TheRock/compiler/amd-llvm/llvm/test/CodeGen/X86/masked_gather.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=x86_64-unknown-linux-gnu  -mattr=+sse4.2 < %s | FileCheck %s --check-prefixes=SSE
; RUN: llc -mtriple=x86_64-unknown-linux-gnu  -mattr=+avx < %s | FileCheck %s --check-prefixes=AVX1
; RUN: llc -mtriple=x86_64-unknown-linux-gnu  -mattr=+avx2 < %s | FileCheck %s --check-prefixes=AVX2,AVX2-NOGATHER
; RUN: llc -mtriple=x86_64-unknown-linux-gnu  -mattr=+avx2,+fast-gather < %s | FileCheck %s --check-prefixes=AVX2,AVX2-GATHER
; RUN: llc -mtriple=x86_64-unknown-linux-gnu  -mattr=+avx512f < %s | FileCheck %s --check-prefixes=AVX512,AVX512F
; RUN: llc -mtriple=x86_64-unknown-linux-gnu  -mattr=+avx512f,+avx512vl < %s | FileCheck %s --check-prefixes=AVX512,AVX512VL

;
; vXf32
;

%struct.a = type { [4 x i32], [4 x i8], %struct.b, i32 }
%struct.b = type { i32, i32 }
@c = external dso_local global %struct.a, align 4

define <4 x float> @gather_v4f32_ptr_v4i32(<4 x ptr> %ptr, <4 x i32> %trigger, <4 x float> %passthru) {
; SSE-LABEL: gather_v4f32_ptr_v4i32:
; SSE:       # %bb.0:
; SSE-NEXT:    pxor %xmm4, %xmm4
; SSE-NEXT:    pcmpeqd %xmm2, %xmm4
; SSE-NEXT:    movmskps %xmm4, %eax
; SSE-NEXT:    testb $1, %al
; SSE-NEXT:    jne .LBB0_1
; SSE-NEXT:  # %bb.2: # %else
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    jne .LBB0_3
; SSE-NEXT:  .LBB0_4: # %else2
; SSE-NEXT:    testb $4, %al
; SSE-NEXT:    jne .LBB0_5
; SSE-NEXT:  .LBB0_6: # %else5
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    jne .LBB0_7
; SSE-NEXT:  .LBB0_8: # %else8
; SSE-NEXT:    movaps %xmm3, %xmm0
; SSE-NEXT:    retq
; SSE-NEXT:  .LBB0_1: # %cond.load
; SSE-NEXT:    movq %xmm0, %rcx
; SSE-NEXT:    movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; SSE-NEXT:    pblendw {{.*#+}} xmm3 = xmm2[0,1],xmm3[2,3,4,5,6,7]
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    je .LBB0_4
; SSE-NEXT:  .LBB0_3: # %cond.load1
; SSE-NEXT:    pextrq $1, %xmm0, %rcx
; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0],mem[0],xmm3[2,3]
; SSE-NEXT:    testb $4, %al
; SSE-NEXT:    je .LBB0_6
; SSE-NEXT:  .LBB0_5: # %cond.load4
; SSE-NEXT:    movq %xmm1, %rcx
; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0,1],mem[0],xmm3[3]
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    je .LBB0_8
; SSE-NEXT:  .LBB0_7: # %cond.load7
; SSE-NEXT:    pextrq $1, %xmm1, %rax
; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0,1,2],mem[0]
; SSE-NEXT:    movaps %xmm3, %xmm0
; SSE-NEXT:    retq
;
; AVX1-LABEL: gather_v4f32_ptr_v4i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
; AVX1-NEXT:    vmovmskps %xmm1, %eax
; AVX1-NEXT:    testb $1, %al
; AVX1-NEXT:    je .LBB0_2
; AVX1-NEXT:  # %bb.1: # %cond.load
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX1-NEXT:    vmovss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
; AVX1-NEXT:  .LBB0_2: # %else
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    je .LBB0_4
; AVX1-NEXT:  # %bb.3: # %cond.load1
; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; AVX1-NEXT:  .LBB0_4: # %else2
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    jne .LBB0_5
; AVX1-NEXT:  # %bb.6: # %else5
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    jne .LBB0_7
; AVX1-NEXT:  .LBB0_8: # %else8
; AVX1-NEXT:    vmovaps %xmm2, %xmm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
; AVX1-NEXT:  .LBB0_5: # %cond.load4
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    je .LBB0_8
; AVX1-NEXT:  .LBB0_7: # %cond.load7
; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; AVX1-NEXT:    vmovaps %xmm2, %xmm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-NOGATHER-LABEL: gather_v4f32_ptr_v4i32:
; AVX2-NOGATHER:       # %bb.0:
; AVX2-NOGATHER-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-NOGATHER-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
; AVX2-NOGATHER-NEXT:    vmovmskps %xmm1, %eax
; AVX2-NOGATHER-NEXT:    testb $1, %al
; AVX2-NOGATHER-NEXT:    je .LBB0_2
; AVX2-NOGATHER-NEXT:  # %bb.1: # %cond.load
; AVX2-NOGATHER-NEXT:    vmovq %xmm0, %rcx
; AVX2-NOGATHER-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX2-NOGATHER-NEXT:    vmovss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
; AVX2-NOGATHER-NEXT:  .LBB0_2: # %else
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    je .LBB0_4
; AVX2-NOGATHER-NEXT:  # %bb.3: # %cond.load1
; AVX2-NOGATHER-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; AVX2-NOGATHER-NEXT:  .LBB0_4: # %else2
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NOGATHER-NEXT:    jne .LBB0_5
; AVX2-NOGATHER-NEXT:  # %bb.6: # %else5
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    jne .LBB0_7
; AVX2-NOGATHER-NEXT:  .LBB0_8: # %else8
; AVX2-NOGATHER-NEXT:    vmovaps %xmm2, %xmm0
; AVX2-NOGATHER-NEXT:    vzeroupper
; AVX2-NOGATHER-NEXT:    retq
; AVX2-NOGATHER-NEXT:  .LBB0_5: # %cond.load4
; AVX2-NOGATHER-NEXT:    vmovq %xmm0, %rcx
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    je .LBB0_8
; AVX2-NOGATHER-NEXT:  .LBB0_7: # %cond.load7
; AVX2-NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; AVX2-NOGATHER-NEXT:    vmovaps %xmm2, %xmm0
; AVX2-NOGATHER-NEXT:    vzeroupper
; AVX2-NOGATHER-NEXT:    retq
;
; AVX2-GATHER-LABEL: gather_v4f32_ptr_v4i32:
; AVX2-GATHER:       # %bb.0:
; AVX2-GATHER-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-GATHER-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
; AVX2-GATHER-NEXT:    vgatherqps %xmm1, (,%ymm0), %xmm2
; AVX2-GATHER-NEXT:    vmovaps %xmm2, %xmm0
; AVX2-GATHER-NEXT:    vzeroupper
; AVX2-GATHER-NEXT:    retq
;
; AVX512F-LABEL: gather_v4f32_ptr_v4i32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512F-NEXT:    vptestnmd %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kmovw %k0, %eax
; AVX512F-NEXT:    testb $1, %al
; AVX512F-NEXT:    je .LBB0_2
; AVX512F-NEXT:  # %bb.1: # %cond.load
; AVX512F-NEXT:    vmovq %xmm0, %rcx
; AVX512F-NEXT:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX512F-NEXT:    vpblendd {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
; AVX512F-NEXT:  .LBB0_2: # %else
; AVX512F-NEXT:    testb $2, %al
; AVX512F-NEXT:    je .LBB0_4
; AVX512F-NEXT:  # %bb.3: # %cond.load1
; AVX512F-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; AVX512F-NEXT:  .LBB0_4: # %else2
; AVX512F-NEXT:    testb $4, %al
; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX512F-NEXT:    jne .LBB0_5
; AVX512F-NEXT:  # %bb.6: # %else5
; AVX512F-NEXT:    testb $8, %al
; AVX512F-NEXT:    jne .LBB0_7
; AVX512F-NEXT:  .LBB0_8: # %else8
; AVX512F-NEXT:    vmovaps %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
; AVX512F-NEXT:  .LBB0_5: # %cond.load4
; AVX512F-NEXT:    vmovq %xmm0, %rcx
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; AVX512F-NEXT:    testb $8, %al
; AVX512F-NEXT:    je .LBB0_8
; AVX512F-NEXT:  .LBB0_7: # %cond.load7
; AVX512F-NEXT:    vpextrq $1, %xmm0, %rax
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; AVX512F-NEXT:    vmovaps %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: gather_v4f32_ptr_v4i32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vptestnmd %xmm1, %xmm1, %k1
; AVX512VL-NEXT:    vgatherqps (,%ymm0), %xmm2 {%k1}
; AVX512VL-NEXT:    vmovaps %xmm2, %xmm0
; AVX512VL-NEXT:    vzeroupper
; AVX512VL-NEXT:    retq
  %mask = icmp eq <4 x i32> %trigger, zeroinitializer
  %res = call <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr> %ptr, i32 4, <4 x i1> %mask, <4 x float> %passthru)
  ret <4 x float> %res
}

define <4 x float> @masked_gather_v4f32_ptr_v4i32(<4 x ptr> %ptr, i32 %trigger, <4 x float> %passthru) {
; SSE-LABEL: masked_gather_v4f32_ptr_v4i32:
; SSE:       # %bb.0:
; SSE-NEXT:    testb $1, %dil
; SSE-NEXT:    jne .LBB1_1
; SSE-NEXT:  # %bb.2: # %else
; SSE-NEXT:    testb $2, %dil
; SSE-NEXT:    jne .LBB1_3
; SSE-NEXT:  .LBB1_4: # %else2
; SSE-NEXT:    testb $4, %dil
; SSE-NEXT:    jne .LBB1_5
; SSE-NEXT:  .LBB1_6: # %else5
; SSE-NEXT:    testb $8, %dil
; SSE-NEXT:    jne .LBB1_7
; SSE-NEXT:  .LBB1_8: # %else8
; SSE-NEXT:    movaps %xmm2, %xmm0
; SSE-NEXT:    retq
; SSE-NEXT:  .LBB1_1: # %cond.load
; SSE-NEXT:    movq %xmm0, %rax
; SSE-NEXT:    movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; SSE-NEXT:    movss {{.*#+}} xmm2 = xmm3[0],xmm2[1,2,3]
; SSE-NEXT:    testb $2, %dil
; SSE-NEXT:    je .LBB1_4
; SSE-NEXT:  .LBB1_3: # %cond.load1
; SSE-NEXT:    pextrq $1, %xmm0, %rax
; SSE-NEXT:    insertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; SSE-NEXT:    testb $4, %dil
; SSE-NEXT:    je .LBB1_6
; SSE-NEXT:  .LBB1_5: # %cond.load4
; SSE-NEXT:    movq %xmm1, %rax
; SSE-NEXT:    insertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; SSE-NEXT:    testb $8, %dil
; SSE-NEXT:    je .LBB1_8
; SSE-NEXT:  .LBB1_7: # %cond.load7
; SSE-NEXT:    pextrq $1, %xmm1, %rax
; SSE-NEXT:    insertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; SSE-NEXT:    movaps %xmm2, %xmm0
; SSE-NEXT:    retq
;
; AVX1-LABEL: masked_gather_v4f32_ptr_v4i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    testb $1, %dil
; AVX1-NEXT:    je .LBB1_2
; AVX1-NEXT:  # %bb.1: # %cond.load
; AVX1-NEXT:    vmovq %xmm0, %rax
; AVX1-NEXT:    vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX1-NEXT:    vmovss {{.*#+}} xmm1 = xmm2[0],xmm1[1,2,3]
; AVX1-NEXT:  .LBB1_2: # %else
; AVX1-NEXT:    testb $2, %dil
; AVX1-NEXT:    je .LBB1_4
; AVX1-NEXT:  # %bb.3: # %cond.load1
; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
; AVX1-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[2,3]
; AVX1-NEXT:  .LBB1_4: # %else2
; AVX1-NEXT:    testb $4, %dil
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    jne .LBB1_5
; AVX1-NEXT:  # %bb.6: # %else5
; AVX1-NEXT:    testb $8, %dil
; AVX1-NEXT:    jne .LBB1_7
; AVX1-NEXT:  .LBB1_8: # %else8
; AVX1-NEXT:    vmovaps %xmm1, %xmm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
; AVX1-NEXT:  .LBB1_5: # %cond.load4
; AVX1-NEXT:    vmovq %xmm0, %rax
; AVX1-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1],mem[0],xmm1[3]
; AVX1-NEXT:    testb $8, %dil
; AVX1-NEXT:    je .LBB1_8
; AVX1-NEXT:  .LBB1_7: # %cond.load7
; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
; AVX1-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],mem[0]
; AVX1-NEXT:    vmovaps %xmm1, %xmm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-NOGATHER-LABEL: masked_gather_v4f32_ptr_v4i32:
; AVX2-NOGATHER:       # %bb.0:
; AVX2-NOGATHER-NEXT:    testb $1, %dil
; AVX2-NOGATHER-NEXT:    je .LBB1_2
; AVX2-NOGATHER-NEXT:  # %bb.1: # %cond.load
; AVX2-NOGATHER-NEXT:    vmovq %xmm0, %rax
; AVX2-NOGATHER-NEXT:    vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX2-NOGATHER-NEXT:    vmovss {{.*#+}} xmm1 = xmm2[0],xmm1[1,2,3]
; AVX2-NOGATHER-NEXT:  .LBB1_2: # %else
; AVX2-NOGATHER-NEXT:    testb $2, %dil
; AVX2-NOGATHER-NEXT:    je .LBB1_4
; AVX2-NOGATHER-NEXT:  # %bb.3: # %cond.load1
; AVX2-NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[2,3]
; AVX2-NOGATHER-NEXT:  .LBB1_4: # %else2
; AVX2-NOGATHER-NEXT:    testb $4, %dil
; AVX2-NOGATHER-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NOGATHER-NEXT:    jne .LBB1_5
; AVX2-NOGATHER-NEXT:  # %bb.6: # %else5
; AVX2-NOGATHER-NEXT:    testb $8, %dil
; AVX2-NOGATHER-NEXT:    jne .LBB1_7
; AVX2-NOGATHER-NEXT:  .LBB1_8: # %else8
; AVX2-NOGATHER-NEXT:    vmovaps %xmm1, %xmm0
; AVX2-NOGATHER-NEXT:    vzeroupper
; AVX2-NOGATHER-NEXT:    retq
; AVX2-NOGATHER-NEXT:  .LBB1_5: # %cond.load4
; AVX2-NOGATHER-NEXT:    vmovq %xmm0, %rax
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1],mem[0],xmm1[3]
; AVX2-NOGATHER-NEXT:    testb $8, %dil
; AVX2-NOGATHER-NEXT:    je .LBB1_8
; AVX2-NOGATHER-NEXT:  .LBB1_7: # %cond.load7
; AVX2-NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],mem[0]
; AVX2-NOGATHER-NEXT:    vmovaps %xmm1, %xmm0
; AVX2-NOGATHER-NEXT:    vzeroupper
; AVX2-NOGATHER-NEXT:    retq
;
; AVX2-GATHER-LABEL: masked_gather_v4f32_ptr_v4i32:
; AVX2-GATHER:       # %bb.0:
; AVX2-GATHER-NEXT:    vmovd %edi, %xmm2
; AVX2-GATHER-NEXT:    vpbroadcastd %xmm2, %xmm2
; AVX2-GATHER-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [1,2,4,8]
; AVX2-GATHER-NEXT:    vpand %xmm3, %xmm2, %xmm2
; AVX2-GATHER-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
; AVX2-GATHER-NEXT:    vgatherqps %xmm2, (,%ymm0), %xmm1
; AVX2-GATHER-NEXT:    vmovaps %xmm1, %xmm0
; AVX2-GATHER-NEXT:    vzeroupper
; AVX2-GATHER-NEXT:    retq
;
; AVX512F-LABEL: masked_gather_v4f32_ptr_v4i32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    testb $1, %dil
; AVX512F-NEXT:    je .LBB1_2
; AVX512F-NEXT:  # %bb.1: # %cond.load
; AVX512F-NEXT:    vmovq %xmm0, %rax
; AVX512F-NEXT:    vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX512F-NEXT:    vmovss {{.*#+}} xmm1 = xmm2[0],xmm1[1,2,3]
; AVX512F-NEXT:  .LBB1_2: # %else
; AVX512F-NEXT:    testb $2, %dil
; AVX512F-NEXT:    je .LBB1_4
; AVX512F-NEXT:  # %bb.3: # %cond.load1
; AVX512F-NEXT:    vpextrq $1, %xmm0, %rax
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[2,3]
; AVX512F-NEXT:  .LBB1_4: # %else2
; AVX512F-NEXT:    testb $4, %dil
; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX512F-NEXT:    jne .LBB1_5
; AVX512F-NEXT:  # %bb.6: # %else5
; AVX512F-NEXT:    testb $8, %dil
; AVX512F-NEXT:    jne .LBB1_7
; AVX512F-NEXT:  .LBB1_8: # %else8
; AVX512F-NEXT:    vmovaps %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
; AVX512F-NEXT:  .LBB1_5: # %cond.load4
; AVX512F-NEXT:    vmovq %xmm0, %rax
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1],mem[0],xmm1[3]
; AVX512F-NEXT:    testb $8, %dil
; AVX512F-NEXT:    je .LBB1_8
; AVX512F-NEXT:  .LBB1_7: # %cond.load7
; AVX512F-NEXT:    vpextrq $1, %xmm0, %rax
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],mem[0]
; AVX512F-NEXT:    vmovaps %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: masked_gather_v4f32_ptr_v4i32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    kmovw %edi, %k1
; AVX512VL-NEXT:    vgatherqps (,%ymm0), %xmm1 {%k1}
; AVX512VL-NEXT:    vmovaps %xmm1, %xmm0
; AVX512VL-NEXT:    vzeroupper
; AVX512VL-NEXT:    retq
  %trigger.trunc = trunc i32 %trigger to i4
  %mask = bitcast i4 %trigger.trunc to <4 x i1>
  %res = call <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr> %ptr, i32 4, <4 x i1> %mask, <4 x float> %passthru)
  ret <4 x float> %res
}

define <4 x float> @gather_v4f32_v4i32_v4i32(ptr %base, <4 x i32> %idx, <4 x i32> %trigger, <4 x float> %passthru) {
; SSE-LABEL: gather_v4f32_v4i32_v4i32:
; SSE:       # %bb.0:
; SSE-NEXT:    movq %rdi, %xmm3
; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,1,0,1]
; SSE-NEXT:    pmovsxdq %xmm0, %xmm4
; SSE-NEXT:    psllq $2, %xmm4
; SSE-NEXT:    paddq %xmm3, %xmm4
; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
; SSE-NEXT:    pmovsxdq %xmm0, %xmm0
; SSE-NEXT:    pxor %xmm5, %xmm5
; SSE-NEXT:    pcmpeqd %xmm1, %xmm5
; SSE-NEXT:    movmskps %xmm5, %eax
; SSE-NEXT:    testb $1, %al
; SSE-NEXT:    je .LBB2_2
; SSE-NEXT:  # %bb.1: # %cond.load
; SSE-NEXT:    movq %xmm4, %rcx
; SSE-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE-NEXT:    pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3,4,5,6,7]
; SSE-NEXT:  .LBB2_2: # %else
; SSE-NEXT:    psllq $2, %xmm0
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    je .LBB2_4
; SSE-NEXT:  # %bb.3: # %cond.load1
; SSE-NEXT:    pextrq $1, %xmm4, %rcx
; SSE-NEXT:    insertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; SSE-NEXT:  .LBB2_4: # %else2
; SSE-NEXT:    paddq %xmm0, %xmm3
; SSE-NEXT:    testb $4, %al
; SSE-NEXT:    jne .LBB2_5
; SSE-NEXT:  # %bb.6: # %else5
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    jne .LBB2_7
; SSE-NEXT:  .LBB2_8: # %else8
; SSE-NEXT:    movaps %xmm2, %xmm0
; SSE-NEXT:    retq
; SSE-NEXT:  .LBB2_5: # %cond.load4
; SSE-NEXT:    movq %xmm3, %rcx
; SSE-NEXT:    insertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    je .LBB2_8
; SSE-NEXT:  .LBB2_7: # %cond.load7
; SSE-NEXT:    pextrq $1, %xmm3, %rax
; SSE-NEXT:    insertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; SSE-NEXT:    movaps %xmm2, %xmm0
; SSE-NEXT:    retq
;
; AVX1-LABEL: gather_v4f32_v4i32_v4i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq %rdi, %xmm3
; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,1,0,1]
; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm0[2,3,2,3]
; AVX1-NEXT:    vpmovsxdq %xmm4, %xmm4
; AVX1-NEXT:    vpsllq $2, %xmm4, %xmm4
; AVX1-NEXT:    vpaddq %xmm4, %xmm3, %xmm4
; AVX1-NEXT:    vpmovsxdq %xmm0, %xmm0
; AVX1-NEXT:    vpsllq $2, %xmm0, %xmm0
; AVX1-NEXT:    vpaddq %xmm0, %xmm3, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
; AVX1-NEXT:    vmovmskps %xmm1, %eax
; AVX1-NEXT:    testb $1, %al
; AVX1-NEXT:    je .LBB2_2
; AVX1-NEXT:  # %bb.1: # %cond.load
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX1-NEXT:    vmovss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
; AVX1-NEXT:  .LBB2_2: # %else
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    je .LBB2_4
; AVX1-NEXT:  # %bb.3: # %cond.load1
; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; AVX1-NEXT:  .LBB2_4: # %else2
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    jne .LBB2_5
; AVX1-NEXT:  # %bb.6: # %else5
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    jne .LBB2_7
; AVX1-NEXT:  .LBB2_8: # %else8
; AVX1-NEXT:    vmovaps %xmm2, %xmm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
; AVX1-NEXT:  .LBB2_5: # %cond.load4
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    je .LBB2_8
; AVX1-NEXT:  .LBB2_7: # %cond.load7
; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; AVX1-NEXT:    vmovaps %xmm2, %xmm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-NOGATHER-LABEL: gather_v4f32_v4i32_v4i32:
; AVX2-NOGATHER:       # %bb.0:
; AVX2-NOGATHER-NEXT:    vmovq %rdi, %xmm3
; AVX2-NOGATHER-NEXT:    vpbroadcastq %xmm3, %ymm3
; AVX2-NOGATHER-NEXT:    vpmovsxdq %xmm0, %ymm0
; AVX2-NOGATHER-NEXT:    vpsllq $2, %ymm0, %ymm0
; AVX2-NOGATHER-NEXT:    vpaddq %ymm0, %ymm3, %ymm0
; AVX2-NOGATHER-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-NOGATHER-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
; AVX2-NOGATHER-NEXT:    vmovmskps %xmm1, %eax
; AVX2-NOGATHER-NEXT:    testb $1, %al
; AVX2-NOGATHER-NEXT:    je .LBB2_2
; AVX2-NOGATHER-NEXT:  # %bb.1: # %cond.load
; AVX2-NOGATHER-NEXT:    vmovq %xmm0, %rcx
; AVX2-NOGATHER-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX2-NOGATHER-NEXT:    vmovss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
; AVX2-NOGATHER-NEXT:  .LBB2_2: # %else
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    je .LBB2_4
; AVX2-NOGATHER-NEXT:  # %bb.3: # %cond.load1
; AVX2-NOGATHER-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; AVX2-NOGATHER-NEXT:  .LBB2_4: # %else2
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NOGATHER-NEXT:    jne .LBB2_5
; AVX2-NOGATHER-NEXT:  # %bb.6: # %else5
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    jne .LBB2_7
; AVX2-NOGATHER-NEXT:  .LBB2_8: # %else8
; AVX2-NOGATHER-NEXT:    vmovaps %xmm2, %xmm0
; AVX2-NOGATHER-NEXT:    vzeroupper
; AVX2-NOGATHER-NEXT:    retq
; AVX2-NOGATHER-NEXT:  .LBB2_5: # %cond.load4
; AVX2-NOGATHER-NEXT:    vmovq %xmm0, %rcx
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    je .LBB2_8
; AVX2-NOGATHER-NEXT:  .LBB2_7: # %cond.load7
; AVX2-NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; AVX2-NOGATHER-NEXT:    vmovaps %xmm2, %xmm0
; AVX2-NOGATHER-NEXT:    vzeroupper
; AVX2-NOGATHER-NEXT:    retq
;
; AVX2-GATHER-LABEL: gather_v4f32_v4i32_v4i32:
; AVX2-GATHER:       # %bb.0:
; AVX2-GATHER-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-GATHER-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
; AVX2-GATHER-NEXT:    vgatherdps %xmm1, (%rdi,%xmm0,4), %xmm2
; AVX2-GATHER-NEXT:    vmovaps %xmm2, %xmm0
; AVX2-GATHER-NEXT:    retq
;
; AVX512F-LABEL: gather_v4f32_v4i32_v4i32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512F-NEXT:    vmovq %rdi, %xmm3
; AVX512F-NEXT:    vpbroadcastq %xmm3, %ymm3
; AVX512F-NEXT:    vpmovsxdq %xmm0, %ymm0
; AVX512F-NEXT:    vpsllq $2, %ymm0, %ymm0
; AVX512F-NEXT:    vpaddq %ymm0, %ymm3, %ymm0
; AVX512F-NEXT:    vptestnmd %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kmovw %k0, %eax
; AVX512F-NEXT:    testb $1, %al
; AVX512F-NEXT:    je .LBB2_2
; AVX512F-NEXT:  # %bb.1: # %cond.load
; AVX512F-NEXT:    vmovq %xmm0, %rcx
; AVX512F-NEXT:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX512F-NEXT:    vpblendd {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
; AVX512F-NEXT:  .LBB2_2: # %else
; AVX512F-NEXT:    testb $2, %al
; AVX512F-NEXT:    je .LBB2_4
; AVX512F-NEXT:  # %bb.3: # %cond.load1
; AVX512F-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; AVX512F-NEXT:  .LBB2_4: # %else2
; AVX512F-NEXT:    testb $4, %al
; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX512F-NEXT:    jne .LBB2_5
; AVX512F-NEXT:  # %bb.6: # %else5
; AVX512F-NEXT:    testb $8, %al
; AVX512F-NEXT:    jne .LBB2_7
; AVX512F-NEXT:  .LBB2_8: # %else8
; AVX512F-NEXT:    vmovaps %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
; AVX512F-NEXT:  .LBB2_5: # %cond.load4
; AVX512F-NEXT:    vmovq %xmm0, %rcx
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; AVX512F-NEXT:    testb $8, %al
; AVX512F-NEXT:    je .LBB2_8
; AVX512F-NEXT:  .LBB2_7: # %cond.load7
; AVX512F-NEXT:    vpextrq $1, %xmm0, %rax
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; AVX512F-NEXT:    vmovaps %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: gather_v4f32_v4i32_v4i32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vptestnmd %xmm1, %xmm1, %k1
; AVX512VL-NEXT:    vgatherdps (%rdi,%xmm0,4), %xmm2 {%k1}
; AVX512VL-NEXT:    vmovaps %xmm2, %xmm0
; AVX512VL-NEXT:    retq
  %vptr0 = insertelement <4 x ptr> undef, ptr %base, i32 0
  %vptr1 = shufflevector <4 x ptr> %vptr0, <4 x ptr> undef, <4 x i32> zeroinitializer
  %vptr2 = getelementptr float, <4 x ptr> %vptr1, <4 x i32> %idx

  %mask = icmp eq <4 x i32> %trigger, zeroinitializer
  %res = call <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr> %vptr2, i32 4, <4 x i1> %mask, <4 x float> %passthru)
  ret <4 x float> %res
}

define <4 x float> @gather_v4f32_v4i64_v4i32(ptr %base, <4 x i64> %idx, <4 x i32> %trigger, <4 x float> %passthru) {
; SSE-LABEL: gather_v4f32_v4i64_v4i32:
; SSE:       # %bb.0:
; SSE-NEXT:    movq %rdi, %xmm4
; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,0,1]
; SSE-NEXT:    psllq $2, %xmm0
; SSE-NEXT:    paddq %xmm4, %xmm0
; SSE-NEXT:    pxor %xmm5, %xmm5
; SSE-NEXT:    pcmpeqd %xmm2, %xmm5
; SSE-NEXT:    movmskps %xmm5, %eax
; SSE-NEXT:    testb $1, %al
; SSE-NEXT:    je .LBB3_2
; SSE-NEXT:  # %bb.1: # %cond.load
; SSE-NEXT:    movq %xmm0, %rcx
; SSE-NEXT:    movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; SSE-NEXT:    pblendw {{.*#+}} xmm3 = xmm2[0,1],xmm3[2,3,4,5,6,7]
; SSE-NEXT:  .LBB3_2: # %else
; SSE-NEXT:    psllq $2, %xmm1
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    je .LBB3_4
; SSE-NEXT:  # %bb.3: # %cond.load1
; SSE-NEXT:    pextrq $1, %xmm0, %rcx
; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0],mem[0],xmm3[2,3]
; SSE-NEXT:  .LBB3_4: # %else2
; SSE-NEXT:    paddq %xmm1, %xmm4
; SSE-NEXT:    testb $4, %al
; SSE-NEXT:    jne .LBB3_5
; SSE-NEXT:  # %bb.6: # %else5
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    jne .LBB3_7
; SSE-NEXT:  .LBB3_8: # %else8
; SSE-NEXT:    movaps %xmm3, %xmm0
; SSE-NEXT:    retq
; SSE-NEXT:  .LBB3_5: # %cond.load4
; SSE-NEXT:    movq %xmm4, %rcx
; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0,1],mem[0],xmm3[3]
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    je .LBB3_8
; SSE-NEXT:  .LBB3_7: # %cond.load7
; SSE-NEXT:    pextrq $1, %xmm4, %rax
; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0,1,2],mem[0]
; SSE-NEXT:    movaps %xmm3, %xmm0
; SSE-NEXT:    retq
;
; AVX1-LABEL: gather_v4f32_v4i64_v4i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
; AVX1-NEXT:    vpsllq $2, %xmm3, %xmm3
; AVX1-NEXT:    vmovq %rdi, %xmm4
; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,1,0,1]
; AVX1-NEXT:    vpaddq %xmm3, %xmm4, %xmm3
; AVX1-NEXT:    vpsllq $2, %xmm0, %xmm0
; AVX1-NEXT:    vpaddq %xmm0, %xmm4, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
; AVX1-NEXT:    vmovmskps %xmm1, %eax
; AVX1-NEXT:    testb $1, %al
; AVX1-NEXT:    je .LBB3_2
; AVX1-NEXT:  # %bb.1: # %cond.load
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX1-NEXT:    vmovss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
; AVX1-NEXT:  .LBB3_2: # %else
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    je .LBB3_4
; AVX1-NEXT:  # %bb.3: # %cond.load1
; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; AVX1-NEXT:  .LBB3_4: # %else2
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    jne .LBB3_5
; AVX1-NEXT:  # %bb.6: # %else5
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    jne .LBB3_7
; AVX1-NEXT:  .LBB3_8: # %else8
; AVX1-NEXT:    vmovaps %xmm2, %xmm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
; AVX1-NEXT:  .LBB3_5: # %cond.load4
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    je .LBB3_8
; AVX1-NEXT:  .LBB3_7: # %cond.load7
; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; AVX1-NEXT:    vmovaps %xmm2, %xmm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-NOGATHER-LABEL: gather_v4f32_v4i64_v4i32:
; AVX2-NOGATHER:       # %bb.0:
; AVX2-NOGATHER-NEXT:    vmovq %rdi, %xmm3
; AVX2-NOGATHER-NEXT:    vpbroadcastq %xmm3, %ymm3
; AVX2-NOGATHER-NEXT:    vpsllq $2, %ymm0, %ymm0
; AVX2-NOGATHER-NEXT:    vpaddq %ymm0, %ymm3, %ymm0
; AVX2-NOGATHER-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-NOGATHER-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
; AVX2-NOGATHER-NEXT:    vmovmskps %xmm1, %eax
; AVX2-NOGATHER-NEXT:    testb $1, %al
; AVX2-NOGATHER-NEXT:    je .LBB3_2
; AVX2-NOGATHER-NEXT:  # %bb.1: # %cond.load
; AVX2-NOGATHER-NEXT:    vmovq %xmm0, %rcx
; AVX2-NOGATHER-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX2-NOGATHER-NEXT:    vmovss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
; AVX2-NOGATHER-NEXT:  .LBB3_2: # %else
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    je .LBB3_4
; AVX2-NOGATHER-NEXT:  # %bb.3: # %cond.load1
; AVX2-NOGATHER-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; AVX2-NOGATHER-NEXT:  .LBB3_4: # %else2
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NOGATHER-NEXT:    jne .LBB3_5
; AVX2-NOGATHER-NEXT:  # %bb.6: # %else5
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    jne .LBB3_7
; AVX2-NOGATHER-NEXT:  .LBB3_8: # %else8
; AVX2-NOGATHER-NEXT:    vmovaps %xmm2, %xmm0
; AVX2-NOGATHER-NEXT:    vzeroupper
; AVX2-NOGATHER-NEXT:    retq
; AVX2-NOGATHER-NEXT:  .LBB3_5: # %cond.load4
; AVX2-NOGATHER-NEXT:    vmovq %xmm0, %rcx
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    je .LBB3_8
; AVX2-NOGATHER-NEXT:  .LBB3_7: # %cond.load7
; AVX2-NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
; AVX2-NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; AVX2-NOGATHER-NEXT:    vmovaps %xmm2, %xmm0
; AVX2-NOGATHER-NEXT:    vzeroupper
; AVX2-NOGATHER-NEXT:    retq
;
; AVX2-GATHER-LABEL: gather_v4f32_v4i64_v4i32:
; AVX2-GATHER:       # %bb.0:
; AVX2-GATHER-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-GATHER-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
; AVX2-GATHER-NEXT:    vgatherqps %xmm1, (%rdi,%ymm0,4), %xmm2
; AVX2-GATHER-NEXT:    vmovaps %xmm2, %xmm0
; AVX2-GATHER-NEXT:    vzeroupper
; AVX2-GATHER-NEXT:    retq
;
; AVX512F-LABEL: gather_v4f32_v4i64_v4i32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512F-NEXT:    vmovq %rdi, %xmm3
; AVX512F-NEXT:    vpbroadcastq %xmm3, %ymm3
; AVX512F-NEXT:    vpsllq $2, %ymm0, %ymm0
; AVX512F-NEXT:    vpaddq %ymm0, %ymm3, %ymm0
; AVX512F-NEXT:    vptestnmd %zmm1, %zmm1, %k0
; AVX512F-NEXT:    kmovw %k0, %eax
; AVX512F-NEXT:    testb $1, %al
; AVX512F-NEXT:    je .LBB3_2
; AVX512F-NEXT:  # %bb.1: # %cond.load
; AVX512F-NEXT:    vmovq %xmm0, %rcx
; AVX512F-NEXT:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX512F-NEXT:    vpblendd {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
; AVX512F-NEXT:  .LBB3_2: # %else
; AVX512F-NEXT:    testb $2, %al
; AVX512F-NEXT:    je .LBB3_4
; AVX512F-NEXT:  # %bb.3: # %cond.load1
; AVX512F-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
; AVX512F-NEXT:  .LBB3_4: # %else2
; AVX512F-NEXT:    testb $4, %al
; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX512F-NEXT:    jne .LBB3_5
; AVX512F-NEXT:  # %bb.6: # %else5
; AVX512F-NEXT:    testb $8, %al
; AVX512F-NEXT:    jne .LBB3_7
; AVX512F-NEXT:  .LBB3_8: # %else8
; AVX512F-NEXT:    vmovaps %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
; AVX512F-NEXT:  .LBB3_5: # %cond.load4
; AVX512F-NEXT:    vmovq %xmm0, %rcx
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
; AVX512F-NEXT:    testb $8, %al
; AVX512F-NEXT:    je .LBB3_8
; AVX512F-NEXT:  .LBB3_7: # %cond.load7
; AVX512F-NEXT:    vpextrq $1, %xmm0, %rax
; AVX512F-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
; AVX512F-NEXT:    vmovaps %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: gather_v4f32_v4i64_v4i32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vptestnmd %xmm1, %xmm1, %k1
; AVX512VL-NEXT:    vgatherqps (%rdi,%ymm0,4), %xmm2 {%k1}
; AVX512VL-NEXT:    vmovaps %xmm2, %xmm0
; AVX512VL-NEXT:    vzeroupper
; AVX512VL-NEXT:    retq
  %vptr0 = insertelement <4 x ptr> undef, ptr %base, i32 0
  %vptr1 = shufflevector <4 x ptr> %vptr0, <4 x ptr> undef, <4 x i32> zeroinitializer
  %vptr2 = getelementptr float, <4 x ptr> %vptr1, <4 x i64> %idx

  %mask = icmp eq <4 x i32> %trigger, zeroinitializer
  %res = call <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr> %vptr2, i32 4, <4 x i1> %mask, <4 x float> %passthru)
  ret <4 x float> %res
}

;
; vXi8
;

define <16 x i8> @gather_v16i8_v16i32_v16i8(ptr %base, <16 x i32> %idx, <16 x i8> %trigger, <16 x i8> %passthru) {
; SSE-LABEL: gather_v16i8_v16i32_v16i8:
; SSE:       # %bb.0:
; SSE-NEXT:    movq %rdi, %xmm6
; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,1,0,1]
; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm0[2,3,2,3]
; SSE-NEXT:    pmovsxdq %xmm0, %xmm0
; SSE-NEXT:    paddq %xmm6, %xmm0
; SSE-NEXT:    pxor %xmm8, %xmm8
; SSE-NEXT:    pcmpeqb %xmm4, %xmm8
; SSE-NEXT:    pmovmskb %xmm8, %eax
; SSE-NEXT:    testb $1, %al
; SSE-NEXT:    je .LBB4_2
; SSE-NEXT:  # %bb.1: # %cond.load
; SSE-NEXT:    movq %xmm0, %rcx
; SSE-NEXT:    pinsrb $0, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_2: # %else
; SSE-NEXT:    pmovsxdq %xmm7, %xmm4
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    je .LBB4_4
; SSE-NEXT:  # %bb.3: # %cond.load1
; SSE-NEXT:    pextrq $1, %xmm0, %rcx
; SSE-NEXT:    pinsrb $1, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_4: # %else2
; SSE-NEXT:    paddq %xmm6, %xmm4
; SSE-NEXT:    testb $4, %al
; SSE-NEXT:    je .LBB4_6
; SSE-NEXT:  # %bb.5: # %cond.load4
; SSE-NEXT:    movq %xmm4, %rcx
; SSE-NEXT:    pinsrb $2, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_6: # %else5
; SSE-NEXT:    pmovsxdq %xmm1, %xmm0
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    je .LBB4_8
; SSE-NEXT:  # %bb.7: # %cond.load7
; SSE-NEXT:    pextrq $1, %xmm4, %rcx
; SSE-NEXT:    pinsrb $3, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_8: # %else8
; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
; SSE-NEXT:    paddq %xmm6, %xmm0
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    je .LBB4_10
; SSE-NEXT:  # %bb.9: # %cond.load10
; SSE-NEXT:    movq %xmm0, %rcx
; SSE-NEXT:    pinsrb $4, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_10: # %else11
; SSE-NEXT:    pmovsxdq %xmm1, %xmm1
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    je .LBB4_12
; SSE-NEXT:  # %bb.11: # %cond.load13
; SSE-NEXT:    pextrq $1, %xmm0, %rcx
; SSE-NEXT:    pinsrb $5, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_12: # %else14
; SSE-NEXT:    paddq %xmm6, %xmm1
; SSE-NEXT:    testb $64, %al
; SSE-NEXT:    je .LBB4_14
; SSE-NEXT:  # %bb.13: # %cond.load16
; SSE-NEXT:    movq %xmm1, %rcx
; SSE-NEXT:    pinsrb $6, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_14: # %else17
; SSE-NEXT:    pmovsxdq %xmm2, %xmm0
; SSE-NEXT:    testb %al, %al
; SSE-NEXT:    jns .LBB4_16
; SSE-NEXT:  # %bb.15: # %cond.load19
; SSE-NEXT:    pextrq $1, %xmm1, %rcx
; SSE-NEXT:    pinsrb $7, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_16: # %else20
; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,3,2,3]
; SSE-NEXT:    paddq %xmm6, %xmm0
; SSE-NEXT:    testl $256, %eax # imm = 0x100
; SSE-NEXT:    je .LBB4_18
; SSE-NEXT:  # %bb.17: # %cond.load22
; SSE-NEXT:    movq %xmm0, %rcx
; SSE-NEXT:    pinsrb $8, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_18: # %else23
; SSE-NEXT:    pmovsxdq %xmm1, %xmm1
; SSE-NEXT:    testl $512, %eax # imm = 0x200
; SSE-NEXT:    je .LBB4_20
; SSE-NEXT:  # %bb.19: # %cond.load25
; SSE-NEXT:    pextrq $1, %xmm0, %rcx
; SSE-NEXT:    pinsrb $9, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_20: # %else26
; SSE-NEXT:    paddq %xmm6, %xmm1
; SSE-NEXT:    testl $1024, %eax # imm = 0x400
; SSE-NEXT:    je .LBB4_22
; SSE-NEXT:  # %bb.21: # %cond.load28
; SSE-NEXT:    movq %xmm1, %rcx
; SSE-NEXT:    pinsrb $10, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_22: # %else29
; SSE-NEXT:    pmovsxdq %xmm3, %xmm0
; SSE-NEXT:    testl $2048, %eax # imm = 0x800
; SSE-NEXT:    je .LBB4_24
; SSE-NEXT:  # %bb.23: # %cond.load31
; SSE-NEXT:    pextrq $1, %xmm1, %rcx
; SSE-NEXT:    pinsrb $11, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_24: # %else32
; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[2,3,2,3]
; SSE-NEXT:    paddq %xmm6, %xmm0
; SSE-NEXT:    testl $4096, %eax # imm = 0x1000
; SSE-NEXT:    je .LBB4_26
; SSE-NEXT:  # %bb.25: # %cond.load34
; SSE-NEXT:    movq %xmm0, %rcx
; SSE-NEXT:    pinsrb $12, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_26: # %else35
; SSE-NEXT:    pmovsxdq %xmm1, %xmm1
; SSE-NEXT:    testl $8192, %eax # imm = 0x2000
; SSE-NEXT:    je .LBB4_28
; SSE-NEXT:  # %bb.27: # %cond.load37
; SSE-NEXT:    pextrq $1, %xmm0, %rcx
; SSE-NEXT:    pinsrb $13, (%rcx), %xmm5
; SSE-NEXT:  .LBB4_28: # %else38
; SSE-NEXT:    paddq %xmm1, %xmm6
; SSE-NEXT:    testl $16384, %eax # imm = 0x4000
; SSE-NEXT:    jne .LBB4_29
; SSE-NEXT:  # %bb.30: # %else41
; SSE-NEXT:    testl $32768, %eax # imm = 0x8000
; SSE-NEXT:    jne .LBB4_31
; SSE-NEXT:  .LBB4_32: # %else44
; SSE-NEXT:    movdqa %xmm5, %xmm0
; SSE-NEXT:    retq
; SSE-NEXT:  .LBB4_29: # %cond.load40
; SSE-NEXT:    movq %xmm6, %rcx
; SSE-NEXT:    pinsrb $14, (%rcx), %xmm5
; SSE-NEXT:    testl $32768, %eax # imm = 0x8000
; SSE-NEXT:    je .LBB4_32
; SSE-NEXT:  .LBB4_31: # %cond.load43
; SSE-NEXT:    pextrq $1, %xmm6, %rax
; SSE-NEXT:    pinsrb $15, (%rax), %xmm5
; SSE-NEXT:    movdqa %xmm5, %xmm0
; SSE-NEXT:    retq
;
; AVX1-LABEL: gather_v16i8_v16i32_v16i8:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vmovq %rdi, %xmm4
; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,1,0,1]
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm0[2,3,2,3]
; AVX1-NEXT:    vpmovsxdq %xmm6, %xmm6
; AVX1-NEXT:    vpaddq %xmm6, %xmm4, %xmm6
; AVX1-NEXT:    vpmovsxdq %xmm0, %xmm0
; AVX1-NEXT:    vpaddq %xmm0, %xmm4, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm2
; AVX1-NEXT:    vpmovmskb %xmm2, %eax
; AVX1-NEXT:    testb $1, %al
; AVX1-NEXT:    je .LBB4_2
; AVX1-NEXT:  # %bb.1: # %cond.load
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $0, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_2: # %else
; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm5[2,3,2,3]
; AVX1-NEXT:    vpmovsxdq %xmm5, %xmm6
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    je .LBB4_4
; AVX1-NEXT:  # %bb.3: # %cond.load1
; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $1, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_4: # %else2
; AVX1-NEXT:    vpmovsxdq %xmm2, %xmm5
; AVX1-NEXT:    vpaddq %xmm6, %xmm4, %xmm2
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    je .LBB4_6
; AVX1-NEXT:  # %bb.5: # %cond.load4
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $2, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_6: # %else5
; AVX1-NEXT:    vpaddq %xmm5, %xmm4, %xmm5
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    je .LBB4_8
; AVX1-NEXT:  # %bb.7: # %cond.load7
; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $3, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_8: # %else8
; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm2, %ymm0
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    je .LBB4_10
; AVX1-NEXT:  # %bb.9: # %cond.load10
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $4, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_10: # %else11
; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
; AVX1-NEXT:    vpmovsxdq %xmm1, %xmm6
; AVX1-NEXT:    testb $32, %al
; AVX1-NEXT:    je .LBB4_12
; AVX1-NEXT:  # %bb.11: # %cond.load13
; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $5, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_12: # %else14
; AVX1-NEXT:    vpmovsxdq %xmm2, %xmm5
; AVX1-NEXT:    vpaddq %xmm6, %xmm4, %xmm2
; AVX1-NEXT:    testb $64, %al
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    je .LBB4_14
; AVX1-NEXT:  # %bb.13: # %cond.load16
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $6, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_14: # %else17
; AVX1-NEXT:    vpaddq %xmm5, %xmm4, %xmm5
; AVX1-NEXT:    testb %al, %al
; AVX1-NEXT:    jns .LBB4_16
; AVX1-NEXT:  # %bb.15: # %cond.load19
; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $7, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_16: # %else20
; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm2, %ymm0
; AVX1-NEXT:    testl $256, %eax # imm = 0x100
; AVX1-NEXT:    je .LBB4_18
; AVX1-NEXT:  # %bb.17: # %cond.load22
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $8, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_18: # %else23
; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
; AVX1-NEXT:    vpmovsxdq %xmm1, %xmm1
; AVX1-NEXT:    testl $512, %eax # imm = 0x200
; AVX1-NEXT:    je .LBB4_20
; AVX1-NEXT:  # %bb.19: # %cond.load25
; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $9, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_20: # %else26
; AVX1-NEXT:    vpmovsxdq %xmm2, %xmm2
; AVX1-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX1-NEXT:    testl $1024, %eax # imm = 0x400
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    je .LBB4_22
; AVX1-NEXT:  # %bb.21: # %cond.load28
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $10, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_22: # %else29
; AVX1-NEXT:    vpaddq %xmm2, %xmm4, %xmm2
; AVX1-NEXT:    testl $2048, %eax # imm = 0x800
; AVX1-NEXT:    je .LBB4_24
; AVX1-NEXT:  # %bb.23: # %cond.load31
; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $11, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_24: # %else32
; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm0
; AVX1-NEXT:    testl $4096, %eax # imm = 0x1000
; AVX1-NEXT:    je .LBB4_26
; AVX1-NEXT:  # %bb.25: # %cond.load34
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $12, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_26: # %else35
; AVX1-NEXT:    testl $8192, %eax # imm = 0x2000
; AVX1-NEXT:    je .LBB4_28
; AVX1-NEXT:  # %bb.27: # %cond.load37
; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $13, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:  .LBB4_28: # %else38
; AVX1-NEXT:    testl $16384, %eax # imm = 0x4000
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    jne .LBB4_29
; AVX1-NEXT:  # %bb.30: # %else41
; AVX1-NEXT:    testl $32768, %eax # imm = 0x8000
; AVX1-NEXT:    jne .LBB4_31
; AVX1-NEXT:  .LBB4_32: # %else44
; AVX1-NEXT:    vmovdqa %xmm3, %xmm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
; AVX1-NEXT:  .LBB4_29: # %cond.load40
; AVX1-NEXT:    vmovq %xmm0, %rcx
; AVX1-NEXT:    vpinsrb $14, (%rcx), %xmm3, %xmm3
; AVX1-NEXT:    testl $32768, %eax # imm = 0x8000
; AVX1-NEXT:    je .LBB4_32
; AVX1-NEXT:  .LBB4_31: # %cond.load43
; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
; AVX1-NEXT:    vpinsrb $15, (%rax), %xmm3, %xmm3
; AVX1-NEXT:    vmovdqa %xmm3, %xmm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: gather_v16i8_v16i32_v16i8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovq %rdi, %xmm4
; AVX2-NEXT:    vpbroadcastq %xmm4, %ymm4
; AVX2-NEXT:    vpmovsxdq %xmm0, %ymm5
; AVX2-NEXT:    vpaddq %ymm5, %ymm4, %ymm5
; AVX2-NEXT:    vpxor %xmm6, %xmm6, %xmm6
; AVX2-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm2
; AVX2-NEXT:    vpmovmskb %xmm2, %eax
; AVX2-NEXT:    testb $1, %al
; AVX2-NEXT:    je .LBB4_2
; AVX2-NEXT:  # %bb.1: # %cond.load
; AVX2-NEXT:    vmovq %xmm5, %rcx
; AVX2-NEXT:    vpinsrb $0, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_2: # %else
; AVX2-NEXT:    testb $2, %al
; AVX2-NEXT:    je .LBB4_4
; AVX2-NEXT:  # %bb.3: # %cond.load1
; AVX2-NEXT:    vpextrq $1, %xmm5, %rcx
; AVX2-NEXT:    vpinsrb $1, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_4: # %else2
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
; AVX2-NEXT:    testb $4, %al
; AVX2-NEXT:    vextracti128 $1, %ymm5, %xmm0
; AVX2-NEXT:    je .LBB4_6
; AVX2-NEXT:  # %bb.5: # %cond.load4
; AVX2-NEXT:    vmovq %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $2, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_6: # %else5
; AVX2-NEXT:    vpmovsxdq %xmm2, %ymm2
; AVX2-NEXT:    testb $8, %al
; AVX2-NEXT:    je .LBB4_8
; AVX2-NEXT:  # %bb.7: # %cond.load7
; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $3, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_8: # %else8
; AVX2-NEXT:    vpaddq %ymm2, %ymm4, %ymm0
; AVX2-NEXT:    testb $16, %al
; AVX2-NEXT:    je .LBB4_10
; AVX2-NEXT:  # %bb.9: # %cond.load10
; AVX2-NEXT:    vmovq %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $4, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_10: # %else11
; AVX2-NEXT:    testb $32, %al
; AVX2-NEXT:    je .LBB4_12
; AVX2-NEXT:  # %bb.11: # %cond.load13
; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $5, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_12: # %else14
; AVX2-NEXT:    testb $64, %al
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT:    je .LBB4_14
; AVX2-NEXT:  # %bb.13: # %cond.load16
; AVX2-NEXT:    vmovq %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $6, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_14: # %else17
; AVX2-NEXT:    vpmovsxdq %xmm1, %ymm2
; AVX2-NEXT:    testb %al, %al
; AVX2-NEXT:    jns .LBB4_16
; AVX2-NEXT:  # %bb.15: # %cond.load19
; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $7, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_16: # %else20
; AVX2-NEXT:    vpaddq %ymm2, %ymm4, %ymm0
; AVX2-NEXT:    testl $256, %eax # imm = 0x100
; AVX2-NEXT:    je .LBB4_18
; AVX2-NEXT:  # %bb.17: # %cond.load22
; AVX2-NEXT:    vmovq %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $8, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_18: # %else23
; AVX2-NEXT:    testl $512, %eax # imm = 0x200
; AVX2-NEXT:    je .LBB4_20
; AVX2-NEXT:  # %bb.19: # %cond.load25
; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $9, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_20: # %else26
; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm1
; AVX2-NEXT:    testl $1024, %eax # imm = 0x400
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT:    je .LBB4_22
; AVX2-NEXT:  # %bb.21: # %cond.load28
; AVX2-NEXT:    vmovq %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $10, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_22: # %else29
; AVX2-NEXT:    vpmovsxdq %xmm1, %ymm1
; AVX2-NEXT:    testl $2048, %eax # imm = 0x800
; AVX2-NEXT:    je .LBB4_24
; AVX2-NEXT:  # %bb.23: # %cond.load31
; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $11, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_24: # %else32
; AVX2-NEXT:    vpaddq %ymm1, %ymm4, %ymm0
; AVX2-NEXT:    testl $4096, %eax # imm = 0x1000
; AVX2-NEXT:    je .LBB4_26
; AVX2-NEXT:  # %bb.25: # %cond.load34
; AVX2-NEXT:    vmovq %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $12, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_26: # %else35
; AVX2-NEXT:    testl $8192, %eax # imm = 0x2000
; AVX2-NEXT:    je .LBB4_28
; AVX2-NEXT:  # %bb.27: # %cond.load37
; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $13, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:  .LBB4_28: # %else38
; AVX2-NEXT:    testl $16384, %eax # imm = 0x4000
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT:    jne .LBB4_29
; AVX2-NEXT:  # %bb.30: # %else41
; AVX2-NEXT:    testl $32768, %eax # imm = 0x8000
; AVX2-NEXT:    jne .LBB4_31
; AVX2-NEXT:  .LBB4_32: # %else44
; AVX2-NEXT:    vmovdqa %xmm3, %xmm0
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
; AVX2-NEXT:  .LBB4_29: # %cond.load40
; AVX2-NEXT:    vmovq %xmm0, %rcx
; AVX2-NEXT:    vpinsrb $14, (%rcx), %xmm3, %xmm3
; AVX2-NEXT:    testl $32768, %eax # imm = 0x8000
; AVX2-NEXT:    je .LBB4_32
; AVX2-NEXT:  .LBB4_31: # %cond.load43
; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
; AVX2-NEXT:    vpinsrb $15, (%rax), %xmm3, %xmm3
; AVX2-NEXT:    vmovdqa %xmm3, %xmm0
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
;
; AVX512-LABEL: gather_v16i8_v16i32_v16i8:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vpbroadcastq %rdi, %zmm3
; AVX512-NEXT:    vpmovsxdq %ymm0, %zmm4
; AVX512-NEXT:    vpaddq %zmm4, %zmm3, %zmm4
; AVX512-NEXT:    vpxor %xmm5, %xmm5, %xmm5
; AVX512-NEXT:    vpcmpeqb %xmm5, %xmm1, %xmm1
; AVX512-NEXT:    vpmovmskb %xmm1, %eax
; AVX512-NEXT:    testb $1, %al
; AVX512-NEXT:    jne .LBB4_1
; AVX512-NEXT:  # %bb.2: # %else
; AVX512-NEXT:    testb $2, %al
; AVX512-NEXT:    jne .LBB4_3
; AVX512-NEXT:  .LBB4_4: # %else2
; AVX512-NEXT:    testb $4, %al
; AVX512-NEXT:    jne .LBB4_5
; AVX512-NEXT:  .LBB4_6: # %else5
; AVX512-NEXT:    testb $8, %al
; AVX512-NEXT:    je .LBB4_8
; AVX512-NEXT:  .LBB4_7: # %cond.load7
; AVX512-NEXT:    vextracti128 $1, %ymm4, %xmm1
; AVX512-NEXT:    vpextrq $1, %xmm1, %rcx
; AVX512-NEXT:    vpinsrb $3, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:  .LBB4_8: # %else8
; AVX512-NEXT:    testb $16, %al
; AVX512-NEXT:    vextracti32x4 $2, %zmm4, %xmm1
; AVX512-NEXT:    je .LBB4_10
; AVX512-NEXT:  # %bb.9: # %cond.load10
; AVX512-NEXT:    vmovq %xmm1, %rcx
; AVX512-NEXT:    vpinsrb $4, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:  .LBB4_10: # %else11
; AVX512-NEXT:    testb $32, %al
; AVX512-NEXT:    je .LBB4_12
; AVX512-NEXT:  # %bb.11: # %cond.load13
; AVX512-NEXT:    vpextrq $1, %xmm1, %rcx
; AVX512-NEXT:    vpinsrb $5, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:  .LBB4_12: # %else14
; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
; AVX512-NEXT:    testb $64, %al
; AVX512-NEXT:    vextracti32x4 $3, %zmm4, %xmm0
; AVX512-NEXT:    je .LBB4_14
; AVX512-NEXT:  # %bb.13: # %cond.load16
; AVX512-NEXT:    vmovq %xmm0, %rcx
; AVX512-NEXT:    vpinsrb $6, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:  .LBB4_14: # %else17
; AVX512-NEXT:    vpmovsxdq %ymm1, %zmm1
; AVX512-NEXT:    testb %al, %al
; AVX512-NEXT:    jns .LBB4_16
; AVX512-NEXT:  # %bb.15: # %cond.load19
; AVX512-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX512-NEXT:    vpinsrb $7, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:  .LBB4_16: # %else20
; AVX512-NEXT:    vpaddq %zmm1, %zmm3, %zmm0
; AVX512-NEXT:    testl $256, %eax # imm = 0x100
; AVX512-NEXT:    jne .LBB4_17
; AVX512-NEXT:  # %bb.18: # %else23
; AVX512-NEXT:    testl $512, %eax # imm = 0x200
; AVX512-NEXT:    jne .LBB4_19
; AVX512-NEXT:  .LBB4_20: # %else26
; AVX512-NEXT:    testl $1024, %eax # imm = 0x400
; AVX512-NEXT:    jne .LBB4_21
; AVX512-NEXT:  .LBB4_22: # %else29
; AVX512-NEXT:    testl $2048, %eax # imm = 0x800
; AVX512-NEXT:    je .LBB4_24
; AVX512-NEXT:  .LBB4_23: # %cond.load31
; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX512-NEXT:    vpextrq $1, %xmm1, %rcx
; AVX512-NEXT:    vpinsrb $11, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:  .LBB4_24: # %else32
; AVX512-NEXT:    testl $4096, %eax # imm = 0x1000
; AVX512-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
; AVX512-NEXT:    je .LBB4_26
; AVX512-NEXT:  # %bb.25: # %cond.load34
; AVX512-NEXT:    vmovq %xmm1, %rcx
; AVX512-NEXT:    vpinsrb $12, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:  .LBB4_26: # %else35
; AVX512-NEXT:    testl $8192, %eax # imm = 0x2000
; AVX512-NEXT:    je .LBB4_28
; AVX512-NEXT:  # %bb.27: # %cond.load37
; AVX512-NEXT:    vpextrq $1, %xmm1, %rcx
; AVX512-NEXT:    vpinsrb $13, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:  .LBB4_28: # %else38
; AVX512-NEXT:    testl $16384, %eax # imm = 0x4000
; AVX512-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
; AVX512-NEXT:    jne .LBB4_29
; AVX512-NEXT:  # %bb.30: # %else41
; AVX512-NEXT:    testl $32768, %eax # imm = 0x8000
; AVX512-NEXT:    jne .LBB4_31
; AVX512-NEXT:  .LBB4_32: # %else44
; AVX512-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
; AVX512-NEXT:  .LBB4_1: # %cond.load
; AVX512-NEXT:    vmovq %xmm4, %rcx
; AVX512-NEXT:    vpinsrb $0, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:    testb $2, %al
; AVX512-NEXT:    je .LBB4_4
; AVX512-NEXT:  .LBB4_3: # %cond.load1
; AVX512-NEXT:    vpextrq $1, %xmm4, %rcx
; AVX512-NEXT:    vpinsrb $1, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:    testb $4, %al
; AVX512-NEXT:    je .LBB4_6
; AVX512-NEXT:  .LBB4_5: # %cond.load4
; AVX512-NEXT:    vextracti128 $1, %ymm4, %xmm1
; AVX512-NEXT:    vmovq %xmm1, %rcx
; AVX512-NEXT:    vpinsrb $2, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:    testb $8, %al
; AVX512-NEXT:    jne .LBB4_7
; AVX512-NEXT:    jmp .LBB4_8
; AVX512-NEXT:  .LBB4_17: # %cond.load22
; AVX512-NEXT:    vmovq %xmm0, %rcx
; AVX512-NEXT:    vpinsrb $8, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:    testl $512, %eax # imm = 0x200
; AVX512-NEXT:    je .LBB4_20
; AVX512-NEXT:  .LBB4_19: # %cond.load25
; AVX512-NEXT:    vpextrq $1, %xmm0, %rcx
; AVX512-NEXT:    vpinsrb $9, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:    testl $1024, %eax # imm = 0x400
; AVX512-NEXT:    je .LBB4_22
; AVX512-NEXT:  .LBB4_21: # %cond.load28
; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX512-NEXT:    vmovq %xmm1, %rcx
; AVX512-NEXT:    vpinsrb $10, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:    testl $2048, %eax # imm = 0x800
; AVX512-NEXT:    jne .LBB4_23
; AVX512-NEXT:    jmp .LBB4_24
; AVX512-NEXT:  .LBB4_29: # %cond.load40
; AVX512-NEXT:    vmovq %xmm0, %rcx
; AVX512-NEXT:    vpinsrb $14, (%rcx), %xmm2, %xmm2
; AVX512-NEXT:    testl $32768, %eax # imm = 0x8000
; AVX512-NEXT:    je .LBB4_32
; AVX512-NEXT:  .LBB4_31: # %cond.load43
; AVX512-NEXT:    vpextrq $1, %xmm0, %rax
; AVX512-NEXT:    vpinsrb $15, (%rax), %xmm2, %xmm2
; AVX512-NEXT:    vmovdqa %xmm2, %xmm0
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %vptr0 = insertelement <16 x ptr> undef, ptr %base, i32 0
  %vptr1 = shufflevector <16 x ptr> %vptr0, <16 x ptr> undef, <16 x i32> zeroinitializer
  %vptr2 = getelementptr i8, <16 x ptr> %vptr1, <16 x i32> %idx

  %mask = icmp eq <16 x i8> %trigger, zeroinitializer
  %res = call <16 x i8> @llvm.masked.gather.v16i8.v16p0(<16 x ptr> %vptr2, i32 4, <16 x i1> %mask, <16 x i8> %passthru)
  ret <16 x i8> %res
}

define <8 x i32> @gather_v8i32_v8i32(<8 x i32> %trigger) {
; SSE-LABEL: gather_v8i32_v8i32:
; SSE:       # %bb.0:
; SSE-NEXT:    movdqa %xmm1, %xmm3
; SSE-NEXT:    movdqa %xmm0, %xmm2
; SSE-NEXT:    pxor %xmm0, %xmm0
; SSE-NEXT:    pcmpeqd %xmm0, %xmm1
; SSE-NEXT:    pcmpeqd %xmm2, %xmm0
; SSE-NEXT:    packssdw %xmm1, %xmm0
; SSE-NEXT:    packsswb %xmm0, %xmm0
; SSE-NEXT:    pmovmskb %xmm0, %eax
; SSE-NEXT:    testb $1, %al
; SSE-NEXT:    je .LBB5_1
; SSE-NEXT:  # %bb.2: # %cond.load
; SSE-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    jne .LBB5_4
; SSE-NEXT:    jmp .LBB5_5
; SSE-NEXT:  .LBB5_1:
; SSE-NEXT:    # implicit-def: $xmm0
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    je .LBB5_5
; SSE-NEXT:  .LBB5_4: # %cond.load1
; SSE-NEXT:    pinsrd $1, c+12(%rip), %xmm0
; SSE-NEXT:  .LBB5_5: # %else2
; SSE-NEXT:    testb $4, %al
; SSE-NEXT:    jne .LBB5_6
; SSE-NEXT:  # %bb.7: # %else5
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    jne .LBB5_8
; SSE-NEXT:  .LBB5_9: # %else8
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    je .LBB5_10
; SSE-NEXT:  .LBB5_11: # %cond.load10
; SSE-NEXT:    pinsrd $0, c+12(%rip), %xmm1
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    jne .LBB5_13
; SSE-NEXT:    jmp .LBB5_14
; SSE-NEXT:  .LBB5_6: # %cond.load4
; SSE-NEXT:    pinsrd $2, c+12(%rip), %xmm0
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    je .LBB5_9
; SSE-NEXT:  .LBB5_8: # %cond.load7
; SSE-NEXT:    pinsrd $3, c+12(%rip), %xmm0
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    jne .LBB5_11
; SSE-NEXT:  .LBB5_10:
; SSE-NEXT:    # implicit-def: $xmm1
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    je .LBB5_14
; SSE-NEXT:  .LBB5_13: # %cond.load13
; SSE-NEXT:    pinsrd $1, c+12(%rip), %xmm1
; SSE-NEXT:  .LBB5_14: # %else14
; SSE-NEXT:    testb $64, %al
; SSE-NEXT:    jne .LBB5_15
; SSE-NEXT:  # %bb.16: # %else17
; SSE-NEXT:    testb $-128, %al
; SSE-NEXT:    je .LBB5_18
; SSE-NEXT:  .LBB5_17: # %cond.load19
; SSE-NEXT:    pinsrd $3, c+12(%rip), %xmm1
; SSE-NEXT:  .LBB5_18: # %else20
; SSE-NEXT:    pxor %xmm4, %xmm4
; SSE-NEXT:    movdqa %xmm2, %xmm5
; SSE-NEXT:    pcmpeqd %xmm4, %xmm5
; SSE-NEXT:    pcmpeqd %xmm3, %xmm4
; SSE-NEXT:    packssdw %xmm4, %xmm5
; SSE-NEXT:    packsswb %xmm5, %xmm5
; SSE-NEXT:    pmovmskb %xmm5, %eax
; SSE-NEXT:    testb $1, %al
; SSE-NEXT:    je .LBB5_19
; SSE-NEXT:  # %bb.20: # %cond.load23
; SSE-NEXT:    movd {{.*#+}} xmm4 = mem[0],zero,zero,zero
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    jne .LBB5_22
; SSE-NEXT:    jmp .LBB5_23
; SSE-NEXT:  .LBB5_15: # %cond.load16
; SSE-NEXT:    pinsrd $2, c+12(%rip), %xmm1
; SSE-NEXT:    testb $-128, %al
; SSE-NEXT:    jne .LBB5_17
; SSE-NEXT:    jmp .LBB5_18
; SSE-NEXT:  .LBB5_19:
; SSE-NEXT:    # implicit-def: $xmm4
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    je .LBB5_23
; SSE-NEXT:  .LBB5_22: # %cond.load28
; SSE-NEXT:    pinsrd $1, c+28(%rip), %xmm4
; SSE-NEXT:  .LBB5_23: # %else31
; SSE-NEXT:    testb $4, %al
; SSE-NEXT:    jne .LBB5_24
; SSE-NEXT:  # %bb.25: # %else36
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    jne .LBB5_26
; SSE-NEXT:  .LBB5_27: # %else41
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    je .LBB5_28
; SSE-NEXT:  .LBB5_29: # %cond.load43
; SSE-NEXT:    pinsrd $0, c+28(%rip), %xmm5
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    jne .LBB5_31
; SSE-NEXT:    jmp .LBB5_32
; SSE-NEXT:  .LBB5_24: # %cond.load33
; SSE-NEXT:    pinsrd $2, c+28(%rip), %xmm4
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    je .LBB5_27
; SSE-NEXT:  .LBB5_26: # %cond.load38
; SSE-NEXT:    pinsrd $3, c+28(%rip), %xmm4
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    jne .LBB5_29
; SSE-NEXT:  .LBB5_28:
; SSE-NEXT:    # implicit-def: $xmm5
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    je .LBB5_32
; SSE-NEXT:  .LBB5_31: # %cond.load48
; SSE-NEXT:    pinsrd $1, c+28(%rip), %xmm5
; SSE-NEXT:  .LBB5_32: # %else51
; SSE-NEXT:    testb $64, %al
; SSE-NEXT:    jne .LBB5_33
; SSE-NEXT:  # %bb.34: # %else56
; SSE-NEXT:    testb $-128, %al
; SSE-NEXT:    je .LBB5_36
; SSE-NEXT:  .LBB5_35: # %cond.load58
; SSE-NEXT:    pinsrd $3, c+28(%rip), %xmm5
; SSE-NEXT:  .LBB5_36: # %else61
; SSE-NEXT:    pxor %xmm6, %xmm6
; SSE-NEXT:    pcmpeqd %xmm6, %xmm2
; SSE-NEXT:    pcmpeqd %xmm6, %xmm3
; SSE-NEXT:    packssdw %xmm3, %xmm2
; SSE-NEXT:    packsswb %xmm2, %xmm2
; SSE-NEXT:    pmovmskb %xmm2, %eax
; SSE-NEXT:    testb $1, %al
; SSE-NEXT:    je .LBB5_37
; SSE-NEXT:  # %bb.38: # %cond.load64
; SSE-NEXT:    movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    jne .LBB5_40
; SSE-NEXT:    jmp .LBB5_41
; SSE-NEXT:  .LBB5_33: # %cond.load53
; SSE-NEXT:    pinsrd $2, c+28(%rip), %xmm5
; SSE-NEXT:    testb $-128, %al
; SSE-NEXT:    jne .LBB5_35
; SSE-NEXT:    jmp .LBB5_36
; SSE-NEXT:  .LBB5_37:
; SSE-NEXT:    # implicit-def: $xmm2
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    je .LBB5_41
; SSE-NEXT:  .LBB5_40: # %cond.load69
; SSE-NEXT:    pinsrd $1, c+28(%rip), %xmm2
; SSE-NEXT:  .LBB5_41: # %else72
; SSE-NEXT:    testb $4, %al
; SSE-NEXT:    jne .LBB5_42
; SSE-NEXT:  # %bb.43: # %else77
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    jne .LBB5_44
; SSE-NEXT:  .LBB5_45: # %else82
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    je .LBB5_46
; SSE-NEXT:  .LBB5_47: # %cond.load84
; SSE-NEXT:    pinsrd $0, c+28(%rip), %xmm3
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    jne .LBB5_49
; SSE-NEXT:    jmp .LBB5_50
; SSE-NEXT:  .LBB5_42: # %cond.load74
; SSE-NEXT:    pinsrd $2, c+28(%rip), %xmm2
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    je .LBB5_45
; SSE-NEXT:  .LBB5_44: # %cond.load79
; SSE-NEXT:    pinsrd $3, c+28(%rip), %xmm2
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    jne .LBB5_47
; SSE-NEXT:  .LBB5_46:
; SSE-NEXT:    # implicit-def: $xmm3
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    je .LBB5_50
; SSE-NEXT:  .LBB5_49: # %cond.load89
; SSE-NEXT:    pinsrd $1, c+28(%rip), %xmm3
; SSE-NEXT:  .LBB5_50: # %else92
; SSE-NEXT:    testb $64, %al
; SSE-NEXT:    je .LBB5_52
; SSE-NEXT:  # %bb.51: # %cond.load94
; SSE-NEXT:    pinsrd $2, c+28(%rip), %xmm3
; SSE-NEXT:  .LBB5_52: # %else97
; SSE-NEXT:    paddd %xmm4, %xmm0
; SSE-NEXT:    paddd %xmm5, %xmm1
; SSE-NEXT:    testb $-128, %al
; SSE-NEXT:    je .LBB5_54
; SSE-NEXT:  # %bb.53: # %cond.load99
; SSE-NEXT:    pinsrd $3, c+28(%rip), %xmm3
; SSE-NEXT:  .LBB5_54: # %else102
; SSE-NEXT:    paddd %xmm3, %xmm1
; SSE-NEXT:    paddd %xmm2, %xmm0
; SSE-NEXT:    retq
;
; AVX1-LABEL: gather_v8i32_v8i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm2, %xmm3
; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm1
; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
; AVX1-NEXT:    vmovmskps %ymm1, %eax
; AVX1-NEXT:    testb $1, %al
; AVX1-NEXT:    # implicit-def: $ymm1
; AVX1-NEXT:    jne .LBB5_1
; AVX1-NEXT:  # %bb.2: # %else
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    jne .LBB5_3
; AVX1-NEXT:  .LBB5_4: # %else2
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    jne .LBB5_5
; AVX1-NEXT:  .LBB5_6: # %else5
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    jne .LBB5_7
; AVX1-NEXT:  .LBB5_8: # %else8
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    jne .LBB5_9
; AVX1-NEXT:  .LBB5_10: # %else11
; AVX1-NEXT:    testb $32, %al
; AVX1-NEXT:    jne .LBB5_11
; AVX1-NEXT:  .LBB5_12: # %else14
; AVX1-NEXT:    testb $64, %al
; AVX1-NEXT:    jne .LBB5_13
; AVX1-NEXT:  .LBB5_14: # %else17
; AVX1-NEXT:    testb $-128, %al
; AVX1-NEXT:    je .LBB5_16
; AVX1-NEXT:  .LBB5_15: # %cond.load19
; AVX1-NEXT:    vbroadcastss c+12(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
; AVX1-NEXT:  .LBB5_16: # %else20
; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm4
; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm0, %xmm3
; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
; AVX1-NEXT:    vmovmskps %ymm3, %eax
; AVX1-NEXT:    testb $1, %al
; AVX1-NEXT:    # implicit-def: $ymm3
; AVX1-NEXT:    jne .LBB5_17
; AVX1-NEXT:  # %bb.18: # %else26
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    jne .LBB5_19
; AVX1-NEXT:  .LBB5_20: # %else31
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    jne .LBB5_21
; AVX1-NEXT:  .LBB5_22: # %else36
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    jne .LBB5_23
; AVX1-NEXT:  .LBB5_24: # %else41
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    jne .LBB5_25
; AVX1-NEXT:  .LBB5_26: # %else46
; AVX1-NEXT:    testb $32, %al
; AVX1-NEXT:    jne .LBB5_27
; AVX1-NEXT:  .LBB5_28: # %else51
; AVX1-NEXT:    testb $64, %al
; AVX1-NEXT:    jne .LBB5_29
; AVX1-NEXT:  .LBB5_30: # %else56
; AVX1-NEXT:    testb $-128, %al
; AVX1-NEXT:    je .LBB5_32
; AVX1-NEXT:  .LBB5_31: # %cond.load58
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm4
; AVX1-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
; AVX1-NEXT:  .LBB5_32: # %else61
; AVX1-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm2, %xmm2
; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
; AVX1-NEXT:    vmovmskps %ymm0, %eax
; AVX1-NEXT:    testb $1, %al
; AVX1-NEXT:    # implicit-def: $ymm0
; AVX1-NEXT:    jne .LBB5_33
; AVX1-NEXT:  # %bb.34: # %else67
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    jne .LBB5_35
; AVX1-NEXT:  .LBB5_36: # %else72
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    jne .LBB5_37
; AVX1-NEXT:  .LBB5_38: # %else77
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    jne .LBB5_39
; AVX1-NEXT:  .LBB5_40: # %else82
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    je .LBB5_42
; AVX1-NEXT:  .LBB5_41: # %cond.load84
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm2
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4],ymm0[5,6,7]
; AVX1-NEXT:  .LBB5_42: # %else87
; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm2
; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
; AVX1-NEXT:    vpaddd %xmm3, %xmm1, %xmm1
; AVX1-NEXT:    testb $32, %al
; AVX1-NEXT:    je .LBB5_44
; AVX1-NEXT:  # %bb.43: # %cond.load89
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5],ymm0[6,7]
; AVX1-NEXT:  .LBB5_44: # %else92
; AVX1-NEXT:    vpaddd %xmm2, %xmm4, %xmm2
; AVX1-NEXT:    testb $64, %al
; AVX1-NEXT:    je .LBB5_46
; AVX1-NEXT:  # %bb.45: # %cond.load94
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm3[6],ymm0[7]
; AVX1-NEXT:  .LBB5_46: # %else97
; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
; AVX1-NEXT:    testb $-128, %al
; AVX1-NEXT:    je .LBB5_48
; AVX1-NEXT:  # %bb.47: # %cond.load99
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm2
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
; AVX1-NEXT:  .LBB5_48: # %else102
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
; AVX1-NEXT:    vpaddd %xmm2, %xmm3, %xmm2
; AVX1-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
; AVX1-NEXT:    retq
; AVX1-NEXT:  .LBB5_1: # %cond.load
; AVX1-NEXT:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    je .LBB5_4
; AVX1-NEXT:  .LBB5_3: # %cond.load1
; AVX1-NEXT:    vpinsrd $1, c+12(%rip), %xmm1, %xmm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    je .LBB5_6
; AVX1-NEXT:  .LBB5_5: # %cond.load4
; AVX1-NEXT:    vpinsrd $2, c+12(%rip), %xmm1, %xmm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    je .LBB5_8
; AVX1-NEXT:  .LBB5_7: # %cond.load7
; AVX1-NEXT:    vpinsrd $3, c+12(%rip), %xmm1, %xmm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    je .LBB5_10
; AVX1-NEXT:  .LBB5_9: # %cond.load10
; AVX1-NEXT:    vbroadcastss c+12(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4],ymm1[5,6,7]
; AVX1-NEXT:    testb $32, %al
; AVX1-NEXT:    je .LBB5_12
; AVX1-NEXT:  .LBB5_11: # %cond.load13
; AVX1-NEXT:    vbroadcastss c+12(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5],ymm1[6,7]
; AVX1-NEXT:    testb $64, %al
; AVX1-NEXT:    je .LBB5_14
; AVX1-NEXT:  .LBB5_13: # %cond.load16
; AVX1-NEXT:    vbroadcastss c+12(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm3[6],ymm1[7]
; AVX1-NEXT:    testb $-128, %al
; AVX1-NEXT:    jne .LBB5_15
; AVX1-NEXT:    jmp .LBB5_16
; AVX1-NEXT:  .LBB5_17: # %cond.load23
; AVX1-NEXT:    vmovd {{.*#+}} xmm3 = mem[0],zero,zero,zero
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    je .LBB5_20
; AVX1-NEXT:  .LBB5_19: # %cond.load28
; AVX1-NEXT:    vpinsrd $1, c+28(%rip), %xmm3, %xmm4
; AVX1-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    je .LBB5_22
; AVX1-NEXT:  .LBB5_21: # %cond.load33
; AVX1-NEXT:    vpinsrd $2, c+28(%rip), %xmm3, %xmm4
; AVX1-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    je .LBB5_24
; AVX1-NEXT:  .LBB5_23: # %cond.load38
; AVX1-NEXT:    vpinsrd $3, c+28(%rip), %xmm3, %xmm4
; AVX1-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    je .LBB5_26
; AVX1-NEXT:  .LBB5_25: # %cond.load43
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm4
; AVX1-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4],ymm3[5,6,7]
; AVX1-NEXT:    testb $32, %al
; AVX1-NEXT:    je .LBB5_28
; AVX1-NEXT:  .LBB5_27: # %cond.load48
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm4
; AVX1-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5],ymm3[6,7]
; AVX1-NEXT:    testb $64, %al
; AVX1-NEXT:    je .LBB5_30
; AVX1-NEXT:  .LBB5_29: # %cond.load53
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm4
; AVX1-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm4[6],ymm3[7]
; AVX1-NEXT:    testb $-128, %al
; AVX1-NEXT:    jne .LBB5_31
; AVX1-NEXT:    jmp .LBB5_32
; AVX1-NEXT:  .LBB5_33: # %cond.load64
; AVX1-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    je .LBB5_36
; AVX1-NEXT:  .LBB5_35: # %cond.load69
; AVX1-NEXT:    vpinsrd $1, c+28(%rip), %xmm0, %xmm2
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    je .LBB5_38
; AVX1-NEXT:  .LBB5_37: # %cond.load74
; AVX1-NEXT:    vpinsrd $2, c+28(%rip), %xmm0, %xmm2
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    je .LBB5_40
; AVX1-NEXT:  .LBB5_39: # %cond.load79
; AVX1-NEXT:    vpinsrd $3, c+28(%rip), %xmm0, %xmm2
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    jne .LBB5_41
; AVX1-NEXT:    jmp .LBB5_42
;
; AVX2-NOGATHER-LABEL: gather_v8i32_v8i32:
; AVX2-NOGATHER:       # %bb.0:
; AVX2-NOGATHER-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; AVX2-NOGATHER-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm1
; AVX2-NOGATHER-NEXT:    vmovmskps %ymm1, %eax
; AVX2-NOGATHER-NEXT:    testb $1, %al
; AVX2-NOGATHER-NEXT:    # implicit-def: $ymm1
; AVX2-NOGATHER-NEXT:    jne .LBB5_1
; AVX2-NOGATHER-NEXT:  # %bb.2: # %else
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_3
; AVX2-NOGATHER-NEXT:  .LBB5_4: # %else2
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_5
; AVX2-NOGATHER-NEXT:  .LBB5_6: # %else5
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_7
; AVX2-NOGATHER-NEXT:  .LBB5_8: # %else8
; AVX2-NOGATHER-NEXT:    testb $16, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_9
; AVX2-NOGATHER-NEXT:  .LBB5_10: # %else11
; AVX2-NOGATHER-NEXT:    testb $32, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_11
; AVX2-NOGATHER-NEXT:  .LBB5_12: # %else14
; AVX2-NOGATHER-NEXT:    testb $64, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_13
; AVX2-NOGATHER-NEXT:  .LBB5_14: # %else17
; AVX2-NOGATHER-NEXT:    testb $-128, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_16
; AVX2-NOGATHER-NEXT:  .LBB5_15: # %cond.load19
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+12(%rip), %ymm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
; AVX2-NOGATHER-NEXT:  .LBB5_16: # %else20
; AVX2-NOGATHER-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX2-NOGATHER-NEXT:    vpcmpeqd %ymm2, %ymm0, %ymm2
; AVX2-NOGATHER-NEXT:    vmovmskps %ymm2, %eax
; AVX2-NOGATHER-NEXT:    testb $1, %al
; AVX2-NOGATHER-NEXT:    # implicit-def: $ymm2
; AVX2-NOGATHER-NEXT:    jne .LBB5_17
; AVX2-NOGATHER-NEXT:  # %bb.18: # %else26
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_19
; AVX2-NOGATHER-NEXT:  .LBB5_20: # %else31
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_21
; AVX2-NOGATHER-NEXT:  .LBB5_22: # %else36
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_23
; AVX2-NOGATHER-NEXT:  .LBB5_24: # %else41
; AVX2-NOGATHER-NEXT:    testb $16, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_25
; AVX2-NOGATHER-NEXT:  .LBB5_26: # %else46
; AVX2-NOGATHER-NEXT:    testb $32, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_27
; AVX2-NOGATHER-NEXT:  .LBB5_28: # %else51
; AVX2-NOGATHER-NEXT:    testb $64, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_29
; AVX2-NOGATHER-NEXT:  .LBB5_30: # %else56
; AVX2-NOGATHER-NEXT:    testb $-128, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_32
; AVX2-NOGATHER-NEXT:  .LBB5_31: # %cond.load58
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
; AVX2-NOGATHER-NEXT:  .LBB5_32: # %else61
; AVX2-NOGATHER-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-NOGATHER-NEXT:    vpcmpeqd %ymm3, %ymm0, %ymm0
; AVX2-NOGATHER-NEXT:    vmovmskps %ymm0, %eax
; AVX2-NOGATHER-NEXT:    testb $1, %al
; AVX2-NOGATHER-NEXT:    # implicit-def: $ymm0
; AVX2-NOGATHER-NEXT:    jne .LBB5_33
; AVX2-NOGATHER-NEXT:  # %bb.34: # %else67
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_35
; AVX2-NOGATHER-NEXT:  .LBB5_36: # %else72
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_37
; AVX2-NOGATHER-NEXT:  .LBB5_38: # %else77
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_39
; AVX2-NOGATHER-NEXT:  .LBB5_40: # %else82
; AVX2-NOGATHER-NEXT:    testb $16, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_41
; AVX2-NOGATHER-NEXT:  .LBB5_42: # %else87
; AVX2-NOGATHER-NEXT:    testb $32, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_43
; AVX2-NOGATHER-NEXT:  .LBB5_44: # %else92
; AVX2-NOGATHER-NEXT:    testb $64, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_46
; AVX2-NOGATHER-NEXT:  .LBB5_45: # %cond.load94
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm3[6],ymm0[7]
; AVX2-NOGATHER-NEXT:  .LBB5_46: # %else97
; AVX2-NOGATHER-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
; AVX2-NOGATHER-NEXT:    testb $-128, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_48
; AVX2-NOGATHER-NEXT:  # %bb.47: # %cond.load99
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
; AVX2-NOGATHER-NEXT:  .LBB5_48: # %else102
; AVX2-NOGATHER-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
; AVX2-NOGATHER-NEXT:    retq
; AVX2-NOGATHER-NEXT:  .LBB5_1: # %cond.load
; AVX2-NOGATHER-NEXT:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_4
; AVX2-NOGATHER-NEXT:  .LBB5_3: # %cond.load1
; AVX2-NOGATHER-NEXT:    vpinsrd $1, c+12(%rip), %xmm1, %xmm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_6
; AVX2-NOGATHER-NEXT:  .LBB5_5: # %cond.load4
; AVX2-NOGATHER-NEXT:    vpinsrd $2, c+12(%rip), %xmm1, %xmm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_8
; AVX2-NOGATHER-NEXT:  .LBB5_7: # %cond.load7
; AVX2-NOGATHER-NEXT:    vpinsrd $3, c+12(%rip), %xmm1, %xmm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $16, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_10
; AVX2-NOGATHER-NEXT:  .LBB5_9: # %cond.load10
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+12(%rip), %ymm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4],ymm1[5,6,7]
; AVX2-NOGATHER-NEXT:    testb $32, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_12
; AVX2-NOGATHER-NEXT:  .LBB5_11: # %cond.load13
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+12(%rip), %ymm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6,7]
; AVX2-NOGATHER-NEXT:    testb $64, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_14
; AVX2-NOGATHER-NEXT:  .LBB5_13: # %cond.load16
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+12(%rip), %ymm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
; AVX2-NOGATHER-NEXT:    testb $-128, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_15
; AVX2-NOGATHER-NEXT:    jmp .LBB5_16
; AVX2-NOGATHER-NEXT:  .LBB5_17: # %cond.load23
; AVX2-NOGATHER-NEXT:    vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_20
; AVX2-NOGATHER-NEXT:  .LBB5_19: # %cond.load28
; AVX2-NOGATHER-NEXT:    vpinsrd $1, c+28(%rip), %xmm2, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_22
; AVX2-NOGATHER-NEXT:  .LBB5_21: # %cond.load33
; AVX2-NOGATHER-NEXT:    vpinsrd $2, c+28(%rip), %xmm2, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_24
; AVX2-NOGATHER-NEXT:  .LBB5_23: # %cond.load38
; AVX2-NOGATHER-NEXT:    vpinsrd $3, c+28(%rip), %xmm2, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $16, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_26
; AVX2-NOGATHER-NEXT:  .LBB5_25: # %cond.load43
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4],ymm2[5,6,7]
; AVX2-NOGATHER-NEXT:    testb $32, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_28
; AVX2-NOGATHER-NEXT:  .LBB5_27: # %cond.load48
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5],ymm2[6,7]
; AVX2-NOGATHER-NEXT:    testb $64, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_30
; AVX2-NOGATHER-NEXT:  .LBB5_29: # %cond.load53
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6],ymm2[7]
; AVX2-NOGATHER-NEXT:    testb $-128, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_31
; AVX2-NOGATHER-NEXT:    jmp .LBB5_32
; AVX2-NOGATHER-NEXT:  .LBB5_33: # %cond.load64
; AVX2-NOGATHER-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_36
; AVX2-NOGATHER-NEXT:  .LBB5_35: # %cond.load69
; AVX2-NOGATHER-NEXT:    vpinsrd $1, c+28(%rip), %xmm0, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_38
; AVX2-NOGATHER-NEXT:  .LBB5_37: # %cond.load74
; AVX2-NOGATHER-NEXT:    vpinsrd $2, c+28(%rip), %xmm0, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_40
; AVX2-NOGATHER-NEXT:  .LBB5_39: # %cond.load79
; AVX2-NOGATHER-NEXT:    vpinsrd $3, c+28(%rip), %xmm0, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $16, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_42
; AVX2-NOGATHER-NEXT:  .LBB5_41: # %cond.load84
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm3[4],ymm0[5,6,7]
; AVX2-NOGATHER-NEXT:    testb $32, %al
; AVX2-NOGATHER-NEXT:    je .LBB5_44
; AVX2-NOGATHER-NEXT:  .LBB5_43: # %cond.load89
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5],ymm0[6,7]
; AVX2-NOGATHER-NEXT:    testb $64, %al
; AVX2-NOGATHER-NEXT:    jne .LBB5_45
; AVX2-NOGATHER-NEXT:    jmp .LBB5_46
;
; AVX2-GATHER-LABEL: gather_v8i32_v8i32:
; AVX2-GATHER:       # %bb.0:
; AVX2-GATHER-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; AVX2-GATHER-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm0
; AVX2-GATHER-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX2-GATHER-NEXT:    vmovdqa %ymm0, %ymm3
; AVX2-GATHER-NEXT:    vpgatherdd %ymm3, c+12(,%ymm1), %ymm2
; AVX2-GATHER-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-GATHER-NEXT:    vpgatherdd %ymm0, c+28(,%ymm1), %ymm3
; AVX2-GATHER-NEXT:    vpaddd %ymm3, %ymm2, %ymm0
; AVX2-GATHER-NEXT:    vpaddd %ymm3, %ymm0, %ymm0
; AVX2-GATHER-NEXT:    retq
;
; AVX512F-LABEL: gather_v8i32_v8i32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
; AVX512F-NEXT:    vptestnmd %zmm0, %zmm0, %k0
; AVX512F-NEXT:    kshiftlw $8, %k0, %k0
; AVX512F-NEXT:    kshiftrw $8, %k0, %k1
; AVX512F-NEXT:    vpxor %xmm0, %xmm0, %xmm0
; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT:    kmovw %k1, %k2
; AVX512F-NEXT:    vpgatherdd c+12(,%zmm0), %zmm1 {%k2}
; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vpgatherdd c+28(,%zmm0), %zmm2 {%k1}
; AVX512F-NEXT:    vpaddd %ymm2, %ymm1, %ymm0
; AVX512F-NEXT:    vpaddd %ymm2, %ymm0, %ymm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: gather_v8i32_v8i32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vptestnmd %ymm0, %ymm0, %k1
; AVX512VL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT:    kmovw %k1, %k2
; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; AVX512VL-NEXT:    vpgatherdd c+12(,%ymm0), %ymm1 {%k2}
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpgatherdd c+28(,%ymm0), %ymm2 {%k1}
; AVX512VL-NEXT:    vpaddd %ymm2, %ymm1, %ymm0
; AVX512VL-NEXT:    vpaddd %ymm2, %ymm0, %ymm0
; AVX512VL-NEXT:    retq
  %1 = icmp eq <8 x i32> %trigger, zeroinitializer
  %2 = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> getelementptr (%struct.a, <8 x ptr> <ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c>, <8 x i64> zeroinitializer, i32 0, <8 x i64> <i64 3, i64 3, i64 3, i64 3, i64 3, i64 3, i64 3, i64 3>), i32 4, <8 x i1> %1, <8 x i32> undef)
  %3 = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> getelementptr (%struct.a, <8 x ptr> <ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c>, <8 x i64> zeroinitializer, i32 3), i32 4, <8 x i1> %1, <8 x i32> undef)
  %4 = add <8 x i32> %2, %3
  %5 = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> getelementptr (%struct.a, <8 x ptr> <ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c>, <8 x i64> zeroinitializer, i32 3), i32 4, <8 x i1> %1, <8 x i32> undef)
  %6 = add <8 x i32> %4, %5
  ret <8 x i32> %6
}

define <8 x i32> @masked_gather_v8i32_v8i32(i8 %trigger) {
; SSE-LABEL: masked_gather_v8i32_v8i32:
; SSE:       # %bb.0:
; SSE-NEXT:    testb $1, %dil
; SSE-NEXT:    je .LBB6_1
; SSE-NEXT:  # %bb.2: # %cond.load
; SSE-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE-NEXT:    testb $2, %dil
; SSE-NEXT:    jne .LBB6_4
; SSE-NEXT:    jmp .LBB6_5
; SSE-NEXT:  .LBB6_1:
; SSE-NEXT:    # implicit-def: $xmm0
; SSE-NEXT:    testb $2, %dil
; SSE-NEXT:    je .LBB6_5
; SSE-NEXT:  .LBB6_4: # %cond.load1
; SSE-NEXT:    pinsrd $1, c+12(%rip), %xmm0
; SSE-NEXT:  .LBB6_5: # %else2
; SSE-NEXT:    movd %edi, %xmm1
; SSE-NEXT:    testb $4, %dil
; SSE-NEXT:    je .LBB6_7
; SSE-NEXT:  # %bb.6: # %cond.load4
; SSE-NEXT:    pinsrd $2, c+12(%rip), %xmm0
; SSE-NEXT:  .LBB6_7: # %else5
; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
; SSE-NEXT:    testb $8, %dil
; SSE-NEXT:    je .LBB6_9
; SSE-NEXT:  # %bb.8: # %cond.load7
; SSE-NEXT:    pinsrd $3, c+12(%rip), %xmm0
; SSE-NEXT:  .LBB6_9: # %else8
; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[0,1,0,1]
; SSE-NEXT:    pmovzxbw {{.*#+}} xmm2 = [1,2,4,8,16,32,64,128]
; SSE-NEXT:    testb $16, %dil
; SSE-NEXT:    je .LBB6_10
; SSE-NEXT:  # %bb.11: # %cond.load10
; SSE-NEXT:    pinsrd $0, c+12(%rip), %xmm1
; SSE-NEXT:    jmp .LBB6_12
; SSE-NEXT:  .LBB6_10:
; SSE-NEXT:    # implicit-def: $xmm1
; SSE-NEXT:  .LBB6_12: # %else11
; SSE-NEXT:    pand %xmm2, %xmm3
; SSE-NEXT:    testb $32, %dil
; SSE-NEXT:    je .LBB6_14
; SSE-NEXT:  # %bb.13: # %cond.load13
; SSE-NEXT:    pinsrd $1, c+12(%rip), %xmm1
; SSE-NEXT:  .LBB6_14: # %else14
; SSE-NEXT:    pcmpeqw %xmm2, %xmm3
; SSE-NEXT:    testb $64, %dil
; SSE-NEXT:    je .LBB6_16
; SSE-NEXT:  # %bb.15: # %cond.load16
; SSE-NEXT:    pinsrd $2, c+12(%rip), %xmm1
; SSE-NEXT:  .LBB6_16: # %else17
; SSE-NEXT:    psrlw $15, %xmm3
; SSE-NEXT:    testb $-128, %dil
; SSE-NEXT:    je .LBB6_18
; SSE-NEXT:  # %bb.17: # %cond.load19
; SSE-NEXT:    pinsrd $3, c+12(%rip), %xmm1
; SSE-NEXT:  .LBB6_18: # %else20
; SSE-NEXT:    psllw $15, %xmm3
; SSE-NEXT:    movdqa %xmm3, %xmm2
; SSE-NEXT:    packsswb %xmm2, %xmm2
; SSE-NEXT:    pmovmskb %xmm2, %eax
; SSE-NEXT:    testb $1, %al
; SSE-NEXT:    je .LBB6_19
; SSE-NEXT:  # %bb.20: # %cond.load23
; SSE-NEXT:    movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    jne .LBB6_22
; SSE-NEXT:    jmp .LBB6_23
; SSE-NEXT:  .LBB6_19:
; SSE-NEXT:    # implicit-def: $xmm2
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    je .LBB6_23
; SSE-NEXT:  .LBB6_22: # %cond.load28
; SSE-NEXT:    pinsrd $1, c+28(%rip), %xmm2
; SSE-NEXT:  .LBB6_23: # %else31
; SSE-NEXT:    testb $4, %al
; SSE-NEXT:    jne .LBB6_24
; SSE-NEXT:  # %bb.25: # %else36
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    jne .LBB6_26
; SSE-NEXT:  .LBB6_27: # %else41
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    je .LBB6_28
; SSE-NEXT:  .LBB6_29: # %cond.load43
; SSE-NEXT:    pinsrd $0, c+28(%rip), %xmm4
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    jne .LBB6_31
; SSE-NEXT:    jmp .LBB6_32
; SSE-NEXT:  .LBB6_24: # %cond.load33
; SSE-NEXT:    pinsrd $2, c+28(%rip), %xmm2
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    je .LBB6_27
; SSE-NEXT:  .LBB6_26: # %cond.load38
; SSE-NEXT:    pinsrd $3, c+28(%rip), %xmm2
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    jne .LBB6_29
; SSE-NEXT:  .LBB6_28:
; SSE-NEXT:    # implicit-def: $xmm4
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    je .LBB6_32
; SSE-NEXT:  .LBB6_31: # %cond.load48
; SSE-NEXT:    pinsrd $1, c+28(%rip), %xmm4
; SSE-NEXT:  .LBB6_32: # %else51
; SSE-NEXT:    testb $64, %al
; SSE-NEXT:    jne .LBB6_33
; SSE-NEXT:  # %bb.34: # %else56
; SSE-NEXT:    testb $-128, %al
; SSE-NEXT:    je .LBB6_36
; SSE-NEXT:  .LBB6_35: # %cond.load58
; SSE-NEXT:    pinsrd $3, c+28(%rip), %xmm4
; SSE-NEXT:  .LBB6_36: # %else61
; SSE-NEXT:    packsswb %xmm3, %xmm3
; SSE-NEXT:    pmovmskb %xmm3, %eax
; SSE-NEXT:    testb $1, %al
; SSE-NEXT:    je .LBB6_37
; SSE-NEXT:  # %bb.38: # %cond.load64
; SSE-NEXT:    movd {{.*#+}} xmm3 = mem[0],zero,zero,zero
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    jne .LBB6_40
; SSE-NEXT:    jmp .LBB6_41
; SSE-NEXT:  .LBB6_33: # %cond.load53
; SSE-NEXT:    pinsrd $2, c+28(%rip), %xmm4
; SSE-NEXT:    testb $-128, %al
; SSE-NEXT:    jne .LBB6_35
; SSE-NEXT:    jmp .LBB6_36
; SSE-NEXT:  .LBB6_37:
; SSE-NEXT:    # implicit-def: $xmm3
; SSE-NEXT:    testb $2, %al
; SSE-NEXT:    je .LBB6_41
; SSE-NEXT:  .LBB6_40: # %cond.load69
; SSE-NEXT:    pinsrd $1, c+28(%rip), %xmm3
; SSE-NEXT:  .LBB6_41: # %else72
; SSE-NEXT:    testb $4, %al
; SSE-NEXT:    jne .LBB6_42
; SSE-NEXT:  # %bb.43: # %else77
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    jne .LBB6_44
; SSE-NEXT:  .LBB6_45: # %else82
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    je .LBB6_46
; SSE-NEXT:  .LBB6_47: # %cond.load84
; SSE-NEXT:    pinsrd $0, c+28(%rip), %xmm5
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    jne .LBB6_49
; SSE-NEXT:    jmp .LBB6_50
; SSE-NEXT:  .LBB6_42: # %cond.load74
; SSE-NEXT:    pinsrd $2, c+28(%rip), %xmm3
; SSE-NEXT:    testb $8, %al
; SSE-NEXT:    je .LBB6_45
; SSE-NEXT:  .LBB6_44: # %cond.load79
; SSE-NEXT:    pinsrd $3, c+28(%rip), %xmm3
; SSE-NEXT:    testb $16, %al
; SSE-NEXT:    jne .LBB6_47
; SSE-NEXT:  .LBB6_46:
; SSE-NEXT:    # implicit-def: $xmm5
; SSE-NEXT:    testb $32, %al
; SSE-NEXT:    je .LBB6_50
; SSE-NEXT:  .LBB6_49: # %cond.load89
; SSE-NEXT:    pinsrd $1, c+28(%rip), %xmm5
; SSE-NEXT:  .LBB6_50: # %else92
; SSE-NEXT:    testb $64, %al
; SSE-NEXT:    je .LBB6_52
; SSE-NEXT:  # %bb.51: # %cond.load94
; SSE-NEXT:    pinsrd $2, c+28(%rip), %xmm5
; SSE-NEXT:  .LBB6_52: # %else97
; SSE-NEXT:    paddd %xmm2, %xmm0
; SSE-NEXT:    paddd %xmm4, %xmm1
; SSE-NEXT:    testb $-128, %al
; SSE-NEXT:    je .LBB6_54
; SSE-NEXT:  # %bb.53: # %cond.load99
; SSE-NEXT:    pinsrd $3, c+28(%rip), %xmm5
; SSE-NEXT:  .LBB6_54: # %else102
; SSE-NEXT:    paddd %xmm5, %xmm1
; SSE-NEXT:    paddd %xmm3, %xmm0
; SSE-NEXT:    retq
;
; AVX1-LABEL: masked_gather_v8i32_v8i32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    testb $1, %dil
; AVX1-NEXT:    # implicit-def: $ymm0
; AVX1-NEXT:    je .LBB6_2
; AVX1-NEXT:  # %bb.1: # %cond.load
; AVX1-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX1-NEXT:  .LBB6_2: # %else
; AVX1-NEXT:    testb $2, %dil
; AVX1-NEXT:    je .LBB6_4
; AVX1-NEXT:  # %bb.3: # %cond.load1
; AVX1-NEXT:    vpinsrd $1, c+12(%rip), %xmm0, %xmm1
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX1-NEXT:  .LBB6_4: # %else2
; AVX1-NEXT:    vmovd %edi, %xmm1
; AVX1-NEXT:    testb $4, %dil
; AVX1-NEXT:    je .LBB6_6
; AVX1-NEXT:  # %bb.5: # %cond.load4
; AVX1-NEXT:    vpinsrd $2, c+12(%rip), %xmm0, %xmm2
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX1-NEXT:  .LBB6_6: # %else5
; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
; AVX1-NEXT:    testb $8, %dil
; AVX1-NEXT:    je .LBB6_8
; AVX1-NEXT:  # %bb.7: # %cond.load7
; AVX1-NEXT:    vpinsrd $3, c+12(%rip), %xmm0, %xmm2
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX1-NEXT:  .LBB6_8: # %else8
; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[0,1,0,1]
; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = [1,2,4,8,16,32,64,128]
; AVX1-NEXT:    testb $16, %dil
; AVX1-NEXT:    je .LBB6_10
; AVX1-NEXT:  # %bb.9: # %cond.load10
; AVX1-NEXT:    vbroadcastss c+12(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm3[4],ymm0[5,6,7]
; AVX1-NEXT:  .LBB6_10: # %else11
; AVX1-NEXT:    vpand %xmm1, %xmm2, %xmm2
; AVX1-NEXT:    testb $32, %dil
; AVX1-NEXT:    je .LBB6_12
; AVX1-NEXT:  # %bb.11: # %cond.load13
; AVX1-NEXT:    vbroadcastss c+12(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5],ymm0[6,7]
; AVX1-NEXT:  .LBB6_12: # %else14
; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm2, %xmm1
; AVX1-NEXT:    testb $64, %dil
; AVX1-NEXT:    je .LBB6_14
; AVX1-NEXT:  # %bb.13: # %cond.load16
; AVX1-NEXT:    vbroadcastss c+12(%rip), %ymm2
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6],ymm0[7]
; AVX1-NEXT:  .LBB6_14: # %else17
; AVX1-NEXT:    vpsrlw $15, %xmm1, %xmm1
; AVX1-NEXT:    testb $-128, %dil
; AVX1-NEXT:    je .LBB6_16
; AVX1-NEXT:  # %bb.15: # %cond.load19
; AVX1-NEXT:    vbroadcastss c+12(%rip), %ymm2
; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
; AVX1-NEXT:  .LBB6_16: # %else20
; AVX1-NEXT:    vpsllw $15, %xmm1, %xmm2
; AVX1-NEXT:    vpacksswb %xmm2, %xmm2, %xmm1
; AVX1-NEXT:    vpmovmskb %xmm1, %eax
; AVX1-NEXT:    testb $1, %al
; AVX1-NEXT:    # implicit-def: $ymm1
; AVX1-NEXT:    jne .LBB6_17
; AVX1-NEXT:  # %bb.18: # %else26
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    jne .LBB6_19
; AVX1-NEXT:  .LBB6_20: # %else31
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    jne .LBB6_21
; AVX1-NEXT:  .LBB6_22: # %else36
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    jne .LBB6_23
; AVX1-NEXT:  .LBB6_24: # %else41
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    jne .LBB6_25
; AVX1-NEXT:  .LBB6_26: # %else46
; AVX1-NEXT:    testb $32, %al
; AVX1-NEXT:    jne .LBB6_27
; AVX1-NEXT:  .LBB6_28: # %else51
; AVX1-NEXT:    testb $64, %al
; AVX1-NEXT:    jne .LBB6_29
; AVX1-NEXT:  .LBB6_30: # %else56
; AVX1-NEXT:    testb $-128, %al
; AVX1-NEXT:    je .LBB6_32
; AVX1-NEXT:  .LBB6_31: # %cond.load58
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
; AVX1-NEXT:  .LBB6_32: # %else61
; AVX1-NEXT:    vpacksswb %xmm2, %xmm2, %xmm2
; AVX1-NEXT:    vpmovmskb %xmm2, %eax
; AVX1-NEXT:    testb $1, %al
; AVX1-NEXT:    # implicit-def: $ymm2
; AVX1-NEXT:    jne .LBB6_33
; AVX1-NEXT:  # %bb.34: # %else67
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    jne .LBB6_35
; AVX1-NEXT:  .LBB6_36: # %else72
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    jne .LBB6_37
; AVX1-NEXT:  .LBB6_38: # %else77
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    jne .LBB6_39
; AVX1-NEXT:  .LBB6_40: # %else82
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    je .LBB6_42
; AVX1-NEXT:  .LBB6_41: # %cond.load84
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4],ymm2[5,6,7]
; AVX1-NEXT:  .LBB6_42: # %else87
; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    testb $32, %al
; AVX1-NEXT:    je .LBB6_44
; AVX1-NEXT:  # %bb.43: # %cond.load89
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm1
; AVX1-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm1[5],ymm2[6,7]
; AVX1-NEXT:  .LBB6_44: # %else92
; AVX1-NEXT:    vpaddd %xmm3, %xmm4, %xmm1
; AVX1-NEXT:    testb $64, %al
; AVX1-NEXT:    je .LBB6_46
; AVX1-NEXT:  # %bb.45: # %cond.load94
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6],ymm2[7]
; AVX1-NEXT:  .LBB6_46: # %else97
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT:    testb $-128, %al
; AVX1-NEXT:    je .LBB6_48
; AVX1-NEXT:  # %bb.47: # %cond.load99
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm1
; AVX1-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm1[7]
; AVX1-NEXT:  .LBB6_48: # %else102
; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm1
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
; AVX1-NEXT:    vpaddd %xmm1, %xmm3, %xmm1
; AVX1-NEXT:    vpaddd %xmm2, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT:    retq
; AVX1-NEXT:  .LBB6_17: # %cond.load23
; AVX1-NEXT:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    je .LBB6_20
; AVX1-NEXT:  .LBB6_19: # %cond.load28
; AVX1-NEXT:    vpinsrd $1, c+28(%rip), %xmm1, %xmm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    je .LBB6_22
; AVX1-NEXT:  .LBB6_21: # %cond.load33
; AVX1-NEXT:    vpinsrd $2, c+28(%rip), %xmm1, %xmm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    je .LBB6_24
; AVX1-NEXT:  .LBB6_23: # %cond.load38
; AVX1-NEXT:    vpinsrd $3, c+28(%rip), %xmm1, %xmm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    je .LBB6_26
; AVX1-NEXT:  .LBB6_25: # %cond.load43
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4],ymm1[5,6,7]
; AVX1-NEXT:    testb $32, %al
; AVX1-NEXT:    je .LBB6_28
; AVX1-NEXT:  .LBB6_27: # %cond.load48
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5],ymm1[6,7]
; AVX1-NEXT:    testb $64, %al
; AVX1-NEXT:    je .LBB6_30
; AVX1-NEXT:  .LBB6_29: # %cond.load53
; AVX1-NEXT:    vbroadcastss c+28(%rip), %ymm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm3[6],ymm1[7]
; AVX1-NEXT:    testb $-128, %al
; AVX1-NEXT:    jne .LBB6_31
; AVX1-NEXT:    jmp .LBB6_32
; AVX1-NEXT:  .LBB6_33: # %cond.load64
; AVX1-NEXT:    vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX1-NEXT:    testb $2, %al
; AVX1-NEXT:    je .LBB6_36
; AVX1-NEXT:  .LBB6_35: # %cond.load69
; AVX1-NEXT:    vpinsrd $1, c+28(%rip), %xmm2, %xmm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
; AVX1-NEXT:    testb $4, %al
; AVX1-NEXT:    je .LBB6_38
; AVX1-NEXT:  .LBB6_37: # %cond.load74
; AVX1-NEXT:    vpinsrd $2, c+28(%rip), %xmm2, %xmm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
; AVX1-NEXT:    testb $8, %al
; AVX1-NEXT:    je .LBB6_40
; AVX1-NEXT:  .LBB6_39: # %cond.load79
; AVX1-NEXT:    vpinsrd $3, c+28(%rip), %xmm2, %xmm3
; AVX1-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
; AVX1-NEXT:    testb $16, %al
; AVX1-NEXT:    jne .LBB6_41
; AVX1-NEXT:    jmp .LBB6_42
;
; AVX2-NOGATHER-LABEL: masked_gather_v8i32_v8i32:
; AVX2-NOGATHER:       # %bb.0:
; AVX2-NOGATHER-NEXT:    testb $1, %dil
; AVX2-NOGATHER-NEXT:    # implicit-def: $ymm0
; AVX2-NOGATHER-NEXT:    jne .LBB6_1
; AVX2-NOGATHER-NEXT:  # %bb.2: # %else
; AVX2-NOGATHER-NEXT:    testb $2, %dil
; AVX2-NOGATHER-NEXT:    jne .LBB6_3
; AVX2-NOGATHER-NEXT:  .LBB6_4: # %else2
; AVX2-NOGATHER-NEXT:    testb $4, %dil
; AVX2-NOGATHER-NEXT:    je .LBB6_6
; AVX2-NOGATHER-NEXT:  .LBB6_5: # %cond.load4
; AVX2-NOGATHER-NEXT:    vpinsrd $2, c+12(%rip), %xmm0, %xmm1
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-NOGATHER-NEXT:  .LBB6_6: # %else5
; AVX2-NOGATHER-NEXT:    vmovd %edi, %xmm1
; AVX2-NOGATHER-NEXT:    testb $8, %dil
; AVX2-NOGATHER-NEXT:    je .LBB6_8
; AVX2-NOGATHER-NEXT:  # %bb.7: # %cond.load7
; AVX2-NOGATHER-NEXT:    vpinsrd $3, c+12(%rip), %xmm0, %xmm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-NOGATHER-NEXT:  .LBB6_8: # %else8
; AVX2-NOGATHER-NEXT:    vpbroadcastb %xmm1, %xmm2
; AVX2-NOGATHER-NEXT:    vpmovzxbw {{.*#+}} xmm1 = [1,2,4,8,16,32,64,128]
; AVX2-NOGATHER-NEXT:    testb $16, %dil
; AVX2-NOGATHER-NEXT:    je .LBB6_10
; AVX2-NOGATHER-NEXT:  # %bb.9: # %cond.load10
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+12(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm3[4],ymm0[5,6,7]
; AVX2-NOGATHER-NEXT:  .LBB6_10: # %else11
; AVX2-NOGATHER-NEXT:    vpand %xmm1, %xmm2, %xmm2
; AVX2-NOGATHER-NEXT:    testb $32, %dil
; AVX2-NOGATHER-NEXT:    je .LBB6_12
; AVX2-NOGATHER-NEXT:  # %bb.11: # %cond.load13
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+12(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5],ymm0[6,7]
; AVX2-NOGATHER-NEXT:  .LBB6_12: # %else14
; AVX2-NOGATHER-NEXT:    vpcmpeqw %xmm1, %xmm2, %xmm1
; AVX2-NOGATHER-NEXT:    testb $64, %dil
; AVX2-NOGATHER-NEXT:    je .LBB6_14
; AVX2-NOGATHER-NEXT:  # %bb.13: # %cond.load16
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+12(%rip), %ymm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6],ymm0[7]
; AVX2-NOGATHER-NEXT:  .LBB6_14: # %else17
; AVX2-NOGATHER-NEXT:    vpsrlw $15, %xmm1, %xmm1
; AVX2-NOGATHER-NEXT:    testb $-128, %dil
; AVX2-NOGATHER-NEXT:    je .LBB6_16
; AVX2-NOGATHER-NEXT:  # %bb.15: # %cond.load19
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+12(%rip), %ymm2
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
; AVX2-NOGATHER-NEXT:  .LBB6_16: # %else20
; AVX2-NOGATHER-NEXT:    vpsllw $15, %xmm1, %xmm2
; AVX2-NOGATHER-NEXT:    vpacksswb %xmm2, %xmm2, %xmm1
; AVX2-NOGATHER-NEXT:    vpmovmskb %xmm1, %eax
; AVX2-NOGATHER-NEXT:    testb $1, %al
; AVX2-NOGATHER-NEXT:    # implicit-def: $ymm1
; AVX2-NOGATHER-NEXT:    jne .LBB6_17
; AVX2-NOGATHER-NEXT:  # %bb.18: # %else26
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_19
; AVX2-NOGATHER-NEXT:  .LBB6_20: # %else31
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_21
; AVX2-NOGATHER-NEXT:  .LBB6_22: # %else36
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_23
; AVX2-NOGATHER-NEXT:  .LBB6_24: # %else41
; AVX2-NOGATHER-NEXT:    testb $16, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_25
; AVX2-NOGATHER-NEXT:  .LBB6_26: # %else46
; AVX2-NOGATHER-NEXT:    testb $32, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_27
; AVX2-NOGATHER-NEXT:  .LBB6_28: # %else51
; AVX2-NOGATHER-NEXT:    testb $64, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_29
; AVX2-NOGATHER-NEXT:  .LBB6_30: # %else56
; AVX2-NOGATHER-NEXT:    testb $-128, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_32
; AVX2-NOGATHER-NEXT:  .LBB6_31: # %cond.load58
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
; AVX2-NOGATHER-NEXT:  .LBB6_32: # %else61
; AVX2-NOGATHER-NEXT:    vpacksswb %xmm2, %xmm2, %xmm2
; AVX2-NOGATHER-NEXT:    vpmovmskb %xmm2, %eax
; AVX2-NOGATHER-NEXT:    testb $1, %al
; AVX2-NOGATHER-NEXT:    # implicit-def: $ymm2
; AVX2-NOGATHER-NEXT:    jne .LBB6_33
; AVX2-NOGATHER-NEXT:  # %bb.34: # %else67
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_35
; AVX2-NOGATHER-NEXT:  .LBB6_36: # %else72
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_37
; AVX2-NOGATHER-NEXT:  .LBB6_38: # %else77
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_39
; AVX2-NOGATHER-NEXT:  .LBB6_40: # %else82
; AVX2-NOGATHER-NEXT:    testb $16, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_41
; AVX2-NOGATHER-NEXT:  .LBB6_42: # %else87
; AVX2-NOGATHER-NEXT:    testb $32, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_43
; AVX2-NOGATHER-NEXT:  .LBB6_44: # %else92
; AVX2-NOGATHER-NEXT:    testb $64, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_46
; AVX2-NOGATHER-NEXT:  .LBB6_45: # %cond.load94
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6],ymm2[7]
; AVX2-NOGATHER-NEXT:  .LBB6_46: # %else97
; AVX2-NOGATHER-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
; AVX2-NOGATHER-NEXT:    testb $-128, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_48
; AVX2-NOGATHER-NEXT:  # %bb.47: # %cond.load99
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm1
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm1[7]
; AVX2-NOGATHER-NEXT:  .LBB6_48: # %else102
; AVX2-NOGATHER-NEXT:    vpaddd %ymm2, %ymm0, %ymm0
; AVX2-NOGATHER-NEXT:    retq
; AVX2-NOGATHER-NEXT:  .LBB6_1: # %cond.load
; AVX2-NOGATHER-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX2-NOGATHER-NEXT:    testb $2, %dil
; AVX2-NOGATHER-NEXT:    je .LBB6_4
; AVX2-NOGATHER-NEXT:  .LBB6_3: # %cond.load1
; AVX2-NOGATHER-NEXT:    vpinsrd $1, c+12(%rip), %xmm0, %xmm1
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $4, %dil
; AVX2-NOGATHER-NEXT:    jne .LBB6_5
; AVX2-NOGATHER-NEXT:    jmp .LBB6_6
; AVX2-NOGATHER-NEXT:  .LBB6_17: # %cond.load23
; AVX2-NOGATHER-NEXT:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_20
; AVX2-NOGATHER-NEXT:  .LBB6_19: # %cond.load28
; AVX2-NOGATHER-NEXT:    vpinsrd $1, c+28(%rip), %xmm1, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_22
; AVX2-NOGATHER-NEXT:  .LBB6_21: # %cond.load33
; AVX2-NOGATHER-NEXT:    vpinsrd $2, c+28(%rip), %xmm1, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_24
; AVX2-NOGATHER-NEXT:  .LBB6_23: # %cond.load38
; AVX2-NOGATHER-NEXT:    vpinsrd $3, c+28(%rip), %xmm1, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $16, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_26
; AVX2-NOGATHER-NEXT:  .LBB6_25: # %cond.load43
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4],ymm1[5,6,7]
; AVX2-NOGATHER-NEXT:    testb $32, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_28
; AVX2-NOGATHER-NEXT:  .LBB6_27: # %cond.load48
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5],ymm1[6,7]
; AVX2-NOGATHER-NEXT:    testb $64, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_30
; AVX2-NOGATHER-NEXT:  .LBB6_29: # %cond.load53
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm3[6],ymm1[7]
; AVX2-NOGATHER-NEXT:    testb $-128, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_31
; AVX2-NOGATHER-NEXT:    jmp .LBB6_32
; AVX2-NOGATHER-NEXT:  .LBB6_33: # %cond.load64
; AVX2-NOGATHER-NEXT:    vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX2-NOGATHER-NEXT:    testb $2, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_36
; AVX2-NOGATHER-NEXT:  .LBB6_35: # %cond.load69
; AVX2-NOGATHER-NEXT:    vpinsrd $1, c+28(%rip), %xmm2, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $4, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_38
; AVX2-NOGATHER-NEXT:  .LBB6_37: # %cond.load74
; AVX2-NOGATHER-NEXT:    vpinsrd $2, c+28(%rip), %xmm2, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $8, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_40
; AVX2-NOGATHER-NEXT:  .LBB6_39: # %cond.load79
; AVX2-NOGATHER-NEXT:    vpinsrd $3, c+28(%rip), %xmm2, %xmm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
; AVX2-NOGATHER-NEXT:    testb $16, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_42
; AVX2-NOGATHER-NEXT:  .LBB6_41: # %cond.load84
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4],ymm2[5,6,7]
; AVX2-NOGATHER-NEXT:    testb $32, %al
; AVX2-NOGATHER-NEXT:    je .LBB6_44
; AVX2-NOGATHER-NEXT:  .LBB6_43: # %cond.load89
; AVX2-NOGATHER-NEXT:    vpbroadcastd c+28(%rip), %ymm3
; AVX2-NOGATHER-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5],ymm2[6,7]
; AVX2-NOGATHER-NEXT:    testb $64, %al
; AVX2-NOGATHER-NEXT:    jne .LBB6_45
; AVX2-NOGATHER-NEXT:    jmp .LBB6_46
;
; AVX2-GATHER-LABEL: masked_gather_v8i32_v8i32:
; AVX2-GATHER:       # %bb.0:
; AVX2-GATHER-NEXT:    vmovd %edi, %xmm0
; AVX2-GATHER-NEXT:    vpbroadcastb %xmm0, %ymm0
; AVX2-GATHER-NEXT:    vpmovzxbd {{.*#+}} ymm1 = [1,2,4,8,16,32,64,128]
; AVX2-GATHER-NEXT:    vpand %ymm1, %ymm0, %ymm0
; AVX2-GATHER-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm0
; AVX2-GATHER-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; AVX2-GATHER-NEXT:    vmovdqa %ymm0, %ymm2
; AVX2-GATHER-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; AVX2-GATHER-NEXT:    vpgatherdd %ymm2, c+12(,%ymm1), %ymm3
; AVX2-GATHER-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX2-GATHER-NEXT:    vpgatherdd %ymm0, c+28(,%ymm1), %ymm2
; AVX2-GATHER-NEXT:    vpaddd %ymm2, %ymm3, %ymm0
; AVX2-GATHER-NEXT:    vpaddd %ymm2, %ymm0, %ymm0
; AVX2-GATHER-NEXT:    retq
;
; AVX512F-LABEL: masked_gather_v8i32_v8i32:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    movzbl %dil, %eax
; AVX512F-NEXT:    kmovw %eax, %k1
; AVX512F-NEXT:    vpxor %xmm0, %xmm0, %xmm0
; AVX512F-NEXT:    kmovw %k1, %k2
; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT:    vpgatherdd c+12(,%zmm0), %zmm1 {%k2}
; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vpgatherdd c+28(,%zmm0), %zmm2 {%k1}
; AVX512F-NEXT:    vpaddd %ymm2, %ymm1, %ymm0
; AVX512F-NEXT:    vpaddd %ymm2, %ymm0, %ymm0
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: masked_gather_v8i32_v8i32:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    kmovw %edi, %k1
; AVX512VL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT:    kmovw %k1, %k2
; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; AVX512VL-NEXT:    vpgatherdd c+12(,%ymm0), %ymm1 {%k2}
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpgatherdd c+28(,%ymm0), %ymm2 {%k1}
; AVX512VL-NEXT:    vpaddd %ymm2, %ymm1, %ymm0
; AVX512VL-NEXT:    vpaddd %ymm2, %ymm0, %ymm0
; AVX512VL-NEXT:    retq
  %1 = bitcast i8 %trigger to <8 x i1>
  %2 = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> getelementptr (%struct.a, <8 x ptr> <ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c>, <8 x i64> zeroinitializer, i32 0, <8 x i64> <i64 3, i64 3, i64 3, i64 3, i64 3, i64 3, i64 3, i64 3>), i32 4, <8 x i1> %1, <8 x i32> undef)
  %3 = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> getelementptr (%struct.a, <8 x ptr> <ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c>, <8 x i64> zeroinitializer, i32 3), i32 4, <8 x i1> %1, <8 x i32> undef)
  %4 = add <8 x i32> %2, %3
  %5 = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> getelementptr (%struct.a, <8 x ptr> <ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c, ptr @c>, <8 x i64> zeroinitializer, i32 3), i32 4, <8 x i1> %1, <8 x i32> undef)
  %6 = add <8 x i32> %4, %5
  ret <8 x i32> %6
}

declare <2 x double> @llvm.masked.gather.v2f64.v2p0(<2 x ptr>, i32, <2 x i1>, <2 x double>)
declare <4 x double> @llvm.masked.gather.v4f64.v4p0(<4 x ptr>, i32, <4 x i1>, <4 x double>)

declare <4 x float> @llvm.masked.gather.v4f32.v4p0(<4 x ptr>, i32, <4 x i1>, <4 x float>)
declare <8 x float> @llvm.masked.gather.v8f32.v8p0(<8 x ptr>, i32, <8 x i1>, <8 x float>)

declare <16 x i8> @llvm.masked.gather.v16i8.v16p0(<16 x ptr>, i32, <16 x i1>, <16 x i8>)

declare <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr>, i32, <8 x i1>, <8 x i32>)