; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx  | FileCheck %s --check-prefixes=AVX
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX2-FP
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX2-FCP
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512vl | FileCheck %s --check-prefixes=AVX512
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512-FCP
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=AVX512DQ
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512vl,+avx512dq,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512DQ-FCP
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512vl,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512BW-FCP
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512vl,+avx512dq,+avx512bw | FileCheck %s --check-prefixes=AVX512DQ-BW
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512vl,+avx512dq,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512DQ-BW-FCP

; These patterns are produced by LoopVectorizer for interleaved loads.

define void @load_i32_stride3_vf2(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr %out.vec2) nounwind {
; SSE-LABEL: load_i32_stride3_vf2:
; SSE:       # %bb.0:
; SSE-NEXT:    movdqa (%rdi), %xmm0
; SSE-NEXT:    movdqa 16(%rdi), %xmm1
; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,3,2,3]
; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,1,1]
; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,1,1]
; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE-NEXT:    movq %xmm2, (%rsi)
; SSE-NEXT:    movq %xmm3, (%rdx)
; SSE-NEXT:    movq %xmm0, (%rcx)
; SSE-NEXT:    retq
;
; AVX-LABEL: load_i32_stride3_vf2:
; AVX:       # %bb.0:
; AVX-NEXT:    vmovaps (%rdi), %xmm0
; AVX-NEXT:    vmovaps 16(%rdi), %xmm1
; AVX-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[0,3,2,3]
; AVX-NEXT:    vblendps {{.*#+}} xmm3 = xmm1[0],xmm0[1],xmm1[2,3]
; AVX-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[1,0,2,3]
; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,2,3]
; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
; AVX-NEXT:    vmovlps %xmm2, (%rsi)
; AVX-NEXT:    vmovlps %xmm3, (%rdx)
; AVX-NEXT:    vmovlps %xmm0, (%rcx)
; AVX-NEXT:    retq
;
; AVX2-LABEL: load_i32_stride3_vf2:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovaps (%rdi), %xmm0
; AVX2-NEXT:    vmovaps 16(%rdi), %xmm1
; AVX2-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[0,3,2,3]
; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,0,2,3]
; AVX2-NEXT:    vbroadcastss 8(%rdi), %xmm3
; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0],xmm1[1],xmm3[2,3]
; AVX2-NEXT:    vmovlps %xmm2, (%rsi)
; AVX2-NEXT:    vmovlps %xmm0, (%rdx)
; AVX2-NEXT:    vmovlps %xmm1, (%rcx)
; AVX2-NEXT:    retq
;
; AVX2-FP-LABEL: load_i32_stride3_vf2:
; AVX2-FP:       # %bb.0:
; AVX2-FP-NEXT:    vmovaps (%rdi), %xmm0
; AVX2-FP-NEXT:    vmovaps 16(%rdi), %xmm1
; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[0,3,2,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,0,2,3]
; AVX2-FP-NEXT:    vbroadcastss 8(%rdi), %xmm3
; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0],xmm1[1],xmm3[2,3]
; AVX2-FP-NEXT:    vmovlps %xmm2, (%rsi)
; AVX2-FP-NEXT:    vmovlps %xmm0, (%rdx)
; AVX2-FP-NEXT:    vmovlps %xmm1, (%rcx)
; AVX2-FP-NEXT:    retq
;
; AVX2-FCP-LABEL: load_i32_stride3_vf2:
; AVX2-FCP:       # %bb.0:
; AVX2-FCP-NEXT:    vmovaps (%rdi), %xmm0
; AVX2-FCP-NEXT:    vmovaps 16(%rdi), %xmm1
; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[0,3,2,3]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,0,2,3]
; AVX2-FCP-NEXT:    vbroadcastss 8(%rdi), %xmm3
; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0],xmm1[1],xmm3[2,3]
; AVX2-FCP-NEXT:    vmovlps %xmm2, (%rsi)
; AVX2-FCP-NEXT:    vmovlps %xmm0, (%rdx)
; AVX2-FCP-NEXT:    vmovlps %xmm1, (%rcx)
; AVX2-FCP-NEXT:    retq
;
; AVX512-LABEL: load_i32_stride3_vf2:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovaps (%rdi), %xmm0
; AVX512-NEXT:    vmovaps 16(%rdi), %xmm1
; AVX512-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[0,3,2,3]
; AVX512-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
; AVX512-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,0,2,3]
; AVX512-NEXT:    vbroadcastss 8(%rdi), %xmm3
; AVX512-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0],xmm1[1],xmm3[2,3]
; AVX512-NEXT:    vmovlps %xmm2, (%rsi)
; AVX512-NEXT:    vmovlps %xmm0, (%rdx)
; AVX512-NEXT:    vmovlps %xmm1, (%rcx)
; AVX512-NEXT:    retq
;
; AVX512-FCP-LABEL: load_i32_stride3_vf2:
; AVX512-FCP:       # %bb.0:
; AVX512-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[0,3,2,3]
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm1 = [1,4,6,7]
; AVX512-FCP-NEXT:    vpermps (%rdi), %ymm1, %ymm1
; AVX512-FCP-NEXT:    vbroadcastss 8(%rdi), %xmm2
; AVX512-FCP-NEXT:    vblendps {{.*#+}} xmm2 = xmm2[0],mem[1],xmm2[2,3]
; AVX512-FCP-NEXT:    vmovlps %xmm0, (%rsi)
; AVX512-FCP-NEXT:    vmovlps %xmm1, (%rdx)
; AVX512-FCP-NEXT:    vmovlps %xmm2, (%rcx)
; AVX512-FCP-NEXT:    vzeroupper
; AVX512-FCP-NEXT:    retq
;
; AVX512DQ-LABEL: load_i32_stride3_vf2:
; AVX512DQ:       # %bb.0:
; AVX512DQ-NEXT:    vmovaps (%rdi), %xmm0
; AVX512DQ-NEXT:    vmovaps 16(%rdi), %xmm1
; AVX512DQ-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[0,3,2,3]
; AVX512DQ-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
; AVX512DQ-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,0,2,3]
; AVX512DQ-NEXT:    vbroadcastss 8(%rdi), %xmm3
; AVX512DQ-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0],xmm1[1],xmm3[2,3]
; AVX512DQ-NEXT:    vmovlps %xmm2, (%rsi)
; AVX512DQ-NEXT:    vmovlps %xmm0, (%rdx)
; AVX512DQ-NEXT:    vmovlps %xmm1, (%rcx)
; AVX512DQ-NEXT:    retq
;
; AVX512DQ-FCP-LABEL: load_i32_stride3_vf2:
; AVX512DQ-FCP:       # %bb.0:
; AVX512DQ-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[0,3,2,3]
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm1 = [1,4,6,7]
; AVX512DQ-FCP-NEXT:    vpermps (%rdi), %ymm1, %ymm1
; AVX512DQ-FCP-NEXT:    vbroadcastss 8(%rdi), %xmm2
; AVX512DQ-FCP-NEXT:    vblendps {{.*#+}} xmm2 = xmm2[0],mem[1],xmm2[2,3]
; AVX512DQ-FCP-NEXT:    vmovlps %xmm0, (%rsi)
; AVX512DQ-FCP-NEXT:    vmovlps %xmm1, (%rdx)
; AVX512DQ-FCP-NEXT:    vmovlps %xmm2, (%rcx)
; AVX512DQ-FCP-NEXT:    vzeroupper
; AVX512DQ-FCP-NEXT:    retq
;
; AVX512BW-LABEL: load_i32_stride3_vf2:
; AVX512BW:       # %bb.0:
; AVX512BW-NEXT:    vmovaps (%rdi), %xmm0
; AVX512BW-NEXT:    vmovaps 16(%rdi), %xmm1
; AVX512BW-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[0,3,2,3]
; AVX512BW-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
; AVX512BW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,0,2,3]
; AVX512BW-NEXT:    vbroadcastss 8(%rdi), %xmm3
; AVX512BW-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0],xmm1[1],xmm3[2,3]
; AVX512BW-NEXT:    vmovlps %xmm2, (%rsi)
; AVX512BW-NEXT:    vmovlps %xmm0, (%rdx)
; AVX512BW-NEXT:    vmovlps %xmm1, (%rcx)
; AVX512BW-NEXT:    retq
;
; AVX512BW-FCP-LABEL: load_i32_stride3_vf2:
; AVX512BW-FCP:       # %bb.0:
; AVX512BW-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[0,3,2,3]
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm1 = [1,4,6,7]
; AVX512BW-FCP-NEXT:    vpermps (%rdi), %ymm1, %ymm1
; AVX512BW-FCP-NEXT:    vbroadcastss 8(%rdi), %xmm2
; AVX512BW-FCP-NEXT:    vblendps {{.*#+}} xmm2 = xmm2[0],mem[1],xmm2[2,3]
; AVX512BW-FCP-NEXT:    vmovlps %xmm0, (%rsi)
; AVX512BW-FCP-NEXT:    vmovlps %xmm1, (%rdx)
; AVX512BW-FCP-NEXT:    vmovlps %xmm2, (%rcx)
; AVX512BW-FCP-NEXT:    vzeroupper
; AVX512BW-FCP-NEXT:    retq
;
; AVX512DQ-BW-LABEL: load_i32_stride3_vf2:
; AVX512DQ-BW:       # %bb.0:
; AVX512DQ-BW-NEXT:    vmovaps (%rdi), %xmm0
; AVX512DQ-BW-NEXT:    vmovaps 16(%rdi), %xmm1
; AVX512DQ-BW-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[0,3,2,3]
; AVX512DQ-BW-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
; AVX512DQ-BW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,0,2,3]
; AVX512DQ-BW-NEXT:    vbroadcastss 8(%rdi), %xmm3
; AVX512DQ-BW-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0],xmm1[1],xmm3[2,3]
; AVX512DQ-BW-NEXT:    vmovlps %xmm2, (%rsi)
; AVX512DQ-BW-NEXT:    vmovlps %xmm0, (%rdx)
; AVX512DQ-BW-NEXT:    vmovlps %xmm1, (%rcx)
; AVX512DQ-BW-NEXT:    retq
;
; AVX512DQ-BW-FCP-LABEL: load_i32_stride3_vf2:
; AVX512DQ-BW-FCP:       # %bb.0:
; AVX512DQ-BW-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[0,3,2,3]
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm1 = [1,4,6,7]
; AVX512DQ-BW-FCP-NEXT:    vpermps (%rdi), %ymm1, %ymm1
; AVX512DQ-BW-FCP-NEXT:    vbroadcastss 8(%rdi), %xmm2
; AVX512DQ-BW-FCP-NEXT:    vblendps {{.*#+}} xmm2 = xmm2[0],mem[1],xmm2[2,3]
; AVX512DQ-BW-FCP-NEXT:    vmovlps %xmm0, (%rsi)
; AVX512DQ-BW-FCP-NEXT:    vmovlps %xmm1, (%rdx)
; AVX512DQ-BW-FCP-NEXT:    vmovlps %xmm2, (%rcx)
; AVX512DQ-BW-FCP-NEXT:    vzeroupper
; AVX512DQ-BW-FCP-NEXT:    retq
  %wide.vec = load <6 x i32>, ptr %in.vec, align 64
  %strided.vec0 = shufflevector <6 x i32> %wide.vec, <6 x i32> poison, <2 x i32> <i32 0, i32 3>
  %strided.vec1 = shufflevector <6 x i32> %wide.vec, <6 x i32> poison, <2 x i32> <i32 1, i32 4>
  %strided.vec2 = shufflevector <6 x i32> %wide.vec, <6 x i32> poison, <2 x i32> <i32 2, i32 5>
  store <2 x i32> %strided.vec0, ptr %out.vec0, align 64
  store <2 x i32> %strided.vec1, ptr %out.vec1, align 64
  store <2 x i32> %strided.vec2, ptr %out.vec2, align 64
  ret void
}

define void @load_i32_stride3_vf4(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr %out.vec2) nounwind {
; SSE-LABEL: load_i32_stride3_vf4:
; SSE:       # %bb.0:
; SSE-NEXT:    movdqa (%rdi), %xmm0
; SSE-NEXT:    movaps 16(%rdi), %xmm1
; SSE-NEXT:    movaps 32(%rdi), %xmm2
; SSE-NEXT:    movdqa %xmm0, %xmm3
; SSE-NEXT:    movaps %xmm1, %xmm4
; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[2,3,2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm1[0,0]
; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm1[1,1,1,1]
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[1,0]
; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,3],xmm1[0,2]
; SSE-NEXT:    shufps {{.*#+}} xmm4 = xmm4[3,1],xmm2[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm4[0,2]
; SSE-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1]
; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,1],xmm2[0,3]
; SSE-NEXT:    movaps %xmm3, (%rsi)
; SSE-NEXT:    movaps %xmm0, (%rdx)
; SSE-NEXT:    movaps %xmm5, (%rcx)
; SSE-NEXT:    retq
;
; AVX-LABEL: load_i32_stride3_vf4:
; AVX:       # %bb.0:
; AVX-NEXT:    vmovaps (%rdi), %xmm0
; AVX-NEXT:    vmovaps 16(%rdi), %xmm1
; AVX-NEXT:    vmovaps 32(%rdi), %xmm2
; AVX-NEXT:    vblendps {{.*#+}} xmm3 = xmm0[0],xmm2[1],xmm0[2,3]
; AVX-NEXT:    vblendps {{.*#+}} xmm3 = xmm3[0,1],xmm1[2],xmm3[3]
; AVX-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,3,2,1]
; AVX-NEXT:    vmovaps 32(%rdi), %xmm4
; AVX-NEXT:    vblendps {{.*#+}} xmm4 = mem[0,1],xmm4[2,3]
; AVX-NEXT:    vblendps {{.*#+}} xmm4 = xmm1[0],xmm4[1,2],xmm1[3]
; AVX-NEXT:    vshufps {{.*#+}} xmm4 = xmm4[1,0,3,2]
; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm2[0,1],xmm0[2],xmm2[3]
; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,1,0,3]
; AVX-NEXT:    vmovaps %xmm3, (%rsi)
; AVX-NEXT:    vmovaps %xmm4, (%rdx)
; AVX-NEXT:    vmovaps %xmm0, (%rcx)
; AVX-NEXT:    retq
;
; AVX2-LABEL: load_i32_stride3_vf4:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovaps (%rdi), %ymm0
; AVX2-NEXT:    vmovaps 32(%rdi), %ymm1
; AVX2-NEXT:    vmovaps {{.*#+}} xmm2 = [0,3,6,1]
; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3,4,5,6,7]
; AVX2-NEXT:    vpermps %ymm3, %ymm2, %ymm2
; AVX2-NEXT:    vmovaps {{.*#+}} xmm3 = [1,4,7,2]
; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2,3],ymm0[4,5,6,7]
; AVX2-NEXT:    vpermps %ymm4, %ymm3, %ymm3
; AVX2-NEXT:    vmovaps {{.*#+}} xmm4 = [2,5,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3],ymm0[4,5,6,7]
; AVX2-NEXT:    vpermps %ymm0, %ymm4, %ymm0
; AVX2-NEXT:    vmovaps %xmm2, (%rsi)
; AVX2-NEXT:    vmovaps %xmm3, (%rdx)
; AVX2-NEXT:    vmovaps %xmm0, (%rcx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
;
; AVX2-FP-LABEL: load_i32_stride3_vf4:
; AVX2-FP:       # %bb.0:
; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm0
; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm1
; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm2 = [0,3,6,1]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3,4,5,6,7]
; AVX2-FP-NEXT:    vpermps %ymm3, %ymm2, %ymm2
; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm3 = [1,4,7,2]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2,3],ymm0[4,5,6,7]
; AVX2-FP-NEXT:    vpermps %ymm4, %ymm3, %ymm3
; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm4 = [2,5,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3],ymm0[4,5,6,7]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm4, %ymm0
; AVX2-FP-NEXT:    vmovaps %xmm2, (%rsi)
; AVX2-FP-NEXT:    vmovaps %xmm3, (%rdx)
; AVX2-FP-NEXT:    vmovaps %xmm0, (%rcx)
; AVX2-FP-NEXT:    vzeroupper
; AVX2-FP-NEXT:    retq
;
; AVX2-FCP-LABEL: load_i32_stride3_vf4:
; AVX2-FCP:       # %bb.0:
; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm0
; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm1
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm2 = [0,3,6,1]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3,4,5,6,7]
; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm2
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm3 = [1,4,7,2]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2,3],ymm0[4,5,6,7]
; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm3, %ymm3
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm4 = [2,5,0,3]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3],ymm0[4,5,6,7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm4, %ymm0
; AVX2-FCP-NEXT:    vmovaps %xmm2, (%rsi)
; AVX2-FCP-NEXT:    vmovaps %xmm3, (%rdx)
; AVX2-FCP-NEXT:    vmovaps %xmm0, (%rcx)
; AVX2-FCP-NEXT:    vzeroupper
; AVX2-FCP-NEXT:    retq
;
; AVX512-LABEL: load_i32_stride3_vf4:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,3,6,9]
; AVX512-NEXT:    vmovaps (%rdi), %zmm1
; AVX512-NEXT:    vpermps %zmm1, %zmm0, %zmm0
; AVX512-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [1,4,7,10]
; AVX512-NEXT:    vpermps %zmm1, %zmm2, %zmm2
; AVX512-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,5,8,11]
; AVX512-NEXT:    vpermps %zmm1, %zmm3, %zmm1
; AVX512-NEXT:    vmovaps %xmm0, (%rsi)
; AVX512-NEXT:    vmovaps %xmm2, (%rdx)
; AVX512-NEXT:    vmovaps %xmm1, (%rcx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
;
; AVX512-FCP-LABEL: load_i32_stride3_vf4:
; AVX512-FCP:       # %bb.0:
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,3,6,9]
; AVX512-FCP-NEXT:    vmovaps (%rdi), %zmm1
; AVX512-FCP-NEXT:    vpermps %zmm1, %zmm0, %zmm0
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [1,4,7,10]
; AVX512-FCP-NEXT:    vpermps %zmm1, %zmm2, %zmm2
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,5,8,11]
; AVX512-FCP-NEXT:    vpermps %zmm1, %zmm3, %zmm1
; AVX512-FCP-NEXT:    vmovaps %xmm0, (%rsi)
; AVX512-FCP-NEXT:    vmovaps %xmm2, (%rdx)
; AVX512-FCP-NEXT:    vmovaps %xmm1, (%rcx)
; AVX512-FCP-NEXT:    vzeroupper
; AVX512-FCP-NEXT:    retq
;
; AVX512DQ-LABEL: load_i32_stride3_vf4:
; AVX512DQ:       # %bb.0:
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,3,6,9]
; AVX512DQ-NEXT:    vmovaps (%rdi), %zmm1
; AVX512DQ-NEXT:    vpermps %zmm1, %zmm0, %zmm0
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [1,4,7,10]
; AVX512DQ-NEXT:    vpermps %zmm1, %zmm2, %zmm2
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,5,8,11]
; AVX512DQ-NEXT:    vpermps %zmm1, %zmm3, %zmm1
; AVX512DQ-NEXT:    vmovaps %xmm0, (%rsi)
; AVX512DQ-NEXT:    vmovaps %xmm2, (%rdx)
; AVX512DQ-NEXT:    vmovaps %xmm1, (%rcx)
; AVX512DQ-NEXT:    vzeroupper
; AVX512DQ-NEXT:    retq
;
; AVX512DQ-FCP-LABEL: load_i32_stride3_vf4:
; AVX512DQ-FCP:       # %bb.0:
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,3,6,9]
; AVX512DQ-FCP-NEXT:    vmovaps (%rdi), %zmm1
; AVX512DQ-FCP-NEXT:    vpermps %zmm1, %zmm0, %zmm0
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [1,4,7,10]
; AVX512DQ-FCP-NEXT:    vpermps %zmm1, %zmm2, %zmm2
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,5,8,11]
; AVX512DQ-FCP-NEXT:    vpermps %zmm1, %zmm3, %zmm1
; AVX512DQ-FCP-NEXT:    vmovaps %xmm0, (%rsi)
; AVX512DQ-FCP-NEXT:    vmovaps %xmm2, (%rdx)
; AVX512DQ-FCP-NEXT:    vmovaps %xmm1, (%rcx)
; AVX512DQ-FCP-NEXT:    vzeroupper
; AVX512DQ-FCP-NEXT:    retq
;
; AVX512BW-LABEL: load_i32_stride3_vf4:
; AVX512BW:       # %bb.0:
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,3,6,9]
; AVX512BW-NEXT:    vmovaps (%rdi), %zmm1
; AVX512BW-NEXT:    vpermps %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [1,4,7,10]
; AVX512BW-NEXT:    vpermps %zmm1, %zmm2, %zmm2
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,5,8,11]
; AVX512BW-NEXT:    vpermps %zmm1, %zmm3, %zmm1
; AVX512BW-NEXT:    vmovaps %xmm0, (%rsi)
; AVX512BW-NEXT:    vmovaps %xmm2, (%rdx)
; AVX512BW-NEXT:    vmovaps %xmm1, (%rcx)
; AVX512BW-NEXT:    vzeroupper
; AVX512BW-NEXT:    retq
;
; AVX512BW-FCP-LABEL: load_i32_stride3_vf4:
; AVX512BW-FCP:       # %bb.0:
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,3,6,9]
; AVX512BW-FCP-NEXT:    vmovaps (%rdi), %zmm1
; AVX512BW-FCP-NEXT:    vpermps %zmm1, %zmm0, %zmm0
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [1,4,7,10]
; AVX512BW-FCP-NEXT:    vpermps %zmm1, %zmm2, %zmm2
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,5,8,11]
; AVX512BW-FCP-NEXT:    vpermps %zmm1, %zmm3, %zmm1
; AVX512BW-FCP-NEXT:    vmovaps %xmm0, (%rsi)
; AVX512BW-FCP-NEXT:    vmovaps %xmm2, (%rdx)
; AVX512BW-FCP-NEXT:    vmovaps %xmm1, (%rcx)
; AVX512BW-FCP-NEXT:    vzeroupper
; AVX512BW-FCP-NEXT:    retq
;
; AVX512DQ-BW-LABEL: load_i32_stride3_vf4:
; AVX512DQ-BW:       # %bb.0:
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,3,6,9]
; AVX512DQ-BW-NEXT:    vmovaps (%rdi), %zmm1
; AVX512DQ-BW-NEXT:    vpermps %zmm1, %zmm0, %zmm0
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [1,4,7,10]
; AVX512DQ-BW-NEXT:    vpermps %zmm1, %zmm2, %zmm2
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,5,8,11]
; AVX512DQ-BW-NEXT:    vpermps %zmm1, %zmm3, %zmm1
; AVX512DQ-BW-NEXT:    vmovaps %xmm0, (%rsi)
; AVX512DQ-BW-NEXT:    vmovaps %xmm2, (%rdx)
; AVX512DQ-BW-NEXT:    vmovaps %xmm1, (%rcx)
; AVX512DQ-BW-NEXT:    vzeroupper
; AVX512DQ-BW-NEXT:    retq
;
; AVX512DQ-BW-FCP-LABEL: load_i32_stride3_vf4:
; AVX512DQ-BW-FCP:       # %bb.0:
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,3,6,9]
; AVX512DQ-BW-FCP-NEXT:    vmovaps (%rdi), %zmm1
; AVX512DQ-BW-FCP-NEXT:    vpermps %zmm1, %zmm0, %zmm0
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [1,4,7,10]
; AVX512DQ-BW-FCP-NEXT:    vpermps %zmm1, %zmm2, %zmm2
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,5,8,11]
; AVX512DQ-BW-FCP-NEXT:    vpermps %zmm1, %zmm3, %zmm1
; AVX512DQ-BW-FCP-NEXT:    vmovaps %xmm0, (%rsi)
; AVX512DQ-BW-FCP-NEXT:    vmovaps %xmm2, (%rdx)
; AVX512DQ-BW-FCP-NEXT:    vmovaps %xmm1, (%rcx)
; AVX512DQ-BW-FCP-NEXT:    vzeroupper
; AVX512DQ-BW-FCP-NEXT:    retq
  %wide.vec = load <12 x i32>, ptr %in.vec, align 64
  %strided.vec0 = shufflevector <12 x i32> %wide.vec, <12 x i32> poison, <4 x i32> <i32 0, i32 3, i32 6, i32 9>
  %strided.vec1 = shufflevector <12 x i32> %wide.vec, <12 x i32> poison, <4 x i32> <i32 1, i32 4, i32 7, i32 10>
  %strided.vec2 = shufflevector <12 x i32> %wide.vec, <12 x i32> poison, <4 x i32> <i32 2, i32 5, i32 8, i32 11>
  store <4 x i32> %strided.vec0, ptr %out.vec0, align 64
  store <4 x i32> %strided.vec1, ptr %out.vec1, align 64
  store <4 x i32> %strided.vec2, ptr %out.vec2, align 64
  ret void
}

define void @load_i32_stride3_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr %out.vec2) nounwind {
; SSE-LABEL: load_i32_stride3_vf8:
; SSE:       # %bb.0:
; SSE-NEXT:    movaps 80(%rdi), %xmm1
; SSE-NEXT:    movaps 64(%rdi), %xmm5
; SSE-NEXT:    movdqa (%rdi), %xmm0
; SSE-NEXT:    movaps 16(%rdi), %xmm7
; SSE-NEXT:    movaps 32(%rdi), %xmm4
; SSE-NEXT:    movdqa 48(%rdi), %xmm2
; SSE-NEXT:    movdqa %xmm0, %xmm3
; SSE-NEXT:    movaps %xmm7, %xmm8
; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm0[2,3,2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm7[0,0]
; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm7[1,1,1,1]
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[2,0],xmm4[1,0]
; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,3],xmm7[0,2]
; SSE-NEXT:    movdqa %xmm2, %xmm7
; SSE-NEXT:    movaps %xmm5, %xmm10
; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm2[2,3,2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm5[0,0]
; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm5[1,1,1,1]
; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[2,0],xmm1[1,0]
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,3],xmm5[0,2]
; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[3,1],xmm1[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm10[0,2]
; SSE-NEXT:    shufps {{.*#+}} xmm8 = xmm8[3,1],xmm4[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm8[0,2]
; SSE-NEXT:    punpckldq {{.*#+}} xmm6 = xmm6[0],xmm9[0],xmm6[1],xmm9[1]
; SSE-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,1],xmm4[0,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm11 = xmm11[0],xmm12[0],xmm11[1],xmm12[1]
; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[0,1],xmm1[0,3]
; SSE-NEXT:    movaps %xmm7, 16(%rsi)
; SSE-NEXT:    movaps %xmm3, (%rsi)
; SSE-NEXT:    movaps %xmm2, 16(%rdx)
; SSE-NEXT:    movaps %xmm0, (%rdx)
; SSE-NEXT:    movaps %xmm11, 16(%rcx)
; SSE-NEXT:    movaps %xmm6, (%rcx)
; SSE-NEXT:    retq
;
; AVX-LABEL: load_i32_stride3_vf8:
; AVX:       # %bb.0:
; AVX-NEXT:    vmovaps 64(%rdi), %ymm0
; AVX-NEXT:    vmovaps 32(%rdi), %ymm1
; AVX-NEXT:    vmovaps (%rdi), %ymm2
; AVX-NEXT:    vblendps {{.*#+}} ymm3 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6],ymm1[7]
; AVX-NEXT:    vmovaps 16(%rdi), %xmm4
; AVX-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[2,1],ymm1[1,3],ymm4[6,5],ymm1[5,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[0,3],ymm4[0,2],ymm3[4,7],ymm4[4,6]
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm0[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm5 = ymm0[1,0],ymm4[2,0],ymm0[5,4],ymm4[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm5[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm5 = ymm0[2,0],ymm4[3,0],ymm0[6,4],ymm4[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm5 = ymm4[0,0],ymm5[2,0],ymm4[4,4],ymm5[6,4]
; AVX-NEXT:    vmovaps 16(%rdi), %xmm6
; AVX-NEXT:    vblendps {{.*#+}} ymm7 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,2],ymm6[0,3],ymm7[5,6],ymm6[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0,1,2,3,4],ymm5[5,6,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm2 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm2 = ymm6[1,0],ymm2[2,0],ymm6[5,4],ymm2[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[2,0],ymm1[0,3],ymm2[6,4],ymm1[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm4[0,1],ymm0[0,3],ymm4[4,5],ymm0[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
; AVX-NEXT:    vmovaps %ymm3, (%rsi)
; AVX-NEXT:    vmovaps %ymm5, (%rdx)
; AVX-NEXT:    vmovaps %ymm0, (%rcx)
; AVX-NEXT:    vzeroupper
; AVX-NEXT:    retq
;
; AVX2-LABEL: load_i32_stride3_vf8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovaps (%rdi), %ymm0
; AVX2-NEXT:    vmovaps 32(%rdi), %ymm1
; AVX2-NEXT:    vmovaps 64(%rdi), %ymm2
; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm4 = [0,3,6,1,4,7,2,5]
; AVX2-NEXT:    vpermps %ymm3, %ymm4, %ymm3
; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0],ymm4[1,2],ymm2[3],ymm4[4,5],ymm2[6],ymm4[7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm5 = [1,4,7,2,5,0,3,6]
; AVX2-NEXT:    vpermps %ymm4, %ymm5, %ymm4
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
; AVX2-NEXT:    vmovaps %ymm3, (%rsi)
; AVX2-NEXT:    vmovaps %ymm4, (%rdx)
; AVX2-NEXT:    vmovaps %ymm0, (%rcx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
;
; AVX2-FP-LABEL: load_i32_stride3_vf8:
; AVX2-FP:       # %bb.0:
; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm0
; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm1
; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm2
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm4 = [0,3,6,1,4,7,2,5]
; AVX2-FP-NEXT:    vpermps %ymm3, %ymm4, %ymm3
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0],ymm4[1,2],ymm2[3],ymm4[4,5],ymm2[6],ymm4[7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm5 = [1,4,7,2,5,0,3,6]
; AVX2-FP-NEXT:    vpermps %ymm4, %ymm5, %ymm4
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
; AVX2-FP-NEXT:    vmovaps %ymm3, (%rsi)
; AVX2-FP-NEXT:    vmovaps %ymm4, (%rdx)
; AVX2-FP-NEXT:    vmovaps %ymm0, (%rcx)
; AVX2-FP-NEXT:    vzeroupper
; AVX2-FP-NEXT:    retq
;
; AVX2-FCP-LABEL: load_i32_stride3_vf8:
; AVX2-FCP:       # %bb.0:
; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm0
; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm1
; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm2
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm4 = [0,3,6,1,4,7,2,5]
; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm4, %ymm3
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0],ymm4[1,2],ymm2[3],ymm4[4,5],ymm2[6],ymm4[7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm5 = [1,4,7,2,5,0,3,6]
; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm5, %ymm4
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2,3],ymm2[4],ymm0[5,6],ymm2[7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,1,4,7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rsi)
; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rdx)
; AVX2-FCP-NEXT:    vmovaps %ymm0, (%rcx)
; AVX2-FCP-NEXT:    vzeroupper
; AVX2-FCP-NEXT:    retq
;
; AVX512-LABEL: load_i32_stride3_vf8:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512-NEXT:    vpmovsxbd {{.*#+}} ymm2 = [0,3,6,9,12,15,18,21]
; AVX512-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
; AVX512-NEXT:    vpmovsxbd {{.*#+}} ymm3 = [1,4,7,10,13,16,19,22]
; AVX512-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512-NEXT:    vpmovsxbd {{.*#+}} ymm4 = [2,5,8,11,14,17,20,23]
; AVX512-NEXT:    vpermi2d %zmm1, %zmm0, %zmm4
; AVX512-NEXT:    vmovdqa %ymm2, (%rsi)
; AVX512-NEXT:    vmovdqa %ymm3, (%rdx)
; AVX512-NEXT:    vmovdqa %ymm4, (%rcx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
;
; AVX512-FCP-LABEL: load_i32_stride3_vf8:
; AVX512-FCP:       # %bb.0:
; AVX512-FCP-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm2 = [0,3,6,9,12,15,18,21]
; AVX512-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm3 = [1,4,7,10,13,16,19,22]
; AVX512-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm4 = [2,5,8,11,14,17,20,23]
; AVX512-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm4
; AVX512-FCP-NEXT:    vmovdqa %ymm2, (%rsi)
; AVX512-FCP-NEXT:    vmovdqa %ymm3, (%rdx)
; AVX512-FCP-NEXT:    vmovdqa %ymm4, (%rcx)
; AVX512-FCP-NEXT:    vzeroupper
; AVX512-FCP-NEXT:    retq
;
; AVX512DQ-LABEL: load_i32_stride3_vf8:
; AVX512DQ:       # %bb.0:
; AVX512DQ-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512DQ-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} ymm2 = [0,3,6,9,12,15,18,21]
; AVX512DQ-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} ymm3 = [1,4,7,10,13,16,19,22]
; AVX512DQ-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} ymm4 = [2,5,8,11,14,17,20,23]
; AVX512DQ-NEXT:    vpermi2d %zmm1, %zmm0, %zmm4
; AVX512DQ-NEXT:    vmovdqa %ymm2, (%rsi)
; AVX512DQ-NEXT:    vmovdqa %ymm3, (%rdx)
; AVX512DQ-NEXT:    vmovdqa %ymm4, (%rcx)
; AVX512DQ-NEXT:    vzeroupper
; AVX512DQ-NEXT:    retq
;
; AVX512DQ-FCP-LABEL: load_i32_stride3_vf8:
; AVX512DQ-FCP:       # %bb.0:
; AVX512DQ-FCP-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512DQ-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm2 = [0,3,6,9,12,15,18,21]
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm3 = [1,4,7,10,13,16,19,22]
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm4 = [2,5,8,11,14,17,20,23]
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm4
; AVX512DQ-FCP-NEXT:    vmovdqa %ymm2, (%rsi)
; AVX512DQ-FCP-NEXT:    vmovdqa %ymm3, (%rdx)
; AVX512DQ-FCP-NEXT:    vmovdqa %ymm4, (%rcx)
; AVX512DQ-FCP-NEXT:    vzeroupper
; AVX512DQ-FCP-NEXT:    retq
;
; AVX512BW-LABEL: load_i32_stride3_vf8:
; AVX512BW:       # %bb.0:
; AVX512BW-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512BW-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} ymm2 = [0,3,6,9,12,15,18,21]
; AVX512BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} ymm3 = [1,4,7,10,13,16,19,22]
; AVX512BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} ymm4 = [2,5,8,11,14,17,20,23]
; AVX512BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm4
; AVX512BW-NEXT:    vmovdqa %ymm2, (%rsi)
; AVX512BW-NEXT:    vmovdqa %ymm3, (%rdx)
; AVX512BW-NEXT:    vmovdqa %ymm4, (%rcx)
; AVX512BW-NEXT:    vzeroupper
; AVX512BW-NEXT:    retq
;
; AVX512BW-FCP-LABEL: load_i32_stride3_vf8:
; AVX512BW-FCP:       # %bb.0:
; AVX512BW-FCP-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512BW-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm2 = [0,3,6,9,12,15,18,21]
; AVX512BW-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm3 = [1,4,7,10,13,16,19,22]
; AVX512BW-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm4 = [2,5,8,11,14,17,20,23]
; AVX512BW-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm4
; AVX512BW-FCP-NEXT:    vmovdqa %ymm2, (%rsi)
; AVX512BW-FCP-NEXT:    vmovdqa %ymm3, (%rdx)
; AVX512BW-FCP-NEXT:    vmovdqa %ymm4, (%rcx)
; AVX512BW-FCP-NEXT:    vzeroupper
; AVX512BW-FCP-NEXT:    retq
;
; AVX512DQ-BW-LABEL: load_i32_stride3_vf8:
; AVX512DQ-BW:       # %bb.0:
; AVX512DQ-BW-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512DQ-BW-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} ymm2 = [0,3,6,9,12,15,18,21]
; AVX512DQ-BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} ymm3 = [1,4,7,10,13,16,19,22]
; AVX512DQ-BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} ymm4 = [2,5,8,11,14,17,20,23]
; AVX512DQ-BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm4
; AVX512DQ-BW-NEXT:    vmovdqa %ymm2, (%rsi)
; AVX512DQ-BW-NEXT:    vmovdqa %ymm3, (%rdx)
; AVX512DQ-BW-NEXT:    vmovdqa %ymm4, (%rcx)
; AVX512DQ-BW-NEXT:    vzeroupper
; AVX512DQ-BW-NEXT:    retq
;
; AVX512DQ-BW-FCP-LABEL: load_i32_stride3_vf8:
; AVX512DQ-BW-FCP:       # %bb.0:
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm2 = [0,3,6,9,12,15,18,21]
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm3 = [1,4,7,10,13,16,19,22]
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm4 = [2,5,8,11,14,17,20,23]
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm4
; AVX512DQ-BW-FCP-NEXT:    vmovdqa %ymm2, (%rsi)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa %ymm3, (%rdx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa %ymm4, (%rcx)
; AVX512DQ-BW-FCP-NEXT:    vzeroupper
; AVX512DQ-BW-FCP-NEXT:    retq
  %wide.vec = load <24 x i32>, ptr %in.vec, align 64
  %strided.vec0 = shufflevector <24 x i32> %wide.vec, <24 x i32> poison, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>
  %strided.vec1 = shufflevector <24 x i32> %wide.vec, <24 x i32> poison, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>
  %strided.vec2 = shufflevector <24 x i32> %wide.vec, <24 x i32> poison, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>
  store <8 x i32> %strided.vec0, ptr %out.vec0, align 64
  store <8 x i32> %strided.vec1, ptr %out.vec1, align 64
  store <8 x i32> %strided.vec2, ptr %out.vec2, align 64
  ret void
}

define void @load_i32_stride3_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr %out.vec2) nounwind {
; SSE-LABEL: load_i32_stride3_vf16:
; SSE:       # %bb.0:
; SSE-NEXT:    movaps 96(%rdi), %xmm6
; SSE-NEXT:    movaps 128(%rdi), %xmm12
; SSE-NEXT:    movaps 112(%rdi), %xmm13
; SSE-NEXT:    movaps 144(%rdi), %xmm11
; SSE-NEXT:    movaps 176(%rdi), %xmm10
; SSE-NEXT:    movaps 160(%rdi), %xmm9
; SSE-NEXT:    movaps (%rdi), %xmm7
; SSE-NEXT:    movaps 16(%rdi), %xmm8
; SSE-NEXT:    movaps 32(%rdi), %xmm3
; SSE-NEXT:    movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 48(%rdi), %xmm15
; SSE-NEXT:    movaps 80(%rdi), %xmm14
; SSE-NEXT:    movaps 64(%rdi), %xmm2
; SSE-NEXT:    movaps %xmm2, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm14[1,0]
; SSE-NEXT:    movaps %xmm15, %xmm5
; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm8, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm3[1,0]
; SSE-NEXT:    movaps %xmm7, %xmm5
; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm9, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm10[1,0]
; SSE-NEXT:    movaps %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm11, %xmm3
; SSE-NEXT:    movaps %xmm11, %xmm4
; SSE-NEXT:    movaps %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm13, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm12[1,0]
; SSE-NEXT:    movaps %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm6, %xmm5
; SSE-NEXT:    movaps %xmm6, %xmm3
; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm15, %xmm11
; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[1,0],xmm2[0,0]
; SSE-NEXT:    movaps %xmm2, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,1],xmm14[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[0,2],xmm0[0,2]
; SSE-NEXT:    movaps %xmm4, %xmm6
; SSE-NEXT:    shufps {{.*#+}} xmm6 = xmm6[1,0],xmm9[0,0]
; SSE-NEXT:    movaps %xmm9, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,1],xmm10[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,2],xmm0[0,2]
; SSE-NEXT:    movaps %xmm3, %xmm1
; SSE-NEXT:    movaps %xmm3, %xmm10
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm13[0,0]
; SSE-NEXT:    movaps %xmm13, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,1],xmm12[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[0,2]
; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm7[2,3,2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[1,0],xmm8[0,0]
; SSE-NEXT:    movaps %xmm8, %xmm12
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm12 = xmm12[3,1],xmm4[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,2],xmm12[0,2]
; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm2[1,1,1,1]
; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm15[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm12[0],xmm2[1],xmm12[1]
; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,1],xmm14[0,3]
; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm8[1,1,1,1]
; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm4[0,3]
; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm9[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Folded Reload
; SSE-NEXT:    # xmm8 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm8 = xmm8[0],xmm3[0],xmm8[1],xmm3[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Folded Reload
; SSE-NEXT:    # xmm8 = xmm8[0,1],mem[0,3]
; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm13[1,1,1,1]
; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm10[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
; SSE-NEXT:    # xmm4 = xmm4[0,1],mem[0,3]
; SSE-NEXT:    movaps %xmm5, 32(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
; SSE-NEXT:    movaps %xmm3, 48(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
; SSE-NEXT:    movaps %xmm3, (%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
; SSE-NEXT:    movaps %xmm3, 16(%rsi)
; SSE-NEXT:    movaps %xmm1, 32(%rdx)
; SSE-NEXT:    movaps %xmm6, 48(%rdx)
; SSE-NEXT:    movaps %xmm7, (%rdx)
; SSE-NEXT:    movaps %xmm11, 16(%rdx)
; SSE-NEXT:    movaps %xmm4, 32(%rcx)
; SSE-NEXT:    movaps %xmm8, 48(%rcx)
; SSE-NEXT:    movaps %xmm0, (%rcx)
; SSE-NEXT:    movaps %xmm2, 16(%rcx)
; SSE-NEXT:    retq
;
; AVX-LABEL: load_i32_stride3_vf16:
; AVX:       # %bb.0:
; AVX-NEXT:    vmovaps 160(%rdi), %ymm0
; AVX-NEXT:    vmovaps 128(%rdi), %ymm1
; AVX-NEXT:    vmovaps 96(%rdi), %ymm2
; AVX-NEXT:    vmovaps 64(%rdi), %ymm3
; AVX-NEXT:    vmovaps 32(%rdi), %ymm4
; AVX-NEXT:    vmovaps (%rdi), %ymm6
; AVX-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0],ymm4[1],ymm6[2,3],ymm4[4],ymm6[5,6],ymm4[7]
; AVX-NEXT:    vmovaps 16(%rdi), %xmm7
; AVX-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[2,1],ymm4[1,3],ymm7[6,5],ymm4[5,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[0,3],ymm7[0,2],ymm5[4,7],ymm7[4,6]
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm7 = ymm3[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm8 = ymm3[1,0],ymm7[2,0],ymm3[5,4],ymm7[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm8 = ymm8[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm8[6,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6],ymm1[7]
; AVX-NEXT:    vmovaps 112(%rdi), %xmm9
; AVX-NEXT:    vshufps {{.*#+}} ymm9 = ymm9[2,1],ymm1[1,3],ymm9[6,5],ymm1[5,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm8 = ymm8[0,3],ymm9[0,2],ymm8[4,7],ymm9[4,6]
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm9 = ymm0[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm10 = ymm0[1,0],ymm9[2,0],ymm0[5,4],ymm9[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm10[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm10 = ymm3[2,0],ymm7[3,0],ymm3[6,4],ymm7[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm10 = ymm7[0,0],ymm10[2,0],ymm7[4,4],ymm10[6,4]
; AVX-NEXT:    vmovaps 16(%rdi), %xmm11
; AVX-NEXT:    vblendps {{.*#+}} ymm12 = ymm6[0,1],ymm4[2],ymm6[3,4],ymm4[5],ymm6[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm12 = ymm12[1,2],ymm11[0,3],ymm12[5,6],ymm11[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm12 = ymm12[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm10 = ymm12[0,1,2,3,4],ymm10[5,6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm12 = ymm0[2,0],ymm9[3,0],ymm0[6,4],ymm9[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm12 = ymm9[0,0],ymm12[2,0],ymm9[4,4],ymm12[6,4]
; AVX-NEXT:    vmovaps 112(%rdi), %xmm13
; AVX-NEXT:    vblendps {{.*#+}} ymm14 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[1,2],ymm13[0,3],ymm14[5,6],ymm13[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm12 = ymm14[0,1,2,3,4],ymm12[5,6,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm6 = ymm4[0,1],ymm6[2],ymm4[3,4],ymm6[5],ymm4[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm11[1,0],ymm6[2,0],ymm11[5,4],ymm6[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm4 = ymm6[2,0],ymm4[0,3],ymm6[6,4],ymm4[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm3 = ymm7[0,1],ymm3[0,3],ymm7[4,5],ymm3[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm3[5,6,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm2 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm2 = ymm13[1,0],ymm2[2,0],ymm13[5,4],ymm2[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[2,0],ymm1[0,3],ymm2[6,4],ymm1[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm9[0,1],ymm0[0,3],ymm9[4,5],ymm0[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
; AVX-NEXT:    vmovaps %ymm8, 32(%rsi)
; AVX-NEXT:    vmovaps %ymm5, (%rsi)
; AVX-NEXT:    vmovaps %ymm12, 32(%rdx)
; AVX-NEXT:    vmovaps %ymm10, (%rdx)
; AVX-NEXT:    vmovaps %ymm0, 32(%rcx)
; AVX-NEXT:    vmovaps %ymm3, (%rcx)
; AVX-NEXT:    vzeroupper
; AVX-NEXT:    retq
;
; AVX2-LABEL: load_i32_stride3_vf16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovaps 160(%rdi), %ymm0
; AVX2-NEXT:    vmovaps 128(%rdi), %ymm1
; AVX2-NEXT:    vmovaps (%rdi), %ymm2
; AVX2-NEXT:    vmovaps 32(%rdi), %ymm3
; AVX2-NEXT:    vmovaps 64(%rdi), %ymm4
; AVX2-NEXT:    vmovaps 96(%rdi), %ymm5
; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6],ymm3[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm4[2],ymm6[3,4],ymm4[5],ymm6[6,7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm7 = [0,3,6,1,4,7,2,5]
; AVX2-NEXT:    vpermps %ymm6, %ymm7, %ymm6
; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0],ymm1[1],ymm5[2,3],ymm1[4],ymm5[5,6],ymm1[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7]
; AVX2-NEXT:    vpermps %ymm8, %ymm7, %ymm7
; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0],ymm8[1,2],ymm4[3],ymm8[4,5],ymm4[6],ymm8[7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm9 = [1,4,7,2,5,0,3,6]
; AVX2-NEXT:    vpermps %ymm8, %ymm9, %ymm8
; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm5[0,1],ymm1[2],ymm5[3,4],ymm1[5],ymm5[6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0],ymm10[1,2],ymm0[3],ymm10[4,5],ymm0[6],ymm10[7]
; AVX2-NEXT:    vpermps %ymm10, %ymm9, %ymm9
; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm3 = [2,5,0,3,6,u,u,u]
; AVX2-NEXT:    vpermps %ymm2, %ymm3, %ymm2
; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm5[2],ymm1[3,4],ymm5[5],ymm1[6,7]
; AVX2-NEXT:    vpermps %ymm1, %ymm3, %ymm1
; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
; AVX2-NEXT:    vmovaps %ymm7, 32(%rsi)
; AVX2-NEXT:    vmovaps %ymm6, (%rsi)
; AVX2-NEXT:    vmovaps %ymm9, 32(%rdx)
; AVX2-NEXT:    vmovaps %ymm8, (%rdx)
; AVX2-NEXT:    vmovaps %ymm0, 32(%rcx)
; AVX2-NEXT:    vmovaps %ymm2, (%rcx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
;
; AVX2-FP-LABEL: load_i32_stride3_vf16:
; AVX2-FP:       # %bb.0:
; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm0
; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm1
; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm2
; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm3
; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm4
; AVX2-FP-NEXT:    vmovaps 96(%rdi), %ymm5
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6],ymm3[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm4[2],ymm6[3,4],ymm4[5],ymm6[6,7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm7 = [0,3,6,1,4,7,2,5]
; AVX2-FP-NEXT:    vpermps %ymm6, %ymm7, %ymm6
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0],ymm1[1],ymm5[2,3],ymm1[4],ymm5[5,6],ymm1[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7]
; AVX2-FP-NEXT:    vpermps %ymm8, %ymm7, %ymm7
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0],ymm8[1,2],ymm4[3],ymm8[4,5],ymm4[6],ymm8[7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm9 = [1,4,7,2,5,0,3,6]
; AVX2-FP-NEXT:    vpermps %ymm8, %ymm9, %ymm8
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm5[0,1],ymm1[2],ymm5[3,4],ymm1[5],ymm5[6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0],ymm10[1,2],ymm0[3],ymm10[4,5],ymm0[6],ymm10[7]
; AVX2-FP-NEXT:    vpermps %ymm10, %ymm9, %ymm9
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm3 = [2,5,0,3,6,u,u,u]
; AVX2-FP-NEXT:    vpermps %ymm2, %ymm3, %ymm2
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm5[2],ymm1[3,4],ymm5[5],ymm1[6,7]
; AVX2-FP-NEXT:    vpermps %ymm1, %ymm3, %ymm1
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%rsi)
; AVX2-FP-NEXT:    vmovaps %ymm6, (%rsi)
; AVX2-FP-NEXT:    vmovaps %ymm9, 32(%rdx)
; AVX2-FP-NEXT:    vmovaps %ymm8, (%rdx)
; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm2, (%rcx)
; AVX2-FP-NEXT:    vzeroupper
; AVX2-FP-NEXT:    retq
;
; AVX2-FCP-LABEL: load_i32_stride3_vf16:
; AVX2-FCP:       # %bb.0:
; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm0
; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm1
; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm2
; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm3
; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm4
; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %ymm5
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6],ymm3[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm4[2],ymm6[3,4],ymm4[5],ymm6[6,7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm7 = [0,3,6,1,4,7,2,5]
; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm7, %ymm6
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0],ymm1[1],ymm5[2,3],ymm1[4],ymm5[5,6],ymm1[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7]
; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm7, %ymm7
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0],ymm8[1,2],ymm4[3],ymm8[4,5],ymm4[6],ymm8[7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm9 = [1,4,7,2,5,0,3,6]
; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm9, %ymm8
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm5[0,1],ymm1[2],ymm5[3,4],ymm1[5],ymm5[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0],ymm10[1,2],ymm0[3],ymm10[4,5],ymm0[6],ymm10[7]
; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm9, %ymm9
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm4[1],ymm2[2,3],ymm4[4],ymm2[5,6],ymm4[7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm3 = [2,5,0,3,6,1,4,7]
; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm3, %ymm2
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm5[2],ymm1[3,4],ymm5[5],ymm1[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6],ymm0[7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm3, %ymm0
; AVX2-FCP-NEXT:    vmovaps %ymm7, 32(%rsi)
; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rsi)
; AVX2-FCP-NEXT:    vmovaps %ymm9, 32(%rdx)
; AVX2-FCP-NEXT:    vmovaps %ymm8, (%rdx)
; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm2, (%rcx)
; AVX2-FCP-NEXT:    vzeroupper
; AVX2-FCP-NEXT:    retq
;
; AVX512-LABEL: load_i32_stride3_vf16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512-NEXT:    vmovdqa64 128(%rdi), %zmm2
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512-NEXT:    vpermi2d %zmm2, %zmm3, %zmm4
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512-NEXT:    vpermi2d %zmm0, %zmm1, %zmm3
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm5 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512-NEXT:    vpermi2d %zmm2, %zmm3, %zmm5
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm0 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512-NEXT:    vpermi2d %zmm2, %zmm3, %zmm0
; AVX512-NEXT:    vmovdqa64 %zmm4, (%rsi)
; AVX512-NEXT:    vmovdqa64 %zmm5, (%rdx)
; AVX512-NEXT:    vmovdqa64 %zmm0, (%rcx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
;
; AVX512-FCP-LABEL: load_i32_stride3_vf16:
; AVX512-FCP:       # %bb.0:
; AVX512-FCP-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm2
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm4
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512-FCP-NEXT:    vpermi2d %zmm0, %zmm1, %zmm3
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm5 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm5
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm0 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm0
; AVX512-FCP-NEXT:    vmovdqa64 %zmm4, (%rsi)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm5, (%rdx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm0, (%rcx)
; AVX512-FCP-NEXT:    vzeroupper
; AVX512-FCP-NEXT:    retq
;
; AVX512DQ-LABEL: load_i32_stride3_vf16:
; AVX512DQ:       # %bb.0:
; AVX512DQ-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512DQ-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512DQ-NEXT:    vmovdqa64 128(%rdi), %zmm2
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-NEXT:    vpermi2d %zmm2, %zmm3, %zmm4
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-NEXT:    vpermi2d %zmm0, %zmm1, %zmm3
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm5 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-NEXT:    vpermi2d %zmm2, %zmm3, %zmm5
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm0 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-NEXT:    vpermi2d %zmm2, %zmm3, %zmm0
; AVX512DQ-NEXT:    vmovdqa64 %zmm4, (%rsi)
; AVX512DQ-NEXT:    vmovdqa64 %zmm5, (%rdx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm0, (%rcx)
; AVX512DQ-NEXT:    vzeroupper
; AVX512DQ-NEXT:    retq
;
; AVX512DQ-FCP-LABEL: load_i32_stride3_vf16:
; AVX512DQ-FCP:       # %bb.0:
; AVX512DQ-FCP-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512DQ-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512DQ-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm2
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm4
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm0, %zmm1, %zmm3
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm5 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm5
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm0 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm0
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm4, (%rsi)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm5, (%rdx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm0, (%rcx)
; AVX512DQ-FCP-NEXT:    vzeroupper
; AVX512DQ-FCP-NEXT:    retq
;
; AVX512BW-LABEL: load_i32_stride3_vf16:
; AVX512BW:       # %bb.0:
; AVX512BW-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512BW-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512BW-NEXT:    vmovdqa64 128(%rdi), %zmm2
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512BW-NEXT:    vpermi2d %zmm2, %zmm3, %zmm4
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512BW-NEXT:    vpermi2d %zmm0, %zmm1, %zmm3
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm5 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512BW-NEXT:    vpermi2d %zmm2, %zmm3, %zmm5
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm0 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512BW-NEXT:    vpermi2d %zmm2, %zmm3, %zmm0
; AVX512BW-NEXT:    vmovdqa64 %zmm4, (%rsi)
; AVX512BW-NEXT:    vmovdqa64 %zmm5, (%rdx)
; AVX512BW-NEXT:    vmovdqa64 %zmm0, (%rcx)
; AVX512BW-NEXT:    vzeroupper
; AVX512BW-NEXT:    retq
;
; AVX512BW-FCP-LABEL: load_i32_stride3_vf16:
; AVX512BW-FCP:       # %bb.0:
; AVX512BW-FCP-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512BW-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512BW-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm2
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512BW-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512BW-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm4
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512BW-FCP-NEXT:    vpermi2d %zmm0, %zmm1, %zmm3
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm5 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512BW-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm5
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512BW-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm0 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512BW-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm0
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm4, (%rsi)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm5, (%rdx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm0, (%rcx)
; AVX512BW-FCP-NEXT:    vzeroupper
; AVX512BW-FCP-NEXT:    retq
;
; AVX512DQ-BW-LABEL: load_i32_stride3_vf16:
; AVX512DQ-BW:       # %bb.0:
; AVX512DQ-BW-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512DQ-BW-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512DQ-BW-NEXT:    vmovdqa64 128(%rdi), %zmm2
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-BW-NEXT:    vpermi2d %zmm2, %zmm3, %zmm4
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-BW-NEXT:    vpermi2d %zmm0, %zmm1, %zmm3
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm5 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-BW-NEXT:    vpermi2d %zmm2, %zmm3, %zmm5
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm0 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-BW-NEXT:    vpermi2d %zmm2, %zmm3, %zmm0
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm4, (%rsi)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm5, (%rdx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm0, (%rcx)
; AVX512DQ-BW-NEXT:    vzeroupper
; AVX512DQ-BW-NEXT:    retq
;
; AVX512DQ-BW-FCP-LABEL: load_i32_stride3_vf16:
; AVX512DQ-BW-FCP:       # %bb.0:
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 (%rdi), %zmm0
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm1
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm2
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm4
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm0, %zmm1, %zmm3
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm5 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm5
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm3 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm1, %zmm0, %zmm3
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm0 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm0
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm4, (%rsi)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm5, (%rdx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm0, (%rcx)
; AVX512DQ-BW-FCP-NEXT:    vzeroupper
; AVX512DQ-BW-FCP-NEXT:    retq
  %wide.vec = load <48 x i32>, ptr %in.vec, align 64
  %strided.vec0 = shufflevector <48 x i32> %wide.vec, <48 x i32> poison, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>
  %strided.vec1 = shufflevector <48 x i32> %wide.vec, <48 x i32> poison, <16 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46>
  %strided.vec2 = shufflevector <48 x i32> %wide.vec, <48 x i32> poison, <16 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47>
  store <16 x i32> %strided.vec0, ptr %out.vec0, align 64
  store <16 x i32> %strided.vec1, ptr %out.vec1, align 64
  store <16 x i32> %strided.vec2, ptr %out.vec2, align 64
  ret void
}

define void @load_i32_stride3_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr %out.vec2) nounwind {
; SSE-LABEL: load_i32_stride3_vf32:
; SSE:       # %bb.0:
; SSE-NEXT:    subq $392, %rsp # imm = 0x188
; SSE-NEXT:    movaps 192(%rdi), %xmm4
; SSE-NEXT:    movaps 224(%rdi), %xmm3
; SSE-NEXT:    movaps 208(%rdi), %xmm14
; SSE-NEXT:    movaps %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 240(%rdi), %xmm7
; SSE-NEXT:    movaps 272(%rdi), %xmm10
; SSE-NEXT:    movaps 256(%rdi), %xmm9
; SSE-NEXT:    movaps (%rdi), %xmm13
; SSE-NEXT:    movaps %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 16(%rdi), %xmm8
; SSE-NEXT:    movaps %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 32(%rdi), %xmm11
; SSE-NEXT:    movaps %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 48(%rdi), %xmm2
; SSE-NEXT:    movaps 80(%rdi), %xmm12
; SSE-NEXT:    movaps 64(%rdi), %xmm5
; SSE-NEXT:    movaps %xmm5, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm12[1,0]
; SSE-NEXT:    movaps %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm2, %xmm1
; SSE-NEXT:    movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm9, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm10[1,0]
; SSE-NEXT:    movaps %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm7, %xmm1
; SSE-NEXT:    movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm8, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm11[1,0]
; SSE-NEXT:    movaps %xmm13, %xmm1
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm14, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm3[1,0]
; SSE-NEXT:    movaps %xmm3, %xmm13
; SSE-NEXT:    movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm4, %xmm1
; SSE-NEXT:    movaps %xmm4, %xmm11
; SSE-NEXT:    movaps %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 176(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 160(%rdi), %xmm0
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps 144(%rdi), %xmm3
; SSE-NEXT:    movaps %xmm3, %xmm1
; SSE-NEXT:    movaps %xmm3, (%rsp) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 368(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 352(%rdi), %xmm0
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps 336(%rdi), %xmm4
; SSE-NEXT:    movaps %xmm4, %xmm1
; SSE-NEXT:    movaps %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 128(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 112(%rdi), %xmm15
; SSE-NEXT:    movaps %xmm15, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps 96(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 320(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 304(%rdi), %xmm6
; SSE-NEXT:    movaps %xmm6, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps 288(%rdi), %xmm8
; SSE-NEXT:    movaps %xmm8, %xmm1
; SSE-NEXT:    movaps %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm5[0,0]
; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[3,1],xmm12[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm5[0,2]
; SSE-NEXT:    movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm7, %xmm14
; SSE-NEXT:    movaps %xmm9, %xmm0
; SSE-NEXT:    movaps %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm14 = xmm14[1,0],xmm9[0,0]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,1],xmm10[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm14 = xmm14[0,2],xmm0[0,2]
; SSE-NEXT:    movaps %xmm11, %xmm10
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[1,0],xmm0[0,0]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,1],xmm13[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[0,2],xmm0[0,2]
; SSE-NEXT:    movaps %xmm3, %xmm9
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm9 = xmm9[1,0],xmm0[0,0]
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = xmm0[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm9 = xmm9[0,2],xmm0[0,2]
; SSE-NEXT:    movaps %xmm4, %xmm7
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[1,0],xmm0[0,0]
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = xmm0[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,2],xmm0[0,2]
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, %xmm11
; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[1,0],xmm15[0,0]
; SSE-NEXT:    movaps %xmm15, %xmm0
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,1],xmm12[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[0,2],xmm0[0,2]
; SSE-NEXT:    movaps %xmm8, %xmm3
; SSE-NEXT:    movaps %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,0],xmm6[0,0]
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
; SSE-NEXT:    # xmm6 = xmm6[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2],xmm6[0,2]
; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm2[2,3,2,3]
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm0[0,0]
; SSE-NEXT:    movaps %xmm0, %xmm13
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm13 = xmm13[3,1],xmm1[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm13[0,2]
; SSE-NEXT:    movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Folded Reload
; SSE-NEXT:    # xmm13 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Folded Reload
; SSE-NEXT:    # xmm8 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm8 = xmm8[0],xmm13[0],xmm8[1],xmm13[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Folded Reload
; SSE-NEXT:    # xmm8 = xmm8[0,1],mem[0,3]
; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm0[1,1,1,1]
; SSE-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm13[0],xmm5[1],xmm13[1]
; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,1],xmm1[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Folded Reload
; SSE-NEXT:    # xmm13 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, (%rsp), %xmm6 # 16-byte Folded Reload
; SSE-NEXT:    # xmm6 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm6 = xmm6[0],xmm13[0],xmm6[1],xmm13[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
; SSE-NEXT:    # xmm6 = xmm6[0,1],mem[0,3]
; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm15[1,1,1,1]
; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm13[0],xmm4[1],xmm13[1]
; SSE-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,1],xmm12[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
; SSE-NEXT:    # xmm15 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Folded Reload
; SSE-NEXT:    # xmm13 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm13 = xmm13[0],xmm15[0],xmm13[1],xmm15[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Folded Reload
; SSE-NEXT:    # xmm13 = xmm13[0,1],mem[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
; SSE-NEXT:    # xmm15 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm15 = xmm15[0],xmm0[0],xmm15[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
; SSE-NEXT:    # xmm15 = xmm15[0,1],mem[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = xmm1[0,1],mem[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
; SSE-NEXT:    # xmm2 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = xmm0[0,1],mem[0,3]
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    movaps %xmm2, 96(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    movaps %xmm2, 32(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    movaps %xmm2, 112(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    movaps %xmm2, 48(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    movaps %xmm2, 64(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    movaps %xmm2, (%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    movaps %xmm2, 80(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    movaps %xmm2, 16(%rsi)
; SSE-NEXT:    movaps %xmm3, 96(%rdx)
; SSE-NEXT:    movaps %xmm11, 32(%rdx)
; SSE-NEXT:    movaps %xmm7, 112(%rdx)
; SSE-NEXT:    movaps %xmm9, 48(%rdx)
; SSE-NEXT:    movaps %xmm10, 64(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    movaps %xmm2, (%rdx)
; SSE-NEXT:    movaps %xmm14, 80(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    movaps %xmm2, 16(%rdx)
; SSE-NEXT:    movaps %xmm0, 96(%rcx)
; SSE-NEXT:    movaps %xmm1, 112(%rcx)
; SSE-NEXT:    movaps %xmm15, 64(%rcx)
; SSE-NEXT:    movaps %xmm13, 80(%rcx)
; SSE-NEXT:    movaps %xmm4, 32(%rcx)
; SSE-NEXT:    movaps %xmm6, 48(%rcx)
; SSE-NEXT:    movaps %xmm5, (%rcx)
; SSE-NEXT:    movaps %xmm8, 16(%rcx)
; SSE-NEXT:    addq $392, %rsp # imm = 0x188
; SSE-NEXT:    retq
;
; AVX-LABEL: load_i32_stride3_vf32:
; AVX:       # %bb.0:
; AVX-NEXT:    subq $392, %rsp # imm = 0x188
; AVX-NEXT:    vmovaps 256(%rdi), %ymm2
; AVX-NEXT:    vmovaps 224(%rdi), %ymm7
; AVX-NEXT:    vmovaps 192(%rdi), %ymm3
; AVX-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 352(%rdi), %ymm4
; AVX-NEXT:    vmovaps 320(%rdi), %ymm5
; AVX-NEXT:    vmovaps 288(%rdi), %ymm6
; AVX-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 160(%rdi), %ymm10
; AVX-NEXT:    vmovaps 128(%rdi), %ymm9
; AVX-NEXT:    vmovaps 96(%rdi), %ymm0
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm9[1],ymm0[2,3],ymm9[4],ymm0[5,6],ymm9[7]
; AVX-NEXT:    vmovaps 112(%rdi), %xmm1
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,1],ymm9[1,3],ymm1[6,5],ymm9[5,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,3],ymm1[0,2],ymm0[4,7],ymm1[4,6]
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm11 = ymm10[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[1,0],ymm11[2,0],ymm10[5,4],ymm11[6,4]
; AVX-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0],ymm5[1],ymm6[2,3],ymm5[4],ymm6[5,6],ymm5[7]
; AVX-NEXT:    vmovaps 304(%rdi), %xmm1
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,1],ymm5[1,3],ymm1[6,5],ymm5[5,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,3],ymm1[0,2],ymm0[4,7],ymm1[4,6]
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm14 = ymm4[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm4[1,0],ymm14[2,0],ymm4[5,4],ymm14[6,4]
; AVX-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0],ymm7[1],ymm3[2,3],ymm7[4],ymm3[5,6],ymm7[7]
; AVX-NEXT:    vmovaps 208(%rdi), %xmm1
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,1],ymm7[1,3],ymm1[6,5],ymm7[5,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,3],ymm1[0,2],ymm0[4,7],ymm1[4,6]
; AVX-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm12 = ymm2[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[1,0],ymm12[2,0],ymm2[5,4],ymm12[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 32(%rdi), %ymm15
; AVX-NEXT:    vmovaps 16(%rdi), %xmm0
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,1],ymm15[1,3],ymm0[6,5],ymm15[5,7]
; AVX-NEXT:    vmovaps (%rdi), %ymm2
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0],ymm15[1],ymm2[2,3],ymm15[4],ymm2[5,6],ymm15[7]
; AVX-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm13 = ymm1[0,3],ymm0[0,2],ymm1[4,7],ymm0[4,6]
; AVX-NEXT:    vmovaps 64(%rdi), %ymm7
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm7[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm8 = ymm7[1,0],ymm0[2,0],ymm7[5,4],ymm0[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm8 = ymm8[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm8[6,7]
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups %ymm11, (%rsp) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm8 = ymm10[2,0],ymm11[3,0],ymm10[6,4],ymm11[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm8 = ymm11[0,0],ymm8[2,0],ymm11[4,4],ymm8[6,4]
; AVX-NEXT:    vmovaps 112(%rdi), %xmm13
; AVX-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm6 # 32-byte Folded Reload
; AVX-NEXT:    # ymm6 = mem[0,1],ymm9[2],mem[3,4],ymm9[5],mem[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,2],ymm13[0,3],ymm6[5,6],ymm13[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1,2,3,4],ymm8[5,6,7]
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm4[2,0],ymm14[3,0],ymm4[6,4],ymm14[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm14[0,0],ymm6[2,0],ymm14[4,4],ymm6[6,4]
; AVX-NEXT:    vmovaps 304(%rdi), %xmm8
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX-NEXT:    vblendps {{.*#+}} ymm10 = ymm3[0,1],ymm5[2],ymm3[3,4],ymm5[5],ymm3[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[1,2],ymm8[0,3],ymm10[5,6],ymm8[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm10[0,1,2,3,4],ymm6[5,6,7]
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm7[2,0],ymm0[3,0],ymm7[6,4],ymm0[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm0[0,0],ymm6[2,0],ymm0[4,4],ymm6[6,4]
; AVX-NEXT:    vmovaps 16(%rdi), %xmm11
; AVX-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0,1],ymm15[2],ymm2[3,4],ymm15[5],ymm2[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[1,2],ymm11[0,3],ymm4[5,6],ymm11[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3,4],ymm6[5,6,7]
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm4 = ymm1[2,0],ymm12[3,0],ymm1[6,4],ymm12[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm4 = ymm12[0,0],ymm4[2,0],ymm12[4,4],ymm4[6,4]
; AVX-NEXT:    vmovaps 208(%rdi), %xmm10
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX-NEXT:    vblendps {{.*#+}} ymm6 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,2],ymm10[0,3],ymm6[5,6],ymm10[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3,4],ymm4[5,6,7]
; AVX-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm6 # 32-byte Folded Reload
; AVX-NEXT:    # ymm6 = ymm9[0,1],mem[2],ymm9[3,4],mem[5],ymm9[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm13[1,0],ymm6[2,0],ymm13[5,4],ymm6[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[2,0],ymm9[0,3],ymm6[6,4],ymm9[4,7]
; AVX-NEXT:    vmovups (%rsp), %ymm9 # 32-byte Reload
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
; AVX-NEXT:    # ymm9 = ymm9[0,1],mem[0,3],ymm9[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm9[5,6,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm9 = ymm5[0,1],ymm3[2],ymm5[3,4],ymm3[5],ymm5[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm8 = ymm8[1,0],ymm9[2,0],ymm8[5,4],ymm9[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm5 = ymm8[2,0],ymm5[0,3],ymm8[6,4],ymm5[4,7]
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm8 # 32-byte Folded Reload
; AVX-NEXT:    # ymm8 = ymm14[0,1],mem[0,3],ymm14[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm8[5,6,7]
; AVX-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm3 # 32-byte Folded Reload
; AVX-NEXT:    # ymm3 = ymm15[0,1],mem[2],ymm15[3,4],mem[5],ymm15[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm3 = ymm11[1,0],ymm3[2,0],ymm11[5,4],ymm3[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[2,0],ymm15[0,3],ymm3[6,4],ymm15[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1],ymm7[0,3],ymm0[4,5],ymm7[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1,2,3,4],ymm0[5,6,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[1,0],ymm1[2,0],ymm10[5,4],ymm1[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0],ymm2[0,3],ymm1[6,4],ymm2[4,7]
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
; AVX-NEXT:    # ymm2 = ymm12[0,1],mem[0,3],ymm12[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm2, (%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm2, 64(%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm2, 96(%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm2, 32(%rsi)
; AVX-NEXT:    vmovaps %ymm4, 64(%rdx)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm2, (%rdx)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm2, 96(%rdx)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm2, 32(%rdx)
; AVX-NEXT:    vmovaps %ymm1, 64(%rcx)
; AVX-NEXT:    vmovaps %ymm0, (%rcx)
; AVX-NEXT:    vmovaps %ymm5, 96(%rcx)
; AVX-NEXT:    vmovaps %ymm6, 32(%rcx)
; AVX-NEXT:    addq $392, %rsp # imm = 0x188
; AVX-NEXT:    vzeroupper
; AVX-NEXT:    retq
;
; AVX2-LABEL: load_i32_stride3_vf32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    subq $40, %rsp
; AVX2-NEXT:    vmovaps 256(%rdi), %ymm15
; AVX2-NEXT:    vmovaps 224(%rdi), %ymm5
; AVX2-NEXT:    vmovaps 192(%rdi), %ymm3
; AVX2-NEXT:    vmovaps 352(%rdi), %ymm4
; AVX2-NEXT:    vmovaps 320(%rdi), %ymm8
; AVX2-NEXT:    vmovaps 288(%rdi), %ymm10
; AVX2-NEXT:    vmovaps 160(%rdi), %ymm13
; AVX2-NEXT:    vmovaps 128(%rdi), %ymm1
; AVX2-NEXT:    vmovaps (%rdi), %ymm6
; AVX2-NEXT:    vmovaps 32(%rdi), %ymm9
; AVX2-NEXT:    vmovaps 64(%rdi), %ymm7
; AVX2-NEXT:    vmovaps 96(%rdi), %ymm0
; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm13[2],ymm11[3,4],ymm13[5],ymm11[6,7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm12 = [0,3,6,1,4,7,2,5]
; AVX2-NEXT:    vpermps %ymm11, %ymm12, %ymm2
; AVX2-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm10[0],ymm8[1],ymm10[2,3],ymm8[4],ymm10[5,6],ymm8[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm4[2],ymm11[3,4],ymm4[5],ymm11[6,7]
; AVX2-NEXT:    vpermps %ymm11, %ymm12, %ymm2
; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm6[0],ymm9[1],ymm6[2,3],ymm9[4],ymm6[5,6],ymm9[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm7[2],ymm11[3,4],ymm7[5],ymm11[6,7]
; AVX2-NEXT:    vpermps %ymm11, %ymm12, %ymm2
; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm3[0],ymm5[1],ymm3[2,3],ymm5[4],ymm3[5,6],ymm5[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1],ymm15[2],ymm14[3,4],ymm15[5],ymm14[6,7]
; AVX2-NEXT:    vpermps %ymm14, %ymm12, %ymm2
; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0],ymm12[1,2],ymm13[3],ymm12[4,5],ymm13[6],ymm12[7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm2 = [1,4,7,2,5,0,3,6]
; AVX2-NEXT:    vpermps %ymm12, %ymm2, %ymm11
; AVX2-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm10[0,1],ymm8[2],ymm10[3,4],ymm8[5],ymm10[6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm4[0],ymm11[1,2],ymm4[3],ymm11[4,5],ymm4[6],ymm11[7]
; AVX2-NEXT:    vpermps %ymm11, %ymm2, %ymm11
; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm6[0,1],ymm9[2],ymm6[3,4],ymm9[5],ymm6[6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm7[0],ymm12[1,2],ymm7[3],ymm12[4,5],ymm7[6],ymm12[7]
; AVX2-NEXT:    vpermps %ymm12, %ymm2, %ymm12
; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm3[0,1],ymm5[2],ymm3[3,4],ymm5[5],ymm3[6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0],ymm14[1,2],ymm15[3],ymm14[4,5],ymm15[6],ymm14[7]
; AVX2-NEXT:    vpermps %ymm14, %ymm2, %ymm2
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm13 = ymm13[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm13[5,6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm10[2],ymm8[3,4],ymm10[5],ymm8[6,7]
; AVX2-NEXT:    vpermps %ymm8, %ymm1, %ymm8
; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm8[0,1,2,3,4],ymm4[5,6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm9[0,1],ymm6[2],ymm9[3,4],ymm6[5],ymm9[6,7]
; AVX2-NEXT:    vpermps %ymm6, %ymm1, %ymm6
; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5,6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2],ymm5[3,4],ymm3[5],ymm5[6,7]
; AVX2-NEXT:    vpermps %ymm3, %ymm1, %ymm1
; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm15[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm3, 64(%rsi)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm3, (%rsi)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm3, 96(%rsi)
; AVX2-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm3, 32(%rsi)
; AVX2-NEXT:    vmovaps %ymm2, 64(%rdx)
; AVX2-NEXT:    vmovaps %ymm12, (%rdx)
; AVX2-NEXT:    vmovaps %ymm11, 96(%rdx)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm2, 32(%rdx)
; AVX2-NEXT:    vmovaps %ymm1, 64(%rcx)
; AVX2-NEXT:    vmovaps %ymm6, (%rcx)
; AVX2-NEXT:    vmovaps %ymm4, 96(%rcx)
; AVX2-NEXT:    vmovaps %ymm0, 32(%rcx)
; AVX2-NEXT:    addq $40, %rsp
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
;
; AVX2-FP-LABEL: load_i32_stride3_vf32:
; AVX2-FP:       # %bb.0:
; AVX2-FP-NEXT:    subq $40, %rsp
; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm15
; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm5
; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm3
; AVX2-FP-NEXT:    vmovaps 352(%rdi), %ymm4
; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm8
; AVX2-FP-NEXT:    vmovaps 288(%rdi), %ymm10
; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm13
; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm1
; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm6
; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm9
; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm7
; AVX2-FP-NEXT:    vmovaps 96(%rdi), %ymm0
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm13[2],ymm11[3,4],ymm13[5],ymm11[6,7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm12 = [0,3,6,1,4,7,2,5]
; AVX2-FP-NEXT:    vpermps %ymm11, %ymm12, %ymm2
; AVX2-FP-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm10[0],ymm8[1],ymm10[2,3],ymm8[4],ymm10[5,6],ymm8[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm4[2],ymm11[3,4],ymm4[5],ymm11[6,7]
; AVX2-FP-NEXT:    vpermps %ymm11, %ymm12, %ymm2
; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm6[0],ymm9[1],ymm6[2,3],ymm9[4],ymm6[5,6],ymm9[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm7[2],ymm11[3,4],ymm7[5],ymm11[6,7]
; AVX2-FP-NEXT:    vpermps %ymm11, %ymm12, %ymm2
; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm3[0],ymm5[1],ymm3[2,3],ymm5[4],ymm3[5,6],ymm5[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1],ymm15[2],ymm14[3,4],ymm15[5],ymm14[6,7]
; AVX2-FP-NEXT:    vpermps %ymm14, %ymm12, %ymm2
; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0],ymm12[1,2],ymm13[3],ymm12[4,5],ymm13[6],ymm12[7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm2 = [1,4,7,2,5,0,3,6]
; AVX2-FP-NEXT:    vpermps %ymm12, %ymm2, %ymm11
; AVX2-FP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm10[0,1],ymm8[2],ymm10[3,4],ymm8[5],ymm10[6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm4[0],ymm11[1,2],ymm4[3],ymm11[4,5],ymm4[6],ymm11[7]
; AVX2-FP-NEXT:    vpermps %ymm11, %ymm2, %ymm11
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm6[0,1],ymm9[2],ymm6[3,4],ymm9[5],ymm6[6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm7[0],ymm12[1,2],ymm7[3],ymm12[4,5],ymm7[6],ymm12[7]
; AVX2-FP-NEXT:    vpermps %ymm12, %ymm2, %ymm12
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm3[0,1],ymm5[2],ymm3[3,4],ymm5[5],ymm3[6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0],ymm14[1,2],ymm15[3],ymm14[4,5],ymm15[6],ymm14[7]
; AVX2-FP-NEXT:    vpermps %ymm14, %ymm2, %ymm2
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm13 = ymm13[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm13[5,6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm10[2],ymm8[3,4],ymm10[5],ymm8[6,7]
; AVX2-FP-NEXT:    vpermps %ymm8, %ymm1, %ymm8
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm8[0,1,2,3,4],ymm4[5,6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm9[0,1],ymm6[2],ymm9[3,4],ymm6[5],ymm9[6,7]
; AVX2-FP-NEXT:    vpermps %ymm6, %ymm1, %ymm6
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5,6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2],ymm5[3,4],ymm3[5],ymm5[6,7]
; AVX2-FP-NEXT:    vpermps %ymm3, %ymm1, %ymm1
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm15[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm3, 64(%rsi)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm3, (%rsi)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm3, 96(%rsi)
; AVX2-FP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rsi)
; AVX2-FP-NEXT:    vmovaps %ymm2, 64(%rdx)
; AVX2-FP-NEXT:    vmovaps %ymm12, (%rdx)
; AVX2-FP-NEXT:    vmovaps %ymm11, 96(%rdx)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm2, 32(%rdx)
; AVX2-FP-NEXT:    vmovaps %ymm1, 64(%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm6, (%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm4, 96(%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rcx)
; AVX2-FP-NEXT:    addq $40, %rsp
; AVX2-FP-NEXT:    vzeroupper
; AVX2-FP-NEXT:    retq
;
; AVX2-FCP-LABEL: load_i32_stride3_vf32:
; AVX2-FCP:       # %bb.0:
; AVX2-FCP-NEXT:    subq $72, %rsp
; AVX2-FCP-NEXT:    vmovaps 256(%rdi), %ymm0
; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm2
; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm1
; AVX2-FCP-NEXT:    vmovaps 352(%rdi), %ymm4
; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm6
; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm7
; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm13
; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm14
; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm8
; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm9
; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm10
; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %ymm15
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm15[0],ymm14[1],ymm15[2,3],ymm14[4],ymm15[5,6],ymm14[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm13[2],ymm3[3,4],ymm13[5],ymm3[6,7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm12 = [0,3,6,1,4,7,2,5]
; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm12, %ymm3
; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0],ymm6[1],ymm7[2,3],ymm6[4],ymm7[5,6],ymm6[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1],ymm4[2],ymm5[3,4],ymm4[5],ymm5[6,7]
; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm12, %ymm3
; AVX2-FCP-NEXT:    vmovups %ymm3, (%rsp) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm8[0],ymm9[1],ymm8[2,3],ymm9[4],ymm8[5,6],ymm9[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7]
; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm12, %ymm3
; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6],ymm2[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm0[2],ymm3[3,4],ymm0[5],ymm3[6,7]
; AVX2-FCP-NEXT:    vmovaps %ymm0, %ymm2
; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm12, %ymm0
; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm15[0,1],ymm14[2],ymm15[3,4],ymm14[5],ymm15[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0],ymm12[1,2],ymm13[3],ymm12[4,5],ymm13[6],ymm12[7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm3 = [1,4,7,2,5,0,3,6]
; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm3, %ymm5
; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0,1],ymm6[2],ymm7[3,4],ymm6[5],ymm7[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm4[0],ymm5[1,2],ymm4[3],ymm5[4,5],ymm4[6],ymm5[7]
; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm3, %ymm5
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm8[0,1],ymm9[2],ymm8[3,4],ymm9[5],ymm8[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm10[0],ymm11[1,2],ymm10[3],ymm11[4,5],ymm10[6],ymm11[7]
; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm3, %ymm11
; AVX2-FCP-NEXT:    vmovaps %ymm1, %ymm0
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm2[0],ymm12[1,2],ymm2[3],ymm12[4,5],ymm2[6],ymm12[7]
; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm3, %ymm3
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm14[0,1],ymm15[2],ymm14[3,4],ymm15[5],ymm14[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0],ymm13[1],ymm12[2,3],ymm13[4],ymm12[5,6],ymm13[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm7[2],ymm6[3,4],ymm7[5],ymm6[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0],ymm4[1],ymm6[2,3],ymm4[4],ymm6[5,6],ymm4[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm9[0,1],ymm8[2],ymm9[3,4],ymm8[5],ymm9[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0],ymm10[1],ymm6[2,3],ymm10[4],ymm6[5,6],ymm10[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6],ymm2[7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm2 = [2,5,0,3,6,1,4,7]
; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm2, %ymm7
; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm2, %ymm4
; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm2, %ymm6
; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm0, 64(%rsi)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm2, (%rsi)
; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm2 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm2, 96(%rsi)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm2, 32(%rsi)
; AVX2-FCP-NEXT:    vmovaps %ymm3, 64(%rdx)
; AVX2-FCP-NEXT:    vmovaps %ymm11, (%rdx)
; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rdx)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%rdx)
; AVX2-FCP-NEXT:    vmovaps %ymm1, 64(%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm4, 96(%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm7, 32(%rcx)
; AVX2-FCP-NEXT:    addq $72, %rsp
; AVX2-FCP-NEXT:    vzeroupper
; AVX2-FCP-NEXT:    retq
;
; AVX512-LABEL: load_i32_stride3_vf32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovdqa64 320(%rdi), %zmm0
; AVX512-NEXT:    vmovdqa64 256(%rdi), %zmm1
; AVX512-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512-NEXT:    vmovdqa64 64(%rdi), %zmm3
; AVX512-NEXT:    vmovdqa64 128(%rdi), %zmm4
; AVX512-NEXT:    vmovdqa64 192(%rdi), %zmm5
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm6 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512-NEXT:    vmovdqa64 %zmm5, %zmm7
; AVX512-NEXT:    vpermt2d %zmm1, %zmm6, %zmm7
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512-NEXT:    vpermt2d %zmm0, %zmm8, %zmm7
; AVX512-NEXT:    vpermi2d %zmm3, %zmm2, %zmm6
; AVX512-NEXT:    vpermt2d %zmm4, %zmm8, %zmm6
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512-NEXT:    vmovdqa64 %zmm1, %zmm9
; AVX512-NEXT:    vpermt2d %zmm5, %zmm8, %zmm9
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512-NEXT:    vpermt2d %zmm0, %zmm10, %zmm9
; AVX512-NEXT:    vpermi2d %zmm2, %zmm3, %zmm8
; AVX512-NEXT:    vpermt2d %zmm4, %zmm10, %zmm8
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512-NEXT:    vpermt2d %zmm1, %zmm10, %zmm5
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512-NEXT:    vpermt2d %zmm0, %zmm1, %zmm5
; AVX512-NEXT:    vpermt2d %zmm3, %zmm10, %zmm2
; AVX512-NEXT:    vpermt2d %zmm4, %zmm1, %zmm2
; AVX512-NEXT:    vmovdqa64 %zmm7, 64(%rsi)
; AVX512-NEXT:    vmovdqa64 %zmm6, (%rsi)
; AVX512-NEXT:    vmovdqa64 %zmm9, 64(%rdx)
; AVX512-NEXT:    vmovdqa64 %zmm8, (%rdx)
; AVX512-NEXT:    vmovdqa64 %zmm5, 64(%rcx)
; AVX512-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
;
; AVX512-FCP-LABEL: load_i32_stride3_vf32:
; AVX512-FCP:       # %bb.0:
; AVX512-FCP-NEXT:    vmovdqa64 320(%rdi), %zmm0
; AVX512-FCP-NEXT:    vmovdqa64 256(%rdi), %zmm1
; AVX512-FCP-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm3
; AVX512-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm4
; AVX512-FCP-NEXT:    vmovdqa64 192(%rdi), %zmm5
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm6 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512-FCP-NEXT:    vmovdqa64 %zmm5, %zmm7
; AVX512-FCP-NEXT:    vpermt2d %zmm1, %zmm6, %zmm7
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512-FCP-NEXT:    vpermt2d %zmm0, %zmm8, %zmm7
; AVX512-FCP-NEXT:    vpermi2d %zmm3, %zmm2, %zmm6
; AVX512-FCP-NEXT:    vpermt2d %zmm4, %zmm8, %zmm6
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512-FCP-NEXT:    vmovdqa64 %zmm1, %zmm9
; AVX512-FCP-NEXT:    vpermt2d %zmm5, %zmm8, %zmm9
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512-FCP-NEXT:    vpermt2d %zmm0, %zmm10, %zmm9
; AVX512-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm8
; AVX512-FCP-NEXT:    vpermt2d %zmm4, %zmm10, %zmm8
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512-FCP-NEXT:    vpermt2d %zmm1, %zmm10, %zmm5
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512-FCP-NEXT:    vpermt2d %zmm0, %zmm1, %zmm5
; AVX512-FCP-NEXT:    vpermt2d %zmm3, %zmm10, %zmm2
; AVX512-FCP-NEXT:    vpermt2d %zmm4, %zmm1, %zmm2
; AVX512-FCP-NEXT:    vmovdqa64 %zmm7, 64(%rsi)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm6, (%rsi)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm9, 64(%rdx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm8, (%rdx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm5, 64(%rcx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512-FCP-NEXT:    vzeroupper
; AVX512-FCP-NEXT:    retq
;
; AVX512DQ-LABEL: load_i32_stride3_vf32:
; AVX512DQ:       # %bb.0:
; AVX512DQ-NEXT:    vmovdqa64 320(%rdi), %zmm0
; AVX512DQ-NEXT:    vmovdqa64 256(%rdi), %zmm1
; AVX512DQ-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512DQ-NEXT:    vmovdqa64 64(%rdi), %zmm3
; AVX512DQ-NEXT:    vmovdqa64 128(%rdi), %zmm4
; AVX512DQ-NEXT:    vmovdqa64 192(%rdi), %zmm5
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm6 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-NEXT:    vmovdqa64 %zmm5, %zmm7
; AVX512DQ-NEXT:    vpermt2d %zmm1, %zmm6, %zmm7
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-NEXT:    vpermt2d %zmm0, %zmm8, %zmm7
; AVX512DQ-NEXT:    vpermi2d %zmm3, %zmm2, %zmm6
; AVX512DQ-NEXT:    vpermt2d %zmm4, %zmm8, %zmm6
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-NEXT:    vmovdqa64 %zmm1, %zmm9
; AVX512DQ-NEXT:    vpermt2d %zmm5, %zmm8, %zmm9
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-NEXT:    vpermt2d %zmm0, %zmm10, %zmm9
; AVX512DQ-NEXT:    vpermi2d %zmm2, %zmm3, %zmm8
; AVX512DQ-NEXT:    vpermt2d %zmm4, %zmm10, %zmm8
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-NEXT:    vpermt2d %zmm1, %zmm10, %zmm5
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-NEXT:    vpermt2d %zmm0, %zmm1, %zmm5
; AVX512DQ-NEXT:    vpermt2d %zmm3, %zmm10, %zmm2
; AVX512DQ-NEXT:    vpermt2d %zmm4, %zmm1, %zmm2
; AVX512DQ-NEXT:    vmovdqa64 %zmm7, 64(%rsi)
; AVX512DQ-NEXT:    vmovdqa64 %zmm6, (%rsi)
; AVX512DQ-NEXT:    vmovdqa64 %zmm9, 64(%rdx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm8, (%rdx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm5, 64(%rcx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512DQ-NEXT:    vzeroupper
; AVX512DQ-NEXT:    retq
;
; AVX512DQ-FCP-LABEL: load_i32_stride3_vf32:
; AVX512DQ-FCP:       # %bb.0:
; AVX512DQ-FCP-NEXT:    vmovdqa64 320(%rdi), %zmm0
; AVX512DQ-FCP-NEXT:    vmovdqa64 256(%rdi), %zmm1
; AVX512DQ-FCP-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512DQ-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm3
; AVX512DQ-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm4
; AVX512DQ-FCP-NEXT:    vmovdqa64 192(%rdi), %zmm5
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm6 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm5, %zmm7
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm1, %zmm6, %zmm7
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm0, %zmm8, %zmm7
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm3, %zmm2, %zmm6
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm4, %zmm8, %zmm6
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm1, %zmm9
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm5, %zmm8, %zmm9
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm0, %zmm10, %zmm9
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm8
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm4, %zmm10, %zmm8
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm1, %zmm10, %zmm5
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm0, %zmm1, %zmm5
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm3, %zmm10, %zmm2
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm4, %zmm1, %zmm2
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm7, 64(%rsi)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm6, (%rsi)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm9, 64(%rdx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm8, (%rdx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm5, 64(%rcx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512DQ-FCP-NEXT:    vzeroupper
; AVX512DQ-FCP-NEXT:    retq
;
; AVX512BW-LABEL: load_i32_stride3_vf32:
; AVX512BW:       # %bb.0:
; AVX512BW-NEXT:    vmovdqa64 320(%rdi), %zmm0
; AVX512BW-NEXT:    vmovdqa64 256(%rdi), %zmm1
; AVX512BW-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512BW-NEXT:    vmovdqa64 64(%rdi), %zmm3
; AVX512BW-NEXT:    vmovdqa64 128(%rdi), %zmm4
; AVX512BW-NEXT:    vmovdqa64 192(%rdi), %zmm5
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm6 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512BW-NEXT:    vmovdqa64 %zmm5, %zmm7
; AVX512BW-NEXT:    vpermt2d %zmm1, %zmm6, %zmm7
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512BW-NEXT:    vpermt2d %zmm0, %zmm8, %zmm7
; AVX512BW-NEXT:    vpermi2d %zmm3, %zmm2, %zmm6
; AVX512BW-NEXT:    vpermt2d %zmm4, %zmm8, %zmm6
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm9
; AVX512BW-NEXT:    vpermt2d %zmm5, %zmm8, %zmm9
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512BW-NEXT:    vpermt2d %zmm0, %zmm10, %zmm9
; AVX512BW-NEXT:    vpermi2d %zmm2, %zmm3, %zmm8
; AVX512BW-NEXT:    vpermt2d %zmm4, %zmm10, %zmm8
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512BW-NEXT:    vpermt2d %zmm1, %zmm10, %zmm5
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512BW-NEXT:    vpermt2d %zmm0, %zmm1, %zmm5
; AVX512BW-NEXT:    vpermt2d %zmm3, %zmm10, %zmm2
; AVX512BW-NEXT:    vpermt2d %zmm4, %zmm1, %zmm2
; AVX512BW-NEXT:    vmovdqa64 %zmm7, 64(%rsi)
; AVX512BW-NEXT:    vmovdqa64 %zmm6, (%rsi)
; AVX512BW-NEXT:    vmovdqa64 %zmm9, 64(%rdx)
; AVX512BW-NEXT:    vmovdqa64 %zmm8, (%rdx)
; AVX512BW-NEXT:    vmovdqa64 %zmm5, 64(%rcx)
; AVX512BW-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512BW-NEXT:    vzeroupper
; AVX512BW-NEXT:    retq
;
; AVX512BW-FCP-LABEL: load_i32_stride3_vf32:
; AVX512BW-FCP:       # %bb.0:
; AVX512BW-FCP-NEXT:    vmovdqa64 320(%rdi), %zmm0
; AVX512BW-FCP-NEXT:    vmovdqa64 256(%rdi), %zmm1
; AVX512BW-FCP-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512BW-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm3
; AVX512BW-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm4
; AVX512BW-FCP-NEXT:    vmovdqa64 192(%rdi), %zmm5
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm6 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm5, %zmm7
; AVX512BW-FCP-NEXT:    vpermt2d %zmm1, %zmm6, %zmm7
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512BW-FCP-NEXT:    vpermt2d %zmm0, %zmm8, %zmm7
; AVX512BW-FCP-NEXT:    vpermi2d %zmm3, %zmm2, %zmm6
; AVX512BW-FCP-NEXT:    vpermt2d %zmm4, %zmm8, %zmm6
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm1, %zmm9
; AVX512BW-FCP-NEXT:    vpermt2d %zmm5, %zmm8, %zmm9
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512BW-FCP-NEXT:    vpermt2d %zmm0, %zmm10, %zmm9
; AVX512BW-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm8
; AVX512BW-FCP-NEXT:    vpermt2d %zmm4, %zmm10, %zmm8
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512BW-FCP-NEXT:    vpermt2d %zmm1, %zmm10, %zmm5
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512BW-FCP-NEXT:    vpermt2d %zmm0, %zmm1, %zmm5
; AVX512BW-FCP-NEXT:    vpermt2d %zmm3, %zmm10, %zmm2
; AVX512BW-FCP-NEXT:    vpermt2d %zmm4, %zmm1, %zmm2
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm7, 64(%rsi)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm6, (%rsi)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm9, 64(%rdx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm8, (%rdx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm5, 64(%rcx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512BW-FCP-NEXT:    vzeroupper
; AVX512BW-FCP-NEXT:    retq
;
; AVX512DQ-BW-LABEL: load_i32_stride3_vf32:
; AVX512DQ-BW:       # %bb.0:
; AVX512DQ-BW-NEXT:    vmovdqa64 320(%rdi), %zmm0
; AVX512DQ-BW-NEXT:    vmovdqa64 256(%rdi), %zmm1
; AVX512DQ-BW-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512DQ-BW-NEXT:    vmovdqa64 64(%rdi), %zmm3
; AVX512DQ-BW-NEXT:    vmovdqa64 128(%rdi), %zmm4
; AVX512DQ-BW-NEXT:    vmovdqa64 192(%rdi), %zmm5
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm6 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm5, %zmm7
; AVX512DQ-BW-NEXT:    vpermt2d %zmm1, %zmm6, %zmm7
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-BW-NEXT:    vpermt2d %zmm0, %zmm8, %zmm7
; AVX512DQ-BW-NEXT:    vpermi2d %zmm3, %zmm2, %zmm6
; AVX512DQ-BW-NEXT:    vpermt2d %zmm4, %zmm8, %zmm6
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm1, %zmm9
; AVX512DQ-BW-NEXT:    vpermt2d %zmm5, %zmm8, %zmm9
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-BW-NEXT:    vpermt2d %zmm0, %zmm10, %zmm9
; AVX512DQ-BW-NEXT:    vpermi2d %zmm2, %zmm3, %zmm8
; AVX512DQ-BW-NEXT:    vpermt2d %zmm4, %zmm10, %zmm8
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-BW-NEXT:    vpermt2d %zmm1, %zmm10, %zmm5
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-BW-NEXT:    vpermt2d %zmm0, %zmm1, %zmm5
; AVX512DQ-BW-NEXT:    vpermt2d %zmm3, %zmm10, %zmm2
; AVX512DQ-BW-NEXT:    vpermt2d %zmm4, %zmm1, %zmm2
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm7, 64(%rsi)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm6, (%rsi)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm9, 64(%rdx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm8, (%rdx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm5, 64(%rcx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512DQ-BW-NEXT:    vzeroupper
; AVX512DQ-BW-NEXT:    retq
;
; AVX512DQ-BW-FCP-LABEL: load_i32_stride3_vf32:
; AVX512DQ-BW-FCP:       # %bb.0:
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 320(%rdi), %zmm0
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 256(%rdi), %zmm1
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm3
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm4
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 192(%rdi), %zmm5
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm6 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm5, %zmm7
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm1, %zmm6, %zmm7
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm0, %zmm8, %zmm7
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm3, %zmm2, %zmm6
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm4, %zmm8, %zmm6
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm8 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm1, %zmm9
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm5, %zmm8, %zmm9
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm0, %zmm10, %zmm9
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm2, %zmm3, %zmm8
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm4, %zmm10, %zmm8
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm10 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm1, %zmm10, %zmm5
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm0, %zmm1, %zmm5
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm3, %zmm10, %zmm2
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm4, %zmm1, %zmm2
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm7, 64(%rsi)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm6, (%rsi)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm9, 64(%rdx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm8, (%rdx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm5, 64(%rcx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512DQ-BW-FCP-NEXT:    vzeroupper
; AVX512DQ-BW-FCP-NEXT:    retq
  %wide.vec = load <96 x i32>, ptr %in.vec, align 64
  %strided.vec0 = shufflevector <96 x i32> %wide.vec, <96 x i32> poison, <32 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45, i32 48, i32 51, i32 54, i32 57, i32 60, i32 63, i32 66, i32 69, i32 72, i32 75, i32 78, i32 81, i32 84, i32 87, i32 90, i32 93>
  %strided.vec1 = shufflevector <96 x i32> %wide.vec, <96 x i32> poison, <32 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46, i32 49, i32 52, i32 55, i32 58, i32 61, i32 64, i32 67, i32 70, i32 73, i32 76, i32 79, i32 82, i32 85, i32 88, i32 91, i32 94>
  %strided.vec2 = shufflevector <96 x i32> %wide.vec, <96 x i32> poison, <32 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47, i32 50, i32 53, i32 56, i32 59, i32 62, i32 65, i32 68, i32 71, i32 74, i32 77, i32 80, i32 83, i32 86, i32 89, i32 92, i32 95>
  store <32 x i32> %strided.vec0, ptr %out.vec0, align 64
  store <32 x i32> %strided.vec1, ptr %out.vec1, align 64
  store <32 x i32> %strided.vec2, ptr %out.vec2, align 64
  ret void
}

define void @load_i32_stride3_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr %out.vec2) nounwind {
; SSE-LABEL: load_i32_stride3_vf64:
; SSE:       # %bb.0:
; SSE-NEXT:    subq $1112, %rsp # imm = 0x458
; SSE-NEXT:    movaps 624(%rdi), %xmm2
; SSE-NEXT:    movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 656(%rdi), %xmm4
; SSE-NEXT:    movaps %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 640(%rdi), %xmm10
; SSE-NEXT:    movaps %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 432(%rdi), %xmm6
; SSE-NEXT:    movaps %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 464(%rdi), %xmm5
; SSE-NEXT:    movaps %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 448(%rdi), %xmm11
; SSE-NEXT:    movaps %xmm11, (%rsp) # 16-byte Spill
; SSE-NEXT:    movaps 240(%rdi), %xmm7
; SSE-NEXT:    movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 272(%rdi), %xmm3
; SSE-NEXT:    movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 256(%rdi), %xmm13
; SSE-NEXT:    movaps %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 48(%rdi), %xmm9
; SSE-NEXT:    movaps 80(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 64(%rdi), %xmm12
; SSE-NEXT:    movaps %xmm12, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps %xmm9, %xmm1
; SSE-NEXT:    movaps %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm13, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm3[1,0]
; SSE-NEXT:    movaps %xmm7, %xmm1
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm11, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm5[1,0]
; SSE-NEXT:    movaps %xmm6, %xmm1
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm10, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm4[1,0]
; SSE-NEXT:    movaps %xmm2, %xmm1
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 16(%rdi), %xmm0
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 32(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps (%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 224(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 208(%rdi), %xmm0
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps 192(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 416(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 400(%rdi), %xmm0
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps 384(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 608(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 592(%rdi), %xmm0
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps 576(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 176(%rdi), %xmm10
; SSE-NEXT:    movaps 160(%rdi), %xmm8
; SSE-NEXT:    movaps %xmm8, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm10[1,0]
; SSE-NEXT:    movaps %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 144(%rdi), %xmm2
; SSE-NEXT:    movaps %xmm2, %xmm1
; SSE-NEXT:    movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 368(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 352(%rdi), %xmm15
; SSE-NEXT:    movaps %xmm15, %xmm0
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps 336(%rdi), %xmm14
; SSE-NEXT:    movaps %xmm14, %xmm1
; SSE-NEXT:    movaps %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 560(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 544(%rdi), %xmm0
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps 528(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 752(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 736(%rdi), %xmm0
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[1,0]
; SSE-NEXT:    movaps 720(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,2]
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 128(%rdi), %xmm6
; SSE-NEXT:    movaps 112(%rdi), %xmm4
; SSE-NEXT:    movaps %xmm4, %xmm1
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm6[1,0]
; SSE-NEXT:    movaps 96(%rdi), %xmm3
; SSE-NEXT:    movaps %xmm3, %xmm7
; SSE-NEXT:    movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,3],xmm1[0,2]
; SSE-NEXT:    movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 320(%rdi), %xmm13
; SSE-NEXT:    movaps 304(%rdi), %xmm11
; SSE-NEXT:    movaps %xmm11, %xmm1
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm13[1,0]
; SSE-NEXT:    movaps %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 288(%rdi), %xmm5
; SSE-NEXT:    movaps %xmm5, %xmm7
; SSE-NEXT:    movaps %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,3],xmm1[0,2]
; SSE-NEXT:    movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 512(%rdi), %xmm0
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 496(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[1,0]
; SSE-NEXT:    movaps 480(%rdi), %xmm7
; SSE-NEXT:    movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,3],xmm1[0,2]
; SSE-NEXT:    movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 704(%rdi), %xmm7
; SSE-NEXT:    movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps 688(%rdi), %xmm1
; SSE-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm7[1,0]
; SSE-NEXT:    movaps 672(%rdi), %xmm7
; SSE-NEXT:    movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,3],xmm1[0,2]
; SSE-NEXT:    movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm9 = xmm9[1,0],xmm12[0,0]
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
; SSE-NEXT:    # xmm12 = xmm12[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm9 = xmm9[0,2],xmm12[0,2]
; SSE-NEXT:    movaps %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm8[0,0]
; SSE-NEXT:    shufps {{.*#+}} xmm8 = xmm8[3,1],xmm10[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm8[0,2]
; SSE-NEXT:    movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,0],xmm4[0,0]
; SSE-NEXT:    shufps {{.*#+}} xmm4 = xmm4[3,1],xmm6[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
; SSE-NEXT:    movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[1,0],xmm1[0,0]
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = xmm1[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,2],xmm1[0,2]
; SSE-NEXT:    movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
; SSE-NEXT:    movaps %xmm9, %xmm0
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm3[0,0]
; SSE-NEXT:    movaps %xmm3, %xmm1
; SSE-NEXT:    movaps %xmm3, %xmm12
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = xmm1[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm14 = xmm14[1,0],xmm15[0,0]
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
; SSE-NEXT:    # xmm15 = xmm15[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm14 = xmm14[0,2],xmm15[0,2]
; SSE-NEXT:    movaps %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[1,0],xmm11[0,0]
; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[3,1],xmm13[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,2],xmm11[0,2]
; SSE-NEXT:    movaps %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    movaps (%rsp), %xmm1 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm1[0,0]
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = xmm1[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
; SSE-NEXT:    movaps %xmm7, %xmm0
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm4[0,0]
; SSE-NEXT:    movaps %xmm4, %xmm1
; SSE-NEXT:    movaps %xmm4, %xmm8
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = xmm1[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm1[0,0]
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = xmm1[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm1[0,0]
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,1],xmm5[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm1[0,0]
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = xmm1[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm14[0,0]
; SSE-NEXT:    movaps %xmm14, %xmm1
; SSE-NEXT:    movaps %xmm14, %xmm3
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = xmm1[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm1[0,0]
; SSE-NEXT:    shufps $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = xmm1[3,1],mem[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm14 = xmm14[1,0],xmm1[0,0]
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,1],xmm4[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm14 = xmm14[0,2],xmm1[0,2]
; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm2[2,3,2,3]
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm0[0,0]
; SSE-NEXT:    movaps %xmm2, %xmm11
; SSE-NEXT:    movaps %xmm0, %xmm1
; SSE-NEXT:    movaps %xmm0, %xmm2
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,1],xmm0[2,3]
; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[0,2],xmm1[0,2]
; SSE-NEXT:    movaps %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[1,1,1,1]
; SSE-NEXT:    punpckldq {{.*#+}} xmm10 = xmm10[0],xmm1[0],xmm10[1],xmm1[1]
; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[0,1],xmm0[0,3]
; SSE-NEXT:    movaps %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = xmm0[0,1],mem[0,3]
; SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
; SSE-NEXT:    # xmm15 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm15 = xmm15[0],xmm0[0],xmm15[1],xmm0[1]
; SSE-NEXT:    shufps {{.*#+}} xmm15 = xmm15[0,1],xmm6[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Folded Reload
; SSE-NEXT:    # xmm13 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm13 = xmm13[0],xmm0[0],xmm13[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Folded Reload
; SSE-NEXT:    # xmm13 = xmm13[0,1],mem[0,3]
; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm12[1,1,1,1]
; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm9[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm12 = xmm12[0],xmm0[0],xmm12[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
; SSE-NEXT:    # xmm12 = xmm12[0,1],mem[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Folded Reload
; SSE-NEXT:    # xmm11 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm11 = xmm11[0],xmm0[0],xmm11[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Folded Reload
; SSE-NEXT:    # xmm11 = xmm11[0,1],mem[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Folded Reload
; SSE-NEXT:    # xmm10 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm10 = xmm10[0],xmm0[0],xmm10[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Folded Reload
; SSE-NEXT:    # xmm10 = xmm10[0,1],mem[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
; SSE-NEXT:    # xmm9 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm9 = xmm9[0],xmm0[0],xmm9[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
; SSE-NEXT:    # xmm9 = xmm9[0,1],mem[0,3]
; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm8[1,1,1,1]
; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm7[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm8 = xmm8[0],xmm0[0],xmm8[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Folded Reload
; SSE-NEXT:    # xmm8 = xmm8[0,1],mem[0,3]
; SSE-NEXT:    pshufd $85, (%rsp), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
; SSE-NEXT:    # xmm7 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm0[0],xmm7[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
; SSE-NEXT:    # xmm7 = xmm7[0,1],mem[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
; SSE-NEXT:    # xmm6 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm6 = xmm6[0],xmm0[0],xmm6[1],xmm0[1]
; SSE-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,1],xmm5[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
; SSE-NEXT:    # xmm5 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm0[0],xmm5[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
; SSE-NEXT:    # xmm5 = xmm5[0,1],mem[0,3]
; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
; SSE-NEXT:    # xmm3 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
; SSE-NEXT:    # xmm3 = xmm3[0,1],mem[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
; SSE-NEXT:    # xmm2 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
; SSE-NEXT:    # xmm2 = xmm2[0,1],mem[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT:    # xmm1 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm4[0,3]
; SSE-NEXT:    pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
; SSE-NEXT:    # xmm4 = mem[1,1,1,1]
; SSE-NEXT:    pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = mem[2,3,2,3]
; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
; SSE-NEXT:    shufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT:    # xmm0 = xmm0[0,1],mem[0,3]
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 224(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 160(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 96(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 32(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 240(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 176(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 112(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 48(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 192(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 128(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 64(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, (%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 208(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 144(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 80(%rsi)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 16(%rsi)
; SSE-NEXT:    movaps %xmm14, 224(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 240(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 192(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 208(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 160(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 176(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 128(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 144(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 96(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 112(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 64(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 80(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 32(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 48(%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, (%rdx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
; SSE-NEXT:    movaps %xmm4, 16(%rdx)
; SSE-NEXT:    movaps %xmm0, 240(%rcx)
; SSE-NEXT:    movaps %xmm1, 224(%rcx)
; SSE-NEXT:    movaps %xmm2, 208(%rcx)
; SSE-NEXT:    movaps %xmm3, 192(%rcx)
; SSE-NEXT:    movaps %xmm5, 176(%rcx)
; SSE-NEXT:    movaps %xmm6, 160(%rcx)
; SSE-NEXT:    movaps %xmm7, 144(%rcx)
; SSE-NEXT:    movaps %xmm8, 128(%rcx)
; SSE-NEXT:    movaps %xmm9, 112(%rcx)
; SSE-NEXT:    movaps %xmm10, 96(%rcx)
; SSE-NEXT:    movaps %xmm11, 80(%rcx)
; SSE-NEXT:    movaps %xmm12, 64(%rcx)
; SSE-NEXT:    movaps %xmm13, 48(%rcx)
; SSE-NEXT:    movaps %xmm15, 32(%rcx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    movaps %xmm0, 16(%rcx)
; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT:    movaps %xmm0, (%rcx)
; SSE-NEXT:    addq $1112, %rsp # imm = 0x458
; SSE-NEXT:    retq
;
; AVX-LABEL: load_i32_stride3_vf64:
; AVX:       # %bb.0:
; AVX-NEXT:    subq $1384, %rsp # imm = 0x568
; AVX-NEXT:    vmovaps 544(%rdi), %ymm2
; AVX-NEXT:    vmovaps 512(%rdi), %ymm3
; AVX-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 480(%rdi), %ymm4
; AVX-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 352(%rdi), %ymm5
; AVX-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 320(%rdi), %ymm6
; AVX-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 288(%rdi), %ymm7
; AVX-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 160(%rdi), %ymm8
; AVX-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 128(%rdi), %ymm9
; AVX-NEXT:    vmovups %ymm9, (%rsp) # 32-byte Spill
; AVX-NEXT:    vmovaps 96(%rdi), %ymm0
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm9[1],ymm0[2,3],ymm9[4],ymm0[5,6],ymm9[7]
; AVX-NEXT:    vmovaps 112(%rdi), %xmm1
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,1],ymm9[1,3],ymm1[6,5],ymm9[5,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,3],ymm1[0,2],ymm0[4,7],ymm1[4,6]
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm8[2,3,0,1]
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[1,0],ymm1[2,0],ymm8[5,4],ymm1[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm7[0],ymm6[1],ymm7[2,3],ymm6[4],ymm7[5,6],ymm6[7]
; AVX-NEXT:    vmovaps 304(%rdi), %xmm1
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,1],ymm6[1,3],ymm1[6,5],ymm6[5,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,3],ymm1[0,2],ymm0[4,7],ymm1[4,6]
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm14 = ymm5[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[1,0],ymm14[2,0],ymm5[5,4],ymm14[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
; AVX-NEXT:    vmovaps 496(%rdi), %xmm1
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,1],ymm3[1,3],ymm1[6,5],ymm3[5,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,3],ymm1[0,2],ymm0[4,7],ymm1[4,6]
; AVX-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm2[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[1,0],ymm3[2,0],ymm2[5,4],ymm3[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 704(%rdi), %ymm2
; AVX-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 688(%rdi), %xmm0
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,1],ymm2[1,3],ymm0[6,5],ymm2[5,7]
; AVX-NEXT:    vmovaps 672(%rdi), %ymm1
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6],ymm2[7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm1[0,3],ymm0[0,2],ymm1[4,7],ymm0[4,6]
; AVX-NEXT:    vmovaps 736(%rdi), %ymm1
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm15 = ymm1[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,0],ymm15[2,0],ymm1[5,4],ymm15[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 32(%rdi), %ymm7
; AVX-NEXT:    vmovaps 16(%rdi), %xmm0
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,1],ymm7[1,3],ymm0[6,5],ymm7[5,7]
; AVX-NEXT:    vmovaps (%rdi), %ymm1
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm7[1],ymm1[2,3],ymm7[4],ymm1[5,6],ymm7[7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm1[0,3],ymm0[0,2],ymm1[4,7],ymm0[4,6]
; AVX-NEXT:    vmovaps 64(%rdi), %ymm4
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm4[2,3,0,1]
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm4[1,0],ymm1[2,0],ymm4[5,4],ymm1[6,4]
; AVX-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 224(%rdi), %ymm6
; AVX-NEXT:    vmovaps 208(%rdi), %xmm0
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,1],ymm6[1,3],ymm0[6,5],ymm6[5,7]
; AVX-NEXT:    vmovaps 192(%rdi), %ymm1
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm6[1],ymm1[2,3],ymm6[4],ymm1[5,6],ymm6[7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm1[0,3],ymm0[0,2],ymm1[4,7],ymm0[4,6]
; AVX-NEXT:    vmovaps 256(%rdi), %ymm5
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm5[2,3,0,1]
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[1,0],ymm1[2,0],ymm5[5,4],ymm1[6,4]
; AVX-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 416(%rdi), %ymm12
; AVX-NEXT:    vmovaps 400(%rdi), %xmm0
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,1],ymm12[1,3],ymm0[6,5],ymm12[5,7]
; AVX-NEXT:    vmovaps 384(%rdi), %ymm1
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm12[1],ymm1[2,3],ymm12[4],ymm1[5,6],ymm12[7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm1[0,3],ymm0[0,2],ymm1[4,7],ymm0[4,6]
; AVX-NEXT:    vmovaps 448(%rdi), %ymm8
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm9 = ymm8[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[1,0],ymm9[2,0],ymm8[5,4],ymm9[6,4]
; AVX-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovaps 608(%rdi), %ymm10
; AVX-NEXT:    vmovaps 592(%rdi), %xmm0
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,1],ymm10[1,3],ymm0[6,5],ymm10[5,7]
; AVX-NEXT:    vmovaps 576(%rdi), %ymm1
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm10[1],ymm1[2,3],ymm10[4],ymm1[5,6],ymm10[7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm1[0,3],ymm0[0,2],ymm1[4,7],ymm0[4,6]
; AVX-NEXT:    vmovaps 640(%rdi), %ymm13
; AVX-NEXT:    vperm2f128 {{.*#+}} ymm11 = ymm13[2,3,0,1]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[1,0],ymm11[2,0],ymm13[5,4],ymm11[6,4]
; AVX-NEXT:    vmovups %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0],ymm1[3,0],ymm0[6,4],ymm1[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm2 = ymm1[0,0],ymm0[2,0],ymm1[4,4],ymm0[6,4]
; AVX-NEXT:    vmovaps 112(%rdi), %xmm0
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
; AVX-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
; AVX-NEXT:    # ymm1 = mem[0,1],ymm1[2],mem[3,4],ymm1[5],mem[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,2],ymm0[0,3],ymm1[5,6],ymm0[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm2[5,6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0],ymm14[3,0],ymm0[6,4],ymm14[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm14[0,0],ymm0[2,0],ymm14[4,4],ymm0[6,4]
; AVX-NEXT:    vmovaps 304(%rdi), %xmm2
; AVX-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
; AVX-NEXT:    # ymm1 = mem[0,1],ymm1[2],mem[3,4],ymm1[5],mem[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,2],ymm2[0,3],ymm1[5,6],ymm2[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0],ymm3[3,0],ymm0[6,4],ymm3[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm3[0,0],ymm0[2,0],ymm3[4,4],ymm0[6,4]
; AVX-NEXT:    vmovaps 496(%rdi), %xmm3
; AVX-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
; AVX-NEXT:    # ymm1 = mem[0,1],ymm1[2],mem[3,4],ymm1[5],mem[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,2],ymm3[0,3],ymm1[5,6],ymm3[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0],ymm15[3,0],ymm0[6,4],ymm15[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm15[0,0],ymm0[2,0],ymm15[4,4],ymm0[6,4]
; AVX-NEXT:    vmovaps 688(%rdi), %xmm3
; AVX-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
; AVX-NEXT:    # ymm1 = mem[0,1],ymm1[2],mem[3,4],ymm1[5],mem[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,2],ymm3[0,3],ymm1[5,6],ymm3[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
; AVX-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm4[2,0],ymm1[3,0],ymm4[6,4],ymm1[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0],ymm0[2,0],ymm1[4,4],ymm0[6,4]
; AVX-NEXT:    vmovaps 16(%rdi), %xmm4
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX-NEXT:    vblendps {{.*#+}} ymm2 = ymm0[0,1],ymm7[2],ymm0[3,4],ymm7[5],ymm0[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[1,2],ymm4[0,3],ymm2[5,6],ymm4[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
; AVX-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[2,0],ymm2[3,0],ymm5[6,4],ymm2[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,0],ymm1[2,0],ymm2[4,4],ymm1[6,4]
; AVX-NEXT:    vmovaps 208(%rdi), %xmm5
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0,1],ymm6[2],ymm1[3,4],ymm6[5],ymm1[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[1,2],ymm5[0,3],ymm3[5,6],ymm5[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
; AVX-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm2 = ymm8[2,0],ymm9[3,0],ymm8[6,4],ymm9[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm3 = ymm9[0,0],ymm2[2,0],ymm9[4,4],ymm2[6,4]
; AVX-NEXT:    vmovaps 400(%rdi), %xmm8
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX-NEXT:    vblendps {{.*#+}} ymm15 = ymm2[0,1],ymm12[2],ymm2[3,4],ymm12[5],ymm2[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm15 = ymm15[1,2],ymm8[0,3],ymm15[5,6],ymm8[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm15 = ymm15[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm3 = ymm15[0,1,2,3,4],ymm3[5,6,7]
; AVX-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vshufps {{.*#+}} ymm3 = ymm13[2,0],ymm11[3,0],ymm13[6,4],ymm11[7,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm15 = ymm11[0,0],ymm3[2,0],ymm11[4,4],ymm3[6,4]
; AVX-NEXT:    vmovaps 592(%rdi), %xmm9
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX-NEXT:    vblendps {{.*#+}} ymm14 = ymm3[0,1],ymm10[2],ymm3[3,4],ymm10[5],ymm3[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[1,2],ymm9[0,3],ymm14[5,6],ymm9[4,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[0,2,3,1,4,6,7,5]
; AVX-NEXT:    vblendps {{.*#+}} ymm13 = ymm14[0,1,2,3,4],ymm15[5,6,7]
; AVX-NEXT:    vmovups %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX-NEXT:    vmovups (%rsp), %ymm15 # 32-byte Reload
; AVX-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm14 # 32-byte Folded Reload
; AVX-NEXT:    # ymm14 = ymm15[0,1],mem[2],ymm15[3,4],mem[5],ymm15[6,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm14 = ymm13[1,0],ymm14[2,0],ymm13[5,4],ymm14[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[2,0],ymm15[0,3],ymm14[6,4],ymm15[4,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm15 # 32-byte Folded Reload
; AVX-NEXT:    # ymm15 = ymm15[0,1],mem[0,3],ymm15[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm13 = ymm14[0,1,2,3,4],ymm15[5,6,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm14 = ymm7[0,1],ymm0[2],ymm7[3,4],ymm0[5],ymm7[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm4[1,0],ymm14[2,0],ymm4[5,4],ymm14[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0],ymm7[0,3],ymm0[6,4],ymm7[4,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm7 # 32-byte Folded Reload
; AVX-NEXT:    # ymm7 = ymm4[0,1],mem[0,3],ymm4[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm7 = ymm0[0,1,2,3,4],ymm7[5,6,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
; AVX-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm0 # 32-byte Folded Reload
; AVX-NEXT:    # ymm0 = ymm14[0,1],mem[2],ymm14[3,4],mem[5],ymm14[6,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm4[1,0],ymm0[2,0],ymm4[5,4],ymm0[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0],ymm14[0,3],ymm0[6,4],ymm14[4,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm14 # 32-byte Folded Reload
; AVX-NEXT:    # ymm14 = ymm4[0,1],mem[0,3],ymm4[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm14 = ymm0[0,1,2,3,4],ymm14[5,6,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1],ymm1[2],ymm6[3,4],ymm1[5],ymm6[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm5[1,0],ymm0[2,0],ymm5[5,4],ymm0[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0],ymm6[0,3],ymm0[6,4],ymm6[4,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
; AVX-NEXT:    # ymm1 = ymm1[0,1],mem[0,3],ymm1[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
; AVX-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm1 # 32-byte Folded Reload
; AVX-NEXT:    # ymm1 = ymm6[0,1],mem[2],ymm6[3,4],mem[5],ymm6[6,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm4[1,0],ymm1[2,0],ymm4[5,4],ymm1[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0],ymm6[0,3],ymm1[6,4],ymm6[4,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
; AVX-NEXT:    # ymm6 = ymm4[0,1],mem[0,3],ymm4[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm6[5,6,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm6 = ymm12[0,1],ymm2[2],ymm12[3,4],ymm2[5],ymm12[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm2 = ymm8[1,0],ymm6[2,0],ymm8[5,4],ymm6[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[2,0],ymm12[0,3],ymm2[6,4],ymm12[4,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm5 # 32-byte Folded Reload
; AVX-NEXT:    # ymm5 = ymm4[0,1],mem[0,3],ymm4[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm5[5,6,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
; AVX-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm5 # 32-byte Folded Reload
; AVX-NEXT:    # ymm5 = ymm6[0,1],mem[2],ymm6[3,4],mem[5],ymm6[6,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vshufps {{.*#+}} ymm5 = ymm4[1,0],ymm5[2,0],ymm4[5,4],ymm5[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[2,0],ymm6[0,3],ymm5[6,4],ymm6[4,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
; AVX-NEXT:    # ymm6 = ymm4[0,1],mem[0,3],ymm4[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm6 = ymm10[0,1],ymm3[2],ymm10[3,4],ymm3[5],ymm10[6,7]
; AVX-NEXT:    vshufps {{.*#+}} ymm3 = ymm9[1,0],ymm6[2,0],ymm9[5,4],ymm6[6,4]
; AVX-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[2,0],ymm10[0,3],ymm3[6,4],ymm10[4,7]
; AVX-NEXT:    vshufps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm4 # 32-byte Folded Reload
; AVX-NEXT:    # ymm4 = ymm11[0,1],mem[0,3],ymm11[4,5],mem[4,7]
; AVX-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7]
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 192(%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 128(%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 64(%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, (%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 224(%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 160(%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 96(%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 32(%rsi)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 192(%rdx)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 128(%rdx)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 64(%rdx)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, (%rdx)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 224(%rdx)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 160(%rdx)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 96(%rdx)
; AVX-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX-NEXT:    vmovaps %ymm4, 32(%rdx)
; AVX-NEXT:    vmovaps %ymm3, 192(%rcx)
; AVX-NEXT:    vmovaps %ymm5, 224(%rcx)
; AVX-NEXT:    vmovaps %ymm2, 128(%rcx)
; AVX-NEXT:    vmovaps %ymm1, 160(%rcx)
; AVX-NEXT:    vmovaps %ymm0, 64(%rcx)
; AVX-NEXT:    vmovaps %ymm14, 96(%rcx)
; AVX-NEXT:    vmovaps %ymm7, (%rcx)
; AVX-NEXT:    vmovaps %ymm13, 32(%rcx)
; AVX-NEXT:    addq $1384, %rsp # imm = 0x568
; AVX-NEXT:    vzeroupper
; AVX-NEXT:    retq
;
; AVX2-LABEL: load_i32_stride3_vf64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    subq $1032, %rsp # imm = 0x408
; AVX2-NEXT:    vmovaps 736(%rdi), %ymm2
; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 704(%rdi), %ymm3
; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 672(%rdi), %ymm4
; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 544(%rdi), %ymm5
; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 512(%rdi), %ymm6
; AVX2-NEXT:    vmovups %ymm6, (%rsp) # 32-byte Spill
; AVX2-NEXT:    vmovaps 480(%rdi), %ymm7
; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 352(%rdi), %ymm8
; AVX2-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 320(%rdi), %ymm10
; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 288(%rdi), %ymm11
; AVX2-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 160(%rdi), %ymm9
; AVX2-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 128(%rdi), %ymm0
; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 96(%rdi), %ymm1
; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6],ymm0[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1],ymm9[2],ymm0[3,4],ymm9[5],ymm0[6,7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm0 = [0,3,6,1,4,7,2,5]
; AVX2-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm11[0],ymm10[1],ymm11[2,3],ymm10[4],ymm11[5,6],ymm10[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm8[2],ymm1[3,4],ymm8[5],ymm1[6,7]
; AVX2-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0],ymm6[1],ymm7[2,3],ymm6[4],ymm7[5,6],ymm6[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm5[2],ymm1[3,4],ymm5[5],ymm1[6,7]
; AVX2-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
; AVX2-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps (%rdi), %ymm1
; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 32(%rdi), %ymm3
; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 64(%rdi), %ymm8
; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6],ymm3[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm8[2],ymm4[3,4],ymm8[5],ymm4[6,7]
; AVX2-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vpermps %ymm4, %ymm0, %ymm4
; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 256(%rdi), %ymm1
; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 224(%rdi), %ymm2
; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 192(%rdi), %ymm3
; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6],ymm2[7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm1[2],ymm6[3,4],ymm1[5],ymm6[6,7]
; AVX2-NEXT:    vpermps %ymm6, %ymm0, %ymm6
; AVX2-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 448(%rdi), %ymm13
; AVX2-NEXT:    vmovaps 416(%rdi), %ymm12
; AVX2-NEXT:    vmovaps 384(%rdi), %ymm14
; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm14[0],ymm12[1],ymm14[2,3],ymm12[4],ymm14[5,6],ymm12[7]
; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1],ymm13[2],ymm10[3,4],ymm13[5],ymm10[6,7]
; AVX2-NEXT:    vpermps %ymm10, %ymm0, %ymm10
; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovaps 640(%rdi), %ymm6
; AVX2-NEXT:    vmovaps 608(%rdi), %ymm5
; AVX2-NEXT:    vmovaps 576(%rdi), %ymm7
; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm7[0],ymm5[1],ymm7[2,3],ymm5[4],ymm7[5,6],ymm5[7]
; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm6[2],ymm15[3,4],ymm6[5],ymm15[6,7]
; AVX2-NEXT:    vpermps %ymm15, %ymm0, %ymm0
; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1],ymm4[2],ymm9[3,4],ymm4[5],ymm9[6,7]
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0],ymm0[1,2],ymm11[3],ymm0[4,5],ymm11[6],ymm0[7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm15 = [1,4,7,2,5,0,3,6]
; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
; AVX2-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
; AVX2-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
; AVX2-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0],ymm0[1,2],ymm8[3],ymm0[4,5],ymm8[6],ymm0[7]
; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0,1],ymm3[2],ymm8[3,4],ymm3[5],ymm8[6,7]
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0],ymm0[1,2],ymm10[3],ymm0[4,5],ymm10[6],ymm0[7]
; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1],ymm12[2],ymm14[3,4],ymm12[5],ymm14[6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm13[0],ymm0[1,2],ymm13[3],ymm0[4,5],ymm13[6],ymm0[7]
; AVX2-NEXT:    vmovaps %ymm13, %ymm14
; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm13
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm7[0,1],ymm5[2],ymm7[3,4],ymm5[5],ymm7[6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0],ymm0[1,2],ymm6[3],ymm0[4,5],ymm6[6],ymm0[7]
; AVX2-NEXT:    vmovaps %ymm6, %ymm7
; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm15
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm4[0,1],ymm9[2],ymm4[3,4],ymm9[5],ymm4[6,7]
; AVX2-NEXT:    vmovaps {{.*#+}} ymm12 = [2,5,0,3,6,u,u,u]
; AVX2-NEXT:    vpermps %ymm0, %ymm12, %ymm0
; AVX2-NEXT:    vshufps {{.*#+}} ymm9 = ymm11[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm9 = ymm9[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm0[0,1,2,3,4],ymm9[5,6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7]
; AVX2-NEXT:    vpermps %ymm0, %ymm12, %ymm0
; AVX2-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm1 = mem[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
; AVX2-NEXT:    vpermps %ymm1, %ymm12, %ymm1
; AVX2-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm2 = mem[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm8[2],ymm3[3,4],ymm8[5],ymm3[6,7]
; AVX2-NEXT:    vpermps %ymm2, %ymm12, %ymm2
; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm10[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
; AVX2-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
; AVX2-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm3 = ymm3[0,1],mem[2],ymm3[3,4],mem[5],ymm3[6,7]
; AVX2-NEXT:    vpermps %ymm3, %ymm12, %ymm3
; AVX2-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm4 = mem[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7]
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm4 = mem[0,1],ymm4[2],mem[3,4],ymm4[5],mem[6,7]
; AVX2-NEXT:    vpermps %ymm4, %ymm12, %ymm4
; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm14[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7]
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm5 = ymm5[0,1],mem[2],ymm5[3,4],mem[5],ymm5[6,7]
; AVX2-NEXT:    vpermps %ymm5, %ymm12, %ymm5
; AVX2-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm6 = mem[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7]
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
; AVX2-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
; AVX2-NEXT:    # ymm6 = ymm6[0,1],mem[2],ymm6[3,4],mem[5],ymm6[6,7]
; AVX2-NEXT:    vpermps %ymm6, %ymm12, %ymm6
; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[0,1,0,3,4,5,4,7]
; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5,6,7]
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 192(%rsi)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 128(%rsi)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 64(%rsi)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, (%rsi)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 224(%rsi)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 160(%rsi)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 96(%rsi)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 32(%rsi)
; AVX2-NEXT:    vmovaps %ymm15, 192(%rdx)
; AVX2-NEXT:    vmovaps %ymm13, 128(%rdx)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 64(%rdx)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, (%rdx)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 224(%rdx)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 160(%rdx)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 96(%rdx)
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-NEXT:    vmovaps %ymm7, 32(%rdx)
; AVX2-NEXT:    vmovaps %ymm6, 192(%rcx)
; AVX2-NEXT:    vmovaps %ymm5, 224(%rcx)
; AVX2-NEXT:    vmovaps %ymm4, 128(%rcx)
; AVX2-NEXT:    vmovaps %ymm3, 160(%rcx)
; AVX2-NEXT:    vmovaps %ymm2, 64(%rcx)
; AVX2-NEXT:    vmovaps %ymm1, 96(%rcx)
; AVX2-NEXT:    vmovaps %ymm0, (%rcx)
; AVX2-NEXT:    vmovaps %ymm9, 32(%rcx)
; AVX2-NEXT:    addq $1032, %rsp # imm = 0x408
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
;
; AVX2-FP-LABEL: load_i32_stride3_vf64:
; AVX2-FP:       # %bb.0:
; AVX2-FP-NEXT:    subq $1032, %rsp # imm = 0x408
; AVX2-FP-NEXT:    vmovaps 736(%rdi), %ymm2
; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 704(%rdi), %ymm3
; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 672(%rdi), %ymm4
; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 544(%rdi), %ymm5
; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 512(%rdi), %ymm6
; AVX2-FP-NEXT:    vmovups %ymm6, (%rsp) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 480(%rdi), %ymm7
; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 352(%rdi), %ymm8
; AVX2-FP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm10
; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 288(%rdi), %ymm11
; AVX2-FP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm9
; AVX2-FP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm0
; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 96(%rdi), %ymm1
; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6],ymm0[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1],ymm9[2],ymm0[3,4],ymm9[5],ymm0[6,7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm0 = [0,3,6,1,4,7,2,5]
; AVX2-FP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm11[0],ymm10[1],ymm11[2,3],ymm10[4],ymm11[5,6],ymm10[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm8[2],ymm1[3,4],ymm8[5],ymm1[6,7]
; AVX2-FP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0],ymm6[1],ymm7[2,3],ymm6[4],ymm7[5,6],ymm6[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm5[2],ymm1[3,4],ymm5[5],ymm1[6,7]
; AVX2-FP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
; AVX2-FP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm1
; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm3
; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm8
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6],ymm3[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm8[2],ymm4[3,4],ymm8[5],ymm4[6,7]
; AVX2-FP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vpermps %ymm4, %ymm0, %ymm4
; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm1
; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm2
; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm3
; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6],ymm2[7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm1[2],ymm6[3,4],ymm1[5],ymm6[6,7]
; AVX2-FP-NEXT:    vpermps %ymm6, %ymm0, %ymm6
; AVX2-FP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 448(%rdi), %ymm13
; AVX2-FP-NEXT:    vmovaps 416(%rdi), %ymm12
; AVX2-FP-NEXT:    vmovaps 384(%rdi), %ymm14
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm14[0],ymm12[1],ymm14[2,3],ymm12[4],ymm14[5,6],ymm12[7]
; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1],ymm13[2],ymm10[3,4],ymm13[5],ymm10[6,7]
; AVX2-FP-NEXT:    vpermps %ymm10, %ymm0, %ymm10
; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovaps 640(%rdi), %ymm6
; AVX2-FP-NEXT:    vmovaps 608(%rdi), %ymm5
; AVX2-FP-NEXT:    vmovaps 576(%rdi), %ymm7
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm7[0],ymm5[1],ymm7[2,3],ymm5[4],ymm7[5,6],ymm5[7]
; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm6[2],ymm15[3,4],ymm6[5],ymm15[6,7]
; AVX2-FP-NEXT:    vpermps %ymm15, %ymm0, %ymm0
; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1],ymm4[2],ymm9[3,4],ymm4[5],ymm9[6,7]
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0],ymm0[1,2],ymm11[3],ymm0[4,5],ymm11[6],ymm0[7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm15 = [1,4,7,2,5,0,3,6]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
; AVX2-FP-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
; AVX2-FP-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
; AVX2-FP-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0],ymm0[1,2],ymm8[3],ymm0[4,5],ymm8[6],ymm0[7]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0,1],ymm3[2],ymm8[3,4],ymm3[5],ymm8[6,7]
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0],ymm0[1,2],ymm10[3],ymm0[4,5],ymm10[6],ymm0[7]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1],ymm12[2],ymm14[3,4],ymm12[5],ymm14[6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm13[0],ymm0[1,2],ymm13[3],ymm0[4,5],ymm13[6],ymm0[7]
; AVX2-FP-NEXT:    vmovaps %ymm13, %ymm14
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm13
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm7[0,1],ymm5[2],ymm7[3,4],ymm5[5],ymm7[6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0],ymm0[1,2],ymm6[3],ymm0[4,5],ymm6[6],ymm0[7]
; AVX2-FP-NEXT:    vmovaps %ymm6, %ymm7
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm15
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm4[0,1],ymm9[2],ymm4[3,4],ymm9[5],ymm4[6,7]
; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm12 = [2,5,0,3,6,u,u,u]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm12, %ymm0
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm9 = ymm11[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm9 = ymm9[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm0[0,1,2,3,4],ymm9[5,6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7]
; AVX2-FP-NEXT:    vpermps %ymm0, %ymm12, %ymm0
; AVX2-FP-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm1 = mem[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
; AVX2-FP-NEXT:    vpermps %ymm1, %ymm12, %ymm1
; AVX2-FP-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm2 = mem[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm8[2],ymm3[3,4],ymm8[5],ymm3[6,7]
; AVX2-FP-NEXT:    vpermps %ymm2, %ymm12, %ymm2
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm10[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
; AVX2-FP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm3 = ymm3[0,1],mem[2],ymm3[3,4],mem[5],ymm3[6,7]
; AVX2-FP-NEXT:    vpermps %ymm3, %ymm12, %ymm3
; AVX2-FP-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm4 = mem[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7]
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm4 = mem[0,1],ymm4[2],mem[3,4],ymm4[5],mem[6,7]
; AVX2-FP-NEXT:    vpermps %ymm4, %ymm12, %ymm4
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm14[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7]
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm5 = ymm5[0,1],mem[2],ymm5[3,4],mem[5],ymm5[6,7]
; AVX2-FP-NEXT:    vpermps %ymm5, %ymm12, %ymm5
; AVX2-FP-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm6 = mem[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7]
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
; AVX2-FP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
; AVX2-FP-NEXT:    # ymm6 = ymm6[0,1],mem[2],ymm6[3,4],mem[5],ymm6[6,7]
; AVX2-FP-NEXT:    vpermps %ymm6, %ymm12, %ymm6
; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[0,1,0,3,4,5,4,7]
; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5,6,7]
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 192(%rsi)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 128(%rsi)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 64(%rsi)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, (%rsi)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 224(%rsi)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 160(%rsi)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 96(%rsi)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%rsi)
; AVX2-FP-NEXT:    vmovaps %ymm15, 192(%rdx)
; AVX2-FP-NEXT:    vmovaps %ymm13, 128(%rdx)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 64(%rdx)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, (%rdx)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 224(%rdx)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 160(%rdx)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 96(%rdx)
; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%rdx)
; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm5, 224(%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm4, 128(%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm3, 160(%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm2, 64(%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm1, 96(%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm0, (%rcx)
; AVX2-FP-NEXT:    vmovaps %ymm9, 32(%rcx)
; AVX2-FP-NEXT:    addq $1032, %rsp # imm = 0x408
; AVX2-FP-NEXT:    vzeroupper
; AVX2-FP-NEXT:    retq
;
; AVX2-FCP-LABEL: load_i32_stride3_vf64:
; AVX2-FCP:       # %bb.0:
; AVX2-FCP-NEXT:    subq $1032, %rsp # imm = 0x408
; AVX2-FCP-NEXT:    vmovaps 736(%rdi), %ymm2
; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 704(%rdi), %ymm3
; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 672(%rdi), %ymm4
; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 544(%rdi), %ymm5
; AVX2-FCP-NEXT:    vmovups %ymm5, (%rsp) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 512(%rdi), %ymm6
; AVX2-FCP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 480(%rdi), %ymm7
; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 352(%rdi), %ymm8
; AVX2-FCP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm9
; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm10
; AVX2-FCP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm11
; AVX2-FCP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm14
; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %ymm13
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm13[0],ymm14[1],ymm13[2,3],ymm14[4],ymm13[5,6],ymm14[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1],ymm11[2],ymm0[3,4],ymm11[5],ymm0[6,7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm0 = [0,3,6,1,4,7,2,5]
; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm10[0],ymm9[1],ymm10[2,3],ymm9[4],ymm10[5,6],ymm9[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm8[2],ymm1[3,4],ymm8[5],ymm1[6,7]
; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0],ymm6[1],ymm7[2,3],ymm6[4],ymm7[5,6],ymm6[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm5[2],ymm1[3,4],ymm5[5],ymm1[6,7]
; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm5
; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm1
; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm3
; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0],ymm1[1],ymm5[2,3],ymm1[4],ymm5[5,6],ymm1[7]
; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm3[2],ymm1[3,4],ymm3[5],ymm1[6,7]
; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 256(%rdi), %ymm2
; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm4
; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm1
; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm4[1],ymm1[2,3],ymm4[4],ymm1[5,6],ymm4[7]
; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 448(%rdi), %ymm1
; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 416(%rdi), %ymm2
; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 384(%rdi), %ymm3
; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6],ymm2[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm1[2],ymm3[3,4],ymm1[5],ymm3[6,7]
; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm0, %ymm1
; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps 640(%rdi), %ymm8
; AVX2-FCP-NEXT:    vmovaps 608(%rdi), %ymm7
; AVX2-FCP-NEXT:    vmovaps 576(%rdi), %ymm11
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm11[0],ymm7[1],ymm11[2,3],ymm7[4],ymm11[5,6],ymm7[7]
; AVX2-FCP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm8[2],ymm15[3,4],ymm8[5],ymm15[6,7]
; AVX2-FCP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm0, %ymm0
; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps %ymm14, %ymm10
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm13[0,1],ymm14[2],ymm13[3,4],ymm14[5],ymm13[6,7]
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm12[0],ymm0[1,2],ymm12[3],ymm0[4,5],ymm12[6],ymm0[7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm15 = [1,4,7,2,5,0,3,6]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovaps %ymm9, %ymm14
; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm0 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm9[2],mem[3,4],ymm9[5],mem[6,7]
; AVX2-FCP-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
; AVX2-FCP-NEXT:    vblendps $73, (%rsp), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
; AVX2-FCP-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm5[0,1],ymm1[2],ymm5[3,4],ymm1[5],ymm5[6,7]
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0],ymm0[1,2],ymm2[3],ymm0[4,5],ymm2[6],ymm0[7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7]
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm4[0],ymm0[1,2],ymm4[3],ymm0[4,5],ymm4[6],ymm0[7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1],ymm5[2],ymm6[3,4],ymm5[5],ymm6[6,7]
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0],ymm0[1,2],ymm9[3],ymm0[4,5],ymm9[6],ymm0[7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1],ymm7[2],ymm11[3,4],ymm7[5],ymm11[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0],ymm0[1,2],ymm8[3],ymm0[4,5],ymm8[6],ymm0[7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm11
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0,1],ymm13[2],ymm10[3,4],ymm13[5],ymm10[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm12[1],ymm0[2,3],ymm12[4],ymm0[5,6],ymm12[7]
; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm8 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm8 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm8[0],ymm2[1],ymm8[2,3],ymm2[4],ymm8[5,6],ymm2[7]
; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm8 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm8 = ymm14[0,1],mem[2],ymm14[3,4],mem[5],ymm14[6,7]
; AVX2-FCP-NEXT:    vblendps $146, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm8 = ymm8[0],mem[1],ymm8[2,3],mem[4],ymm8[5,6],mem[7]
; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm2 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm3[2],mem[3,4],ymm3[5],mem[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm4[1],ymm2[2,3],ymm4[4],ymm2[5,6],ymm4[7]
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm7 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm7 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
; AVX2-FCP-NEXT:    vblendps $146, (%rsp), %ymm7, %ymm7 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm7 = ymm7[0],mem[1],ymm7[2,3],mem[4],ymm7[5,6],mem[7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1],ymm6[2],ymm5[3,4],ymm6[5],ymm5[6,7]
; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm9[1],ymm1[2,3],ymm9[4],ymm1[5,6],ymm9[7]
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm6 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm6 = ymm3[0,1],mem[2],ymm3[3,4],mem[5],ymm3[6,7]
; AVX2-FCP-NEXT:    vblendps $146, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm6 = ymm6[0],mem[1],ymm6[2,3],mem[4],ymm6[5,6],mem[7]
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm3 = mem[0,1],ymm3[2],mem[3,4],ymm3[5],mem[6,7]
; AVX2-FCP-NEXT:    vblendps $146, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
; AVX2-FCP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3],mem[4],ymm3[5,6],mem[7]
; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm5 = [2,5,0,3,6,1,4,7]
; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm5, %ymm4
; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm5, %ymm8
; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm2
; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm5, %ymm7
; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm5, %ymm1
; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm5, %ymm6
; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm5, %ymm3
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 192(%rsi)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%rsi)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%rsi)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, (%rsi)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 224(%rsi)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 160(%rsi)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rsi)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%rsi)
; AVX2-FCP-NEXT:    vmovaps %ymm11, 192(%rdx)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%rdx)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%rdx)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, (%rdx)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 224(%rdx)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 160(%rdx)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rdx)
; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%rdx)
; AVX2-FCP-NEXT:    vmovaps %ymm3, 192(%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm6, 224(%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm1, 128(%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm7, 160(%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm2, 64(%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm8, 96(%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rcx)
; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%rcx)
; AVX2-FCP-NEXT:    addq $1032, %rsp # imm = 0x408
; AVX2-FCP-NEXT:    vzeroupper
; AVX2-FCP-NEXT:    retq
;
; AVX512-LABEL: load_i32_stride3_vf64:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vmovdqa64 704(%rdi), %zmm4
; AVX512-NEXT:    vmovdqa64 640(%rdi), %zmm5
; AVX512-NEXT:    vmovdqa64 576(%rdi), %zmm0
; AVX512-NEXT:    vmovdqa64 512(%rdi), %zmm6
; AVX512-NEXT:    vmovdqa64 448(%rdi), %zmm7
; AVX512-NEXT:    vmovdqa64 384(%rdi), %zmm1
; AVX512-NEXT:    vmovdqa64 320(%rdi), %zmm8
; AVX512-NEXT:    vmovdqa64 256(%rdi), %zmm9
; AVX512-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512-NEXT:    vmovdqa64 64(%rdi), %zmm10
; AVX512-NEXT:    vmovdqa64 128(%rdi), %zmm11
; AVX512-NEXT:    vmovdqa64 192(%rdi), %zmm3
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm12 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512-NEXT:    vmovdqa64 %zmm3, %zmm13
; AVX512-NEXT:    vpermt2d %zmm9, %zmm12, %zmm13
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512-NEXT:    vpermt2d %zmm8, %zmm14, %zmm13
; AVX512-NEXT:    vmovdqa64 %zmm1, %zmm15
; AVX512-NEXT:    vpermt2d %zmm7, %zmm12, %zmm15
; AVX512-NEXT:    vpermt2d %zmm6, %zmm14, %zmm15
; AVX512-NEXT:    vmovdqa64 %zmm0, %zmm16
; AVX512-NEXT:    vpermt2d %zmm5, %zmm12, %zmm16
; AVX512-NEXT:    vpermt2d %zmm4, %zmm14, %zmm16
; AVX512-NEXT:    vpermi2d %zmm10, %zmm2, %zmm12
; AVX512-NEXT:    vpermt2d %zmm11, %zmm14, %zmm12
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512-NEXT:    vmovdqa64 %zmm7, %zmm17
; AVX512-NEXT:    vpermt2d %zmm1, %zmm14, %zmm17
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512-NEXT:    vpermt2d %zmm6, %zmm18, %zmm17
; AVX512-NEXT:    vmovdqa64 %zmm9, %zmm19
; AVX512-NEXT:    vpermt2d %zmm3, %zmm14, %zmm19
; AVX512-NEXT:    vpermt2d %zmm8, %zmm18, %zmm19
; AVX512-NEXT:    vmovdqa64 %zmm5, %zmm20
; AVX512-NEXT:    vpermt2d %zmm0, %zmm14, %zmm20
; AVX512-NEXT:    vpermt2d %zmm4, %zmm18, %zmm20
; AVX512-NEXT:    vpermi2d %zmm2, %zmm10, %zmm14
; AVX512-NEXT:    vpermt2d %zmm11, %zmm18, %zmm14
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512-NEXT:    vpermt2d %zmm9, %zmm18, %zmm3
; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm9 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512-NEXT:    vpermt2d %zmm8, %zmm9, %zmm3
; AVX512-NEXT:    vpermt2d %zmm5, %zmm18, %zmm0
; AVX512-NEXT:    vpermt2d %zmm4, %zmm9, %zmm0
; AVX512-NEXT:    vpermt2d %zmm7, %zmm18, %zmm1
; AVX512-NEXT:    vpermt2d %zmm6, %zmm9, %zmm1
; AVX512-NEXT:    vpermt2d %zmm10, %zmm18, %zmm2
; AVX512-NEXT:    vpermt2d %zmm11, %zmm9, %zmm2
; AVX512-NEXT:    vmovdqa64 %zmm16, 192(%rsi)
; AVX512-NEXT:    vmovdqa64 %zmm15, 128(%rsi)
; AVX512-NEXT:    vmovdqa64 %zmm13, 64(%rsi)
; AVX512-NEXT:    vmovdqa64 %zmm12, (%rsi)
; AVX512-NEXT:    vmovdqa64 %zmm20, 192(%rdx)
; AVX512-NEXT:    vmovdqa64 %zmm14, (%rdx)
; AVX512-NEXT:    vmovdqa64 %zmm19, 64(%rdx)
; AVX512-NEXT:    vmovdqa64 %zmm17, 128(%rdx)
; AVX512-NEXT:    vmovdqa64 %zmm1, 128(%rcx)
; AVX512-NEXT:    vmovdqa64 %zmm0, 192(%rcx)
; AVX512-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512-NEXT:    vmovdqa64 %zmm3, 64(%rcx)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
;
; AVX512-FCP-LABEL: load_i32_stride3_vf64:
; AVX512-FCP:       # %bb.0:
; AVX512-FCP-NEXT:    vmovdqa64 704(%rdi), %zmm4
; AVX512-FCP-NEXT:    vmovdqa64 640(%rdi), %zmm5
; AVX512-FCP-NEXT:    vmovdqa64 576(%rdi), %zmm0
; AVX512-FCP-NEXT:    vmovdqa64 512(%rdi), %zmm6
; AVX512-FCP-NEXT:    vmovdqa64 448(%rdi), %zmm7
; AVX512-FCP-NEXT:    vmovdqa64 384(%rdi), %zmm1
; AVX512-FCP-NEXT:    vmovdqa64 320(%rdi), %zmm8
; AVX512-FCP-NEXT:    vmovdqa64 256(%rdi), %zmm9
; AVX512-FCP-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm10
; AVX512-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm11
; AVX512-FCP-NEXT:    vmovdqa64 192(%rdi), %zmm3
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm12 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512-FCP-NEXT:    vmovdqa64 %zmm3, %zmm13
; AVX512-FCP-NEXT:    vpermt2d %zmm9, %zmm12, %zmm13
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512-FCP-NEXT:    vpermt2d %zmm8, %zmm14, %zmm13
; AVX512-FCP-NEXT:    vmovdqa64 %zmm1, %zmm15
; AVX512-FCP-NEXT:    vpermt2d %zmm7, %zmm12, %zmm15
; AVX512-FCP-NEXT:    vpermt2d %zmm6, %zmm14, %zmm15
; AVX512-FCP-NEXT:    vmovdqa64 %zmm0, %zmm16
; AVX512-FCP-NEXT:    vpermt2d %zmm5, %zmm12, %zmm16
; AVX512-FCP-NEXT:    vpermt2d %zmm4, %zmm14, %zmm16
; AVX512-FCP-NEXT:    vpermi2d %zmm10, %zmm2, %zmm12
; AVX512-FCP-NEXT:    vpermt2d %zmm11, %zmm14, %zmm12
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512-FCP-NEXT:    vmovdqa64 %zmm7, %zmm17
; AVX512-FCP-NEXT:    vpermt2d %zmm1, %zmm14, %zmm17
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512-FCP-NEXT:    vpermt2d %zmm6, %zmm18, %zmm17
; AVX512-FCP-NEXT:    vmovdqa64 %zmm9, %zmm19
; AVX512-FCP-NEXT:    vpermt2d %zmm3, %zmm14, %zmm19
; AVX512-FCP-NEXT:    vpermt2d %zmm8, %zmm18, %zmm19
; AVX512-FCP-NEXT:    vmovdqa64 %zmm5, %zmm20
; AVX512-FCP-NEXT:    vpermt2d %zmm0, %zmm14, %zmm20
; AVX512-FCP-NEXT:    vpermt2d %zmm4, %zmm18, %zmm20
; AVX512-FCP-NEXT:    vpermi2d %zmm2, %zmm10, %zmm14
; AVX512-FCP-NEXT:    vpermt2d %zmm11, %zmm18, %zmm14
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512-FCP-NEXT:    vpermt2d %zmm9, %zmm18, %zmm3
; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm9 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512-FCP-NEXT:    vpermt2d %zmm8, %zmm9, %zmm3
; AVX512-FCP-NEXT:    vpermt2d %zmm5, %zmm18, %zmm0
; AVX512-FCP-NEXT:    vpermt2d %zmm4, %zmm9, %zmm0
; AVX512-FCP-NEXT:    vpermt2d %zmm7, %zmm18, %zmm1
; AVX512-FCP-NEXT:    vpermt2d %zmm6, %zmm9, %zmm1
; AVX512-FCP-NEXT:    vpermt2d %zmm10, %zmm18, %zmm2
; AVX512-FCP-NEXT:    vpermt2d %zmm11, %zmm9, %zmm2
; AVX512-FCP-NEXT:    vmovdqa64 %zmm16, 192(%rsi)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm15, 128(%rsi)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm13, 64(%rsi)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm12, (%rsi)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm20, 192(%rdx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm14, (%rdx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm19, 64(%rdx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm17, 128(%rdx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm1, 128(%rcx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm0, 192(%rcx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512-FCP-NEXT:    vmovdqa64 %zmm3, 64(%rcx)
; AVX512-FCP-NEXT:    vzeroupper
; AVX512-FCP-NEXT:    retq
;
; AVX512DQ-LABEL: load_i32_stride3_vf64:
; AVX512DQ:       # %bb.0:
; AVX512DQ-NEXT:    vmovdqa64 704(%rdi), %zmm4
; AVX512DQ-NEXT:    vmovdqa64 640(%rdi), %zmm5
; AVX512DQ-NEXT:    vmovdqa64 576(%rdi), %zmm0
; AVX512DQ-NEXT:    vmovdqa64 512(%rdi), %zmm6
; AVX512DQ-NEXT:    vmovdqa64 448(%rdi), %zmm7
; AVX512DQ-NEXT:    vmovdqa64 384(%rdi), %zmm1
; AVX512DQ-NEXT:    vmovdqa64 320(%rdi), %zmm8
; AVX512DQ-NEXT:    vmovdqa64 256(%rdi), %zmm9
; AVX512DQ-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512DQ-NEXT:    vmovdqa64 64(%rdi), %zmm10
; AVX512DQ-NEXT:    vmovdqa64 128(%rdi), %zmm11
; AVX512DQ-NEXT:    vmovdqa64 192(%rdi), %zmm3
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm12 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-NEXT:    vmovdqa64 %zmm3, %zmm13
; AVX512DQ-NEXT:    vpermt2d %zmm9, %zmm12, %zmm13
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-NEXT:    vpermt2d %zmm8, %zmm14, %zmm13
; AVX512DQ-NEXT:    vmovdqa64 %zmm1, %zmm15
; AVX512DQ-NEXT:    vpermt2d %zmm7, %zmm12, %zmm15
; AVX512DQ-NEXT:    vpermt2d %zmm6, %zmm14, %zmm15
; AVX512DQ-NEXT:    vmovdqa64 %zmm0, %zmm16
; AVX512DQ-NEXT:    vpermt2d %zmm5, %zmm12, %zmm16
; AVX512DQ-NEXT:    vpermt2d %zmm4, %zmm14, %zmm16
; AVX512DQ-NEXT:    vpermi2d %zmm10, %zmm2, %zmm12
; AVX512DQ-NEXT:    vpermt2d %zmm11, %zmm14, %zmm12
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-NEXT:    vmovdqa64 %zmm7, %zmm17
; AVX512DQ-NEXT:    vpermt2d %zmm1, %zmm14, %zmm17
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-NEXT:    vpermt2d %zmm6, %zmm18, %zmm17
; AVX512DQ-NEXT:    vmovdqa64 %zmm9, %zmm19
; AVX512DQ-NEXT:    vpermt2d %zmm3, %zmm14, %zmm19
; AVX512DQ-NEXT:    vpermt2d %zmm8, %zmm18, %zmm19
; AVX512DQ-NEXT:    vmovdqa64 %zmm5, %zmm20
; AVX512DQ-NEXT:    vpermt2d %zmm0, %zmm14, %zmm20
; AVX512DQ-NEXT:    vpermt2d %zmm4, %zmm18, %zmm20
; AVX512DQ-NEXT:    vpermi2d %zmm2, %zmm10, %zmm14
; AVX512DQ-NEXT:    vpermt2d %zmm11, %zmm18, %zmm14
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-NEXT:    vpermt2d %zmm9, %zmm18, %zmm3
; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm9 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-NEXT:    vpermt2d %zmm8, %zmm9, %zmm3
; AVX512DQ-NEXT:    vpermt2d %zmm5, %zmm18, %zmm0
; AVX512DQ-NEXT:    vpermt2d %zmm4, %zmm9, %zmm0
; AVX512DQ-NEXT:    vpermt2d %zmm7, %zmm18, %zmm1
; AVX512DQ-NEXT:    vpermt2d %zmm6, %zmm9, %zmm1
; AVX512DQ-NEXT:    vpermt2d %zmm10, %zmm18, %zmm2
; AVX512DQ-NEXT:    vpermt2d %zmm11, %zmm9, %zmm2
; AVX512DQ-NEXT:    vmovdqa64 %zmm16, 192(%rsi)
; AVX512DQ-NEXT:    vmovdqa64 %zmm15, 128(%rsi)
; AVX512DQ-NEXT:    vmovdqa64 %zmm13, 64(%rsi)
; AVX512DQ-NEXT:    vmovdqa64 %zmm12, (%rsi)
; AVX512DQ-NEXT:    vmovdqa64 %zmm20, 192(%rdx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm14, (%rdx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm19, 64(%rdx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm17, 128(%rdx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm1, 128(%rcx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm0, 192(%rcx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512DQ-NEXT:    vmovdqa64 %zmm3, 64(%rcx)
; AVX512DQ-NEXT:    vzeroupper
; AVX512DQ-NEXT:    retq
;
; AVX512DQ-FCP-LABEL: load_i32_stride3_vf64:
; AVX512DQ-FCP:       # %bb.0:
; AVX512DQ-FCP-NEXT:    vmovdqa64 704(%rdi), %zmm4
; AVX512DQ-FCP-NEXT:    vmovdqa64 640(%rdi), %zmm5
; AVX512DQ-FCP-NEXT:    vmovdqa64 576(%rdi), %zmm0
; AVX512DQ-FCP-NEXT:    vmovdqa64 512(%rdi), %zmm6
; AVX512DQ-FCP-NEXT:    vmovdqa64 448(%rdi), %zmm7
; AVX512DQ-FCP-NEXT:    vmovdqa64 384(%rdi), %zmm1
; AVX512DQ-FCP-NEXT:    vmovdqa64 320(%rdi), %zmm8
; AVX512DQ-FCP-NEXT:    vmovdqa64 256(%rdi), %zmm9
; AVX512DQ-FCP-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512DQ-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm10
; AVX512DQ-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm11
; AVX512DQ-FCP-NEXT:    vmovdqa64 192(%rdi), %zmm3
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm12 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm3, %zmm13
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm9, %zmm12, %zmm13
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm8, %zmm14, %zmm13
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm1, %zmm15
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm7, %zmm12, %zmm15
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm6, %zmm14, %zmm15
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm0, %zmm16
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm5, %zmm12, %zmm16
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm4, %zmm14, %zmm16
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm10, %zmm2, %zmm12
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm11, %zmm14, %zmm12
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm7, %zmm17
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm1, %zmm14, %zmm17
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm6, %zmm18, %zmm17
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm9, %zmm19
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm3, %zmm14, %zmm19
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm8, %zmm18, %zmm19
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm5, %zmm20
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm0, %zmm14, %zmm20
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm4, %zmm18, %zmm20
; AVX512DQ-FCP-NEXT:    vpermi2d %zmm2, %zmm10, %zmm14
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm11, %zmm18, %zmm14
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm9, %zmm18, %zmm3
; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm9 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm8, %zmm9, %zmm3
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm5, %zmm18, %zmm0
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm4, %zmm9, %zmm0
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm7, %zmm18, %zmm1
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm6, %zmm9, %zmm1
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm10, %zmm18, %zmm2
; AVX512DQ-FCP-NEXT:    vpermt2d %zmm11, %zmm9, %zmm2
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm16, 192(%rsi)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm15, 128(%rsi)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm13, 64(%rsi)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm12, (%rsi)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm20, 192(%rdx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm14, (%rdx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm19, 64(%rdx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm17, 128(%rdx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm1, 128(%rcx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm0, 192(%rcx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm3, 64(%rcx)
; AVX512DQ-FCP-NEXT:    vzeroupper
; AVX512DQ-FCP-NEXT:    retq
;
; AVX512BW-LABEL: load_i32_stride3_vf64:
; AVX512BW:       # %bb.0:
; AVX512BW-NEXT:    vmovdqa64 704(%rdi), %zmm4
; AVX512BW-NEXT:    vmovdqa64 640(%rdi), %zmm5
; AVX512BW-NEXT:    vmovdqa64 576(%rdi), %zmm0
; AVX512BW-NEXT:    vmovdqa64 512(%rdi), %zmm6
; AVX512BW-NEXT:    vmovdqa64 448(%rdi), %zmm7
; AVX512BW-NEXT:    vmovdqa64 384(%rdi), %zmm1
; AVX512BW-NEXT:    vmovdqa64 320(%rdi), %zmm8
; AVX512BW-NEXT:    vmovdqa64 256(%rdi), %zmm9
; AVX512BW-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512BW-NEXT:    vmovdqa64 64(%rdi), %zmm10
; AVX512BW-NEXT:    vmovdqa64 128(%rdi), %zmm11
; AVX512BW-NEXT:    vmovdqa64 192(%rdi), %zmm3
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm12 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512BW-NEXT:    vmovdqa64 %zmm3, %zmm13
; AVX512BW-NEXT:    vpermt2d %zmm9, %zmm12, %zmm13
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512BW-NEXT:    vpermt2d %zmm8, %zmm14, %zmm13
; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm15
; AVX512BW-NEXT:    vpermt2d %zmm7, %zmm12, %zmm15
; AVX512BW-NEXT:    vpermt2d %zmm6, %zmm14, %zmm15
; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm16
; AVX512BW-NEXT:    vpermt2d %zmm5, %zmm12, %zmm16
; AVX512BW-NEXT:    vpermt2d %zmm4, %zmm14, %zmm16
; AVX512BW-NEXT:    vpermi2d %zmm10, %zmm2, %zmm12
; AVX512BW-NEXT:    vpermt2d %zmm11, %zmm14, %zmm12
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512BW-NEXT:    vmovdqa64 %zmm7, %zmm17
; AVX512BW-NEXT:    vpermt2d %zmm1, %zmm14, %zmm17
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512BW-NEXT:    vpermt2d %zmm6, %zmm18, %zmm17
; AVX512BW-NEXT:    vmovdqa64 %zmm9, %zmm19
; AVX512BW-NEXT:    vpermt2d %zmm3, %zmm14, %zmm19
; AVX512BW-NEXT:    vpermt2d %zmm8, %zmm18, %zmm19
; AVX512BW-NEXT:    vmovdqa64 %zmm5, %zmm20
; AVX512BW-NEXT:    vpermt2d %zmm0, %zmm14, %zmm20
; AVX512BW-NEXT:    vpermt2d %zmm4, %zmm18, %zmm20
; AVX512BW-NEXT:    vpermi2d %zmm2, %zmm10, %zmm14
; AVX512BW-NEXT:    vpermt2d %zmm11, %zmm18, %zmm14
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512BW-NEXT:    vpermt2d %zmm9, %zmm18, %zmm3
; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm9 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512BW-NEXT:    vpermt2d %zmm8, %zmm9, %zmm3
; AVX512BW-NEXT:    vpermt2d %zmm5, %zmm18, %zmm0
; AVX512BW-NEXT:    vpermt2d %zmm4, %zmm9, %zmm0
; AVX512BW-NEXT:    vpermt2d %zmm7, %zmm18, %zmm1
; AVX512BW-NEXT:    vpermt2d %zmm6, %zmm9, %zmm1
; AVX512BW-NEXT:    vpermt2d %zmm10, %zmm18, %zmm2
; AVX512BW-NEXT:    vpermt2d %zmm11, %zmm9, %zmm2
; AVX512BW-NEXT:    vmovdqa64 %zmm16, 192(%rsi)
; AVX512BW-NEXT:    vmovdqa64 %zmm15, 128(%rsi)
; AVX512BW-NEXT:    vmovdqa64 %zmm13, 64(%rsi)
; AVX512BW-NEXT:    vmovdqa64 %zmm12, (%rsi)
; AVX512BW-NEXT:    vmovdqa64 %zmm20, 192(%rdx)
; AVX512BW-NEXT:    vmovdqa64 %zmm14, (%rdx)
; AVX512BW-NEXT:    vmovdqa64 %zmm19, 64(%rdx)
; AVX512BW-NEXT:    vmovdqa64 %zmm17, 128(%rdx)
; AVX512BW-NEXT:    vmovdqa64 %zmm1, 128(%rcx)
; AVX512BW-NEXT:    vmovdqa64 %zmm0, 192(%rcx)
; AVX512BW-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512BW-NEXT:    vmovdqa64 %zmm3, 64(%rcx)
; AVX512BW-NEXT:    vzeroupper
; AVX512BW-NEXT:    retq
;
; AVX512BW-FCP-LABEL: load_i32_stride3_vf64:
; AVX512BW-FCP:       # %bb.0:
; AVX512BW-FCP-NEXT:    vmovdqa64 704(%rdi), %zmm4
; AVX512BW-FCP-NEXT:    vmovdqa64 640(%rdi), %zmm5
; AVX512BW-FCP-NEXT:    vmovdqa64 576(%rdi), %zmm0
; AVX512BW-FCP-NEXT:    vmovdqa64 512(%rdi), %zmm6
; AVX512BW-FCP-NEXT:    vmovdqa64 448(%rdi), %zmm7
; AVX512BW-FCP-NEXT:    vmovdqa64 384(%rdi), %zmm1
; AVX512BW-FCP-NEXT:    vmovdqa64 320(%rdi), %zmm8
; AVX512BW-FCP-NEXT:    vmovdqa64 256(%rdi), %zmm9
; AVX512BW-FCP-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512BW-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm10
; AVX512BW-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm11
; AVX512BW-FCP-NEXT:    vmovdqa64 192(%rdi), %zmm3
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm12 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm3, %zmm13
; AVX512BW-FCP-NEXT:    vpermt2d %zmm9, %zmm12, %zmm13
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512BW-FCP-NEXT:    vpermt2d %zmm8, %zmm14, %zmm13
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm1, %zmm15
; AVX512BW-FCP-NEXT:    vpermt2d %zmm7, %zmm12, %zmm15
; AVX512BW-FCP-NEXT:    vpermt2d %zmm6, %zmm14, %zmm15
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm0, %zmm16
; AVX512BW-FCP-NEXT:    vpermt2d %zmm5, %zmm12, %zmm16
; AVX512BW-FCP-NEXT:    vpermt2d %zmm4, %zmm14, %zmm16
; AVX512BW-FCP-NEXT:    vpermi2d %zmm10, %zmm2, %zmm12
; AVX512BW-FCP-NEXT:    vpermt2d %zmm11, %zmm14, %zmm12
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm7, %zmm17
; AVX512BW-FCP-NEXT:    vpermt2d %zmm1, %zmm14, %zmm17
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512BW-FCP-NEXT:    vpermt2d %zmm6, %zmm18, %zmm17
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm9, %zmm19
; AVX512BW-FCP-NEXT:    vpermt2d %zmm3, %zmm14, %zmm19
; AVX512BW-FCP-NEXT:    vpermt2d %zmm8, %zmm18, %zmm19
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm5, %zmm20
; AVX512BW-FCP-NEXT:    vpermt2d %zmm0, %zmm14, %zmm20
; AVX512BW-FCP-NEXT:    vpermt2d %zmm4, %zmm18, %zmm20
; AVX512BW-FCP-NEXT:    vpermi2d %zmm2, %zmm10, %zmm14
; AVX512BW-FCP-NEXT:    vpermt2d %zmm11, %zmm18, %zmm14
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512BW-FCP-NEXT:    vpermt2d %zmm9, %zmm18, %zmm3
; AVX512BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm9 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512BW-FCP-NEXT:    vpermt2d %zmm8, %zmm9, %zmm3
; AVX512BW-FCP-NEXT:    vpermt2d %zmm5, %zmm18, %zmm0
; AVX512BW-FCP-NEXT:    vpermt2d %zmm4, %zmm9, %zmm0
; AVX512BW-FCP-NEXT:    vpermt2d %zmm7, %zmm18, %zmm1
; AVX512BW-FCP-NEXT:    vpermt2d %zmm6, %zmm9, %zmm1
; AVX512BW-FCP-NEXT:    vpermt2d %zmm10, %zmm18, %zmm2
; AVX512BW-FCP-NEXT:    vpermt2d %zmm11, %zmm9, %zmm2
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm16, 192(%rsi)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm15, 128(%rsi)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm13, 64(%rsi)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm12, (%rsi)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm20, 192(%rdx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm14, (%rdx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm19, 64(%rdx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm17, 128(%rdx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm1, 128(%rcx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm0, 192(%rcx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm3, 64(%rcx)
; AVX512BW-FCP-NEXT:    vzeroupper
; AVX512BW-FCP-NEXT:    retq
;
; AVX512DQ-BW-LABEL: load_i32_stride3_vf64:
; AVX512DQ-BW:       # %bb.0:
; AVX512DQ-BW-NEXT:    vmovdqa64 704(%rdi), %zmm4
; AVX512DQ-BW-NEXT:    vmovdqa64 640(%rdi), %zmm5
; AVX512DQ-BW-NEXT:    vmovdqa64 576(%rdi), %zmm0
; AVX512DQ-BW-NEXT:    vmovdqa64 512(%rdi), %zmm6
; AVX512DQ-BW-NEXT:    vmovdqa64 448(%rdi), %zmm7
; AVX512DQ-BW-NEXT:    vmovdqa64 384(%rdi), %zmm1
; AVX512DQ-BW-NEXT:    vmovdqa64 320(%rdi), %zmm8
; AVX512DQ-BW-NEXT:    vmovdqa64 256(%rdi), %zmm9
; AVX512DQ-BW-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512DQ-BW-NEXT:    vmovdqa64 64(%rdi), %zmm10
; AVX512DQ-BW-NEXT:    vmovdqa64 128(%rdi), %zmm11
; AVX512DQ-BW-NEXT:    vmovdqa64 192(%rdi), %zmm3
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm12 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm3, %zmm13
; AVX512DQ-BW-NEXT:    vpermt2d %zmm9, %zmm12, %zmm13
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-BW-NEXT:    vpermt2d %zmm8, %zmm14, %zmm13
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm1, %zmm15
; AVX512DQ-BW-NEXT:    vpermt2d %zmm7, %zmm12, %zmm15
; AVX512DQ-BW-NEXT:    vpermt2d %zmm6, %zmm14, %zmm15
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm0, %zmm16
; AVX512DQ-BW-NEXT:    vpermt2d %zmm5, %zmm12, %zmm16
; AVX512DQ-BW-NEXT:    vpermt2d %zmm4, %zmm14, %zmm16
; AVX512DQ-BW-NEXT:    vpermi2d %zmm10, %zmm2, %zmm12
; AVX512DQ-BW-NEXT:    vpermt2d %zmm11, %zmm14, %zmm12
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm7, %zmm17
; AVX512DQ-BW-NEXT:    vpermt2d %zmm1, %zmm14, %zmm17
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-BW-NEXT:    vpermt2d %zmm6, %zmm18, %zmm17
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm9, %zmm19
; AVX512DQ-BW-NEXT:    vpermt2d %zmm3, %zmm14, %zmm19
; AVX512DQ-BW-NEXT:    vpermt2d %zmm8, %zmm18, %zmm19
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm5, %zmm20
; AVX512DQ-BW-NEXT:    vpermt2d %zmm0, %zmm14, %zmm20
; AVX512DQ-BW-NEXT:    vpermt2d %zmm4, %zmm18, %zmm20
; AVX512DQ-BW-NEXT:    vpermi2d %zmm2, %zmm10, %zmm14
; AVX512DQ-BW-NEXT:    vpermt2d %zmm11, %zmm18, %zmm14
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-BW-NEXT:    vpermt2d %zmm9, %zmm18, %zmm3
; AVX512DQ-BW-NEXT:    vpmovsxbd {{.*#+}} zmm9 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-BW-NEXT:    vpermt2d %zmm8, %zmm9, %zmm3
; AVX512DQ-BW-NEXT:    vpermt2d %zmm5, %zmm18, %zmm0
; AVX512DQ-BW-NEXT:    vpermt2d %zmm4, %zmm9, %zmm0
; AVX512DQ-BW-NEXT:    vpermt2d %zmm7, %zmm18, %zmm1
; AVX512DQ-BW-NEXT:    vpermt2d %zmm6, %zmm9, %zmm1
; AVX512DQ-BW-NEXT:    vpermt2d %zmm10, %zmm18, %zmm2
; AVX512DQ-BW-NEXT:    vpermt2d %zmm11, %zmm9, %zmm2
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm16, 192(%rsi)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm15, 128(%rsi)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm13, 64(%rsi)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm12, (%rsi)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm20, 192(%rdx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm14, (%rdx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm19, 64(%rdx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm17, 128(%rdx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm1, 128(%rcx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm0, 192(%rcx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm3, 64(%rcx)
; AVX512DQ-BW-NEXT:    vzeroupper
; AVX512DQ-BW-NEXT:    retq
;
; AVX512DQ-BW-FCP-LABEL: load_i32_stride3_vf64:
; AVX512DQ-BW-FCP:       # %bb.0:
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 704(%rdi), %zmm4
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 640(%rdi), %zmm5
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 576(%rdi), %zmm0
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 512(%rdi), %zmm6
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 448(%rdi), %zmm7
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 384(%rdi), %zmm1
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 320(%rdi), %zmm8
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 256(%rdi), %zmm9
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 (%rdi), %zmm2
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 64(%rdi), %zmm10
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 128(%rdi), %zmm11
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 192(%rdi), %zmm3
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm12 = [0,3,6,9,12,15,18,21,24,27,30,0,0,0,0,0]
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm3, %zmm13
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm9, %zmm12, %zmm13
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [0,1,2,3,4,5,6,7,8,9,10,17,20,23,26,29]
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm8, %zmm14, %zmm13
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm1, %zmm15
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm7, %zmm12, %zmm15
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm6, %zmm14, %zmm15
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm0, %zmm16
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm5, %zmm12, %zmm16
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm4, %zmm14, %zmm16
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm10, %zmm2, %zmm12
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm11, %zmm14, %zmm12
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm14 = [17,20,23,26,29,0,3,6,9,12,15,0,0,0,0,0]
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm7, %zmm17
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm1, %zmm14, %zmm17
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [0,1,2,3,4,5,6,7,8,9,10,18,21,24,27,30]
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm6, %zmm18, %zmm17
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm9, %zmm19
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm3, %zmm14, %zmm19
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm8, %zmm18, %zmm19
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm5, %zmm20
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm0, %zmm14, %zmm20
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm4, %zmm18, %zmm20
; AVX512DQ-BW-FCP-NEXT:    vpermi2d %zmm2, %zmm10, %zmm14
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm11, %zmm18, %zmm14
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [2,5,8,11,14,17,20,23,26,29,0,0,0,0,0,0]
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm9, %zmm18, %zmm3
; AVX512DQ-BW-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm9 = [0,1,2,3,4,5,6,7,8,9,16,19,22,25,28,31]
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm8, %zmm9, %zmm3
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm5, %zmm18, %zmm0
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm4, %zmm9, %zmm0
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm7, %zmm18, %zmm1
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm6, %zmm9, %zmm1
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm10, %zmm18, %zmm2
; AVX512DQ-BW-FCP-NEXT:    vpermt2d %zmm11, %zmm9, %zmm2
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm16, 192(%rsi)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm15, 128(%rsi)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm13, 64(%rsi)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm12, (%rsi)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm20, 192(%rdx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm14, (%rdx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm19, 64(%rdx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm17, 128(%rdx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm1, 128(%rcx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm0, 192(%rcx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm2, (%rcx)
; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm3, 64(%rcx)
; AVX512DQ-BW-FCP-NEXT:    vzeroupper
; AVX512DQ-BW-FCP-NEXT:    retq
  %wide.vec = load <192 x i32>, ptr %in.vec, align 64
  %strided.vec0 = shufflevector <192 x i32> %wide.vec, <192 x i32> poison, <64 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45, i32 48, i32 51, i32 54, i32 57, i32 60, i32 63, i32 66, i32 69, i32 72, i32 75, i32 78, i32 81, i32 84, i32 87, i32 90, i32 93, i32 96, i32 99, i32 102, i32 105, i32 108, i32 111, i32 114, i32 117, i32 120, i32 123, i32 126, i32 129, i32 132, i32 135, i32 138, i32 141, i32 144, i32 147, i32 150, i32 153, i32 156, i32 159, i32 162, i32 165, i32 168, i32 171, i32 174, i32 177, i32 180, i32 183, i32 186, i32 189>
  %strided.vec1 = shufflevector <192 x i32> %wide.vec, <192 x i32> poison, <64 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46, i32 49, i32 52, i32 55, i32 58, i32 61, i32 64, i32 67, i32 70, i32 73, i32 76, i32 79, i32 82, i32 85, i32 88, i32 91, i32 94, i32 97, i32 100, i32 103, i32 106, i32 109, i32 112, i32 115, i32 118, i32 121, i32 124, i32 127, i32 130, i32 133, i32 136, i32 139, i32 142, i32 145, i32 148, i32 151, i32 154, i32 157, i32 160, i32 163, i32 166, i32 169, i32 172, i32 175, i32 178, i32 181, i32 184, i32 187, i32 190>
  %strided.vec2 = shufflevector <192 x i32> %wide.vec, <192 x i32> poison, <64 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47, i32 50, i32 53, i32 56, i32 59, i32 62, i32 65, i32 68, i32 71, i32 74, i32 77, i32 80, i32 83, i32 86, i32 89, i32 92, i32 95, i32 98, i32 101, i32 104, i32 107, i32 110, i32 113, i32 116, i32 119, i32 122, i32 125, i32 128, i32 131, i32 134, i32 137, i32 140, i32 143, i32 146, i32 149, i32 152, i32 155, i32 158, i32 161, i32 164, i32 167, i32 170, i32 173, i32 176, i32 179, i32 182, i32 185, i32 188, i32 191>
  store <64 x i32> %strided.vec0, ptr %out.vec0, align 64
  store <64 x i32> %strided.vec1, ptr %out.vec1, align 64
  store <64 x i32> %strided.vec2, ptr %out.vec2, align 64
  ret void
}