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $2, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $5, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $4, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $6, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    shrl $7, %eax
; SSE2-SSSE3-NEXT:    movzwl %ax, %eax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm3
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
; SSE2-SSSE3-NEXT:    movdqa %xmm1, %xmm0
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
; SSE2-SSSE3-NEXT:    pslld $31, %xmm0
; SSE2-SSSE3-NEXT:    psrad $31, %xmm0
; SSE2-SSSE3-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    pslld $31, %xmm1
; SSE2-SSSE3-NEXT:    psrad $31, %xmm1
; SSE2-SSSE3-NEXT:    retq
;
; AVX1-LABEL: ext_i8_8i32:
; AVX1:       # BB#0:
; AVX1-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX1-NEXT:    movsbq -{{[0-9]+}}(%rsp), %rax
; AVX1-NEXT:    movq %rax, %rcx
; AVX1-NEXT:    shlq $58, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    movq %rax, %rdx
; AVX1-NEXT:    shlq $59, %rdx
; AVX1-NEXT:    sarq $63, %rdx
; AVX1-NEXT:    vmovd %edx, %xmm0
; AVX1-NEXT:    vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movq %rax, %rcx
; AVX1-NEXT:    shlq $57, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vpinsrd $2, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movq %rax, %rcx
; AVX1-NEXT:    shrq $7, %rcx
; AVX1-NEXT:    vpinsrd $3, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movq %rax, %rcx
; AVX1-NEXT:    shlq $62, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    movq %rax, %rdx
; AVX1-NEXT:    shlq $63, %rdx
; AVX1-NEXT:    sarq $63, %rdx
; AVX1-NEXT:    vmovd %edx, %xmm1
; AVX1-NEXT:    vpinsrd $1, %ecx, %xmm1, %xmm1
; AVX1-NEXT:    movq %rax, %rcx
; AVX1-NEXT:    shlq $61, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vpinsrd $2, %ecx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $60, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrd $3, %eax, %xmm1, %xmm1
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: ext_i8_8i32:
; AVX2:       # BB#0:
; AVX2-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    movsbq -{{[0-9]+}}(%rsp), %rax
; AVX2-NEXT:    movq %rax, %rcx
; AVX2-NEXT:    shlq $58, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    movq %rax, %rdx
; AVX2-NEXT:    shlq $59, %rdx
; AVX2-NEXT:    sarq $63, %rdx
; AVX2-NEXT:    vmovd %edx, %xmm0
; AVX2-NEXT:    vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movq %rax, %rcx
; AVX2-NEXT:    shlq $57, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vpinsrd $2, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movq %rax, %rcx
; AVX2-NEXT:    shrq $7, %rcx
; AVX2-NEXT:    vpinsrd $3, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movq %rax, %rcx
; AVX2-NEXT:    shlq $62, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    movq %rax, %rdx
; AVX2-NEXT:    shlq $63, %rdx
; AVX2-NEXT:    sarq $63, %rdx
; AVX2-NEXT:    vmovd %edx, %xmm1
; AVX2-NEXT:    vpinsrd $1, %ecx, %xmm1, %xmm1
; AVX2-NEXT:    movq %rax, %rcx
; AVX2-NEXT:    shlq $61, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vpinsrd $2, %ecx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $60, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrd $3, %eax, %xmm1, %xmm1
; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
; AVX2-NEXT:    retq
;
; AVX512-LABEL: ext_i8_8i32:
; AVX512:       # BB#0:
; AVX512-NEXT:    kmovd %edi, %k1
; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; AVX512-NEXT:    retq
  %1 = bitcast i8 %a0 to <8 x i1>
  %2 = sext <8 x i1> %1 to <8 x i32>
  ret <8 x i32> %2
}

define <16 x i16> @ext_i16_16i16(i16 %a0) {
; SSE2-SSSE3-LABEL: ext_i16_16i16:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $7, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $6, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $5, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $4, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $3, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $2, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm3
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $11, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $10, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $9, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm3
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $8, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $13, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $12, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm3
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $14, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    shrl $15, %eax
; SSE2-SSSE3-NEXT:    movzwl %ax, %eax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm4
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3],xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE2-SSSE3-NEXT:    movdqa %xmm1, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-SSSE3-NEXT:    psllw $15, %xmm0
; SSE2-SSSE3-NEXT:    psraw $15, %xmm0
; SSE2-SSSE3-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
; SSE2-SSSE3-NEXT:    psllw $15, %xmm1
; SSE2-SSSE3-NEXT:    psraw $15, %xmm1
; SSE2-SSSE3-NEXT:    retq
;
; AVX1-LABEL: ext_i16_16i16:
; AVX1:       # BB#0:
; AVX1-NEXT:    pushq %rbp
; AVX1-NEXT:  .Lcfi0:
; AVX1-NEXT:    .cfi_def_cfa_offset 16
; AVX1-NEXT:    pushq %r15
; AVX1-NEXT:  .Lcfi1:
; AVX1-NEXT:    .cfi_def_cfa_offset 24
; AVX1-NEXT:    pushq %r14
; AVX1-NEXT:  .Lcfi2:
; AVX1-NEXT:    .cfi_def_cfa_offset 32
; AVX1-NEXT:    pushq %r13
; AVX1-NEXT:  .Lcfi3:
; AVX1-NEXT:    .cfi_def_cfa_offset 40
; AVX1-NEXT:    pushq %r12
; AVX1-NEXT:  .Lcfi4:
; AVX1-NEXT:    .cfi_def_cfa_offset 48
; AVX1-NEXT:    pushq %rbx
; AVX1-NEXT:  .Lcfi5:
; AVX1-NEXT:    .cfi_def_cfa_offset 56
; AVX1-NEXT:  .Lcfi6:
; AVX1-NEXT:    .cfi_offset %rbx, -56
; AVX1-NEXT:  .Lcfi7:
; AVX1-NEXT:    .cfi_offset %r12, -48
; AVX1-NEXT:  .Lcfi8:
; AVX1-NEXT:    .cfi_offset %r13, -40
; AVX1-NEXT:  .Lcfi9:
; AVX1-NEXT:    .cfi_offset %r14, -32
; AVX1-NEXT:  .Lcfi10:
; AVX1-NEXT:    .cfi_offset %r15, -24
; AVX1-NEXT:  .Lcfi11:
; AVX1-NEXT:    .cfi_offset %rbp, -16
; AVX1-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; AVX1-NEXT:    movswq -{{[0-9]+}}(%rsp), %rax
; AVX1-NEXT:    movq %rax, %rcx
; AVX1-NEXT:    shlq $55, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vmovd %ecx, %xmm0
; AVX1-NEXT:    movq %rax, %r8
; AVX1-NEXT:    movq %rax, %r10
; AVX1-NEXT:    movq %rax, %r11
; AVX1-NEXT:    movq %rax, %r14
; AVX1-NEXT:    movq %rax, %r15
; AVX1-NEXT:    movq %rax, %r9
; AVX1-NEXT:    movq %rax, %r12
; AVX1-NEXT:    movq %rax, %r13
; AVX1-NEXT:    movq %rax, %rbx
; AVX1-NEXT:    movq %rax, %rdi
; AVX1-NEXT:    movq %rax, %rcx
; AVX1-NEXT:    movq %rax, %rdx
; AVX1-NEXT:    movq %rax, %rsi
; AVX1-NEXT:    movsbq %al, %rbp
; AVX1-NEXT:    shlq $54, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrw $1, %eax, %xmm0, %xmm0
; AVX1-NEXT:    shlq $53, %r8
; AVX1-NEXT:    sarq $63, %r8
; AVX1-NEXT:    vpinsrw $2, %r8d, %xmm0, %xmm0
; AVX1-NEXT:    shlq $52, %r10
; AVX1-NEXT:    sarq $63, %r10
; AVX1-NEXT:    vpinsrw $3, %r10d, %xmm0, %xmm0
; AVX1-NEXT:    shlq $51, %r11
; AVX1-NEXT:    sarq $63, %r11
; AVX1-NEXT:    vpinsrw $4, %r11d, %xmm0, %xmm0
; AVX1-NEXT:    shlq $50, %r14
; AVX1-NEXT:    sarq $63, %r14
; AVX1-NEXT:    vpinsrw $5, %r14d, %xmm0, %xmm0
; AVX1-NEXT:    shlq $49, %r15
; AVX1-NEXT:    sarq $63, %r15
; AVX1-NEXT:    vpinsrw $6, %r15d, %xmm0, %xmm0
; AVX1-NEXT:    shrq $15, %r9
; AVX1-NEXT:    vpinsrw $7, %r9d, %xmm0, %xmm0
; AVX1-NEXT:    shlq $63, %r13
; AVX1-NEXT:    sarq $63, %r13
; AVX1-NEXT:    vmovd %r13d, %xmm1
; AVX1-NEXT:    shlq $62, %r12
; AVX1-NEXT:    sarq $63, %r12
; AVX1-NEXT:    vpinsrw $1, %r12d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $61, %rbx
; AVX1-NEXT:    sarq $63, %rbx
; AVX1-NEXT:    vpinsrw $2, %ebx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $60, %rdi
; AVX1-NEXT:    sarq $63, %rdi
; AVX1-NEXT:    vpinsrw $3, %edi, %xmm1, %xmm1
; AVX1-NEXT:    shlq $59, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vpinsrw $4, %ecx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $58, %rdx
; AVX1-NEXT:    sarq $63, %rdx
; AVX1-NEXT:    vpinsrw $5, %edx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $57, %rsi
; AVX1-NEXT:    sarq $63, %rsi
; AVX1-NEXT:    vpinsrw $6, %esi, %xmm1, %xmm1
; AVX1-NEXT:    shrq $7, %rbp
; AVX1-NEXT:    vpinsrw $7, %ebp, %xmm1, %xmm1
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT:    popq %rbx
; AVX1-NEXT:    popq %r12
; AVX1-NEXT:    popq %r13
; AVX1-NEXT:    popq %r14
; AVX1-NEXT:    popq %r15
; AVX1-NEXT:    popq %rbp
; AVX1-NEXT:    retq
;
; AVX2-LABEL: ext_i16_16i16:
; AVX2:       # BB#0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:  .Lcfi0:
; AVX2-NEXT:    .cfi_def_cfa_offset 16
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:  .Lcfi1:
; AVX2-NEXT:    .cfi_def_cfa_offset 24
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:  .Lcfi2:
; AVX2-NEXT:    .cfi_def_cfa_offset 32
; AVX2-NEXT:    pushq %r13
; AVX2-NEXT:  .Lcfi3:
; AVX2-NEXT:    .cfi_def_cfa_offset 40
; AVX2-NEXT:    pushq %r12
; AVX2-NEXT:  .Lcfi4:
; AVX2-NEXT:    .cfi_def_cfa_offset 48
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:  .Lcfi5:
; AVX2-NEXT:    .cfi_def_cfa_offset 56
; AVX2-NEXT:  .Lcfi6:
; AVX2-NEXT:    .cfi_offset %rbx, -56
; AVX2-NEXT:  .Lcfi7:
; AVX2-NEXT:    .cfi_offset %r12, -48
; AVX2-NEXT:  .Lcfi8:
; AVX2-NEXT:    .cfi_offset %r13, -40
; AVX2-NEXT:  .Lcfi9:
; AVX2-NEXT:    .cfi_offset %r14, -32
; AVX2-NEXT:  .Lcfi10:
; AVX2-NEXT:    .cfi_offset %r15, -24
; AVX2-NEXT:  .Lcfi11:
; AVX2-NEXT:    .cfi_offset %rbp, -16
; AVX2-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    movswq -{{[0-9]+}}(%rsp), %rax
; AVX2-NEXT:    movq %rax, %rcx
; AVX2-NEXT:    shlq $55, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vmovd %ecx, %xmm0
; AVX2-NEXT:    movq %rax, %r8
; AVX2-NEXT:    movq %rax, %r10
; AVX2-NEXT:    movq %rax, %r11
; AVX2-NEXT:    movq %rax, %r14
; AVX2-NEXT:    movq %rax, %r15
; AVX2-NEXT:    movq %rax, %r9
; AVX2-NEXT:    movq %rax, %r12
; AVX2-NEXT:    movq %rax, %r13
; AVX2-NEXT:    movq %rax, %rbx
; AVX2-NEXT:    movq %rax, %rdi
; AVX2-NEXT:    movq %rax, %rcx
; AVX2-NEXT:    movq %rax, %rdx
; AVX2-NEXT:    movq %rax, %rsi
; AVX2-NEXT:    movsbq %al, %rbp
; AVX2-NEXT:    shlq $54, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrw $1, %eax, %xmm0, %xmm0
; AVX2-NEXT:    shlq $53, %r8
; AVX2-NEXT:    sarq $63, %r8
; AVX2-NEXT:    vpinsrw $2, %r8d, %xmm0, %xmm0
; AVX2-NEXT:    shlq $52, %r10
; AVX2-NEXT:    sarq $63, %r10
; AVX2-NEXT:    vpinsrw $3, %r10d, %xmm0, %xmm0
; AVX2-NEXT:    shlq $51, %r11
; AVX2-NEXT:    sarq $63, %r11
; AVX2-NEXT:    vpinsrw $4, %r11d, %xmm0, %xmm0
; AVX2-NEXT:    shlq $50, %r14
; AVX2-NEXT:    sarq $63, %r14
; AVX2-NEXT:    vpinsrw $5, %r14d, %xmm0, %xmm0
; AVX2-NEXT:    shlq $49, %r15
; AVX2-NEXT:    sarq $63, %r15
; AVX2-NEXT:    vpinsrw $6, %r15d, %xmm0, %xmm0
; AVX2-NEXT:    shrq $15, %r9
; AVX2-NEXT:    vpinsrw $7, %r9d, %xmm0, %xmm0
; AVX2-NEXT:    shlq $63, %r13
; AVX2-NEXT:    sarq $63, %r13
; AVX2-NEXT:    vmovd %r13d, %xmm1
; AVX2-NEXT:    shlq $62, %r12
; AVX2-NEXT:    sarq $63, %r12
; AVX2-NEXT:    vpinsrw $1, %r12d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $61, %rbx
; AVX2-NEXT:    sarq $63, %rbx
; AVX2-NEXT:    vpinsrw $2, %ebx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $60, %rdi
; AVX2-NEXT:    sarq $63, %rdi
; AVX2-NEXT:    vpinsrw $3, %edi, %xmm1, %xmm1
; AVX2-NEXT:    shlq $59, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vpinsrw $4, %ecx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $58, %rdx
; AVX2-NEXT:    sarq $63, %rdx
; AVX2-NEXT:    vpinsrw $5, %edx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $57, %rsi
; AVX2-NEXT:    sarq $63, %rsi
; AVX2-NEXT:    vpinsrw $6, %esi, %xmm1, %xmm1
; AVX2-NEXT:    shrq $7, %rbp
; AVX2-NEXT:    vpinsrw $7, %ebp, %xmm1, %xmm1
; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r12
; AVX2-NEXT:    popq %r13
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512-LABEL: ext_i16_16i16:
; AVX512:       # BB#0:
; AVX512-NEXT:    kmovd %edi, %k0
; AVX512-NEXT:    vpmovm2w %k0, %ymm0
; AVX512-NEXT:    retq
  %1 = bitcast i16 %a0 to <16 x i1>
  %2 = sext <16 x i1> %1 to <16 x i16>
  ret <16 x i16> %2
}

define <32 x i8> @ext_i32_32i8(i32 %a0) {
; SSE2-SSSE3-LABEL: ext_i32_32i8:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    pushq %rbp
; SSE2-SSSE3-NEXT:  .Lcfi12:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 16
; SSE2-SSSE3-NEXT:    pushq %r15
; SSE2-SSSE3-NEXT:  .Lcfi13:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 24
; SSE2-SSSE3-NEXT:    pushq %r14
; SSE2-SSSE3-NEXT:  .Lcfi14:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 32
; SSE2-SSSE3-NEXT:    pushq %r13
; SSE2-SSSE3-NEXT:  .Lcfi15:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 40
; SSE2-SSSE3-NEXT:    pushq %r12
; SSE2-SSSE3-NEXT:  .Lcfi16:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 48
; SSE2-SSSE3-NEXT:    pushq %rbx
; SSE2-SSSE3-NEXT:  .Lcfi17:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 56
; SSE2-SSSE3-NEXT:  .Lcfi18:
; SSE2-SSSE3-NEXT:    .cfi_offset %rbx, -56
; SSE2-SSSE3-NEXT:  .Lcfi19:
; SSE2-SSSE3-NEXT:    .cfi_offset %r12, -48
; SSE2-SSSE3-NEXT:  .Lcfi20:
; SSE2-SSSE3-NEXT:    .cfi_offset %r13, -40
; SSE2-SSSE3-NEXT:  .Lcfi21:
; SSE2-SSSE3-NEXT:    .cfi_offset %r14, -32
; SSE2-SSSE3-NEXT:  .Lcfi22:
; SSE2-SSSE3-NEXT:    .cfi_offset %r15, -24
; SSE2-SSSE3-NEXT:  .Lcfi23:
; SSE2-SSSE3-NEXT:    .cfi_offset %rbp, -16
; SSE2-SSSE3-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    shrl $16, %edi
; SSE2-SSSE3-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movswq -{{[0-9]+}}(%rsp), %rbx
; SSE2-SSSE3-NEXT:    movq %rbx, %r8
; SSE2-SSSE3-NEXT:    movq %rbx, %r9
; SSE2-SSSE3-NEXT:    movq %rbx, %r10
; SSE2-SSSE3-NEXT:    movq %rbx, %r11
; SSE2-SSSE3-NEXT:    movq %rbx, %r14
; SSE2-SSSE3-NEXT:    movq %rbx, %r15
; SSE2-SSSE3-NEXT:    movq %rbx, %r12
; SSE2-SSSE3-NEXT:    movq %rbx, %r13
; SSE2-SSSE3-NEXT:    movq %rbx, %rdi
; SSE2-SSSE3-NEXT:    movq %rbx, %rcx
; SSE2-SSSE3-NEXT:    movq %rbx, %rdx
; SSE2-SSSE3-NEXT:    movq %rbx, %rbp
; SSE2-SSSE3-NEXT:    movq %rbx, %rsi
; SSE2-SSSE3-NEXT:    movq %rbx, %rax
; SSE2-SSSE3-NEXT:    shrq $15, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm0
; SSE2-SSSE3-NEXT:    movq %rbx, %rax
; SSE2-SSSE3-NEXT:    movsbq %bl, %rbx
; SSE2-SSSE3-NEXT:    shlq $49, %r8
; SSE2-SSSE3-NEXT:    sarq $63, %r8
; SSE2-SSSE3-NEXT:    movd %r8d, %xmm15
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0],xmm0[0],xmm15[1],xmm0[1],xmm15[2],xmm0[2],xmm15[3],xmm0[3],xmm15[4],xmm0[4],xmm15[5],xmm0[5],xmm15[6],xmm0[6],xmm15[7],xmm0[7]
; SSE2-SSSE3-NEXT:    shlq $50, %r9
; SSE2-SSSE3-NEXT:    sarq $63, %r9
; SSE2-SSSE3-NEXT:    movd %r9d, %xmm8
; SSE2-SSSE3-NEXT:    shlq $51, %r10
; SSE2-SSSE3-NEXT:    sarq $63, %r10
; SSE2-SSSE3-NEXT:    movd %r10d, %xmm3
; SSE2-SSSE3-NEXT:    shlq $52, %r11
; SSE2-SSSE3-NEXT:    sarq $63, %r11
; SSE2-SSSE3-NEXT:    movd %r11d, %xmm9
; SSE2-SSSE3-NEXT:    shlq $53, %r14
; SSE2-SSSE3-NEXT:    sarq $63, %r14
; SSE2-SSSE3-NEXT:    movd %r14d, %xmm6
; SSE2-SSSE3-NEXT:    shlq $54, %r15
; SSE2-SSSE3-NEXT:    sarq $63, %r15
; SSE2-SSSE3-NEXT:    movd %r15d, %xmm10
; SSE2-SSSE3-NEXT:    shlq $55, %r12
; SSE2-SSSE3-NEXT:    sarq $63, %r12
; SSE2-SSSE3-NEXT:    movd %r12d, %xmm1
; SSE2-SSSE3-NEXT:    shlq $60, %r13
; SSE2-SSSE3-NEXT:    sarq $63, %r13
; SSE2-SSSE3-NEXT:    movd %r13d, %xmm11
; SSE2-SSSE3-NEXT:    shlq $61, %rdi
; SSE2-SSSE3-NEXT:    sarq $63, %rdi
; SSE2-SSSE3-NEXT:    movd %edi, %xmm5
; SSE2-SSSE3-NEXT:    shlq $62, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm12
; SSE2-SSSE3-NEXT:    shlq $63, %rdx
; SSE2-SSSE3-NEXT:    sarq $63, %rdx
; SSE2-SSSE3-NEXT:    movd %edx, %xmm0
; SSE2-SSSE3-NEXT:    shlq $58, %rbp
; SSE2-SSSE3-NEXT:    sarq $63, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm13
; SSE2-SSSE3-NEXT:    shlq $59, %rsi
; SSE2-SSSE3-NEXT:    sarq $63, %rsi
; SSE2-SSSE3-NEXT:    movd %esi, %xmm7
; SSE2-SSSE3-NEXT:    shlq $57, %rax
; SSE2-SSSE3-NEXT:    sarq $63, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm4
; SSE2-SSSE3-NEXT:    shrq $7, %rbx
; SSE2-SSSE3-NEXT:    movd %ebx, %xmm14
; SSE2-SSSE3-NEXT:    movswq -{{[0-9]+}}(%rsp), %rsi
; SSE2-SSSE3-NEXT:    movq %rsi, %r8
; SSE2-SSSE3-NEXT:    movq %rsi, %r9
; SSE2-SSSE3-NEXT:    movq %rsi, %r10
; SSE2-SSSE3-NEXT:    movq %rsi, %r11
; SSE2-SSSE3-NEXT:    movq %rsi, %r14
; SSE2-SSSE3-NEXT:    movq %rsi, %r15
; SSE2-SSSE3-NEXT:    movq %rsi, %r12
; SSE2-SSSE3-NEXT:    movq %rsi, %r13
; SSE2-SSSE3-NEXT:    movq %rsi, %rbx
; SSE2-SSSE3-NEXT:    movq %rsi, %rax
; SSE2-SSSE3-NEXT:    movq %rsi, %rcx
; SSE2-SSSE3-NEXT:    movq %rsi, %rdx
; SSE2-SSSE3-NEXT:    movq %rsi, %rdi
; SSE2-SSSE3-NEXT:    movq %rsi, %rbp
; SSE2-SSSE3-NEXT:    shrq $15, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm2
; SSE2-SSSE3-NEXT:    movq %rsi, %rbp
; SSE2-SSSE3-NEXT:    movsbq %sil, %rsi
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm8[0],xmm3[1],xmm8[1],xmm3[2],xmm8[2],xmm3[3],xmm8[3],xmm3[4],xmm8[4],xmm3[5],xmm8[5],xmm3[6],xmm8[6],xmm3[7],xmm8[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm15[0],xmm3[1],xmm15[1],xmm3[2],xmm15[2],xmm3[3],xmm15[3]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm9[0],xmm6[1],xmm9[1],xmm6[2],xmm9[2],xmm6[3],xmm9[3],xmm6[4],xmm9[4],xmm6[5],xmm9[5],xmm6[6],xmm9[6],xmm6[7],xmm9[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm10[0],xmm1[1],xmm10[1],xmm1[2],xmm10[2],xmm1[3],xmm10[3],xmm1[4],xmm10[4],xmm1[5],xmm10[5],xmm1[6],xmm10[6],xmm1[7],xmm10[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm6[0],xmm1[1],xmm6[1],xmm1[2],xmm6[2],xmm1[3],xmm6[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm11[0],xmm5[1],xmm11[1],xmm5[2],xmm11[2],xmm5[3],xmm11[3],xmm5[4],xmm11[4],xmm5[5],xmm11[5],xmm5[6],xmm11[6],xmm5[7],xmm11[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm12[0],xmm0[1],xmm12[1],xmm0[2],xmm12[2],xmm0[3],xmm12[3],xmm0[4],xmm12[4],xmm0[5],xmm12[5],xmm0[6],xmm12[6],xmm0[7],xmm12[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm13[0],xmm7[1],xmm13[1],xmm7[2],xmm13[2],xmm7[3],xmm13[3],xmm7[4],xmm13[4],xmm7[5],xmm13[5],xmm7[6],xmm13[6],xmm7[7],xmm13[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm14[0],xmm4[1],xmm14[1],xmm4[2],xmm14[2],xmm4[3],xmm14[3],xmm4[4],xmm14[4],xmm4[5],xmm14[5],xmm4[6],xmm14[6],xmm4[7],xmm14[7]
; SSE2-SSSE3-NEXT:    shlq $49, %r8
; SSE2-SSSE3-NEXT:    sarq $63, %r8
; SSE2-SSSE3-NEXT:    movd %r8d, %xmm3
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm7 = xmm7[0],xmm4[0],xmm7[1],xmm4[1],xmm7[2],xmm4[2],xmm7[3],xmm4[3]
; SSE2-SSSE3-NEXT:    shlq $50, %r9
; SSE2-SSSE3-NEXT:    sarq $63, %r9
; SSE2-SSSE3-NEXT:    movd %r9d, %xmm4
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[1],xmm7[1]
; SSE2-SSSE3-NEXT:    shlq $51, %r10
; SSE2-SSSE3-NEXT:    sarq $63, %r10
; SSE2-SSSE3-NEXT:    movd %r10d, %xmm5
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE2-SSSE3-NEXT:    shlq $52, %r11
; SSE2-SSSE3-NEXT:    sarq $63, %r11
; SSE2-SSSE3-NEXT:    movd %r11d, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
; SSE2-SSSE3-NEXT:    shlq $53, %r14
; SSE2-SSSE3-NEXT:    sarq $63, %r14
; SSE2-SSSE3-NEXT:    movd %r14d, %xmm2
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3],xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
; SSE2-SSSE3-NEXT:    shlq $54, %r15
; SSE2-SSSE3-NEXT:    sarq $63, %r15
; SSE2-SSSE3-NEXT:    movd %r15d, %xmm4
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm3[0],xmm5[1],xmm3[1],xmm5[2],xmm3[2],xmm5[3],xmm3[3]
; SSE2-SSSE3-NEXT:    shlq $55, %r12
; SSE2-SSSE3-NEXT:    sarq $63, %r12
; SSE2-SSSE3-NEXT:    movd %r12d, %xmm3
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
; SSE2-SSSE3-NEXT:    shlq $60, %r13
; SSE2-SSSE3-NEXT:    sarq $63, %r13
; SSE2-SSSE3-NEXT:    movd %r13d, %xmm6
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
; SSE2-SSSE3-NEXT:    shlq $61, %rbx
; SSE2-SSSE3-NEXT:    sarq $63, %rbx
; SSE2-SSSE3-NEXT:    movd %ebx, %xmm4
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
; SSE2-SSSE3-NEXT:    shlq $62, %rax
; SSE2-SSSE3-NEXT:    sarq $63, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm2
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
; SSE2-SSSE3-NEXT:    shlq $63, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1],xmm4[2],xmm6[2],xmm4[3],xmm6[3],xmm4[4],xmm6[4],xmm4[5],xmm6[5],xmm4[6],xmm6[6],xmm4[7],xmm6[7]
; SSE2-SSSE3-NEXT:    shlq $58, %rdx
; SSE2-SSSE3-NEXT:    sarq $63, %rdx
; SSE2-SSSE3-NEXT:    movd %edx, %xmm5
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
; SSE2-SSSE3-NEXT:    shlq $59, %rdi
; SSE2-SSSE3-NEXT:    sarq $63, %rdi
; SSE2-SSSE3-NEXT:    movd %edi, %xmm2
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1],xmm2[2],xmm5[2],xmm2[3],xmm5[3],xmm2[4],xmm5[4],xmm2[5],xmm5[5],xmm2[6],xmm5[6],xmm2[7],xmm5[7]
; SSE2-SSSE3-NEXT:    shlq $57, %rbp
; SSE2-SSSE3-NEXT:    sarq $63, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm4
; SSE2-SSSE3-NEXT:    shrq $7, %rsi
; SSE2-SSSE3-NEXT:    movd %esi, %xmm5
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3],xmm4[4],xmm5[4],xmm4[5],xmm5[5],xmm4[6],xmm5[6],xmm4[7],xmm5[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
; SSE2-SSSE3-NEXT:    popq %rbx
; SSE2-SSSE3-NEXT:    popq %r12
; SSE2-SSSE3-NEXT:    popq %r13
; SSE2-SSSE3-NEXT:    popq %r14
; SSE2-SSSE3-NEXT:    popq %r15
; SSE2-SSSE3-NEXT:    popq %rbp
; SSE2-SSSE3-NEXT:    retq
;
; AVX1-LABEL: ext_i32_32i8:
; AVX1:       # BB#0:
; AVX1-NEXT:    pushq %rbp
; AVX1-NEXT:  .Lcfi12:
; AVX1-NEXT:    .cfi_def_cfa_offset 16
; AVX1-NEXT:  .Lcfi13:
; AVX1-NEXT:    .cfi_offset %rbp, -16
; AVX1-NEXT:    movq %rsp, %rbp
; AVX1-NEXT:  .Lcfi14:
; AVX1-NEXT:    .cfi_def_cfa_register %rbp
; AVX1-NEXT:    pushq %r15
; AVX1-NEXT:    pushq %r14
; AVX1-NEXT:    pushq %r13
; AVX1-NEXT:    pushq %r12
; AVX1-NEXT:    pushq %rbx
; AVX1-NEXT:    andq $-32, %rsp
; AVX1-NEXT:    subq $64, %rsp
; AVX1-NEXT:  .Lcfi15:
; AVX1-NEXT:    .cfi_offset %rbx, -56
; AVX1-NEXT:  .Lcfi16:
; AVX1-NEXT:    .cfi_offset %r12, -48
; AVX1-NEXT:  .Lcfi17:
; AVX1-NEXT:    .cfi_offset %r13, -40
; AVX1-NEXT:  .Lcfi18:
; AVX1-NEXT:    .cfi_offset %r14, -32
; AVX1-NEXT:  .Lcfi19:
; AVX1-NEXT:    .cfi_offset %r15, -24
; AVX1-NEXT:    movl %edi, (%rsp)
; AVX1-NEXT:    movslq (%rsp), %rdx
; AVX1-NEXT:    movq %rdx, %rcx
; AVX1-NEXT:    shlq $47, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vmovd %ecx, %xmm0
; AVX1-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    movq %rdx, %r8
; AVX1-NEXT:    movq %rdx, %rcx
; AVX1-NEXT:    movq %rdx, %rdi
; AVX1-NEXT:    movq %rdx, %r13
; AVX1-NEXT:    movq %rdx, %rsi
; AVX1-NEXT:    movq %rdx, %r10
; AVX1-NEXT:    movq %rdx, %r11
; AVX1-NEXT:    movq %rdx, %r9
; AVX1-NEXT:    movq %rdx, %rbx
; AVX1-NEXT:    movq %rdx, %r14
; AVX1-NEXT:    movq %rdx, %r15
; AVX1-NEXT:    movq %rdx, %r12
; AVX1-NEXT:    movq %rdx, %rax
; AVX1-NEXT:    shlq $46, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX1-NEXT:    shlq $45, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    shlq $44, %r8
; AVX1-NEXT:    sarq $63, %r8
; AVX1-NEXT:    vpinsrb $3, %r8d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r8
; AVX1-NEXT:    shlq $43, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vpinsrb $4, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %rcx
; AVX1-NEXT:    shlq $42, %rdi
; AVX1-NEXT:    sarq $63, %rdi
; AVX1-NEXT:    vpinsrb $5, %edi, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %rdi
; AVX1-NEXT:    shlq $41, %r13
; AVX1-NEXT:    sarq $63, %r13
; AVX1-NEXT:    vpinsrb $6, %r13d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r13
; AVX1-NEXT:    shlq $40, %rsi
; AVX1-NEXT:    sarq $63, %rsi
; AVX1-NEXT:    vpinsrb $7, %esi, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %rsi
; AVX1-NEXT:    shlq $39, %r10
; AVX1-NEXT:    sarq $63, %r10
; AVX1-NEXT:    vpinsrb $8, %r10d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r10
; AVX1-NEXT:    shlq $38, %r11
; AVX1-NEXT:    sarq $63, %r11
; AVX1-NEXT:    vpinsrb $9, %r11d, %xmm0, %xmm0
; AVX1-NEXT:    movsbq %dl, %rax
; AVX1-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    shlq $37, %r9
; AVX1-NEXT:    sarq $63, %r9
; AVX1-NEXT:    vpinsrb $10, %r9d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r9
; AVX1-NEXT:    shlq $36, %rbx
; AVX1-NEXT:    sarq $63, %rbx
; AVX1-NEXT:    vpinsrb $11, %ebx, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %rbx
; AVX1-NEXT:    shlq $35, %r14
; AVX1-NEXT:    sarq $63, %r14
; AVX1-NEXT:    vpinsrb $12, %r14d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r14
; AVX1-NEXT:    shlq $34, %r15
; AVX1-NEXT:    sarq $63, %r15
; AVX1-NEXT:    vpinsrb $13, %r15d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r15
; AVX1-NEXT:    shlq $33, %r12
; AVX1-NEXT:    sarq $63, %r12
; AVX1-NEXT:    vpinsrb $14, %r12d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r12
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX1-NEXT:    shrq $31, %rax
; AVX1-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %rax
; AVX1-NEXT:    shlq $63, %r8
; AVX1-NEXT:    sarq $63, %r8
; AVX1-NEXT:    vmovd %r8d, %xmm1
; AVX1-NEXT:    movq %rdx, %r8
; AVX1-NEXT:    movswq %dx, %rdx
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
; AVX1-NEXT:    shlq $62, %r11
; AVX1-NEXT:    sarq $63, %r11
; AVX1-NEXT:    vpinsrb $1, %r11d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $61, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vpinsrb $2, %ecx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $60, %rdi
; AVX1-NEXT:    sarq $63, %rdi
; AVX1-NEXT:    vpinsrb $3, %edi, %xmm1, %xmm1
; AVX1-NEXT:    shlq $59, %r13
; AVX1-NEXT:    sarq $63, %r13
; AVX1-NEXT:    vpinsrb $4, %r13d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $58, %rsi
; AVX1-NEXT:    sarq $63, %rsi
; AVX1-NEXT:    vpinsrb $5, %esi, %xmm1, %xmm1
; AVX1-NEXT:    shlq $57, %r10
; AVX1-NEXT:    sarq $63, %r10
; AVX1-NEXT:    vpinsrb $6, %r10d, %xmm1, %xmm1
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
; AVX1-NEXT:    shrq $7, %rcx
; AVX1-NEXT:    vpinsrb $7, %ecx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $55, %r9
; AVX1-NEXT:    sarq $63, %r9
; AVX1-NEXT:    vpinsrb $8, %r9d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $54, %rbx
; AVX1-NEXT:    sarq $63, %rbx
; AVX1-NEXT:    vpinsrb $9, %ebx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $53, %r14
; AVX1-NEXT:    sarq $63, %r14
; AVX1-NEXT:    vpinsrb $10, %r14d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $52, %r15
; AVX1-NEXT:    sarq $63, %r15
; AVX1-NEXT:    vpinsrb $11, %r15d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $51, %r12
; AVX1-NEXT:    sarq $63, %r12
; AVX1-NEXT:    vpinsrb $12, %r12d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $50, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
; AVX1-NEXT:    shlq $49, %r8
; AVX1-NEXT:    sarq $63, %r8
; AVX1-NEXT:    vpinsrb $14, %r8d, %xmm1, %xmm1
; AVX1-NEXT:    shrq $15, %rdx
; AVX1-NEXT:    vpinsrb $15, %edx, %xmm1, %xmm1
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT:    leaq -40(%rbp), %rsp
; AVX1-NEXT:    popq %rbx
; AVX1-NEXT:    popq %r12
; AVX1-NEXT:    popq %r13
; AVX1-NEXT:    popq %r14
; AVX1-NEXT:    popq %r15
; AVX1-NEXT:    popq %rbp
; AVX1-NEXT:    retq
;
; AVX2-LABEL: ext_i32_32i8:
; AVX2:       # BB#0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:  .Lcfi12:
; AVX2-NEXT:    .cfi_def_cfa_offset 16
; AVX2-NEXT:  .Lcfi13:
; AVX2-NEXT:    .cfi_offset %rbp, -16
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:  .Lcfi14:
; AVX2-NEXT:    .cfi_def_cfa_register %rbp
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %r13
; AVX2-NEXT:    pushq %r12
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $64, %rsp
; AVX2-NEXT:  .Lcfi15:
; AVX2-NEXT:    .cfi_offset %rbx, -56
; AVX2-NEXT:  .Lcfi16:
; AVX2-NEXT:    .cfi_offset %r12, -48
; AVX2-NEXT:  .Lcfi17:
; AVX2-NEXT:    .cfi_offset %r13, -40
; AVX2-NEXT:  .Lcfi18:
; AVX2-NEXT:    .cfi_offset %r14, -32
; AVX2-NEXT:  .Lcfi19:
; AVX2-NEXT:    .cfi_offset %r15, -24
; AVX2-NEXT:    movl %edi, (%rsp)
; AVX2-NEXT:    movslq (%rsp), %rdx
; AVX2-NEXT:    movq %rdx, %rcx
; AVX2-NEXT:    shlq $47, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vmovd %ecx, %xmm0
; AVX2-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    movq %rdx, %r8
; AVX2-NEXT:    movq %rdx, %rcx
; AVX2-NEXT:    movq %rdx, %rdi
; AVX2-NEXT:    movq %rdx, %r13
; AVX2-NEXT:    movq %rdx, %rsi
; AVX2-NEXT:    movq %rdx, %r10
; AVX2-NEXT:    movq %rdx, %r11
; AVX2-NEXT:    movq %rdx, %r9
; AVX2-NEXT:    movq %rdx, %rbx
; AVX2-NEXT:    movq %rdx, %r14
; AVX2-NEXT:    movq %rdx, %r15
; AVX2-NEXT:    movq %rdx, %r12
; AVX2-NEXT:    movq %rdx, %rax
; AVX2-NEXT:    shlq $46, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX2-NEXT:    shlq $45, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    shlq $44, %r8
; AVX2-NEXT:    sarq $63, %r8
; AVX2-NEXT:    vpinsrb $3, %r8d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r8
; AVX2-NEXT:    shlq $43, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vpinsrb $4, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %rcx
; AVX2-NEXT:    shlq $42, %rdi
; AVX2-NEXT:    sarq $63, %rdi
; AVX2-NEXT:    vpinsrb $5, %edi, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %rdi
; AVX2-NEXT:    shlq $41, %r13
; AVX2-NEXT:    sarq $63, %r13
; AVX2-NEXT:    vpinsrb $6, %r13d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r13
; AVX2-NEXT:    shlq $40, %rsi
; AVX2-NEXT:    sarq $63, %rsi
; AVX2-NEXT:    vpinsrb $7, %esi, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %rsi
; AVX2-NEXT:    shlq $39, %r10
; AVX2-NEXT:    sarq $63, %r10
; AVX2-NEXT:    vpinsrb $8, %r10d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r10
; AVX2-NEXT:    shlq $38, %r11
; AVX2-NEXT:    sarq $63, %r11
; AVX2-NEXT:    vpinsrb $9, %r11d, %xmm0, %xmm0
; AVX2-NEXT:    movsbq %dl, %rax
; AVX2-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    shlq $37, %r9
; AVX2-NEXT:    sarq $63, %r9
; AVX2-NEXT:    vpinsrb $10, %r9d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r9
; AVX2-NEXT:    shlq $36, %rbx
; AVX2-NEXT:    sarq $63, %rbx
; AVX2-NEXT:    vpinsrb $11, %ebx, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %rbx
; AVX2-NEXT:    shlq $35, %r14
; AVX2-NEXT:    sarq $63, %r14
; AVX2-NEXT:    vpinsrb $12, %r14d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r14
; AVX2-NEXT:    shlq $34, %r15
; AVX2-NEXT:    sarq $63, %r15
; AVX2-NEXT:    vpinsrb $13, %r15d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r15
; AVX2-NEXT:    shlq $33, %r12
; AVX2-NEXT:    sarq $63, %r12
; AVX2-NEXT:    vpinsrb $14, %r12d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r12
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX2-NEXT:    shrq $31, %rax
; AVX2-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %rax
; AVX2-NEXT:    shlq $63, %r8
; AVX2-NEXT:    sarq $63, %r8
; AVX2-NEXT:    vmovd %r8d, %xmm1
; AVX2-NEXT:    movq %rdx, %r8
; AVX2-NEXT:    movswq %dx, %rdx
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
; AVX2-NEXT:    shlq $62, %r11
; AVX2-NEXT:    sarq $63, %r11
; AVX2-NEXT:    vpinsrb $1, %r11d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $61, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vpinsrb $2, %ecx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $60, %rdi
; AVX2-NEXT:    sarq $63, %rdi
; AVX2-NEXT:    vpinsrb $3, %edi, %xmm1, %xmm1
; AVX2-NEXT:    shlq $59, %r13
; AVX2-NEXT:    sarq $63, %r13
; AVX2-NEXT:    vpinsrb $4, %r13d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $58, %rsi
; AVX2-NEXT:    sarq $63, %rsi
; AVX2-NEXT:    vpinsrb $5, %esi, %xmm1, %xmm1
; AVX2-NEXT:    shlq $57, %r10
; AVX2-NEXT:    sarq $63, %r10
; AVX2-NEXT:    vpinsrb $6, %r10d, %xmm1, %xmm1
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
; AVX2-NEXT:    shrq $7, %rcx
; AVX2-NEXT:    vpinsrb $7, %ecx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $55, %r9
; AVX2-NEXT:    sarq $63, %r9
; AVX2-NEXT:    vpinsrb $8, %r9d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $54, %rbx
; AVX2-NEXT:    sarq $63, %rbx
; AVX2-NEXT:    vpinsrb $9, %ebx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $53, %r14
; AVX2-NEXT:    sarq $63, %r14
; AVX2-NEXT:    vpinsrb $10, %r14d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $52, %r15
; AVX2-NEXT:    sarq $63, %r15
; AVX2-NEXT:    vpinsrb $11, %r15d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $51, %r12
; AVX2-NEXT:    sarq $63, %r12
; AVX2-NEXT:    vpinsrb $12, %r12d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $50, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
; AVX2-NEXT:    shlq $49, %r8
; AVX2-NEXT:    sarq $63, %r8
; AVX2-NEXT:    vpinsrb $14, %r8d, %xmm1, %xmm1
; AVX2-NEXT:    shrq $15, %rdx
; AVX2-NEXT:    vpinsrb $15, %edx, %xmm1, %xmm1
; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
; AVX2-NEXT:    leaq -40(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r12
; AVX2-NEXT:    popq %r13
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512-LABEL: ext_i32_32i8:
; AVX512:       # BB#0:
; AVX512-NEXT:    kmovd %edi, %k0
; AVX512-NEXT:    vpmovm2b %k0, %ymm0
; AVX512-NEXT:    retq
  %1 = bitcast i32 %a0 to <32 x i1>
  %2 = sext <32 x i1> %1 to <32 x i8>
  ret <32 x i8> %2
}

;
; 512-bit vectors
;

define <8 x i64> @ext_i8_8i64(i8 %a0) {
; SSE2-SSSE3-LABEL: ext_i8_8i64:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $3, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $2, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm3
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $5, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $4, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $6, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    shrl $7, %eax
; SSE2-SSSE3-NEXT:    movzwl %ax, %eax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm2
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm1[0]
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,1,0,3]
; SSE2-SSSE3-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,5,6,7]
; SSE2-SSSE3-NEXT:    psllq $63, %xmm0
; SSE2-SSSE3-NEXT:    psrad $31, %xmm0
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,1,1,3]
; SSE2-SSSE3-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,6,7]
; SSE2-SSSE3-NEXT:    psllq $63, %xmm1
; SSE2-SSSE3-NEXT:    psrad $31, %xmm1
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[2,1,2,3]
; SSE2-SSSE3-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,5,6,7]
; SSE2-SSSE3-NEXT:    psllq $63, %xmm2
; SSE2-SSSE3-NEXT:    psrad $31, %xmm2
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[3,1,3,3]
; SSE2-SSSE3-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,5,6,7]
; SSE2-SSSE3-NEXT:    psllq $63, %xmm3
; SSE2-SSSE3-NEXT:    psrad $31, %xmm3
; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
; SSE2-SSSE3-NEXT:    retq
;
; AVX1-LABEL: ext_i8_8i64:
; AVX1:       # BB#0:
; AVX1-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX1-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    movl %eax, %edx
; AVX1-NEXT:    andl $1, %edx
; AVX1-NEXT:    vmovd %edx, %xmm0
; AVX1-NEXT:    vpinsrw $1, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $2, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrw $2, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $3, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrw $3, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $4, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrw $4, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $5, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $6, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    shrl $7, %eax
; AVX1-NEXT:    movzwl %ax, %eax
; AVX1-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm1
; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
; AVX1-NEXT:    vpslld $31, %xmm0, %xmm0
; AVX1-NEXT:    vpsrad $31, %xmm0, %xmm0
; AVX1-NEXT:    vpmovsxdq %xmm0, %xmm2
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
; AVX1-NEXT:    vpmovsxdq %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; AVX1-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX1-NEXT:    vpsrad $31, %xmm1, %xmm1
; AVX1-NEXT:    vpmovsxdq %xmm1, %xmm2
; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
; AVX1-NEXT:    vpmovsxdq %xmm1, %xmm1
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
; AVX1-NEXT:    retq
;
; AVX2-LABEL: ext_i8_8i64:
; AVX2:       # BB#0:
; AVX2-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    movl %eax, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    vmovd %edx, %xmm0
; AVX2-NEXT:    vpinsrw $1, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrw $2, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $3, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrw $3, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $4, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrw $4, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $5, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $6, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    shrl $7, %eax
; AVX2-NEXT:    movzwl %ax, %eax
; AVX2-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm1
; AVX2-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
; AVX2-NEXT:    vpslld $31, %xmm0, %xmm0
; AVX2-NEXT:    vpsrad $31, %xmm0, %xmm0
; AVX2-NEXT:    vpmovsxdq %xmm0, %ymm0
; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; AVX2-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX2-NEXT:    vpsrad $31, %xmm1, %xmm1
; AVX2-NEXT:    vpmovsxdq %xmm1, %ymm1
; AVX2-NEXT:    retq
;
; AVX512-LABEL: ext_i8_8i64:
; AVX512:       # BB#0:
; AVX512-NEXT:    kmovd %edi, %k1
; AVX512-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
; AVX512-NEXT:    retq
  %1 = bitcast i8 %a0 to <8 x i1>
  %2 = sext <8 x i1> %1 to <8 x i64>
  ret <8 x i64> %2
}

define <16 x i32> @ext_i16_16i32(i16 %a0) {
; SSE2-SSSE3-LABEL: ext_i16_16i32:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $7, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $6, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $5, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $4, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $3, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $2, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm3
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $11, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $10, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $9, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $8, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $13, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $12, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $14, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    shrl $15, %eax
; SSE2-SSSE3-NEXT:    movzwl %ax, %eax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm4
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3],xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm0[0]
; SSE2-SSSE3-NEXT:    movdqa %xmm3, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movdqa %xmm1, %xmm0
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
; SSE2-SSSE3-NEXT:    pslld $31, %xmm0
; SSE2-SSSE3-NEXT:    psrad $31, %xmm0
; SSE2-SSSE3-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    pslld $31, %xmm1
; SSE2-SSSE3-NEXT:    psrad $31, %xmm1
; SSE2-SSSE3-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm0[8],xmm3[9],xmm0[9],xmm3[10],xmm0[10],xmm3[11],xmm0[11],xmm3[12],xmm0[12],xmm3[13],xmm0[13],xmm3[14],xmm0[14],xmm3[15],xmm0[15]
; SSE2-SSSE3-NEXT:    movdqa %xmm3, %xmm2
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
; SSE2-SSSE3-NEXT:    pslld $31, %xmm2
; SSE2-SSSE3-NEXT:    psrad $31, %xmm2
; SSE2-SSSE3-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
; SSE2-SSSE3-NEXT:    pslld $31, %xmm3
; SSE2-SSSE3-NEXT:    psrad $31, %xmm3
; SSE2-SSSE3-NEXT:    retq
;
; AVX1-LABEL: ext_i16_16i32:
; AVX1:       # BB#0:
; AVX1-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; AVX1-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    movl %eax, %edx
; AVX1-NEXT:    andl $1, %edx
; AVX1-NEXT:    vmovd %edx, %xmm0
; AVX1-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $2, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $3, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $3, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $4, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $4, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $5, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $5, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $6, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $6, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $7, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $7, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $8, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $8, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $9, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $9, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $10, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $10, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $11, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $11, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $12, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $12, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $13, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $13, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movl %eax, %ecx
; AVX1-NEXT:    shrl $14, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $14, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    shrl $15, %eax
; AVX1-NEXT:    movzwl %ax, %eax
; AVX1-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm1
; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
; AVX1-NEXT:    vpslld $31, %xmm2, %xmm2
; AVX1-NEXT:    vpsrad $31, %xmm2, %xmm2
; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
; AVX1-NEXT:    vpslld $31, %xmm0, %xmm0
; AVX1-NEXT:    vpsrad $31, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
; AVX1-NEXT:    vpslld $31, %xmm2, %xmm2
; AVX1-NEXT:    vpsrad $31, %xmm2, %xmm2
; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; AVX1-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX1-NEXT:    vpsrad $31, %xmm1, %xmm1
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
; AVX1-NEXT:    retq
;
; AVX2-LABEL: ext_i16_16i32:
; AVX2:       # BB#0:
; AVX2-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    movl %eax, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    vmovd %edx, %xmm0
; AVX2-NEXT:    vpinsrb $1, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $3, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $3, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $4, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $4, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $5, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $5, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $6, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $6, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $7, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $7, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $8, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $8, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $9, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $9, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $10, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $10, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $11, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $11, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $12, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $12, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $13, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $13, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    shrl $14, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $14, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    shrl $15, %eax
; AVX2-NEXT:    movzwl %ax, %eax
; AVX2-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm1
; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX2-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
; AVX2-NEXT:    vpslld $31, %ymm0, %ymm0
; AVX2-NEXT:    vpsrad $31, %ymm0, %ymm0
; AVX2-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX2-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
; AVX2-NEXT:    vpslld $31, %ymm1, %ymm1
; AVX2-NEXT:    vpsrad $31, %ymm1, %ymm1
; AVX2-NEXT:    retq
;
; AVX512-LABEL: ext_i16_16i32:
; AVX512:       # BB#0:
; AVX512-NEXT:    kmovd %edi, %k1
; AVX512-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
; AVX512-NEXT:    retq
  %1 = bitcast i16 %a0 to <16 x i1>
  %2 = sext <16 x i1> %1 to <16 x i32>
  ret <16 x i32> %2
}

define <32 x i16> @ext_i32_32i16(i32 %a0) {
; SSE2-SSSE3-LABEL: ext_i32_32i16:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    movl %edi, %eax
; SSE2-SSSE3-NEXT:    shrl $16, %eax
; SSE2-SSSE3-NEXT:    movw %ax, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $7, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $6, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $5, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $4, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $3, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $2, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm3
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $11, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $10, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $9, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $8, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $13, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $12, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $14, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    shrl $15, %eax
; SSE2-SSSE3-NEXT:    movzwl %ax, %eax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm4
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3],xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm0[0]
; SSE2-SSSE3-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $7, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $6, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $5, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $4, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $3, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $2, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm4
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3],xmm4[4],xmm0[4],xmm4[5],xmm0[5],xmm4[6],xmm0[6],xmm4[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $11, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $10, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $9, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm4
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $8, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $13, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $12, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm4
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1],xmm4[2],xmm2[2],xmm4[3],xmm2[3],xmm4[4],xmm2[4],xmm4[5],xmm2[5],xmm4[6],xmm2[6],xmm4[7],xmm2[7]
; SSE2-SSSE3-NEXT:    movl %eax, %ecx
; SSE2-SSSE3-NEXT:    shrl $14, %ecx
; SSE2-SSSE3-NEXT:    andl $1, %ecx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    shrl $15, %eax
; SSE2-SSSE3-NEXT:    movzwl %ax, %eax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm5
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1],xmm2[2],xmm5[2],xmm2[3],xmm5[3],xmm2[4],xmm5[4],xmm2[5],xmm5[5],xmm2[6],xmm5[6],xmm2[7],xmm5[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1],xmm4[2],xmm2[2],xmm4[3],xmm2[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE2-SSSE3-NEXT:    movdqa %xmm1, %xmm0
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-SSSE3-NEXT:    psllw $15, %xmm0
; SSE2-SSSE3-NEXT:    psraw $15, %xmm0
; SSE2-SSSE3-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
; SSE2-SSSE3-NEXT:    psllw $15, %xmm1
; SSE2-SSSE3-NEXT:    psraw $15, %xmm1
; SSE2-SSSE3-NEXT:    movdqa %xmm3, %xmm2
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
; SSE2-SSSE3-NEXT:    psllw $15, %xmm2
; SSE2-SSSE3-NEXT:    psraw $15, %xmm2
; SSE2-SSSE3-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm0[8],xmm3[9],xmm0[9],xmm3[10],xmm0[10],xmm3[11],xmm0[11],xmm3[12],xmm0[12],xmm3[13],xmm0[13],xmm3[14],xmm0[14],xmm3[15],xmm0[15]
; SSE2-SSSE3-NEXT:    psllw $15, %xmm3
; SSE2-SSSE3-NEXT:    psraw $15, %xmm3
; SSE2-SSSE3-NEXT:    retq
;
; AVX1-LABEL: ext_i32_32i16:
; AVX1:       # BB#0:
; AVX1-NEXT:    pushq %rbp
; AVX1-NEXT:  .Lcfi20:
; AVX1-NEXT:    .cfi_def_cfa_offset 16
; AVX1-NEXT:  .Lcfi21:
; AVX1-NEXT:    .cfi_offset %rbp, -16
; AVX1-NEXT:    movq %rsp, %rbp
; AVX1-NEXT:  .Lcfi22:
; AVX1-NEXT:    .cfi_def_cfa_register %rbp
; AVX1-NEXT:    pushq %r15
; AVX1-NEXT:    pushq %r14
; AVX1-NEXT:    pushq %r13
; AVX1-NEXT:    pushq %r12
; AVX1-NEXT:    pushq %rbx
; AVX1-NEXT:    andq $-32, %rsp
; AVX1-NEXT:    subq $128, %rsp
; AVX1-NEXT:  .Lcfi23:
; AVX1-NEXT:    .cfi_offset %rbx, -56
; AVX1-NEXT:  .Lcfi24:
; AVX1-NEXT:    .cfi_offset %r12, -48
; AVX1-NEXT:  .Lcfi25:
; AVX1-NEXT:    .cfi_offset %r13, -40
; AVX1-NEXT:  .Lcfi26:
; AVX1-NEXT:    .cfi_offset %r14, -32
; AVX1-NEXT:  .Lcfi27:
; AVX1-NEXT:    .cfi_offset %r15, -24
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX1-NEXT:    movl %edi, %r13d
; AVX1-NEXT:    movl %edi, %r12d
; AVX1-NEXT:    movl %edi, %r15d
; AVX1-NEXT:    movl %edi, %r14d
; AVX1-NEXT:    movl %edi, %ebx
; AVX1-NEXT:    movl %edi, %r11d
; AVX1-NEXT:    movl %edi, %r10d
; AVX1-NEXT:    movl %edi, %r9d
; AVX1-NEXT:    movl %edi, %r8d
; AVX1-NEXT:    movl %edi, %esi
; AVX1-NEXT:    movl %edi, %edx
; AVX1-NEXT:    movl %edi, %ecx
; AVX1-NEXT:    movl %edi, %eax
; AVX1-NEXT:    andl $1, %edi
; AVX1-NEXT:    vmovd %edi, %xmm0
; AVX1-NEXT:    shrl %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
; AVX1-NEXT:    shrl $2, %ecx
; AVX1-NEXT:    andl $1, %ecx
; AVX1-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    shrl $3, %edx
; AVX1-NEXT:    andl $1, %edx
; AVX1-NEXT:    vpinsrb $3, %edx, %xmm0, %xmm0
; AVX1-NEXT:    shrl $4, %esi
; AVX1-NEXT:    andl $1, %esi
; AVX1-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
; AVX1-NEXT:    shrl $5, %r8d
; AVX1-NEXT:    andl $1, %r8d
; AVX1-NEXT:    vpinsrb $5, %r8d, %xmm0, %xmm0
; AVX1-NEXT:    shrl $6, %r9d
; AVX1-NEXT:    andl $1, %r9d
; AVX1-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
; AVX1-NEXT:    shrl $7, %r10d
; AVX1-NEXT:    andl $1, %r10d
; AVX1-NEXT:    vpinsrb $7, %r10d, %xmm0, %xmm0
; AVX1-NEXT:    shrl $8, %r11d
; AVX1-NEXT:    andl $1, %r11d
; AVX1-NEXT:    vpinsrb $8, %r11d, %xmm0, %xmm0
; AVX1-NEXT:    shrl $9, %ebx
; AVX1-NEXT:    andl $1, %ebx
; AVX1-NEXT:    vpinsrb $9, %ebx, %xmm0, %xmm0
; AVX1-NEXT:    shrl $10, %r14d
; AVX1-NEXT:    andl $1, %r14d
; AVX1-NEXT:    vpinsrb $10, %r14d, %xmm0, %xmm0
; AVX1-NEXT:    shrl $11, %r15d
; AVX1-NEXT:    andl $1, %r15d
; AVX1-NEXT:    vpinsrb $11, %r15d, %xmm0, %xmm0
; AVX1-NEXT:    shrl $12, %r12d
; AVX1-NEXT:    andl $1, %r12d
; AVX1-NEXT:    vpinsrb $12, %r12d, %xmm0, %xmm0
; AVX1-NEXT:    shrl $13, %r13d
; AVX1-NEXT:    andl $1, %r13d
; AVX1-NEXT:    vpinsrb $13, %r13d, %xmm0, %xmm0
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $14, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $15, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $16, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vmovd %eax, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $17, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $1, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $18, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $19, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $3, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $20, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $21, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $22, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $23, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $7, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $24, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $25, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $9, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $26, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $27, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $11, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $28, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $29, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $30, %eax
; AVX1-NEXT:    andl $1, %eax
; AVX1-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX1-NEXT:    shrl $31, %eax
; AVX1-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm1
; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVX1-NEXT:    vpsllw $15, %xmm2, %xmm2
; AVX1-NEXT:    vpsraw $15, %xmm2, %xmm2
; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-NEXT:    vpsllw $15, %xmm0, %xmm0
; AVX1-NEXT:    vpsraw $15, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX1-NEXT:    vpsllw $15, %xmm2, %xmm2
; AVX1-NEXT:    vpsraw $15, %xmm2, %xmm2
; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-NEXT:    vpsllw $15, %xmm1, %xmm1
; AVX1-NEXT:    vpsraw $15, %xmm1, %xmm1
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
; AVX1-NEXT:    leaq -40(%rbp), %rsp
; AVX1-NEXT:    popq %rbx
; AVX1-NEXT:    popq %r12
; AVX1-NEXT:    popq %r13
; AVX1-NEXT:    popq %r14
; AVX1-NEXT:    popq %r15
; AVX1-NEXT:    popq %rbp
; AVX1-NEXT:    retq
;
; AVX2-LABEL: ext_i32_32i16:
; AVX2:       # BB#0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:  .Lcfi20:
; AVX2-NEXT:    .cfi_def_cfa_offset 16
; AVX2-NEXT:  .Lcfi21:
; AVX2-NEXT:    .cfi_offset %rbp, -16
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:  .Lcfi22:
; AVX2-NEXT:    .cfi_def_cfa_register %rbp
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %r13
; AVX2-NEXT:    pushq %r12
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $128, %rsp
; AVX2-NEXT:  .Lcfi23:
; AVX2-NEXT:    .cfi_offset %rbx, -56
; AVX2-NEXT:  .Lcfi24:
; AVX2-NEXT:    .cfi_offset %r12, -48
; AVX2-NEXT:  .Lcfi25:
; AVX2-NEXT:    .cfi_offset %r13, -40
; AVX2-NEXT:  .Lcfi26:
; AVX2-NEXT:    .cfi_offset %r14, -32
; AVX2-NEXT:  .Lcfi27:
; AVX2-NEXT:    .cfi_offset %r15, -24
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp) # 4-byte Spill
; AVX2-NEXT:    movl %edi, %r13d
; AVX2-NEXT:    movl %edi, %r12d
; AVX2-NEXT:    movl %edi, %r15d
; AVX2-NEXT:    movl %edi, %r14d
; AVX2-NEXT:    movl %edi, %ebx
; AVX2-NEXT:    movl %edi, %r11d
; AVX2-NEXT:    movl %edi, %r10d
; AVX2-NEXT:    movl %edi, %r9d
; AVX2-NEXT:    movl %edi, %r8d
; AVX2-NEXT:    movl %edi, %esi
; AVX2-NEXT:    movl %edi, %edx
; AVX2-NEXT:    movl %edi, %ecx
; AVX2-NEXT:    movl %edi, %eax
; AVX2-NEXT:    andl $1, %edi
; AVX2-NEXT:    vmovd %edi, %xmm0
; AVX2-NEXT:    shrl %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
; AVX2-NEXT:    shrl $2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    shrl $3, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    vpinsrb $3, %edx, %xmm0, %xmm0
; AVX2-NEXT:    shrl $4, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    vpinsrb $4, %esi, %xmm0, %xmm0
; AVX2-NEXT:    shrl $5, %r8d
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    vpinsrb $5, %r8d, %xmm0, %xmm0
; AVX2-NEXT:    shrl $6, %r9d
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
; AVX2-NEXT:    shrl $7, %r10d
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    vpinsrb $7, %r10d, %xmm0, %xmm0
; AVX2-NEXT:    shrl $8, %r11d
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    vpinsrb $8, %r11d, %xmm0, %xmm0
; AVX2-NEXT:    shrl $9, %ebx
; AVX2-NEXT:    andl $1, %ebx
; AVX2-NEXT:    vpinsrb $9, %ebx, %xmm0, %xmm0
; AVX2-NEXT:    shrl $10, %r14d
; AVX2-NEXT:    andl $1, %r14d
; AVX2-NEXT:    vpinsrb $10, %r14d, %xmm0, %xmm0
; AVX2-NEXT:    shrl $11, %r15d
; AVX2-NEXT:    andl $1, %r15d
; AVX2-NEXT:    vpinsrb $11, %r15d, %xmm0, %xmm0
; AVX2-NEXT:    shrl $12, %r12d
; AVX2-NEXT:    andl $1, %r12d
; AVX2-NEXT:    vpinsrb $12, %r12d, %xmm0, %xmm0
; AVX2-NEXT:    shrl $13, %r13d
; AVX2-NEXT:    andl $1, %r13d
; AVX2-NEXT:    vpinsrb $13, %r13d, %xmm0, %xmm0
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $14, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $15, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $16, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vmovd %eax, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $17, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $1, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $18, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $19, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $3, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $20, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $21, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $22, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $23, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $7, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $24, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $25, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $9, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $26, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $27, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $11, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $28, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $29, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $30, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
; AVX2-NEXT:    shrl $31, %eax
; AVX2-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm1
; AVX2-NEXT:    vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX2-NEXT:    vpsllw $15, %ymm0, %ymm0
; AVX2-NEXT:    vpsraw $15, %ymm0, %ymm0
; AVX2-NEXT:    vpmovzxbw {{.*#+}} 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,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX2-NEXT:    vpsllw $15, %ymm1, %ymm1
; AVX2-NEXT:    vpsraw $15, %ymm1, %ymm1
; AVX2-NEXT:    leaq -40(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r12
; AVX2-NEXT:    popq %r13
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512-LABEL: ext_i32_32i16:
; AVX512:       # BB#0:
; AVX512-NEXT:    kmovd %edi, %k0
; AVX512-NEXT:    vpmovm2w %k0, %zmm0
; AVX512-NEXT:    retq
  %1 = bitcast i32 %a0 to <32 x i1>
  %2 = sext <32 x i1> %1 to <32 x i16>
  ret <32 x i16> %2
}

define <64 x i8> @ext_i64_64i8(i64 %a0) {
; SSE2-SSSE3-LABEL: ext_i64_64i8:
; SSE2-SSSE3:       # BB#0:
; SSE2-SSSE3-NEXT:    pushq %rbp
; SSE2-SSSE3-NEXT:  .Lcfi24:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 16
; SSE2-SSSE3-NEXT:    pushq %r15
; SSE2-SSSE3-NEXT:  .Lcfi25:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 24
; SSE2-SSSE3-NEXT:    pushq %r14
; SSE2-SSSE3-NEXT:  .Lcfi26:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 32
; SSE2-SSSE3-NEXT:    pushq %r13
; SSE2-SSSE3-NEXT:  .Lcfi27:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 40
; SSE2-SSSE3-NEXT:    pushq %r12
; SSE2-SSSE3-NEXT:  .Lcfi28:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 48
; SSE2-SSSE3-NEXT:    pushq %rbx
; SSE2-SSSE3-NEXT:  .Lcfi29:
; SSE2-SSSE3-NEXT:    .cfi_def_cfa_offset 56
; SSE2-SSSE3-NEXT:  .Lcfi30:
; SSE2-SSSE3-NEXT:    .cfi_offset %rbx, -56
; SSE2-SSSE3-NEXT:  .Lcfi31:
; SSE2-SSSE3-NEXT:    .cfi_offset %r12, -48
; SSE2-SSSE3-NEXT:  .Lcfi32:
; SSE2-SSSE3-NEXT:    .cfi_offset %r13, -40
; SSE2-SSSE3-NEXT:  .Lcfi33:
; SSE2-SSSE3-NEXT:    .cfi_offset %r14, -32
; SSE2-SSSE3-NEXT:  .Lcfi34:
; SSE2-SSSE3-NEXT:    .cfi_offset %r15, -24
; SSE2-SSSE3-NEXT:  .Lcfi35:
; SSE2-SSSE3-NEXT:    .cfi_offset %rbp, -16
; SSE2-SSSE3-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movq %rdi, %rax
; SSE2-SSSE3-NEXT:    shrq $32, %rax
; SSE2-SSSE3-NEXT:    movw %ax, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movq %rdi, %rax
; SSE2-SSSE3-NEXT:    shrq $48, %rax
; SSE2-SSSE3-NEXT:    movw %ax, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    shrl $16, %edi
; SSE2-SSSE3-NEXT:    movw %di, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT:    movswq -{{[0-9]+}}(%rsp), %rbx
; SSE2-SSSE3-NEXT:    movq %rbx, %r8
; SSE2-SSSE3-NEXT:    movq %rbx, %r9
; SSE2-SSSE3-NEXT:    movq %rbx, %r10
; SSE2-SSSE3-NEXT:    movq %rbx, %r11
; SSE2-SSSE3-NEXT:    movq %rbx, %r14
; SSE2-SSSE3-NEXT:    movq %rbx, %r15
; SSE2-SSSE3-NEXT:    movq %rbx, %r12
; SSE2-SSSE3-NEXT:    movq %rbx, %r13
; SSE2-SSSE3-NEXT:    movq %rbx, %rdi
; SSE2-SSSE3-NEXT:    movq %rbx, %rcx
; SSE2-SSSE3-NEXT:    movq %rbx, %rdx
; SSE2-SSSE3-NEXT:    movq %rbx, %rsi
; SSE2-SSSE3-NEXT:    movq %rbx, %rbp
; SSE2-SSSE3-NEXT:    movq %rbx, %rax
; SSE2-SSSE3-NEXT:    shrq $15, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm0
; SSE2-SSSE3-NEXT:    movq %rbx, %rax
; SSE2-SSSE3-NEXT:    movsbq %bl, %rbx
; SSE2-SSSE3-NEXT:    shlq $49, %r8
; SSE2-SSSE3-NEXT:    sarq $63, %r8
; SSE2-SSSE3-NEXT:    movd %r8d, %xmm15
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0],xmm0[0],xmm15[1],xmm0[1],xmm15[2],xmm0[2],xmm15[3],xmm0[3],xmm15[4],xmm0[4],xmm15[5],xmm0[5],xmm15[6],xmm0[6],xmm15[7],xmm0[7]
; SSE2-SSSE3-NEXT:    shlq $50, %r9
; SSE2-SSSE3-NEXT:    sarq $63, %r9
; SSE2-SSSE3-NEXT:    movd %r9d, %xmm8
; SSE2-SSSE3-NEXT:    shlq $51, %r10
; SSE2-SSSE3-NEXT:    sarq $63, %r10
; SSE2-SSSE3-NEXT:    movd %r10d, %xmm2
; SSE2-SSSE3-NEXT:    shlq $52, %r11
; SSE2-SSSE3-NEXT:    sarq $63, %r11
; SSE2-SSSE3-NEXT:    movd %r11d, %xmm9
; SSE2-SSSE3-NEXT:    shlq $53, %r14
; SSE2-SSSE3-NEXT:    sarq $63, %r14
; SSE2-SSSE3-NEXT:    movd %r14d, %xmm6
; SSE2-SSSE3-NEXT:    shlq $54, %r15
; SSE2-SSSE3-NEXT:    sarq $63, %r15
; SSE2-SSSE3-NEXT:    movd %r15d, %xmm10
; SSE2-SSSE3-NEXT:    shlq $55, %r12
; SSE2-SSSE3-NEXT:    sarq $63, %r12
; SSE2-SSSE3-NEXT:    movd %r12d, %xmm4
; SSE2-SSSE3-NEXT:    shlq $60, %r13
; SSE2-SSSE3-NEXT:    sarq $63, %r13
; SSE2-SSSE3-NEXT:    movd %r13d, %xmm11
; SSE2-SSSE3-NEXT:    shlq $61, %rdi
; SSE2-SSSE3-NEXT:    sarq $63, %rdi
; SSE2-SSSE3-NEXT:    movd %edi, %xmm5
; SSE2-SSSE3-NEXT:    shlq $62, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm12
; SSE2-SSSE3-NEXT:    shlq $63, %rdx
; SSE2-SSSE3-NEXT:    sarq $63, %rdx
; SSE2-SSSE3-NEXT:    movd %edx, %xmm0
; SSE2-SSSE3-NEXT:    shlq $58, %rsi
; SSE2-SSSE3-NEXT:    sarq $63, %rsi
; SSE2-SSSE3-NEXT:    movd %esi, %xmm13
; SSE2-SSSE3-NEXT:    shlq $59, %rbp
; SSE2-SSSE3-NEXT:    sarq $63, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm7
; SSE2-SSSE3-NEXT:    shlq $57, %rax
; SSE2-SSSE3-NEXT:    sarq $63, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm3
; SSE2-SSSE3-NEXT:    shrq $7, %rbx
; SSE2-SSSE3-NEXT:    movd %ebx, %xmm14
; SSE2-SSSE3-NEXT:    movswq -{{[0-9]+}}(%rsp), %rsi
; SSE2-SSSE3-NEXT:    movq %rsi, %r8
; SSE2-SSSE3-NEXT:    movq %rsi, %r9
; SSE2-SSSE3-NEXT:    movq %rsi, %r10
; SSE2-SSSE3-NEXT:    movq %rsi, %r11
; SSE2-SSSE3-NEXT:    movq %rsi, %r14
; SSE2-SSSE3-NEXT:    movq %rsi, %r15
; SSE2-SSSE3-NEXT:    movq %rsi, %r12
; SSE2-SSSE3-NEXT:    movq %rsi, %r13
; SSE2-SSSE3-NEXT:    movq %rsi, %rbx
; SSE2-SSSE3-NEXT:    movq %rsi, %rax
; SSE2-SSSE3-NEXT:    movq %rsi, %rcx
; SSE2-SSSE3-NEXT:    movq %rsi, %rdx
; SSE2-SSSE3-NEXT:    movq %rsi, %rdi
; SSE2-SSSE3-NEXT:    movq %rsi, %rbp
; SSE2-SSSE3-NEXT:    shrq $15, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm1
; SSE2-SSSE3-NEXT:    movq %rsi, %rbp
; SSE2-SSSE3-NEXT:    movsbq %sil, %rsi
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm8[0],xmm2[1],xmm8[1],xmm2[2],xmm8[2],xmm2[3],xmm8[3],xmm2[4],xmm8[4],xmm2[5],xmm8[5],xmm2[6],xmm8[6],xmm2[7],xmm8[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm15[0],xmm2[1],xmm15[1],xmm2[2],xmm15[2],xmm2[3],xmm15[3]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm9[0],xmm6[1],xmm9[1],xmm6[2],xmm9[2],xmm6[3],xmm9[3],xmm6[4],xmm9[4],xmm6[5],xmm9[5],xmm6[6],xmm9[6],xmm6[7],xmm9[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm10[0],xmm4[1],xmm10[1],xmm4[2],xmm10[2],xmm4[3],xmm10[3],xmm4[4],xmm10[4],xmm4[5],xmm10[5],xmm4[6],xmm10[6],xmm4[7],xmm10[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1],xmm4[2],xmm6[2],xmm4[3],xmm6[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm11[0],xmm5[1],xmm11[1],xmm5[2],xmm11[2],xmm5[3],xmm11[3],xmm5[4],xmm11[4],xmm5[5],xmm11[5],xmm5[6],xmm11[6],xmm5[7],xmm11[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm12[0],xmm0[1],xmm12[1],xmm0[2],xmm12[2],xmm0[3],xmm12[3],xmm0[4],xmm12[4],xmm0[5],xmm12[5],xmm0[6],xmm12[6],xmm0[7],xmm12[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm13[0],xmm7[1],xmm13[1],xmm7[2],xmm13[2],xmm7[3],xmm13[3],xmm7[4],xmm13[4],xmm7[5],xmm13[5],xmm7[6],xmm13[6],xmm7[7],xmm13[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm14[0],xmm3[1],xmm14[1],xmm3[2],xmm14[2],xmm3[3],xmm14[3],xmm3[4],xmm14[4],xmm3[5],xmm14[5],xmm3[6],xmm14[6],xmm3[7],xmm14[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm7 = xmm7[0],xmm3[0],xmm7[1],xmm3[1],xmm7[2],xmm3[2],xmm7[3],xmm3[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[1],xmm7[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm4[0]
; SSE2-SSSE3-NEXT:    shlq $49, %r8
; SSE2-SSSE3-NEXT:    sarq $63, %r8
; SSE2-SSSE3-NEXT:    movd %r8d, %xmm13
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm13 = xmm13[0],xmm1[0],xmm13[1],xmm1[1],xmm13[2],xmm1[2],xmm13[3],xmm1[3],xmm13[4],xmm1[4],xmm13[5],xmm1[5],xmm13[6],xmm1[6],xmm13[7],xmm1[7]
; SSE2-SSSE3-NEXT:    shlq $50, %r9
; SSE2-SSSE3-NEXT:    sarq $63, %r9
; SSE2-SSSE3-NEXT:    movd %r9d, %xmm1
; SSE2-SSSE3-NEXT:    shlq $51, %r10
; SSE2-SSSE3-NEXT:    sarq $63, %r10
; SSE2-SSSE3-NEXT:    movd %r10d, %xmm3
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3],xmm3[4],xmm1[4],xmm3[5],xmm1[5],xmm3[6],xmm1[6],xmm3[7],xmm1[7]
; SSE2-SSSE3-NEXT:    shlq $52, %r11
; SSE2-SSSE3-NEXT:    sarq $63, %r11
; SSE2-SSSE3-NEXT:    movd %r11d, %xmm8
; SSE2-SSSE3-NEXT:    shlq $53, %r14
; SSE2-SSSE3-NEXT:    sarq $63, %r14
; SSE2-SSSE3-NEXT:    movd %r14d, %xmm15
; SSE2-SSSE3-NEXT:    shlq $54, %r15
; SSE2-SSSE3-NEXT:    sarq $63, %r15
; SSE2-SSSE3-NEXT:    movd %r15d, %xmm9
; SSE2-SSSE3-NEXT:    shlq $55, %r12
; SSE2-SSSE3-NEXT:    sarq $63, %r12
; SSE2-SSSE3-NEXT:    movd %r12d, %xmm4
; SSE2-SSSE3-NEXT:    shlq $60, %r13
; SSE2-SSSE3-NEXT:    sarq $63, %r13
; SSE2-SSSE3-NEXT:    movd %r13d, %xmm10
; SSE2-SSSE3-NEXT:    shlq $61, %rbx
; SSE2-SSSE3-NEXT:    sarq $63, %rbx
; SSE2-SSSE3-NEXT:    movd %ebx, %xmm7
; SSE2-SSSE3-NEXT:    shlq $62, %rax
; SSE2-SSSE3-NEXT:    sarq $63, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm11
; SSE2-SSSE3-NEXT:    shlq $63, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm2
; SSE2-SSSE3-NEXT:    shlq $58, %rdx
; SSE2-SSSE3-NEXT:    sarq $63, %rdx
; SSE2-SSSE3-NEXT:    movd %edx, %xmm12
; SSE2-SSSE3-NEXT:    shlq $59, %rdi
; SSE2-SSSE3-NEXT:    sarq $63, %rdi
; SSE2-SSSE3-NEXT:    movd %edi, %xmm5
; SSE2-SSSE3-NEXT:    shlq $57, %rbp
; SSE2-SSSE3-NEXT:    sarq $63, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm1
; SSE2-SSSE3-NEXT:    shrq $7, %rsi
; SSE2-SSSE3-NEXT:    movd %esi, %xmm14
; SSE2-SSSE3-NEXT:    movswq -{{[0-9]+}}(%rsp), %rsi
; SSE2-SSSE3-NEXT:    movq %rsi, %r8
; SSE2-SSSE3-NEXT:    movq %rsi, %r9
; SSE2-SSSE3-NEXT:    movq %rsi, %r10
; SSE2-SSSE3-NEXT:    movq %rsi, %r11
; SSE2-SSSE3-NEXT:    movq %rsi, %r14
; SSE2-SSSE3-NEXT:    movq %rsi, %r15
; SSE2-SSSE3-NEXT:    movq %rsi, %r12
; SSE2-SSSE3-NEXT:    movq %rsi, %r13
; SSE2-SSSE3-NEXT:    movq %rsi, %rbx
; SSE2-SSSE3-NEXT:    movq %rsi, %rax
; SSE2-SSSE3-NEXT:    movq %rsi, %rcx
; SSE2-SSSE3-NEXT:    movq %rsi, %rdx
; SSE2-SSSE3-NEXT:    movq %rsi, %rdi
; SSE2-SSSE3-NEXT:    movq %rsi, %rbp
; SSE2-SSSE3-NEXT:    shrq $15, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm6
; SSE2-SSSE3-NEXT:    movq %rsi, %rbp
; SSE2-SSSE3-NEXT:    movsbq %sil, %rsi
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm13[0],xmm3[1],xmm13[1],xmm3[2],xmm13[2],xmm3[3],xmm13[3]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0],xmm8[0],xmm15[1],xmm8[1],xmm15[2],xmm8[2],xmm15[3],xmm8[3],xmm15[4],xmm8[4],xmm15[5],xmm8[5],xmm15[6],xmm8[6],xmm15[7],xmm8[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm9[0],xmm4[1],xmm9[1],xmm4[2],xmm9[2],xmm4[3],xmm9[3],xmm4[4],xmm9[4],xmm4[5],xmm9[5],xmm4[6],xmm9[6],xmm4[7],xmm9[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm15[0],xmm4[1],xmm15[1],xmm4[2],xmm15[2],xmm4[3],xmm15[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm10[0],xmm7[1],xmm10[1],xmm7[2],xmm10[2],xmm7[3],xmm10[3],xmm7[4],xmm10[4],xmm7[5],xmm10[5],xmm7[6],xmm10[6],xmm7[7],xmm10[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm11[0],xmm2[1],xmm11[1],xmm2[2],xmm11[2],xmm2[3],xmm11[3],xmm2[4],xmm11[4],xmm2[5],xmm11[5],xmm2[6],xmm11[6],xmm2[7],xmm11[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm7[0],xmm2[1],xmm7[1],xmm2[2],xmm7[2],xmm2[3],xmm7[3]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm12[0],xmm5[1],xmm12[1],xmm5[2],xmm12[2],xmm5[3],xmm12[3],xmm5[4],xmm12[4],xmm5[5],xmm12[5],xmm5[6],xmm12[6],xmm5[7],xmm12[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm14[0],xmm1[1],xmm14[1],xmm1[2],xmm14[2],xmm1[3],xmm14[3],xmm1[4],xmm14[4],xmm1[5],xmm14[5],xmm1[6],xmm14[6],xmm1[7],xmm14[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm1[0],xmm5[1],xmm1[1],xmm5[2],xmm1[2],xmm5[3],xmm1[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm4[0]
; SSE2-SSSE3-NEXT:    shlq $49, %r8
; SSE2-SSSE3-NEXT:    sarq $63, %r8
; SSE2-SSSE3-NEXT:    movd %r8d, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm6[0],xmm1[1],xmm6[1],xmm1[2],xmm6[2],xmm1[3],xmm6[3],xmm1[4],xmm6[4],xmm1[5],xmm6[5],xmm1[6],xmm6[6],xmm1[7],xmm6[7]
; SSE2-SSSE3-NEXT:    shlq $50, %r9
; SSE2-SSSE3-NEXT:    sarq $63, %r9
; SSE2-SSSE3-NEXT:    movd %r9d, %xmm3
; SSE2-SSSE3-NEXT:    shlq $51, %r10
; SSE2-SSSE3-NEXT:    sarq $63, %r10
; SSE2-SSSE3-NEXT:    movd %r10d, %xmm4
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3],xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1],xmm4[2],xmm1[2],xmm4[3],xmm1[3]
; SSE2-SSSE3-NEXT:    shlq $52, %r11
; SSE2-SSSE3-NEXT:    sarq $63, %r11
; SSE2-SSSE3-NEXT:    movd %r11d, %xmm8
; SSE2-SSSE3-NEXT:    shlq $53, %r14
; SSE2-SSSE3-NEXT:    sarq $63, %r14
; SSE2-SSSE3-NEXT:    movd %r14d, %xmm13
; SSE2-SSSE3-NEXT:    shlq $54, %r15
; SSE2-SSSE3-NEXT:    sarq $63, %r15
; SSE2-SSSE3-NEXT:    movd %r15d, %xmm9
; SSE2-SSSE3-NEXT:    shlq $55, %r12
; SSE2-SSSE3-NEXT:    sarq $63, %r12
; SSE2-SSSE3-NEXT:    movd %r12d, %xmm1
; SSE2-SSSE3-NEXT:    shlq $60, %r13
; SSE2-SSSE3-NEXT:    sarq $63, %r13
; SSE2-SSSE3-NEXT:    movd %r13d, %xmm10
; SSE2-SSSE3-NEXT:    shlq $61, %rbx
; SSE2-SSSE3-NEXT:    sarq $63, %rbx
; SSE2-SSSE3-NEXT:    movd %ebx, %xmm15
; SSE2-SSSE3-NEXT:    shlq $62, %rax
; SSE2-SSSE3-NEXT:    sarq $63, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm11
; SSE2-SSSE3-NEXT:    shlq $63, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm3
; SSE2-SSSE3-NEXT:    shlq $58, %rdx
; SSE2-SSSE3-NEXT:    sarq $63, %rdx
; SSE2-SSSE3-NEXT:    movd %edx, %xmm12
; SSE2-SSSE3-NEXT:    shlq $59, %rdi
; SSE2-SSSE3-NEXT:    sarq $63, %rdi
; SSE2-SSSE3-NEXT:    movd %edi, %xmm5
; SSE2-SSSE3-NEXT:    shlq $57, %rbp
; SSE2-SSSE3-NEXT:    sarq $63, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm6
; SSE2-SSSE3-NEXT:    shrq $7, %rsi
; SSE2-SSSE3-NEXT:    movd %esi, %xmm14
; SSE2-SSSE3-NEXT:    movswq -{{[0-9]+}}(%rsp), %rsi
; SSE2-SSSE3-NEXT:    movq %rsi, %r8
; SSE2-SSSE3-NEXT:    movq %rsi, %r9
; SSE2-SSSE3-NEXT:    movq %rsi, %r10
; SSE2-SSSE3-NEXT:    movq %rsi, %r11
; SSE2-SSSE3-NEXT:    movq %rsi, %r14
; SSE2-SSSE3-NEXT:    movq %rsi, %r15
; SSE2-SSSE3-NEXT:    movq %rsi, %r12
; SSE2-SSSE3-NEXT:    movq %rsi, %r13
; SSE2-SSSE3-NEXT:    movq %rsi, %rbx
; SSE2-SSSE3-NEXT:    movq %rsi, %rax
; SSE2-SSSE3-NEXT:    movq %rsi, %rcx
; SSE2-SSSE3-NEXT:    movq %rsi, %rdx
; SSE2-SSSE3-NEXT:    movq %rsi, %rdi
; SSE2-SSSE3-NEXT:    movq %rsi, %rbp
; SSE2-SSSE3-NEXT:    shrq $15, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm7
; SSE2-SSSE3-NEXT:    movq %rsi, %rbp
; SSE2-SSSE3-NEXT:    movsbq %sil, %rsi
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm13 = xmm13[0],xmm8[0],xmm13[1],xmm8[1],xmm13[2],xmm8[2],xmm13[3],xmm8[3],xmm13[4],xmm8[4],xmm13[5],xmm8[5],xmm13[6],xmm8[6],xmm13[7],xmm8[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm9[0],xmm1[1],xmm9[1],xmm1[2],xmm9[2],xmm1[3],xmm9[3],xmm1[4],xmm9[4],xmm1[5],xmm9[5],xmm1[6],xmm9[6],xmm1[7],xmm9[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm13[0],xmm1[1],xmm13[1],xmm1[2],xmm13[2],xmm1[3],xmm13[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0],xmm10[0],xmm15[1],xmm10[1],xmm15[2],xmm10[2],xmm15[3],xmm10[3],xmm15[4],xmm10[4],xmm15[5],xmm10[5],xmm15[6],xmm10[6],xmm15[7],xmm10[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm11[0],xmm3[1],xmm11[1],xmm3[2],xmm11[2],xmm3[3],xmm11[3],xmm3[4],xmm11[4],xmm3[5],xmm11[5],xmm3[6],xmm11[6],xmm3[7],xmm11[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm15[0],xmm3[1],xmm15[1],xmm3[2],xmm15[2],xmm3[3],xmm15[3]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm12[0],xmm5[1],xmm12[1],xmm5[2],xmm12[2],xmm5[3],xmm12[3],xmm5[4],xmm12[4],xmm5[5],xmm12[5],xmm5[6],xmm12[6],xmm5[7],xmm12[7]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm14[0],xmm6[1],xmm14[1],xmm6[2],xmm14[2],xmm6[3],xmm14[3],xmm6[4],xmm14[4],xmm6[5],xmm14[5],xmm6[6],xmm14[6],xmm6[7],xmm14[7]
; SSE2-SSSE3-NEXT:    shlq $49, %r8
; SSE2-SSSE3-NEXT:    sarq $63, %r8
; SSE2-SSSE3-NEXT:    movd %r8d, %xmm4
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3]
; SSE2-SSSE3-NEXT:    shlq $50, %r9
; SSE2-SSSE3-NEXT:    sarq $63, %r9
; SSE2-SSSE3-NEXT:    movd %r9d, %xmm6
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
; SSE2-SSSE3-NEXT:    shlq $51, %r10
; SSE2-SSSE3-NEXT:    sarq $63, %r10
; SSE2-SSSE3-NEXT:    movd %r10d, %xmm5
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm1[0]
; SSE2-SSSE3-NEXT:    shlq $52, %r11
; SSE2-SSSE3-NEXT:    sarq $63, %r11
; SSE2-SSSE3-NEXT:    movd %r11d, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm7[0],xmm4[1],xmm7[1],xmm4[2],xmm7[2],xmm4[3],xmm7[3],xmm4[4],xmm7[4],xmm4[5],xmm7[5],xmm4[6],xmm7[6],xmm4[7],xmm7[7]
; SSE2-SSSE3-NEXT:    shlq $53, %r14
; SSE2-SSSE3-NEXT:    sarq $63, %r14
; SSE2-SSSE3-NEXT:    movd %r14d, %xmm7
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3],xmm5[4],xmm6[4],xmm5[5],xmm6[5],xmm5[6],xmm6[6],xmm5[7],xmm6[7]
; SSE2-SSSE3-NEXT:    shlq $54, %r15
; SSE2-SSSE3-NEXT:    sarq $63, %r15
; SSE2-SSSE3-NEXT:    movd %r15d, %xmm6
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
; SSE2-SSSE3-NEXT:    shlq $55, %r12
; SSE2-SSSE3-NEXT:    sarq $63, %r12
; SSE2-SSSE3-NEXT:    movd %r12d, %xmm4
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm1[0],xmm7[1],xmm1[1],xmm7[2],xmm1[2],xmm7[3],xmm1[3],xmm7[4],xmm1[4],xmm7[5],xmm1[5],xmm7[6],xmm1[6],xmm7[7],xmm1[7]
; SSE2-SSSE3-NEXT:    shlq $60, %r13
; SSE2-SSSE3-NEXT:    sarq $63, %r13
; SSE2-SSSE3-NEXT:    movd %r13d, %xmm8
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1],xmm4[2],xmm6[2],xmm4[3],xmm6[3],xmm4[4],xmm6[4],xmm4[5],xmm6[5],xmm4[6],xmm6[6],xmm4[7],xmm6[7]
; SSE2-SSSE3-NEXT:    shlq $61, %rbx
; SSE2-SSSE3-NEXT:    sarq $63, %rbx
; SSE2-SSSE3-NEXT:    movd %ebx, %xmm6
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm7[0],xmm4[1],xmm7[1],xmm4[2],xmm7[2],xmm4[3],xmm7[3]
; SSE2-SSSE3-NEXT:    shlq $62, %rax
; SSE2-SSSE3-NEXT:    sarq $63, %rax
; SSE2-SSSE3-NEXT:    movd %eax, %xmm7
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
; SSE2-SSSE3-NEXT:    shlq $63, %rcx
; SSE2-SSSE3-NEXT:    sarq $63, %rcx
; SSE2-SSSE3-NEXT:    movd %ecx, %xmm1
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm8[0],xmm6[1],xmm8[1],xmm6[2],xmm8[2],xmm6[3],xmm8[3],xmm6[4],xmm8[4],xmm6[5],xmm8[5],xmm6[6],xmm8[6],xmm6[7],xmm8[7]
; SSE2-SSSE3-NEXT:    shlq $58, %rdx
; SSE2-SSSE3-NEXT:    sarq $63, %rdx
; SSE2-SSSE3-NEXT:    movd %edx, %xmm5
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm7[0],xmm1[1],xmm7[1],xmm1[2],xmm7[2],xmm1[3],xmm7[3],xmm1[4],xmm7[4],xmm1[5],xmm7[5],xmm1[6],xmm7[6],xmm1[7],xmm7[7]
; SSE2-SSSE3-NEXT:    shlq $59, %rdi
; SSE2-SSSE3-NEXT:    sarq $63, %rdi
; SSE2-SSSE3-NEXT:    movd %edi, %xmm7
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm6[0],xmm1[1],xmm6[1],xmm1[2],xmm6[2],xmm1[3],xmm6[3]
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm5[0],xmm7[1],xmm5[1],xmm7[2],xmm5[2],xmm7[3],xmm5[3],xmm7[4],xmm5[4],xmm7[5],xmm5[5],xmm7[6],xmm5[6],xmm7[7],xmm5[7]
; SSE2-SSSE3-NEXT:    shlq $57, %rbp
; SSE2-SSSE3-NEXT:    sarq $63, %rbp
; SSE2-SSSE3-NEXT:    movd %ebp, %xmm5
; SSE2-SSSE3-NEXT:    shrq $7, %rsi
; SSE2-SSSE3-NEXT:    movd %esi, %xmm6
; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3],xmm5[4],xmm6[4],xmm5[5],xmm6[5],xmm5[6],xmm6[6],xmm5[7],xmm6[7]
; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm7 = xmm7[0],xmm5[0],xmm7[1],xmm5[1],xmm7[2],xmm5[2],xmm7[3],xmm5[3]
; SSE2-SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm7[0],xmm1[1],xmm7[1]
; SSE2-SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm4[0]
; SSE2-SSSE3-NEXT:    popq %rbx
; SSE2-SSSE3-NEXT:    popq %r12
; SSE2-SSSE3-NEXT:    popq %r13
; SSE2-SSSE3-NEXT:    popq %r14
; SSE2-SSSE3-NEXT:    popq %r15
; SSE2-SSSE3-NEXT:    popq %rbp
; SSE2-SSSE3-NEXT:    retq
;
; AVX1-LABEL: ext_i64_64i8:
; AVX1:       # BB#0:
; AVX1-NEXT:    pushq %rbp
; AVX1-NEXT:  .Lcfi28:
; AVX1-NEXT:    .cfi_def_cfa_offset 16
; AVX1-NEXT:  .Lcfi29:
; AVX1-NEXT:    .cfi_offset %rbp, -16
; AVX1-NEXT:    movq %rsp, %rbp
; AVX1-NEXT:  .Lcfi30:
; AVX1-NEXT:    .cfi_def_cfa_register %rbp
; AVX1-NEXT:    pushq %r15
; AVX1-NEXT:    pushq %r14
; AVX1-NEXT:    pushq %r13
; AVX1-NEXT:    pushq %r12
; AVX1-NEXT:    pushq %rbx
; AVX1-NEXT:    andq $-32, %rsp
; AVX1-NEXT:    subq $128, %rsp
; AVX1-NEXT:  .Lcfi31:
; AVX1-NEXT:    .cfi_offset %rbx, -56
; AVX1-NEXT:  .Lcfi32:
; AVX1-NEXT:    .cfi_offset %r12, -48
; AVX1-NEXT:  .Lcfi33:
; AVX1-NEXT:    .cfi_offset %r13, -40
; AVX1-NEXT:  .Lcfi34:
; AVX1-NEXT:    .cfi_offset %r14, -32
; AVX1-NEXT:  .Lcfi35:
; AVX1-NEXT:    .cfi_offset %r15, -24
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
; AVX1-NEXT:    shrq $32, %rdi
; AVX1-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
; AVX1-NEXT:    movslq {{[0-9]+}}(%rsp), %rdx
; AVX1-NEXT:    movq %rdx, %rcx
; AVX1-NEXT:    shlq $47, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vmovd %ecx, %xmm0
; AVX1-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    movq %rdx, %r8
; AVX1-NEXT:    movq %rdx, %rcx
; AVX1-NEXT:    movq %rdx, %rdi
; AVX1-NEXT:    movq %rdx, %r13
; AVX1-NEXT:    movq %rdx, %rsi
; AVX1-NEXT:    movq %rdx, %r10
; AVX1-NEXT:    movq %rdx, %r11
; AVX1-NEXT:    movq %rdx, %r9
; AVX1-NEXT:    movq %rdx, %rbx
; AVX1-NEXT:    movq %rdx, %r14
; AVX1-NEXT:    movq %rdx, %r15
; AVX1-NEXT:    movq %rdx, %r12
; AVX1-NEXT:    movq %rdx, %rax
; AVX1-NEXT:    shlq $46, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX1-NEXT:    shlq $45, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    shlq $44, %r8
; AVX1-NEXT:    sarq $63, %r8
; AVX1-NEXT:    vpinsrb $3, %r8d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r8
; AVX1-NEXT:    shlq $43, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vpinsrb $4, %ecx, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %rcx
; AVX1-NEXT:    shlq $42, %rdi
; AVX1-NEXT:    sarq $63, %rdi
; AVX1-NEXT:    vpinsrb $5, %edi, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %rdi
; AVX1-NEXT:    shlq $41, %r13
; AVX1-NEXT:    sarq $63, %r13
; AVX1-NEXT:    vpinsrb $6, %r13d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r13
; AVX1-NEXT:    shlq $40, %rsi
; AVX1-NEXT:    sarq $63, %rsi
; AVX1-NEXT:    vpinsrb $7, %esi, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %rsi
; AVX1-NEXT:    shlq $39, %r10
; AVX1-NEXT:    sarq $63, %r10
; AVX1-NEXT:    vpinsrb $8, %r10d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r10
; AVX1-NEXT:    shlq $38, %r11
; AVX1-NEXT:    sarq $63, %r11
; AVX1-NEXT:    vpinsrb $9, %r11d, %xmm0, %xmm0
; AVX1-NEXT:    movsbq %dl, %rax
; AVX1-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    shlq $37, %r9
; AVX1-NEXT:    sarq $63, %r9
; AVX1-NEXT:    vpinsrb $10, %r9d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r9
; AVX1-NEXT:    shlq $36, %rbx
; AVX1-NEXT:    sarq $63, %rbx
; AVX1-NEXT:    vpinsrb $11, %ebx, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %rbx
; AVX1-NEXT:    shlq $35, %r14
; AVX1-NEXT:    sarq $63, %r14
; AVX1-NEXT:    vpinsrb $12, %r14d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r14
; AVX1-NEXT:    shlq $34, %r15
; AVX1-NEXT:    sarq $63, %r15
; AVX1-NEXT:    vpinsrb $13, %r15d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r15
; AVX1-NEXT:    shlq $33, %r12
; AVX1-NEXT:    sarq $63, %r12
; AVX1-NEXT:    vpinsrb $14, %r12d, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %r12
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX1-NEXT:    shrq $31, %rax
; AVX1-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
; AVX1-NEXT:    movq %rdx, %rax
; AVX1-NEXT:    shlq $63, %r8
; AVX1-NEXT:    sarq $63, %r8
; AVX1-NEXT:    vmovd %r8d, %xmm1
; AVX1-NEXT:    movq %rdx, %r8
; AVX1-NEXT:    movswq %dx, %rdx
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
; AVX1-NEXT:    shlq $62, %r11
; AVX1-NEXT:    sarq $63, %r11
; AVX1-NEXT:    vpinsrb $1, %r11d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $61, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vpinsrb $2, %ecx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $60, %rdi
; AVX1-NEXT:    sarq $63, %rdi
; AVX1-NEXT:    vpinsrb $3, %edi, %xmm1, %xmm1
; AVX1-NEXT:    shlq $59, %r13
; AVX1-NEXT:    sarq $63, %r13
; AVX1-NEXT:    vpinsrb $4, %r13d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $58, %rsi
; AVX1-NEXT:    sarq $63, %rsi
; AVX1-NEXT:    vpinsrb $5, %esi, %xmm1, %xmm1
; AVX1-NEXT:    shlq $57, %r10
; AVX1-NEXT:    sarq $63, %r10
; AVX1-NEXT:    vpinsrb $6, %r10d, %xmm1, %xmm1
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
; AVX1-NEXT:    shrq $7, %rcx
; AVX1-NEXT:    vpinsrb $7, %ecx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $55, %r9
; AVX1-NEXT:    sarq $63, %r9
; AVX1-NEXT:    vpinsrb $8, %r9d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $54, %rbx
; AVX1-NEXT:    sarq $63, %rbx
; AVX1-NEXT:    vpinsrb $9, %ebx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $53, %r14
; AVX1-NEXT:    sarq $63, %r14
; AVX1-NEXT:    vpinsrb $10, %r14d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $52, %r15
; AVX1-NEXT:    sarq $63, %r15
; AVX1-NEXT:    vpinsrb $11, %r15d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $51, %r12
; AVX1-NEXT:    sarq $63, %r12
; AVX1-NEXT:    vpinsrb $12, %r12d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $50, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
; AVX1-NEXT:    shlq $49, %r8
; AVX1-NEXT:    sarq $63, %r8
; AVX1-NEXT:    vpinsrb $14, %r8d, %xmm1, %xmm1
; AVX1-NEXT:    shrq $15, %rdx
; AVX1-NEXT:    vpinsrb $15, %edx, %xmm1, %xmm1
; AVX1-NEXT:    movslq {{[0-9]+}}(%rsp), %rdx
; AVX1-NEXT:    movq %rdx, %rcx
; AVX1-NEXT:    shlq $47, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vmovd %ecx, %xmm2
; AVX1-NEXT:    movq %rdx, %r13
; AVX1-NEXT:    movq %rdx, %rcx
; AVX1-NEXT:    movq %rdx, %r9
; AVX1-NEXT:    movq %rdx, %r12
; AVX1-NEXT:    movq %rdx, %rdi
; AVX1-NEXT:    movq %rdx, %rbx
; AVX1-NEXT:    movq %rdx, %r8
; AVX1-NEXT:    movq %rdx, %r10
; AVX1-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    movq %rdx, %rsi
; AVX1-NEXT:    movq %rdx, %r11
; AVX1-NEXT:    movq %rdx, %r14
; AVX1-NEXT:    movq %rdx, %r15
; AVX1-NEXT:    movq %rdx, %rax
; AVX1-NEXT:    shlq $46, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrb $1, %eax, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    shlq $45, %r13
; AVX1-NEXT:    sarq $63, %r13
; AVX1-NEXT:    vpinsrb $2, %r13d, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %r13
; AVX1-NEXT:    shlq $44, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vpinsrb $3, %ecx, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %rcx
; AVX1-NEXT:    shlq $43, %r9
; AVX1-NEXT:    sarq $63, %r9
; AVX1-NEXT:    vpinsrb $4, %r9d, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %r9
; AVX1-NEXT:    shlq $42, %r12
; AVX1-NEXT:    sarq $63, %r12
; AVX1-NEXT:    vpinsrb $5, %r12d, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %r12
; AVX1-NEXT:    shlq $41, %rdi
; AVX1-NEXT:    sarq $63, %rdi
; AVX1-NEXT:    vpinsrb $6, %edi, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %rdi
; AVX1-NEXT:    shlq $40, %rbx
; AVX1-NEXT:    sarq $63, %rbx
; AVX1-NEXT:    vpinsrb $7, %ebx, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %rbx
; AVX1-NEXT:    shlq $39, %r8
; AVX1-NEXT:    sarq $63, %r8
; AVX1-NEXT:    vpinsrb $8, %r8d, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %r8
; AVX1-NEXT:    shlq $38, %r10
; AVX1-NEXT:    sarq $63, %r10
; AVX1-NEXT:    vpinsrb $9, %r10d, %xmm2, %xmm2
; AVX1-NEXT:    movsbq %dl, %rax
; AVX1-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX1-NEXT:    shlq $37, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrb $10, %eax, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %r10
; AVX1-NEXT:    shlq $36, %rsi
; AVX1-NEXT:    sarq $63, %rsi
; AVX1-NEXT:    vpinsrb $11, %esi, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %rsi
; AVX1-NEXT:    shlq $35, %r11
; AVX1-NEXT:    sarq $63, %r11
; AVX1-NEXT:    vpinsrb $12, %r11d, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %r11
; AVX1-NEXT:    shlq $34, %r14
; AVX1-NEXT:    sarq $63, %r14
; AVX1-NEXT:    vpinsrb $13, %r14d, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %r14
; AVX1-NEXT:    shlq $33, %r15
; AVX1-NEXT:    sarq $63, %r15
; AVX1-NEXT:    vpinsrb $14, %r15d, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %r15
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX1-NEXT:    shrq $31, %rax
; AVX1-NEXT:    vpinsrb $15, %eax, %xmm2, %xmm2
; AVX1-NEXT:    movq %rdx, %rax
; AVX1-NEXT:    shlq $63, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vmovd %ecx, %xmm3
; AVX1-NEXT:    movq %rdx, %rcx
; AVX1-NEXT:    movswq %dx, %rdx
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT:    shlq $62, %r13
; AVX1-NEXT:    sarq $63, %r13
; AVX1-NEXT:    vpinsrb $1, %r13d, %xmm3, %xmm1
; AVX1-NEXT:    shlq $61, %r9
; AVX1-NEXT:    sarq $63, %r9
; AVX1-NEXT:    vpinsrb $2, %r9d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $60, %r12
; AVX1-NEXT:    sarq $63, %r12
; AVX1-NEXT:    vpinsrb $3, %r12d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $59, %rdi
; AVX1-NEXT:    sarq $63, %rdi
; AVX1-NEXT:    vpinsrb $4, %edi, %xmm1, %xmm1
; AVX1-NEXT:    shlq $58, %rbx
; AVX1-NEXT:    sarq $63, %rbx
; AVX1-NEXT:    vpinsrb $5, %ebx, %xmm1, %xmm1
; AVX1-NEXT:    shlq $57, %r8
; AVX1-NEXT:    sarq $63, %r8
; AVX1-NEXT:    vpinsrb $6, %r8d, %xmm1, %xmm1
; AVX1-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
; AVX1-NEXT:    shrq $7, %rdi
; AVX1-NEXT:    vpinsrb $7, %edi, %xmm1, %xmm1
; AVX1-NEXT:    shlq $55, %r10
; AVX1-NEXT:    sarq $63, %r10
; AVX1-NEXT:    vpinsrb $8, %r10d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $54, %rsi
; AVX1-NEXT:    sarq $63, %rsi
; AVX1-NEXT:    vpinsrb $9, %esi, %xmm1, %xmm1
; AVX1-NEXT:    shlq $53, %r11
; AVX1-NEXT:    sarq $63, %r11
; AVX1-NEXT:    vpinsrb $10, %r11d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $52, %r14
; AVX1-NEXT:    sarq $63, %r14
; AVX1-NEXT:    vpinsrb $11, %r14d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $51, %r15
; AVX1-NEXT:    sarq $63, %r15
; AVX1-NEXT:    vpinsrb $12, %r15d, %xmm1, %xmm1
; AVX1-NEXT:    shlq $50, %rax
; AVX1-NEXT:    sarq $63, %rax
; AVX1-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
; AVX1-NEXT:    shlq $49, %rcx
; AVX1-NEXT:    sarq $63, %rcx
; AVX1-NEXT:    vpinsrb $14, %ecx, %xmm1, %xmm1
; AVX1-NEXT:    shrq $15, %rdx
; AVX1-NEXT:    vpinsrb $15, %edx, %xmm1, %xmm1
; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
; AVX1-NEXT:    leaq -40(%rbp), %rsp
; AVX1-NEXT:    popq %rbx
; AVX1-NEXT:    popq %r12
; AVX1-NEXT:    popq %r13
; AVX1-NEXT:    popq %r14
; AVX1-NEXT:    popq %r15
; AVX1-NEXT:    popq %rbp
; AVX1-NEXT:    retq
;
; AVX2-LABEL: ext_i64_64i8:
; AVX2:       # BB#0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:  .Lcfi28:
; AVX2-NEXT:    .cfi_def_cfa_offset 16
; AVX2-NEXT:  .Lcfi29:
; AVX2-NEXT:    .cfi_offset %rbp, -16
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:  .Lcfi30:
; AVX2-NEXT:    .cfi_def_cfa_register %rbp
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %r13
; AVX2-NEXT:    pushq %r12
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $128, %rsp
; AVX2-NEXT:  .Lcfi31:
; AVX2-NEXT:    .cfi_offset %rbx, -56
; AVX2-NEXT:  .Lcfi32:
; AVX2-NEXT:    .cfi_offset %r12, -48
; AVX2-NEXT:  .Lcfi33:
; AVX2-NEXT:    .cfi_offset %r13, -40
; AVX2-NEXT:  .Lcfi34:
; AVX2-NEXT:    .cfi_offset %r14, -32
; AVX2-NEXT:  .Lcfi35:
; AVX2-NEXT:    .cfi_offset %r15, -24
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    shrq $32, %rdi
; AVX2-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    movslq {{[0-9]+}}(%rsp), %rdx
; AVX2-NEXT:    movq %rdx, %rcx
; AVX2-NEXT:    shlq $47, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vmovd %ecx, %xmm0
; AVX2-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    movq %rdx, %r8
; AVX2-NEXT:    movq %rdx, %rcx
; AVX2-NEXT:    movq %rdx, %rdi
; AVX2-NEXT:    movq %rdx, %r13
; AVX2-NEXT:    movq %rdx, %rsi
; AVX2-NEXT:    movq %rdx, %r10
; AVX2-NEXT:    movq %rdx, %r11
; AVX2-NEXT:    movq %rdx, %r9
; AVX2-NEXT:    movq %rdx, %rbx
; AVX2-NEXT:    movq %rdx, %r14
; AVX2-NEXT:    movq %rdx, %r15
; AVX2-NEXT:    movq %rdx, %r12
; AVX2-NEXT:    movq %rdx, %rax
; AVX2-NEXT:    shlq $46, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX2-NEXT:    shlq $45, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    shlq $44, %r8
; AVX2-NEXT:    sarq $63, %r8
; AVX2-NEXT:    vpinsrb $3, %r8d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r8
; AVX2-NEXT:    shlq $43, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vpinsrb $4, %ecx, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %rcx
; AVX2-NEXT:    shlq $42, %rdi
; AVX2-NEXT:    sarq $63, %rdi
; AVX2-NEXT:    vpinsrb $5, %edi, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %rdi
; AVX2-NEXT:    shlq $41, %r13
; AVX2-NEXT:    sarq $63, %r13
; AVX2-NEXT:    vpinsrb $6, %r13d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r13
; AVX2-NEXT:    shlq $40, %rsi
; AVX2-NEXT:    sarq $63, %rsi
; AVX2-NEXT:    vpinsrb $7, %esi, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %rsi
; AVX2-NEXT:    shlq $39, %r10
; AVX2-NEXT:    sarq $63, %r10
; AVX2-NEXT:    vpinsrb $8, %r10d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r10
; AVX2-NEXT:    shlq $38, %r11
; AVX2-NEXT:    sarq $63, %r11
; AVX2-NEXT:    vpinsrb $9, %r11d, %xmm0, %xmm0
; AVX2-NEXT:    movsbq %dl, %rax
; AVX2-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    shlq $37, %r9
; AVX2-NEXT:    sarq $63, %r9
; AVX2-NEXT:    vpinsrb $10, %r9d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r9
; AVX2-NEXT:    shlq $36, %rbx
; AVX2-NEXT:    sarq $63, %rbx
; AVX2-NEXT:    vpinsrb $11, %ebx, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %rbx
; AVX2-NEXT:    shlq $35, %r14
; AVX2-NEXT:    sarq $63, %r14
; AVX2-NEXT:    vpinsrb $12, %r14d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r14
; AVX2-NEXT:    shlq $34, %r15
; AVX2-NEXT:    sarq $63, %r15
; AVX2-NEXT:    vpinsrb $13, %r15d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r15
; AVX2-NEXT:    shlq $33, %r12
; AVX2-NEXT:    sarq $63, %r12
; AVX2-NEXT:    vpinsrb $14, %r12d, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %r12
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX2-NEXT:    shrq $31, %rax
; AVX2-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
; AVX2-NEXT:    movq %rdx, %rax
; AVX2-NEXT:    shlq $63, %r8
; AVX2-NEXT:    sarq $63, %r8
; AVX2-NEXT:    vmovd %r8d, %xmm1
; AVX2-NEXT:    movq %rdx, %r8
; AVX2-NEXT:    movswq %dx, %rdx
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
; AVX2-NEXT:    shlq $62, %r11
; AVX2-NEXT:    sarq $63, %r11
; AVX2-NEXT:    vpinsrb $1, %r11d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $61, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vpinsrb $2, %ecx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $60, %rdi
; AVX2-NEXT:    sarq $63, %rdi
; AVX2-NEXT:    vpinsrb $3, %edi, %xmm1, %xmm1
; AVX2-NEXT:    shlq $59, %r13
; AVX2-NEXT:    sarq $63, %r13
; AVX2-NEXT:    vpinsrb $4, %r13d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $58, %rsi
; AVX2-NEXT:    sarq $63, %rsi
; AVX2-NEXT:    vpinsrb $5, %esi, %xmm1, %xmm1
; AVX2-NEXT:    shlq $57, %r10
; AVX2-NEXT:    sarq $63, %r10
; AVX2-NEXT:    vpinsrb $6, %r10d, %xmm1, %xmm1
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
; AVX2-NEXT:    shrq $7, %rcx
; AVX2-NEXT:    vpinsrb $7, %ecx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $55, %r9
; AVX2-NEXT:    sarq $63, %r9
; AVX2-NEXT:    vpinsrb $8, %r9d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $54, %rbx
; AVX2-NEXT:    sarq $63, %rbx
; AVX2-NEXT:    vpinsrb $9, %ebx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $53, %r14
; AVX2-NEXT:    sarq $63, %r14
; AVX2-NEXT:    vpinsrb $10, %r14d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $52, %r15
; AVX2-NEXT:    sarq $63, %r15
; AVX2-NEXT:    vpinsrb $11, %r15d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $51, %r12
; AVX2-NEXT:    sarq $63, %r12
; AVX2-NEXT:    vpinsrb $12, %r12d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $50, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
; AVX2-NEXT:    shlq $49, %r8
; AVX2-NEXT:    sarq $63, %r8
; AVX2-NEXT:    vpinsrb $14, %r8d, %xmm1, %xmm1
; AVX2-NEXT:    shrq $15, %rdx
; AVX2-NEXT:    vpinsrb $15, %edx, %xmm1, %xmm1
; AVX2-NEXT:    movslq {{[0-9]+}}(%rsp), %rdx
; AVX2-NEXT:    movq %rdx, %rcx
; AVX2-NEXT:    shlq $47, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vmovd %ecx, %xmm2
; AVX2-NEXT:    movq %rdx, %r13
; AVX2-NEXT:    movq %rdx, %rcx
; AVX2-NEXT:    movq %rdx, %r9
; AVX2-NEXT:    movq %rdx, %r12
; AVX2-NEXT:    movq %rdx, %rdi
; AVX2-NEXT:    movq %rdx, %rbx
; AVX2-NEXT:    movq %rdx, %r8
; AVX2-NEXT:    movq %rdx, %r10
; AVX2-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    movq %rdx, %rsi
; AVX2-NEXT:    movq %rdx, %r11
; AVX2-NEXT:    movq %rdx, %r14
; AVX2-NEXT:    movq %rdx, %r15
; AVX2-NEXT:    movq %rdx, %rax
; AVX2-NEXT:    shlq $46, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrb $1, %eax, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    shlq $45, %r13
; AVX2-NEXT:    sarq $63, %r13
; AVX2-NEXT:    vpinsrb $2, %r13d, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %r13
; AVX2-NEXT:    shlq $44, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vpinsrb $3, %ecx, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %rcx
; AVX2-NEXT:    shlq $43, %r9
; AVX2-NEXT:    sarq $63, %r9
; AVX2-NEXT:    vpinsrb $4, %r9d, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %r9
; AVX2-NEXT:    shlq $42, %r12
; AVX2-NEXT:    sarq $63, %r12
; AVX2-NEXT:    vpinsrb $5, %r12d, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %r12
; AVX2-NEXT:    shlq $41, %rdi
; AVX2-NEXT:    sarq $63, %rdi
; AVX2-NEXT:    vpinsrb $6, %edi, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %rdi
; AVX2-NEXT:    shlq $40, %rbx
; AVX2-NEXT:    sarq $63, %rbx
; AVX2-NEXT:    vpinsrb $7, %ebx, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %rbx
; AVX2-NEXT:    shlq $39, %r8
; AVX2-NEXT:    sarq $63, %r8
; AVX2-NEXT:    vpinsrb $8, %r8d, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %r8
; AVX2-NEXT:    shlq $38, %r10
; AVX2-NEXT:    sarq $63, %r10
; AVX2-NEXT:    vpinsrb $9, %r10d, %xmm2, %xmm2
; AVX2-NEXT:    movsbq %dl, %rax
; AVX2-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX2-NEXT:    shlq $37, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrb $10, %eax, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %r10
; AVX2-NEXT:    shlq $36, %rsi
; AVX2-NEXT:    sarq $63, %rsi
; AVX2-NEXT:    vpinsrb $11, %esi, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %rsi
; AVX2-NEXT:    shlq $35, %r11
; AVX2-NEXT:    sarq $63, %r11
; AVX2-NEXT:    vpinsrb $12, %r11d, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %r11
; AVX2-NEXT:    shlq $34, %r14
; AVX2-NEXT:    sarq $63, %r14
; AVX2-NEXT:    vpinsrb $13, %r14d, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %r14
; AVX2-NEXT:    shlq $33, %r15
; AVX2-NEXT:    sarq $63, %r15
; AVX2-NEXT:    vpinsrb $14, %r15d, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %r15
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
; AVX2-NEXT:    shrq $31, %rax
; AVX2-NEXT:    vpinsrb $15, %eax, %xmm2, %xmm2
; AVX2-NEXT:    movq %rdx, %rax
; AVX2-NEXT:    shlq $63, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vmovd %ecx, %xmm3
; AVX2-NEXT:    movq %rdx, %rcx
; AVX2-NEXT:    movswq %dx, %rdx
; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
; AVX2-NEXT:    shlq $62, %r13
; AVX2-NEXT:    sarq $63, %r13
; AVX2-NEXT:    vpinsrb $1, %r13d, %xmm3, %xmm1
; AVX2-NEXT:    shlq $61, %r9
; AVX2-NEXT:    sarq $63, %r9
; AVX2-NEXT:    vpinsrb $2, %r9d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $60, %r12
; AVX2-NEXT:    sarq $63, %r12
; AVX2-NEXT:    vpinsrb $3, %r12d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $59, %rdi
; AVX2-NEXT:    sarq $63, %rdi
; AVX2-NEXT:    vpinsrb $4, %edi, %xmm1, %xmm1
; AVX2-NEXT:    shlq $58, %rbx
; AVX2-NEXT:    sarq $63, %rbx
; AVX2-NEXT:    vpinsrb $5, %ebx, %xmm1, %xmm1
; AVX2-NEXT:    shlq $57, %r8
; AVX2-NEXT:    sarq $63, %r8
; AVX2-NEXT:    vpinsrb $6, %r8d, %xmm1, %xmm1
; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
; AVX2-NEXT:    shrq $7, %rdi
; AVX2-NEXT:    vpinsrb $7, %edi, %xmm1, %xmm1
; AVX2-NEXT:    shlq $55, %r10
; AVX2-NEXT:    sarq $63, %r10
; AVX2-NEXT:    vpinsrb $8, %r10d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $54, %rsi
; AVX2-NEXT:    sarq $63, %rsi
; AVX2-NEXT:    vpinsrb $9, %esi, %xmm1, %xmm1
; AVX2-NEXT:    shlq $53, %r11
; AVX2-NEXT:    sarq $63, %r11
; AVX2-NEXT:    vpinsrb $10, %r11d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $52, %r14
; AVX2-NEXT:    sarq $63, %r14
; AVX2-NEXT:    vpinsrb $11, %r14d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $51, %r15
; AVX2-NEXT:    sarq $63, %r15
; AVX2-NEXT:    vpinsrb $12, %r15d, %xmm1, %xmm1
; AVX2-NEXT:    shlq $50, %rax
; AVX2-NEXT:    sarq $63, %rax
; AVX2-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
; AVX2-NEXT:    shlq $49, %rcx
; AVX2-NEXT:    sarq $63, %rcx
; AVX2-NEXT:    vpinsrb $14, %ecx, %xmm1, %xmm1
; AVX2-NEXT:    shrq $15, %rdx
; AVX2-NEXT:    vpinsrb $15, %edx, %xmm1, %xmm1
; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
; AVX2-NEXT:    leaq -40(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r12
; AVX2-NEXT:    popq %r13
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512-LABEL: ext_i64_64i8:
; AVX512:       # BB#0:
; AVX512-NEXT:    kmovq %rdi, %k0
; AVX512-NEXT:    vpmovm2b %k0, %zmm0
; AVX512-NEXT:    retq
  %1 = bitcast i64 %a0 to <64 x i1>
  %2 = sext <64 x i1> %1 to <64 x i8>
  ret <64 x i8> %2
}