llvm/test/CodeGen/X86/subvectorwise-store-of-vector-splat.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=-sse2 | FileCheck %s --check-prefixes=ALL,SCALAR
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2 | FileCheck %s --check-prefixes=ALL,SSE,SSE2,SSE2-ONLY
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+sse3 | FileCheck %s --check-prefixes=ALL,SSE,SSE2,SSE3
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+ssse3 | FileCheck %s --check-prefixes=ALL,SSE,SSE2,SSSE3,SSSE3-ONLY
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+sse4.1 | FileCheck %s --check-prefixes=ALL,SSE,SSE2,SSSE3,SSE41
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+sse4.2 | FileCheck %s --check-prefixes=ALL,SSE,SSE2,SSSE3,SSE42
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx | FileCheck %s --check-prefixes=ALL,SSE,AVX,AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 | FileCheck %s --check-prefixes=ALL,SSE,AVX,AVX2,AVX2-ONLY
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512vl | FileCheck %s --check-prefixes=ALL,SSE,AVX,AVX2,AVX512,AVX512F
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=ALL,SSE,AVX,AVX2,AVX512,AVX512BW

define void @vec32_v2i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec32_v2i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %ecx
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %cl, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %cl, 3(%rdx)
; SCALAR-NEXT:    movb %al, 2(%rdx)
; SCALAR-NEXT:    retq
;
; SSE-LABEL: vec32_v2i8:
; SSE:       # %bb.0:
; SSE-NEXT:    movl (%rdi), %eax
; SSE-NEXT:    notl %eax
; SSE-NEXT:    movw %ax, (%rsi)
; SSE-NEXT:    movw %ax, (%rdx)
; SSE-NEXT:    movw %ax, 2(%rdx)
; SSE-NEXT:    retq
  %in.subvec.not = load <2 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i8> %in.subvec.not, <i8 -1, i8 -1>
  store <2 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 0
  store <2 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 1
  store <2 x i8> %in.subvec, ptr %out.subvec1.ptr, align 2
  ret void
}

define void @vec64_v2i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec64_v2i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %ecx
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %cl, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %cl, 3(%rdx)
; SCALAR-NEXT:    movb %al, 2(%rdx)
; SCALAR-NEXT:    movb %cl, 5(%rdx)
; SCALAR-NEXT:    movb %al, 4(%rdx)
; SCALAR-NEXT:    movb %cl, 7(%rdx)
; SCALAR-NEXT:    movb %al, 6(%rdx)
; SCALAR-NEXT:    retq
;
; SSE-LABEL: vec64_v2i8:
; SSE:       # %bb.0:
; SSE-NEXT:    movl (%rdi), %eax
; SSE-NEXT:    notl %eax
; SSE-NEXT:    movw %ax, (%rsi)
; SSE-NEXT:    movw %ax, (%rdx)
; SSE-NEXT:    movw %ax, 2(%rdx)
; SSE-NEXT:    movw %ax, 4(%rdx)
; SSE-NEXT:    movw %ax, 6(%rdx)
; SSE-NEXT:    retq
  %in.subvec.not = load <2 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i8> %in.subvec.not, <i8 -1, i8 -1>
  store <2 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 0
  store <2 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 1
  store <2 x i8> %in.subvec, ptr %out.subvec1.ptr, align 2
  %out.subvec2.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 2
  store <2 x i8> %in.subvec, ptr %out.subvec2.ptr, align 4
  %out.subvec3.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 3
  store <2 x i8> %in.subvec, ptr %out.subvec3.ptr, align 2
  ret void
}

define void @vec64_v2i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec64_v2i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzwl 2(%rdi), %eax
; SCALAR-NEXT:    movl (%rdi), %ecx
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    movw %ax, 2(%rsi)
; SCALAR-NEXT:    movw %cx, (%rsi)
; SCALAR-NEXT:    movw %ax, 2(%rdx)
; SCALAR-NEXT:    movw %cx, (%rdx)
; SCALAR-NEXT:    movw %ax, 6(%rdx)
; SCALAR-NEXT:    movw %cx, 4(%rdx)
; SCALAR-NEXT:    retq
;
; SSE-LABEL: vec64_v2i16:
; SSE:       # %bb.0:
; SSE-NEXT:    movl (%rdi), %eax
; SSE-NEXT:    notl %eax
; SSE-NEXT:    movl %eax, (%rsi)
; SSE-NEXT:    movl %eax, (%rdx)
; SSE-NEXT:    movl %eax, 4(%rdx)
; SSE-NEXT:    retq
  %in.subvec.not = load <2 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i16> %in.subvec.not, <i16 -1, i16 -1>
  store <2 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 0
  store <2 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 1
  store <2 x i16> %in.subvec, ptr %out.subvec1.ptr, align 4
  ret void
}

define void @vec64_v4i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec64_v4i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl 3(%rdi), %eax
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %r8d
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, 3(%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %dil, 1(%rsi)
; SCALAR-NEXT:    movb %r8b, (%rsi)
; SCALAR-NEXT:    movb %al, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movb %dil, 1(%rdx)
; SCALAR-NEXT:    movb %r8b, (%rdx)
; SCALAR-NEXT:    movb %al, 7(%rdx)
; SCALAR-NEXT:    movb %cl, 6(%rdx)
; SCALAR-NEXT:    movb %dil, 5(%rdx)
; SCALAR-NEXT:    movb %r8b, 4(%rdx)
; SCALAR-NEXT:    retq
;
; SSE-LABEL: vec64_v4i8:
; SSE:       # %bb.0:
; SSE-NEXT:    movl (%rdi), %eax
; SSE-NEXT:    notl %eax
; SSE-NEXT:    movl %eax, (%rsi)
; SSE-NEXT:    movl %eax, (%rdx)
; SSE-NEXT:    movl %eax, 4(%rdx)
; SSE-NEXT:    retq
  %in.subvec.not = load <4 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1>
  store <4 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 0
  store <4 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 1
  store <4 x i8> %in.subvec, ptr %out.subvec1.ptr, align 4
  ret void
}

define void @vec128_v2i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec128_v2i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %ecx
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %cl, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %cl, 3(%rdx)
; SCALAR-NEXT:    movb %al, 2(%rdx)
; SCALAR-NEXT:    movb %cl, 5(%rdx)
; SCALAR-NEXT:    movb %al, 4(%rdx)
; SCALAR-NEXT:    movb %cl, 7(%rdx)
; SCALAR-NEXT:    movb %al, 6(%rdx)
; SCALAR-NEXT:    movb %cl, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    movb %cl, 11(%rdx)
; SCALAR-NEXT:    movb %al, 10(%rdx)
; SCALAR-NEXT:    movb %cl, 13(%rdx)
; SCALAR-NEXT:    movb %al, 12(%rdx)
; SCALAR-NEXT:    movb %cl, 15(%rdx)
; SCALAR-NEXT:    movb %al, 14(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec128_v2i8:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSE2-ONLY-NEXT:    movd %xmm0, %eax
; SSE2-ONLY-NEXT:    movw %ax, (%rsi)
; SSE2-ONLY-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE2-ONLY-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-ONLY-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec128_v2i8:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE3-NEXT:    pxor (%rdi), %xmm0
; SSE3-NEXT:    movd %xmm0, %eax
; SSE3-NEXT:    movw %ax, (%rsi)
; SSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE3-NEXT:    movdqa %xmm0, (%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec128_v2i8:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSSE3-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSSE3-ONLY-NEXT:    movd %xmm0, %eax
; SSSE3-ONLY-NEXT:    movw %ax, (%rsi)
; SSSE3-ONLY-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSSE3-ONLY-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSSE3-ONLY-NEXT:    movdqa %xmm0, (%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec128_v2i8:
; SSE41:       # %bb.0:
; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE41-NEXT:    pxor (%rdi), %xmm0
; SSE41-NEXT:    pextrw $0, %xmm0, (%rsi)
; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE41-NEXT:    movdqa %xmm0, (%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec128_v2i8:
; SSE42:       # %bb.0:
; SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE42-NEXT:    pxor (%rdi), %xmm0
; SSE42-NEXT:    pextrw $0, %xmm0, (%rsi)
; SSE42-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE42-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE42-NEXT:    movdqa %xmm0, (%rdx)
; SSE42-NEXT:    retq
;
; AVX1-LABEL: vec128_v2i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec128_v2i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVX2-NEXT:    vpbroadcastw %xmm0, %xmm0
; AVX2-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-NEXT:    retq
  %in.subvec.not = load <2 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i8> %in.subvec.not, <i8 -1, i8 -1>
  store <2 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 0
  store <2 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 1
  store <2 x i8> %in.subvec, ptr %out.subvec1.ptr, align 2
  %out.subvec2.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 2
  store <2 x i8> %in.subvec, ptr %out.subvec2.ptr, align 4
  %out.subvec3.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 3
  store <2 x i8> %in.subvec, ptr %out.subvec3.ptr, align 2
  %out.subvec4.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 4
  store <2 x i8> %in.subvec, ptr %out.subvec4.ptr, align 8
  %out.subvec5.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 5
  store <2 x i8> %in.subvec, ptr %out.subvec5.ptr, align 2
  %out.subvec6.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 6
  store <2 x i8> %in.subvec, ptr %out.subvec6.ptr, align 4
  %out.subvec7.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 7
  store <2 x i8> %in.subvec, ptr %out.subvec7.ptr, align 2
  ret void
}

define void @vec128_v2i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec128_v2i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzwl 2(%rdi), %eax
; SCALAR-NEXT:    movl (%rdi), %ecx
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    movw %ax, 2(%rsi)
; SCALAR-NEXT:    movw %cx, (%rsi)
; SCALAR-NEXT:    movw %ax, 2(%rdx)
; SCALAR-NEXT:    movw %cx, (%rdx)
; SCALAR-NEXT:    movw %ax, 6(%rdx)
; SCALAR-NEXT:    movw %cx, 4(%rdx)
; SCALAR-NEXT:    movw %ax, 10(%rdx)
; SCALAR-NEXT:    movw %cx, 8(%rdx)
; SCALAR-NEXT:    movw %ax, 14(%rdx)
; SCALAR-NEXT:    movw %cx, 12(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec128_v2i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movd %xmm0, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec128_v2i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vmovd %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec128_v2i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-NEXT:    vmovd %xmm0, (%rsi)
; AVX2-NEXT:    vpbroadcastd %xmm0, %xmm0
; AVX2-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-NEXT:    retq
  %in.subvec.not = load <2 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i16> %in.subvec.not, <i16 -1, i16 -1>
  store <2 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 0
  store <2 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 1
  store <2 x i16> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 2
  store <2 x i16> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 3
  store <2 x i16> %in.subvec, ptr %out.subvec3.ptr, align 4
  ret void
}

define void @vec128_v2i32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec128_v2i32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movl %ecx, 4(%rsi)
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movl %ecx, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movl %ecx, 12(%rdx)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec128_v2i32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec128_v2i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec128_v2i32:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec128_v2i32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %xmm0
; AVX512-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX512-NEXT:    retq
  %in.subvec.not = load <2 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i32> %in.subvec.not, <i32 -1, i32 -1>
  store <2 x i32> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 0
  store <2 x i32> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 1
  store <2 x i32> %in.subvec, ptr %out.subvec1.ptr, align 8
  ret void
}

define void @vec128_v2f32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec128_v2f32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movl %ecx, 4(%rsi)
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movl %ecx, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movl %ecx, 12(%rdx)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec128_v2f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec128_v2f32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec128_v2f32:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec128_v2f32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %xmm0
; AVX512-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX512-NEXT:    retq
  %in.subvec.not = load <2 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <2 x i32> %in.subvec.not, <i32 -1, i32 -1>
  %in.subvec = bitcast <2 x i32> %in.subvec.int to <2 x float>
  store <2 x float> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 0
  store <2 x float> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 1
  store <2 x float> %in.subvec, ptr %out.subvec1.ptr, align 8
  ret void
}

define void @vec128_v4i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec128_v4i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl 3(%rdi), %eax
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %r8d
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, 3(%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %dil, 1(%rsi)
; SCALAR-NEXT:    movb %r8b, (%rsi)
; SCALAR-NEXT:    movb %al, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movb %dil, 1(%rdx)
; SCALAR-NEXT:    movb %r8b, (%rdx)
; SCALAR-NEXT:    movb %al, 7(%rdx)
; SCALAR-NEXT:    movb %cl, 6(%rdx)
; SCALAR-NEXT:    movb %dil, 5(%rdx)
; SCALAR-NEXT:    movb %r8b, 4(%rdx)
; SCALAR-NEXT:    movb %al, 11(%rdx)
; SCALAR-NEXT:    movb %cl, 10(%rdx)
; SCALAR-NEXT:    movb %dil, 9(%rdx)
; SCALAR-NEXT:    movb %r8b, 8(%rdx)
; SCALAR-NEXT:    movb %al, 15(%rdx)
; SCALAR-NEXT:    movb %cl, 14(%rdx)
; SCALAR-NEXT:    movb %dil, 13(%rdx)
; SCALAR-NEXT:    movb %r8b, 12(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec128_v4i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movd %xmm0, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec128_v4i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vmovd %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec128_v4i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-NEXT:    vmovd %xmm0, (%rsi)
; AVX2-NEXT:    vpbroadcastd %xmm0, %xmm0
; AVX2-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-NEXT:    retq
  %in.subvec.not = load <4 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1>
  store <4 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 0
  store <4 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 1
  store <4 x i8> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 2
  store <4 x i8> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 3
  store <4 x i8> %in.subvec, ptr %out.subvec3.ptr, align 4
  ret void
}

define void @vec128_v4i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec128_v4i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzwl 6(%rdi), %eax
; SCALAR-NEXT:    movzwl 2(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %r8d
; SCALAR-NEXT:    movl 4(%rdi), %edi
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    movw %ax, 6(%rsi)
; SCALAR-NEXT:    movw %di, 4(%rsi)
; SCALAR-NEXT:    movw %cx, 2(%rsi)
; SCALAR-NEXT:    movw %r8w, (%rsi)
; SCALAR-NEXT:    movw %ax, 6(%rdx)
; SCALAR-NEXT:    movw %di, 4(%rdx)
; SCALAR-NEXT:    movw %cx, 2(%rdx)
; SCALAR-NEXT:    movw %r8w, (%rdx)
; SCALAR-NEXT:    movw %ax, 14(%rdx)
; SCALAR-NEXT:    movw %di, 12(%rdx)
; SCALAR-NEXT:    movw %cx, 10(%rdx)
; SCALAR-NEXT:    movw %r8w, 8(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec128_v4i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec128_v4i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec128_v4i16:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec128_v4i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %xmm0
; AVX512-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX512-NEXT:    retq
  %in.subvec.not = load <4 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1, i16 -1>
  store <4 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 0
  store <4 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 1
  store <4 x i16> %in.subvec, ptr %out.subvec1.ptr, align 8
  ret void
}

define void @vec128_v8i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec128_v8i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzbl 7(%rdi), %ebx
; SCALAR-NEXT:    movzbl 6(%rdi), %r11d
; SCALAR-NEXT:    movzbl 5(%rdi), %r10d
; SCALAR-NEXT:    movzbl 4(%rdi), %r9d
; SCALAR-NEXT:    movzbl 3(%rdi), %r8d
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    notb %r10b
; SCALAR-NEXT:    notb %r11b
; SCALAR-NEXT:    notb %bl
; SCALAR-NEXT:    movb %bl, 7(%rsi)
; SCALAR-NEXT:    movb %r11b, 6(%rsi)
; SCALAR-NEXT:    movb %r10b, 5(%rsi)
; SCALAR-NEXT:    movb %r9b, 4(%rsi)
; SCALAR-NEXT:    movb %r8b, 3(%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %dil, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %bl, 7(%rdx)
; SCALAR-NEXT:    movb %r11b, 6(%rdx)
; SCALAR-NEXT:    movb %r10b, 5(%rdx)
; SCALAR-NEXT:    movb %r9b, 4(%rdx)
; SCALAR-NEXT:    movb %r8b, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movb %dil, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %bl, 15(%rdx)
; SCALAR-NEXT:    movb %r11b, 14(%rdx)
; SCALAR-NEXT:    movb %r10b, 13(%rdx)
; SCALAR-NEXT:    movb %r9b, 12(%rdx)
; SCALAR-NEXT:    movb %r8b, 11(%rdx)
; SCALAR-NEXT:    movb %cl, 10(%rdx)
; SCALAR-NEXT:    movb %dil, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec128_v8i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec128_v8i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec128_v8i8:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec128_v8i8:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %xmm0
; AVX512-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX512-NEXT:    retq
  %in.subvec.not = load <8 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <8 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <8 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 0
  store <8 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 1
  store <8 x i8> %in.subvec, ptr %out.subvec1.ptr, align 8
  ret void
}

define void @vec256_v2i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v2i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %ecx
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %cl, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %cl, 3(%rdx)
; SCALAR-NEXT:    movb %al, 2(%rdx)
; SCALAR-NEXT:    movb %cl, 5(%rdx)
; SCALAR-NEXT:    movb %al, 4(%rdx)
; SCALAR-NEXT:    movb %cl, 7(%rdx)
; SCALAR-NEXT:    movb %al, 6(%rdx)
; SCALAR-NEXT:    movb %cl, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    movb %cl, 11(%rdx)
; SCALAR-NEXT:    movb %al, 10(%rdx)
; SCALAR-NEXT:    movb %cl, 13(%rdx)
; SCALAR-NEXT:    movb %al, 12(%rdx)
; SCALAR-NEXT:    movb %cl, 15(%rdx)
; SCALAR-NEXT:    movb %al, 14(%rdx)
; SCALAR-NEXT:    movb %cl, 17(%rdx)
; SCALAR-NEXT:    movb %al, 16(%rdx)
; SCALAR-NEXT:    movb %cl, 19(%rdx)
; SCALAR-NEXT:    movb %al, 18(%rdx)
; SCALAR-NEXT:    movb %cl, 21(%rdx)
; SCALAR-NEXT:    movb %al, 20(%rdx)
; SCALAR-NEXT:    movb %cl, 23(%rdx)
; SCALAR-NEXT:    movb %al, 22(%rdx)
; SCALAR-NEXT:    movb %cl, 25(%rdx)
; SCALAR-NEXT:    movb %al, 24(%rdx)
; SCALAR-NEXT:    movb %cl, 27(%rdx)
; SCALAR-NEXT:    movb %al, 26(%rdx)
; SCALAR-NEXT:    movb %cl, 29(%rdx)
; SCALAR-NEXT:    movb %al, 28(%rdx)
; SCALAR-NEXT:    movb %cl, 31(%rdx)
; SCALAR-NEXT:    movb %al, 30(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec256_v2i8:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSE2-ONLY-NEXT:    movd %xmm0, %eax
; SSE2-ONLY-NEXT:    movw %ax, (%rsi)
; SSE2-ONLY-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE2-ONLY-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-ONLY-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-ONLY-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec256_v2i8:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE3-NEXT:    pxor (%rdi), %xmm0
; SSE3-NEXT:    movd %xmm0, %eax
; SSE3-NEXT:    movw %ax, (%rsi)
; SSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE3-NEXT:    movdqa %xmm0, (%rdx)
; SSE3-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec256_v2i8:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSSE3-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSSE3-ONLY-NEXT:    movd %xmm0, %eax
; SSSE3-ONLY-NEXT:    movw %ax, (%rsi)
; SSSE3-ONLY-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSSE3-ONLY-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSSE3-ONLY-NEXT:    movdqa %xmm0, (%rdx)
; SSSE3-ONLY-NEXT:    movdqa %xmm0, 16(%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec256_v2i8:
; SSE41:       # %bb.0:
; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE41-NEXT:    pxor (%rdi), %xmm0
; SSE41-NEXT:    pextrw $0, %xmm0, (%rsi)
; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE41-NEXT:    movdqa %xmm0, (%rdx)
; SSE41-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec256_v2i8:
; SSE42:       # %bb.0:
; SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE42-NEXT:    pxor (%rdi), %xmm0
; SSE42-NEXT:    pextrw $0, %xmm0, (%rsi)
; SSE42-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE42-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE42-NEXT:    movdqa %xmm0, (%rdx)
; SSE42-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE42-NEXT:    retq
;
; AVX1-LABEL: vec256_v2i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec256_v2i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVX2-NEXT:    vpbroadcastw %xmm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <2 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i8> %in.subvec.not, <i8 -1, i8 -1>
  store <2 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 0
  store <2 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 1
  store <2 x i8> %in.subvec, ptr %out.subvec1.ptr, align 2
  %out.subvec2.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 2
  store <2 x i8> %in.subvec, ptr %out.subvec2.ptr, align 4
  %out.subvec3.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 3
  store <2 x i8> %in.subvec, ptr %out.subvec3.ptr, align 2
  %out.subvec4.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 4
  store <2 x i8> %in.subvec, ptr %out.subvec4.ptr, align 8
  %out.subvec5.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 5
  store <2 x i8> %in.subvec, ptr %out.subvec5.ptr, align 2
  %out.subvec6.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 6
  store <2 x i8> %in.subvec, ptr %out.subvec6.ptr, align 4
  %out.subvec7.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 7
  store <2 x i8> %in.subvec, ptr %out.subvec7.ptr, align 2
  %out.subvec8.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 8
  store <2 x i8> %in.subvec, ptr %out.subvec8.ptr, align 16
  %out.subvec9.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 9
  store <2 x i8> %in.subvec, ptr %out.subvec9.ptr, align 2
  %out.subvec10.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 10
  store <2 x i8> %in.subvec, ptr %out.subvec10.ptr, align 4
  %out.subvec11.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 11
  store <2 x i8> %in.subvec, ptr %out.subvec11.ptr, align 2
  %out.subvec12.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 12
  store <2 x i8> %in.subvec, ptr %out.subvec12.ptr, align 8
  %out.subvec13.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 13
  store <2 x i8> %in.subvec, ptr %out.subvec13.ptr, align 2
  %out.subvec14.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 14
  store <2 x i8> %in.subvec, ptr %out.subvec14.ptr, align 4
  %out.subvec15.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 15
  store <2 x i8> %in.subvec, ptr %out.subvec15.ptr, align 2
  ret void
}

define void @vec256_v2i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v2i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzwl 2(%rdi), %eax
; SCALAR-NEXT:    movl (%rdi), %ecx
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    movw %ax, 2(%rsi)
; SCALAR-NEXT:    movw %cx, (%rsi)
; SCALAR-NEXT:    movw %ax, 2(%rdx)
; SCALAR-NEXT:    movw %cx, (%rdx)
; SCALAR-NEXT:    movw %ax, 6(%rdx)
; SCALAR-NEXT:    movw %cx, 4(%rdx)
; SCALAR-NEXT:    movw %ax, 10(%rdx)
; SCALAR-NEXT:    movw %cx, 8(%rdx)
; SCALAR-NEXT:    movw %ax, 14(%rdx)
; SCALAR-NEXT:    movw %cx, 12(%rdx)
; SCALAR-NEXT:    movw %ax, 18(%rdx)
; SCALAR-NEXT:    movw %cx, 16(%rdx)
; SCALAR-NEXT:    movw %ax, 22(%rdx)
; SCALAR-NEXT:    movw %cx, 20(%rdx)
; SCALAR-NEXT:    movw %ax, 26(%rdx)
; SCALAR-NEXT:    movw %cx, 24(%rdx)
; SCALAR-NEXT:    movw %ax, 30(%rdx)
; SCALAR-NEXT:    movw %cx, 28(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v2i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movd %xmm0, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec256_v2i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vmovd %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec256_v2i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-NEXT:    vmovd %xmm0, (%rsi)
; AVX2-NEXT:    vpbroadcastd %xmm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <2 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i16> %in.subvec.not, <i16 -1, i16 -1>
  store <2 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 0
  store <2 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 1
  store <2 x i16> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 2
  store <2 x i16> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 3
  store <2 x i16> %in.subvec, ptr %out.subvec3.ptr, align 4
  %out.subvec4.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 4
  store <2 x i16> %in.subvec, ptr %out.subvec4.ptr, align 16
  %out.subvec5.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 5
  store <2 x i16> %in.subvec, ptr %out.subvec5.ptr, align 4
  %out.subvec6.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 6
  store <2 x i16> %in.subvec, ptr %out.subvec6.ptr, align 8
  %out.subvec7.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 7
  store <2 x i16> %in.subvec, ptr %out.subvec7.ptr, align 4
  ret void
}

define void @vec256_v2i32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v2i32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movl %ecx, 4(%rsi)
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movl %ecx, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movl %ecx, 12(%rdx)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    movl %ecx, 20(%rdx)
; SCALAR-NEXT:    movl %eax, 16(%rdx)
; SCALAR-NEXT:    movl %ecx, 28(%rdx)
; SCALAR-NEXT:    movl %eax, 24(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v2i32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec256_v2i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec256_v2i32:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec256_v2i32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX512-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <2 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i32> %in.subvec.not, <i32 -1, i32 -1>
  store <2 x i32> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 0
  store <2 x i32> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 1
  store <2 x i32> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 2
  store <2 x i32> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 3
  store <2 x i32> %in.subvec, ptr %out.subvec3.ptr, align 8
  ret void
}

define void @vec256_v2f32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v2f32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movl %ecx, 4(%rsi)
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movl %ecx, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movl %ecx, 12(%rdx)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    movl %ecx, 20(%rdx)
; SCALAR-NEXT:    movl %eax, 16(%rdx)
; SCALAR-NEXT:    movl %ecx, 28(%rdx)
; SCALAR-NEXT:    movl %eax, 24(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v2f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec256_v2f32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec256_v2f32:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec256_v2f32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX512-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <2 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <2 x i32> %in.subvec.not, <i32 -1, i32 -1>
  %in.subvec = bitcast <2 x i32> %in.subvec.int to <2 x float>
  store <2 x float> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 0
  store <2 x float> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 1
  store <2 x float> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 2
  store <2 x float> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 3
  store <2 x float> %in.subvec, ptr %out.subvec3.ptr, align 8
  ret void
}

define void @vec256_v2i64(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v2i64:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 24(%rdx)
; SCALAR-NEXT:    movq %rax, 16(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v2i64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec256_v2i64:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <2 x i64>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i64> %in.subvec.not, <i64 -1, i64 -1>
  store <2 x i64> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i64>, ptr %out.vec.ptr, i64 0
  store <2 x i64> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i64>, ptr %out.vec.ptr, i64 1
  store <2 x i64> %in.subvec, ptr %out.subvec1.ptr, align 16
  ret void
}

define void @vec256_v2f64(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v2f64:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 24(%rdx)
; SCALAR-NEXT:    movq %rax, 16(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v2f64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec256_v2f64:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <2 x i64>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <2 x i64> %in.subvec.not, <i64 -1, i64 -1>
  %in.subvec = bitcast <2 x i64> %in.subvec.int to <2 x double>
  store <2 x double> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x double>, ptr %out.vec.ptr, i64 0
  store <2 x double> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x double>, ptr %out.vec.ptr, i64 1
  store <2 x double> %in.subvec, ptr %out.subvec1.ptr, align 16
  ret void
}

define void @vec256_v4i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v4i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl 3(%rdi), %r8d
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    movb %r8b, 3(%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %dil, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %r8b, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movb %dil, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %r8b, 7(%rdx)
; SCALAR-NEXT:    movb %cl, 6(%rdx)
; SCALAR-NEXT:    movb %dil, 5(%rdx)
; SCALAR-NEXT:    movb %al, 4(%rdx)
; SCALAR-NEXT:    movb %r8b, 11(%rdx)
; SCALAR-NEXT:    movb %cl, 10(%rdx)
; SCALAR-NEXT:    movb %dil, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    movb %r8b, 15(%rdx)
; SCALAR-NEXT:    movb %cl, 14(%rdx)
; SCALAR-NEXT:    movb %dil, 13(%rdx)
; SCALAR-NEXT:    movb %al, 12(%rdx)
; SCALAR-NEXT:    movb %r8b, 19(%rdx)
; SCALAR-NEXT:    movb %cl, 18(%rdx)
; SCALAR-NEXT:    movb %dil, 17(%rdx)
; SCALAR-NEXT:    movb %al, 16(%rdx)
; SCALAR-NEXT:    movb %r8b, 23(%rdx)
; SCALAR-NEXT:    movb %cl, 22(%rdx)
; SCALAR-NEXT:    movb %dil, 21(%rdx)
; SCALAR-NEXT:    movb %al, 20(%rdx)
; SCALAR-NEXT:    movb %r8b, 27(%rdx)
; SCALAR-NEXT:    movb %cl, 26(%rdx)
; SCALAR-NEXT:    movb %dil, 25(%rdx)
; SCALAR-NEXT:    movb %al, 24(%rdx)
; SCALAR-NEXT:    movb %r8b, 31(%rdx)
; SCALAR-NEXT:    movb %cl, 30(%rdx)
; SCALAR-NEXT:    movb %dil, 29(%rdx)
; SCALAR-NEXT:    movb %al, 28(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v4i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movd %xmm0, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec256_v4i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vmovd %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec256_v4i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-NEXT:    vmovd %xmm0, (%rsi)
; AVX2-NEXT:    vpbroadcastd %xmm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <4 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1>
  store <4 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 0
  store <4 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 1
  store <4 x i8> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 2
  store <4 x i8> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 3
  store <4 x i8> %in.subvec, ptr %out.subvec3.ptr, align 4
  %out.subvec4.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 4
  store <4 x i8> %in.subvec, ptr %out.subvec4.ptr, align 16
  %out.subvec5.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 5
  store <4 x i8> %in.subvec, ptr %out.subvec5.ptr, align 4
  %out.subvec6.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 6
  store <4 x i8> %in.subvec, ptr %out.subvec6.ptr, align 8
  %out.subvec7.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 7
  store <4 x i8> %in.subvec, ptr %out.subvec7.ptr, align 4
  ret void
}

define void @vec256_v4i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v4i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzwl 6(%rdi), %r8d
; SCALAR-NEXT:    movzwl 2(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %edi
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    movw %r8w, 6(%rsi)
; SCALAR-NEXT:    movw %di, 4(%rsi)
; SCALAR-NEXT:    movw %cx, 2(%rsi)
; SCALAR-NEXT:    movw %ax, (%rsi)
; SCALAR-NEXT:    movw %r8w, 6(%rdx)
; SCALAR-NEXT:    movw %di, 4(%rdx)
; SCALAR-NEXT:    movw %cx, 2(%rdx)
; SCALAR-NEXT:    movw %ax, (%rdx)
; SCALAR-NEXT:    movw %r8w, 14(%rdx)
; SCALAR-NEXT:    movw %di, 12(%rdx)
; SCALAR-NEXT:    movw %cx, 10(%rdx)
; SCALAR-NEXT:    movw %ax, 8(%rdx)
; SCALAR-NEXT:    movw %r8w, 22(%rdx)
; SCALAR-NEXT:    movw %di, 20(%rdx)
; SCALAR-NEXT:    movw %cx, 18(%rdx)
; SCALAR-NEXT:    movw %ax, 16(%rdx)
; SCALAR-NEXT:    movw %r8w, 30(%rdx)
; SCALAR-NEXT:    movw %di, 28(%rdx)
; SCALAR-NEXT:    movw %cx, 26(%rdx)
; SCALAR-NEXT:    movw %ax, 24(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v4i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec256_v4i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec256_v4i16:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec256_v4i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX512-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <4 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1, i16 -1>
  store <4 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 0
  store <4 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 1
  store <4 x i16> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 2
  store <4 x i16> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 3
  store <4 x i16> %in.subvec, ptr %out.subvec3.ptr, align 8
  ret void
}

define void @vec256_v4i32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v4i32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movaps (%rdi), %xmm0
; SCALAR-NEXT:    xorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SCALAR-NEXT:    movaps %xmm0, (%rsi)
; SCALAR-NEXT:    movaps %xmm0, (%rdx)
; SCALAR-NEXT:    movaps %xmm0, 16(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v4i32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec256_v4i32:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <4 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1, i32 -1>
  store <4 x i32> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i32>, ptr %out.vec.ptr, i64 0
  store <4 x i32> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i32>, ptr %out.vec.ptr, i64 1
  store <4 x i32> %in.subvec, ptr %out.subvec1.ptr, align 16
  ret void
}

define void @vec256_v4f32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v4f32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movaps (%rdi), %xmm0
; SCALAR-NEXT:    xorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SCALAR-NEXT:    movaps %xmm0, (%rsi)
; SCALAR-NEXT:    movaps %xmm0, (%rdx)
; SCALAR-NEXT:    movaps %xmm0, 16(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v4f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec256_v4f32:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <4 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <4 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1, i32 -1>
  %in.subvec = bitcast <4 x i32> %in.subvec.int to <4 x float>
  store <4 x float> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x float>, ptr %out.vec.ptr, i64 0
  store <4 x float> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x float>, ptr %out.vec.ptr, i64 1
  store <4 x float> %in.subvec, ptr %out.subvec1.ptr, align 16
  ret void
}

define void @vec256_v8i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v8i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzbl 7(%rdi), %ebx
; SCALAR-NEXT:    movzbl 6(%rdi), %r11d
; SCALAR-NEXT:    movzbl 5(%rdi), %r10d
; SCALAR-NEXT:    movzbl 4(%rdi), %r9d
; SCALAR-NEXT:    movzbl 3(%rdi), %r8d
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    notb %r10b
; SCALAR-NEXT:    notb %r11b
; SCALAR-NEXT:    notb %bl
; SCALAR-NEXT:    movb %bl, 7(%rsi)
; SCALAR-NEXT:    movb %r11b, 6(%rsi)
; SCALAR-NEXT:    movb %r10b, 5(%rsi)
; SCALAR-NEXT:    movb %r9b, 4(%rsi)
; SCALAR-NEXT:    movb %r8b, 3(%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %dil, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %bl, 7(%rdx)
; SCALAR-NEXT:    movb %r11b, 6(%rdx)
; SCALAR-NEXT:    movb %r10b, 5(%rdx)
; SCALAR-NEXT:    movb %r9b, 4(%rdx)
; SCALAR-NEXT:    movb %r8b, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movb %dil, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %bl, 15(%rdx)
; SCALAR-NEXT:    movb %r11b, 14(%rdx)
; SCALAR-NEXT:    movb %r10b, 13(%rdx)
; SCALAR-NEXT:    movb %r9b, 12(%rdx)
; SCALAR-NEXT:    movb %r8b, 11(%rdx)
; SCALAR-NEXT:    movb %cl, 10(%rdx)
; SCALAR-NEXT:    movb %dil, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    movb %bl, 23(%rdx)
; SCALAR-NEXT:    movb %r11b, 22(%rdx)
; SCALAR-NEXT:    movb %r10b, 21(%rdx)
; SCALAR-NEXT:    movb %r9b, 20(%rdx)
; SCALAR-NEXT:    movb %r8b, 19(%rdx)
; SCALAR-NEXT:    movb %cl, 18(%rdx)
; SCALAR-NEXT:    movb %dil, 17(%rdx)
; SCALAR-NEXT:    movb %al, 16(%rdx)
; SCALAR-NEXT:    movb %bl, 31(%rdx)
; SCALAR-NEXT:    movb %r11b, 30(%rdx)
; SCALAR-NEXT:    movb %r10b, 29(%rdx)
; SCALAR-NEXT:    movb %r9b, 28(%rdx)
; SCALAR-NEXT:    movb %r8b, 27(%rdx)
; SCALAR-NEXT:    movb %cl, 26(%rdx)
; SCALAR-NEXT:    movb %dil, 25(%rdx)
; SCALAR-NEXT:    movb %al, 24(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v8i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec256_v8i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec256_v8i8:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec256_v8i8:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX512-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <8 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <8 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <8 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 0
  store <8 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 1
  store <8 x i8> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 2
  store <8 x i8> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 3
  store <8 x i8> %in.subvec, ptr %out.subvec3.ptr, align 8
  ret void
}

define void @vec256_v8i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v8i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzwl 14(%rdi), %ebx
; SCALAR-NEXT:    movl 12(%rdi), %r11d
; SCALAR-NEXT:    movzwl 10(%rdi), %r10d
; SCALAR-NEXT:    movl 8(%rdi), %r9d
; SCALAR-NEXT:    movzwl 6(%rdi), %r8d
; SCALAR-NEXT:    movzwl 2(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %edi
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    notl %r9d
; SCALAR-NEXT:    notl %r10d
; SCALAR-NEXT:    notl %r11d
; SCALAR-NEXT:    notl %ebx
; SCALAR-NEXT:    movw %bx, 14(%rsi)
; SCALAR-NEXT:    movw %r11w, 12(%rsi)
; SCALAR-NEXT:    movw %r10w, 10(%rsi)
; SCALAR-NEXT:    movw %r9w, 8(%rsi)
; SCALAR-NEXT:    movw %r8w, 6(%rsi)
; SCALAR-NEXT:    movw %di, 4(%rsi)
; SCALAR-NEXT:    movw %cx, 2(%rsi)
; SCALAR-NEXT:    movw %ax, (%rsi)
; SCALAR-NEXT:    movw %bx, 14(%rdx)
; SCALAR-NEXT:    movw %r11w, 12(%rdx)
; SCALAR-NEXT:    movw %r10w, 10(%rdx)
; SCALAR-NEXT:    movw %r9w, 8(%rdx)
; SCALAR-NEXT:    movw %r8w, 6(%rdx)
; SCALAR-NEXT:    movw %di, 4(%rdx)
; SCALAR-NEXT:    movw %cx, 2(%rdx)
; SCALAR-NEXT:    movw %ax, (%rdx)
; SCALAR-NEXT:    movw %bx, 30(%rdx)
; SCALAR-NEXT:    movw %r11w, 28(%rdx)
; SCALAR-NEXT:    movw %r10w, 26(%rdx)
; SCALAR-NEXT:    movw %r9w, 24(%rdx)
; SCALAR-NEXT:    movw %r8w, 22(%rdx)
; SCALAR-NEXT:    movw %di, 20(%rdx)
; SCALAR-NEXT:    movw %cx, 18(%rdx)
; SCALAR-NEXT:    movw %ax, 16(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v8i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec256_v8i16:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <8 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <8 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
  store <8 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <8 x i16>, ptr %out.vec.ptr, i64 0
  store <8 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <8 x i16>, ptr %out.vec.ptr, i64 1
  store <8 x i16> %in.subvec, ptr %out.subvec1.ptr, align 16
  ret void
}

define void @vec256_v16i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec256_v16i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbp
; SCALAR-NEXT:    pushq %r15
; SCALAR-NEXT:    pushq %r14
; SCALAR-NEXT:    pushq %r13
; SCALAR-NEXT:    pushq %r12
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzbl 15(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 14(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 13(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 12(%rdi), %r15d
; SCALAR-NEXT:    movzbl 11(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 10(%rdi), %ebp
; SCALAR-NEXT:    movzbl 9(%rdi), %r14d
; SCALAR-NEXT:    movzbl 8(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 7(%rdi), %r12d
; SCALAR-NEXT:    movzbl 6(%rdi), %r10d
; SCALAR-NEXT:    movzbl 5(%rdi), %r9d
; SCALAR-NEXT:    movzbl 4(%rdi), %ebx
; SCALAR-NEXT:    movzbl 3(%rdi), %r8d
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %r13d
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r13b
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    movb %r8b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %bl
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    movb %r9b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r10b
; SCALAR-NEXT:    movb %r10b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r12b
; SCALAR-NEXT:    movb %r12b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r11d # 1-byte Folded Reload
; SCALAR-NEXT:    notb %r11b
; SCALAR-NEXT:    movl %r14d, %r10d
; SCALAR-NEXT:    notb %r10b
; SCALAR-NEXT:    notb %bpl
; SCALAR-NEXT:    movl %ebp, %r14d
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 1-byte Folded Reload
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    movb %r8b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movl %r15d, %edi
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    movb %dil, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r9d # 1-byte Folded Reload
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ebp # 1-byte Folded Reload
; SCALAR-NEXT:    notb %bpl
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r15d # 1-byte Folded Reload
; SCALAR-NEXT:    notb %r15b
; SCALAR-NEXT:    movb %r15b, 15(%rsi)
; SCALAR-NEXT:    movb %bpl, 14(%rsi)
; SCALAR-NEXT:    movb %bpl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movl %r9d, %eax
; SCALAR-NEXT:    movb %r9b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %r9b, 13(%rsi)
; SCALAR-NEXT:    movb %dil, 12(%rsi)
; SCALAR-NEXT:    movb %r8b, 11(%rsi)
; SCALAR-NEXT:    movb %r14b, 10(%rsi)
; SCALAR-NEXT:    movb %r10b, 9(%rsi)
; SCALAR-NEXT:    movl %r10d, %r8d
; SCALAR-NEXT:    movb %r11b, 8(%rsi)
; SCALAR-NEXT:    movl %r11d, %r9d
; SCALAR-NEXT:    movb %r11b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %r12b, 7(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %cl, 6(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %dil, 5(%rsi)
; SCALAR-NEXT:    movb %bl, 4(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %cl, 3(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %r13b, 1(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r10d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r10b, (%rsi)
; SCALAR-NEXT:    movb %r15b, 15(%rdx)
; SCALAR-NEXT:    movl %r15d, %r11d
; SCALAR-NEXT:    movb %bpl, 14(%rdx)
; SCALAR-NEXT:    movb %al, 13(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r12d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r12b, 12(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r15d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r15b, 11(%rdx)
; SCALAR-NEXT:    movb %r14b, 10(%rdx)
; SCALAR-NEXT:    movb %r8b, 9(%rdx)
; SCALAR-NEXT:    movb %r9b, 8(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r9d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r9b, 7(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 6(%rdx)
; SCALAR-NEXT:    movb %dil, 5(%rdx)
; SCALAR-NEXT:    movb %bl, 4(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movb %r13b, 1(%rdx)
; SCALAR-NEXT:    movl %r10d, %edi
; SCALAR-NEXT:    movb %r10b, (%rdx)
; SCALAR-NEXT:    movb %r11b, 31(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r10d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r10b, 30(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r10d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r10b, 29(%rdx)
; SCALAR-NEXT:    movb %r12b, 28(%rdx)
; SCALAR-NEXT:    movb %r15b, 27(%rdx)
; SCALAR-NEXT:    movb %r14b, 26(%rdx)
; SCALAR-NEXT:    movb %r8b, 25(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r10d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r10b, 24(%rdx)
; SCALAR-NEXT:    movb %r9b, 23(%rdx)
; SCALAR-NEXT:    movb %al, 22(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 21(%rdx)
; SCALAR-NEXT:    movb %bl, 20(%rdx)
; SCALAR-NEXT:    movb %sil, 19(%rdx)
; SCALAR-NEXT:    movb %cl, 18(%rdx)
; SCALAR-NEXT:    movb %r13b, 17(%rdx)
; SCALAR-NEXT:    movb %dil, 16(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    popq %r12
; SCALAR-NEXT:    popq %r13
; SCALAR-NEXT:    popq %r14
; SCALAR-NEXT:    popq %r15
; SCALAR-NEXT:    popq %rbp
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec256_v16i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec256_v16i8:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <16 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <16 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <16 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <16 x i8>, ptr %out.vec.ptr, i64 0
  store <16 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <16 x i8>, ptr %out.vec.ptr, i64 1
  store <16 x i8> %in.subvec, ptr %out.subvec1.ptr, align 16
  ret void
}

define void @vec384_v2i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v2i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %ecx
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %cl, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %cl, 3(%rdx)
; SCALAR-NEXT:    movb %al, 2(%rdx)
; SCALAR-NEXT:    movb %cl, 5(%rdx)
; SCALAR-NEXT:    movb %al, 4(%rdx)
; SCALAR-NEXT:    movb %cl, 7(%rdx)
; SCALAR-NEXT:    movb %al, 6(%rdx)
; SCALAR-NEXT:    movb %cl, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    movb %cl, 11(%rdx)
; SCALAR-NEXT:    movb %al, 10(%rdx)
; SCALAR-NEXT:    movb %cl, 13(%rdx)
; SCALAR-NEXT:    movb %al, 12(%rdx)
; SCALAR-NEXT:    movb %cl, 15(%rdx)
; SCALAR-NEXT:    movb %al, 14(%rdx)
; SCALAR-NEXT:    movb %cl, 17(%rdx)
; SCALAR-NEXT:    movb %al, 16(%rdx)
; SCALAR-NEXT:    movb %cl, 19(%rdx)
; SCALAR-NEXT:    movb %al, 18(%rdx)
; SCALAR-NEXT:    movb %cl, 21(%rdx)
; SCALAR-NEXT:    movb %al, 20(%rdx)
; SCALAR-NEXT:    movb %cl, 23(%rdx)
; SCALAR-NEXT:    movb %al, 22(%rdx)
; SCALAR-NEXT:    movb %cl, 25(%rdx)
; SCALAR-NEXT:    movb %al, 24(%rdx)
; SCALAR-NEXT:    movb %cl, 27(%rdx)
; SCALAR-NEXT:    movb %al, 26(%rdx)
; SCALAR-NEXT:    movb %cl, 29(%rdx)
; SCALAR-NEXT:    movb %al, 28(%rdx)
; SCALAR-NEXT:    movb %cl, 31(%rdx)
; SCALAR-NEXT:    movb %al, 30(%rdx)
; SCALAR-NEXT:    movb %cl, 33(%rdx)
; SCALAR-NEXT:    movb %al, 32(%rdx)
; SCALAR-NEXT:    movb %cl, 35(%rdx)
; SCALAR-NEXT:    movb %al, 34(%rdx)
; SCALAR-NEXT:    movb %cl, 37(%rdx)
; SCALAR-NEXT:    movb %al, 36(%rdx)
; SCALAR-NEXT:    movb %cl, 39(%rdx)
; SCALAR-NEXT:    movb %al, 38(%rdx)
; SCALAR-NEXT:    movb %cl, 41(%rdx)
; SCALAR-NEXT:    movb %al, 40(%rdx)
; SCALAR-NEXT:    movb %cl, 43(%rdx)
; SCALAR-NEXT:    movb %al, 42(%rdx)
; SCALAR-NEXT:    movb %cl, 45(%rdx)
; SCALAR-NEXT:    movb %al, 44(%rdx)
; SCALAR-NEXT:    movb %cl, 47(%rdx)
; SCALAR-NEXT:    movb %al, 46(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec384_v2i8:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSE2-ONLY-NEXT:    movd %xmm0, %eax
; SSE2-ONLY-NEXT:    movw %ax, (%rsi)
; SSE2-ONLY-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE2-ONLY-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-ONLY-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-ONLY-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-ONLY-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec384_v2i8:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE3-NEXT:    pxor (%rdi), %xmm0
; SSE3-NEXT:    movd %xmm0, %eax
; SSE3-NEXT:    movw %ax, (%rsi)
; SSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE3-NEXT:    movdqa %xmm0, (%rdx)
; SSE3-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE3-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec384_v2i8:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSSE3-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSSE3-ONLY-NEXT:    movd %xmm0, %eax
; SSSE3-ONLY-NEXT:    movw %ax, (%rsi)
; SSSE3-ONLY-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSSE3-ONLY-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSSE3-ONLY-NEXT:    movdqa %xmm0, (%rdx)
; SSSE3-ONLY-NEXT:    movdqa %xmm0, 16(%rdx)
; SSSE3-ONLY-NEXT:    movdqa %xmm0, 32(%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec384_v2i8:
; SSE41:       # %bb.0:
; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE41-NEXT:    pxor (%rdi), %xmm0
; SSE41-NEXT:    pextrw $0, %xmm0, (%rsi)
; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE41-NEXT:    movdqa %xmm0, (%rdx)
; SSE41-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE41-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec384_v2i8:
; SSE42:       # %bb.0:
; SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE42-NEXT:    pxor (%rdi), %xmm0
; SSE42-NEXT:    pextrw $0, %xmm0, (%rsi)
; SSE42-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE42-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE42-NEXT:    movdqa %xmm0, (%rdx)
; SSE42-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE42-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE42-NEXT:    retq
;
; AVX1-LABEL: vec384_v2i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm1
; AVX1-NEXT:    vmovaps %ymm1, (%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec384_v2i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVX2-NEXT:    vpbroadcastw %xmm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <2 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i8> %in.subvec.not, <i8 -1, i8 -1>
  store <2 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 0
  store <2 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 1
  store <2 x i8> %in.subvec, ptr %out.subvec1.ptr, align 2
  %out.subvec2.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 2
  store <2 x i8> %in.subvec, ptr %out.subvec2.ptr, align 4
  %out.subvec3.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 3
  store <2 x i8> %in.subvec, ptr %out.subvec3.ptr, align 2
  %out.subvec4.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 4
  store <2 x i8> %in.subvec, ptr %out.subvec4.ptr, align 8
  %out.subvec5.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 5
  store <2 x i8> %in.subvec, ptr %out.subvec5.ptr, align 2
  %out.subvec6.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 6
  store <2 x i8> %in.subvec, ptr %out.subvec6.ptr, align 4
  %out.subvec7.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 7
  store <2 x i8> %in.subvec, ptr %out.subvec7.ptr, align 2
  %out.subvec8.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 8
  store <2 x i8> %in.subvec, ptr %out.subvec8.ptr, align 16
  %out.subvec9.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 9
  store <2 x i8> %in.subvec, ptr %out.subvec9.ptr, align 2
  %out.subvec10.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 10
  store <2 x i8> %in.subvec, ptr %out.subvec10.ptr, align 4
  %out.subvec11.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 11
  store <2 x i8> %in.subvec, ptr %out.subvec11.ptr, align 2
  %out.subvec12.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 12
  store <2 x i8> %in.subvec, ptr %out.subvec12.ptr, align 8
  %out.subvec13.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 13
  store <2 x i8> %in.subvec, ptr %out.subvec13.ptr, align 2
  %out.subvec14.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 14
  store <2 x i8> %in.subvec, ptr %out.subvec14.ptr, align 4
  %out.subvec15.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 15
  store <2 x i8> %in.subvec, ptr %out.subvec15.ptr, align 2
  %out.subvec16.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 16
  store <2 x i8> %in.subvec, ptr %out.subvec16.ptr, align 32
  %out.subvec17.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 17
  store <2 x i8> %in.subvec, ptr %out.subvec17.ptr, align 2
  %out.subvec18.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 18
  store <2 x i8> %in.subvec, ptr %out.subvec18.ptr, align 4
  %out.subvec19.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 19
  store <2 x i8> %in.subvec, ptr %out.subvec19.ptr, align 2
  %out.subvec20.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 20
  store <2 x i8> %in.subvec, ptr %out.subvec20.ptr, align 8
  %out.subvec21.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 21
  store <2 x i8> %in.subvec, ptr %out.subvec21.ptr, align 2
  %out.subvec22.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 22
  store <2 x i8> %in.subvec, ptr %out.subvec22.ptr, align 4
  %out.subvec23.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 23
  store <2 x i8> %in.subvec, ptr %out.subvec23.ptr, align 2
  ret void
}

define void @vec384_v2i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v2i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzwl 2(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movw %cx, 2(%rsi)
; SCALAR-NEXT:    movw %ax, (%rsi)
; SCALAR-NEXT:    movw %cx, 2(%rdx)
; SCALAR-NEXT:    movw %ax, (%rdx)
; SCALAR-NEXT:    movw %cx, 6(%rdx)
; SCALAR-NEXT:    movw %ax, 4(%rdx)
; SCALAR-NEXT:    movw %cx, 10(%rdx)
; SCALAR-NEXT:    movw %ax, 8(%rdx)
; SCALAR-NEXT:    movw %cx, 14(%rdx)
; SCALAR-NEXT:    movw %ax, 12(%rdx)
; SCALAR-NEXT:    movw %cx, 18(%rdx)
; SCALAR-NEXT:    movw %ax, 16(%rdx)
; SCALAR-NEXT:    movw %cx, 22(%rdx)
; SCALAR-NEXT:    movw %ax, 20(%rdx)
; SCALAR-NEXT:    movw %cx, 26(%rdx)
; SCALAR-NEXT:    movw %ax, 24(%rdx)
; SCALAR-NEXT:    movw %cx, 30(%rdx)
; SCALAR-NEXT:    movw %ax, 28(%rdx)
; SCALAR-NEXT:    movw %cx, 34(%rdx)
; SCALAR-NEXT:    movw %ax, 32(%rdx)
; SCALAR-NEXT:    movw %cx, 38(%rdx)
; SCALAR-NEXT:    movw %ax, 36(%rdx)
; SCALAR-NEXT:    movw %cx, 42(%rdx)
; SCALAR-NEXT:    movw %ax, 40(%rdx)
; SCALAR-NEXT:    movw %cx, 46(%rdx)
; SCALAR-NEXT:    movw %ax, 44(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v2i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movd %xmm0, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v2i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vmovd %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec384_v2i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-NEXT:    vmovd %xmm0, (%rsi)
; AVX2-NEXT:    vpbroadcastd %xmm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <2 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i16> %in.subvec.not, <i16 -1, i16 -1>
  store <2 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 0
  store <2 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 1
  store <2 x i16> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 2
  store <2 x i16> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 3
  store <2 x i16> %in.subvec, ptr %out.subvec3.ptr, align 4
  %out.subvec4.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 4
  store <2 x i16> %in.subvec, ptr %out.subvec4.ptr, align 16
  %out.subvec5.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 5
  store <2 x i16> %in.subvec, ptr %out.subvec5.ptr, align 4
  %out.subvec6.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 6
  store <2 x i16> %in.subvec, ptr %out.subvec6.ptr, align 8
  %out.subvec7.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 7
  store <2 x i16> %in.subvec, ptr %out.subvec7.ptr, align 4
  %out.subvec8.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 8
  store <2 x i16> %in.subvec, ptr %out.subvec8.ptr, align 32
  %out.subvec9.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 9
  store <2 x i16> %in.subvec, ptr %out.subvec9.ptr, align 4
  %out.subvec10.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 10
  store <2 x i16> %in.subvec, ptr %out.subvec10.ptr, align 8
  %out.subvec11.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 11
  store <2 x i16> %in.subvec, ptr %out.subvec11.ptr, align 4
  ret void
}

define void @vec384_v2i32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v2i32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movl %ecx, 4(%rsi)
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movl %ecx, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movl %ecx, 12(%rdx)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    movl %ecx, 20(%rdx)
; SCALAR-NEXT:    movl %eax, 16(%rdx)
; SCALAR-NEXT:    movl %ecx, 28(%rdx)
; SCALAR-NEXT:    movl %eax, 24(%rdx)
; SCALAR-NEXT:    movl %ecx, 36(%rdx)
; SCALAR-NEXT:    movl %eax, 32(%rdx)
; SCALAR-NEXT:    movl %ecx, 44(%rdx)
; SCALAR-NEXT:    movl %eax, 40(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v2i32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v2i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm1
; AVX1-NEXT:    vmovaps %ymm1, (%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec384_v2i32:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec384_v2i32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX512-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX512-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <2 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i32> %in.subvec.not, <i32 -1, i32 -1>
  store <2 x i32> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 0
  store <2 x i32> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 1
  store <2 x i32> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 2
  store <2 x i32> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 3
  store <2 x i32> %in.subvec, ptr %out.subvec3.ptr, align 8
  %out.subvec4.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 4
  store <2 x i32> %in.subvec, ptr %out.subvec4.ptr, align 32
  %out.subvec5.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 5
  store <2 x i32> %in.subvec, ptr %out.subvec5.ptr, align 8
  ret void
}

define void @vec384_v2f32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v2f32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movl %ecx, 4(%rsi)
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movl %ecx, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movl %ecx, 12(%rdx)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    movl %ecx, 20(%rdx)
; SCALAR-NEXT:    movl %eax, 16(%rdx)
; SCALAR-NEXT:    movl %ecx, 28(%rdx)
; SCALAR-NEXT:    movl %eax, 24(%rdx)
; SCALAR-NEXT:    movl %ecx, 36(%rdx)
; SCALAR-NEXT:    movl %eax, 32(%rdx)
; SCALAR-NEXT:    movl %ecx, 44(%rdx)
; SCALAR-NEXT:    movl %eax, 40(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v2f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v2f32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm1
; AVX1-NEXT:    vmovaps %ymm1, (%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec384_v2f32:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec384_v2f32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX512-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX512-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <2 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <2 x i32> %in.subvec.not, <i32 -1, i32 -1>
  %in.subvec = bitcast <2 x i32> %in.subvec.int to <2 x float>
  store <2 x float> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 0
  store <2 x float> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 1
  store <2 x float> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 2
  store <2 x float> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 3
  store <2 x float> %in.subvec, ptr %out.subvec3.ptr, align 8
  %out.subvec4.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 4
  store <2 x float> %in.subvec, ptr %out.subvec4.ptr, align 32
  %out.subvec5.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 5
  store <2 x float> %in.subvec, ptr %out.subvec5.ptr, align 8
  ret void
}

define void @vec384_v2i64(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v2i64:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 24(%rdx)
; SCALAR-NEXT:    movq %rax, 16(%rdx)
; SCALAR-NEXT:    movq %rcx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v2i64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec384_v2i64:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <2 x i64>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i64> %in.subvec.not, <i64 -1, i64 -1>
  store <2 x i64> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i64>, ptr %out.vec.ptr, i64 0
  store <2 x i64> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i64>, ptr %out.vec.ptr, i64 1
  store <2 x i64> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <2 x i64>, ptr %out.vec.ptr, i64 2
  store <2 x i64> %in.subvec, ptr %out.subvec2.ptr, align 32
  ret void
}

define void @vec384_v2f64(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v2f64:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 24(%rdx)
; SCALAR-NEXT:    movq %rax, 16(%rdx)
; SCALAR-NEXT:    movq %rcx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v2f64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec384_v2f64:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <2 x i64>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <2 x i64> %in.subvec.not, <i64 -1, i64 -1>
  %in.subvec = bitcast <2 x i64> %in.subvec.int to <2 x double>
  store <2 x double> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x double>, ptr %out.vec.ptr, i64 0
  store <2 x double> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x double>, ptr %out.vec.ptr, i64 1
  store <2 x double> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <2 x double>, ptr %out.vec.ptr, i64 2
  store <2 x double> %in.subvec, ptr %out.subvec2.ptr, align 32
  ret void
}

define void @vec384_v3i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v3i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl %eax, %ecx
; SCALAR-NEXT:    shrl $16, %ecx
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    movw %ax, (%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movw %ax, (%rdx)
; SCALAR-NEXT:    movb %cl, 6(%rdx)
; SCALAR-NEXT:    movw %ax, 4(%rdx)
; SCALAR-NEXT:    movb %cl, 10(%rdx)
; SCALAR-NEXT:    movw %ax, 8(%rdx)
; SCALAR-NEXT:    movb %cl, 14(%rdx)
; SCALAR-NEXT:    movw %ax, 12(%rdx)
; SCALAR-NEXT:    movb %cl, 18(%rdx)
; SCALAR-NEXT:    movw %ax, 16(%rdx)
; SCALAR-NEXT:    movb %cl, 22(%rdx)
; SCALAR-NEXT:    movw %ax, 20(%rdx)
; SCALAR-NEXT:    movb %cl, 26(%rdx)
; SCALAR-NEXT:    movw %ax, 24(%rdx)
; SCALAR-NEXT:    movb %cl, 30(%rdx)
; SCALAR-NEXT:    movw %ax, 28(%rdx)
; SCALAR-NEXT:    movb %cl, 34(%rdx)
; SCALAR-NEXT:    movw %ax, 32(%rdx)
; SCALAR-NEXT:    movb %cl, 38(%rdx)
; SCALAR-NEXT:    movw %ax, 36(%rdx)
; SCALAR-NEXT:    movb %cl, 42(%rdx)
; SCALAR-NEXT:    movw %ax, 40(%rdx)
; SCALAR-NEXT:    movb %cl, 46(%rdx)
; SCALAR-NEXT:    movw %ax, 44(%rdx)
; SCALAR-NEXT:    movb %cl, 50(%rdx)
; SCALAR-NEXT:    movw %ax, 48(%rdx)
; SCALAR-NEXT:    movb %cl, 54(%rdx)
; SCALAR-NEXT:    movw %ax, 52(%rdx)
; SCALAR-NEXT:    movb %cl, 58(%rdx)
; SCALAR-NEXT:    movw %ax, 56(%rdx)
; SCALAR-NEXT:    movb %cl, 62(%rdx)
; SCALAR-NEXT:    movw %ax, 60(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec384_v3i8:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    movl (%rdi), %eax
; SSE2-ONLY-NEXT:    notl %eax
; SSE2-ONLY-NEXT:    movw %ax, (%rsi)
; SSE2-ONLY-NEXT:    movl %eax, %ecx
; SSE2-ONLY-NEXT:    shrl $16, %ecx
; SSE2-ONLY-NEXT:    movb %cl, 2(%rsi)
; SSE2-ONLY-NEXT:    movb %cl, 2(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, (%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 6(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 4(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 10(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 8(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 14(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 12(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 18(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 16(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 22(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 20(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 26(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 24(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 30(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 28(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 34(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 32(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 38(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 36(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 42(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 40(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 46(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 44(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 50(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 48(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 54(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 52(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 58(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 56(%rdx)
; SSE2-ONLY-NEXT:    movb %cl, 62(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 60(%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec384_v3i8:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movl (%rdi), %eax
; SSE3-NEXT:    notl %eax
; SSE3-NEXT:    movw %ax, (%rsi)
; SSE3-NEXT:    movl %eax, %ecx
; SSE3-NEXT:    shrl $16, %ecx
; SSE3-NEXT:    movb %cl, 2(%rsi)
; SSE3-NEXT:    movb %cl, 2(%rdx)
; SSE3-NEXT:    movw %ax, (%rdx)
; SSE3-NEXT:    movb %cl, 6(%rdx)
; SSE3-NEXT:    movw %ax, 4(%rdx)
; SSE3-NEXT:    movb %cl, 10(%rdx)
; SSE3-NEXT:    movw %ax, 8(%rdx)
; SSE3-NEXT:    movb %cl, 14(%rdx)
; SSE3-NEXT:    movw %ax, 12(%rdx)
; SSE3-NEXT:    movb %cl, 18(%rdx)
; SSE3-NEXT:    movw %ax, 16(%rdx)
; SSE3-NEXT:    movb %cl, 22(%rdx)
; SSE3-NEXT:    movw %ax, 20(%rdx)
; SSE3-NEXT:    movb %cl, 26(%rdx)
; SSE3-NEXT:    movw %ax, 24(%rdx)
; SSE3-NEXT:    movb %cl, 30(%rdx)
; SSE3-NEXT:    movw %ax, 28(%rdx)
; SSE3-NEXT:    movb %cl, 34(%rdx)
; SSE3-NEXT:    movw %ax, 32(%rdx)
; SSE3-NEXT:    movb %cl, 38(%rdx)
; SSE3-NEXT:    movw %ax, 36(%rdx)
; SSE3-NEXT:    movb %cl, 42(%rdx)
; SSE3-NEXT:    movw %ax, 40(%rdx)
; SSE3-NEXT:    movb %cl, 46(%rdx)
; SSE3-NEXT:    movw %ax, 44(%rdx)
; SSE3-NEXT:    movb %cl, 50(%rdx)
; SSE3-NEXT:    movw %ax, 48(%rdx)
; SSE3-NEXT:    movb %cl, 54(%rdx)
; SSE3-NEXT:    movw %ax, 52(%rdx)
; SSE3-NEXT:    movb %cl, 58(%rdx)
; SSE3-NEXT:    movw %ax, 56(%rdx)
; SSE3-NEXT:    movb %cl, 62(%rdx)
; SSE3-NEXT:    movw %ax, 60(%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec384_v3i8:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    movl (%rdi), %eax
; SSSE3-ONLY-NEXT:    notl %eax
; SSSE3-ONLY-NEXT:    movw %ax, (%rsi)
; SSSE3-ONLY-NEXT:    movl %eax, %ecx
; SSSE3-ONLY-NEXT:    shrl $16, %ecx
; SSSE3-ONLY-NEXT:    movb %cl, 2(%rsi)
; SSSE3-ONLY-NEXT:    movb %cl, 2(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, (%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 6(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 4(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 10(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 8(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 14(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 12(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 18(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 16(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 22(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 20(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 26(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 24(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 30(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 28(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 34(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 32(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 38(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 36(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 42(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 40(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 46(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 44(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 50(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 48(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 54(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 52(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 58(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 56(%rdx)
; SSSE3-ONLY-NEXT:    movb %cl, 62(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 60(%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec384_v3i8:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE41-NEXT:    pxor %xmm1, %xmm0
; SSE41-NEXT:    pextrb $2, %xmm0, 2(%rsi)
; SSE41-NEXT:    movd %xmm0, %eax
; SSE41-NEXT:    movw %ax, (%rsi)
; SSE41-NEXT:    pextrb $2, %xmm0, 2(%rdx)
; SSE41-NEXT:    movw %ax, (%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 6(%rdx)
; SSE41-NEXT:    movw %ax, 4(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 10(%rdx)
; SSE41-NEXT:    movw %ax, 8(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 14(%rdx)
; SSE41-NEXT:    movw %ax, 12(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 18(%rdx)
; SSE41-NEXT:    movw %ax, 16(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 22(%rdx)
; SSE41-NEXT:    movw %ax, 20(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 26(%rdx)
; SSE41-NEXT:    movw %ax, 24(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 30(%rdx)
; SSE41-NEXT:    movw %ax, 28(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 34(%rdx)
; SSE41-NEXT:    movw %ax, 32(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 38(%rdx)
; SSE41-NEXT:    movw %ax, 36(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 42(%rdx)
; SSE41-NEXT:    movw %ax, 40(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 46(%rdx)
; SSE41-NEXT:    movw %ax, 44(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 50(%rdx)
; SSE41-NEXT:    movw %ax, 48(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 54(%rdx)
; SSE41-NEXT:    movw %ax, 52(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 58(%rdx)
; SSE41-NEXT:    movw %ax, 56(%rdx)
; SSE41-NEXT:    pextrb $2, %xmm0, 62(%rdx)
; SSE41-NEXT:    movw %ax, 60(%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec384_v3i8:
; SSE42:       # %bb.0:
; SSE42-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE42-NEXT:    pxor %xmm1, %xmm0
; SSE42-NEXT:    pextrb $2, %xmm0, 2(%rsi)
; SSE42-NEXT:    movd %xmm0, %eax
; SSE42-NEXT:    movw %ax, (%rsi)
; SSE42-NEXT:    pextrb $2, %xmm0, 2(%rdx)
; SSE42-NEXT:    movw %ax, (%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 6(%rdx)
; SSE42-NEXT:    movw %ax, 4(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 10(%rdx)
; SSE42-NEXT:    movw %ax, 8(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 14(%rdx)
; SSE42-NEXT:    movw %ax, 12(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 18(%rdx)
; SSE42-NEXT:    movw %ax, 16(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 22(%rdx)
; SSE42-NEXT:    movw %ax, 20(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 26(%rdx)
; SSE42-NEXT:    movw %ax, 24(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 30(%rdx)
; SSE42-NEXT:    movw %ax, 28(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 34(%rdx)
; SSE42-NEXT:    movw %ax, 32(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 38(%rdx)
; SSE42-NEXT:    movw %ax, 36(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 42(%rdx)
; SSE42-NEXT:    movw %ax, 40(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 46(%rdx)
; SSE42-NEXT:    movw %ax, 44(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 50(%rdx)
; SSE42-NEXT:    movw %ax, 48(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 54(%rdx)
; SSE42-NEXT:    movw %ax, 52(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 58(%rdx)
; SSE42-NEXT:    movw %ax, 56(%rdx)
; SSE42-NEXT:    pextrb $2, %xmm0, 62(%rdx)
; SSE42-NEXT:    movw %ax, 60(%rdx)
; SSE42-NEXT:    retq
;
; AVX1-LABEL: vec384_v3i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vpextrb $2, %xmm0, 2(%rsi)
; AVX1-NEXT:    vmovd %xmm0, %eax
; AVX1-NEXT:    movw %ax, (%rsi)
; AVX1-NEXT:    vpextrb $2, %xmm0, 2(%rdx)
; AVX1-NEXT:    movw %ax, (%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 6(%rdx)
; AVX1-NEXT:    movw %ax, 4(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 10(%rdx)
; AVX1-NEXT:    movw %ax, 8(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 14(%rdx)
; AVX1-NEXT:    movw %ax, 12(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 18(%rdx)
; AVX1-NEXT:    movw %ax, 16(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 22(%rdx)
; AVX1-NEXT:    movw %ax, 20(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 26(%rdx)
; AVX1-NEXT:    movw %ax, 24(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 30(%rdx)
; AVX1-NEXT:    movw %ax, 28(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 34(%rdx)
; AVX1-NEXT:    movw %ax, 32(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 38(%rdx)
; AVX1-NEXT:    movw %ax, 36(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 42(%rdx)
; AVX1-NEXT:    movw %ax, 40(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 46(%rdx)
; AVX1-NEXT:    movw %ax, 44(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 50(%rdx)
; AVX1-NEXT:    movw %ax, 48(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 54(%rdx)
; AVX1-NEXT:    movw %ax, 52(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 58(%rdx)
; AVX1-NEXT:    movw %ax, 56(%rdx)
; AVX1-NEXT:    vpextrb $2, %xmm0, 62(%rdx)
; AVX1-NEXT:    movw %ax, 60(%rdx)
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec384_v3i8:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 2(%rsi)
; AVX2-ONLY-NEXT:    vmovd %xmm0, %eax
; AVX2-ONLY-NEXT:    movw %ax, (%rsi)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 2(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, (%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 6(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 4(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 10(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 8(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 14(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 12(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 18(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 16(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 22(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 20(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 26(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 24(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 30(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 28(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 34(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 32(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 38(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 36(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 42(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 40(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 46(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 44(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 50(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 48(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 54(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 52(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 58(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 56(%rdx)
; AVX2-ONLY-NEXT:    vpextrb $2, %xmm0, 62(%rdx)
; AVX2-ONLY-NEXT:    movw %ax, 60(%rdx)
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec384_v3i8:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vpextrb $2, %xmm0, 2(%rsi)
; AVX512-NEXT:    vmovd %xmm0, %eax
; AVX512-NEXT:    movw %ax, (%rsi)
; AVX512-NEXT:    vpextrb $2, %xmm0, 2(%rdx)
; AVX512-NEXT:    movw %ax, (%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 6(%rdx)
; AVX512-NEXT:    movw %ax, 4(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 10(%rdx)
; AVX512-NEXT:    movw %ax, 8(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 14(%rdx)
; AVX512-NEXT:    movw %ax, 12(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 18(%rdx)
; AVX512-NEXT:    movw %ax, 16(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 22(%rdx)
; AVX512-NEXT:    movw %ax, 20(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 26(%rdx)
; AVX512-NEXT:    movw %ax, 24(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 30(%rdx)
; AVX512-NEXT:    movw %ax, 28(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 34(%rdx)
; AVX512-NEXT:    movw %ax, 32(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 38(%rdx)
; AVX512-NEXT:    movw %ax, 36(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 42(%rdx)
; AVX512-NEXT:    movw %ax, 40(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 46(%rdx)
; AVX512-NEXT:    movw %ax, 44(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 50(%rdx)
; AVX512-NEXT:    movw %ax, 48(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 54(%rdx)
; AVX512-NEXT:    movw %ax, 52(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 58(%rdx)
; AVX512-NEXT:    movw %ax, 56(%rdx)
; AVX512-NEXT:    vpextrb $2, %xmm0, 62(%rdx)
; AVX512-NEXT:    movw %ax, 60(%rdx)
; AVX512-NEXT:    retq
  %in.subvec.not = load <3 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <3 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1>
  store <3 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 0
  store <3 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 1
  store <3 x i8> %in.subvec, ptr %out.subvec1.ptr, align 1
  %out.subvec2.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 2
  store <3 x i8> %in.subvec, ptr %out.subvec2.ptr, align 2
  %out.subvec3.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 3
  store <3 x i8> %in.subvec, ptr %out.subvec3.ptr, align 1
  %out.subvec4.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 4
  store <3 x i8> %in.subvec, ptr %out.subvec4.ptr, align 4
  %out.subvec5.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 5
  store <3 x i8> %in.subvec, ptr %out.subvec5.ptr, align 1
  %out.subvec6.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 6
  store <3 x i8> %in.subvec, ptr %out.subvec6.ptr, align 2
  %out.subvec7.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 7
  store <3 x i8> %in.subvec, ptr %out.subvec7.ptr, align 1
  %out.subvec8.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 8
  store <3 x i8> %in.subvec, ptr %out.subvec8.ptr, align 8
  %out.subvec9.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 9
  store <3 x i8> %in.subvec, ptr %out.subvec9.ptr, align 1
  %out.subvec10.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 10
  store <3 x i8> %in.subvec, ptr %out.subvec10.ptr, align 2
  %out.subvec11.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 11
  store <3 x i8> %in.subvec, ptr %out.subvec11.ptr, align 1
  %out.subvec12.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 12
  store <3 x i8> %in.subvec, ptr %out.subvec12.ptr, align 4
  %out.subvec13.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 13
  store <3 x i8> %in.subvec, ptr %out.subvec13.ptr, align 1
  %out.subvec14.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 14
  store <3 x i8> %in.subvec, ptr %out.subvec14.ptr, align 2
  %out.subvec15.ptr = getelementptr <3 x i8>, ptr %out.vec.ptr, i64 15
  store <3 x i8> %in.subvec, ptr %out.subvec15.ptr, align 1
  ret void
}

define void @vec384_v3i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v3i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq %rax, %rcx
; SCALAR-NEXT:    shrq $32, %rcx
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movw %cx, 4(%rsi)
; SCALAR-NEXT:    movw %cx, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movw %cx, 12(%rdx)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    movw %cx, 20(%rdx)
; SCALAR-NEXT:    movl %eax, 16(%rdx)
; SCALAR-NEXT:    movw %cx, 28(%rdx)
; SCALAR-NEXT:    movl %eax, 24(%rdx)
; SCALAR-NEXT:    movw %cx, 36(%rdx)
; SCALAR-NEXT:    movl %eax, 32(%rdx)
; SCALAR-NEXT:    movw %cx, 44(%rdx)
; SCALAR-NEXT:    movl %eax, 40(%rdx)
; SCALAR-NEXT:    movw %cx, 52(%rdx)
; SCALAR-NEXT:    movl %eax, 48(%rdx)
; SCALAR-NEXT:    movw %cx, 60(%rdx)
; SCALAR-NEXT:    movl %eax, 56(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec384_v3i16:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-ONLY-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-ONLY-NEXT:    pxor %xmm0, %xmm1
; SSE2-ONLY-NEXT:    movd %xmm1, (%rsi)
; SSE2-ONLY-NEXT:    pextrw $2, %xmm1, %eax
; SSE2-ONLY-NEXT:    movw %ax, 4(%rsi)
; SSE2-ONLY-NEXT:    movw %ax, 4(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, (%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 12(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 20(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 16(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 28(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 36(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 32(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 44(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 52(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 48(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 60(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec384_v3i16:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE3-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE3-NEXT:    pxor %xmm0, %xmm1
; SSE3-NEXT:    movd %xmm1, (%rsi)
; SSE3-NEXT:    pextrw $2, %xmm1, %eax
; SSE3-NEXT:    movw %ax, 4(%rsi)
; SSE3-NEXT:    movw %ax, 4(%rdx)
; SSE3-NEXT:    movd %xmm1, (%rdx)
; SSE3-NEXT:    movw %ax, 12(%rdx)
; SSE3-NEXT:    movd %xmm1, 8(%rdx)
; SSE3-NEXT:    movw %ax, 20(%rdx)
; SSE3-NEXT:    movd %xmm1, 16(%rdx)
; SSE3-NEXT:    movw %ax, 28(%rdx)
; SSE3-NEXT:    movd %xmm1, 24(%rdx)
; SSE3-NEXT:    movw %ax, 36(%rdx)
; SSE3-NEXT:    movd %xmm1, 32(%rdx)
; SSE3-NEXT:    movw %ax, 44(%rdx)
; SSE3-NEXT:    movd %xmm1, 40(%rdx)
; SSE3-NEXT:    movw %ax, 52(%rdx)
; SSE3-NEXT:    movd %xmm1, 48(%rdx)
; SSE3-NEXT:    movw %ax, 60(%rdx)
; SSE3-NEXT:    movd %xmm1, 56(%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec384_v3i16:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSSE3-ONLY-NEXT:    pcmpeqd %xmm1, %xmm1
; SSSE3-ONLY-NEXT:    pxor %xmm0, %xmm1
; SSSE3-ONLY-NEXT:    movd %xmm1, (%rsi)
; SSSE3-ONLY-NEXT:    pextrw $2, %xmm1, %eax
; SSSE3-ONLY-NEXT:    movw %ax, 4(%rsi)
; SSSE3-ONLY-NEXT:    movw %ax, 4(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, (%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 12(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 20(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 16(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 28(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 36(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 32(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 44(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 52(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 48(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 60(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec384_v3i16:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE41-NEXT:    pxor %xmm0, %xmm1
; SSE41-NEXT:    pextrw $2, %xmm1, 4(%rsi)
; SSE41-NEXT:    movd %xmm1, (%rsi)
; SSE41-NEXT:    pextrw $2, %xmm1, 4(%rdx)
; SSE41-NEXT:    movd %xmm1, (%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 12(%rdx)
; SSE41-NEXT:    movd %xmm1, 8(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 20(%rdx)
; SSE41-NEXT:    movd %xmm1, 16(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 28(%rdx)
; SSE41-NEXT:    movd %xmm1, 24(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 36(%rdx)
; SSE41-NEXT:    movd %xmm1, 32(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 44(%rdx)
; SSE41-NEXT:    movd %xmm1, 40(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 52(%rdx)
; SSE41-NEXT:    movd %xmm1, 48(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 60(%rdx)
; SSE41-NEXT:    movd %xmm1, 56(%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec384_v3i16:
; SSE42:       # %bb.0:
; SSE42-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE42-NEXT:    pxor %xmm0, %xmm1
; SSE42-NEXT:    pextrw $2, %xmm1, 4(%rsi)
; SSE42-NEXT:    movd %xmm1, (%rsi)
; SSE42-NEXT:    pextrw $2, %xmm1, 4(%rdx)
; SSE42-NEXT:    movd %xmm1, (%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 12(%rdx)
; SSE42-NEXT:    movd %xmm1, 8(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 20(%rdx)
; SSE42-NEXT:    movd %xmm1, 16(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 28(%rdx)
; SSE42-NEXT:    movd %xmm1, 24(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 36(%rdx)
; SSE42-NEXT:    movd %xmm1, 32(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 44(%rdx)
; SSE42-NEXT:    movd %xmm1, 40(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 52(%rdx)
; SSE42-NEXT:    movd %xmm1, 48(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 60(%rdx)
; SSE42-NEXT:    movd %xmm1, 56(%rdx)
; SSE42-NEXT:    retq
;
; AVX1-LABEL: vec384_v3i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vpextrw $2, %xmm0, 4(%rsi)
; AVX1-NEXT:    vmovd %xmm0, (%rsi)
; AVX1-NEXT:    vpextrw $2, %xmm0, 4(%rdx)
; AVX1-NEXT:    vmovd %xmm0, (%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 12(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 8(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 20(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 16(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 28(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 24(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 36(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 32(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 44(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 40(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 52(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 48(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 60(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 56(%rdx)
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec384_v3i16:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 4(%rsi)
; AVX2-ONLY-NEXT:    vmovd %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 4(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, (%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 12(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 8(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 20(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 16(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 28(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 24(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 36(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 44(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 40(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 52(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 48(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 60(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 56(%rdx)
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec384_v3i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vpextrw $2, %xmm0, 4(%rsi)
; AVX512-NEXT:    vmovd %xmm0, (%rsi)
; AVX512-NEXT:    vpextrw $2, %xmm0, 4(%rdx)
; AVX512-NEXT:    vmovd %xmm0, (%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 12(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 8(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 20(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 16(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 28(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 24(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 36(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 32(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 44(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 40(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 52(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 48(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 60(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 56(%rdx)
; AVX512-NEXT:    retq
  %in.subvec.not = load <3 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <3 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1>
  store <3 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <3 x i16>, ptr %out.vec.ptr, i64 0
  store <3 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <3 x i16>, ptr %out.vec.ptr, i64 1
  store <3 x i16> %in.subvec, ptr %out.subvec1.ptr, align 2
  %out.subvec2.ptr = getelementptr <3 x i16>, ptr %out.vec.ptr, i64 2
  store <3 x i16> %in.subvec, ptr %out.subvec2.ptr, align 4
  %out.subvec3.ptr = getelementptr <3 x i16>, ptr %out.vec.ptr, i64 3
  store <3 x i16> %in.subvec, ptr %out.subvec3.ptr, align 2
  %out.subvec4.ptr = getelementptr <3 x i16>, ptr %out.vec.ptr, i64 4
  store <3 x i16> %in.subvec, ptr %out.subvec4.ptr, align 8
  %out.subvec5.ptr = getelementptr <3 x i16>, ptr %out.vec.ptr, i64 5
  store <3 x i16> %in.subvec, ptr %out.subvec5.ptr, align 2
  %out.subvec6.ptr = getelementptr <3 x i16>, ptr %out.vec.ptr, i64 6
  store <3 x i16> %in.subvec, ptr %out.subvec6.ptr, align 4
  %out.subvec7.ptr = getelementptr <3 x i16>, ptr %out.vec.ptr, i64 7
  store <3 x i16> %in.subvec, ptr %out.subvec7.ptr, align 2
  ret void
}

define void @vec384_v3i32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v3i32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl 8(%rdi), %eax
; SCALAR-NEXT:    movq (%rdi), %rcx
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    movl %eax, 8(%rsi)
; SCALAR-NEXT:    movq %rcx, (%rsi)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    movq %rcx, (%rdx)
; SCALAR-NEXT:    movl %eax, 24(%rdx)
; SCALAR-NEXT:    movq %rcx, 16(%rdx)
; SCALAR-NEXT:    movl %eax, 40(%rdx)
; SCALAR-NEXT:    movq %rcx, 32(%rdx)
; SCALAR-NEXT:    movl %eax, 56(%rdx)
; SCALAR-NEXT:    movq %rcx, 48(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec384_v3i32:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSE2-ONLY-NEXT:    movq %xmm0, (%rsi)
; SSE2-ONLY-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE2-ONLY-NEXT:    movd %xmm1, 8(%rsi)
; SSE2-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, (%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 16(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 32(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 48(%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec384_v3i32:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE3-NEXT:    pxor (%rdi), %xmm0
; SSE3-NEXT:    movq %xmm0, (%rsi)
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-NEXT:    movd %xmm1, 8(%rsi)
; SSE3-NEXT:    movd %xmm1, 8(%rdx)
; SSE3-NEXT:    movq %xmm0, (%rdx)
; SSE3-NEXT:    movd %xmm1, 24(%rdx)
; SSE3-NEXT:    movq %xmm0, 16(%rdx)
; SSE3-NEXT:    movd %xmm1, 40(%rdx)
; SSE3-NEXT:    movq %xmm0, 32(%rdx)
; SSE3-NEXT:    movd %xmm1, 56(%rdx)
; SSE3-NEXT:    movq %xmm0, 48(%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec384_v3i32:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSSE3-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSSE3-ONLY-NEXT:    movq %xmm0, (%rsi)
; SSSE3-ONLY-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSSE3-ONLY-NEXT:    movd %xmm1, 8(%rsi)
; SSSE3-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, (%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 16(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 32(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 48(%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec384_v3i32:
; SSE41:       # %bb.0:
; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE41-NEXT:    pxor (%rdi), %xmm0
; SSE41-NEXT:    pextrd $2, %xmm0, 8(%rsi)
; SSE41-NEXT:    movq %xmm0, (%rsi)
; SSE41-NEXT:    pextrd $2, %xmm0, 8(%rdx)
; SSE41-NEXT:    movq %xmm0, (%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 24(%rdx)
; SSE41-NEXT:    movq %xmm0, 16(%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 40(%rdx)
; SSE41-NEXT:    movq %xmm0, 32(%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 56(%rdx)
; SSE41-NEXT:    movq %xmm0, 48(%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec384_v3i32:
; SSE42:       # %bb.0:
; SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE42-NEXT:    pxor (%rdi), %xmm0
; SSE42-NEXT:    pextrd $2, %xmm0, 8(%rsi)
; SSE42-NEXT:    movq %xmm0, (%rsi)
; SSE42-NEXT:    pextrd $2, %xmm0, 8(%rdx)
; SSE42-NEXT:    movq %xmm0, (%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 24(%rdx)
; SSE42-NEXT:    movq %xmm0, 16(%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 40(%rdx)
; SSE42-NEXT:    movq %xmm0, 32(%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 56(%rdx)
; SSE42-NEXT:    movq %xmm0, 48(%rdx)
; SSE42-NEXT:    retq
;
; AVX-LABEL: vec384_v3i32:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vpextrd $2, %xmm0, 8(%rsi)
; AVX-NEXT:    vmovq %xmm0, (%rsi)
; AVX-NEXT:    vpextrd $2, %xmm0, 8(%rdx)
; AVX-NEXT:    vmovq %xmm0, (%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 24(%rdx)
; AVX-NEXT:    vmovq %xmm0, 16(%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 40(%rdx)
; AVX-NEXT:    vmovq %xmm0, 32(%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 56(%rdx)
; AVX-NEXT:    vmovq %xmm0, 48(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <3 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <3 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1>
  store <3 x i32> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <3 x i32>, ptr %out.vec.ptr, i64 0
  store <3 x i32> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <3 x i32>, ptr %out.vec.ptr, i64 1
  store <3 x i32> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <3 x i32>, ptr %out.vec.ptr, i64 2
  store <3 x i32> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <3 x i32>, ptr %out.vec.ptr, i64 3
  store <3 x i32> %in.subvec, ptr %out.subvec3.ptr, align 4
  ret void
}

define void @vec384_v3f32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v3f32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl 8(%rdi), %eax
; SCALAR-NEXT:    movq (%rdi), %rcx
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    movl %eax, 8(%rsi)
; SCALAR-NEXT:    movq %rcx, (%rsi)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    movq %rcx, (%rdx)
; SCALAR-NEXT:    movl %eax, 24(%rdx)
; SCALAR-NEXT:    movq %rcx, 16(%rdx)
; SCALAR-NEXT:    movl %eax, 40(%rdx)
; SCALAR-NEXT:    movq %rcx, 32(%rdx)
; SCALAR-NEXT:    movl %eax, 56(%rdx)
; SCALAR-NEXT:    movq %rcx, 48(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec384_v3f32:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSE2-ONLY-NEXT:    movq %xmm0, (%rsi)
; SSE2-ONLY-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE2-ONLY-NEXT:    movd %xmm1, 8(%rsi)
; SSE2-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, (%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 16(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 32(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 48(%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec384_v3f32:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE3-NEXT:    pxor (%rdi), %xmm0
; SSE3-NEXT:    movq %xmm0, (%rsi)
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-NEXT:    movd %xmm1, 8(%rsi)
; SSE3-NEXT:    movd %xmm1, 8(%rdx)
; SSE3-NEXT:    movq %xmm0, (%rdx)
; SSE3-NEXT:    movd %xmm1, 24(%rdx)
; SSE3-NEXT:    movq %xmm0, 16(%rdx)
; SSE3-NEXT:    movd %xmm1, 40(%rdx)
; SSE3-NEXT:    movq %xmm0, 32(%rdx)
; SSE3-NEXT:    movd %xmm1, 56(%rdx)
; SSE3-NEXT:    movq %xmm0, 48(%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec384_v3f32:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSSE3-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSSE3-ONLY-NEXT:    movq %xmm0, (%rsi)
; SSSE3-ONLY-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSSE3-ONLY-NEXT:    movd %xmm1, 8(%rsi)
; SSSE3-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, (%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 16(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 32(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 48(%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec384_v3f32:
; SSE41:       # %bb.0:
; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE41-NEXT:    pxor (%rdi), %xmm0
; SSE41-NEXT:    pextrd $2, %xmm0, 8(%rsi)
; SSE41-NEXT:    movq %xmm0, (%rsi)
; SSE41-NEXT:    pextrd $2, %xmm0, 8(%rdx)
; SSE41-NEXT:    movq %xmm0, (%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 24(%rdx)
; SSE41-NEXT:    movq %xmm0, 16(%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 40(%rdx)
; SSE41-NEXT:    movq %xmm0, 32(%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 56(%rdx)
; SSE41-NEXT:    movq %xmm0, 48(%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec384_v3f32:
; SSE42:       # %bb.0:
; SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE42-NEXT:    pxor (%rdi), %xmm0
; SSE42-NEXT:    pextrd $2, %xmm0, 8(%rsi)
; SSE42-NEXT:    movq %xmm0, (%rsi)
; SSE42-NEXT:    pextrd $2, %xmm0, 8(%rdx)
; SSE42-NEXT:    movq %xmm0, (%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 24(%rdx)
; SSE42-NEXT:    movq %xmm0, 16(%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 40(%rdx)
; SSE42-NEXT:    movq %xmm0, 32(%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 56(%rdx)
; SSE42-NEXT:    movq %xmm0, 48(%rdx)
; SSE42-NEXT:    retq
;
; AVX-LABEL: vec384_v3f32:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vpextrd $2, %xmm0, 8(%rsi)
; AVX-NEXT:    vmovq %xmm0, (%rsi)
; AVX-NEXT:    vpextrd $2, %xmm0, 8(%rdx)
; AVX-NEXT:    vmovq %xmm0, (%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 24(%rdx)
; AVX-NEXT:    vmovq %xmm0, 16(%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 40(%rdx)
; AVX-NEXT:    vmovq %xmm0, 32(%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 56(%rdx)
; AVX-NEXT:    vmovq %xmm0, 48(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <3 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <3 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1>
  %in.subvec = bitcast <3 x i32> %in.subvec.int to <3 x float>
  store <3 x float> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <3 x float>, ptr %out.vec.ptr, i64 0
  store <3 x float> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <3 x float>, ptr %out.vec.ptr, i64 1
  store <3 x float> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <3 x float>, ptr %out.vec.ptr, i64 2
  store <3 x float> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <3 x float>, ptr %out.vec.ptr, i64 3
  store <3 x float> %in.subvec, ptr %out.subvec3.ptr, align 4
  ret void
}

define void @vec384_v3i64(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v3i64:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    movq 16(%rdi), %rdi
; SCALAR-NEXT:    notq %rdi
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rdi, 16(%rsi)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rdi, 16(%rdx)
; SCALAR-NEXT:    movq %rdi, 48(%rdx)
; SCALAR-NEXT:    movq %rcx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v3i64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq 16(%rdi), %rax
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    notq %rax
; SSE2-NEXT:    movq %rax, 16(%rsi)
; SSE2-NEXT:    movq %rax, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movq %rax, 48(%rdx)
; SSE2-NEXT:    movdqu %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v3i64:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT:    vmovlps %xmm1, 16(%rsi)
; AVX1-NEXT:    vmovaps %xmm0, (%rsi)
; AVX1-NEXT:    vmovlps %xmm1, 16(%rdx)
; AVX1-NEXT:    vmovaps %xmm0, (%rdx)
; AVX1-NEXT:    vmovlps %xmm1, 48(%rdx)
; AVX1-NEXT:    vmovups %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec384_v3i64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vmovq %xmm1, 16(%rsi)
; AVX2-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX2-NEXT:    vmovq %xmm1, 16(%rdx)
; AVX2-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-NEXT:    vmovq %xmm1, 48(%rdx)
; AVX2-NEXT:    vmovdqu %xmm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <3 x i64>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <3 x i64> %in.subvec.not, <i64 -1, i64 -1, i64 -1>
  store <3 x i64> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <3 x i64>, ptr %out.vec.ptr, i64 0
  store <3 x i64> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <3 x i64>, ptr %out.vec.ptr, i64 1
  store <3 x i64> %in.subvec, ptr %out.subvec1.ptr, align 8
  ret void
}

define void @vec384_v3f64(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v3f64:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    movq 16(%rdi), %rdi
; SCALAR-NEXT:    notq %rdi
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rdi, 16(%rsi)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rdi, 16(%rdx)
; SCALAR-NEXT:    movq %rdi, 48(%rdx)
; SCALAR-NEXT:    movq %rcx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v3f64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq 16(%rdi), %rax
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    notq %rax
; SSE2-NEXT:    movq %rax, 16(%rsi)
; SSE2-NEXT:    movq %rax, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movq %rax, 48(%rdx)
; SSE2-NEXT:    movdqu %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v3f64:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT:    vmovlps %xmm1, 16(%rsi)
; AVX1-NEXT:    vmovaps %xmm0, (%rsi)
; AVX1-NEXT:    vmovlps %xmm1, 16(%rdx)
; AVX1-NEXT:    vmovaps %xmm0, (%rdx)
; AVX1-NEXT:    vmovlps %xmm1, 48(%rdx)
; AVX1-NEXT:    vmovups %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec384_v3f64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vmovq %xmm1, 16(%rsi)
; AVX2-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX2-NEXT:    vmovq %xmm1, 16(%rdx)
; AVX2-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-NEXT:    vmovq %xmm1, 48(%rdx)
; AVX2-NEXT:    vmovdqu %xmm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <3 x i64>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <3 x i64> %in.subvec.not, <i64 -1, i64 -1, i64 -1>
  %in.subvec = bitcast <3 x i64> %in.subvec.int to <3 x double>
  store <3 x double> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <3 x double>, ptr %out.vec.ptr, i64 0
  store <3 x double> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <3 x double>, ptr %out.vec.ptr, i64 1
  store <3 x double> %in.subvec, ptr %out.subvec1.ptr, align 8
  ret void
}

define void @vec384_v4i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v4i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl 3(%rdi), %r8d
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    movb %r8b, 3(%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %dil, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %r8b, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movb %dil, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %r8b, 7(%rdx)
; SCALAR-NEXT:    movb %cl, 6(%rdx)
; SCALAR-NEXT:    movb %dil, 5(%rdx)
; SCALAR-NEXT:    movb %al, 4(%rdx)
; SCALAR-NEXT:    movb %r8b, 11(%rdx)
; SCALAR-NEXT:    movb %cl, 10(%rdx)
; SCALAR-NEXT:    movb %dil, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    movb %r8b, 15(%rdx)
; SCALAR-NEXT:    movb %cl, 14(%rdx)
; SCALAR-NEXT:    movb %dil, 13(%rdx)
; SCALAR-NEXT:    movb %al, 12(%rdx)
; SCALAR-NEXT:    movb %r8b, 19(%rdx)
; SCALAR-NEXT:    movb %cl, 18(%rdx)
; SCALAR-NEXT:    movb %dil, 17(%rdx)
; SCALAR-NEXT:    movb %al, 16(%rdx)
; SCALAR-NEXT:    movb %r8b, 23(%rdx)
; SCALAR-NEXT:    movb %cl, 22(%rdx)
; SCALAR-NEXT:    movb %dil, 21(%rdx)
; SCALAR-NEXT:    movb %al, 20(%rdx)
; SCALAR-NEXT:    movb %r8b, 27(%rdx)
; SCALAR-NEXT:    movb %cl, 26(%rdx)
; SCALAR-NEXT:    movb %dil, 25(%rdx)
; SCALAR-NEXT:    movb %al, 24(%rdx)
; SCALAR-NEXT:    movb %r8b, 31(%rdx)
; SCALAR-NEXT:    movb %cl, 30(%rdx)
; SCALAR-NEXT:    movb %dil, 29(%rdx)
; SCALAR-NEXT:    movb %al, 28(%rdx)
; SCALAR-NEXT:    movb %r8b, 35(%rdx)
; SCALAR-NEXT:    movb %cl, 34(%rdx)
; SCALAR-NEXT:    movb %dil, 33(%rdx)
; SCALAR-NEXT:    movb %al, 32(%rdx)
; SCALAR-NEXT:    movb %r8b, 39(%rdx)
; SCALAR-NEXT:    movb %cl, 38(%rdx)
; SCALAR-NEXT:    movb %dil, 37(%rdx)
; SCALAR-NEXT:    movb %al, 36(%rdx)
; SCALAR-NEXT:    movb %r8b, 43(%rdx)
; SCALAR-NEXT:    movb %cl, 42(%rdx)
; SCALAR-NEXT:    movb %dil, 41(%rdx)
; SCALAR-NEXT:    movb %al, 40(%rdx)
; SCALAR-NEXT:    movb %r8b, 47(%rdx)
; SCALAR-NEXT:    movb %cl, 46(%rdx)
; SCALAR-NEXT:    movb %dil, 45(%rdx)
; SCALAR-NEXT:    movb %al, 44(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v4i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movd %xmm0, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v4i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vmovd %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec384_v4i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-NEXT:    vmovd %xmm0, (%rsi)
; AVX2-NEXT:    vpbroadcastd %xmm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <4 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1>
  store <4 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 0
  store <4 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 1
  store <4 x i8> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 2
  store <4 x i8> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 3
  store <4 x i8> %in.subvec, ptr %out.subvec3.ptr, align 4
  %out.subvec4.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 4
  store <4 x i8> %in.subvec, ptr %out.subvec4.ptr, align 16
  %out.subvec5.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 5
  store <4 x i8> %in.subvec, ptr %out.subvec5.ptr, align 4
  %out.subvec6.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 6
  store <4 x i8> %in.subvec, ptr %out.subvec6.ptr, align 8
  %out.subvec7.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 7
  store <4 x i8> %in.subvec, ptr %out.subvec7.ptr, align 4
  %out.subvec8.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 8
  store <4 x i8> %in.subvec, ptr %out.subvec8.ptr, align 32
  %out.subvec9.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 9
  store <4 x i8> %in.subvec, ptr %out.subvec9.ptr, align 4
  %out.subvec10.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 10
  store <4 x i8> %in.subvec, ptr %out.subvec10.ptr, align 8
  %out.subvec11.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 11
  store <4 x i8> %in.subvec, ptr %out.subvec11.ptr, align 4
  ret void
}

define void @vec384_v4i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v4i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzwl 6(%rdi), %r8d
; SCALAR-NEXT:    movzwl 2(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %edi
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    movw %r8w, 6(%rsi)
; SCALAR-NEXT:    movw %di, 4(%rsi)
; SCALAR-NEXT:    movw %cx, 2(%rsi)
; SCALAR-NEXT:    movw %ax, (%rsi)
; SCALAR-NEXT:    movw %r8w, 6(%rdx)
; SCALAR-NEXT:    movw %di, 4(%rdx)
; SCALAR-NEXT:    movw %cx, 2(%rdx)
; SCALAR-NEXT:    movw %ax, (%rdx)
; SCALAR-NEXT:    movw %r8w, 14(%rdx)
; SCALAR-NEXT:    movw %di, 12(%rdx)
; SCALAR-NEXT:    movw %cx, 10(%rdx)
; SCALAR-NEXT:    movw %ax, 8(%rdx)
; SCALAR-NEXT:    movw %r8w, 22(%rdx)
; SCALAR-NEXT:    movw %di, 20(%rdx)
; SCALAR-NEXT:    movw %cx, 18(%rdx)
; SCALAR-NEXT:    movw %ax, 16(%rdx)
; SCALAR-NEXT:    movw %r8w, 30(%rdx)
; SCALAR-NEXT:    movw %di, 28(%rdx)
; SCALAR-NEXT:    movw %cx, 26(%rdx)
; SCALAR-NEXT:    movw %ax, 24(%rdx)
; SCALAR-NEXT:    movw %r8w, 38(%rdx)
; SCALAR-NEXT:    movw %di, 36(%rdx)
; SCALAR-NEXT:    movw %cx, 34(%rdx)
; SCALAR-NEXT:    movw %ax, 32(%rdx)
; SCALAR-NEXT:    movw %r8w, 46(%rdx)
; SCALAR-NEXT:    movw %di, 44(%rdx)
; SCALAR-NEXT:    movw %cx, 42(%rdx)
; SCALAR-NEXT:    movw %ax, 40(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v4i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v4i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm1
; AVX1-NEXT:    vmovaps %ymm1, (%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec384_v4i16:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec384_v4i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX512-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX512-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <4 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1, i16 -1>
  store <4 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 0
  store <4 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 1
  store <4 x i16> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 2
  store <4 x i16> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 3
  store <4 x i16> %in.subvec, ptr %out.subvec3.ptr, align 8
  %out.subvec4.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 4
  store <4 x i16> %in.subvec, ptr %out.subvec4.ptr, align 32
  %out.subvec5.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 5
  store <4 x i16> %in.subvec, ptr %out.subvec5.ptr, align 8
  ret void
}

define void @vec384_v4i32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v4i32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movaps (%rdi), %xmm0
; SCALAR-NEXT:    xorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SCALAR-NEXT:    movaps %xmm0, (%rsi)
; SCALAR-NEXT:    movaps %xmm0, (%rdx)
; SCALAR-NEXT:    movaps %xmm0, 16(%rdx)
; SCALAR-NEXT:    movaps %xmm0, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v4i32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec384_v4i32:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <4 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1, i32 -1>
  store <4 x i32> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i32>, ptr %out.vec.ptr, i64 0
  store <4 x i32> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i32>, ptr %out.vec.ptr, i64 1
  store <4 x i32> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <4 x i32>, ptr %out.vec.ptr, i64 2
  store <4 x i32> %in.subvec, ptr %out.subvec2.ptr, align 32
  ret void
}

define void @vec384_v4f32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v4f32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movaps (%rdi), %xmm0
; SCALAR-NEXT:    xorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SCALAR-NEXT:    movaps %xmm0, (%rsi)
; SCALAR-NEXT:    movaps %xmm0, (%rdx)
; SCALAR-NEXT:    movaps %xmm0, 16(%rdx)
; SCALAR-NEXT:    movaps %xmm0, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v4f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec384_v4f32:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <4 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <4 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1, i32 -1>
  %in.subvec = bitcast <4 x i32> %in.subvec.int to <4 x float>
  store <4 x float> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x float>, ptr %out.vec.ptr, i64 0
  store <4 x float> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x float>, ptr %out.vec.ptr, i64 1
  store <4 x float> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <4 x float>, ptr %out.vec.ptr, i64 2
  store <4 x float> %in.subvec, ptr %out.subvec2.ptr, align 32
  ret void
}

define void @vec384_v6i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v6i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq %rax, %rcx
; SCALAR-NEXT:    shrq $32, %rcx
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movw %cx, 4(%rsi)
; SCALAR-NEXT:    movw %cx, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movw %cx, 12(%rdx)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    movw %cx, 20(%rdx)
; SCALAR-NEXT:    movl %eax, 16(%rdx)
; SCALAR-NEXT:    movw %cx, 28(%rdx)
; SCALAR-NEXT:    movl %eax, 24(%rdx)
; SCALAR-NEXT:    movw %cx, 36(%rdx)
; SCALAR-NEXT:    movl %eax, 32(%rdx)
; SCALAR-NEXT:    movw %cx, 44(%rdx)
; SCALAR-NEXT:    movl %eax, 40(%rdx)
; SCALAR-NEXT:    movw %cx, 52(%rdx)
; SCALAR-NEXT:    movl %eax, 48(%rdx)
; SCALAR-NEXT:    movw %cx, 60(%rdx)
; SCALAR-NEXT:    movl %eax, 56(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec384_v6i8:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-ONLY-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-ONLY-NEXT:    pxor %xmm0, %xmm1
; SSE2-ONLY-NEXT:    movd %xmm1, (%rsi)
; SSE2-ONLY-NEXT:    pextrw $2, %xmm1, %eax
; SSE2-ONLY-NEXT:    movw %ax, 4(%rsi)
; SSE2-ONLY-NEXT:    movw %ax, 4(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, (%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 12(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 20(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 16(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 28(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 36(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 32(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 44(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 52(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 48(%rdx)
; SSE2-ONLY-NEXT:    movw %ax, 60(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec384_v6i8:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE3-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE3-NEXT:    pxor %xmm0, %xmm1
; SSE3-NEXT:    movd %xmm1, (%rsi)
; SSE3-NEXT:    pextrw $2, %xmm1, %eax
; SSE3-NEXT:    movw %ax, 4(%rsi)
; SSE3-NEXT:    movw %ax, 4(%rdx)
; SSE3-NEXT:    movd %xmm1, (%rdx)
; SSE3-NEXT:    movw %ax, 12(%rdx)
; SSE3-NEXT:    movd %xmm1, 8(%rdx)
; SSE3-NEXT:    movw %ax, 20(%rdx)
; SSE3-NEXT:    movd %xmm1, 16(%rdx)
; SSE3-NEXT:    movw %ax, 28(%rdx)
; SSE3-NEXT:    movd %xmm1, 24(%rdx)
; SSE3-NEXT:    movw %ax, 36(%rdx)
; SSE3-NEXT:    movd %xmm1, 32(%rdx)
; SSE3-NEXT:    movw %ax, 44(%rdx)
; SSE3-NEXT:    movd %xmm1, 40(%rdx)
; SSE3-NEXT:    movw %ax, 52(%rdx)
; SSE3-NEXT:    movd %xmm1, 48(%rdx)
; SSE3-NEXT:    movw %ax, 60(%rdx)
; SSE3-NEXT:    movd %xmm1, 56(%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec384_v6i8:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSSE3-ONLY-NEXT:    pcmpeqd %xmm1, %xmm1
; SSSE3-ONLY-NEXT:    pxor %xmm0, %xmm1
; SSSE3-ONLY-NEXT:    movd %xmm1, (%rsi)
; SSSE3-ONLY-NEXT:    pextrw $2, %xmm1, %eax
; SSSE3-ONLY-NEXT:    movw %ax, 4(%rsi)
; SSSE3-ONLY-NEXT:    movw %ax, 4(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, (%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 12(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 20(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 16(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 28(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 36(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 32(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 44(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 52(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 48(%rdx)
; SSSE3-ONLY-NEXT:    movw %ax, 60(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec384_v6i8:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE41-NEXT:    pxor %xmm0, %xmm1
; SSE41-NEXT:    pextrw $2, %xmm1, 4(%rsi)
; SSE41-NEXT:    movd %xmm1, (%rsi)
; SSE41-NEXT:    pextrw $2, %xmm1, 4(%rdx)
; SSE41-NEXT:    movd %xmm1, (%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 12(%rdx)
; SSE41-NEXT:    movd %xmm1, 8(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 20(%rdx)
; SSE41-NEXT:    movd %xmm1, 16(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 28(%rdx)
; SSE41-NEXT:    movd %xmm1, 24(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 36(%rdx)
; SSE41-NEXT:    movd %xmm1, 32(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 44(%rdx)
; SSE41-NEXT:    movd %xmm1, 40(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 52(%rdx)
; SSE41-NEXT:    movd %xmm1, 48(%rdx)
; SSE41-NEXT:    pextrw $2, %xmm1, 60(%rdx)
; SSE41-NEXT:    movd %xmm1, 56(%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec384_v6i8:
; SSE42:       # %bb.0:
; SSE42-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE42-NEXT:    pxor %xmm0, %xmm1
; SSE42-NEXT:    pextrw $2, %xmm1, 4(%rsi)
; SSE42-NEXT:    movd %xmm1, (%rsi)
; SSE42-NEXT:    pextrw $2, %xmm1, 4(%rdx)
; SSE42-NEXT:    movd %xmm1, (%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 12(%rdx)
; SSE42-NEXT:    movd %xmm1, 8(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 20(%rdx)
; SSE42-NEXT:    movd %xmm1, 16(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 28(%rdx)
; SSE42-NEXT:    movd %xmm1, 24(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 36(%rdx)
; SSE42-NEXT:    movd %xmm1, 32(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 44(%rdx)
; SSE42-NEXT:    movd %xmm1, 40(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 52(%rdx)
; SSE42-NEXT:    movd %xmm1, 48(%rdx)
; SSE42-NEXT:    pextrw $2, %xmm1, 60(%rdx)
; SSE42-NEXT:    movd %xmm1, 56(%rdx)
; SSE42-NEXT:    retq
;
; AVX1-LABEL: vec384_v6i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vpextrw $2, %xmm0, 4(%rsi)
; AVX1-NEXT:    vmovd %xmm0, (%rsi)
; AVX1-NEXT:    vpextrw $2, %xmm0, 4(%rdx)
; AVX1-NEXT:    vmovd %xmm0, (%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 12(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 8(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 20(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 16(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 28(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 24(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 36(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 32(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 44(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 40(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 52(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 48(%rdx)
; AVX1-NEXT:    vpextrw $2, %xmm0, 60(%rdx)
; AVX1-NEXT:    vmovd %xmm0, 56(%rdx)
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec384_v6i8:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 4(%rsi)
; AVX2-ONLY-NEXT:    vmovd %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 4(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, (%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 12(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 8(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 20(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 16(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 28(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 24(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 36(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 44(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 40(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 52(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 48(%rdx)
; AVX2-ONLY-NEXT:    vpextrw $2, %xmm0, 60(%rdx)
; AVX2-ONLY-NEXT:    vmovd %xmm0, 56(%rdx)
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec384_v6i8:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vpextrw $2, %xmm0, 4(%rsi)
; AVX512-NEXT:    vmovd %xmm0, (%rsi)
; AVX512-NEXT:    vpextrw $2, %xmm0, 4(%rdx)
; AVX512-NEXT:    vmovd %xmm0, (%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 12(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 8(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 20(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 16(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 28(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 24(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 36(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 32(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 44(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 40(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 52(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 48(%rdx)
; AVX512-NEXT:    vpextrw $2, %xmm0, 60(%rdx)
; AVX512-NEXT:    vmovd %xmm0, 56(%rdx)
; AVX512-NEXT:    retq
  %in.subvec.not = load <6 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <6 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <6 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <6 x i8>, ptr %out.vec.ptr, i64 0
  store <6 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <6 x i8>, ptr %out.vec.ptr, i64 1
  store <6 x i8> %in.subvec, ptr %out.subvec1.ptr, align 2
  %out.subvec2.ptr = getelementptr <6 x i8>, ptr %out.vec.ptr, i64 2
  store <6 x i8> %in.subvec, ptr %out.subvec2.ptr, align 4
  %out.subvec3.ptr = getelementptr <6 x i8>, ptr %out.vec.ptr, i64 3
  store <6 x i8> %in.subvec, ptr %out.subvec3.ptr, align 2
  %out.subvec4.ptr = getelementptr <6 x i8>, ptr %out.vec.ptr, i64 4
  store <6 x i8> %in.subvec, ptr %out.subvec4.ptr, align 8
  %out.subvec5.ptr = getelementptr <6 x i8>, ptr %out.vec.ptr, i64 5
  store <6 x i8> %in.subvec, ptr %out.subvec5.ptr, align 2
  %out.subvec6.ptr = getelementptr <6 x i8>, ptr %out.vec.ptr, i64 6
  store <6 x i8> %in.subvec, ptr %out.subvec6.ptr, align 4
  %out.subvec7.ptr = getelementptr <6 x i8>, ptr %out.vec.ptr, i64 7
  store <6 x i8> %in.subvec, ptr %out.subvec7.ptr, align 2
  ret void
}

define void @vec384_v6i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v6i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movl 8(%rdi), %ecx
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movl %ecx, 8(%rsi)
; SCALAR-NEXT:    movl %ecx, 8(%rdx)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movl %ecx, 24(%rdx)
; SCALAR-NEXT:    movq %rax, 16(%rdx)
; SCALAR-NEXT:    movl %ecx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    movl %ecx, 56(%rdx)
; SCALAR-NEXT:    movq %rax, 48(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec384_v6i16:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSE2-ONLY-NEXT:    movq %xmm0, (%rsi)
; SSE2-ONLY-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE2-ONLY-NEXT:    movd %xmm1, 8(%rsi)
; SSE2-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, (%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 16(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 32(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 48(%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec384_v6i16:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE3-NEXT:    pxor (%rdi), %xmm0
; SSE3-NEXT:    movq %xmm0, (%rsi)
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-NEXT:    movd %xmm1, 8(%rsi)
; SSE3-NEXT:    movd %xmm1, 8(%rdx)
; SSE3-NEXT:    movq %xmm0, (%rdx)
; SSE3-NEXT:    movd %xmm1, 24(%rdx)
; SSE3-NEXT:    movq %xmm0, 16(%rdx)
; SSE3-NEXT:    movd %xmm1, 40(%rdx)
; SSE3-NEXT:    movq %xmm0, 32(%rdx)
; SSE3-NEXT:    movd %xmm1, 56(%rdx)
; SSE3-NEXT:    movq %xmm0, 48(%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec384_v6i16:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSSE3-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSSE3-ONLY-NEXT:    movq %xmm0, (%rsi)
; SSSE3-ONLY-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSSE3-ONLY-NEXT:    movd %xmm1, 8(%rsi)
; SSSE3-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, (%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 16(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 32(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 48(%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec384_v6i16:
; SSE41:       # %bb.0:
; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE41-NEXT:    pxor (%rdi), %xmm0
; SSE41-NEXT:    pextrd $2, %xmm0, 8(%rsi)
; SSE41-NEXT:    movq %xmm0, (%rsi)
; SSE41-NEXT:    pextrd $2, %xmm0, 8(%rdx)
; SSE41-NEXT:    movq %xmm0, (%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 24(%rdx)
; SSE41-NEXT:    movq %xmm0, 16(%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 40(%rdx)
; SSE41-NEXT:    movq %xmm0, 32(%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 56(%rdx)
; SSE41-NEXT:    movq %xmm0, 48(%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec384_v6i16:
; SSE42:       # %bb.0:
; SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE42-NEXT:    pxor (%rdi), %xmm0
; SSE42-NEXT:    pextrd $2, %xmm0, 8(%rsi)
; SSE42-NEXT:    movq %xmm0, (%rsi)
; SSE42-NEXT:    pextrd $2, %xmm0, 8(%rdx)
; SSE42-NEXT:    movq %xmm0, (%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 24(%rdx)
; SSE42-NEXT:    movq %xmm0, 16(%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 40(%rdx)
; SSE42-NEXT:    movq %xmm0, 32(%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 56(%rdx)
; SSE42-NEXT:    movq %xmm0, 48(%rdx)
; SSE42-NEXT:    retq
;
; AVX-LABEL: vec384_v6i16:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vpextrd $2, %xmm0, 8(%rsi)
; AVX-NEXT:    vmovq %xmm0, (%rsi)
; AVX-NEXT:    vpextrd $2, %xmm0, 8(%rdx)
; AVX-NEXT:    vmovq %xmm0, (%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 24(%rdx)
; AVX-NEXT:    vmovq %xmm0, 16(%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 40(%rdx)
; AVX-NEXT:    vmovq %xmm0, 32(%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 56(%rdx)
; AVX-NEXT:    vmovq %xmm0, 48(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <6 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <6 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
  store <6 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <6 x i16>, ptr %out.vec.ptr, i64 0
  store <6 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <6 x i16>, ptr %out.vec.ptr, i64 1
  store <6 x i16> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <6 x i16>, ptr %out.vec.ptr, i64 2
  store <6 x i16> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <6 x i16>, ptr %out.vec.ptr, i64 3
  store <6 x i16> %in.subvec, ptr %out.subvec3.ptr, align 4
  ret void
}

define void @vec384_v6i32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v6i32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    movq 16(%rdi), %rdi
; SCALAR-NEXT:    notq %rdi
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rdi, 16(%rsi)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rdi, 16(%rdx)
; SCALAR-NEXT:    movq %rdi, 48(%rdx)
; SCALAR-NEXT:    movq %rcx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v6i32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    movdqa 16(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movq %xmm1, 16(%rsi)
; SSE2-NEXT:    movq %xmm1, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movq %xmm1, 48(%rdx)
; SSE2-NEXT:    movdqu %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v6i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT:    vmovlps %xmm1, 16(%rsi)
; AVX1-NEXT:    vmovaps %xmm0, (%rsi)
; AVX1-NEXT:    vmovlps %xmm1, 16(%rdx)
; AVX1-NEXT:    vmovaps %xmm0, (%rdx)
; AVX1-NEXT:    vmovlps %xmm1, 48(%rdx)
; AVX1-NEXT:    vmovups %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec384_v6i32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vmovq %xmm1, 16(%rsi)
; AVX2-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX2-NEXT:    vmovq %xmm1, 16(%rdx)
; AVX2-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-NEXT:    vmovq %xmm1, 48(%rdx)
; AVX2-NEXT:    vmovdqu %xmm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <6 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <6 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
  store <6 x i32> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <6 x i32>, ptr %out.vec.ptr, i64 0
  store <6 x i32> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <6 x i32>, ptr %out.vec.ptr, i64 1
  store <6 x i32> %in.subvec, ptr %out.subvec1.ptr, align 8
  ret void
}

define void @vec384_v6f32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v6f32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    movq 16(%rdi), %rdi
; SCALAR-NEXT:    notq %rdi
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rdi, 16(%rsi)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rdi, 16(%rdx)
; SCALAR-NEXT:    movq %rdi, 48(%rdx)
; SCALAR-NEXT:    movq %rcx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v6f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    movdqa 16(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movq %xmm1, 16(%rsi)
; SSE2-NEXT:    movq %xmm1, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movq %xmm1, 48(%rdx)
; SSE2-NEXT:    movdqu %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v6f32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT:    vmovlps %xmm1, 16(%rsi)
; AVX1-NEXT:    vmovaps %xmm0, (%rsi)
; AVX1-NEXT:    vmovlps %xmm1, 16(%rdx)
; AVX1-NEXT:    vmovaps %xmm0, (%rdx)
; AVX1-NEXT:    vmovlps %xmm1, 48(%rdx)
; AVX1-NEXT:    vmovups %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec384_v6f32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vmovq %xmm1, 16(%rsi)
; AVX2-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX2-NEXT:    vmovq %xmm1, 16(%rdx)
; AVX2-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-NEXT:    vmovq %xmm1, 48(%rdx)
; AVX2-NEXT:    vmovdqu %xmm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <6 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <6 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
  %in.subvec = bitcast <6 x i32> %in.subvec.int to <6 x float>
  store <6 x float> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <6 x float>, ptr %out.vec.ptr, i64 0
  store <6 x float> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <6 x float>, ptr %out.vec.ptr, i64 1
  store <6 x float> %in.subvec, ptr %out.subvec1.ptr, align 8
  ret void
}

define void @vec384_v8i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v8i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzbl 7(%rdi), %ebx
; SCALAR-NEXT:    movzbl 6(%rdi), %r11d
; SCALAR-NEXT:    movzbl 5(%rdi), %r10d
; SCALAR-NEXT:    movzbl 4(%rdi), %r9d
; SCALAR-NEXT:    movzbl 3(%rdi), %r8d
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    notb %r10b
; SCALAR-NEXT:    notb %r11b
; SCALAR-NEXT:    notb %bl
; SCALAR-NEXT:    movb %bl, 7(%rsi)
; SCALAR-NEXT:    movb %r11b, 6(%rsi)
; SCALAR-NEXT:    movb %r10b, 5(%rsi)
; SCALAR-NEXT:    movb %r9b, 4(%rsi)
; SCALAR-NEXT:    movb %r8b, 3(%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %dil, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %bl, 7(%rdx)
; SCALAR-NEXT:    movb %r11b, 6(%rdx)
; SCALAR-NEXT:    movb %r10b, 5(%rdx)
; SCALAR-NEXT:    movb %r9b, 4(%rdx)
; SCALAR-NEXT:    movb %r8b, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movb %dil, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %bl, 15(%rdx)
; SCALAR-NEXT:    movb %r11b, 14(%rdx)
; SCALAR-NEXT:    movb %r10b, 13(%rdx)
; SCALAR-NEXT:    movb %r9b, 12(%rdx)
; SCALAR-NEXT:    movb %r8b, 11(%rdx)
; SCALAR-NEXT:    movb %cl, 10(%rdx)
; SCALAR-NEXT:    movb %dil, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    movb %bl, 23(%rdx)
; SCALAR-NEXT:    movb %r11b, 22(%rdx)
; SCALAR-NEXT:    movb %r10b, 21(%rdx)
; SCALAR-NEXT:    movb %r9b, 20(%rdx)
; SCALAR-NEXT:    movb %r8b, 19(%rdx)
; SCALAR-NEXT:    movb %cl, 18(%rdx)
; SCALAR-NEXT:    movb %dil, 17(%rdx)
; SCALAR-NEXT:    movb %al, 16(%rdx)
; SCALAR-NEXT:    movb %bl, 31(%rdx)
; SCALAR-NEXT:    movb %r11b, 30(%rdx)
; SCALAR-NEXT:    movb %r10b, 29(%rdx)
; SCALAR-NEXT:    movb %r9b, 28(%rdx)
; SCALAR-NEXT:    movb %r8b, 27(%rdx)
; SCALAR-NEXT:    movb %cl, 26(%rdx)
; SCALAR-NEXT:    movb %dil, 25(%rdx)
; SCALAR-NEXT:    movb %al, 24(%rdx)
; SCALAR-NEXT:    movb %bl, 39(%rdx)
; SCALAR-NEXT:    movb %r11b, 38(%rdx)
; SCALAR-NEXT:    movb %r10b, 37(%rdx)
; SCALAR-NEXT:    movb %r9b, 36(%rdx)
; SCALAR-NEXT:    movb %r8b, 35(%rdx)
; SCALAR-NEXT:    movb %cl, 34(%rdx)
; SCALAR-NEXT:    movb %dil, 33(%rdx)
; SCALAR-NEXT:    movb %al, 32(%rdx)
; SCALAR-NEXT:    movb %bl, 47(%rdx)
; SCALAR-NEXT:    movb %r11b, 46(%rdx)
; SCALAR-NEXT:    movb %r10b, 45(%rdx)
; SCALAR-NEXT:    movb %r9b, 44(%rdx)
; SCALAR-NEXT:    movb %r8b, 43(%rdx)
; SCALAR-NEXT:    movb %cl, 42(%rdx)
; SCALAR-NEXT:    movb %dil, 41(%rdx)
; SCALAR-NEXT:    movb %al, 40(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v8i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v8i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm1
; AVX1-NEXT:    vmovaps %ymm1, (%rdx)
; AVX1-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec384_v8i8:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec384_v8i8:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX512-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX512-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <8 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <8 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <8 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 0
  store <8 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 1
  store <8 x i8> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 2
  store <8 x i8> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 3
  store <8 x i8> %in.subvec, ptr %out.subvec3.ptr, align 8
  %out.subvec4.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 4
  store <8 x i8> %in.subvec, ptr %out.subvec4.ptr, align 32
  %out.subvec5.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 5
  store <8 x i8> %in.subvec, ptr %out.subvec5.ptr, align 8
  ret void
}

define void @vec384_v8i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v8i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzwl 14(%rdi), %ebx
; SCALAR-NEXT:    movl 12(%rdi), %r11d
; SCALAR-NEXT:    movzwl 10(%rdi), %r10d
; SCALAR-NEXT:    movl 8(%rdi), %r9d
; SCALAR-NEXT:    movzwl 6(%rdi), %r8d
; SCALAR-NEXT:    movzwl 2(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %edi
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    notl %r9d
; SCALAR-NEXT:    notl %r10d
; SCALAR-NEXT:    notl %r11d
; SCALAR-NEXT:    notl %ebx
; SCALAR-NEXT:    movw %bx, 14(%rsi)
; SCALAR-NEXT:    movw %r11w, 12(%rsi)
; SCALAR-NEXT:    movw %r10w, 10(%rsi)
; SCALAR-NEXT:    movw %r9w, 8(%rsi)
; SCALAR-NEXT:    movw %r8w, 6(%rsi)
; SCALAR-NEXT:    movw %di, 4(%rsi)
; SCALAR-NEXT:    movw %cx, 2(%rsi)
; SCALAR-NEXT:    movw %ax, (%rsi)
; SCALAR-NEXT:    movw %bx, 14(%rdx)
; SCALAR-NEXT:    movw %r11w, 12(%rdx)
; SCALAR-NEXT:    movw %r10w, 10(%rdx)
; SCALAR-NEXT:    movw %r9w, 8(%rdx)
; SCALAR-NEXT:    movw %r8w, 6(%rdx)
; SCALAR-NEXT:    movw %di, 4(%rdx)
; SCALAR-NEXT:    movw %cx, 2(%rdx)
; SCALAR-NEXT:    movw %ax, (%rdx)
; SCALAR-NEXT:    movw %bx, 30(%rdx)
; SCALAR-NEXT:    movw %r11w, 28(%rdx)
; SCALAR-NEXT:    movw %r10w, 26(%rdx)
; SCALAR-NEXT:    movw %r9w, 24(%rdx)
; SCALAR-NEXT:    movw %r8w, 22(%rdx)
; SCALAR-NEXT:    movw %di, 20(%rdx)
; SCALAR-NEXT:    movw %cx, 18(%rdx)
; SCALAR-NEXT:    movw %ax, 16(%rdx)
; SCALAR-NEXT:    movw %bx, 46(%rdx)
; SCALAR-NEXT:    movw %r11w, 44(%rdx)
; SCALAR-NEXT:    movw %r10w, 42(%rdx)
; SCALAR-NEXT:    movw %r9w, 40(%rdx)
; SCALAR-NEXT:    movw %r8w, 38(%rdx)
; SCALAR-NEXT:    movw %di, 36(%rdx)
; SCALAR-NEXT:    movw %cx, 34(%rdx)
; SCALAR-NEXT:    movw %ax, 32(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v8i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec384_v8i16:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <8 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <8 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
  store <8 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <8 x i16>, ptr %out.vec.ptr, i64 0
  store <8 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <8 x i16>, ptr %out.vec.ptr, i64 1
  store <8 x i16> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <8 x i16>, ptr %out.vec.ptr, i64 2
  store <8 x i16> %in.subvec, ptr %out.subvec2.ptr, align 32
  ret void
}

define void @vec384_v12i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v12i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movl 8(%rdi), %ecx
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movl %ecx, 8(%rsi)
; SCALAR-NEXT:    movl %ecx, 8(%rdx)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movl %ecx, 24(%rdx)
; SCALAR-NEXT:    movq %rax, 16(%rdx)
; SCALAR-NEXT:    movl %ecx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    movl %ecx, 56(%rdx)
; SCALAR-NEXT:    movq %rax, 48(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec384_v12i8:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSE2-ONLY-NEXT:    movq %xmm0, (%rsi)
; SSE2-ONLY-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE2-ONLY-NEXT:    movd %xmm1, 8(%rsi)
; SSE2-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, (%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 16(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 32(%rdx)
; SSE2-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSE2-ONLY-NEXT:    movq %xmm0, 48(%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec384_v12i8:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE3-NEXT:    pxor (%rdi), %xmm0
; SSE3-NEXT:    movq %xmm0, (%rsi)
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-NEXT:    movd %xmm1, 8(%rsi)
; SSE3-NEXT:    movd %xmm1, 8(%rdx)
; SSE3-NEXT:    movq %xmm0, (%rdx)
; SSE3-NEXT:    movd %xmm1, 24(%rdx)
; SSE3-NEXT:    movq %xmm0, 16(%rdx)
; SSE3-NEXT:    movd %xmm1, 40(%rdx)
; SSE3-NEXT:    movq %xmm0, 32(%rdx)
; SSE3-NEXT:    movd %xmm1, 56(%rdx)
; SSE3-NEXT:    movq %xmm0, 48(%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec384_v12i8:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSSE3-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSSE3-ONLY-NEXT:    movq %xmm0, (%rsi)
; SSSE3-ONLY-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSSE3-ONLY-NEXT:    movd %xmm1, 8(%rsi)
; SSSE3-ONLY-NEXT:    movd %xmm1, 8(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, (%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 24(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 16(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 40(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 32(%rdx)
; SSSE3-ONLY-NEXT:    movd %xmm1, 56(%rdx)
; SSSE3-ONLY-NEXT:    movq %xmm0, 48(%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec384_v12i8:
; SSE41:       # %bb.0:
; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE41-NEXT:    pxor (%rdi), %xmm0
; SSE41-NEXT:    pextrd $2, %xmm0, 8(%rsi)
; SSE41-NEXT:    movq %xmm0, (%rsi)
; SSE41-NEXT:    pextrd $2, %xmm0, 8(%rdx)
; SSE41-NEXT:    movq %xmm0, (%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 24(%rdx)
; SSE41-NEXT:    movq %xmm0, 16(%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 40(%rdx)
; SSE41-NEXT:    movq %xmm0, 32(%rdx)
; SSE41-NEXT:    pextrd $2, %xmm0, 56(%rdx)
; SSE41-NEXT:    movq %xmm0, 48(%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec384_v12i8:
; SSE42:       # %bb.0:
; SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE42-NEXT:    pxor (%rdi), %xmm0
; SSE42-NEXT:    pextrd $2, %xmm0, 8(%rsi)
; SSE42-NEXT:    movq %xmm0, (%rsi)
; SSE42-NEXT:    pextrd $2, %xmm0, 8(%rdx)
; SSE42-NEXT:    movq %xmm0, (%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 24(%rdx)
; SSE42-NEXT:    movq %xmm0, 16(%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 40(%rdx)
; SSE42-NEXT:    movq %xmm0, 32(%rdx)
; SSE42-NEXT:    pextrd $2, %xmm0, 56(%rdx)
; SSE42-NEXT:    movq %xmm0, 48(%rdx)
; SSE42-NEXT:    retq
;
; AVX-LABEL: vec384_v12i8:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vpextrd $2, %xmm0, 8(%rsi)
; AVX-NEXT:    vmovq %xmm0, (%rsi)
; AVX-NEXT:    vpextrd $2, %xmm0, 8(%rdx)
; AVX-NEXT:    vmovq %xmm0, (%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 24(%rdx)
; AVX-NEXT:    vmovq %xmm0, 16(%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 40(%rdx)
; AVX-NEXT:    vmovq %xmm0, 32(%rdx)
; AVX-NEXT:    vpextrd $2, %xmm0, 56(%rdx)
; AVX-NEXT:    vmovq %xmm0, 48(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <12 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <12 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <12 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <12 x i8>, ptr %out.vec.ptr, i64 0
  store <12 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <12 x i8>, ptr %out.vec.ptr, i64 1
  store <12 x i8> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <12 x i8>, ptr %out.vec.ptr, i64 2
  store <12 x i8> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <12 x i8>, ptr %out.vec.ptr, i64 3
  store <12 x i8> %in.subvec, ptr %out.subvec3.ptr, align 4
  ret void
}

define void @vec384_v12i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v12i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    movq 16(%rdi), %rdi
; SCALAR-NEXT:    notq %rdi
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rdi, 16(%rsi)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rdi, 16(%rdx)
; SCALAR-NEXT:    movq %rdi, 48(%rdx)
; SCALAR-NEXT:    movq %rcx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v12i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    movdqa 16(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movq %xmm1, 16(%rsi)
; SSE2-NEXT:    movq %xmm1, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movq %xmm1, 48(%rdx)
; SSE2-NEXT:    movdqu %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v12i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT:    vmovlps %xmm1, 16(%rsi)
; AVX1-NEXT:    vmovaps %xmm0, (%rsi)
; AVX1-NEXT:    vmovlps %xmm1, 16(%rdx)
; AVX1-NEXT:    vmovaps %xmm0, (%rdx)
; AVX1-NEXT:    vmovlps %xmm1, 48(%rdx)
; AVX1-NEXT:    vmovups %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec384_v12i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vmovq %xmm1, 16(%rsi)
; AVX2-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX2-NEXT:    vmovq %xmm1, 16(%rdx)
; AVX2-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-NEXT:    vmovq %xmm1, 48(%rdx)
; AVX2-NEXT:    vmovdqu %xmm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <12 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <12 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
  store <12 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <12 x i16>, ptr %out.vec.ptr, i64 0
  store <12 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <12 x i16>, ptr %out.vec.ptr, i64 1
  store <12 x i16> %in.subvec, ptr %out.subvec1.ptr, align 8
  ret void
}

define void @vec384_v16i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v16i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbp
; SCALAR-NEXT:    pushq %r15
; SCALAR-NEXT:    pushq %r14
; SCALAR-NEXT:    pushq %r13
; SCALAR-NEXT:    pushq %r12
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzbl 15(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 14(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 13(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 12(%rdi), %r11d
; SCALAR-NEXT:    movzbl 11(%rdi), %r13d
; SCALAR-NEXT:    movzbl 10(%rdi), %r12d
; SCALAR-NEXT:    movzbl 9(%rdi), %ebp
; SCALAR-NEXT:    movzbl 8(%rdi), %r14d
; SCALAR-NEXT:    movzbl 7(%rdi), %ebx
; SCALAR-NEXT:    movzbl 6(%rdi), %r10d
; SCALAR-NEXT:    movzbl 5(%rdi), %r15d
; SCALAR-NEXT:    movzbl 4(%rdi), %r9d
; SCALAR-NEXT:    movzbl 3(%rdi), %r8d
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    movb %dil, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    movb %r8b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    movb %r9b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movl %r15d, %r9d
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    notb %r10b
; SCALAR-NEXT:    notb %bl
; SCALAR-NEXT:    movb %bl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r14b
; SCALAR-NEXT:    notb %bpl
; SCALAR-NEXT:    movl %ebp, %r15d
; SCALAR-NEXT:    notb %r12b
; SCALAR-NEXT:    movb %r12b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r13b
; SCALAR-NEXT:    movb %r13b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r11b
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 1-byte Folded Reload
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    movb %r8b, 15(%rsi)
; SCALAR-NEXT:    movb %cl, 14(%rsi)
; SCALAR-NEXT:    movl %edi, %eax
; SCALAR-NEXT:    movb %dil, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %dil, 13(%rsi)
; SCALAR-NEXT:    movb %r11b, 12(%rsi)
; SCALAR-NEXT:    movl %r11d, %ebp
; SCALAR-NEXT:    movb %r11b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %r13b, 11(%rsi)
; SCALAR-NEXT:    movb %r12b, 10(%rsi)
; SCALAR-NEXT:    movb %r15b, 9(%rsi)
; SCALAR-NEXT:    movb %r14b, 8(%rsi)
; SCALAR-NEXT:    movb %bl, 7(%rsi)
; SCALAR-NEXT:    movb %r10b, 6(%rsi)
; SCALAR-NEXT:    movl %r10d, %ebx
; SCALAR-NEXT:    movb %r10b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %r9b, 5(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r11d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r11b, 4(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r12d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r12b, 3(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r13d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r13b, 1(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r10d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r10b, (%rsi)
; SCALAR-NEXT:    movb %r8b, 15(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %dil, 14(%rdx)
; SCALAR-NEXT:    movb %al, 13(%rdx)
; SCALAR-NEXT:    movb %bpl, 12(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 11(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 10(%rdx)
; SCALAR-NEXT:    movb %r15b, 9(%rdx)
; SCALAR-NEXT:    movb %r14b, 8(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ebp # 1-byte Folded Reload
; SCALAR-NEXT:    movb %bpl, 7(%rdx)
; SCALAR-NEXT:    movb %bl, 6(%rdx)
; SCALAR-NEXT:    movb %r9b, 5(%rdx)
; SCALAR-NEXT:    movb %r9b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %r11b, 4(%rdx)
; SCALAR-NEXT:    movb %r12b, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movl %r13d, %ebx
; SCALAR-NEXT:    movb %r13b, 1(%rdx)
; SCALAR-NEXT:    movl %r10d, %esi
; SCALAR-NEXT:    movb %r10b, (%rdx)
; SCALAR-NEXT:    movb %r8b, 31(%rdx)
; SCALAR-NEXT:    movb %dil, 30(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 29(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r11d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r11b, 28(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r13d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r13b, 27(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r12d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r12b, 26(%rdx)
; SCALAR-NEXT:    movb %r15b, 25(%rdx)
; SCALAR-NEXT:    movb %r14b, 24(%rdx)
; SCALAR-NEXT:    movb %bpl, 23(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r10d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r10b, 22(%rdx)
; SCALAR-NEXT:    movb %r9b, 21(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r9d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r9b, 20(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %dil, 19(%rdx)
; SCALAR-NEXT:    movb %cl, 18(%rdx)
; SCALAR-NEXT:    movb %bl, 17(%rdx)
; SCALAR-NEXT:    movb %sil, 16(%rdx)
; SCALAR-NEXT:    movb %r8b, 47(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r8b, 46(%rdx)
; SCALAR-NEXT:    movb %al, 45(%rdx)
; SCALAR-NEXT:    movb %r11b, 44(%rdx)
; SCALAR-NEXT:    movb %r13b, 43(%rdx)
; SCALAR-NEXT:    movb %r12b, 42(%rdx)
; SCALAR-NEXT:    movb %r15b, 41(%rdx)
; SCALAR-NEXT:    movb %r14b, 40(%rdx)
; SCALAR-NEXT:    movb %bpl, 39(%rdx)
; SCALAR-NEXT:    movb %r10b, 38(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 37(%rdx)
; SCALAR-NEXT:    movb %r9b, 36(%rdx)
; SCALAR-NEXT:    movb %dil, 35(%rdx)
; SCALAR-NEXT:    movb %cl, 34(%rdx)
; SCALAR-NEXT:    movb %bl, 33(%rdx)
; SCALAR-NEXT:    movb %sil, 32(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    popq %r12
; SCALAR-NEXT:    popq %r13
; SCALAR-NEXT:    popq %r14
; SCALAR-NEXT:    popq %r15
; SCALAR-NEXT:    popq %rbp
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v16i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec384_v16i8:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <16 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <16 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <16 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <16 x i8>, ptr %out.vec.ptr, i64 0
  store <16 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <16 x i8>, ptr %out.vec.ptr, i64 1
  store <16 x i8> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <16 x i8>, ptr %out.vec.ptr, i64 2
  store <16 x i8> %in.subvec, ptr %out.subvec2.ptr, align 32
  ret void
}

define void @vec384_v24i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec384_v24i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    movq 16(%rdi), %rdi
; SCALAR-NEXT:    notq %rdi
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rdi, 16(%rsi)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rdi, 16(%rdx)
; SCALAR-NEXT:    movq %rdi, 48(%rdx)
; SCALAR-NEXT:    movq %rcx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec384_v24i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    movdqa 16(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movq %xmm1, 16(%rsi)
; SSE2-NEXT:    movq %xmm1, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movq %xmm1, 48(%rdx)
; SSE2-NEXT:    movdqu %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec384_v24i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT:    vmovlps %xmm1, 16(%rsi)
; AVX1-NEXT:    vmovaps %xmm0, (%rsi)
; AVX1-NEXT:    vmovlps %xmm1, 16(%rdx)
; AVX1-NEXT:    vmovaps %xmm0, (%rdx)
; AVX1-NEXT:    vmovlps %xmm1, 48(%rdx)
; AVX1-NEXT:    vmovups %xmm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec384_v24i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vmovq %xmm1, 16(%rsi)
; AVX2-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX2-NEXT:    vmovq %xmm1, 16(%rdx)
; AVX2-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX2-NEXT:    vmovq %xmm1, 48(%rdx)
; AVX2-NEXT:    vmovdqu %xmm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <24 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <24 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <24 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <24 x i8>, ptr %out.vec.ptr, i64 0
  store <24 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <24 x i8>, ptr %out.vec.ptr, i64 1
  store <24 x i8> %in.subvec, ptr %out.subvec1.ptr, align 8
  ret void
}

define void @vec512_v2i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v2i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %ecx
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %cl, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %cl, 3(%rdx)
; SCALAR-NEXT:    movb %al, 2(%rdx)
; SCALAR-NEXT:    movb %cl, 5(%rdx)
; SCALAR-NEXT:    movb %al, 4(%rdx)
; SCALAR-NEXT:    movb %cl, 7(%rdx)
; SCALAR-NEXT:    movb %al, 6(%rdx)
; SCALAR-NEXT:    movb %cl, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    movb %cl, 11(%rdx)
; SCALAR-NEXT:    movb %al, 10(%rdx)
; SCALAR-NEXT:    movb %cl, 13(%rdx)
; SCALAR-NEXT:    movb %al, 12(%rdx)
; SCALAR-NEXT:    movb %cl, 15(%rdx)
; SCALAR-NEXT:    movb %al, 14(%rdx)
; SCALAR-NEXT:    movb %cl, 17(%rdx)
; SCALAR-NEXT:    movb %al, 16(%rdx)
; SCALAR-NEXT:    movb %cl, 19(%rdx)
; SCALAR-NEXT:    movb %al, 18(%rdx)
; SCALAR-NEXT:    movb %cl, 21(%rdx)
; SCALAR-NEXT:    movb %al, 20(%rdx)
; SCALAR-NEXT:    movb %cl, 23(%rdx)
; SCALAR-NEXT:    movb %al, 22(%rdx)
; SCALAR-NEXT:    movb %cl, 25(%rdx)
; SCALAR-NEXT:    movb %al, 24(%rdx)
; SCALAR-NEXT:    movb %cl, 27(%rdx)
; SCALAR-NEXT:    movb %al, 26(%rdx)
; SCALAR-NEXT:    movb %cl, 29(%rdx)
; SCALAR-NEXT:    movb %al, 28(%rdx)
; SCALAR-NEXT:    movb %cl, 31(%rdx)
; SCALAR-NEXT:    movb %al, 30(%rdx)
; SCALAR-NEXT:    movb %cl, 33(%rdx)
; SCALAR-NEXT:    movb %al, 32(%rdx)
; SCALAR-NEXT:    movb %cl, 35(%rdx)
; SCALAR-NEXT:    movb %al, 34(%rdx)
; SCALAR-NEXT:    movb %cl, 37(%rdx)
; SCALAR-NEXT:    movb %al, 36(%rdx)
; SCALAR-NEXT:    movb %cl, 39(%rdx)
; SCALAR-NEXT:    movb %al, 38(%rdx)
; SCALAR-NEXT:    movb %cl, 41(%rdx)
; SCALAR-NEXT:    movb %al, 40(%rdx)
; SCALAR-NEXT:    movb %cl, 43(%rdx)
; SCALAR-NEXT:    movb %al, 42(%rdx)
; SCALAR-NEXT:    movb %cl, 45(%rdx)
; SCALAR-NEXT:    movb %al, 44(%rdx)
; SCALAR-NEXT:    movb %cl, 47(%rdx)
; SCALAR-NEXT:    movb %al, 46(%rdx)
; SCALAR-NEXT:    movb %cl, 49(%rdx)
; SCALAR-NEXT:    movb %al, 48(%rdx)
; SCALAR-NEXT:    movb %cl, 51(%rdx)
; SCALAR-NEXT:    movb %al, 50(%rdx)
; SCALAR-NEXT:    movb %cl, 53(%rdx)
; SCALAR-NEXT:    movb %al, 52(%rdx)
; SCALAR-NEXT:    movb %cl, 55(%rdx)
; SCALAR-NEXT:    movb %al, 54(%rdx)
; SCALAR-NEXT:    movb %cl, 57(%rdx)
; SCALAR-NEXT:    movb %al, 56(%rdx)
; SCALAR-NEXT:    movb %cl, 59(%rdx)
; SCALAR-NEXT:    movb %al, 58(%rdx)
; SCALAR-NEXT:    movb %cl, 61(%rdx)
; SCALAR-NEXT:    movb %al, 60(%rdx)
; SCALAR-NEXT:    movb %cl, 63(%rdx)
; SCALAR-NEXT:    movb %al, 62(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-ONLY-LABEL: vec512_v2i8:
; SSE2-ONLY:       # %bb.0:
; SSE2-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSE2-ONLY-NEXT:    movd %xmm0, %eax
; SSE2-ONLY-NEXT:    movw %ax, (%rsi)
; SSE2-ONLY-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE2-ONLY-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-ONLY-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-ONLY-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-ONLY-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-ONLY-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-ONLY-NEXT:    retq
;
; SSE3-LABEL: vec512_v2i8:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE3-NEXT:    pxor (%rdi), %xmm0
; SSE3-NEXT:    movd %xmm0, %eax
; SSE3-NEXT:    movw %ax, (%rsi)
; SSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE3-NEXT:    movdqa %xmm0, (%rdx)
; SSE3-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE3-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE3-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE3-NEXT:    retq
;
; SSSE3-ONLY-LABEL: vec512_v2i8:
; SSSE3-ONLY:       # %bb.0:
; SSSE3-ONLY-NEXT:    pcmpeqd %xmm0, %xmm0
; SSSE3-ONLY-NEXT:    pxor (%rdi), %xmm0
; SSSE3-ONLY-NEXT:    movd %xmm0, %eax
; SSSE3-ONLY-NEXT:    movw %ax, (%rsi)
; SSSE3-ONLY-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSSE3-ONLY-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSSE3-ONLY-NEXT:    movdqa %xmm0, (%rdx)
; SSSE3-ONLY-NEXT:    movdqa %xmm0, 16(%rdx)
; SSSE3-ONLY-NEXT:    movdqa %xmm0, 32(%rdx)
; SSSE3-ONLY-NEXT:    movdqa %xmm0, 48(%rdx)
; SSSE3-ONLY-NEXT:    retq
;
; SSE41-LABEL: vec512_v2i8:
; SSE41:       # %bb.0:
; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE41-NEXT:    pxor (%rdi), %xmm0
; SSE41-NEXT:    pextrw $0, %xmm0, (%rsi)
; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE41-NEXT:    movdqa %xmm0, (%rdx)
; SSE41-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE41-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE41-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE41-NEXT:    retq
;
; SSE42-LABEL: vec512_v2i8:
; SSE42:       # %bb.0:
; SSE42-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE42-NEXT:    pxor (%rdi), %xmm0
; SSE42-NEXT:    pextrw $0, %xmm0, (%rsi)
; SSE42-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; SSE42-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE42-NEXT:    movdqa %xmm0, (%rdx)
; SSE42-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE42-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE42-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE42-NEXT:    retq
;
; AVX1-LABEL: vec512_v2i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec512_v2i8:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastw %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512F-LABEL: vec512_v2i8:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512F-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX512F-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVX512F-NEXT:    vpbroadcastw %xmm0, %ymm0
; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
; AVX512F-NEXT:    vmovdqa64 %zmm0, (%rdx)
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512BW-LABEL: vec512_v2i8:
; AVX512BW:       # %bb.0:
; AVX512BW-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512BW-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX512BW-NEXT:    vpextrw $0, %xmm0, (%rsi)
; AVX512BW-NEXT:    vpbroadcastw %xmm0, %zmm0
; AVX512BW-NEXT:    vmovdqa64 %zmm0, (%rdx)
; AVX512BW-NEXT:    vzeroupper
; AVX512BW-NEXT:    retq
  %in.subvec.not = load <2 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i8> %in.subvec.not, <i8 -1, i8 -1>
  store <2 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 0
  store <2 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 1
  store <2 x i8> %in.subvec, ptr %out.subvec1.ptr, align 2
  %out.subvec2.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 2
  store <2 x i8> %in.subvec, ptr %out.subvec2.ptr, align 4
  %out.subvec3.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 3
  store <2 x i8> %in.subvec, ptr %out.subvec3.ptr, align 2
  %out.subvec4.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 4
  store <2 x i8> %in.subvec, ptr %out.subvec4.ptr, align 8
  %out.subvec5.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 5
  store <2 x i8> %in.subvec, ptr %out.subvec5.ptr, align 2
  %out.subvec6.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 6
  store <2 x i8> %in.subvec, ptr %out.subvec6.ptr, align 4
  %out.subvec7.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 7
  store <2 x i8> %in.subvec, ptr %out.subvec7.ptr, align 2
  %out.subvec8.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 8
  store <2 x i8> %in.subvec, ptr %out.subvec8.ptr, align 16
  %out.subvec9.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 9
  store <2 x i8> %in.subvec, ptr %out.subvec9.ptr, align 2
  %out.subvec10.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 10
  store <2 x i8> %in.subvec, ptr %out.subvec10.ptr, align 4
  %out.subvec11.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 11
  store <2 x i8> %in.subvec, ptr %out.subvec11.ptr, align 2
  %out.subvec12.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 12
  store <2 x i8> %in.subvec, ptr %out.subvec12.ptr, align 8
  %out.subvec13.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 13
  store <2 x i8> %in.subvec, ptr %out.subvec13.ptr, align 2
  %out.subvec14.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 14
  store <2 x i8> %in.subvec, ptr %out.subvec14.ptr, align 4
  %out.subvec15.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 15
  store <2 x i8> %in.subvec, ptr %out.subvec15.ptr, align 2
  %out.subvec16.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 16
  store <2 x i8> %in.subvec, ptr %out.subvec16.ptr, align 32
  %out.subvec17.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 17
  store <2 x i8> %in.subvec, ptr %out.subvec17.ptr, align 2
  %out.subvec18.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 18
  store <2 x i8> %in.subvec, ptr %out.subvec18.ptr, align 4
  %out.subvec19.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 19
  store <2 x i8> %in.subvec, ptr %out.subvec19.ptr, align 2
  %out.subvec20.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 20
  store <2 x i8> %in.subvec, ptr %out.subvec20.ptr, align 8
  %out.subvec21.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 21
  store <2 x i8> %in.subvec, ptr %out.subvec21.ptr, align 2
  %out.subvec22.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 22
  store <2 x i8> %in.subvec, ptr %out.subvec22.ptr, align 4
  %out.subvec23.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 23
  store <2 x i8> %in.subvec, ptr %out.subvec23.ptr, align 2
  %out.subvec24.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 24
  store <2 x i8> %in.subvec, ptr %out.subvec24.ptr, align 16
  %out.subvec25.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 25
  store <2 x i8> %in.subvec, ptr %out.subvec25.ptr, align 2
  %out.subvec26.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 26
  store <2 x i8> %in.subvec, ptr %out.subvec26.ptr, align 4
  %out.subvec27.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 27
  store <2 x i8> %in.subvec, ptr %out.subvec27.ptr, align 2
  %out.subvec28.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 28
  store <2 x i8> %in.subvec, ptr %out.subvec28.ptr, align 8
  %out.subvec29.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 29
  store <2 x i8> %in.subvec, ptr %out.subvec29.ptr, align 2
  %out.subvec30.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 30
  store <2 x i8> %in.subvec, ptr %out.subvec30.ptr, align 4
  %out.subvec31.ptr = getelementptr <2 x i8>, ptr %out.vec.ptr, i64 31
  store <2 x i8> %in.subvec, ptr %out.subvec31.ptr, align 2
  ret void
}

define void @vec512_v2i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v2i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzwl 2(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movw %cx, 2(%rsi)
; SCALAR-NEXT:    movw %ax, (%rsi)
; SCALAR-NEXT:    movw %cx, 2(%rdx)
; SCALAR-NEXT:    movw %ax, (%rdx)
; SCALAR-NEXT:    movw %cx, 6(%rdx)
; SCALAR-NEXT:    movw %ax, 4(%rdx)
; SCALAR-NEXT:    movw %cx, 10(%rdx)
; SCALAR-NEXT:    movw %ax, 8(%rdx)
; SCALAR-NEXT:    movw %cx, 14(%rdx)
; SCALAR-NEXT:    movw %ax, 12(%rdx)
; SCALAR-NEXT:    movw %cx, 18(%rdx)
; SCALAR-NEXT:    movw %ax, 16(%rdx)
; SCALAR-NEXT:    movw %cx, 22(%rdx)
; SCALAR-NEXT:    movw %ax, 20(%rdx)
; SCALAR-NEXT:    movw %cx, 26(%rdx)
; SCALAR-NEXT:    movw %ax, 24(%rdx)
; SCALAR-NEXT:    movw %cx, 30(%rdx)
; SCALAR-NEXT:    movw %ax, 28(%rdx)
; SCALAR-NEXT:    movw %cx, 34(%rdx)
; SCALAR-NEXT:    movw %ax, 32(%rdx)
; SCALAR-NEXT:    movw %cx, 38(%rdx)
; SCALAR-NEXT:    movw %ax, 36(%rdx)
; SCALAR-NEXT:    movw %cx, 42(%rdx)
; SCALAR-NEXT:    movw %ax, 40(%rdx)
; SCALAR-NEXT:    movw %cx, 46(%rdx)
; SCALAR-NEXT:    movw %ax, 44(%rdx)
; SCALAR-NEXT:    movw %cx, 50(%rdx)
; SCALAR-NEXT:    movw %ax, 48(%rdx)
; SCALAR-NEXT:    movw %cx, 54(%rdx)
; SCALAR-NEXT:    movw %ax, 52(%rdx)
; SCALAR-NEXT:    movw %cx, 58(%rdx)
; SCALAR-NEXT:    movw %ax, 56(%rdx)
; SCALAR-NEXT:    movw %cx, 62(%rdx)
; SCALAR-NEXT:    movw %ax, 60(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v2i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movd %xmm0, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v2i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vmovd %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec512_v2i16:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovd %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastd %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec512_v2i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX512-NEXT:    vmovd %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastd %xmm0, %zmm0
; AVX512-NEXT:    vmovdqa64 %zmm0, (%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <2 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i16> %in.subvec.not, <i16 -1, i16 -1>
  store <2 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 0
  store <2 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 1
  store <2 x i16> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 2
  store <2 x i16> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 3
  store <2 x i16> %in.subvec, ptr %out.subvec3.ptr, align 4
  %out.subvec4.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 4
  store <2 x i16> %in.subvec, ptr %out.subvec4.ptr, align 16
  %out.subvec5.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 5
  store <2 x i16> %in.subvec, ptr %out.subvec5.ptr, align 4
  %out.subvec6.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 6
  store <2 x i16> %in.subvec, ptr %out.subvec6.ptr, align 8
  %out.subvec7.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 7
  store <2 x i16> %in.subvec, ptr %out.subvec7.ptr, align 4
  %out.subvec8.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 8
  store <2 x i16> %in.subvec, ptr %out.subvec8.ptr, align 32
  %out.subvec9.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 9
  store <2 x i16> %in.subvec, ptr %out.subvec9.ptr, align 4
  %out.subvec10.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 10
  store <2 x i16> %in.subvec, ptr %out.subvec10.ptr, align 8
  %out.subvec11.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 11
  store <2 x i16> %in.subvec, ptr %out.subvec11.ptr, align 4
  %out.subvec12.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 12
  store <2 x i16> %in.subvec, ptr %out.subvec12.ptr, align 16
  %out.subvec13.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 13
  store <2 x i16> %in.subvec, ptr %out.subvec13.ptr, align 4
  %out.subvec14.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 14
  store <2 x i16> %in.subvec, ptr %out.subvec14.ptr, align 8
  %out.subvec15.ptr = getelementptr <2 x i16>, ptr %out.vec.ptr, i64 15
  store <2 x i16> %in.subvec, ptr %out.subvec15.ptr, align 4
  ret void
}

define void @vec512_v2i32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v2i32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movl %ecx, 4(%rsi)
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movl %ecx, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movl %ecx, 12(%rdx)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    movl %ecx, 20(%rdx)
; SCALAR-NEXT:    movl %eax, 16(%rdx)
; SCALAR-NEXT:    movl %ecx, 28(%rdx)
; SCALAR-NEXT:    movl %eax, 24(%rdx)
; SCALAR-NEXT:    movl %ecx, 36(%rdx)
; SCALAR-NEXT:    movl %eax, 32(%rdx)
; SCALAR-NEXT:    movl %ecx, 44(%rdx)
; SCALAR-NEXT:    movl %eax, 40(%rdx)
; SCALAR-NEXT:    movl %ecx, 52(%rdx)
; SCALAR-NEXT:    movl %eax, 48(%rdx)
; SCALAR-NEXT:    movl %ecx, 60(%rdx)
; SCALAR-NEXT:    movl %eax, 56(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v2i32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v2i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec512_v2i32:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec512_v2i32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %zmm0
; AVX512-NEXT:    vmovdqa64 %zmm0, (%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <2 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i32> %in.subvec.not, <i32 -1, i32 -1>
  store <2 x i32> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 0
  store <2 x i32> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 1
  store <2 x i32> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 2
  store <2 x i32> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 3
  store <2 x i32> %in.subvec, ptr %out.subvec3.ptr, align 8
  %out.subvec4.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 4
  store <2 x i32> %in.subvec, ptr %out.subvec4.ptr, align 32
  %out.subvec5.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 5
  store <2 x i32> %in.subvec, ptr %out.subvec5.ptr, align 8
  %out.subvec6.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 6
  store <2 x i32> %in.subvec, ptr %out.subvec6.ptr, align 16
  %out.subvec7.ptr = getelementptr <2 x i32>, ptr %out.vec.ptr, i64 7
  store <2 x i32> %in.subvec, ptr %out.subvec7.ptr, align 8
  ret void
}

define void @vec512_v2f32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v2f32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %ecx
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movl %ecx, 4(%rsi)
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movl %ecx, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movl %ecx, 12(%rdx)
; SCALAR-NEXT:    movl %eax, 8(%rdx)
; SCALAR-NEXT:    movl %ecx, 20(%rdx)
; SCALAR-NEXT:    movl %eax, 16(%rdx)
; SCALAR-NEXT:    movl %ecx, 28(%rdx)
; SCALAR-NEXT:    movl %eax, 24(%rdx)
; SCALAR-NEXT:    movl %ecx, 36(%rdx)
; SCALAR-NEXT:    movl %eax, 32(%rdx)
; SCALAR-NEXT:    movl %ecx, 44(%rdx)
; SCALAR-NEXT:    movl %eax, 40(%rdx)
; SCALAR-NEXT:    movl %ecx, 52(%rdx)
; SCALAR-NEXT:    movl %eax, 48(%rdx)
; SCALAR-NEXT:    movl %ecx, 60(%rdx)
; SCALAR-NEXT:    movl %eax, 56(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v2f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v2f32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec512_v2f32:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec512_v2f32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %zmm0
; AVX512-NEXT:    vmovdqa64 %zmm0, (%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <2 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <2 x i32> %in.subvec.not, <i32 -1, i32 -1>
  %in.subvec = bitcast <2 x i32> %in.subvec.int to <2 x float>
  store <2 x float> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 0
  store <2 x float> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 1
  store <2 x float> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 2
  store <2 x float> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 3
  store <2 x float> %in.subvec, ptr %out.subvec3.ptr, align 8
  %out.subvec4.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 4
  store <2 x float> %in.subvec, ptr %out.subvec4.ptr, align 32
  %out.subvec5.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 5
  store <2 x float> %in.subvec, ptr %out.subvec5.ptr, align 8
  %out.subvec6.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 6
  store <2 x float> %in.subvec, ptr %out.subvec6.ptr, align 16
  %out.subvec7.ptr = getelementptr <2 x float>, ptr %out.vec.ptr, i64 7
  store <2 x float> %in.subvec, ptr %out.subvec7.ptr, align 8
  ret void
}

define void @vec512_v2i64(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v2i64:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 24(%rdx)
; SCALAR-NEXT:    movq %rax, 16(%rdx)
; SCALAR-NEXT:    movq %rcx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    movq %rcx, 56(%rdx)
; SCALAR-NEXT:    movq %rax, 48(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v2i64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec512_v2i64:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 48(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <2 x i64>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i64> %in.subvec.not, <i64 -1, i64 -1>
  store <2 x i64> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i64>, ptr %out.vec.ptr, i64 0
  store <2 x i64> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i64>, ptr %out.vec.ptr, i64 1
  store <2 x i64> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <2 x i64>, ptr %out.vec.ptr, i64 2
  store <2 x i64> %in.subvec, ptr %out.subvec2.ptr, align 32
  %out.subvec3.ptr = getelementptr <2 x i64>, ptr %out.vec.ptr, i64 3
  store <2 x i64> %in.subvec, ptr %out.subvec3.ptr, align 16
  ret void
}

define void @vec512_v2f64(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v2f64:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq (%rdi), %rax
; SCALAR-NEXT:    movq 8(%rdi), %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    movq %rcx, 8(%rsi)
; SCALAR-NEXT:    movq %rax, (%rsi)
; SCALAR-NEXT:    movq %rcx, 8(%rdx)
; SCALAR-NEXT:    movq %rax, (%rdx)
; SCALAR-NEXT:    movq %rcx, 24(%rdx)
; SCALAR-NEXT:    movq %rax, 16(%rdx)
; SCALAR-NEXT:    movq %rcx, 40(%rdx)
; SCALAR-NEXT:    movq %rax, 32(%rdx)
; SCALAR-NEXT:    movq %rcx, 56(%rdx)
; SCALAR-NEXT:    movq %rax, 48(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v2f64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec512_v2f64:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 48(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <2 x i64>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <2 x i64> %in.subvec.not, <i64 -1, i64 -1>
  %in.subvec = bitcast <2 x i64> %in.subvec.int to <2 x double>
  store <2 x double> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x double>, ptr %out.vec.ptr, i64 0
  store <2 x double> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x double>, ptr %out.vec.ptr, i64 1
  store <2 x double> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <2 x double>, ptr %out.vec.ptr, i64 2
  store <2 x double> %in.subvec, ptr %out.subvec2.ptr, align 32
  %out.subvec3.ptr = getelementptr <2 x double>, ptr %out.vec.ptr, i64 3
  store <2 x double> %in.subvec, ptr %out.subvec3.ptr, align 16
  ret void
}

define void @vec512_v2i128(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; ALL-LABEL: vec512_v2i128:
; ALL:       # %bb.0:
; ALL-NEXT:    movq 16(%rdi), %rax
; ALL-NEXT:    movq 24(%rdi), %rcx
; ALL-NEXT:    movq (%rdi), %r8
; ALL-NEXT:    movq 8(%rdi), %rdi
; ALL-NEXT:    notq %rdi
; ALL-NEXT:    notq %r8
; ALL-NEXT:    notq %rcx
; ALL-NEXT:    notq %rax
; ALL-NEXT:    movq %rax, 16(%rsi)
; ALL-NEXT:    movq %rcx, 24(%rsi)
; ALL-NEXT:    movq %r8, (%rsi)
; ALL-NEXT:    movq %rdi, 8(%rsi)
; ALL-NEXT:    movq %rax, 16(%rdx)
; ALL-NEXT:    movq %rcx, 24(%rdx)
; ALL-NEXT:    movq %r8, (%rdx)
; ALL-NEXT:    movq %rdi, 8(%rdx)
; ALL-NEXT:    movq %rax, 48(%rdx)
; ALL-NEXT:    movq %rcx, 56(%rdx)
; ALL-NEXT:    movq %r8, 32(%rdx)
; ALL-NEXT:    movq %rdi, 40(%rdx)
; ALL-NEXT:    retq
  %in.subvec.not = load <2 x i128>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <2 x i128> %in.subvec.not, <i128 -1, i128 -1>
  store <2 x i128> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <2 x i128>, ptr %out.vec.ptr, i64 0
  store <2 x i128> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <2 x i128>, ptr %out.vec.ptr, i64 1
  store <2 x i128> %in.subvec, ptr %out.subvec1.ptr, align 32
  ret void
}

define void @vec512_v4i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v4i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzbl 3(%rdi), %r8d
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    movb %r8b, 3(%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %dil, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %r8b, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movb %dil, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %r8b, 7(%rdx)
; SCALAR-NEXT:    movb %cl, 6(%rdx)
; SCALAR-NEXT:    movb %dil, 5(%rdx)
; SCALAR-NEXT:    movb %al, 4(%rdx)
; SCALAR-NEXT:    movb %r8b, 11(%rdx)
; SCALAR-NEXT:    movb %cl, 10(%rdx)
; SCALAR-NEXT:    movb %dil, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    movb %r8b, 15(%rdx)
; SCALAR-NEXT:    movb %cl, 14(%rdx)
; SCALAR-NEXT:    movb %dil, 13(%rdx)
; SCALAR-NEXT:    movb %al, 12(%rdx)
; SCALAR-NEXT:    movb %r8b, 19(%rdx)
; SCALAR-NEXT:    movb %cl, 18(%rdx)
; SCALAR-NEXT:    movb %dil, 17(%rdx)
; SCALAR-NEXT:    movb %al, 16(%rdx)
; SCALAR-NEXT:    movb %r8b, 23(%rdx)
; SCALAR-NEXT:    movb %cl, 22(%rdx)
; SCALAR-NEXT:    movb %dil, 21(%rdx)
; SCALAR-NEXT:    movb %al, 20(%rdx)
; SCALAR-NEXT:    movb %r8b, 27(%rdx)
; SCALAR-NEXT:    movb %cl, 26(%rdx)
; SCALAR-NEXT:    movb %dil, 25(%rdx)
; SCALAR-NEXT:    movb %al, 24(%rdx)
; SCALAR-NEXT:    movb %r8b, 31(%rdx)
; SCALAR-NEXT:    movb %cl, 30(%rdx)
; SCALAR-NEXT:    movb %dil, 29(%rdx)
; SCALAR-NEXT:    movb %al, 28(%rdx)
; SCALAR-NEXT:    movb %r8b, 35(%rdx)
; SCALAR-NEXT:    movb %cl, 34(%rdx)
; SCALAR-NEXT:    movb %dil, 33(%rdx)
; SCALAR-NEXT:    movb %al, 32(%rdx)
; SCALAR-NEXT:    movb %r8b, 39(%rdx)
; SCALAR-NEXT:    movb %cl, 38(%rdx)
; SCALAR-NEXT:    movb %dil, 37(%rdx)
; SCALAR-NEXT:    movb %al, 36(%rdx)
; SCALAR-NEXT:    movb %r8b, 43(%rdx)
; SCALAR-NEXT:    movb %cl, 42(%rdx)
; SCALAR-NEXT:    movb %dil, 41(%rdx)
; SCALAR-NEXT:    movb %al, 40(%rdx)
; SCALAR-NEXT:    movb %r8b, 47(%rdx)
; SCALAR-NEXT:    movb %cl, 46(%rdx)
; SCALAR-NEXT:    movb %dil, 45(%rdx)
; SCALAR-NEXT:    movb %al, 44(%rdx)
; SCALAR-NEXT:    movb %r8b, 51(%rdx)
; SCALAR-NEXT:    movb %cl, 50(%rdx)
; SCALAR-NEXT:    movb %dil, 49(%rdx)
; SCALAR-NEXT:    movb %al, 48(%rdx)
; SCALAR-NEXT:    movb %r8b, 55(%rdx)
; SCALAR-NEXT:    movb %cl, 54(%rdx)
; SCALAR-NEXT:    movb %dil, 53(%rdx)
; SCALAR-NEXT:    movb %al, 52(%rdx)
; SCALAR-NEXT:    movb %r8b, 59(%rdx)
; SCALAR-NEXT:    movb %cl, 58(%rdx)
; SCALAR-NEXT:    movb %dil, 57(%rdx)
; SCALAR-NEXT:    movb %al, 56(%rdx)
; SCALAR-NEXT:    movb %r8b, 63(%rdx)
; SCALAR-NEXT:    movb %cl, 62(%rdx)
; SCALAR-NEXT:    movb %dil, 61(%rdx)
; SCALAR-NEXT:    movb %al, 60(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v4i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movd %xmm0, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v4i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX1-NEXT:    vmovd %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec512_v4i8:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovd %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastd %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec512_v4i8:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX512-NEXT:    vmovd %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastd %xmm0, %zmm0
; AVX512-NEXT:    vmovdqa64 %zmm0, (%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <4 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1>
  store <4 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 0
  store <4 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 1
  store <4 x i8> %in.subvec, ptr %out.subvec1.ptr, align 4
  %out.subvec2.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 2
  store <4 x i8> %in.subvec, ptr %out.subvec2.ptr, align 8
  %out.subvec3.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 3
  store <4 x i8> %in.subvec, ptr %out.subvec3.ptr, align 4
  %out.subvec4.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 4
  store <4 x i8> %in.subvec, ptr %out.subvec4.ptr, align 16
  %out.subvec5.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 5
  store <4 x i8> %in.subvec, ptr %out.subvec5.ptr, align 4
  %out.subvec6.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 6
  store <4 x i8> %in.subvec, ptr %out.subvec6.ptr, align 8
  %out.subvec7.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 7
  store <4 x i8> %in.subvec, ptr %out.subvec7.ptr, align 4
  %out.subvec8.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 8
  store <4 x i8> %in.subvec, ptr %out.subvec8.ptr, align 32
  %out.subvec9.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 9
  store <4 x i8> %in.subvec, ptr %out.subvec9.ptr, align 4
  %out.subvec10.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 10
  store <4 x i8> %in.subvec, ptr %out.subvec10.ptr, align 8
  %out.subvec11.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 11
  store <4 x i8> %in.subvec, ptr %out.subvec11.ptr, align 4
  %out.subvec12.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 12
  store <4 x i8> %in.subvec, ptr %out.subvec12.ptr, align 16
  %out.subvec13.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 13
  store <4 x i8> %in.subvec, ptr %out.subvec13.ptr, align 4
  %out.subvec14.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 14
  store <4 x i8> %in.subvec, ptr %out.subvec14.ptr, align 8
  %out.subvec15.ptr = getelementptr <4 x i8>, ptr %out.vec.ptr, i64 15
  store <4 x i8> %in.subvec, ptr %out.subvec15.ptr, align 4
  ret void
}

define void @vec512_v4i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v4i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movzwl 6(%rdi), %r8d
; SCALAR-NEXT:    movzwl 2(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %edi
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    movw %r8w, 6(%rsi)
; SCALAR-NEXT:    movw %di, 4(%rsi)
; SCALAR-NEXT:    movw %cx, 2(%rsi)
; SCALAR-NEXT:    movw %ax, (%rsi)
; SCALAR-NEXT:    movw %r8w, 6(%rdx)
; SCALAR-NEXT:    movw %di, 4(%rdx)
; SCALAR-NEXT:    movw %cx, 2(%rdx)
; SCALAR-NEXT:    movw %ax, (%rdx)
; SCALAR-NEXT:    movw %r8w, 14(%rdx)
; SCALAR-NEXT:    movw %di, 12(%rdx)
; SCALAR-NEXT:    movw %cx, 10(%rdx)
; SCALAR-NEXT:    movw %ax, 8(%rdx)
; SCALAR-NEXT:    movw %r8w, 22(%rdx)
; SCALAR-NEXT:    movw %di, 20(%rdx)
; SCALAR-NEXT:    movw %cx, 18(%rdx)
; SCALAR-NEXT:    movw %ax, 16(%rdx)
; SCALAR-NEXT:    movw %r8w, 30(%rdx)
; SCALAR-NEXT:    movw %di, 28(%rdx)
; SCALAR-NEXT:    movw %cx, 26(%rdx)
; SCALAR-NEXT:    movw %ax, 24(%rdx)
; SCALAR-NEXT:    movw %r8w, 38(%rdx)
; SCALAR-NEXT:    movw %di, 36(%rdx)
; SCALAR-NEXT:    movw %cx, 34(%rdx)
; SCALAR-NEXT:    movw %ax, 32(%rdx)
; SCALAR-NEXT:    movw %r8w, 46(%rdx)
; SCALAR-NEXT:    movw %di, 44(%rdx)
; SCALAR-NEXT:    movw %cx, 42(%rdx)
; SCALAR-NEXT:    movw %ax, 40(%rdx)
; SCALAR-NEXT:    movw %r8w, 54(%rdx)
; SCALAR-NEXT:    movw %di, 52(%rdx)
; SCALAR-NEXT:    movw %cx, 50(%rdx)
; SCALAR-NEXT:    movw %ax, 48(%rdx)
; SCALAR-NEXT:    movw %r8w, 62(%rdx)
; SCALAR-NEXT:    movw %di, 60(%rdx)
; SCALAR-NEXT:    movw %cx, 58(%rdx)
; SCALAR-NEXT:    movw %ax, 56(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v4i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v4i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec512_v4i16:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec512_v4i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %zmm0
; AVX512-NEXT:    vmovdqa64 %zmm0, (%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <4 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1, i16 -1>
  store <4 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 0
  store <4 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 1
  store <4 x i16> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 2
  store <4 x i16> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 3
  store <4 x i16> %in.subvec, ptr %out.subvec3.ptr, align 8
  %out.subvec4.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 4
  store <4 x i16> %in.subvec, ptr %out.subvec4.ptr, align 32
  %out.subvec5.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 5
  store <4 x i16> %in.subvec, ptr %out.subvec5.ptr, align 8
  %out.subvec6.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 6
  store <4 x i16> %in.subvec, ptr %out.subvec6.ptr, align 16
  %out.subvec7.ptr = getelementptr <4 x i16>, ptr %out.vec.ptr, i64 7
  store <4 x i16> %in.subvec, ptr %out.subvec7.ptr, align 8
  ret void
}

define void @vec512_v4i32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v4i32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movaps (%rdi), %xmm0
; SCALAR-NEXT:    xorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SCALAR-NEXT:    movaps %xmm0, (%rsi)
; SCALAR-NEXT:    movaps %xmm0, (%rdx)
; SCALAR-NEXT:    movaps %xmm0, 16(%rdx)
; SCALAR-NEXT:    movaps %xmm0, 32(%rdx)
; SCALAR-NEXT:    movaps %xmm0, 48(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v4i32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec512_v4i32:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 48(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <4 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1, i32 -1>
  store <4 x i32> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i32>, ptr %out.vec.ptr, i64 0
  store <4 x i32> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i32>, ptr %out.vec.ptr, i64 1
  store <4 x i32> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <4 x i32>, ptr %out.vec.ptr, i64 2
  store <4 x i32> %in.subvec, ptr %out.subvec2.ptr, align 32
  %out.subvec3.ptr = getelementptr <4 x i32>, ptr %out.vec.ptr, i64 3
  store <4 x i32> %in.subvec, ptr %out.subvec3.ptr, align 16
  ret void
}

define void @vec512_v4f32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v4f32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movaps (%rdi), %xmm0
; SCALAR-NEXT:    xorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SCALAR-NEXT:    movaps %xmm0, (%rsi)
; SCALAR-NEXT:    movaps %xmm0, (%rdx)
; SCALAR-NEXT:    movaps %xmm0, 16(%rdx)
; SCALAR-NEXT:    movaps %xmm0, 32(%rdx)
; SCALAR-NEXT:    movaps %xmm0, 48(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v4f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec512_v4f32:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 48(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <4 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <4 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1, i32 -1>
  %in.subvec = bitcast <4 x i32> %in.subvec.int to <4 x float>
  store <4 x float> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x float>, ptr %out.vec.ptr, i64 0
  store <4 x float> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x float>, ptr %out.vec.ptr, i64 1
  store <4 x float> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <4 x float>, ptr %out.vec.ptr, i64 2
  store <4 x float> %in.subvec, ptr %out.subvec2.ptr, align 32
  %out.subvec3.ptr = getelementptr <4 x float>, ptr %out.vec.ptr, i64 3
  store <4 x float> %in.subvec, ptr %out.subvec3.ptr, align 16
  ret void
}

define void @vec512_v4i64(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v4i64:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq 24(%rdi), %rax
; SCALAR-NEXT:    movq 16(%rdi), %rcx
; SCALAR-NEXT:    movq (%rdi), %r8
; SCALAR-NEXT:    movq 8(%rdi), %rdi
; SCALAR-NEXT:    notq %r8
; SCALAR-NEXT:    notq %rdi
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    movq %rax, 24(%rsi)
; SCALAR-NEXT:    movq %rcx, 16(%rsi)
; SCALAR-NEXT:    movq %rdi, 8(%rsi)
; SCALAR-NEXT:    movq %r8, (%rsi)
; SCALAR-NEXT:    movq %rax, 24(%rdx)
; SCALAR-NEXT:    movq %rcx, 16(%rdx)
; SCALAR-NEXT:    movq %rdi, 8(%rdx)
; SCALAR-NEXT:    movq %r8, (%rdx)
; SCALAR-NEXT:    movq %rax, 56(%rdx)
; SCALAR-NEXT:    movq %rcx, 48(%rdx)
; SCALAR-NEXT:    movq %rdi, 40(%rdx)
; SCALAR-NEXT:    movq %r8, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v4i64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    movdqa 16(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm1, 16(%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm1, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm1, 48(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v4i64:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rsi)
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec512_v4i64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rsi)
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <4 x i64>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <4 x i64> %in.subvec.not, <i64 -1, i64 -1, i64 -1, i64 -1>
  store <4 x i64> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x i64>, ptr %out.vec.ptr, i64 0
  store <4 x i64> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x i64>, ptr %out.vec.ptr, i64 1
  store <4 x i64> %in.subvec, ptr %out.subvec1.ptr, align 32
  ret void
}

define void @vec512_v4f64(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v4f64:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    movq 24(%rdi), %rax
; SCALAR-NEXT:    movq 16(%rdi), %rcx
; SCALAR-NEXT:    movq (%rdi), %r8
; SCALAR-NEXT:    movq 8(%rdi), %rdi
; SCALAR-NEXT:    notq %r8
; SCALAR-NEXT:    notq %rdi
; SCALAR-NEXT:    notq %rcx
; SCALAR-NEXT:    notq %rax
; SCALAR-NEXT:    movq %rax, 24(%rsi)
; SCALAR-NEXT:    movq %rcx, 16(%rsi)
; SCALAR-NEXT:    movq %rdi, 8(%rsi)
; SCALAR-NEXT:    movq %r8, (%rsi)
; SCALAR-NEXT:    movq %rax, 24(%rdx)
; SCALAR-NEXT:    movq %rcx, 16(%rdx)
; SCALAR-NEXT:    movq %rdi, 8(%rdx)
; SCALAR-NEXT:    movq %r8, (%rdx)
; SCALAR-NEXT:    movq %rax, 56(%rdx)
; SCALAR-NEXT:    movq %rcx, 48(%rdx)
; SCALAR-NEXT:    movq %rdi, 40(%rdx)
; SCALAR-NEXT:    movq %r8, 32(%rdx)
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v4f64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    movdqa 16(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm1, 16(%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm1, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm1, 48(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v4f64:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rsi)
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec512_v4f64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rsi)
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <4 x i64>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <4 x i64> %in.subvec.not, <i64 -1, i64 -1, i64 -1, i64 -1>
  %in.subvec = bitcast <4 x i64> %in.subvec.int to <4 x double>
  store <4 x double> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <4 x double>, ptr %out.vec.ptr, i64 0
  store <4 x double> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <4 x double>, ptr %out.vec.ptr, i64 1
  store <4 x double> %in.subvec, ptr %out.subvec1.ptr, align 32
  ret void
}

define void @vec512_v8i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v8i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzbl 7(%rdi), %ebx
; SCALAR-NEXT:    movzbl 6(%rdi), %r11d
; SCALAR-NEXT:    movzbl 5(%rdi), %r10d
; SCALAR-NEXT:    movzbl 4(%rdi), %r9d
; SCALAR-NEXT:    movzbl 3(%rdi), %r8d
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    notb %r10b
; SCALAR-NEXT:    notb %r11b
; SCALAR-NEXT:    notb %bl
; SCALAR-NEXT:    movb %bl, 7(%rsi)
; SCALAR-NEXT:    movb %r11b, 6(%rsi)
; SCALAR-NEXT:    movb %r10b, 5(%rsi)
; SCALAR-NEXT:    movb %r9b, 4(%rsi)
; SCALAR-NEXT:    movb %r8b, 3(%rsi)
; SCALAR-NEXT:    movb %cl, 2(%rsi)
; SCALAR-NEXT:    movb %dil, 1(%rsi)
; SCALAR-NEXT:    movb %al, (%rsi)
; SCALAR-NEXT:    movb %bl, 7(%rdx)
; SCALAR-NEXT:    movb %r11b, 6(%rdx)
; SCALAR-NEXT:    movb %r10b, 5(%rdx)
; SCALAR-NEXT:    movb %r9b, 4(%rdx)
; SCALAR-NEXT:    movb %r8b, 3(%rdx)
; SCALAR-NEXT:    movb %cl, 2(%rdx)
; SCALAR-NEXT:    movb %dil, 1(%rdx)
; SCALAR-NEXT:    movb %al, (%rdx)
; SCALAR-NEXT:    movb %bl, 15(%rdx)
; SCALAR-NEXT:    movb %r11b, 14(%rdx)
; SCALAR-NEXT:    movb %r10b, 13(%rdx)
; SCALAR-NEXT:    movb %r9b, 12(%rdx)
; SCALAR-NEXT:    movb %r8b, 11(%rdx)
; SCALAR-NEXT:    movb %cl, 10(%rdx)
; SCALAR-NEXT:    movb %dil, 9(%rdx)
; SCALAR-NEXT:    movb %al, 8(%rdx)
; SCALAR-NEXT:    movb %bl, 23(%rdx)
; SCALAR-NEXT:    movb %r11b, 22(%rdx)
; SCALAR-NEXT:    movb %r10b, 21(%rdx)
; SCALAR-NEXT:    movb %r9b, 20(%rdx)
; SCALAR-NEXT:    movb %r8b, 19(%rdx)
; SCALAR-NEXT:    movb %cl, 18(%rdx)
; SCALAR-NEXT:    movb %dil, 17(%rdx)
; SCALAR-NEXT:    movb %al, 16(%rdx)
; SCALAR-NEXT:    movb %bl, 31(%rdx)
; SCALAR-NEXT:    movb %r11b, 30(%rdx)
; SCALAR-NEXT:    movb %r10b, 29(%rdx)
; SCALAR-NEXT:    movb %r9b, 28(%rdx)
; SCALAR-NEXT:    movb %r8b, 27(%rdx)
; SCALAR-NEXT:    movb %cl, 26(%rdx)
; SCALAR-NEXT:    movb %dil, 25(%rdx)
; SCALAR-NEXT:    movb %al, 24(%rdx)
; SCALAR-NEXT:    movb %bl, 39(%rdx)
; SCALAR-NEXT:    movb %r11b, 38(%rdx)
; SCALAR-NEXT:    movb %r10b, 37(%rdx)
; SCALAR-NEXT:    movb %r9b, 36(%rdx)
; SCALAR-NEXT:    movb %r8b, 35(%rdx)
; SCALAR-NEXT:    movb %cl, 34(%rdx)
; SCALAR-NEXT:    movb %dil, 33(%rdx)
; SCALAR-NEXT:    movb %al, 32(%rdx)
; SCALAR-NEXT:    movb %bl, 47(%rdx)
; SCALAR-NEXT:    movb %r11b, 46(%rdx)
; SCALAR-NEXT:    movb %r10b, 45(%rdx)
; SCALAR-NEXT:    movb %r9b, 44(%rdx)
; SCALAR-NEXT:    movb %r8b, 43(%rdx)
; SCALAR-NEXT:    movb %cl, 42(%rdx)
; SCALAR-NEXT:    movb %dil, 41(%rdx)
; SCALAR-NEXT:    movb %al, 40(%rdx)
; SCALAR-NEXT:    movb %bl, 55(%rdx)
; SCALAR-NEXT:    movb %r11b, 54(%rdx)
; SCALAR-NEXT:    movb %r10b, 53(%rdx)
; SCALAR-NEXT:    movb %r9b, 52(%rdx)
; SCALAR-NEXT:    movb %r8b, 51(%rdx)
; SCALAR-NEXT:    movb %cl, 50(%rdx)
; SCALAR-NEXT:    movb %dil, 49(%rdx)
; SCALAR-NEXT:    movb %al, 48(%rdx)
; SCALAR-NEXT:    movb %bl, 63(%rdx)
; SCALAR-NEXT:    movb %r11b, 62(%rdx)
; SCALAR-NEXT:    movb %r10b, 61(%rdx)
; SCALAR-NEXT:    movb %r9b, 60(%rdx)
; SCALAR-NEXT:    movb %r8b, 59(%rdx)
; SCALAR-NEXT:    movb %cl, 58(%rdx)
; SCALAR-NEXT:    movb %dil, 57(%rdx)
; SCALAR-NEXT:    movb %al, 56(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v8i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    movq %xmm1, (%rsi)
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v8i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vmovq %xmm0, (%rsi)
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-ONLY-LABEL: vec512_v8i8:
; AVX2-ONLY:       # %bb.0:
; AVX2-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-ONLY-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-ONLY-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; AVX2-ONLY-NEXT:    vmovq %xmm0, (%rsi)
; AVX2-ONLY-NEXT:    vpbroadcastq %xmm0, %ymm0
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-ONLY-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-ONLY-NEXT:    vzeroupper
; AVX2-ONLY-NEXT:    retq
;
; AVX512-LABEL: vec512_v8i8:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    vmovq %xmm0, (%rsi)
; AVX512-NEXT:    vpbroadcastq %xmm0, %zmm0
; AVX512-NEXT:    vmovdqa64 %zmm0, (%rdx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %in.subvec.not = load <8 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <8 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <8 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 0
  store <8 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 1
  store <8 x i8> %in.subvec, ptr %out.subvec1.ptr, align 8
  %out.subvec2.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 2
  store <8 x i8> %in.subvec, ptr %out.subvec2.ptr, align 16
  %out.subvec3.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 3
  store <8 x i8> %in.subvec, ptr %out.subvec3.ptr, align 8
  %out.subvec4.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 4
  store <8 x i8> %in.subvec, ptr %out.subvec4.ptr, align 32
  %out.subvec5.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 5
  store <8 x i8> %in.subvec, ptr %out.subvec5.ptr, align 8
  %out.subvec6.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 6
  store <8 x i8> %in.subvec, ptr %out.subvec6.ptr, align 16
  %out.subvec7.ptr = getelementptr <8 x i8>, ptr %out.vec.ptr, i64 7
  store <8 x i8> %in.subvec, ptr %out.subvec7.ptr, align 8
  ret void
}

define void @vec512_v8i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v8i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzwl 14(%rdi), %ebx
; SCALAR-NEXT:    movl 12(%rdi), %r11d
; SCALAR-NEXT:    movzwl 10(%rdi), %r10d
; SCALAR-NEXT:    movl 8(%rdi), %r9d
; SCALAR-NEXT:    movzwl 6(%rdi), %r8d
; SCALAR-NEXT:    movzwl 2(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %edi
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    notl %r9d
; SCALAR-NEXT:    notl %r10d
; SCALAR-NEXT:    notl %r11d
; SCALAR-NEXT:    notl %ebx
; SCALAR-NEXT:    movw %bx, 14(%rsi)
; SCALAR-NEXT:    movw %r11w, 12(%rsi)
; SCALAR-NEXT:    movw %r10w, 10(%rsi)
; SCALAR-NEXT:    movw %r9w, 8(%rsi)
; SCALAR-NEXT:    movw %r8w, 6(%rsi)
; SCALAR-NEXT:    movw %di, 4(%rsi)
; SCALAR-NEXT:    movw %cx, 2(%rsi)
; SCALAR-NEXT:    movw %ax, (%rsi)
; SCALAR-NEXT:    movw %bx, 14(%rdx)
; SCALAR-NEXT:    movw %r11w, 12(%rdx)
; SCALAR-NEXT:    movw %r10w, 10(%rdx)
; SCALAR-NEXT:    movw %r9w, 8(%rdx)
; SCALAR-NEXT:    movw %r8w, 6(%rdx)
; SCALAR-NEXT:    movw %di, 4(%rdx)
; SCALAR-NEXT:    movw %cx, 2(%rdx)
; SCALAR-NEXT:    movw %ax, (%rdx)
; SCALAR-NEXT:    movw %bx, 30(%rdx)
; SCALAR-NEXT:    movw %r11w, 28(%rdx)
; SCALAR-NEXT:    movw %r10w, 26(%rdx)
; SCALAR-NEXT:    movw %r9w, 24(%rdx)
; SCALAR-NEXT:    movw %r8w, 22(%rdx)
; SCALAR-NEXT:    movw %di, 20(%rdx)
; SCALAR-NEXT:    movw %cx, 18(%rdx)
; SCALAR-NEXT:    movw %ax, 16(%rdx)
; SCALAR-NEXT:    movw %bx, 46(%rdx)
; SCALAR-NEXT:    movw %r11w, 44(%rdx)
; SCALAR-NEXT:    movw %r10w, 42(%rdx)
; SCALAR-NEXT:    movw %r9w, 40(%rdx)
; SCALAR-NEXT:    movw %r8w, 38(%rdx)
; SCALAR-NEXT:    movw %di, 36(%rdx)
; SCALAR-NEXT:    movw %cx, 34(%rdx)
; SCALAR-NEXT:    movw %ax, 32(%rdx)
; SCALAR-NEXT:    movw %bx, 62(%rdx)
; SCALAR-NEXT:    movw %r11w, 60(%rdx)
; SCALAR-NEXT:    movw %r10w, 58(%rdx)
; SCALAR-NEXT:    movw %r9w, 56(%rdx)
; SCALAR-NEXT:    movw %r8w, 54(%rdx)
; SCALAR-NEXT:    movw %di, 52(%rdx)
; SCALAR-NEXT:    movw %cx, 50(%rdx)
; SCALAR-NEXT:    movw %ax, 48(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v8i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec512_v8i16:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 48(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <8 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <8 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
  store <8 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <8 x i16>, ptr %out.vec.ptr, i64 0
  store <8 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <8 x i16>, ptr %out.vec.ptr, i64 1
  store <8 x i16> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <8 x i16>, ptr %out.vec.ptr, i64 2
  store <8 x i16> %in.subvec, ptr %out.subvec2.ptr, align 32
  %out.subvec3.ptr = getelementptr <8 x i16>, ptr %out.vec.ptr, i64 3
  store <8 x i16> %in.subvec, ptr %out.subvec3.ptr, align 16
  ret void
}

define void @vec512_v8i32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v8i32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movl 28(%rdi), %ebx
; SCALAR-NEXT:    movl 24(%rdi), %r11d
; SCALAR-NEXT:    movl 20(%rdi), %r10d
; SCALAR-NEXT:    movl 16(%rdi), %r9d
; SCALAR-NEXT:    movl 12(%rdi), %r8d
; SCALAR-NEXT:    movl 8(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %edi
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    notl %r9d
; SCALAR-NEXT:    notl %r10d
; SCALAR-NEXT:    notl %r11d
; SCALAR-NEXT:    notl %ebx
; SCALAR-NEXT:    movl %ebx, 28(%rsi)
; SCALAR-NEXT:    movl %r11d, 24(%rsi)
; SCALAR-NEXT:    movl %r10d, 20(%rsi)
; SCALAR-NEXT:    movl %r9d, 16(%rsi)
; SCALAR-NEXT:    movl %r8d, 12(%rsi)
; SCALAR-NEXT:    movl %ecx, 8(%rsi)
; SCALAR-NEXT:    movl %edi, 4(%rsi)
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movl %ebx, 28(%rdx)
; SCALAR-NEXT:    movl %r11d, 24(%rdx)
; SCALAR-NEXT:    movl %r10d, 20(%rdx)
; SCALAR-NEXT:    movl %r9d, 16(%rdx)
; SCALAR-NEXT:    movl %r8d, 12(%rdx)
; SCALAR-NEXT:    movl %ecx, 8(%rdx)
; SCALAR-NEXT:    movl %edi, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movl %ebx, 60(%rdx)
; SCALAR-NEXT:    movl %r11d, 56(%rdx)
; SCALAR-NEXT:    movl %r10d, 52(%rdx)
; SCALAR-NEXT:    movl %r9d, 48(%rdx)
; SCALAR-NEXT:    movl %r8d, 44(%rdx)
; SCALAR-NEXT:    movl %ecx, 40(%rdx)
; SCALAR-NEXT:    movl %edi, 36(%rdx)
; SCALAR-NEXT:    movl %eax, 32(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v8i32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    movdqa 16(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm1, 16(%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm1, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm1, 48(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v8i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rsi)
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec512_v8i32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rsi)
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <8 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <8 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
  store <8 x i32> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <8 x i32>, ptr %out.vec.ptr, i64 0
  store <8 x i32> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <8 x i32>, ptr %out.vec.ptr, i64 1
  store <8 x i32> %in.subvec, ptr %out.subvec1.ptr, align 32
  ret void
}

define void @vec512_v8f32(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v8f32:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movl 28(%rdi), %ebx
; SCALAR-NEXT:    movl 24(%rdi), %r11d
; SCALAR-NEXT:    movl 20(%rdi), %r10d
; SCALAR-NEXT:    movl 16(%rdi), %r9d
; SCALAR-NEXT:    movl 12(%rdi), %r8d
; SCALAR-NEXT:    movl 8(%rdi), %ecx
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %edi
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    notl %r9d
; SCALAR-NEXT:    notl %r10d
; SCALAR-NEXT:    notl %r11d
; SCALAR-NEXT:    notl %ebx
; SCALAR-NEXT:    movl %ebx, 28(%rsi)
; SCALAR-NEXT:    movl %r11d, 24(%rsi)
; SCALAR-NEXT:    movl %r10d, 20(%rsi)
; SCALAR-NEXT:    movl %r9d, 16(%rsi)
; SCALAR-NEXT:    movl %r8d, 12(%rsi)
; SCALAR-NEXT:    movl %ecx, 8(%rsi)
; SCALAR-NEXT:    movl %edi, 4(%rsi)
; SCALAR-NEXT:    movl %eax, (%rsi)
; SCALAR-NEXT:    movl %ebx, 28(%rdx)
; SCALAR-NEXT:    movl %r11d, 24(%rdx)
; SCALAR-NEXT:    movl %r10d, 20(%rdx)
; SCALAR-NEXT:    movl %r9d, 16(%rdx)
; SCALAR-NEXT:    movl %r8d, 12(%rdx)
; SCALAR-NEXT:    movl %ecx, 8(%rdx)
; SCALAR-NEXT:    movl %edi, 4(%rdx)
; SCALAR-NEXT:    movl %eax, (%rdx)
; SCALAR-NEXT:    movl %ebx, 60(%rdx)
; SCALAR-NEXT:    movl %r11d, 56(%rdx)
; SCALAR-NEXT:    movl %r10d, 52(%rdx)
; SCALAR-NEXT:    movl %r9d, 48(%rdx)
; SCALAR-NEXT:    movl %r8d, 44(%rdx)
; SCALAR-NEXT:    movl %ecx, 40(%rdx)
; SCALAR-NEXT:    movl %edi, 36(%rdx)
; SCALAR-NEXT:    movl %eax, 32(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v8f32:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    movdqa 16(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm1, 16(%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm1, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm1, 48(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v8f32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rsi)
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec512_v8f32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rsi)
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <8 x i32>, ptr %in.subvec.ptr, align 64
  %in.subvec.int = xor <8 x i32> %in.subvec.not, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
  %in.subvec = bitcast <8 x i32> %in.subvec.int to <8 x float>
  store <8 x float> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <8 x float>, ptr %out.vec.ptr, i64 0
  store <8 x float> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <8 x float>, ptr %out.vec.ptr, i64 1
  store <8 x float> %in.subvec, ptr %out.subvec1.ptr, align 32
  ret void
}

define void @vec512_v16i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v16i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbp
; SCALAR-NEXT:    pushq %r15
; SCALAR-NEXT:    pushq %r14
; SCALAR-NEXT:    pushq %r13
; SCALAR-NEXT:    pushq %r12
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzbl 15(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 14(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 13(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 12(%rdi), %r10d
; SCALAR-NEXT:    movzbl 11(%rdi), %r13d
; SCALAR-NEXT:    movzbl 10(%rdi), %r12d
; SCALAR-NEXT:    movzbl 9(%rdi), %r15d
; SCALAR-NEXT:    movzbl 8(%rdi), %r14d
; SCALAR-NEXT:    movzbl 7(%rdi), %ebp
; SCALAR-NEXT:    movzbl 6(%rdi), %r11d
; SCALAR-NEXT:    movzbl 5(%rdi), %ebx
; SCALAR-NEXT:    movzbl 4(%rdi), %r9d
; SCALAR-NEXT:    movzbl 3(%rdi), %r8d
; SCALAR-NEXT:    movzbl 2(%rdi), %ecx
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %edi
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    movb %dil, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    movb %r8b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    movb %r9b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movl %ebx, %r9d
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    notb %r11b
; SCALAR-NEXT:    movl %r11d, %ebx
; SCALAR-NEXT:    notb %bpl
; SCALAR-NEXT:    notb %r14b
; SCALAR-NEXT:    notb %r15b
; SCALAR-NEXT:    movb %r15b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r12b
; SCALAR-NEXT:    movb %r12b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r13b
; SCALAR-NEXT:    notb %r10b
; SCALAR-NEXT:    movb %r10b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 1-byte Folded Reload
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r11d # 1-byte Folded Reload
; SCALAR-NEXT:    notb %r11b
; SCALAR-NEXT:    movb %r11b, 15(%rsi)
; SCALAR-NEXT:    movb %r11b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %r8b, 14(%rsi)
; SCALAR-NEXT:    movb %r8b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movl %edi, %eax
; SCALAR-NEXT:    movb %dil, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %dil, 13(%rsi)
; SCALAR-NEXT:    movb %r10b, 12(%rsi)
; SCALAR-NEXT:    movb %r13b, 11(%rsi)
; SCALAR-NEXT:    movb %r12b, 10(%rsi)
; SCALAR-NEXT:    movb %r15b, 9(%rsi)
; SCALAR-NEXT:    movb %r14b, 8(%rsi)
; SCALAR-NEXT:    movl %r14d, %r12d
; SCALAR-NEXT:    movb %bpl, 7(%rsi)
; SCALAR-NEXT:    movl %ebp, %r14d
; SCALAR-NEXT:    movb %bpl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %bl, 6(%rsi)
; SCALAR-NEXT:    movb %bl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %r9b, 5(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %cl, 4(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ebp # 1-byte Folded Reload
; SCALAR-NEXT:    movb %bpl, 3(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %dil, 2(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %cl, 1(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r10d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r10b, (%rsi)
; SCALAR-NEXT:    movb %r11b, 15(%rdx)
; SCALAR-NEXT:    movb %r8b, 14(%rdx)
; SCALAR-NEXT:    movb %al, 13(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 12(%rdx)
; SCALAR-NEXT:    movb %r13b, 11(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r15d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r15b, 10(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 9(%rdx)
; SCALAR-NEXT:    movb %r12b, 8(%rdx)
; SCALAR-NEXT:    movb %r14b, 7(%rdx)
; SCALAR-NEXT:    movb %bl, 6(%rdx)
; SCALAR-NEXT:    movb %r9b, 5(%rdx)
; SCALAR-NEXT:    movl %r9d, %r11d
; SCALAR-NEXT:    movb %r9b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r8b, 4(%rdx)
; SCALAR-NEXT:    movb %bpl, 3(%rdx)
; SCALAR-NEXT:    movb %dil, 2(%rdx)
; SCALAR-NEXT:    movb %cl, 1(%rdx)
; SCALAR-NEXT:    movl %ecx, %r14d
; SCALAR-NEXT:    movl %r10d, %esi
; SCALAR-NEXT:    movb %r10b, (%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %cl, 31(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r9d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r9b, 30(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %dil, 29(%rdx)
; SCALAR-NEXT:    movb %al, 28(%rdx)
; SCALAR-NEXT:    movl %eax, %r10d
; SCALAR-NEXT:    movb %r13b, 27(%rdx)
; SCALAR-NEXT:    movb %r15b, 26(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r15d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r15b, 25(%rdx)
; SCALAR-NEXT:    movl %r12d, %ebp
; SCALAR-NEXT:    movb %r12b, 24(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ebx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %bl, 23(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 22(%rdx)
; SCALAR-NEXT:    movb %r11b, 21(%rdx)
; SCALAR-NEXT:    movb %r8b, 20(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r8b, 19(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r8b, 18(%rdx)
; SCALAR-NEXT:    movb %r14b, 17(%rdx)
; SCALAR-NEXT:    movb %sil, 16(%rdx)
; SCALAR-NEXT:    movl %esi, %r11d
; SCALAR-NEXT:    movb %cl, 47(%rdx)
; SCALAR-NEXT:    movb %r9b, 46(%rdx)
; SCALAR-NEXT:    movb %dil, 45(%rdx)
; SCALAR-NEXT:    movb %r10b, 44(%rdx)
; SCALAR-NEXT:    movb %r13b, 43(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r12d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r12b, 42(%rdx)
; SCALAR-NEXT:    movb %r15b, 41(%rdx)
; SCALAR-NEXT:    movl %ebp, %r14d
; SCALAR-NEXT:    movb %bpl, 40(%rdx)
; SCALAR-NEXT:    movl %ebx, %ebp
; SCALAR-NEXT:    movb %bl, 39(%rdx)
; SCALAR-NEXT:    movl %eax, %ebx
; SCALAR-NEXT:    movb %al, 38(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %cl, 37(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 36(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 35(%rdx)
; SCALAR-NEXT:    movb %r8b, 34(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r9d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r9b, 33(%rdx)
; SCALAR-NEXT:    movb %r11b, 32(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r11d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r11b, 63(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r11d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r11b, 62(%rdx)
; SCALAR-NEXT:    movb %dil, 61(%rdx)
; SCALAR-NEXT:    movb %r10b, 60(%rdx)
; SCALAR-NEXT:    movb %r13b, 59(%rdx)
; SCALAR-NEXT:    movb %r12b, 58(%rdx)
; SCALAR-NEXT:    movb %r15b, 57(%rdx)
; SCALAR-NEXT:    movb %r14b, 56(%rdx)
; SCALAR-NEXT:    movb %bpl, 55(%rdx)
; SCALAR-NEXT:    movb %bl, 54(%rdx)
; SCALAR-NEXT:    movb %cl, 53(%rdx)
; SCALAR-NEXT:    movb %al, 52(%rdx)
; SCALAR-NEXT:    movb %sil, 51(%rdx)
; SCALAR-NEXT:    movb %r8b, 50(%rdx)
; SCALAR-NEXT:    movb %r9b, 49(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 48(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    popq %r12
; SCALAR-NEXT:    popq %r13
; SCALAR-NEXT:    popq %r14
; SCALAR-NEXT:    popq %r15
; SCALAR-NEXT:    popq %rbp
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v16i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm0, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 48(%rdx)
; SSE2-NEXT:    retq
;
; AVX-LABEL: vec512_v16i8:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    vpxor (%rdi), %xmm0, %xmm0
; AVX-NEXT:    vmovdqa %xmm0, (%rsi)
; AVX-NEXT:    vmovdqa %xmm0, (%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 16(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 32(%rdx)
; AVX-NEXT:    vmovdqa %xmm0, 48(%rdx)
; AVX-NEXT:    retq
  %in.subvec.not = load <16 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <16 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <16 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <16 x i8>, ptr %out.vec.ptr, i64 0
  store <16 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <16 x i8>, ptr %out.vec.ptr, i64 1
  store <16 x i8> %in.subvec, ptr %out.subvec1.ptr, align 16
  %out.subvec2.ptr = getelementptr <16 x i8>, ptr %out.vec.ptr, i64 2
  store <16 x i8> %in.subvec, ptr %out.subvec2.ptr, align 32
  %out.subvec3.ptr = getelementptr <16 x i8>, ptr %out.vec.ptr, i64 3
  store <16 x i8> %in.subvec, ptr %out.subvec3.ptr, align 16
  ret void
}

define void @vec512_v16i16(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v16i16:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbp
; SCALAR-NEXT:    pushq %r15
; SCALAR-NEXT:    pushq %r14
; SCALAR-NEXT:    pushq %r13
; SCALAR-NEXT:    pushq %r12
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzwl 30(%rdi), %eax
; SCALAR-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    movl 28(%rdi), %eax
; SCALAR-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    movzwl 26(%rdi), %eax
; SCALAR-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    movl 24(%rdi), %r13d
; SCALAR-NEXT:    movzwl 22(%rdi), %r12d
; SCALAR-NEXT:    movl 20(%rdi), %r15d
; SCALAR-NEXT:    movzwl 18(%rdi), %r14d
; SCALAR-NEXT:    movl 16(%rdi), %ebx
; SCALAR-NEXT:    movzwl 14(%rdi), %r11d
; SCALAR-NEXT:    movl 12(%rdi), %r10d
; SCALAR-NEXT:    movzwl 10(%rdi), %r9d
; SCALAR-NEXT:    movl 8(%rdi), %r8d
; SCALAR-NEXT:    movzwl 6(%rdi), %ecx
; SCALAR-NEXT:    movzwl 2(%rdi), %ebp
; SCALAR-NEXT:    movl (%rdi), %eax
; SCALAR-NEXT:    movl 4(%rdi), %edi
; SCALAR-NEXT:    notl %eax
; SCALAR-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    notl %ebp
; SCALAR-NEXT:    movl %ebp, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    movl %edi, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    notl %ecx
; SCALAR-NEXT:    movl %ecx, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    movl %r8d, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    notl %r9d
; SCALAR-NEXT:    movl %r9d, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    movl %r10d, %edi
; SCALAR-NEXT:    notl %edi
; SCALAR-NEXT:    movl %edi, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    notl %r11d
; SCALAR-NEXT:    movl %r11d, %r9d
; SCALAR-NEXT:    notl %ebx
; SCALAR-NEXT:    movl %ebx, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    notl %r14d
; SCALAR-NEXT:    notl %r15d
; SCALAR-NEXT:    notl %r12d
; SCALAR-NEXT:    notl %r13d
; SCALAR-NEXT:    movl %r13d, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %r10d # 4-byte Reload
; SCALAR-NEXT:    notl %r10d
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %r11d # 4-byte Reload
; SCALAR-NEXT:    notl %r11d
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 4-byte Reload
; SCALAR-NEXT:    notl %r8d
; SCALAR-NEXT:    movl %r8d, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    movw %r8w, 30(%rsi)
; SCALAR-NEXT:    movw %r11w, 28(%rsi)
; SCALAR-NEXT:    movl %r11d, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    movw %r10w, 26(%rsi)
; SCALAR-NEXT:    movl %r10d, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; SCALAR-NEXT:    movw %r13w, 24(%rsi)
; SCALAR-NEXT:    movw %r12w, 22(%rsi)
; SCALAR-NEXT:    movw %r15w, 20(%rsi)
; SCALAR-NEXT:    movw %r14w, 18(%rsi)
; SCALAR-NEXT:    movw %bx, 16(%rsi)
; SCALAR-NEXT:    movw %r9w, 14(%rsi)
; SCALAR-NEXT:    movw %di, 12(%rsi)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %ebp # 4-byte Reload
; SCALAR-NEXT:    movw %bp, 10(%rsi)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 4-byte Reload
; SCALAR-NEXT:    movw %di, 8(%rsi)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 4-byte Reload
; SCALAR-NEXT:    movw %cx, 6(%rsi)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 4-byte Reload
; SCALAR-NEXT:    movw %r8w, 4(%rsi)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 4-byte Reload
; SCALAR-NEXT:    movw %ax, 2(%rsi)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %ebx # 4-byte Reload
; SCALAR-NEXT:    movw %bx, (%rsi)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %r13d # 4-byte Reload
; SCALAR-NEXT:    movw %r13w, 30(%rdx)
; SCALAR-NEXT:    movw %r11w, 28(%rdx)
; SCALAR-NEXT:    movw %r10w, 26(%rdx)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 4-byte Reload
; SCALAR-NEXT:    movw %si, 24(%rdx)
; SCALAR-NEXT:    movw %r12w, 22(%rdx)
; SCALAR-NEXT:    movw %r15w, 20(%rdx)
; SCALAR-NEXT:    movw %r14w, 18(%rdx)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %r11d # 4-byte Reload
; SCALAR-NEXT:    movw %r11w, 16(%rdx)
; SCALAR-NEXT:    movw %r9w, 14(%rdx)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %r10d # 4-byte Reload
; SCALAR-NEXT:    movw %r10w, 12(%rdx)
; SCALAR-NEXT:    movw %bp, 10(%rdx)
; SCALAR-NEXT:    movw %di, 8(%rdx)
; SCALAR-NEXT:    movw %cx, 6(%rdx)
; SCALAR-NEXT:    movw %r8w, 4(%rdx)
; SCALAR-NEXT:    movw %ax, 2(%rdx)
; SCALAR-NEXT:    movl %ebx, %esi
; SCALAR-NEXT:    movw %si, (%rdx)
; SCALAR-NEXT:    movw %r13w, 62(%rdx)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %ebx # 4-byte Reload
; SCALAR-NEXT:    movw %bx, 60(%rdx)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %ebx # 4-byte Reload
; SCALAR-NEXT:    movw %bx, 58(%rdx)
; SCALAR-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %ebx # 4-byte Reload
; SCALAR-NEXT:    movw %bx, 56(%rdx)
; SCALAR-NEXT:    movw %r12w, 54(%rdx)
; SCALAR-NEXT:    movw %r15w, 52(%rdx)
; SCALAR-NEXT:    movw %r14w, 50(%rdx)
; SCALAR-NEXT:    movw %r11w, 48(%rdx)
; SCALAR-NEXT:    movw %r9w, 46(%rdx)
; SCALAR-NEXT:    movw %r10w, 44(%rdx)
; SCALAR-NEXT:    movw %bp, 42(%rdx)
; SCALAR-NEXT:    movw %di, 40(%rdx)
; SCALAR-NEXT:    movw %cx, 38(%rdx)
; SCALAR-NEXT:    movw %r8w, 36(%rdx)
; SCALAR-NEXT:    movw %ax, 34(%rdx)
; SCALAR-NEXT:    movw %si, 32(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    popq %r12
; SCALAR-NEXT:    popq %r13
; SCALAR-NEXT:    popq %r14
; SCALAR-NEXT:    popq %r15
; SCALAR-NEXT:    popq %rbp
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v16i16:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    movdqa 16(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm1, 16(%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm1, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm1, 48(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v16i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rsi)
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec512_v16i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rsi)
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <16 x i16>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <16 x i16> %in.subvec.not, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
  store <16 x i16> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <16 x i16>, ptr %out.vec.ptr, i64 0
  store <16 x i16> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <16 x i16>, ptr %out.vec.ptr, i64 1
  store <16 x i16> %in.subvec, ptr %out.subvec1.ptr, align 32
  ret void
}

define void @vec512_v32i8(ptr %in.subvec.ptr, ptr %out.subvec.ptr, ptr %out.vec.ptr) nounwind {
; SCALAR-LABEL: vec512_v32i8:
; SCALAR:       # %bb.0:
; SCALAR-NEXT:    pushq %rbp
; SCALAR-NEXT:    pushq %r15
; SCALAR-NEXT:    pushq %r14
; SCALAR-NEXT:    pushq %r13
; SCALAR-NEXT:    pushq %r12
; SCALAR-NEXT:    pushq %rbx
; SCALAR-NEXT:    movzbl 16(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 15(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 14(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 13(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 12(%rdi), %r13d
; SCALAR-NEXT:    movzbl 11(%rdi), %eax
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 10(%rdi), %r12d
; SCALAR-NEXT:    movzbl 9(%rdi), %r15d
; SCALAR-NEXT:    movzbl 8(%rdi), %r14d
; SCALAR-NEXT:    movzbl 7(%rdi), %ebp
; SCALAR-NEXT:    movzbl 6(%rdi), %ebx
; SCALAR-NEXT:    movzbl 5(%rdi), %r11d
; SCALAR-NEXT:    movzbl 4(%rdi), %r10d
; SCALAR-NEXT:    movzbl 3(%rdi), %r9d
; SCALAR-NEXT:    movzbl 2(%rdi), %r8d
; SCALAR-NEXT:    movzbl (%rdi), %eax
; SCALAR-NEXT:    movzbl 1(%rdi), %ecx
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    movb %r8b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    movb %r9b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r10b
; SCALAR-NEXT:    movb %r10b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r11b
; SCALAR-NEXT:    movb %r11b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %bl
; SCALAR-NEXT:    movb %bl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %bpl
; SCALAR-NEXT:    movb %bpl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r14b
; SCALAR-NEXT:    movb %r14b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r15b
; SCALAR-NEXT:    movb %r15b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r12b
; SCALAR-NEXT:    movb %r12b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r11d # 1-byte Folded Reload
; SCALAR-NEXT:    notb %r11b
; SCALAR-NEXT:    movb %r11b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb %r13b
; SCALAR-NEXT:    movb %r13b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    notb {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Folded Spill
; SCALAR-NEXT:    notb {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Folded Spill
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 1-byte Folded Reload
; SCALAR-NEXT:    notb %r8b
; SCALAR-NEXT:    notb {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Folded Spill
; SCALAR-NEXT:    movzbl 17(%rdi), %eax
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 18(%rdi), %eax
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 19(%rdi), %eax
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 20(%rdi), %eax
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 21(%rdi), %ebp
; SCALAR-NEXT:    notb %bpl
; SCALAR-NEXT:    movb %bpl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 22(%rdi), %ebx
; SCALAR-NEXT:    notb %bl
; SCALAR-NEXT:    movb %bl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 23(%rdi), %r10d
; SCALAR-NEXT:    notb %r10b
; SCALAR-NEXT:    movb %r10b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 24(%rdi), %r9d
; SCALAR-NEXT:    notb %r9b
; SCALAR-NEXT:    movb %r9b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 25(%rdi), %ecx
; SCALAR-NEXT:    notb %cl
; SCALAR-NEXT:    movb %cl, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 26(%rdi), %r14d
; SCALAR-NEXT:    notb %r14b
; SCALAR-NEXT:    movb %r14b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 27(%rdi), %r15d
; SCALAR-NEXT:    notb %r15b
; SCALAR-NEXT:    movb %r15b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 28(%rdi), %r12d
; SCALAR-NEXT:    notb %r12b
; SCALAR-NEXT:    movb %r12b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 29(%rdi), %r13d
; SCALAR-NEXT:    notb %r13b
; SCALAR-NEXT:    movb %r13b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 30(%rdi), %eax
; SCALAR-NEXT:    notb %al
; SCALAR-NEXT:    movb %al, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl 31(%rdi), %edi
; SCALAR-NEXT:    notb %dil
; SCALAR-NEXT:    movb %dil, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movb %dil, 31(%rsi)
; SCALAR-NEXT:    movb %al, 30(%rsi)
; SCALAR-NEXT:    movb %r13b, 29(%rsi)
; SCALAR-NEXT:    movb %r12b, 28(%rsi)
; SCALAR-NEXT:    movb %r15b, 27(%rsi)
; SCALAR-NEXT:    movb %r14b, 26(%rsi)
; SCALAR-NEXT:    movb %cl, 25(%rsi)
; SCALAR-NEXT:    movb %r9b, 24(%rsi)
; SCALAR-NEXT:    movb %r10b, 23(%rsi)
; SCALAR-NEXT:    movb %bl, 22(%rsi)
; SCALAR-NEXT:    movb %bpl, 21(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ebp # 1-byte Folded Reload
; SCALAR-NEXT:    movb %bpl, 20(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 19(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 18(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 17(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %cl, 16(%rsi)
; SCALAR-NEXT:    movb %r8b, 15(%rsi)
; SCALAR-NEXT:    movl %r8d, %r14d
; SCALAR-NEXT:    movb %r8b, {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Spill
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ebx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %bl, 14(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 13(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 12(%rsi)
; SCALAR-NEXT:    movb %r11b, 11(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %dil, 10(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %dil, 9(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %dil, 8(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r11d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r11b, 7(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r13d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r13b, 6(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r10d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r10b, 5(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r12d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r12b, 4(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r9d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r9b, 3(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r15d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r15b, 2(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r8d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r8b, 1(%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %edi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %dil, (%rsi)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 31(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 30(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 29(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 28(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 27(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 26(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 25(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 24(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 23(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 22(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 21(%rdx)
; SCALAR-NEXT:    movb %bpl, 20(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 19(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 18(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 17(%rdx)
; SCALAR-NEXT:    movb %cl, 16(%rdx)
; SCALAR-NEXT:    movb %r14b, 15(%rdx)
; SCALAR-NEXT:    movb %bl, 14(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %cl, 13(%rdx)
; SCALAR-NEXT:    movb %al, 12(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %esi # 1-byte Folded Reload
; SCALAR-NEXT:    movb %sil, 11(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ebx # 1-byte Folded Reload
; SCALAR-NEXT:    movb %bl, 10(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r14d # 1-byte Folded Reload
; SCALAR-NEXT:    movb %r14b, 9(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ebp # 1-byte Folded Reload
; SCALAR-NEXT:    movb %bpl, 8(%rdx)
; SCALAR-NEXT:    movb %r11b, 7(%rdx)
; SCALAR-NEXT:    movb %r13b, 6(%rdx)
; SCALAR-NEXT:    movb %r10b, 5(%rdx)
; SCALAR-NEXT:    movb %r12b, 4(%rdx)
; SCALAR-NEXT:    movb %r9b, 3(%rdx)
; SCALAR-NEXT:    movb %r15b, 2(%rdx)
; SCALAR-NEXT:    movb %r8b, 1(%rdx)
; SCALAR-NEXT:    movb %dil, (%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 63(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 62(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 61(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 60(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 59(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 58(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 57(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 56(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 55(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 54(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 53(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 52(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 51(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 50(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 49(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 48(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 47(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 46(%rdx)
; SCALAR-NEXT:    movb %cl, 45(%rdx)
; SCALAR-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; SCALAR-NEXT:    movb %al, 44(%rdx)
; SCALAR-NEXT:    movb %sil, 43(%rdx)
; SCALAR-NEXT:    movb %bl, 42(%rdx)
; SCALAR-NEXT:    movb %r14b, 41(%rdx)
; SCALAR-NEXT:    movb %bpl, 40(%rdx)
; SCALAR-NEXT:    movb %r11b, 39(%rdx)
; SCALAR-NEXT:    movb %r13b, 38(%rdx)
; SCALAR-NEXT:    movb %r10b, 37(%rdx)
; SCALAR-NEXT:    movb %r12b, 36(%rdx)
; SCALAR-NEXT:    movb %r9b, 35(%rdx)
; SCALAR-NEXT:    movb %r15b, 34(%rdx)
; SCALAR-NEXT:    movb %r8b, 33(%rdx)
; SCALAR-NEXT:    movb %dil, 32(%rdx)
; SCALAR-NEXT:    popq %rbx
; SCALAR-NEXT:    popq %r12
; SCALAR-NEXT:    popq %r13
; SCALAR-NEXT:    popq %r14
; SCALAR-NEXT:    popq %r15
; SCALAR-NEXT:    popq %rbp
; SCALAR-NEXT:    retq
;
; SSE2-LABEL: vec512_v32i8:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
; SSE2-NEXT:    movdqa 16(%rdi), %xmm1
; SSE2-NEXT:    pxor %xmm0, %xmm1
; SSE2-NEXT:    pxor (%rdi), %xmm0
; SSE2-NEXT:    movdqa %xmm0, (%rsi)
; SSE2-NEXT:    movdqa %xmm1, 16(%rsi)
; SSE2-NEXT:    movdqa %xmm0, (%rdx)
; SSE2-NEXT:    movdqa %xmm1, 16(%rdx)
; SSE2-NEXT:    movdqa %xmm1, 48(%rdx)
; SSE2-NEXT:    movdqa %xmm0, 32(%rdx)
; SSE2-NEXT:    retq
;
; AVX1-LABEL: vec512_v32i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vcmptrueps %ymm0, %ymm0, %ymm0
; AVX1-NEXT:    vxorps (%rdi), %ymm0, %ymm0
; AVX1-NEXT:    vmovaps %ymm0, (%rsi)
; AVX1-NEXT:    vmovaps %ymm0, (%rdx)
; AVX1-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: vec512_v32i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    vpxor (%rdi), %ymm0, %ymm0
; AVX2-NEXT:    vmovdqa %ymm0, (%rsi)
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vmovdqa %ymm0, 32(%rdx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %in.subvec.not = load <32 x i8>, ptr %in.subvec.ptr, align 64
  %in.subvec = xor <32 x i8> %in.subvec.not, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  store <32 x i8> %in.subvec, ptr %out.subvec.ptr, align 64
  %out.subvec0.ptr = getelementptr <32 x i8>, ptr %out.vec.ptr, i64 0
  store <32 x i8> %in.subvec, ptr %out.subvec0.ptr, align 64
  %out.subvec1.ptr = getelementptr <32 x i8>, ptr %out.vec.ptr, i64 1
  store <32 x i8> %in.subvec, ptr %out.subvec1.ptr, align 32
  ret void
}
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; SSSE3: {{.*}}