llvm/test/CodeGen/X86/vector-compress.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc < %s -mtriple=x86_64 -mcpu=x86-64-v3                     | FileCheck %s --check-prefixes=CHECK,AVX2
; RUN: llc < %s -mtriple=x86_64 -mcpu=knl                           | FileCheck %s --check-prefixes=CHECK,AVX512F
; RUN: llc < %s -mtriple=x86_64 -mcpu=x86-64-v4                     | FileCheck %s --check-prefixes=CHECK,AVX512VL,AVX512VL-ONLY
; RUN: llc < %s -mtriple=x86_64 -mcpu=x86-64-v4 -mattr=+avx512vbmi2 | FileCheck %s --check-prefixes=CHECK,AVX512VL,AVX512VL-VBMI2

define <4 x i32> @test_compress_v4i32(<4 x i32> %vec, <4 x i1> %mask, <4 x i32> %passthru) nounwind {
; AVX2-LABEL: test_compress_v4i32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX2-NEXT:    vpsrad $31, %xmm1, %xmm1
; AVX2-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vpextrd $1, %xmm1, %eax
; AVX2-NEXT:    vmovd %xmm1, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    movl %esi, %edi
; AVX2-NEXT:    subl %eax, %edi
; AVX2-NEXT:    vpextrd $2, %xmm1, %edx
; AVX2-NEXT:    subl %edx, %edi
; AVX2-NEXT:    vpextrd $3, %xmm1, %ecx
; AVX2-NEXT:    subl %ecx, %edi
; AVX2-NEXT:    andl $3, %edi
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rsi, %rax
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rax, %rdx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    vextractps $3, %xmm0, %r8d
; AVX2-NEXT:    cmpq $4, %rcx
; AVX2-NEXT:    cmovbl -24(%rsp,%rdi,4), %r8d
; AVX2-NEXT:    vmovss %xmm0, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vextractps $1, %xmm0, -24(%rsp,%rsi,4)
; AVX2-NEXT:    vextractps $2, %xmm0, -24(%rsp,%rax,4)
; AVX2-NEXT:    andl $3, %edx
; AVX2-NEXT:    vextractps $3, %xmm0, -24(%rsp,%rdx,4)
; AVX2-NEXT:    cmpq $3, %rcx
; AVX2-NEXT:    movl $3, %eax
; AVX2-NEXT:    cmovbq %rcx, %rax
; AVX2-NEXT:    andl $3, %eax
; AVX2-NEXT:    movl %r8d, -24(%rsp,%rax,4)
; AVX2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v4i32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm2 killed $xmm2 def $zmm2
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
; AVX512F-NEXT:    kshiftrw $12, %k0, %k1
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v4i32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovd2m %xmm1, %k1
; AVX512VL-NEXT:    vpcompressd %xmm0, %xmm2 {%k1}
; AVX512VL-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512VL-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> %mask, <4 x i32> %passthru)
    ret <4 x i32> %out
}

define <4 x float> @test_compress_v4f32(<4 x float> %vec, <4 x i1> %mask, <4 x float> %passthru) nounwind {
; AVX2-LABEL: test_compress_v4f32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX2-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vpsrad $31, %xmm1, %xmm1
; AVX2-NEXT:    vpextrd $1, %xmm1, %edx
; AVX2-NEXT:    vmovd %xmm1, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    movl %esi, %edi
; AVX2-NEXT:    subl %edx, %edi
; AVX2-NEXT:    vpextrd $2, %xmm1, %ecx
; AVX2-NEXT:    subl %ecx, %edi
; AVX2-NEXT:    vpextrd $3, %xmm1, %eax
; AVX2-NEXT:    subl %eax, %edi
; AVX2-NEXT:    andl $3, %edi
; AVX2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX2-NEXT:    vmovss %xmm0, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vextractps $1, %xmm0, -24(%rsp,%rsi,4)
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rsi, %rdx
; AVX2-NEXT:    vextractps $2, %xmm0, -24(%rsp,%rdx,4)
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
; AVX2-NEXT:    vmovss %xmm0, -24(%rsp,%rcx,4)
; AVX2-NEXT:    cmpq $4, %rax
; AVX2-NEXT:    jae .LBB1_2
; AVX2-NEXT:  # %bb.1:
; AVX2-NEXT:    vmovaps %xmm1, %xmm0
; AVX2-NEXT:  .LBB1_2:
; AVX2-NEXT:    cmpq $3, %rax
; AVX2-NEXT:    movl $3, %ecx
; AVX2-NEXT:    cmovbq %rax, %rcx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    vmovss %xmm0, -24(%rsp,%rcx,4)
; AVX2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v4f32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm2 killed $xmm2 def $zmm2
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
; AVX512F-NEXT:    kshiftrw $12, %k0, %k1
; AVX512F-NEXT:    vcompressps %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v4f32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovd2m %xmm1, %k1
; AVX512VL-NEXT:    vcompressps %xmm0, %xmm2 {%k1}
; AVX512VL-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512VL-NEXT:    retq
    %out = call <4 x float> @llvm.experimental.vector.compress(<4 x float> %vec, <4 x i1> %mask, <4 x float> %passthru)
    ret <4 x float> %out
}

define <2 x i64> @test_compress_v2i64(<2 x i64> %vec, <2 x i1> %mask, <2 x i64> %passthru) nounwind {
; AVX2-LABEL: test_compress_v2i64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpsllq $63, %xmm1, %xmm1
; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
; AVX2-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vpextrq $1, %xmm1, %rax
; AVX2-NEXT:    vmovq %xmm1, %rcx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    movq %rcx, %rdx
; AVX2-NEXT:    subq %rax, %rdx
; AVX2-NEXT:    movl %edx, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpextrq $1, %xmm0, %rsi
; AVX2-NEXT:    cmpq $2, %rdx
; AVX2-NEXT:    cmovbq -24(%rsp,%rax,8), %rsi
; AVX2-NEXT:    vmovq %xmm0, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    vpextrq $1, %xmm0, -24(%rsp,%rax,8)
; AVX2-NEXT:    cmpq $1, %rdx
; AVX2-NEXT:    movl $1, %eax
; AVX2-NEXT:    cmovbq %rdx, %rax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    movq %rsi, -24(%rsp,%rax,8)
; AVX2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v2i64:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm2 killed $xmm2 def $zmm2
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vpsllq $63, %xmm1, %xmm1
; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kshiftlw $14, %k0, %k0
; AVX512F-NEXT:    kshiftrw $14, %k0, %k1
; AVX512F-NEXT:    vpcompressq %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v2i64:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpsllq $63, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovq2m %xmm1, %k1
; AVX512VL-NEXT:    vpcompressq %xmm0, %xmm2 {%k1}
; AVX512VL-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512VL-NEXT:    retq
    %out = call <2 x i64> @llvm.experimental.vector.compress(<2 x i64> %vec, <2 x i1> %mask, <2 x i64> %passthru)
    ret <2 x i64> %out
}

define <2 x double> @test_compress_v2f64(<2 x double> %vec, <2 x i1> %mask, <2 x double> %passthru) nounwind {
; AVX2-LABEL: test_compress_v2f64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpsllq $63, %xmm1, %xmm1
; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
; AVX2-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vpextrq $1, %xmm1, %rcx
; AVX2-NEXT:    vmovq %xmm1, %rdx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    movq %rdx, %rax
; AVX2-NEXT:    subq %rcx, %rax
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
; AVX2-NEXT:    vmovlpd %xmm0, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    movl %edx, %ecx
; AVX2-NEXT:    vmovhpd %xmm0, -24(%rsp,%rcx,8)
; AVX2-NEXT:    cmpq $2, %rax
; AVX2-NEXT:    jb .LBB3_2
; AVX2-NEXT:  # %bb.1:
; AVX2-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX2-NEXT:  .LBB3_2:
; AVX2-NEXT:    cmpq $1, %rax
; AVX2-NEXT:    movl $1, %ecx
; AVX2-NEXT:    cmovbq %rax, %rcx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vmovsd %xmm1, -24(%rsp,%rcx,8)
; AVX2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v2f64:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm2 killed $xmm2 def $zmm2
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vpsllq $63, %xmm1, %xmm1
; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kshiftlw $14, %k0, %k0
; AVX512F-NEXT:    kshiftrw $14, %k0, %k1
; AVX512F-NEXT:    vcompresspd %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v2f64:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpsllq $63, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovq2m %xmm1, %k1
; AVX512VL-NEXT:    vcompresspd %xmm0, %xmm2 {%k1}
; AVX512VL-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512VL-NEXT:    retq
    %out = call <2 x double> @llvm.experimental.vector.compress(<2 x double> %vec, <2 x i1> %mask, <2 x double> %passthru)
    ret <2 x double> %out
}

define <8 x i32> @test_compress_v8i32(<8 x i32> %vec, <8 x i1> %mask, <8 x i32> %passthru) nounwind {
; AVX2-LABEL: test_compress_v8i32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $32, %rsp
; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX2-NEXT:    vpslld $31, %ymm1, %ymm1
; AVX2-NEXT:    vmovaps %ymm2, (%rsp)
; AVX2-NEXT:    vpsrad $31, %ymm1, %ymm2
; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm1
; AVX2-NEXT:    vpackssdw %xmm1, %xmm2, %xmm3
; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
; AVX2-NEXT:    vpslld $31, %ymm3, %ymm3
; AVX2-NEXT:    vpsrld $31, %ymm3, %ymm3
; AVX2-NEXT:    vextracti128 $1, %ymm3, %xmm4
; AVX2-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[2,3,2,3]
; AVX2-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpextrd $1, %xmm3, %ecx
; AVX2-NEXT:    vmovd %xmm3, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    vpextrd $1, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vmovd %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    vpextrd $2, %xmm2, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    addq %rcx, %rsi
; AVX2-NEXT:    vpextrd $3, %xmm2, %edi
; AVX2-NEXT:    andl $1, %edi
; AVX2-NEXT:    addq %rsi, %rdi
; AVX2-NEXT:    vmovd %xmm1, %r8d
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addq %rdi, %r8
; AVX2-NEXT:    vpextrd $1, %xmm1, %r9d
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addq %r8, %r9
; AVX2-NEXT:    vpextrd $2, %xmm1, %r10d
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    addq %r9, %r10
; AVX2-NEXT:    vpextrd $3, %xmm1, %r11d
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    addq %r10, %r11
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vextractps $3, %xmm1, %ebx
; AVX2-NEXT:    cmpq $8, %r11
; AVX2-NEXT:    cmovbl (%rsp,%rax,4), %ebx
; AVX2-NEXT:    vmovss %xmm0, (%rsp)
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rdx,4)
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rsi,4)
; AVX2-NEXT:    andl $7, %edi
; AVX2-NEXT:    vmovss %xmm1, (%rsp,%rdi,4)
; AVX2-NEXT:    andl $7, %r8d
; AVX2-NEXT:    vextractps $1, %xmm1, (%rsp,%r8,4)
; AVX2-NEXT:    andl $7, %r9d
; AVX2-NEXT:    vextractps $2, %xmm1, (%rsp,%r9,4)
; AVX2-NEXT:    andl $7, %r10d
; AVX2-NEXT:    vextractps $3, %xmm1, (%rsp,%r10,4)
; AVX2-NEXT:    cmpq $7, %r11
; AVX2-NEXT:    movl $7, %eax
; AVX2-NEXT:    cmovbq %r11, %rax
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    movl %ebx, (%rsp,%rax,4)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    leaq -8(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v8i32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $ymm2 killed $ymm2 def $zmm2
; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
; AVX512F-NEXT:    vpmovsxwq %xmm1, %zmm1
; AVX512F-NEXT:    vpsllq $63, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa %ymm2, %ymm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v8i32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpsllw $15, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovw2m %xmm1, %k1
; AVX512VL-NEXT:    vpcompressd %ymm0, %ymm2 {%k1}
; AVX512VL-NEXT:    vmovdqa %ymm2, %ymm0
; AVX512VL-NEXT:    retq
    %out = call <8 x i32> @llvm.experimental.vector.compress(<8 x i32> %vec, <8 x i1> %mask, <8 x i32> %passthru)
    ret <8 x i32> %out
}

define <8 x float> @test_compress_v8f32(<8 x float> %vec, <8 x i1> %mask, <8 x float> %passthru) nounwind {
; AVX2-LABEL: test_compress_v8f32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $32, %rsp
; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX2-NEXT:    vpslld $31, %ymm1, %ymm1
; AVX2-NEXT:    vmovaps %ymm2, (%rsp)
; AVX2-NEXT:    vpsrad $31, %ymm1, %ymm3
; AVX2-NEXT:    vextracti128 $1, %ymm3, %xmm1
; AVX2-NEXT:    vpackssdw %xmm1, %xmm3, %xmm2
; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; AVX2-NEXT:    vpslld $31, %ymm2, %ymm2
; AVX2-NEXT:    vpsrld $31, %ymm2, %ymm2
; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm4
; AVX2-NEXT:    vpaddd %xmm4, %xmm2, %xmm2
; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm2[2,3,2,3]
; AVX2-NEXT:    vpaddd %xmm4, %xmm2, %xmm2
; AVX2-NEXT:    vpextrd $1, %xmm2, %eax
; AVX2-NEXT:    vmovd %xmm2, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    andl $7, %ecx
; AVX2-NEXT:    vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX2-NEXT:    vmovss %xmm0, (%rsp)
; AVX2-NEXT:    vmovd %xmm3, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    vpextrd $1, %xmm3, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    vpextrd $2, %xmm3, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    vpextrd $3, %xmm3, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vmovd %xmm1, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $7, %ecx
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    vpextrd $1, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    vpextrd $2, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    vpextrd $3, %xmm1, %eax
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rdx, %rax
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $7, %edx
; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rdx,4)
; AVX2-NEXT:    cmpq $8, %rax
; AVX2-NEXT:    jae .LBB5_2
; AVX2-NEXT:  # %bb.1:
; AVX2-NEXT:    vmovaps %xmm2, %xmm0
; AVX2-NEXT:  .LBB5_2:
; AVX2-NEXT:    cmpq $7, %rax
; AVX2-NEXT:    movl $7, %ecx
; AVX2-NEXT:    cmovbq %rax, %rcx
; AVX2-NEXT:    andl $7, %ecx
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    movq %rbp, %rsp
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v8f32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $ymm2 killed $ymm2 def $zmm2
; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
; AVX512F-NEXT:    vpmovsxwq %xmm1, %zmm1
; AVX512F-NEXT:    vpsllq $63, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vcompressps %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa %ymm2, %ymm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v8f32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpsllw $15, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovw2m %xmm1, %k1
; AVX512VL-NEXT:    vcompressps %ymm0, %ymm2 {%k1}
; AVX512VL-NEXT:    vmovdqa %ymm2, %ymm0
; AVX512VL-NEXT:    retq
    %out = call <8 x float> @llvm.experimental.vector.compress(<8 x float> %vec, <8 x i1> %mask, <8 x float> %passthru)
    ret <8 x float> %out
}

define <4 x i64> @test_compress_v4i64(<4 x i64> %vec, <4 x i1> %mask, <4 x i64> %passthru) nounwind {
; AVX2-LABEL: test_compress_v4i64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $32, %rsp
; AVX2-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX2-NEXT:    vpsrad $31, %xmm1, %xmm1
; AVX2-NEXT:    vpmovsxdq %xmm1, %ymm1
; AVX2-NEXT:    vmovaps %ymm2, (%rsp)
; AVX2-NEXT:    vpsrlq $63, %ymm1, %ymm2
; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
; AVX2-NEXT:    vpaddq %xmm3, %xmm2, %xmm2
; AVX2-NEXT:    vpextrq $1, %xmm2, %rax
; AVX2-NEXT:    vmovd %xmm2, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    vpextrq $1, %xmm1, %rax
; AVX2-NEXT:    vmovq %xmm1, %rdx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    movq %rdx, %rsi
; AVX2-NEXT:    subq %rax, %rsi
; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm1
; AVX2-NEXT:    vmovq %xmm1, %rax
; AVX2-NEXT:    movq %rsi, %rdi
; AVX2-NEXT:    subq %rax, %rdi
; AVX2-NEXT:    vpextrq $1, %xmm1, %rax
; AVX2-NEXT:    movq %rdi, %r8
; AVX2-NEXT:    subq %rax, %r8
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vpextrq $1, %xmm1, %rax
; AVX2-NEXT:    cmpq $4, %r8
; AVX2-NEXT:    cmovbq (%rsp,%rcx,8), %rax
; AVX2-NEXT:    vmovq %xmm0, (%rsp)
; AVX2-NEXT:    movl %edx, %ecx
; AVX2-NEXT:    vpextrq $1, %xmm0, (%rsp,%rcx,8)
; AVX2-NEXT:    andl $3, %esi
; AVX2-NEXT:    vmovq %xmm1, (%rsp,%rsi,8)
; AVX2-NEXT:    andl $3, %edi
; AVX2-NEXT:    vpextrq $1, %xmm1, (%rsp,%rdi,8)
; AVX2-NEXT:    cmpq $3, %r8
; AVX2-NEXT:    movl $3, %ecx
; AVX2-NEXT:    cmovbq %r8, %rcx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    movq %rax, (%rsp,%rcx,8)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    movq %rbp, %rsp
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v4i64:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $ymm2 killed $ymm2 def $zmm2
; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
; AVX512F-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
; AVX512F-NEXT:    kshiftrw $12, %k0, %k1
; AVX512F-NEXT:    vpcompressq %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa %ymm2, %ymm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v4i64:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovd2m %xmm1, %k1
; AVX512VL-NEXT:    vpcompressq %ymm0, %ymm2 {%k1}
; AVX512VL-NEXT:    vmovdqa %ymm2, %ymm0
; AVX512VL-NEXT:    retq
    %out = call <4 x i64> @llvm.experimental.vector.compress(<4 x i64> %vec, <4 x i1> %mask, <4 x i64> %passthru)
    ret <4 x i64> %out
}

define <4 x double> @test_compress_v4f64(<4 x double> %vec, <4 x i1> %mask, <4 x double> %passthru) nounwind {
; AVX2-LABEL: test_compress_v4f64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $32, %rsp
; AVX2-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX2-NEXT:    vpsrad $31, %xmm1, %xmm1
; AVX2-NEXT:    vmovaps %ymm2, (%rsp)
; AVX2-NEXT:    vpmovsxdq %xmm1, %ymm2
; AVX2-NEXT:    vpsrlq $63, %ymm2, %ymm1
; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm3
; AVX2-NEXT:    vpaddq %xmm3, %xmm1, %xmm1
; AVX2-NEXT:    vpextrq $1, %xmm1, %rax
; AVX2-NEXT:    vmovd %xmm1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
; AVX2-NEXT:    vmovlpd %xmm0, (%rsp)
; AVX2-NEXT:    vmovq %xmm2, %rax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    vmovhpd %xmm0, (%rsp,%rcx,8)
; AVX2-NEXT:    vpextrq $1, %xmm2, %rcx
; AVX2-NEXT:    subq %rcx, %rax
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovlpd %xmm0, (%rsp,%rcx,8)
; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm2
; AVX2-NEXT:    vmovq %xmm2, %rcx
; AVX2-NEXT:    subq %rcx, %rax
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    vmovhpd %xmm0, (%rsp,%rcx,8)
; AVX2-NEXT:    vpextrq $1, %xmm2, %rcx
; AVX2-NEXT:    subq %rcx, %rax
; AVX2-NEXT:    cmpq $4, %rax
; AVX2-NEXT:    jb .LBB7_2
; AVX2-NEXT:  # %bb.1:
; AVX2-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX2-NEXT:  .LBB7_2:
; AVX2-NEXT:    cmpq $3, %rax
; AVX2-NEXT:    movl $3, %ecx
; AVX2-NEXT:    cmovbq %rax, %rcx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    vmovsd %xmm1, (%rsp,%rcx,8)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    movq %rbp, %rsp
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v4f64:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $ymm2 killed $ymm2 def $zmm2
; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
; AVX512F-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
; AVX512F-NEXT:    kshiftrw $12, %k0, %k1
; AVX512F-NEXT:    vcompresspd %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa %ymm2, %ymm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v4f64:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovd2m %xmm1, %k1
; AVX512VL-NEXT:    vcompresspd %ymm0, %ymm2 {%k1}
; AVX512VL-NEXT:    vmovdqa %ymm2, %ymm0
; AVX512VL-NEXT:    retq
    %out = call <4 x double> @llvm.experimental.vector.compress(<4 x double> %vec, <4 x i1> %mask, <4 x double> %passthru)
    ret <4 x double> %out
}

define <16 x i32> @test_compress_v16i32(<16 x i32> %vec, <16 x i1> %mask, <16 x i32> %passthru) nounwind {
; AVX2-LABEL: test_compress_v16i32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %r13
; AVX2-NEXT:    pushq %r12
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $160, %rsp
; AVX2-NEXT:    vmovaps %ymm4, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovaps %ymm3, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm4 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero
; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,2,3]
; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero
; AVX2-NEXT:    vpaddd %ymm3, %ymm4, %ymm3
; AVX2-NEXT:    vextracti128 $1, %ymm3, %xmm4
; AVX2-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[2,3,2,3]
; AVX2-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpextrd $1, %xmm3, %eax
; AVX2-NEXT:    vmovd %xmm3, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    vpextrb $4, %xmm2, %edx
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    vpextrb $1, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vmovd %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $2, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $3, %xmm2, %r9d
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addq %rcx, %r9
; AVX2-NEXT:    addq %r9, %rdx
; AVX2-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $5, %xmm2, %edi
; AVX2-NEXT:    andl $1, %edi
; AVX2-NEXT:    addq %rdx, %rdi
; AVX2-NEXT:    vpextrb $6, %xmm2, %r10d
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    addq %rdi, %r10
; AVX2-NEXT:    vpextrb $7, %xmm2, %r11d
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    addq %r10, %r11
; AVX2-NEXT:    vpextrb $8, %xmm2, %ebx
; AVX2-NEXT:    andl $1, %ebx
; AVX2-NEXT:    addq %r11, %rbx
; AVX2-NEXT:    vpextrb $9, %xmm2, %r14d
; AVX2-NEXT:    andl $1, %r14d
; AVX2-NEXT:    addq %rbx, %r14
; AVX2-NEXT:    vpextrb $10, %xmm2, %r15d
; AVX2-NEXT:    andl $1, %r15d
; AVX2-NEXT:    addq %r14, %r15
; AVX2-NEXT:    vpextrb $11, %xmm2, %r12d
; AVX2-NEXT:    andl $1, %r12d
; AVX2-NEXT:    addq %r15, %r12
; AVX2-NEXT:    vpextrb $12, %xmm2, %r13d
; AVX2-NEXT:    andl $1, %r13d
; AVX2-NEXT:    addq %r12, %r13
; AVX2-NEXT:    vpextrb $13, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %r13, %rdx
; AVX2-NEXT:    vpextrb $14, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    vpextrb $15, %xmm2, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    addq %rcx, %rsi
; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
; AVX2-NEXT:    vextractps $3, %xmm2, %eax
; AVX2-NEXT:    cmpq $16, %rsi
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
; AVX2-NEXT:    cmovbl 32(%rsp,%r8,4), %eax
; AVX2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
; AVX2-NEXT:    vextractps $1, %xmm0, 32(%rsp,%r8,4)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
; AVX2-NEXT:    vextractps $2, %xmm0, 32(%rsp,%r8,4)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
; AVX2-NEXT:    vextractps $3, %xmm0, 32(%rsp,%r8,4)
; AVX2-NEXT:    andl $15, %r9d
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovss %xmm0, 32(%rsp,%r9,4)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, 32(%rsp,%rax,4)
; AVX2-NEXT:    andl $15, %edi
; AVX2-NEXT:    vextractps $2, %xmm0, 32(%rsp,%rdi,4)
; AVX2-NEXT:    andl $15, %r10d
; AVX2-NEXT:    vextractps $3, %xmm0, 32(%rsp,%r10,4)
; AVX2-NEXT:    andl $15, %r11d
; AVX2-NEXT:    vmovss %xmm1, 32(%rsp,%r11,4)
; AVX2-NEXT:    andl $15, %ebx
; AVX2-NEXT:    vextractps $1, %xmm1, 32(%rsp,%rbx,4)
; AVX2-NEXT:    andl $15, %r14d
; AVX2-NEXT:    vextractps $2, %xmm1, 32(%rsp,%r14,4)
; AVX2-NEXT:    andl $15, %r15d
; AVX2-NEXT:    vextractps $3, %xmm1, 32(%rsp,%r15,4)
; AVX2-NEXT:    andl $15, %r12d
; AVX2-NEXT:    vmovss %xmm2, 32(%rsp,%r12,4)
; AVX2-NEXT:    andl $15, %r13d
; AVX2-NEXT:    vextractps $1, %xmm2, 32(%rsp,%r13,4)
; AVX2-NEXT:    andl $15, %edx
; AVX2-NEXT:    vextractps $2, %xmm2, 32(%rsp,%rdx,4)
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vextractps $3, %xmm2, 32(%rsp,%rcx,4)
; AVX2-NEXT:    cmpq $15, %rsi
; AVX2-NEXT:    movl $15, %eax
; AVX2-NEXT:    cmovbq %rsi, %rax
; AVX2-NEXT:    shll $2, %eax
; AVX2-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 4-byte Reload
; AVX2-NEXT:    movl %ecx, 32(%rsp,%rax)
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm0
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm1
; AVX2-NEXT:    leaq -40(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r12
; AVX2-NEXT:    popq %r13
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v16i32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
; AVX512F-NEXT:    vpslld $31, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa64 %zmm2, %zmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v16i32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpsllw $7, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovb2m %xmm1, %k1
; AVX512VL-NEXT:    vpcompressd %zmm0, %zmm2 {%k1}
; AVX512VL-NEXT:    vmovdqa64 %zmm2, %zmm0
; AVX512VL-NEXT:    retq
    %out = call <16 x i32> @llvm.experimental.vector.compress(<16 x i32> %vec, <16 x i1> %mask, <16 x i32> %passthru)
    ret <16 x i32> %out
}

define <16 x float> @test_compress_v16f32(<16 x float> %vec, <16 x i1> %mask, <16 x float> %passthru) nounwind {
; AVX2-LABEL: test_compress_v16f32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $64, %rsp
; AVX2-NEXT:    vmovaps %ymm4, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovaps %ymm3, (%rsp)
; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm4 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero
; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,2,3]
; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero
; AVX2-NEXT:    vpaddd %ymm3, %ymm4, %ymm3
; AVX2-NEXT:    vextracti128 $1, %ymm3, %xmm4
; AVX2-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[2,3,2,3]
; AVX2-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpextrd $1, %xmm3, %eax
; AVX2-NEXT:    vmovd %xmm3, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; AVX2-NEXT:    vmovss %xmm0, (%rsp)
; AVX2-NEXT:    vmovd %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    vpextrb $1, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    vpextrb $2, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    vpextrb $3, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vpextrb $4, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    vpextrb $5, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    vpextrb $6, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    vpextrb $7, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vpextrb $8, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vmovss %xmm1, (%rsp,%rcx,4)
; AVX2-NEXT:    vpextrb $9, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vextractps $1, %xmm1, (%rsp,%rax,4)
; AVX2-NEXT:    vpextrb $10, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vextractps $2, %xmm1, (%rsp,%rcx,4)
; AVX2-NEXT:    vpextrb $11, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vextractps $3, %xmm1, (%rsp,%rax,4)
; AVX2-NEXT:    vpextrb $12, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    vpextrb $13, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rdx, %rax
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $15, %edx
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rdx,4)
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    vpextrb $14, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    vpextrb $15, %xmm2, %ecx
; AVX2-NEXT:    addq %rax, %rdx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $15, %edx
; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rdx,4)
; AVX2-NEXT:    cmpq $16, %rcx
; AVX2-NEXT:    jae .LBB9_2
; AVX2-NEXT:  # %bb.1:
; AVX2-NEXT:    vmovaps %xmm3, %xmm0
; AVX2-NEXT:  .LBB9_2:
; AVX2-NEXT:    cmpq $15, %rcx
; AVX2-NEXT:    movl $15, %eax
; AVX2-NEXT:    cmovbq %rcx, %rax
; AVX2-NEXT:    shll $2, %eax
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rax)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm1
; AVX2-NEXT:    movq %rbp, %rsp
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v16f32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
; AVX512F-NEXT:    vpslld $31, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vcompressps %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa64 %zmm2, %zmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v16f32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpsllw $7, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovb2m %xmm1, %k1
; AVX512VL-NEXT:    vcompressps %zmm0, %zmm2 {%k1}
; AVX512VL-NEXT:    vmovdqa64 %zmm2, %zmm0
; AVX512VL-NEXT:    retq
    %out = call <16 x float> @llvm.experimental.vector.compress(<16 x float> %vec, <16 x i1> %mask, <16 x float> %passthru)
    ret <16 x float> %out
}

define <8 x i64> @test_compress_v8i64(<8 x i64> %vec, <8 x i1> %mask, <8 x i64> %passthru) nounwind {
; AVX2-LABEL: test_compress_v8i64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $64, %rsp
; AVX2-NEXT:    vmovaps %ymm4, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovaps %ymm3, (%rsp)
; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm4 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero
; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,2,3]
; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero
; AVX2-NEXT:    vpaddq %ymm3, %ymm4, %ymm3
; AVX2-NEXT:    vextracti128 $1, %ymm3, %xmm4
; AVX2-NEXT:    vpaddq %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpextrq $1, %xmm3, %rax
; AVX2-NEXT:    vmovd %xmm3, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vpextrw $4, %xmm2, %eax
; AVX2-NEXT:    andl $7, %ecx
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpextrw $1, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    vmovd %xmm2, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    addq %rsi, %rdx
; AVX2-NEXT:    vpextrw $2, %xmm2, %r8d
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addq %rdx, %r8
; AVX2-NEXT:    vpextrw $3, %xmm2, %r9d
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addq %r8, %r9
; AVX2-NEXT:    addq %r9, %rax
; AVX2-NEXT:    vpextrw $5, %xmm2, %edi
; AVX2-NEXT:    andl $1, %edi
; AVX2-NEXT:    addq %rax, %rdi
; AVX2-NEXT:    vpextrw $6, %xmm2, %r10d
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    addq %rdi, %r10
; AVX2-NEXT:    vpextrw $7, %xmm2, %r11d
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    addq %r10, %r11
; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
; AVX2-NEXT:    vpextrq $1, %xmm2, %rbx
; AVX2-NEXT:    cmpq $8, %r11
; AVX2-NEXT:    cmovbq (%rsp,%rcx,8), %rbx
; AVX2-NEXT:    vmovq %xmm0, (%rsp)
; AVX2-NEXT:    vpextrq $1, %xmm0, (%rsp,%rsi,8)
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovq %xmm0, (%rsp,%rdx,8)
; AVX2-NEXT:    vpextrq $1, %xmm0, (%rsp,%r8,8)
; AVX2-NEXT:    andl $7, %r9d
; AVX2-NEXT:    vmovq %xmm1, (%rsp,%r9,8)
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    vpextrq $1, %xmm1, (%rsp,%rax,8)
; AVX2-NEXT:    andl $7, %edi
; AVX2-NEXT:    vmovq %xmm2, (%rsp,%rdi,8)
; AVX2-NEXT:    andl $7, %r10d
; AVX2-NEXT:    vpextrq $1, %xmm2, (%rsp,%r10,8)
; AVX2-NEXT:    cmpq $7, %r11
; AVX2-NEXT:    movl $7, %eax
; AVX2-NEXT:    cmovbq %r11, %rax
; AVX2-NEXT:    shll $3, %eax
; AVX2-NEXT:    movq %rbx, (%rsp,%rax)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm1
; AVX2-NEXT:    leaq -8(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v8i64:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpmovsxwq %xmm1, %zmm1
; AVX512F-NEXT:    vpsllq $63, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vpcompressq %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa64 %zmm2, %zmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v8i64:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpsllw $15, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovw2m %xmm1, %k1
; AVX512VL-NEXT:    vpcompressq %zmm0, %zmm2 {%k1}
; AVX512VL-NEXT:    vmovdqa64 %zmm2, %zmm0
; AVX512VL-NEXT:    retq
    %out = call <8 x i64> @llvm.experimental.vector.compress(<8 x i64> %vec, <8 x i1> %mask, <8 x i64> %passthru)
    ret <8 x i64> %out
}

define <8 x double> @test_compress_v8f64(<8 x double> %vec, <8 x i1> %mask, <8 x double> %passthru) nounwind {
; AVX2-LABEL: test_compress_v8f64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $64, %rsp
; AVX2-NEXT:    vmovaps %ymm4, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovaps %ymm3, (%rsp)
; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm4 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero
; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,2,3]
; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero
; AVX2-NEXT:    vpaddq %ymm3, %ymm4, %ymm3
; AVX2-NEXT:    vextracti128 $1, %ymm3, %xmm4
; AVX2-NEXT:    vpaddq %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpextrq $1, %xmm3, %rax
; AVX2-NEXT:    vmovd %xmm3, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    andl $7, %ecx
; AVX2-NEXT:    vmovsd {{.*#+}} xmm3 = mem[0],zero
; AVX2-NEXT:    vmovlps %xmm0, (%rsp)
; AVX2-NEXT:    vmovd %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vmovhps %xmm0, (%rsp,%rax,8)
; AVX2-NEXT:    vpextrw $1, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovlps %xmm0, (%rsp,%rcx,8)
; AVX2-NEXT:    vpextrw $2, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    vmovhps %xmm0, (%rsp,%rax,8)
; AVX2-NEXT:    vpextrw $3, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    vmovlpd %xmm1, (%rsp,%rax,8)
; AVX2-NEXT:    vpextrw $4, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    vpextrw $5, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    vmovhpd %xmm1, (%rsp,%rax,8)
; AVX2-NEXT:    vpextrw $6, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $7, %ecx
; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm0
; AVX2-NEXT:    vmovlpd %xmm0, (%rsp,%rcx,8)
; AVX2-NEXT:    vpextrw $7, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rdx, %rax
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $7, %edx
; AVX2-NEXT:    vmovhpd %xmm0, (%rsp,%rdx,8)
; AVX2-NEXT:    cmpq $8, %rax
; AVX2-NEXT:    jb .LBB11_2
; AVX2-NEXT:  # %bb.1:
; AVX2-NEXT:    vshufpd {{.*#+}} xmm3 = xmm0[1,0]
; AVX2-NEXT:  .LBB11_2:
; AVX2-NEXT:    cmpq $7, %rax
; AVX2-NEXT:    movl $7, %ecx
; AVX2-NEXT:    cmovbq %rax, %rcx
; AVX2-NEXT:    shll $3, %ecx
; AVX2-NEXT:    vmovsd %xmm3, (%rsp,%rcx)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm1
; AVX2-NEXT:    movq %rbp, %rsp
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v8f64:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpmovsxwq %xmm1, %zmm1
; AVX512F-NEXT:    vpsllq $63, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vcompresspd %zmm0, %zmm2 {%k1}
; AVX512F-NEXT:    vmovdqa64 %zmm2, %zmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v8f64:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpsllw $15, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovw2m %xmm1, %k1
; AVX512VL-NEXT:    vcompresspd %zmm0, %zmm2 {%k1}
; AVX512VL-NEXT:    vmovdqa64 %zmm2, %zmm0
; AVX512VL-NEXT:    retq
    %out = call <8 x double> @llvm.experimental.vector.compress(<8 x double> %vec, <8 x i1> %mask, <8 x double> %passthru)
    ret <8 x double> %out
}

define <16 x i8> @test_compress_v16i8(<16 x i8> %vec, <16 x i1> %mask, <16 x i8> %passthru) nounwind {
; AVX2-LABEL: test_compress_v16i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %r13
; AVX2-NEXT:    pushq %r12
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    vpsllw $7, %xmm1, %xmm1
; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-NEXT:    vpcmpgtb %xmm1, %xmm3, %xmm1
; AVX2-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vpextrb $1, %xmm1, %r13d
; AVX2-NEXT:    vmovd %xmm1, %edi
; AVX2-NEXT:    movl %edi, %eax
; AVX2-NEXT:    andb $1, %al
; AVX2-NEXT:    vpextrb $2, %xmm1, %esi
; AVX2-NEXT:    subb %r13b, %al
; AVX2-NEXT:    subb %sil, %al
; AVX2-NEXT:    vpextrb $3, %xmm1, %ebp
; AVX2-NEXT:    subb %bpl, %al
; AVX2-NEXT:    vpextrb $4, %xmm1, %r12d
; AVX2-NEXT:    subb %r12b, %al
; AVX2-NEXT:    vpextrb $5, %xmm1, %r15d
; AVX2-NEXT:    vpextrb $6, %xmm1, %r14d
; AVX2-NEXT:    subb %r15b, %al
; AVX2-NEXT:    subb %r14b, %al
; AVX2-NEXT:    vpextrb $7, %xmm1, %ebx
; AVX2-NEXT:    subb %bl, %al
; AVX2-NEXT:    vpextrb $8, %xmm1, %r11d
; AVX2-NEXT:    subb %r11b, %al
; AVX2-NEXT:    vpextrb $9, %xmm1, %r10d
; AVX2-NEXT:    vpextrb $10, %xmm1, %r9d
; AVX2-NEXT:    subb %r10b, %al
; AVX2-NEXT:    subb %r9b, %al
; AVX2-NEXT:    vpextrb $11, %xmm1, %r8d
; AVX2-NEXT:    subb %r8b, %al
; AVX2-NEXT:    vpextrb $12, %xmm1, %ecx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    subb %cl, %al
; AVX2-NEXT:    vpextrb $13, %xmm1, %edx
; AVX2-NEXT:    vpextrb $14, %xmm1, %ecx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    subb %dl, %al
; AVX2-NEXT:    subb %cl, %al
; AVX2-NEXT:    vpextrb $15, %xmm1, %ecx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    subb %cl, %al
; AVX2-NEXT:    movzbl %al, %eax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    movzbl -56(%rsp,%rax), %eax
; AVX2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX2-NEXT:    vpextrb $0, %xmm0, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    andl $1, %edi
; AVX2-NEXT:    vpextrb $1, %xmm0, -56(%rsp,%rdi)
; AVX2-NEXT:    andl $1, %r13d
; AVX2-NEXT:    addq %rdi, %r13
; AVX2-NEXT:    vpextrb $2, %xmm0, -56(%rsp,%r13)
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    addq %r13, %rsi
; AVX2-NEXT:    vpextrb $3, %xmm0, -56(%rsp,%rsi)
; AVX2-NEXT:    andl $1, %ebp
; AVX2-NEXT:    addq %rsi, %rbp
; AVX2-NEXT:    vpextrb $4, %xmm0, -56(%rsp,%rbp)
; AVX2-NEXT:    andl $1, %r12d
; AVX2-NEXT:    addq %rbp, %r12
; AVX2-NEXT:    movl %r12d, %eax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vpextrb $5, %xmm0, -56(%rsp,%rax)
; AVX2-NEXT:    andl $1, %r15d
; AVX2-NEXT:    addq %r12, %r15
; AVX2-NEXT:    andl $1, %r14d
; AVX2-NEXT:    addq %r15, %r14
; AVX2-NEXT:    # kill: def $r15d killed $r15d killed $r15 def $r15
; AVX2-NEXT:    andl $15, %r15d
; AVX2-NEXT:    vpextrb $6, %xmm0, -56(%rsp,%r15)
; AVX2-NEXT:    movl %r14d, %eax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vpextrb $7, %xmm0, -56(%rsp,%rax)
; AVX2-NEXT:    andl $1, %ebx
; AVX2-NEXT:    addq %r14, %rbx
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    addq %rbx, %r11
; AVX2-NEXT:    # kill: def $ebx killed $ebx killed $rbx def $rbx
; AVX2-NEXT:    andl $15, %ebx
; AVX2-NEXT:    vpextrb $8, %xmm0, -56(%rsp,%rbx)
; AVX2-NEXT:    movl %r11d, %eax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vpextrb $9, %xmm0, -56(%rsp,%rax)
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    addq %r11, %r10
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addq %r10, %r9
; AVX2-NEXT:    # kill: def $r10d killed $r10d killed $r10 def $r10
; AVX2-NEXT:    andl $15, %r10d
; AVX2-NEXT:    vpextrb $10, %xmm0, -56(%rsp,%r10)
; AVX2-NEXT:    movl %r9d, %eax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vpextrb $11, %xmm0, -56(%rsp,%rax)
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addq %r9, %r8
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %r8, %rcx
; AVX2-NEXT:    # kill: def $r8d killed $r8d killed $r8 def $r8
; AVX2-NEXT:    andl $15, %r8d
; AVX2-NEXT:    vpextrb $12, %xmm0, -56(%rsp,%r8)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vpextrb $13, %xmm0, -56(%rsp,%rax)
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $15, %edx
; AVX2-NEXT:    vpextrb $14, %xmm0, -56(%rsp,%rdx)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vpextrb $15, %xmm0, -56(%rsp,%rax)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    cmpq $15, %rax
; AVX2-NEXT:    movq %rax, %rdx
; AVX2-NEXT:    vpextrb $15, %xmm0, %eax
; AVX2-NEXT:    cmovbel {{[-0-9]+}}(%r{{[sb]}}p), %eax # 4-byte Folded Reload
; AVX2-NEXT:    movl $15, %ecx
; AVX2-NEXT:    cmovbq %rdx, %rcx
; AVX2-NEXT:    movb %al, -56(%rsp,%rcx)
; AVX2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r12
; AVX2-NEXT:    popq %r13
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v16i8:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
; AVX512F-NEXT:    vpslld $31, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm1 {%k1}
; AVX512F-NEXT:    vpmovdb %zmm1, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-ONLY-LABEL: test_compress_v16i8:
; AVX512VL-ONLY:       # %bb.0:
; AVX512VL-ONLY-NEXT:    vpsllw $7, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpmovb2m %xmm1, %k1
; AVX512VL-ONLY-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512VL-ONLY-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
; AVX512VL-ONLY-NEXT:    vpcompressd %zmm0, %zmm1 {%k1}
; AVX512VL-ONLY-NEXT:    vpmovdb %zmm1, %xmm0
; AVX512VL-ONLY-NEXT:    vzeroupper
; AVX512VL-ONLY-NEXT:    retq
;
; AVX512VL-VBMI2-LABEL: test_compress_v16i8:
; AVX512VL-VBMI2:       # %bb.0:
; AVX512VL-VBMI2-NEXT:    vpsllw $7, %xmm1, %xmm1
; AVX512VL-VBMI2-NEXT:    vpmovb2m %xmm1, %k1
; AVX512VL-VBMI2-NEXT:    vpcompressb %xmm0, %xmm2 {%k1}
; AVX512VL-VBMI2-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512VL-VBMI2-NEXT:    retq
    %out = call <16 x i8> @llvm.experimental.vector.compress(<16 x i8> %vec, <16 x i1> %mask, <16 x i8> %passthru)
    ret <16 x i8> %out
}

define <8 x i16> @test_compress_v8i16(<8 x i16> %vec, <8 x i1> %mask, <8 x i16> %passthru) nounwind {
; AVX2-LABEL: test_compress_v8i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    vpsllw $15, %xmm1, %xmm1
; AVX2-NEXT:    vpsraw $15, %xmm1, %xmm1
; AVX2-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vpextrw $1, %xmm1, %r9d
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    vmovd %xmm1, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    leaq (%rsi,%r9), %rcx
; AVX2-NEXT:    vpextrw $2, %xmm1, %r10d
; AVX2-NEXT:    movl %r10d, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    vpextrw $3, %xmm1, %r8d
; AVX2-NEXT:    movl %r8d, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rax, %rdx
; AVX2-NEXT:    vpextrw $4, %xmm1, %r11d
; AVX2-NEXT:    movl %r11d, %edi
; AVX2-NEXT:    andl $1, %edi
; AVX2-NEXT:    addq %rdx, %rdi
; AVX2-NEXT:    movl %esi, %ebx
; AVX2-NEXT:    addl %r9d, %ebx
; AVX2-NEXT:    vpextrw $5, %xmm1, %ebp
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addl %r10d, %r8d
; AVX2-NEXT:    movl %ebp, %r9d
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addq %rdi, %r9
; AVX2-NEXT:    addl %ebx, %r8d
; AVX2-NEXT:    vpextrw $6, %xmm1, %ebx
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    andl $1, %ebp
; AVX2-NEXT:    addl %r11d, %ebp
; AVX2-NEXT:    movl %ebx, %r10d
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    addq %r9, %r10
; AVX2-NEXT:    andl $1, %ebx
; AVX2-NEXT:    addl %ebp, %ebx
; AVX2-NEXT:    vpextrw $7, %xmm1, %r14d
; AVX2-NEXT:    addl %r8d, %ebx
; AVX2-NEXT:    movl %r14d, %r8d
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addq %r10, %r8
; AVX2-NEXT:    andl $1, %r14d
; AVX2-NEXT:    addl %ebx, %r14d
; AVX2-NEXT:    andl $7, %r14d
; AVX2-NEXT:    vpextrw $7, %xmm0, %r11d
; AVX2-NEXT:    cmpq $8, %r8
; AVX2-NEXT:    cmovbw -16(%rsp,%r14,2), %r11w
; AVX2-NEXT:    vpextrw $0, %xmm0, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vpextrw $1, %xmm0, -16(%rsp,%rsi,2)
; AVX2-NEXT:    andl $7, %ecx
; AVX2-NEXT:    vpextrw $2, %xmm0, -16(%rsp,%rcx,2)
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    vpextrw $3, %xmm0, -16(%rsp,%rax,2)
; AVX2-NEXT:    andl $7, %edx
; AVX2-NEXT:    vpextrw $4, %xmm0, -16(%rsp,%rdx,2)
; AVX2-NEXT:    andl $7, %edi
; AVX2-NEXT:    vpextrw $5, %xmm0, -16(%rsp,%rdi,2)
; AVX2-NEXT:    andl $7, %r9d
; AVX2-NEXT:    vpextrw $6, %xmm0, -16(%rsp,%r9,2)
; AVX2-NEXT:    andl $7, %r10d
; AVX2-NEXT:    vpextrw $7, %xmm0, -16(%rsp,%r10,2)
; AVX2-NEXT:    cmpq $7, %r8
; AVX2-NEXT:    movl $7, %eax
; AVX2-NEXT:    cmovbq %r8, %rax
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    movw %r11w, -16(%rsp,%rax,2)
; AVX2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v8i16:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpmovsxwq %xmm1, %zmm1
; AVX512F-NEXT:    vpsllq $63, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
; AVX512F-NEXT:    vpmovzxwq {{.*#+}} zmm1 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero
; AVX512F-NEXT:    vpcompressq %zmm0, %zmm1 {%k1}
; AVX512F-NEXT:    vpmovqw %zmm1, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-ONLY-LABEL: test_compress_v8i16:
; AVX512VL-ONLY:       # %bb.0:
; AVX512VL-ONLY-NEXT:    vpsllw $15, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpmovw2m %xmm1, %k1
; AVX512VL-ONLY-NEXT:    vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
; AVX512VL-ONLY-NEXT:    vpmovzxwq {{.*#+}} zmm1 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero
; AVX512VL-ONLY-NEXT:    vpcompressq %zmm0, %zmm1 {%k1}
; AVX512VL-ONLY-NEXT:    vpmovqw %zmm1, %xmm0
; AVX512VL-ONLY-NEXT:    vzeroupper
; AVX512VL-ONLY-NEXT:    retq
;
; AVX512VL-VBMI2-LABEL: test_compress_v8i16:
; AVX512VL-VBMI2:       # %bb.0:
; AVX512VL-VBMI2-NEXT:    vpsllw $15, %xmm1, %xmm1
; AVX512VL-VBMI2-NEXT:    vpmovw2m %xmm1, %k1
; AVX512VL-VBMI2-NEXT:    vpcompressw %xmm0, %xmm2 {%k1}
; AVX512VL-VBMI2-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512VL-VBMI2-NEXT:    retq
    %out = call <8 x i16> @llvm.experimental.vector.compress(<8 x i16> %vec, <8 x i1> %mask, <8 x i16> %passthru)
    ret <8 x i16> %out
}

define <32 x i8> @test_compress_v32i8(<32 x i8> %vec, <32 x i1> %mask, <32 x i8> %passthru) nounwind {
; AVX2-LABEL: test_compress_v32i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $32, %rsp
; AVX2-NEXT:    vpsllw $7, %ymm1, %ymm1
; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-NEXT:    vmovaps %ymm2, (%rsp)
; AVX2-NEXT:    vpcmpgtb %ymm1, %ymm3, %ymm2
; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm1
; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX2-NEXT:    vpand %xmm3, %xmm1, %xmm4
; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm3
; AVX2-NEXT:    vpaddb %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpshufb {{.*#+}} xmm4 = xmm3[8,9,10,11,12,13,14,15,12,13],zero,zero,xmm3[14,15],zero,zero
; AVX2-NEXT:    vpaddb %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,1,1]
; AVX2-NEXT:    vpaddb %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpsrld $16, %xmm3, %xmm4
; AVX2-NEXT:    vpaddb %xmm4, %xmm3, %xmm3
; AVX2-NEXT:    vpextrb $1, %xmm3, %eax
; AVX2-NEXT:    vmovd %xmm3, %ecx
; AVX2-NEXT:    addb %al, %cl
; AVX2-NEXT:    movzbl %cl, %eax
; AVX2-NEXT:    andl $31, %eax
; AVX2-NEXT:    movzbl (%rsp,%rax), %eax
; AVX2-NEXT:    vpextrb $0, %xmm0, (%rsp)
; AVX2-NEXT:    vmovd %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpextrb $1, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $1, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    vpextrb $2, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $2, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    vpextrb $3, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $3, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    vpextrb $4, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $4, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    vpextrb $5, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $5, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    vpextrb $6, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $6, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $7, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $7, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $8, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $8, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $9, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $9, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $10, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $10, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $11, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $11, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $12, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $12, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $13, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $13, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $14, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $14, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $15, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $15, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vmovd %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vpextrb $0, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $1, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $1, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $2, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $2, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $3, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $3, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $4, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $4, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $5, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $5, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $6, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $6, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $7, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $7, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $8, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $8, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $9, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $9, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $10, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $10, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $11, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $11, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $12, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $12, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $13, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrb $13, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    vpextrb $14, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $14, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    movl %ecx, %edx
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrb $15, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    vpextrb $15, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    vpextrb $15, %xmm0, %ecx
; AVX2-NEXT:    cmpq $32, %rdx
; AVX2-NEXT:    cmovbl %eax, %ecx
; AVX2-NEXT:    cmpq $31, %rdx
; AVX2-NEXT:    movl $31, %eax
; AVX2-NEXT:    cmovbq %rdx, %rax
; AVX2-NEXT:    movb %cl, (%rsp,%rax)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    movq %rbp, %rsp
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v32i8:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    pushq %rbp
; AVX512F-NEXT:    movq %rsp, %rbp
; AVX512F-NEXT:    andq $-32, %rsp
; AVX512F-NEXT:    subq $32, %rsp
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm3
; AVX512F-NEXT:    vpmovsxbd %xmm3, %zmm5
; AVX512F-NEXT:    vpslld $31, %zmm5, %zmm5
; AVX512F-NEXT:    vptestmd %zmm5, %zmm5, %k1
; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm5
; AVX512F-NEXT:    vpslld $31, %zmm5, %zmm5
; AVX512F-NEXT:    vptestmd %zmm5, %zmm5, %k2
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm5 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512F-NEXT:    vpcompressd %zmm5, %zmm5 {%k2} {z}
; AVX512F-NEXT:    vpmovdb %zmm5, (%rsp)
; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm5 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512F-NEXT:    vpand %xmm5, %xmm1, %xmm1
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm6
; AVX512F-NEXT:    vpaddd %ymm6, %ymm1, %ymm1
; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm6
; AVX512F-NEXT:    vpaddd %xmm6, %xmm1, %xmm1
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm6 = xmm1[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm6, %xmm1, %xmm1
; AVX512F-NEXT:    vpextrd $1, %xmm1, %eax
; AVX512F-NEXT:    vmovd %xmm1, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    andl $31, %ecx
; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT:    vpmovdb %zmm0, (%rsp,%rcx)
; AVX512F-NEXT:    vpand %xmm5, %xmm3, %xmm0
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512F-NEXT:    vpslld $31, %zmm4, %zmm1
; AVX512F-NEXT:    vpsrad $31, %zmm1, %zmm1
; AVX512F-NEXT:    vpsubd %zmm1, %zmm0, %zmm0
; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
; AVX512F-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX512F-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512F-NEXT:    vpextrd $1, %xmm0, %eax
; AVX512F-NEXT:    vmovd %xmm0, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    vpbroadcastd %ecx, %zmm0
; AVX512F-NEXT:    vpcmpnleud {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %k1
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm1 {%k1} {z} = -1
; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
; AVX512F-NEXT:    vpcmpnleud {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %k1
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1
; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
; AVX512F-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
; AVX512F-NEXT:    vpblendvb %ymm0, (%rsp), %ymm2, %ymm0
; AVX512F-NEXT:    movq %rbp, %rsp
; AVX512F-NEXT:    popq %rbp
; AVX512F-NEXT:    retq
;
; AVX512VL-ONLY-LABEL: test_compress_v32i8:
; AVX512VL-ONLY:       # %bb.0:
; AVX512VL-ONLY-NEXT:    pushq %rbp
; AVX512VL-ONLY-NEXT:    movq %rsp, %rbp
; AVX512VL-ONLY-NEXT:    andq $-32, %rsp
; AVX512VL-ONLY-NEXT:    subq $32, %rsp
; AVX512VL-ONLY-NEXT:    vpsllw $7, %ymm1, %ymm1
; AVX512VL-ONLY-NEXT:    vpmovb2m %ymm1, %k5
; AVX512VL-ONLY-NEXT:    kshiftrd $31, %k5, %k0
; AVX512VL-ONLY-NEXT:    kshiftrd $30, %k5, %k1
; AVX512VL-ONLY-NEXT:    kshiftrd $29, %k5, %k2
; AVX512VL-ONLY-NEXT:    kshiftrd $28, %k5, %k3
; AVX512VL-ONLY-NEXT:    kshiftrd $27, %k5, %k4
; AVX512VL-ONLY-NEXT:    kshiftrd $2, %k5, %k6
; AVX512VL-ONLY-NEXT:    kshiftrd $1, %k5, %k7
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    vmovaps %ymm2, (%rsp)
; AVX512VL-ONLY-NEXT:    vmovdqu8 {{.*#+}} ymm1 {%k5} {z} = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512VL-ONLY-NEXT:    vextracti128 $1, %ymm1, %xmm2
; AVX512VL-ONLY-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
; AVX512VL-ONLY-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,1,1]
; AVX512VL-ONLY-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpsrld $16, %xmm1, %xmm2
; AVX512VL-ONLY-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpextrb $1, %xmm1, %eax
; AVX512VL-ONLY-NEXT:    vmovd %xmm1, %edx
; AVX512VL-ONLY-NEXT:    addb %al, %dl
; AVX512VL-ONLY-NEXT:    movzbl %dl, %eax
; AVX512VL-ONLY-NEXT:    andl $31, %eax
; AVX512VL-ONLY-NEXT:    movzbl (%rsp,%rax), %eax
; AVX512VL-ONLY-NEXT:    vpextrb $0, %xmm0, (%rsp)
; AVX512VL-ONLY-NEXT:    kmovd %k5, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $1, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $2, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k6, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $3, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kshiftrd $26, %k5, %k6
; AVX512VL-ONLY-NEXT:    kshiftrd $3, %k5, %k7
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $4, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $4, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $5, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $5, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $6, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $7, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $6, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $7, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $8, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $9, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $8, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $10, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $9, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $10, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $11, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $12, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $11, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $13, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $12, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $13, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $14, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $15, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $14, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $16, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $15, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $17, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX512VL-ONLY-NEXT:    vpextrb $0, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $18, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $1, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $2, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $19, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $20, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $3, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $21, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $4, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $5, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $22, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $23, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $6, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrd $25, %k5, %k7
; AVX512VL-ONLY-NEXT:    kshiftrd $24, %k5, %k5
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $7, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $8, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k5, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $9, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k6, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $10, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k4, %ecx
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $11, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k3, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $12, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k2, %ecx
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $13, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $14, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k1, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $15, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k0, %ecx
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    vpextrb $15, %xmm0, %edx
; AVX512VL-ONLY-NEXT:    cmpq $32, %rcx
; AVX512VL-ONLY-NEXT:    cmovbl %eax, %edx
; AVX512VL-ONLY-NEXT:    cmpq $31, %rcx
; AVX512VL-ONLY-NEXT:    movl $31, %eax
; AVX512VL-ONLY-NEXT:    cmovbq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movb %dl, (%rsp,%rax)
; AVX512VL-ONLY-NEXT:    vmovaps (%rsp), %ymm0
; AVX512VL-ONLY-NEXT:    movq %rbp, %rsp
; AVX512VL-ONLY-NEXT:    popq %rbp
; AVX512VL-ONLY-NEXT:    retq
;
; AVX512VL-VBMI2-LABEL: test_compress_v32i8:
; AVX512VL-VBMI2:       # %bb.0:
; AVX512VL-VBMI2-NEXT:    vpsllw $7, %ymm1, %ymm1
; AVX512VL-VBMI2-NEXT:    vpmovb2m %ymm1, %k1
; AVX512VL-VBMI2-NEXT:    vpcompressb %ymm0, %ymm2 {%k1}
; AVX512VL-VBMI2-NEXT:    vmovdqa %ymm2, %ymm0
; AVX512VL-VBMI2-NEXT:    retq
    %out = call <32 x i8> @llvm.experimental.vector.compress(<32 x i8> %vec, <32 x i1> %mask, <32 x i8> %passthru)
    ret <32 x i8> %out
}

define <16 x i16> @test_compress_v16i16(<16 x i16> %vec, <16 x i1> %mask, <16 x i16> %passthru) nounwind {
; AVX2-LABEL: test_compress_v16i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %r13
; AVX2-NEXT:    pushq %r12
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $96, %rsp
; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX2-NEXT:    vpsllw $15, %ymm1, %ymm3
; AVX2-NEXT:    vpsraw $15, %ymm3, %ymm1
; AVX2-NEXT:    vmovaps %ymm2, (%rsp)
; AVX2-NEXT:    vpsrlw $15, %ymm3, %ymm2
; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
; AVX2-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[2,3,2,3]
; AVX2-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[1,1,1,1]
; AVX2-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
; AVX2-NEXT:    vpextrw $1, %xmm2, %eax
; AVX2-NEXT:    vmovd %xmm2, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    vpextrw $6, %xmm1, %edx
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    vpextrw $1, %xmm1, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vmovd %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrw $2, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrw $3, %xmm1, %r8d
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addq %rcx, %r8
; AVX2-NEXT:    vpextrw $4, %xmm1, %r10d
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    addq %r8, %r10
; AVX2-NEXT:    vpextrw $5, %xmm1, %r11d
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    addq %r10, %r11
; AVX2-NEXT:    addq %r11, %rdx
; AVX2-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrw $7, %xmm1, %r9d
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addq %rdx, %r9
; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm1
; AVX2-NEXT:    vmovd %xmm1, %ebx
; AVX2-NEXT:    andl $1, %ebx
; AVX2-NEXT:    addq %r9, %rbx
; AVX2-NEXT:    vpextrw $1, %xmm1, %r14d
; AVX2-NEXT:    andl $1, %r14d
; AVX2-NEXT:    addq %rbx, %r14
; AVX2-NEXT:    vpextrw $2, %xmm1, %r15d
; AVX2-NEXT:    andl $1, %r15d
; AVX2-NEXT:    addq %r14, %r15
; AVX2-NEXT:    vpextrw $3, %xmm1, %r12d
; AVX2-NEXT:    andl $1, %r12d
; AVX2-NEXT:    addq %r15, %r12
; AVX2-NEXT:    vpextrw $4, %xmm1, %r13d
; AVX2-NEXT:    andl $1, %r13d
; AVX2-NEXT:    addq %r12, %r13
; AVX2-NEXT:    vpextrw $5, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %r13, %rdx
; AVX2-NEXT:    vpextrw $6, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    vpextrw $7, %xmm1, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    addq %rcx, %rsi
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vpextrw $7, %xmm1, %eax
; AVX2-NEXT:    cmpq $16, %rsi
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
; AVX2-NEXT:    cmovbw (%rsp,%rdi,2), %ax
; AVX2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX2-NEXT:    vpextrw $0, %xmm0, (%rsp)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
; AVX2-NEXT:    vpextrw $1, %xmm0, (%rsp,%rdi,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
; AVX2-NEXT:    vpextrw $2, %xmm0, (%rsp,%rdi,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
; AVX2-NEXT:    vpextrw $3, %xmm0, (%rsp,%rdi,2)
; AVX2-NEXT:    vpextrw $4, %xmm0, (%rsp,%r8,2)
; AVX2-NEXT:    andl $15, %r10d
; AVX2-NEXT:    vpextrw $5, %xmm0, (%rsp,%r10,2)
; AVX2-NEXT:    andl $15, %r11d
; AVX2-NEXT:    vpextrw $6, %xmm0, (%rsp,%r11,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    vpextrw $7, %xmm0, (%rsp,%rax,2)
; AVX2-NEXT:    andl $15, %r9d
; AVX2-NEXT:    vpextrw $0, %xmm1, (%rsp,%r9,2)
; AVX2-NEXT:    andl $15, %ebx
; AVX2-NEXT:    vpextrw $1, %xmm1, (%rsp,%rbx,2)
; AVX2-NEXT:    andl $15, %r14d
; AVX2-NEXT:    vpextrw $2, %xmm1, (%rsp,%r14,2)
; AVX2-NEXT:    andl $15, %r15d
; AVX2-NEXT:    vpextrw $3, %xmm1, (%rsp,%r15,2)
; AVX2-NEXT:    andl $15, %r12d
; AVX2-NEXT:    vpextrw $4, %xmm1, (%rsp,%r12,2)
; AVX2-NEXT:    andl $15, %r13d
; AVX2-NEXT:    vpextrw $5, %xmm1, (%rsp,%r13,2)
; AVX2-NEXT:    andl $15, %edx
; AVX2-NEXT:    vpextrw $6, %xmm1, (%rsp,%rdx,2)
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vpextrw $7, %xmm1, (%rsp,%rcx,2)
; AVX2-NEXT:    cmpq $15, %rsi
; AVX2-NEXT:    movl $15, %eax
; AVX2-NEXT:    cmovbq %rsi, %rax
; AVX2-NEXT:    andl $15, %eax
; AVX2-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 4-byte Reload
; AVX2-NEXT:    movw %cx, (%rsp,%rax,2)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    leaq -40(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r12
; AVX2-NEXT:    popq %r13
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v16i16:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm1
; AVX512F-NEXT:    vpslld $31, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm1 {%k1}
; AVX512F-NEXT:    vpmovdw %zmm1, %ymm0
; AVX512F-NEXT:    retq
;
; AVX512VL-ONLY-LABEL: test_compress_v16i16:
; AVX512VL-ONLY:       # %bb.0:
; AVX512VL-ONLY-NEXT:    vpsllw $7, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpmovb2m %xmm1, %k1
; AVX512VL-ONLY-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512VL-ONLY-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
; AVX512VL-ONLY-NEXT:    vpcompressd %zmm0, %zmm1 {%k1}
; AVX512VL-ONLY-NEXT:    vpmovdw %zmm1, %ymm0
; AVX512VL-ONLY-NEXT:    retq
;
; AVX512VL-VBMI2-LABEL: test_compress_v16i16:
; AVX512VL-VBMI2:       # %bb.0:
; AVX512VL-VBMI2-NEXT:    vpsllw $7, %xmm1, %xmm1
; AVX512VL-VBMI2-NEXT:    vpmovb2m %xmm1, %k1
; AVX512VL-VBMI2-NEXT:    vpcompressw %ymm0, %ymm2 {%k1}
; AVX512VL-VBMI2-NEXT:    vmovdqa %ymm2, %ymm0
; AVX512VL-VBMI2-NEXT:    retq
    %out = call <16 x i16> @llvm.experimental.vector.compress(<16 x i16> %vec, <16 x i1> %mask, <16 x i16> %passthru)
    ret <16 x i16> %out
}

define <64 x i8> @test_compress_v64i8(<64 x i8> %vec, <64 x i1> %mask, <64 x i8> %passthru) nounwind {
; AVX2-LABEL: test_compress_v64i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %r13
; AVX2-NEXT:    pushq %r12
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $64, %rsp
; AVX2-NEXT:    movl 360(%rbp), %eax
; AVX2-NEXT:    movl 352(%rbp), %r10d
; AVX2-NEXT:    vmovd %r10d, %xmm4
; AVX2-NEXT:    vpinsrb $1, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 368(%rbp), %eax
; AVX2-NEXT:    vpinsrb $2, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 376(%rbp), %eax
; AVX2-NEXT:    vpinsrb $3, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 384(%rbp), %eax
; AVX2-NEXT:    vpinsrb $4, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 392(%rbp), %eax
; AVX2-NEXT:    vpinsrb $5, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 400(%rbp), %eax
; AVX2-NEXT:    vpinsrb $6, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 408(%rbp), %eax
; AVX2-NEXT:    vpinsrb $7, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 416(%rbp), %eax
; AVX2-NEXT:    vpinsrb $8, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 424(%rbp), %eax
; AVX2-NEXT:    vpinsrb $9, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 432(%rbp), %eax
; AVX2-NEXT:    vpinsrb $10, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 440(%rbp), %eax
; AVX2-NEXT:    vpinsrb $11, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 448(%rbp), %eax
; AVX2-NEXT:    vpinsrb $12, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 456(%rbp), %eax
; AVX2-NEXT:    vpinsrb $13, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 464(%rbp), %eax
; AVX2-NEXT:    vpinsrb $14, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 472(%rbp), %eax
; AVX2-NEXT:    vpinsrb $15, %eax, %xmm4, %xmm4
; AVX2-NEXT:    movl 224(%rbp), %eax
; AVX2-NEXT:    vmovd %eax, %xmm5
; AVX2-NEXT:    movl 232(%rbp), %eax
; AVX2-NEXT:    vpinsrb $1, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 240(%rbp), %eax
; AVX2-NEXT:    vpinsrb $2, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 248(%rbp), %eax
; AVX2-NEXT:    vpinsrb $3, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 256(%rbp), %eax
; AVX2-NEXT:    vpinsrb $4, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 264(%rbp), %eax
; AVX2-NEXT:    vpinsrb $5, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 272(%rbp), %eax
; AVX2-NEXT:    vpinsrb $6, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 280(%rbp), %eax
; AVX2-NEXT:    vpinsrb $7, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 288(%rbp), %eax
; AVX2-NEXT:    vpinsrb $8, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 296(%rbp), %eax
; AVX2-NEXT:    vpinsrb $9, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 304(%rbp), %eax
; AVX2-NEXT:    vpinsrb $10, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 312(%rbp), %eax
; AVX2-NEXT:    vpinsrb $11, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 320(%rbp), %eax
; AVX2-NEXT:    vpinsrb $12, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 328(%rbp), %eax
; AVX2-NEXT:    vpinsrb $13, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 336(%rbp), %eax
; AVX2-NEXT:    vpinsrb $14, %eax, %xmm5, %xmm5
; AVX2-NEXT:    movl 344(%rbp), %eax
; AVX2-NEXT:    vpinsrb $15, %eax, %xmm5, %xmm5
; AVX2-NEXT:    # kill: def $ecx killed $ecx def $rcx
; AVX2-NEXT:    # kill: def $edx killed $edx def $rdx
; AVX2-NEXT:    # kill: def $esi killed $esi def $rsi
; AVX2-NEXT:    movl %edi, %r14d
; AVX2-NEXT:    # kill: def $r9d killed $r9d def $r9
; AVX2-NEXT:    # kill: def $r8d killed $r8d def $r8
; AVX2-NEXT:    vmovd %edi, %xmm6
; AVX2-NEXT:    vpinsrb $1, %esi, %xmm6, %xmm6
; AVX2-NEXT:    vpinsrb $2, %edx, %xmm6, %xmm6
; AVX2-NEXT:    vpinsrb $3, %ecx, %xmm6, %xmm6
; AVX2-NEXT:    vpinsrb $4, %r8d, %xmm6, %xmm6
; AVX2-NEXT:    vpinsrb $5, %r9d, %xmm6, %xmm6
; AVX2-NEXT:    movl 16(%rbp), %edi
; AVX2-NEXT:    vpinsrb $6, %edi, %xmm6, %xmm6
; AVX2-NEXT:    movl 24(%rbp), %r10d
; AVX2-NEXT:    vpinsrb $7, %r10d, %xmm6, %xmm6
; AVX2-NEXT:    movl 32(%rbp), %r11d
; AVX2-NEXT:    vpinsrb $8, %r11d, %xmm6, %xmm6
; AVX2-NEXT:    movl 40(%rbp), %ebx
; AVX2-NEXT:    vpinsrb $9, %ebx, %xmm6, %xmm6
; AVX2-NEXT:    movl 48(%rbp), %r15d
; AVX2-NEXT:    vpinsrb $10, %r15d, %xmm6, %xmm6
; AVX2-NEXT:    movl 56(%rbp), %r12d
; AVX2-NEXT:    vpinsrb $11, %r12d, %xmm6, %xmm6
; AVX2-NEXT:    movl 64(%rbp), %eax
; AVX2-NEXT:    vpinsrb $12, %eax, %xmm6, %xmm6
; AVX2-NEXT:    movl 72(%rbp), %eax
; AVX2-NEXT:    vpinsrb $13, %eax, %xmm6, %xmm6
; AVX2-NEXT:    movl 80(%rbp), %eax
; AVX2-NEXT:    vpinsrb $14, %eax, %xmm6, %xmm6
; AVX2-NEXT:    movl 88(%rbp), %eax
; AVX2-NEXT:    vpinsrb $15, %eax, %xmm6, %xmm6
; AVX2-NEXT:    movl 96(%rbp), %eax
; AVX2-NEXT:    vmovd %eax, %xmm7
; AVX2-NEXT:    movl 104(%rbp), %eax
; AVX2-NEXT:    vpinsrb $1, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 112(%rbp), %eax
; AVX2-NEXT:    vpinsrb $2, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 120(%rbp), %eax
; AVX2-NEXT:    vpinsrb $3, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 128(%rbp), %eax
; AVX2-NEXT:    vpinsrb $4, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 136(%rbp), %eax
; AVX2-NEXT:    vpinsrb $5, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 144(%rbp), %eax
; AVX2-NEXT:    vpinsrb $6, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 152(%rbp), %eax
; AVX2-NEXT:    vpinsrb $7, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 160(%rbp), %eax
; AVX2-NEXT:    vpinsrb $8, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 168(%rbp), %eax
; AVX2-NEXT:    vpinsrb $9, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 176(%rbp), %eax
; AVX2-NEXT:    vpinsrb $10, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 184(%rbp), %eax
; AVX2-NEXT:    vpinsrb $11, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 192(%rbp), %eax
; AVX2-NEXT:    vpinsrb $12, %eax, %xmm7, %xmm7
; AVX2-NEXT:    movl 200(%rbp), %eax
; AVX2-NEXT:    vpinsrb $13, %eax, %xmm7, %xmm7
; AVX2-NEXT:    vinserti128 $1, %xmm4, %ymm5, %ymm4
; AVX2-NEXT:    movl 208(%rbp), %eax
; AVX2-NEXT:    vpinsrb $14, %eax, %xmm7, %xmm5
; AVX2-NEXT:    movl 216(%rbp), %eax
; AVX2-NEXT:    vpinsrb $15, %eax, %xmm5, %xmm5
; AVX2-NEXT:    vinserti128 $1, %xmm5, %ymm6, %ymm5
; AVX2-NEXT:    vmovaps %ymm3, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovaps %ymm2, (%rsp)
; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX2-NEXT:    vpand %ymm2, %ymm5, %ymm3
; AVX2-NEXT:    vpand %ymm2, %ymm4, %ymm2
; AVX2-NEXT:    vpaddb %ymm2, %ymm3, %ymm2
; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
; AVX2-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
; AVX2-NEXT:    vpshufb {{.*#+}} xmm3 = xmm2[8,9,10,11,12,13,14,15,12,13],zero,zero,xmm2[14,15],zero,zero
; AVX2-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[1,1,1,1]
; AVX2-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
; AVX2-NEXT:    vpsrld $16, %xmm2, %xmm3
; AVX2-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
; AVX2-NEXT:    vpextrb $1, %xmm2, %eax
; AVX2-NEXT:    vmovd %xmm2, %r13d
; AVX2-NEXT:    addb %al, %r13b
; AVX2-NEXT:    movzbl %r13b, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    movzbl (%rsp,%rax), %r13d
; AVX2-NEXT:    vpextrb $0, %xmm0, (%rsp)
; AVX2-NEXT:    andl $1, %r14d
; AVX2-NEXT:    vpextrb $1, %xmm0, (%rsp,%r14)
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    addq %r14, %rsi
; AVX2-NEXT:    vpextrb $2, %xmm0, (%rsp,%rsi)
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rsi, %rdx
; AVX2-NEXT:    vpextrb $3, %xmm0, (%rsp,%rdx)
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    vpextrb $4, %xmm0, (%rsp,%rcx)
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addq %rcx, %r8
; AVX2-NEXT:    movl %r8d, %eax
; AVX2-NEXT:    vpextrb $5, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addq %r8, %r9
; AVX2-NEXT:    movl %r9d, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $6, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    andl $1, %edi
; AVX2-NEXT:    addq %r9, %rdi
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    addq %rdi, %r10
; AVX2-NEXT:    # kill: def $edi killed $edi killed $rdi def $rdi
; AVX2-NEXT:    andl $63, %edi
; AVX2-NEXT:    vpextrb $7, %xmm0, (%rsp,%rdi)
; AVX2-NEXT:    movl %r10d, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $8, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    addq %r10, %r11
; AVX2-NEXT:    andl $1, %ebx
; AVX2-NEXT:    addq %r11, %rbx
; AVX2-NEXT:    # kill: def $r11d killed $r11d killed $r11 def $r11
; AVX2-NEXT:    andl $63, %r11d
; AVX2-NEXT:    vpextrb $9, %xmm0, (%rsp,%r11)
; AVX2-NEXT:    movl %ebx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $10, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    andl $1, %r15d
; AVX2-NEXT:    addq %rbx, %r15
; AVX2-NEXT:    andl $1, %r12d
; AVX2-NEXT:    addq %r15, %r12
; AVX2-NEXT:    # kill: def $r15d killed $r15d killed $r15 def $r15
; AVX2-NEXT:    andl $63, %r15d
; AVX2-NEXT:    vpextrb $11, %xmm0, (%rsp,%r15)
; AVX2-NEXT:    movl %r12d, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $12, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 64(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %r12, %rax
; AVX2-NEXT:    movl 72(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $13, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $14, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 80(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 88(%rbp), %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rax, %rdx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $15, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 96(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    movl %edx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vpextrb $0, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $1, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 104(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 112(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $2, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $3, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 120(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 128(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $4, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $5, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 136(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 144(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $6, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $7, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 152(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 160(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $8, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $9, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 168(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 176(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $10, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $11, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 184(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 192(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $12, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $13, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 200(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 208(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $14, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $15, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 216(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 224(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $0, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $1, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl 232(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 240(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $2, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $3, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl 248(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 256(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $4, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $5, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl 264(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 272(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $6, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $7, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl 280(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 288(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $8, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $9, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl 296(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 304(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $10, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $11, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl 312(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 320(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $12, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $13, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl 328(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 336(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $14, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $15, %xmm1, (%rsp,%rax)
; AVX2-NEXT:    movl 344(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 352(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm0
; AVX2-NEXT:    vpextrb $0, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $1, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 360(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 368(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $2, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $3, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 376(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 384(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $4, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $5, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 392(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 400(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $6, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $7, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 408(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 416(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $8, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $9, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 424(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 432(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $10, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $11, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 440(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 448(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $12, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $13, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 456(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movl 464(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $14, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    movq %rcx, %rdx
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vpextrb $15, %xmm0, (%rsp,%rax)
; AVX2-NEXT:    movl 472(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpextrb $15, %xmm0, %eax
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    cmpq $64, %rcx
; AVX2-NEXT:    cmovbl %r13d, %eax
; AVX2-NEXT:    cmpq $63, %rcx
; AVX2-NEXT:    movq %rcx, %rdx
; AVX2-NEXT:    movl $63, %ecx
; AVX2-NEXT:    cmovbq %rdx, %rcx
; AVX2-NEXT:    movb %al, (%rsp,%rcx)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm1
; AVX2-NEXT:    leaq -40(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r12
; AVX2-NEXT:    popq %r13
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v64i8:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    pushq %rbp
; AVX512F-NEXT:    movq %rsp, %rbp
; AVX512F-NEXT:    andq $-64, %rsp
; AVX512F-NEXT:    addq $-128, %rsp
; AVX512F-NEXT:    vmovdqu64 352(%rbp), %zmm2
; AVX512F-NEXT:    vmovdqu64 416(%rbp), %zmm3
; AVX512F-NEXT:    vpmovqw %zmm2, %xmm2
; AVX512F-NEXT:    vpmovqw %zmm3, %xmm3
; AVX512F-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
; AVX512F-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512F-NEXT:    vptestmd %zmm4, %zmm2, %k1
; AVX512F-NEXT:    vmovdqu64 224(%rbp), %zmm2
; AVX512F-NEXT:    vmovdqu64 288(%rbp), %zmm3
; AVX512F-NEXT:    vpmovqw %zmm2, %xmm2
; AVX512F-NEXT:    vpmovqw %zmm3, %xmm3
; AVX512F-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm3
; AVX512F-NEXT:    vmovdqu64 96(%rbp), %zmm2
; AVX512F-NEXT:    vmovdqu64 160(%rbp), %zmm5
; AVX512F-NEXT:    vpmovqw %zmm2, %xmm2
; AVX512F-NEXT:    vpmovqw %zmm5, %xmm5
; AVX512F-NEXT:    vinserti128 $1, %xmm5, %ymm2, %ymm2
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
; AVX512F-NEXT:    vmovd %edi, %xmm5
; AVX512F-NEXT:    vpinsrb $1, %esi, %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $2, %edx, %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $3, %ecx, %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $4, %r8d, %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $5, %r9d, %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $6, 16(%rbp), %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $7, 24(%rbp), %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $8, 32(%rbp), %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $9, 40(%rbp), %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $10, 48(%rbp), %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $11, 56(%rbp), %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $12, 64(%rbp), %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $13, 72(%rbp), %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $14, 80(%rbp), %xmm5, %xmm5
; AVX512F-NEXT:    vpinsrb $15, 88(%rbp), %xmm5, %xmm5
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero,xmm5[4],zero,zero,zero,xmm5[5],zero,zero,zero,xmm5[6],zero,zero,zero,xmm5[7],zero,zero,zero,xmm5[8],zero,zero,zero,xmm5[9],zero,zero,zero,xmm5[10],zero,zero,zero,xmm5[11],zero,zero,zero,xmm5[12],zero,zero,zero,xmm5[13],zero,zero,zero,xmm5[14],zero,zero,zero,xmm5[15],zero,zero,zero
; AVX512F-NEXT:    vptestmd %zmm4, %zmm5, %k3
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm5 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512F-NEXT:    vpcompressd %zmm5, %zmm5 {%k3} {z}
; AVX512F-NEXT:    vpmovdb %zmm5, (%rsp)
; AVX512F-NEXT:    vptestmd %zmm4, %zmm2, %k2
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm2 {%k3} {z} = -1
; AVX512F-NEXT:    vpsrld $31, %zmm2, %zmm5
; AVX512F-NEXT:    vextracti64x4 $1, %zmm5, %ymm6
; AVX512F-NEXT:    vpaddd %ymm6, %ymm5, %ymm5
; AVX512F-NEXT:    vextracti128 $1, %ymm5, %xmm6
; AVX512F-NEXT:    vpaddd %xmm6, %xmm5, %xmm5
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm6 = xmm5[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm6, %xmm5, %xmm5
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm3[0],zero,ymm3[1],zero,ymm3[2],zero,ymm3[3],zero,ymm3[4],zero,ymm3[5],zero,ymm3[6],zero,ymm3[7],zero,ymm3[8],zero,ymm3[9],zero,ymm3[10],zero,ymm3[11],zero,ymm3[12],zero,ymm3[13],zero,ymm3[14],zero,ymm3[15],zero
; AVX512F-NEXT:    vpextrd $1, %xmm5, %eax
; AVX512F-NEXT:    vmovd %xmm5, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    andl $31, %ecx
; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm5
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero,xmm5[4],zero,zero,zero,xmm5[5],zero,zero,zero,xmm5[6],zero,zero,zero,xmm5[7],zero,zero,zero,xmm5[8],zero,zero,zero,xmm5[9],zero,zero,zero,xmm5[10],zero,zero,zero,xmm5[11],zero,zero,zero,xmm5[12],zero,zero,zero,xmm5[13],zero,zero,zero,xmm5[14],zero,zero,zero,xmm5[15],zero,zero,zero
; AVX512F-NEXT:    vpcompressd %zmm5, %zmm5 {%k2} {z}
; AVX512F-NEXT:    vpmovdb %zmm5, (%rsp,%rcx)
; AVX512F-NEXT:    vptestmd %zmm4, %zmm3, %k3
; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero,xmm3[8],zero,zero,zero,xmm3[9],zero,zero,zero,xmm3[10],zero,zero,zero,xmm3[11],zero,zero,zero,xmm3[12],zero,zero,zero,xmm3[13],zero,zero,zero,xmm3[14],zero,zero,zero,xmm3[15],zero,zero,zero
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm0 {%k3} {z}
; AVX512F-NEXT:    vpmovdb %zmm0, {{[0-9]+}}(%rsp)
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm0 {%k3} {z} = -1
; AVX512F-NEXT:    vpsrld $31, %zmm0, %zmm0
; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
; AVX512F-NEXT:    vpaddd %ymm4, %ymm0, %ymm4
; AVX512F-NEXT:    vextracti128 $1, %ymm4, %xmm5
; AVX512F-NEXT:    vpaddd %xmm5, %xmm4, %xmm4
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm5 = xmm4[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm5, %xmm4, %xmm4
; AVX512F-NEXT:    vpextrd $1, %xmm4, %eax
; AVX512F-NEXT:    vmovd %xmm4, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    andl $31, %ecx
; AVX512F-NEXT:    vextracti128 $1, %ymm3, %xmm3
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero,xmm3[8],zero,zero,zero,xmm3[9],zero,zero,zero,xmm3[10],zero,zero,zero,xmm3[11],zero,zero,zero,xmm3[12],zero,zero,zero,xmm3[13],zero,zero,zero,xmm3[14],zero,zero,zero,xmm3[15],zero,zero,zero
; AVX512F-NEXT:    vpcompressd %zmm3, %zmm3 {%k1} {z}
; AVX512F-NEXT:    vpmovdb %zmm3, 32(%rsp,%rcx)
; AVX512F-NEXT:    vmovdqa (%rsp), %ymm3
; AVX512F-NEXT:    vmovdqa %ymm3, {{[0-9]+}}(%rsp)
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm3 {%k2} {z} = -1
; AVX512F-NEXT:    vpsrld $31, %zmm3, %zmm4
; AVX512F-NEXT:    vpsubd %zmm2, %zmm4, %zmm4
; AVX512F-NEXT:    vextracti64x4 $1, %zmm4, %ymm5
; AVX512F-NEXT:    vpaddd %ymm5, %ymm4, %ymm4
; AVX512F-NEXT:    vextracti128 $1, %ymm4, %xmm5
; AVX512F-NEXT:    vpaddd %xmm5, %xmm4, %xmm4
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm5 = xmm4[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm5, %xmm4, %xmm4
; AVX512F-NEXT:    vpextrd $1, %xmm4, %eax
; AVX512F-NEXT:    vmovd %xmm4, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    andl $63, %ecx
; AVX512F-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm4
; AVX512F-NEXT:    vmovdqa %ymm4, 64(%rsp,%rcx)
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm4 {%k1} {z} = -1
; AVX512F-NEXT:    vpsrld $31, %zmm4, %zmm4
; AVX512F-NEXT:    vpsubd %zmm3, %zmm4, %zmm3
; AVX512F-NEXT:    vpsubd %zmm2, %zmm0, %zmm0
; AVX512F-NEXT:    vpaddd %zmm3, %zmm0, %zmm0
; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
; AVX512F-NEXT:    vpaddd %ymm2, %ymm0, %ymm0
; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm2
; AVX512F-NEXT:    vpaddd %xmm2, %xmm0, %xmm0
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm2, %xmm0, %xmm0
; AVX512F-NEXT:    vpextrd $1, %xmm0, %eax
; AVX512F-NEXT:    vmovd %xmm0, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    vpbroadcastd %ecx, %zmm0
; AVX512F-NEXT:    vpcmpnleud {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %k1
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm2 {%k1} {z} = -1
; AVX512F-NEXT:    vpcmpnleud {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %k1
; AVX512F-NEXT:    vpmovdb %zmm2, %xmm2
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm3 {%k1} {z} = -1
; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
; AVX512F-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
; AVX512F-NEXT:    vpblendvb %ymm2, {{[0-9]+}}(%rsp), %ymm3, %ymm2
; AVX512F-NEXT:    vpcmpnleud {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %k1
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm3 {%k1} {z} = -1
; AVX512F-NEXT:    vpcmpnleud {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %k1
; AVX512F-NEXT:    vpmovdb %zmm3, %xmm0
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm3 {%k1} {z} = -1
; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
; AVX512F-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
; AVX512F-NEXT:    vpblendvb %ymm0, {{[0-9]+}}(%rsp), %ymm1, %ymm0
; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512F-NEXT:    movq %rbp, %rsp
; AVX512F-NEXT:    popq %rbp
; AVX512F-NEXT:    retq
;
; AVX512VL-ONLY-LABEL: test_compress_v64i8:
; AVX512VL-ONLY:       # %bb.0:
; AVX512VL-ONLY-NEXT:    pushq %rbp
; AVX512VL-ONLY-NEXT:    movq %rsp, %rbp
; AVX512VL-ONLY-NEXT:    andq $-64, %rsp
; AVX512VL-ONLY-NEXT:    subq $64, %rsp
; AVX512VL-ONLY-NEXT:    vpsllw $7, %zmm1, %zmm1
; AVX512VL-ONLY-NEXT:    vpmovb2m %zmm1, %k5
; AVX512VL-ONLY-NEXT:    kshiftrq $63, %k5, %k0
; AVX512VL-ONLY-NEXT:    kshiftrq $62, %k5, %k1
; AVX512VL-ONLY-NEXT:    kshiftrq $61, %k5, %k2
; AVX512VL-ONLY-NEXT:    kshiftrq $60, %k5, %k3
; AVX512VL-ONLY-NEXT:    kshiftrq $59, %k5, %k4
; AVX512VL-ONLY-NEXT:    kshiftrq $2, %k5, %k6
; AVX512VL-ONLY-NEXT:    vmovaps %zmm2, (%rsp)
; AVX512VL-ONLY-NEXT:    kshiftrq $1, %k5, %k7
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    vmovdqu8 {{.*#+}} zmm1 {%k5} {z} = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512VL-ONLY-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
; AVX512VL-ONLY-NEXT:    vpaddb %ymm2, %ymm1, %ymm1
; AVX512VL-ONLY-NEXT:    vextracti128 $1, %ymm1, %xmm2
; AVX512VL-ONLY-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
; AVX512VL-ONLY-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,1,1]
; AVX512VL-ONLY-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpsrld $16, %xmm1, %xmm2
; AVX512VL-ONLY-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpextrb $1, %xmm1, %eax
; AVX512VL-ONLY-NEXT:    vmovd %xmm1, %edx
; AVX512VL-ONLY-NEXT:    addb %al, %dl
; AVX512VL-ONLY-NEXT:    movzbl %dl, %eax
; AVX512VL-ONLY-NEXT:    andl $63, %eax
; AVX512VL-ONLY-NEXT:    movzbl (%rsp,%rax), %eax
; AVX512VL-ONLY-NEXT:    vpextrb $0, %xmm0, (%rsp)
; AVX512VL-ONLY-NEXT:    kmovd %k5, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $1, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $2, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k6, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $3, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kshiftrq $5, %k5, %k6
; AVX512VL-ONLY-NEXT:    kshiftrq $3, %k5, %k7
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $4, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $4, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $5, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k6, %ecx
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $6, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kshiftrq $58, %k5, %k6
; AVX512VL-ONLY-NEXT:    kshiftrq $6, %k5, %k7
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $7, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $7, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $8, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $9, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $8, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $10, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $9, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $10, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $11, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $12, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $11, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $13, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $12, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $13, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $14, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $15, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $14, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $16, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $15, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $17, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX512VL-ONLY-NEXT:    vpextrb $0, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $18, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $1, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $2, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $19, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $20, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $3, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $21, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $4, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $5, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $22, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $23, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $6, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $24, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $7, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $8, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $25, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $26, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $9, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $27, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $10, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $11, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $28, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $29, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $12, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $30, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $13, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $14, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $31, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $32, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $15, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $33, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
; AVX512VL-ONLY-NEXT:    vpextrb $0, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $1, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $34, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $35, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $2, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $36, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $3, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $4, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $37, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $38, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $5, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $39, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $6, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $7, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $40, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $41, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $8, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $42, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $9, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $10, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $43, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $44, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $11, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $45, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $12, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $13, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $46, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $47, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $14, %xmm1, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $48, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $15, %xmm1, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $49, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
; AVX512VL-ONLY-NEXT:    vpextrb $0, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $50, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $1, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $2, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $51, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $52, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $3, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $53, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $4, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $5, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $54, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %edx
; AVX512VL-ONLY-NEXT:    kshiftrq $55, %k5, %k7
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $6, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    kshiftrq $57, %k5, %k7
; AVX512VL-ONLY-NEXT:    kshiftrq $56, %k5, %k5
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $7, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $8, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k5, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $9, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k7, %ecx
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k6, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $10, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k4, %ecx
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $11, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k3, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $12, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k2, %ecx
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $13, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    movl %ecx, %edx
; AVX512VL-ONLY-NEXT:    andl $63, %edx
; AVX512VL-ONLY-NEXT:    vpextrb $14, %xmm0, (%rsp,%rdx)
; AVX512VL-ONLY-NEXT:    kmovd %k1, %edx
; AVX512VL-ONLY-NEXT:    movzbl %dl, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rcx, %rdx
; AVX512VL-ONLY-NEXT:    movl %edx, %ecx
; AVX512VL-ONLY-NEXT:    andl $63, %ecx
; AVX512VL-ONLY-NEXT:    vpextrb $15, %xmm0, (%rsp,%rcx)
; AVX512VL-ONLY-NEXT:    kmovd %k0, %ecx
; AVX512VL-ONLY-NEXT:    movzbl %cl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    vpextrb $15, %xmm0, %edx
; AVX512VL-ONLY-NEXT:    cmpq $64, %rcx
; AVX512VL-ONLY-NEXT:    cmovbl %eax, %edx
; AVX512VL-ONLY-NEXT:    cmpq $63, %rcx
; AVX512VL-ONLY-NEXT:    movl $63, %eax
; AVX512VL-ONLY-NEXT:    cmovbq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movb %dl, (%rsp,%rax)
; AVX512VL-ONLY-NEXT:    vmovaps (%rsp), %zmm0
; AVX512VL-ONLY-NEXT:    movq %rbp, %rsp
; AVX512VL-ONLY-NEXT:    popq %rbp
; AVX512VL-ONLY-NEXT:    retq
;
; AVX512VL-VBMI2-LABEL: test_compress_v64i8:
; AVX512VL-VBMI2:       # %bb.0:
; AVX512VL-VBMI2-NEXT:    vpsllw $7, %zmm1, %zmm1
; AVX512VL-VBMI2-NEXT:    vpmovb2m %zmm1, %k1
; AVX512VL-VBMI2-NEXT:    vpcompressb %zmm0, %zmm2 {%k1}
; AVX512VL-VBMI2-NEXT:    vmovdqa64 %zmm2, %zmm0
; AVX512VL-VBMI2-NEXT:    retq
    %out = call <64 x i8> @llvm.experimental.vector.compress(<64 x i8> %vec, <64 x i1> %mask, <64 x i8> %passthru)
    ret <64 x i8> %out
}

define <32 x i16> @test_compress_v32i16(<32 x i16> %vec, <32 x i1> %mask, <32 x i16> %passthru) nounwind {
; AVX2-LABEL: test_compress_v32i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %r13
; AVX2-NEXT:    pushq %r12
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $256, %rsp # imm = 0x100
; AVX2-NEXT:    vmovaps %ymm4, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovaps %ymm3, (%rsp)
; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX2-NEXT:    vpand %xmm4, %xmm3, %xmm5
; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm5 = xmm5[0],zero,xmm5[1],zero,xmm5[2],zero,xmm5[3],zero,xmm5[4],zero,xmm5[5],zero,xmm5[6],zero,xmm5[7],zero,xmm5[8],zero,xmm5[9],zero,xmm5[10],zero,xmm5[11],zero,xmm5[12],zero,xmm5[13],zero,xmm5[14],zero,xmm5[15],zero
; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm4
; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
; AVX2-NEXT:    vpaddw %ymm5, %ymm4, %ymm4
; AVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
; AVX2-NEXT:    vpaddw %xmm5, %xmm4, %xmm4
; AVX2-NEXT:    vpshufd {{.*#+}} xmm5 = xmm4[2,3,2,3]
; AVX2-NEXT:    vpaddw %xmm5, %xmm4, %xmm4
; AVX2-NEXT:    vpshufd {{.*#+}} xmm5 = xmm4[1,1,1,1]
; AVX2-NEXT:    vpaddw %xmm5, %xmm4, %xmm4
; AVX2-NEXT:    vpextrw $1, %xmm4, %eax
; AVX2-NEXT:    vmovd %xmm4, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $4, %xmm2, %eax
; AVX2-NEXT:    vpextrb $1, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vmovd %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $2, %xmm2, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rcx, %rdx
; AVX2-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $3, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $5, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $6, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $7, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $8, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $9, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $10, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $11, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $12, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $13, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $14, %xmm2, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $15, %xmm2, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vmovd %xmm3, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $1, %xmm3, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $2, %xmm3, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $3, %xmm3, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT:    vpextrb $4, %xmm3, %r15d
; AVX2-NEXT:    andl $1, %r15d
; AVX2-NEXT:    addq %rcx, %r15
; AVX2-NEXT:    vpextrb $5, %xmm3, %r14d
; AVX2-NEXT:    andl $1, %r14d
; AVX2-NEXT:    addq %r15, %r14
; AVX2-NEXT:    vpextrb $6, %xmm3, %ebx
; AVX2-NEXT:    andl $1, %ebx
; AVX2-NEXT:    addq %r14, %rbx
; AVX2-NEXT:    vpextrb $7, %xmm3, %r11d
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    addq %rbx, %r11
; AVX2-NEXT:    vpextrb $8, %xmm3, %r10d
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    addq %r11, %r10
; AVX2-NEXT:    vpextrb $9, %xmm3, %r9d
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addq %r10, %r9
; AVX2-NEXT:    vpextrb $10, %xmm3, %r8d
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addq %r9, %r8
; AVX2-NEXT:    vpextrb $11, %xmm3, %edi
; AVX2-NEXT:    andl $1, %edi
; AVX2-NEXT:    addq %r8, %rdi
; AVX2-NEXT:    vpextrb $12, %xmm3, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    addq %rdi, %rsi
; AVX2-NEXT:    vpextrb $13, %xmm3, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rsi, %rdx
; AVX2-NEXT:    vpextrb $14, %xmm3, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rdx, %rcx
; AVX2-NEXT:    vpextrb $15, %xmm3, %r12d
; AVX2-NEXT:    andl $1, %r12d
; AVX2-NEXT:    addq %rcx, %r12
; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
; AVX2-NEXT:    vpextrw $7, %xmm2, %eax
; AVX2-NEXT:    cmpq $32, %r12
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    cmovbw (%rsp,%r13,2), %ax
; AVX2-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX2-NEXT:    vpextrw $0, %xmm0, (%rsp)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    vpextrw $1, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    vpextrw $2, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    vpextrw $3, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $4, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $5, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $6, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $7, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vpextrw $0, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $1, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $2, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $3, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $4, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $5, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $6, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $7, %xmm0, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $0, %xmm1, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $1, %xmm1, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $2, %xmm1, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX2-NEXT:    andl $31, %r13d
; AVX2-NEXT:    vpextrw $3, %xmm1, (%rsp,%r13,2)
; AVX2-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; AVX2-NEXT:    andl $31, %eax
; AVX2-NEXT:    vpextrw $4, %xmm1, (%rsp,%rax,2)
; AVX2-NEXT:    andl $31, %r15d
; AVX2-NEXT:    vpextrw $5, %xmm1, (%rsp,%r15,2)
; AVX2-NEXT:    andl $31, %r14d
; AVX2-NEXT:    vpextrw $6, %xmm1, (%rsp,%r14,2)
; AVX2-NEXT:    andl $31, %ebx
; AVX2-NEXT:    vpextrw $7, %xmm1, (%rsp,%rbx,2)
; AVX2-NEXT:    andl $31, %r11d
; AVX2-NEXT:    vpextrw $0, %xmm2, (%rsp,%r11,2)
; AVX2-NEXT:    andl $31, %r10d
; AVX2-NEXT:    vpextrw $1, %xmm2, (%rsp,%r10,2)
; AVX2-NEXT:    andl $31, %r9d
; AVX2-NEXT:    vpextrw $2, %xmm2, (%rsp,%r9,2)
; AVX2-NEXT:    andl $31, %r8d
; AVX2-NEXT:    vpextrw $3, %xmm2, (%rsp,%r8,2)
; AVX2-NEXT:    andl $31, %edi
; AVX2-NEXT:    vpextrw $4, %xmm2, (%rsp,%rdi,2)
; AVX2-NEXT:    andl $31, %esi
; AVX2-NEXT:    vpextrw $5, %xmm2, (%rsp,%rsi,2)
; AVX2-NEXT:    andl $31, %edx
; AVX2-NEXT:    vpextrw $6, %xmm2, (%rsp,%rdx,2)
; AVX2-NEXT:    andl $31, %ecx
; AVX2-NEXT:    vpextrw $7, %xmm2, (%rsp,%rcx,2)
; AVX2-NEXT:    cmpq $31, %r12
; AVX2-NEXT:    movl $31, %eax
; AVX2-NEXT:    cmovbq %r12, %rax
; AVX2-NEXT:    addl %eax, %eax
; AVX2-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 4-byte Reload
; AVX2-NEXT:    movw %cx, (%rsp,%rax)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm1
; AVX2-NEXT:    leaq -40(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r12
; AVX2-NEXT:    popq %r13
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v32i16:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    pushq %rbp
; AVX512F-NEXT:    movq %rsp, %rbp
; AVX512F-NEXT:    andq $-64, %rsp
; AVX512F-NEXT:    subq $64, %rsp
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm3
; AVX512F-NEXT:    vpmovsxbd %xmm3, %zmm5
; AVX512F-NEXT:    vpslld $31, %zmm5, %zmm5
; AVX512F-NEXT:    vptestmd %zmm5, %zmm5, %k1
; AVX512F-NEXT:    vpmovsxbd %xmm1, %zmm5
; AVX512F-NEXT:    vpslld $31, %zmm5, %zmm5
; AVX512F-NEXT:    vptestmd %zmm5, %zmm5, %k2
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm5 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512F-NEXT:    vpcompressd %zmm5, %zmm5 {%k2} {z}
; AVX512F-NEXT:    vpmovdw %zmm5, (%rsp)
; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm5 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512F-NEXT:    vpand %xmm5, %xmm1, %xmm1
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm6
; AVX512F-NEXT:    vpaddd %ymm6, %ymm1, %ymm1
; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm6
; AVX512F-NEXT:    vpaddd %xmm6, %xmm1, %xmm1
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm6 = xmm1[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm6, %xmm1, %xmm1
; AVX512F-NEXT:    vpextrd $1, %xmm1, %eax
; AVX512F-NEXT:    vmovd %xmm1, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    andl $31, %ecx
; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT:    vpmovdw %zmm0, (%rsp,%rcx,2)
; AVX512F-NEXT:    vpand %xmm5, %xmm3, %xmm0
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512F-NEXT:    vpslld $31, %zmm4, %zmm1
; AVX512F-NEXT:    vpsrad $31, %zmm1, %zmm1
; AVX512F-NEXT:    vpsubd %zmm1, %zmm0, %zmm0
; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
; AVX512F-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX512F-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512F-NEXT:    vpextrd $1, %xmm0, %eax
; AVX512F-NEXT:    vmovd %xmm0, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    vpbroadcastd %ecx, %zmm0
; AVX512F-NEXT:    vpcmpnleud {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %k1
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm1 {%k1} {z} = -1
; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm3
; AVX512F-NEXT:    vpblendvb %ymm1, {{[0-9]+}}(%rsp), %ymm3, %ymm1
; AVX512F-NEXT:    vpcmpnleud {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %k1
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1
; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
; AVX512F-NEXT:    vpblendvb %ymm0, (%rsp), %ymm2, %ymm0
; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; AVX512F-NEXT:    movq %rbp, %rsp
; AVX512F-NEXT:    popq %rbp
; AVX512F-NEXT:    retq
;
; AVX512VL-ONLY-LABEL: test_compress_v32i16:
; AVX512VL-ONLY:       # %bb.0:
; AVX512VL-ONLY-NEXT:    pushq %rbp
; AVX512VL-ONLY-NEXT:    movq %rsp, %rbp
; AVX512VL-ONLY-NEXT:    pushq %r15
; AVX512VL-ONLY-NEXT:    pushq %r14
; AVX512VL-ONLY-NEXT:    pushq %r13
; AVX512VL-ONLY-NEXT:    pushq %r12
; AVX512VL-ONLY-NEXT:    pushq %rbx
; AVX512VL-ONLY-NEXT:    andq $-64, %rsp
; AVX512VL-ONLY-NEXT:    subq $320, %rsp # imm = 0x140
; AVX512VL-ONLY-NEXT:    vpsllw $7, %ymm1, %ymm1
; AVX512VL-ONLY-NEXT:    vpmovb2m %ymm1, %k7
; AVX512VL-ONLY-NEXT:    vpmovm2w %k7, %zmm1
; AVX512VL-ONLY-NEXT:    vpsrlw $15, %zmm1, %zmm1
; AVX512VL-ONLY-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
; AVX512VL-ONLY-NEXT:    vpaddw %ymm3, %ymm1, %ymm1
; AVX512VL-ONLY-NEXT:    vextracti128 $1, %ymm1, %xmm3
; AVX512VL-ONLY-NEXT:    vpaddw %xmm3, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[2,3,2,3]
; AVX512VL-ONLY-NEXT:    vpaddw %xmm3, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[1,1,1,1]
; AVX512VL-ONLY-NEXT:    vpaddw %xmm3, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    kshiftrd $31, %k7, %k0
; AVX512VL-ONLY-NEXT:    kshiftrd $30, %k7, %k1
; AVX512VL-ONLY-NEXT:    kshiftrd $29, %k7, %k2
; AVX512VL-ONLY-NEXT:    kshiftrd $28, %k7, %k3
; AVX512VL-ONLY-NEXT:    kshiftrd $27, %k7, %k4
; AVX512VL-ONLY-NEXT:    kshiftrd $26, %k7, %k5
; AVX512VL-ONLY-NEXT:    kshiftrd $23, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $22, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $21, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $20, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $19, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $18, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $17, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $16, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $15, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $14, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $13, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    kshiftrd $12, %k7, %k6
; AVX512VL-ONLY-NEXT:    vpextrw $1, %xmm1, %eax
; AVX512VL-ONLY-NEXT:    vmovd %xmm1, %ecx
; AVX512VL-ONLY-NEXT:    addl %eax, %ecx
; AVX512VL-ONLY-NEXT:    kmovd %k6, %r11d
; AVX512VL-ONLY-NEXT:    kshiftrd $11, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %ebx
; AVX512VL-ONLY-NEXT:    kshiftrd $10, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %r14d
; AVX512VL-ONLY-NEXT:    kshiftrd $9, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %r15d
; AVX512VL-ONLY-NEXT:    kshiftrd $8, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %r12d
; AVX512VL-ONLY-NEXT:    kshiftrd $7, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %r13d
; AVX512VL-ONLY-NEXT:    kshiftrd $6, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %r10d
; AVX512VL-ONLY-NEXT:    kshiftrd $5, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %r9d
; AVX512VL-ONLY-NEXT:    kshiftrd $4, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %r8d
; AVX512VL-ONLY-NEXT:    kshiftrd $3, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %edi
; AVX512VL-ONLY-NEXT:    kshiftrd $2, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %esi
; AVX512VL-ONLY-NEXT:    kshiftrd $1, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k6, %edx
; AVX512VL-ONLY-NEXT:    kshiftrd $25, %k7, %k6
; AVX512VL-ONLY-NEXT:    kmovd %k7, %eax
; AVX512VL-ONLY-NEXT:    kshiftrd $24, %k7, %k7
; AVX512VL-ONLY-NEXT:    vmovaps %zmm2, (%rsp)
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %al, %eax
; AVX512VL-ONLY-NEXT:    andl $1, %eax
; AVX512VL-ONLY-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %dl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rax, %rcx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %sil, %eax
; AVX512VL-ONLY-NEXT:    andl $1, %eax
; AVX512VL-ONLY-NEXT:    addq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %dil, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rax, %rcx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %r8b, %eax
; AVX512VL-ONLY-NEXT:    andl $1, %eax
; AVX512VL-ONLY-NEXT:    addq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %r9b, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rax, %rcx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %r10b, %eax
; AVX512VL-ONLY-NEXT:    andl $1, %eax
; AVX512VL-ONLY-NEXT:    addq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %r13b, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rax, %rcx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %r12b, %eax
; AVX512VL-ONLY-NEXT:    andl $1, %eax
; AVX512VL-ONLY-NEXT:    addq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %r15b, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rax, %rcx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %r14b, %eax
; AVX512VL-ONLY-NEXT:    andl $1, %eax
; AVX512VL-ONLY-NEXT:    addq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %bl, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rax, %rcx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl %r11b, %eax
; AVX512VL-ONLY-NEXT:    andl $1, %eax
; AVX512VL-ONLY-NEXT:    addq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rax, %rcx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %eax
; AVX512VL-ONLY-NEXT:    addq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rax, %rcx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %eax
; AVX512VL-ONLY-NEXT:    addq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rax, %rcx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %eax
; AVX512VL-ONLY-NEXT:    addq %rcx, %rax
; AVX512VL-ONLY-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rax, %rcx
; AVX512VL-ONLY-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r12d # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %r12d
; AVX512VL-ONLY-NEXT:    addq %rcx, %r12
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r15d # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %r15d
; AVX512VL-ONLY-NEXT:    addq %r12, %r15
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %r14d # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %r14d
; AVX512VL-ONLY-NEXT:    addq %r15, %r14
; AVX512VL-ONLY-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ebx # 1-byte Folded Reload
; AVX512VL-ONLY-NEXT:    andl $1, %ebx
; AVX512VL-ONLY-NEXT:    addq %r14, %rbx
; AVX512VL-ONLY-NEXT:    kmovd %k7, %eax
; AVX512VL-ONLY-NEXT:    movzbl %al, %r11d
; AVX512VL-ONLY-NEXT:    andl $1, %r11d
; AVX512VL-ONLY-NEXT:    addq %rbx, %r11
; AVX512VL-ONLY-NEXT:    kmovd %k6, %eax
; AVX512VL-ONLY-NEXT:    movzbl %al, %r10d
; AVX512VL-ONLY-NEXT:    andl $1, %r10d
; AVX512VL-ONLY-NEXT:    addq %r11, %r10
; AVX512VL-ONLY-NEXT:    kmovd %k5, %eax
; AVX512VL-ONLY-NEXT:    movzbl %al, %r9d
; AVX512VL-ONLY-NEXT:    andl $1, %r9d
; AVX512VL-ONLY-NEXT:    addq %r10, %r9
; AVX512VL-ONLY-NEXT:    kmovd %k4, %eax
; AVX512VL-ONLY-NEXT:    movzbl %al, %edi
; AVX512VL-ONLY-NEXT:    andl $1, %edi
; AVX512VL-ONLY-NEXT:    addq %r9, %rdi
; AVX512VL-ONLY-NEXT:    kmovd %k3, %eax
; AVX512VL-ONLY-NEXT:    movzbl %al, %esi
; AVX512VL-ONLY-NEXT:    andl $1, %esi
; AVX512VL-ONLY-NEXT:    addq %rdi, %rsi
; AVX512VL-ONLY-NEXT:    kmovd %k2, %eax
; AVX512VL-ONLY-NEXT:    movzbl %al, %edx
; AVX512VL-ONLY-NEXT:    andl $1, %edx
; AVX512VL-ONLY-NEXT:    addq %rsi, %rdx
; AVX512VL-ONLY-NEXT:    kmovd %k1, %eax
; AVX512VL-ONLY-NEXT:    movzbl %al, %ecx
; AVX512VL-ONLY-NEXT:    andl $1, %ecx
; AVX512VL-ONLY-NEXT:    addq %rdx, %rcx
; AVX512VL-ONLY-NEXT:    kmovd %k0, %r8d
; AVX512VL-ONLY-NEXT:    movzbl %r8b, %r8d
; AVX512VL-ONLY-NEXT:    andl $1, %r8d
; AVX512VL-ONLY-NEXT:    addq %rcx, %r8
; AVX512VL-ONLY-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
; AVX512VL-ONLY-NEXT:    vpextrw $7, %xmm1, %eax
; AVX512VL-ONLY-NEXT:    cmpq $32, %r8
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    cmovbw (%rsp,%r13,2), %ax
; AVX512VL-ONLY-NEXT:    movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; AVX512VL-ONLY-NEXT:    vpextrw $0, %xmm0, (%rsp)
; AVX512VL-ONLY-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %r13d # 4-byte Reload
; AVX512VL-ONLY-NEXT:    vpextrw $1, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movzwl {{[-0-9]+}}(%r{{[sb]}}p), %r13d # 2-byte Folded Reload
; AVX512VL-ONLY-NEXT:    vpextrw $2, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movzwl {{[-0-9]+}}(%r{{[sb]}}p), %r13d # 2-byte Folded Reload
; AVX512VL-ONLY-NEXT:    vpextrw $3, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movzwl {{[-0-9]+}}(%r{{[sb]}}p), %r13d # 2-byte Folded Reload
; AVX512VL-ONLY-NEXT:    vpextrw $4, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movzwl {{[-0-9]+}}(%r{{[sb]}}p), %r13d # 2-byte Folded Reload
; AVX512VL-ONLY-NEXT:    vpextrw $5, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $6, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $7, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vextracti128 $1, %ymm0, %xmm2
; AVX512VL-ONLY-NEXT:    vpextrw $0, %xmm2, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $1, %xmm2, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $2, %xmm2, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $3, %xmm2, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $4, %xmm2, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $5, %xmm2, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $6, %xmm2, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $7, %xmm2, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vextracti32x4 $2, %zmm0, %xmm0
; AVX512VL-ONLY-NEXT:    vpextrw $0, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $1, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $2, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %r13d
; AVX512VL-ONLY-NEXT:    vpextrw $3, %xmm0, (%rsp,%r13,2)
; AVX512VL-ONLY-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; AVX512VL-ONLY-NEXT:    andl $31, %eax
; AVX512VL-ONLY-NEXT:    vpextrw $4, %xmm0, (%rsp,%rax,2)
; AVX512VL-ONLY-NEXT:    andl $31, %r12d
; AVX512VL-ONLY-NEXT:    vpextrw $5, %xmm0, (%rsp,%r12,2)
; AVX512VL-ONLY-NEXT:    andl $31, %r15d
; AVX512VL-ONLY-NEXT:    vpextrw $6, %xmm0, (%rsp,%r15,2)
; AVX512VL-ONLY-NEXT:    andl $31, %r14d
; AVX512VL-ONLY-NEXT:    vpextrw $7, %xmm0, (%rsp,%r14,2)
; AVX512VL-ONLY-NEXT:    andl $31, %ebx
; AVX512VL-ONLY-NEXT:    vpextrw $0, %xmm1, (%rsp,%rbx,2)
; AVX512VL-ONLY-NEXT:    andl $31, %r11d
; AVX512VL-ONLY-NEXT:    vpextrw $1, %xmm1, (%rsp,%r11,2)
; AVX512VL-ONLY-NEXT:    andl $31, %r10d
; AVX512VL-ONLY-NEXT:    vpextrw $2, %xmm1, (%rsp,%r10,2)
; AVX512VL-ONLY-NEXT:    andl $31, %r9d
; AVX512VL-ONLY-NEXT:    vpextrw $3, %xmm1, (%rsp,%r9,2)
; AVX512VL-ONLY-NEXT:    andl $31, %edi
; AVX512VL-ONLY-NEXT:    vpextrw $4, %xmm1, (%rsp,%rdi,2)
; AVX512VL-ONLY-NEXT:    andl $31, %esi
; AVX512VL-ONLY-NEXT:    vpextrw $5, %xmm1, (%rsp,%rsi,2)
; AVX512VL-ONLY-NEXT:    andl $31, %edx
; AVX512VL-ONLY-NEXT:    vpextrw $6, %xmm1, (%rsp,%rdx,2)
; AVX512VL-ONLY-NEXT:    andl $31, %ecx
; AVX512VL-ONLY-NEXT:    vpextrw $7, %xmm1, (%rsp,%rcx,2)
; AVX512VL-ONLY-NEXT:    cmpq $31, %r8
; AVX512VL-ONLY-NEXT:    movl $31, %eax
; AVX512VL-ONLY-NEXT:    cmovbq %r8, %rax
; AVX512VL-ONLY-NEXT:    andl $31, %eax
; AVX512VL-ONLY-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 4-byte Reload
; AVX512VL-ONLY-NEXT:    movw %cx, (%rsp,%rax,2)
; AVX512VL-ONLY-NEXT:    vmovaps (%rsp), %zmm0
; AVX512VL-ONLY-NEXT:    leaq -40(%rbp), %rsp
; AVX512VL-ONLY-NEXT:    popq %rbx
; AVX512VL-ONLY-NEXT:    popq %r12
; AVX512VL-ONLY-NEXT:    popq %r13
; AVX512VL-ONLY-NEXT:    popq %r14
; AVX512VL-ONLY-NEXT:    popq %r15
; AVX512VL-ONLY-NEXT:    popq %rbp
; AVX512VL-ONLY-NEXT:    retq
;
; AVX512VL-VBMI2-LABEL: test_compress_v32i16:
; AVX512VL-VBMI2:       # %bb.0:
; AVX512VL-VBMI2-NEXT:    vpsllw $7, %ymm1, %ymm1
; AVX512VL-VBMI2-NEXT:    vpmovb2m %ymm1, %k1
; AVX512VL-VBMI2-NEXT:    vpcompressw %zmm0, %zmm2 {%k1}
; AVX512VL-VBMI2-NEXT:    vmovdqa64 %zmm2, %zmm0
; AVX512VL-VBMI2-NEXT:    retq
    %out = call <32 x i16> @llvm.experimental.vector.compress(<32 x i16> %vec, <32 x i1> %mask, <32 x i16> %passthru)
    ret <32 x i16> %out
}

define <64 x i32> @test_compress_large(<64 x i1> %mask, <64 x i32> %vec, <64 x i32> %passthru) nounwind {
; AVX2-LABEL: test_compress_large:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $256, %rsp # imm = 0x100
; AVX2-NEXT:    # kill: def $esi killed $esi def $rsi
; AVX2-NEXT:    vmovss %xmm0, (%rsp)
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rsi,4)
; AVX2-NEXT:    # kill: def $ecx killed $ecx def $rcx
; AVX2-NEXT:    # kill: def $edx killed $edx def $rdx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addl %esi, %edx
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rdx,4)
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %edx, %ecx
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    # kill: def $r9d killed $r9d def $r9
; AVX2-NEXT:    # kill: def $r8d killed $r8d def $r8
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addl %ecx, %r8d
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%r8,4)
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addl %r8d, %r9d
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%r9,4)
; AVX2-NEXT:    movl 16(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %r9d, %eax
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 24(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 32(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 40(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vmovss %xmm1, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $1, %xmm1, (%rsp,%rax,4)
; AVX2-NEXT:    movl 48(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 56(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm1, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm1, (%rsp,%rax,4)
; AVX2-NEXT:    movl 64(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 72(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 80(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 88(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 96(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vmovss %xmm2, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 104(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    movl 112(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractps $1, %xmm2, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 120(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm2, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm2, (%rsp,%rax,4)
; AVX2-NEXT:    movl 128(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 136(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    movl 144(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 152(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 160(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 168(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vmovss %xmm3, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $1, %xmm3, (%rsp,%rax,4)
; AVX2-NEXT:    movl 176(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 184(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm3, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm3, (%rsp,%rax,4)
; AVX2-NEXT:    movl 192(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 200(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractf128 $1, %ymm3, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 208(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 216(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 224(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vmovss %xmm4, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 232(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    movl 240(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractps $1, %xmm4, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 248(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm4, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm4, (%rsp,%rax,4)
; AVX2-NEXT:    movl 256(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractf128 $1, %ymm4, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 264(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    movl 272(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 280(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 288(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 296(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vmovss %xmm5, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $1, %xmm5, (%rsp,%rax,4)
; AVX2-NEXT:    movl 304(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 312(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm5, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm5, (%rsp,%rax,4)
; AVX2-NEXT:    movl 320(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 328(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractf128 $1, %ymm5, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 336(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 344(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 352(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vmovss %xmm6, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 360(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    movl 368(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractps $1, %xmm6, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 376(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm6, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm6, (%rsp,%rax,4)
; AVX2-NEXT:    movl 384(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractf128 $1, %ymm6, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 392(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    movl 400(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    movl 408(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl 416(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 424(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vmovss %xmm7, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $1, %xmm7, (%rsp,%rax,4)
; AVX2-NEXT:    movl 432(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 440(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $2, %xmm7, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $3, %xmm7, (%rsp,%rax,4)
; AVX2-NEXT:    movl 448(%rbp), %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addl %ecx, %eax
; AVX2-NEXT:    movl 456(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %eax, %ecx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractf128 $1, %ymm7, %xmm0
; AVX2-NEXT:    vmovss %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movl %ecx, %eax
; AVX2-NEXT:    andl $63, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, (%rsp,%rax,4)
; AVX2-NEXT:    movq %rdi, %rax
; AVX2-NEXT:    movl 464(%rbp), %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addl %ecx, %edx
; AVX2-NEXT:    movl 472(%rbp), %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addl %edx, %ecx
; AVX2-NEXT:    # kill: def $edx killed $edx killed $rdx def $rdx
; AVX2-NEXT:    andl $63, %edx
; AVX2-NEXT:    vextractps $2, %xmm0, (%rsp,%rdx,4)
; AVX2-NEXT:    andl $63, %ecx
; AVX2-NEXT:    vextractps $3, %xmm0, (%rsp,%rcx,4)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm1
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm2
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm3
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm4
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm5
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm6
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm7
; AVX2-NEXT:    vmovaps %ymm7, 224(%rdi)
; AVX2-NEXT:    vmovaps %ymm6, 192(%rdi)
; AVX2-NEXT:    vmovaps %ymm5, 160(%rdi)
; AVX2-NEXT:    vmovaps %ymm4, 128(%rdi)
; AVX2-NEXT:    vmovaps %ymm3, 96(%rdi)
; AVX2-NEXT:    vmovaps %ymm2, 64(%rdi)
; AVX2-NEXT:    vmovaps %ymm1, 32(%rdi)
; AVX2-NEXT:    vmovaps %ymm0, (%rdi)
; AVX2-NEXT:    movq %rbp, %rsp
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_large:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    pushq %rbp
; AVX512F-NEXT:    movq %rsp, %rbp
; AVX512F-NEXT:    andq $-64, %rsp
; AVX512F-NEXT:    subq $512, %rsp # imm = 0x200
; AVX512F-NEXT:    vmovdqu64 352(%rbp), %zmm4
; AVX512F-NEXT:    vmovdqu64 416(%rbp), %zmm5
; AVX512F-NEXT:    vpmovqw %zmm4, %xmm4
; AVX512F-NEXT:    vpmovqw %zmm5, %xmm5
; AVX512F-NEXT:    vinserti128 $1, %xmm5, %ymm4, %ymm4
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
; AVX512F-NEXT:    vpbroadcastd {{.*#+}} zmm5 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512F-NEXT:    vptestmd %zmm5, %zmm4, %k1
; AVX512F-NEXT:    vmovdqu64 224(%rbp), %zmm4
; AVX512F-NEXT:    vmovdqu64 288(%rbp), %zmm6
; AVX512F-NEXT:    vpmovqw %zmm4, %xmm4
; AVX512F-NEXT:    vpmovqw %zmm6, %xmm6
; AVX512F-NEXT:    vinserti128 $1, %xmm6, %ymm4, %ymm4
; AVX512F-NEXT:    vmovdqu64 96(%rbp), %zmm6
; AVX512F-NEXT:    vmovdqu64 160(%rbp), %zmm7
; AVX512F-NEXT:    vpmovqw %zmm6, %xmm6
; AVX512F-NEXT:    vpmovqw %zmm7, %xmm7
; AVX512F-NEXT:    vinserti128 $1, %xmm7, %ymm6, %ymm6
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm6[0],zero,ymm6[1],zero,ymm6[2],zero,ymm6[3],zero,ymm6[4],zero,ymm6[5],zero,ymm6[6],zero,ymm6[7],zero,ymm6[8],zero,ymm6[9],zero,ymm6[10],zero,ymm6[11],zero,ymm6[12],zero,ymm6[13],zero,ymm6[14],zero,ymm6[15],zero
; AVX512F-NEXT:    vptestmd %zmm5, %zmm6, %k2
; AVX512F-NEXT:    vmovd %edi, %xmm6
; AVX512F-NEXT:    vpinsrb $1, %esi, %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $2, %edx, %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $3, %ecx, %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $4, %r8d, %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $5, %r9d, %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $6, 16(%rbp), %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $7, 24(%rbp), %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $8, 32(%rbp), %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $9, 40(%rbp), %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $10, 48(%rbp), %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $11, 56(%rbp), %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $12, 64(%rbp), %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $13, 72(%rbp), %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $14, 80(%rbp), %xmm6, %xmm6
; AVX512F-NEXT:    vpinsrb $15, 88(%rbp), %xmm6, %xmm6
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero,xmm6[2],zero,zero,zero,xmm6[3],zero,zero,zero,xmm6[4],zero,zero,zero,xmm6[5],zero,zero,zero,xmm6[6],zero,zero,zero,xmm6[7],zero,zero,zero,xmm6[8],zero,zero,zero,xmm6[9],zero,zero,zero,xmm6[10],zero,zero,zero,xmm6[11],zero,zero,zero,xmm6[12],zero,zero,zero,xmm6[13],zero,zero,zero,xmm6[14],zero,zero,zero,xmm6[15],zero,zero,zero
; AVX512F-NEXT:    vptestmd %zmm5, %zmm6, %k3
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm0 {%k3} {z}
; AVX512F-NEXT:    vmovdqa64 %zmm0, (%rsp)
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm0 {%k3} {z} = -1
; AVX512F-NEXT:    vpsrld $31, %zmm0, %zmm6
; AVX512F-NEXT:    vextracti64x4 $1, %zmm6, %ymm7
; AVX512F-NEXT:    vpaddd %ymm7, %ymm6, %ymm6
; AVX512F-NEXT:    vextracti128 $1, %ymm6, %xmm7
; AVX512F-NEXT:    vpaddd %xmm7, %xmm6, %xmm6
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm7 = xmm6[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm7, %xmm6, %xmm6
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
; AVX512F-NEXT:    vpextrd $1, %xmm6, %eax
; AVX512F-NEXT:    vmovd %xmm6, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    andl $31, %ecx
; AVX512F-NEXT:    vpcompressd %zmm1, %zmm1 {%k2} {z}
; AVX512F-NEXT:    vmovdqa64 %zmm1, (%rsp,%rcx,4)
; AVX512F-NEXT:    vptestmd %zmm5, %zmm4, %k3
; AVX512F-NEXT:    vpcompressd %zmm2, %zmm1 {%k3} {z}
; AVX512F-NEXT:    vmovdqa64 %zmm1, {{[0-9]+}}(%rsp)
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm1 {%k3} {z} = -1
; AVX512F-NEXT:    vpsrld $31, %zmm1, %zmm1
; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
; AVX512F-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm2
; AVX512F-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpextrd $1, %xmm1, %eax
; AVX512F-NEXT:    vmovd %xmm1, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    andl $31, %ecx
; AVX512F-NEXT:    vpcompressd %zmm3, %zmm1 {%k1} {z}
; AVX512F-NEXT:    vmovdqa64 %zmm1, 128(%rsp,%rcx,4)
; AVX512F-NEXT:    vmovdqa64 (%rsp), %zmm1
; AVX512F-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm2
; AVX512F-NEXT:    vmovdqa64 %zmm1, {{[0-9]+}}(%rsp)
; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm1 {%k2} {z} = -1
; AVX512F-NEXT:    vpsrld $31, %zmm1, %zmm1
; AVX512F-NEXT:    vpsubd %zmm0, %zmm1, %zmm0
; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
; AVX512F-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX512F-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512F-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX512F-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512F-NEXT:    vpextrd $1, %xmm0, %eax
; AVX512F-NEXT:    vmovd %xmm0, %ecx
; AVX512F-NEXT:    addl %eax, %ecx
; AVX512F-NEXT:    andl $63, %ecx
; AVX512F-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm0
; AVX512F-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm1
; AVX512F-NEXT:    vmovaps %zmm0, 256(%rsp,%rcx,4)
; AVX512F-NEXT:    vmovaps %zmm2, {{[0-9]+}}(%rsp)
; AVX512F-NEXT:    vmovaps %zmm1, 320(%rsp,%rcx,4)
; AVX512F-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm0
; AVX512F-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm1
; AVX512F-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm2
; AVX512F-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm3
; AVX512F-NEXT:    movq %rbp, %rsp
; AVX512F-NEXT:    popq %rbp
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_large:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    pushq %rbp
; AVX512VL-NEXT:    movq %rsp, %rbp
; AVX512VL-NEXT:    andq $-64, %rsp
; AVX512VL-NEXT:    subq $512, %rsp # imm = 0x200
; AVX512VL-NEXT:    vpsllw $7, %zmm0, %zmm0
; AVX512VL-NEXT:    vpmovb2m %zmm0, %k3
; AVX512VL-NEXT:    kshiftrq $48, %k3, %k1
; AVX512VL-NEXT:    vpcompressd %zmm1, %zmm0 {%k3} {z}
; AVX512VL-NEXT:    vmovdqa64 %zmm0, (%rsp)
; AVX512VL-NEXT:    vpmovm2d %k3, %zmm0
; AVX512VL-NEXT:    vpsrld $31, %zmm0, %zmm1
; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm5
; AVX512VL-NEXT:    vpaddd %ymm5, %ymm1, %ymm1
; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm5
; AVX512VL-NEXT:    vpaddd %xmm5, %xmm1, %xmm1
; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm5 = xmm1[2,3,2,3]
; AVX512VL-NEXT:    vpaddd %xmm5, %xmm1, %xmm1
; AVX512VL-NEXT:    vpextrd $1, %xmm1, %eax
; AVX512VL-NEXT:    vmovd %xmm1, %ecx
; AVX512VL-NEXT:    addl %eax, %ecx
; AVX512VL-NEXT:    andl $31, %ecx
; AVX512VL-NEXT:    kshiftrq $16, %k3, %k2
; AVX512VL-NEXT:    vpcompressd %zmm2, %zmm1 {%k2} {z}
; AVX512VL-NEXT:    vmovdqa64 %zmm1, (%rsp,%rcx,4)
; AVX512VL-NEXT:    kshiftrq $32, %k3, %k3
; AVX512VL-NEXT:    vpcompressd %zmm3, %zmm1 {%k3} {z}
; AVX512VL-NEXT:    vmovdqa64 %zmm1, {{[0-9]+}}(%rsp)
; AVX512VL-NEXT:    vpmovm2d %k3, %zmm1
; AVX512VL-NEXT:    vpsrld $31, %zmm1, %zmm1
; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
; AVX512VL-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm2
; AVX512VL-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
; AVX512VL-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT:    vpextrd $1, %xmm1, %eax
; AVX512VL-NEXT:    vmovd %xmm1, %ecx
; AVX512VL-NEXT:    addl %eax, %ecx
; AVX512VL-NEXT:    andl $31, %ecx
; AVX512VL-NEXT:    vpcompressd %zmm4, %zmm1 {%k1} {z}
; AVX512VL-NEXT:    vmovdqa64 %zmm1, 128(%rsp,%rcx,4)
; AVX512VL-NEXT:    vmovaps (%rsp), %zmm1
; AVX512VL-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm2
; AVX512VL-NEXT:    vmovaps %zmm1, {{[0-9]+}}(%rsp)
; AVX512VL-NEXT:    vpmovm2d %k2, %zmm1
; AVX512VL-NEXT:    vpsrld $31, %zmm1, %zmm1
; AVX512VL-NEXT:    vpsubd %zmm0, %zmm1, %zmm0
; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
; AVX512VL-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX512VL-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX512VL-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512VL-NEXT:    vpextrd $1, %xmm0, %eax
; AVX512VL-NEXT:    vmovd %xmm0, %ecx
; AVX512VL-NEXT:    addl %eax, %ecx
; AVX512VL-NEXT:    andl $63, %ecx
; AVX512VL-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm0
; AVX512VL-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm1
; AVX512VL-NEXT:    vmovaps %zmm0, 256(%rsp,%rcx,4)
; AVX512VL-NEXT:    vmovaps %zmm2, {{[0-9]+}}(%rsp)
; AVX512VL-NEXT:    vmovaps %zmm1, 320(%rsp,%rcx,4)
; AVX512VL-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm0
; AVX512VL-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm1
; AVX512VL-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm2
; AVX512VL-NEXT:    vmovaps {{[0-9]+}}(%rsp), %zmm3
; AVX512VL-NEXT:    movq %rbp, %rsp
; AVX512VL-NEXT:    popq %rbp
; AVX512VL-NEXT:    retq
    %out = call <64 x i32> @llvm.experimental.vector.compress(<64 x i32> %vec, <64 x i1> %mask, <64 x i32> undef)
    ret <64 x i32> %out
}

define <8 x i64> @test_compress_knownbits_zext_v8i16_8i64(<8 x i16> %vec, <8 x  i1> %mask, <8 x i64> %passthru) nounwind {
; AVX2-LABEL: test_compress_knownbits_zext_v8i16_8i64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $64, %rsp
; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm0[2,3,2,3]
; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero
; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm5 = [3,3,3,3]
; AVX2-NEXT:    vandps %ymm5, %ymm2, %ymm2
; AVX2-NEXT:    vandps %ymm5, %ymm3, %ymm3
; AVX2-NEXT:    vmovaps %ymm3, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovaps %ymm2, (%rsp)
; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm2
; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero
; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,2,3]
; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero
; AVX2-NEXT:    vpaddq %ymm2, %ymm3, %ymm2
; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
; AVX2-NEXT:    vpaddq %xmm3, %xmm2, %xmm2
; AVX2-NEXT:    vpextrq $1, %xmm2, %rax
; AVX2-NEXT:    vmovd %xmm2, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vpextrw $4, %xmm1, %eax
; AVX2-NEXT:    andl $7, %ecx
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpextrw $1, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    vmovd %xmm1, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    addq %rsi, %rdx
; AVX2-NEXT:    vpextrw $2, %xmm1, %r8d
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addq %rdx, %r8
; AVX2-NEXT:    vpextrw $3, %xmm1, %r9d
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addq %r8, %r9
; AVX2-NEXT:    addq %r9, %rax
; AVX2-NEXT:    vpextrw $5, %xmm1, %edi
; AVX2-NEXT:    andl $1, %edi
; AVX2-NEXT:    addq %rax, %rdi
; AVX2-NEXT:    vpextrw $6, %xmm1, %r10d
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    addq %rdi, %r10
; AVX2-NEXT:    vpextrw $7, %xmm1, %r11d
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    addq %r10, %r11
; AVX2-NEXT:    vextracti128 $1, %ymm4, %xmm1
; AVX2-NEXT:    vpextrq $1, %xmm1, %rbx
; AVX2-NEXT:    cmpq $8, %r11
; AVX2-NEXT:    cmovbq (%rsp,%rcx,8), %rbx
; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
; AVX2-NEXT:    vmovq %xmm0, (%rsp)
; AVX2-NEXT:    vpextrq $1, %xmm0, (%rsp,%rsi,8)
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovq %xmm0, (%rsp,%rdx,8)
; AVX2-NEXT:    vpextrq $1, %xmm0, (%rsp,%r8,8)
; AVX2-NEXT:    andl $7, %r9d
; AVX2-NEXT:    vmovq %xmm4, (%rsp,%r9,8)
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    vpextrq $1, %xmm4, (%rsp,%rax,8)
; AVX2-NEXT:    andl $7, %edi
; AVX2-NEXT:    vmovq %xmm1, (%rsp,%rdi,8)
; AVX2-NEXT:    andl $7, %r10d
; AVX2-NEXT:    vpextrq $1, %xmm1, (%rsp,%r10,8)
; AVX2-NEXT:    cmpq $7, %r11
; AVX2-NEXT:    movl $7, %eax
; AVX2-NEXT:    cmovbq %r11, %rax
; AVX2-NEXT:    shll $3, %eax
; AVX2-NEXT:    movq %rbx, (%rsp,%rax)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm1
; AVX2-NEXT:    leaq -8(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_knownbits_zext_v8i16_8i64:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpmovsxwq %xmm1, %zmm1
; AVX512F-NEXT:    vpsllq $63, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vpmovzxwq {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
; AVX512F-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm2, %zmm0
; AVX512F-NEXT:    vpcompressq %zmm1, %zmm0 {%k1}
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_knownbits_zext_v8i16_8i64:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpsllw $15, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovw2m %xmm1, %k1
; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
; AVX512VL-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm2, %zmm0
; AVX512VL-NEXT:    vpcompressq %zmm1, %zmm0 {%k1}
; AVX512VL-NEXT:    retq
  %xvec = zext <8 x i16> %vec to <8 x i64> ;  0 -> 65535
  %xpassthru = and <8 x i64> %passthru, splat (i64 3) ; 0 -> 3
  %out = call <8 x i64> @llvm.experimental.vector.compress(<8 x i64> %xvec, <8 x  i1> %mask, <8 x i64> %xpassthru)
  %res = and <8 x i64> %out, splat (i64 65535) ; unnecessary - %out guaranteed to be 0 -> 65535
  ret <8 x i64> %res
}

define <8 x i64> @test_compress_knownbits_sext_v8i16_8i64(<8 x i16> %vec, <8 x i1> %mask, <8 x i64> %passthru) nounwind {
; AVX2-LABEL: test_compress_knownbits_sext_v8i16_8i64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    movq %rsp, %rbp
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    andq $-32, %rsp
; AVX2-NEXT:    subq $64, %rsp
; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm0[2,3,2,3]
; AVX2-NEXT:    vpmovsxwq %xmm4, %ymm4
; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm5 = [3,3,3,3]
; AVX2-NEXT:    vandps %ymm5, %ymm2, %ymm2
; AVX2-NEXT:    vandps %ymm5, %ymm3, %ymm3
; AVX2-NEXT:    vmovaps %ymm3, {{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovaps %ymm2, (%rsp)
; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm2
; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero
; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,2,3]
; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero
; AVX2-NEXT:    vpaddq %ymm2, %ymm3, %ymm2
; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
; AVX2-NEXT:    vpaddq %xmm3, %xmm2, %xmm2
; AVX2-NEXT:    vpextrq $1, %xmm2, %rax
; AVX2-NEXT:    vmovd %xmm2, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vpextrw $4, %xmm1, %eax
; AVX2-NEXT:    andl $7, %ecx
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpextrw $1, %xmm1, %edx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    vmovd %xmm1, %esi
; AVX2-NEXT:    andl $1, %esi
; AVX2-NEXT:    addq %rsi, %rdx
; AVX2-NEXT:    vpextrw $2, %xmm1, %r8d
; AVX2-NEXT:    andl $1, %r8d
; AVX2-NEXT:    addq %rdx, %r8
; AVX2-NEXT:    vpextrw $3, %xmm1, %r9d
; AVX2-NEXT:    andl $1, %r9d
; AVX2-NEXT:    addq %r8, %r9
; AVX2-NEXT:    addq %r9, %rax
; AVX2-NEXT:    vpextrw $5, %xmm1, %edi
; AVX2-NEXT:    andl $1, %edi
; AVX2-NEXT:    addq %rax, %rdi
; AVX2-NEXT:    vpextrw $6, %xmm1, %r10d
; AVX2-NEXT:    andl $1, %r10d
; AVX2-NEXT:    addq %rdi, %r10
; AVX2-NEXT:    vpextrw $7, %xmm1, %r11d
; AVX2-NEXT:    andl $1, %r11d
; AVX2-NEXT:    addq %r10, %r11
; AVX2-NEXT:    vextracti128 $1, %ymm4, %xmm1
; AVX2-NEXT:    vpextrq $1, %xmm1, %rbx
; AVX2-NEXT:    cmpq $8, %r11
; AVX2-NEXT:    cmovbq (%rsp,%rcx,8), %rbx
; AVX2-NEXT:    vpmovsxwq %xmm0, %ymm0
; AVX2-NEXT:    vmovq %xmm0, (%rsp)
; AVX2-NEXT:    vpextrq $1, %xmm0, (%rsp,%rsi,8)
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovq %xmm0, (%rsp,%rdx,8)
; AVX2-NEXT:    vpextrq $1, %xmm0, (%rsp,%r8,8)
; AVX2-NEXT:    andl $7, %r9d
; AVX2-NEXT:    vmovq %xmm4, (%rsp,%r9,8)
; AVX2-NEXT:    andl $7, %eax
; AVX2-NEXT:    vpextrq $1, %xmm4, (%rsp,%rax,8)
; AVX2-NEXT:    andl $7, %edi
; AVX2-NEXT:    vmovq %xmm1, (%rsp,%rdi,8)
; AVX2-NEXT:    andl $7, %r10d
; AVX2-NEXT:    vpextrq $1, %xmm1, (%rsp,%r10,8)
; AVX2-NEXT:    cmpq $7, %r11
; AVX2-NEXT:    movl $7, %eax
; AVX2-NEXT:    cmovbq %r11, %rax
; AVX2-NEXT:    shll $3, %eax
; AVX2-NEXT:    movq %rbx, (%rsp,%rax)
; AVX2-NEXT:    vmovaps (%rsp), %ymm0
; AVX2-NEXT:    vmovaps {{[0-9]+}}(%rsp), %ymm1
; AVX2-NEXT:    leaq -8(%rbp), %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_knownbits_sext_v8i16_8i64:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpmovsxwq %xmm1, %zmm1
; AVX512F-NEXT:    vpsllq $63, %zmm1, %zmm1
; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k1
; AVX512F-NEXT:    vpmovsxwq %xmm0, %zmm1
; AVX512F-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm2, %zmm0
; AVX512F-NEXT:    vpcompressq %zmm1, %zmm0 {%k1}
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_knownbits_sext_v8i16_8i64:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpsllw $15, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovw2m %xmm1, %k1
; AVX512VL-NEXT:    vpmovsxwq %xmm0, %zmm1
; AVX512VL-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm2, %zmm0
; AVX512VL-NEXT:    vpcompressq %zmm1, %zmm0 {%k1}
; AVX512VL-NEXT:    retq
  %xvec = sext <8 x i16> %vec to <8 x i64> ; sign extend vec
  %xpassthru = and <8 x i64> %passthru, splat(i64 3)
  %out = call <8 x i64> @llvm.experimental.vector.compress(<8 x i64> %xvec, <8 x i1> %mask, <8 x i64> %xpassthru)
  %shl = shl <8 x i64> %out, splat(i64 48)
  %res = ashr <8 x i64> %shl, splat(i64 48)
  ret <8 x i64> %res
}

define <4 x i32> @test_compress_all_const() nounwind {
; AVX2-LABEL: test_compress_all_const:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovsd {{.*#+}} xmm0 = [5,9,0,0]
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_all_const:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovsd {{.*#+}} xmm0 = [5,9,0,0]
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_all_const:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vmovddup {{.*#+}} xmm0 = [5,9,5,9]
; AVX512VL-NEXT:    # xmm0 = mem[0,0]
; AVX512VL-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> <i32 3, i32 5, i32 7, i32 9>,
                                                <4 x i1>   <i1 0,  i1 1,  i1 0,  i1 1>,
                                                <4 x i32> undef)
    ret <4 x i32> %out
}

define <4 x i32> @test_compress_const_mask(<4 x i32> %vec) nounwind {
; CHECK-LABEL: test_compress_const_mask:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,3,2,3]
; CHECK-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> <i1 1, i1 undef, i1 0, i1 1>, <4 x i32> undef)
    ret <4 x i32> %out
}

define <4 x i32> @test_compress_const_mask_passthrough(<4 x i32> %vec, <4 x i32> %passthru) nounwind {
; CHECK-LABEL: test_compress_const_mask_passthrough:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,3],xmm1[2,3]
; CHECK-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> <i1 1, i1 undef, i1 0, i1 1>, <4 x i32> %passthru)
    ret <4 x i32> %out
}

define <4 x i32> @test_compress_const_mask_const_passthrough(<4 x i32> %vec) nounwind {
; CHECK-LABEL: test_compress_const_mask_const_passthrough:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,3,2,3]
; CHECK-NEXT:    movabsq $34359738375, %rax # imm = 0x800000007
; CHECK-NEXT:    vpinsrq $1, %rax, %xmm0, %xmm0
; CHECK-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> <i1 1, i1 0, i1 0, i1 1>, <4 x i32> <i32 5, i32 6, i32 7, i32 8>)
    ret <4 x i32> %out
}

; We pass a placeholder value for the const_mask* tests to check that they are converted to a no-op by simply copying
; the second vector input register to the return register or doing nothing.
define <4 x i32> @test_compress_const_splat1_mask(<4 x i32> %ignore, <4 x i32> %vec) nounwind {
; CHECK-LABEL: test_compress_const_splat1_mask:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> splat (i1 -1), <4 x i32> undef)
    ret <4 x i32> %out
}
define <4 x i32> @test_compress_const_splat0_mask(<4 x i32> %ignore, <4 x i32> %vec) nounwind {
; CHECK-LABEL: test_compress_const_splat0_mask:
; CHECK:       # %bb.0:
; CHECK-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> splat (i1 0), <4 x i32> undef)
    ret <4 x i32> %out
}
define <4 x i32> @test_compress_undef_mask(<4 x i32> %ignore, <4 x i32> %vec) nounwind {
; CHECK-LABEL: test_compress_undef_mask:
; CHECK:       # %bb.0:
; CHECK-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> undef, <4 x i32> undef)
    ret <4 x i32> %out
}
define <4 x i32> @test_compress_const_splat0_mask_with_passthru(<4 x i32> %ignore, <4 x i32> %vec, <4 x i32> %passthru) nounwind {
; CHECK-LABEL: test_compress_const_splat0_mask_with_passthru:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> splat (i1 0), <4 x i32> %passthru)
    ret <4 x i32> %out
}
define <4 x i32> @test_compress_const_splat0_mask_without_passthru(<4 x i32> %ignore, <4 x i32> %vec) nounwind {
; CHECK-LABEL: test_compress_const_splat0_mask_without_passthru:
; CHECK:       # %bb.0:
; CHECK-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> splat (i1 0), <4 x i32> undef)
    ret <4 x i32> %out
}

define <4 x i8> @test_compress_small(<4 x i8> %vec, <4 x i1> %mask) nounwind {
; AVX2-LABEL: test_compress_small:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,4,8,12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
; AVX2-NEXT:    vpsllw $7, %xmm1, %xmm1
; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX2-NEXT:    vpcmpgtb %xmm1, %xmm2, %xmm1
; AVX2-NEXT:    vpextrb $0, %xmm0, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovd %xmm1, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpextrb $1, %xmm0, -24(%rsp,%rax)
; AVX2-NEXT:    vpextrb $1, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vpextrb $2, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $2, %xmm1, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    vpextrb $3, %xmm0, -24(%rsp,%rax)
; AVX2-NEXT:    vpextrb $3, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vpextrb $4, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    andl $15, %ecx
; AVX2-NEXT:    vpextrb $5, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $6, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $7, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $8, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $9, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $10, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $11, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $12, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $13, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $14, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vpextrb $15, %xmm0, -24(%rsp,%rcx)
; AVX2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_small:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
; AVX512F-NEXT:    kshiftrw $12, %k0, %k1
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-ONLY-LABEL: test_compress_small:
; AVX512VL-ONLY:       # %bb.0:
; AVX512VL-ONLY-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512VL-ONLY-NEXT:    vpmovd2m %xmm1, %k1
; AVX512VL-ONLY-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512VL-ONLY-NEXT:    vpcompressd %zmm0, %zmm0 {%k1} {z}
; AVX512VL-ONLY-NEXT:    vpmovdb %zmm0, %xmm0
; AVX512VL-ONLY-NEXT:    vzeroupper
; AVX512VL-ONLY-NEXT:    retq
;
; AVX512VL-VBMI2-LABEL: test_compress_small:
; AVX512VL-VBMI2:       # %bb.0:
; AVX512VL-VBMI2-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512VL-VBMI2-NEXT:    vpmovd2m %xmm1, %k1
; AVX512VL-VBMI2-NEXT:    vpcompressb %xmm0, %xmm0 {%k1} {z}
; AVX512VL-VBMI2-NEXT:    retq
    %out = call <4 x i8> @llvm.experimental.vector.compress(<4 x i8> %vec, <4 x i1> %mask, <4 x i8> undef)
    ret <4 x i8> %out
}

define <4 x i4> @test_compress_illegal_element_type(<4 x i4> %vec, <4 x i1> %mask) nounwind {
; AVX2-LABEL: test_compress_illegal_element_type:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX2-NEXT:    vpsrad $31, %xmm1, %xmm1
; AVX2-NEXT:    vmovss %xmm0, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovd %xmm1, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, -24(%rsp,%rax,4)
; AVX2-NEXT:    vpextrd $1, %xmm1, %ecx
; AVX2-NEXT:    subl %ecx, %eax
; AVX2-NEXT:    movl %eax, %ecx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    vextractps $2, %xmm0, -24(%rsp,%rcx,4)
; AVX2-NEXT:    vpextrd $2, %xmm1, %ecx
; AVX2-NEXT:    subl %ecx, %eax
; AVX2-NEXT:    andl $3, %eax
; AVX2-NEXT:    vextractps $3, %xmm0, -24(%rsp,%rax,4)
; AVX2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_illegal_element_type:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
; AVX512F-NEXT:    kshiftrw $12, %k0, %k1
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_illegal_element_type:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovd2m %xmm1, %k1
; AVX512VL-NEXT:    vpcompressd %xmm0, %xmm0 {%k1} {z}
; AVX512VL-NEXT:    retq
    %out = call <4 x i4> @llvm.experimental.vector.compress(<4 x i4> %vec, <4 x i1> %mask, <4 x i4> undef)
    ret <4 x i4> %out
}

define <3 x i32> @test_compress_narrow(<3 x i32> %vec, <3 x i1> %mask) nounwind {
; AVX2-LABEL: test_compress_narrow:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovd %edi, %xmm1
; AVX2-NEXT:    vpinsrd $1, %esi, %xmm1, %xmm1
; AVX2-NEXT:    vpinsrd $2, %edx, %xmm1, %xmm1
; AVX2-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX2-NEXT:    vpsrad $31, %xmm1, %xmm1
; AVX2-NEXT:    vmovss %xmm0, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovd %xmm1, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, -24(%rsp,%rax,4)
; AVX2-NEXT:    vpextrd $1, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    vpextrd $2, %xmm1, %edx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addl %ecx, %edx
; AVX2-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    vextractps $2, %xmm0, -24(%rsp,%rcx,4)
; AVX2-NEXT:    andl $3, %edx
; AVX2-NEXT:    vextractps $3, %xmm0, -24(%rsp,%rdx,4)
; AVX2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_narrow:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovd %edi, %xmm1
; AVX512F-NEXT:    vpinsrb $1, %esi, %xmm1, %xmm1
; AVX512F-NEXT:    vpinsrb $2, %edx, %xmm1, %xmm1
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
; AVX512F-NEXT:    movb $7, %al
; AVX512F-NEXT:    kmovw %eax, %k1
; AVX512F-NEXT:    vptestmd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm1, %k0 {%k1}
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
; AVX512F-NEXT:    kshiftrw $12, %k0, %k1
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_narrow:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vmovd %edi, %xmm1
; AVX512VL-NEXT:    vpinsrb $1, %esi, %xmm1, %xmm1
; AVX512VL-NEXT:    vpinsrb $2, %edx, %xmm1, %xmm1
; AVX512VL-NEXT:    movb $7, %al
; AVX512VL-NEXT:    kmovd %eax, %k1
; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
; AVX512VL-NEXT:    vptestmd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm1, %k1 {%k1}
; AVX512VL-NEXT:    vpcompressd %xmm0, %xmm0 {%k1} {z}
; AVX512VL-NEXT:    retq
    %out = call <3 x i32> @llvm.experimental.vector.compress(<3 x i32> %vec, <3 x i1> %mask, <3 x i32> undef)
    ret <3 x i32> %out
}

define <3 x i3> @test_compress_narrow_illegal_element_type(<3 x i3> %vec, <3 x i1> %mask) nounwind {
; AVX2-LABEL: test_compress_narrow_illegal_element_type:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovd %ecx, %xmm0
; AVX2-NEXT:    vpinsrd $1, %r8d, %xmm0, %xmm0
; AVX2-NEXT:    vpslld $31, %xmm0, %xmm0
; AVX2-NEXT:    vpsrad $31, %xmm0, %xmm0
; AVX2-NEXT:    movl %edi, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovd %xmm0, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    movl %esi, -24(%rsp,%rax,4)
; AVX2-NEXT:    vpextrd $1, %xmm0, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    andl $3, %ecx
; AVX2-NEXT:    movl %edx, -24(%rsp,%rcx,4)
; AVX2-NEXT:    vmovdqa -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    vmovd %xmm0, %eax
; AVX2-NEXT:    vpextrb $4, %xmm0, %edx
; AVX2-NEXT:    vpextrb $8, %xmm0, %ecx
; AVX2-NEXT:    # kill: def $al killed $al killed $eax
; AVX2-NEXT:    # kill: def $dl killed $dl killed $edx
; AVX2-NEXT:    # kill: def $cl killed $cl killed $ecx
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_narrow_illegal_element_type:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovd %ecx, %xmm0
; AVX512F-NEXT:    vpinsrb $1, %r8d, %xmm0, %xmm0
; AVX512F-NEXT:    vpinsrb $2, %r9d, %xmm0, %xmm0
; AVX512F-NEXT:    vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
; AVX512F-NEXT:    movb $7, %al
; AVX512F-NEXT:    kmovw %eax, %k1
; AVX512F-NEXT:    vptestmd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %k0 {%k1}
; AVX512F-NEXT:    vmovd %edi, %xmm0
; AVX512F-NEXT:    vpinsrd $1, %esi, %xmm0, %xmm0
; AVX512F-NEXT:    vpinsrd $2, %edx, %xmm0, %xmm0
; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
; AVX512F-NEXT:    kshiftrw $12, %k0, %k1
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT:    vmovd %xmm0, %eax
; AVX512F-NEXT:    vpextrb $4, %xmm0, %edx
; AVX512F-NEXT:    vpextrb $8, %xmm0, %ecx
; AVX512F-NEXT:    # kill: def $al killed $al killed $eax
; AVX512F-NEXT:    # kill: def $dl killed $dl killed $edx
; AVX512F-NEXT:    # kill: def $cl killed $cl killed $ecx
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_narrow_illegal_element_type:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    movb $7, %al
; AVX512VL-NEXT:    kmovd %eax, %k1
; AVX512VL-NEXT:    vmovd %ecx, %xmm0
; AVX512VL-NEXT:    vpinsrb $1, %r8d, %xmm0, %xmm0
; AVX512VL-NEXT:    vpinsrb $2, %r9d, %xmm0, %xmm0
; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
; AVX512VL-NEXT:    vptestmd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %k1 {%k1}
; AVX512VL-NEXT:    vmovd %edi, %xmm0
; AVX512VL-NEXT:    vpinsrd $1, %esi, %xmm0, %xmm0
; AVX512VL-NEXT:    vpinsrd $2, %edx, %xmm0, %xmm0
; AVX512VL-NEXT:    vpcompressd %xmm0, %xmm0 {%k1} {z}
; AVX512VL-NEXT:    vmovd %xmm0, %eax
; AVX512VL-NEXT:    vpextrb $4, %xmm0, %edx
; AVX512VL-NEXT:    vpextrb $8, %xmm0, %ecx
; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
; AVX512VL-NEXT:    # kill: def $dl killed $dl killed $edx
; AVX512VL-NEXT:    # kill: def $cl killed $cl killed $ecx
; AVX512VL-NEXT:    retq
    %out = call <3 x i3> @llvm.experimental.vector.compress(<3 x i3> %vec, <3 x i1> %mask, <3 x i3> undef)
    ret <3 x i3> %out
}

define <4 x i32> @test_compress_v4i32_zero_passthru(<4 x i32> %vec, <4 x i1> %mask) nounwind {
; AVX2-LABEL: test_compress_v4i32_zero_passthru:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX2-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; AVX2-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vpsrad $31, %xmm1, %xmm1
; AVX2-NEXT:    vmovss %xmm0, -{{[0-9]+}}(%rsp)
; AVX2-NEXT:    vmovd %xmm1, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vextractps $1, %xmm0, -24(%rsp,%rax,4)
; AVX2-NEXT:    vpextrd $1, %xmm1, %ecx
; AVX2-NEXT:    andl $1, %ecx
; AVX2-NEXT:    addq %rax, %rcx
; AVX2-NEXT:    vextractps $2, %xmm0, -24(%rsp,%rcx,4)
; AVX2-NEXT:    vpextrd $2, %xmm1, %eax
; AVX2-NEXT:    andl $1, %eax
; AVX2-NEXT:    vpextrd $3, %xmm1, %edx
; AVX2-NEXT:    addq %rcx, %rax
; AVX2-NEXT:    andl $1, %edx
; AVX2-NEXT:    addq %rax, %rdx
; AVX2-NEXT:    # kill: def $eax killed $eax killed $rax def $rax
; AVX2-NEXT:    andl $3, %eax
; AVX2-NEXT:    vextractps $3, %xmm0, -24(%rsp,%rax,4)
; AVX2-NEXT:    cmpq $3, %rdx
; AVX2-NEXT:    movl $3, %eax
; AVX2-NEXT:    cmovbq %rdx, %rax
; AVX2-NEXT:    andl $3, %eax
; AVX2-NEXT:    vextractps $3, %xmm0, %ecx
; AVX2-NEXT:    xorl %esi, %esi
; AVX2-NEXT:    cmpq $4, %rdx
; AVX2-NEXT:    cmovael %ecx, %esi
; AVX2-NEXT:    movl %esi, -24(%rsp,%rax,4)
; AVX2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
; AVX2-NEXT:    retq
;
; AVX512F-LABEL: test_compress_v4i32_zero_passthru:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
; AVX512F-NEXT:    kshiftrw $12, %k0, %k1
; AVX512F-NEXT:    vpcompressd %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: test_compress_v4i32_zero_passthru:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
; AVX512VL-NEXT:    vpmovd2m %xmm1, %k1
; AVX512VL-NEXT:    vpcompressd %xmm0, %xmm0 {%k1} {z}
; AVX512VL-NEXT:    retq
    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> %mask, <4 x i32> zeroinitializer)
    ret <4 x i32> %out
}