[llvm] 2c0b888 - [X86] combineX86ShuffleChain - prefer combining to X86ISD::SHUF128 if PERMQ operands are splittable (#133900)

via llvm-commits llvm-commits at lists.llvm.org
Tue Apr 1 05:47:56 PDT 2025


Author: Simon Pilgrim
Date: 2025-04-01T13:47:52+01:00
New Revision: 2c0b888359c6c5976054bb423ba1d7b37bae9f1a

URL: https://github.com/llvm/llvm-project/commit/2c0b888359c6c5976054bb423ba1d7b37bae9f1a
DIFF: https://github.com/llvm/llvm-project/commit/2c0b888359c6c5976054bb423ba1d7b37bae9f1a.diff

LOG: [X86] combineX86ShuffleChain - prefer combining to X86ISD::SHUF128 if PERMQ operands are splittable (#133900)

If the 512-bit unary shuffle is a concatenation of 128/256-bit subvectors then we're better off using a X86ISD::SHUF128 node so we can fold the concatenation into the shuffle as well.

Added: 
    

Modified: 
    llvm/lib/Target/X86/X86ISelLowering.cpp
    llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-7.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index b1745e5a30d7b..78762774f65e8 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -39797,7 +39797,8 @@ static SDValue combineX86ShuffleChain(
     // FIXME: Is there a better way to do this? is256BitLaneRepeatedShuffleMask
     // doesn't work because our mask is for 128 bits and we don't have an MVT
     // to match that.
-    bool PreferPERMQ = UnaryShuffle && isUndefOrInRange(ScaledMask[0], 0, 2) &&
+    bool PreferPERMQ = UnaryShuffle && !isFreeToSplitVector(V1, DAG) &&
+                       isUndefOrInRange(ScaledMask[0], 0, 2) &&
                        isUndefOrInRange(ScaledMask[1], 0, 2) &&
                        isUndefOrInRange(ScaledMask[2], 2, 4) &&
                        isUndefOrInRange(ScaledMask[3], 2, 4) &&

diff  --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-7.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-7.ll
index 6a5dbbc56d9bc..02ec9fc66feab 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-7.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-7.ll
@@ -4145,8 +4145,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512-NEXT:    vpor %xmm8, %xmm11, %xmm8
 ; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm10[0],xmm9[0],xmm10[1],xmm9[1],xmm10[2],xmm9[2],xmm10[3],xmm9[3],xmm10[4],xmm9[4],xmm10[5],xmm9[5],xmm10[6],xmm9[6],xmm10[7],xmm9[7]
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512-NEXT:    vinserti32x4 $2, %xmm8, %zmm11, %zmm8
-; AVX512-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm11[0,1,0,1],zmm8[0,1,0,1]
 ; AVX512-NEXT:    vmovdqa (%rcx), %xmm14
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm11 = xmm14[u,u,u],zero,xmm14[7],zero,xmm14[5,u,u,u],zero,xmm14[8],zero,xmm14[6,u,u]
 ; AVX512-NEXT:    vmovdqa (%rdx), %xmm15
@@ -4154,8 +4153,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512-NEXT:    vpor %xmm11, %xmm12, %xmm11
 ; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm15[0],xmm14[0],xmm15[1],xmm14[1],xmm15[2],xmm14[2],xmm15[3],xmm14[3],xmm15[4],xmm14[4],xmm15[5],xmm14[5],xmm15[6],xmm14[6],xmm15[7],xmm14[7]
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512-NEXT:    vinserti32x4 $2, %xmm11, %zmm12, %zmm11
-; AVX512-NEXT:    vpermq {{.*#+}} zmm16 = zmm11[0,1,0,1,4,5,4,5]
+; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm16 = zmm12[0,1,0,1],zmm11[0,1,0,1]
 ; AVX512-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 ^ (mem & (zmm16 ^ zmm8))
 ; AVX512-NEXT:    vmovdqa (%r9), %xmm11
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm8 = zero,xmm11[4,u,u,u],zero,xmm11[7],zero,xmm11[5,u,u,u],zero,xmm11[8],zero,xmm11[6]
@@ -4164,8 +4162,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512-NEXT:    vpor %xmm8, %xmm13, %xmm8
 ; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm13 = xmm12[0],xmm11[0],xmm12[1],xmm11[1],xmm12[2],xmm11[2],xmm12[3],xmm11[3],xmm12[4],xmm11[4],xmm12[5],xmm11[5],xmm12[6],xmm11[6],xmm12[7],xmm11[7]
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm13 = xmm13[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512-NEXT:    vinserti32x4 $2, %xmm8, %zmm13, %zmm8
-; AVX512-NEXT:    vpermq {{.*#+}} zmm17 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm17 = zmm13[0,1,0,1],zmm8[0,1,0,1]
 ; AVX512-NEXT:    vmovdqa (%r10), %xmm13
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm8 = xmm13[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
 ; AVX512-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm13[1,1,0,0,4,5,6,7]
@@ -4281,8 +4278,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512-FCP-NEXT:    vpor %xmm8, %xmm11, %xmm8
 ; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm10[0],xmm9[0],xmm10[1],xmm9[1],xmm10[2],xmm9[2],xmm10[3],xmm9[3],xmm10[4],xmm9[4],xmm10[5],xmm9[5],xmm10[6],xmm9[6],xmm10[7],xmm9[7]
 ; AVX512-FCP-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512-FCP-NEXT:    vinserti32x4 $2, %xmm8, %zmm11, %zmm8
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm11[0,1,0,1],zmm8[0,1,0,1]
 ; AVX512-FCP-NEXT:    vmovdqa (%rcx), %xmm12
 ; AVX512-FCP-NEXT:    vpshufb {{.*#+}} xmm11 = xmm12[u,u,u],zero,xmm12[7],zero,xmm12[5,u,u,u],zero,xmm12[8],zero,xmm12[6,u,u]
 ; AVX512-FCP-NEXT:    vmovdqa (%rdx), %xmm14
@@ -4290,8 +4286,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512-FCP-NEXT:    vpor %xmm11, %xmm13, %xmm11
 ; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm13 = xmm14[0],xmm12[0],xmm14[1],xmm12[1],xmm14[2],xmm12[2],xmm14[3],xmm12[3],xmm14[4],xmm12[4],xmm14[5],xmm12[5],xmm14[6],xmm12[6],xmm14[7],xmm12[7]
 ; AVX512-FCP-NEXT:    vpshufb {{.*#+}} xmm13 = xmm13[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512-FCP-NEXT:    vinserti32x4 $2, %xmm11, %zmm13, %zmm11
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} zmm16 = zmm11[0,1,0,1,4,5,4,5]
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm16 = zmm13[0,1,0,1],zmm11[0,1,0,1]
 ; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 ^ (mem & (zmm16 ^ zmm8))
 ; AVX512-FCP-NEXT:    vmovdqa (%r10), %xmm11
 ; AVX512-FCP-NEXT:    vpshuflw {{.*#+}} xmm8 = xmm11[1,1,0,0,4,5,6,7]
@@ -4305,8 +4300,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512-FCP-NEXT:    vpor %xmm0, %xmm8, %xmm0
 ; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm8 = xmm15[0],xmm13[0],xmm15[1],xmm13[1],xmm15[2],xmm13[2],xmm15[3],xmm13[3],xmm15[4],xmm13[4],xmm15[5],xmm13[5],xmm15[6],xmm13[6],xmm15[7],xmm13[7]
 ; AVX512-FCP-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512-FCP-NEXT:    vinserti32x4 $2, %xmm0, %zmm8, %zmm0
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} zmm8 = zmm0[0,1,0,1,4,5,4,5]
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm8[0,1,0,1],zmm0[0,1,0,1]
 ; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm8 = zmm17 ^ (mem & (zmm8 ^ zmm17))
 ; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm8 = zmm8 ^ (mem & (zmm8 ^ zmm16))
 ; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm4[14,u,u],zero,zero,zero,zero,ymm4[15,u,u],zero,zero,zero,zero,ymm4[16,u,u],zero,zero,zero,zero,ymm4[17,u,u],zero,zero,zero,zero,ymm4[18]
@@ -4417,8 +4411,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-NEXT:    vpor %xmm8, %xmm11, %xmm8
 ; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm10[0],xmm9[0],xmm10[1],xmm9[1],xmm10[2],xmm9[2],xmm10[3],xmm9[3],xmm10[4],xmm9[4],xmm10[5],xmm9[5],xmm10[6],xmm9[6],xmm10[7],xmm9[7]
 ; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512DQ-NEXT:    vinserti32x4 $2, %xmm8, %zmm11, %zmm8
-; AVX512DQ-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm11[0,1,0,1],zmm8[0,1,0,1]
 ; AVX512DQ-NEXT:    vmovdqa (%rcx), %xmm14
 ; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm11 = xmm14[u,u,u],zero,xmm14[7],zero,xmm14[5,u,u,u],zero,xmm14[8],zero,xmm14[6,u,u]
 ; AVX512DQ-NEXT:    vmovdqa (%rdx), %xmm15
@@ -4426,8 +4419,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-NEXT:    vpor %xmm11, %xmm12, %xmm11
 ; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm15[0],xmm14[0],xmm15[1],xmm14[1],xmm15[2],xmm14[2],xmm15[3],xmm14[3],xmm15[4],xmm14[4],xmm15[5],xmm14[5],xmm15[6],xmm14[6],xmm15[7],xmm14[7]
 ; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512DQ-NEXT:    vinserti32x4 $2, %xmm11, %zmm12, %zmm11
-; AVX512DQ-NEXT:    vpermq {{.*#+}} zmm16 = zmm11[0,1,0,1,4,5,4,5]
+; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm16 = zmm12[0,1,0,1],zmm11[0,1,0,1]
 ; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 ^ (mem & (zmm16 ^ zmm8))
 ; AVX512DQ-NEXT:    vmovdqa (%r9), %xmm11
 ; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm8 = zero,xmm11[4,u,u,u],zero,xmm11[7],zero,xmm11[5,u,u,u],zero,xmm11[8],zero,xmm11[6]
@@ -4436,8 +4428,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-NEXT:    vpor %xmm8, %xmm13, %xmm8
 ; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm13 = xmm12[0],xmm11[0],xmm12[1],xmm11[1],xmm12[2],xmm11[2],xmm12[3],xmm11[3],xmm12[4],xmm11[4],xmm12[5],xmm11[5],xmm12[6],xmm11[6],xmm12[7],xmm11[7]
 ; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm13 = xmm13[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512DQ-NEXT:    vinserti32x4 $2, %xmm8, %zmm13, %zmm8
-; AVX512DQ-NEXT:    vpermq {{.*#+}} zmm17 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm17 = zmm13[0,1,0,1],zmm8[0,1,0,1]
 ; AVX512DQ-NEXT:    vmovdqa (%r10), %xmm13
 ; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm8 = xmm13[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
 ; AVX512DQ-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm13[1,1,0,0,4,5,6,7]
@@ -4553,8 +4544,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-FCP-NEXT:    vpor %xmm8, %xmm11, %xmm8
 ; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm10[0],xmm9[0],xmm10[1],xmm9[1],xmm10[2],xmm9[2],xmm10[3],xmm9[3],xmm10[4],xmm9[4],xmm10[5],xmm9[5],xmm10[6],xmm9[6],xmm10[7],xmm9[7]
 ; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512DQ-FCP-NEXT:    vinserti32x4 $2, %xmm8, %zmm11, %zmm8
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm11[0,1,0,1],zmm8[0,1,0,1]
 ; AVX512DQ-FCP-NEXT:    vmovdqa (%rcx), %xmm12
 ; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} xmm11 = xmm12[u,u,u],zero,xmm12[7],zero,xmm12[5,u,u,u],zero,xmm12[8],zero,xmm12[6,u,u]
 ; AVX512DQ-FCP-NEXT:    vmovdqa (%rdx), %xmm14
@@ -4562,8 +4552,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-FCP-NEXT:    vpor %xmm11, %xmm13, %xmm11
 ; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm13 = xmm14[0],xmm12[0],xmm14[1],xmm12[1],xmm14[2],xmm12[2],xmm14[3],xmm12[3],xmm14[4],xmm12[4],xmm14[5],xmm12[5],xmm14[6],xmm12[6],xmm14[7],xmm12[7]
 ; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} xmm13 = xmm13[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512DQ-FCP-NEXT:    vinserti32x4 $2, %xmm11, %zmm13, %zmm11
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} zmm16 = zmm11[0,1,0,1,4,5,4,5]
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm16 = zmm13[0,1,0,1],zmm11[0,1,0,1]
 ; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 ^ (mem & (zmm16 ^ zmm8))
 ; AVX512DQ-FCP-NEXT:    vmovdqa (%r10), %xmm11
 ; AVX512DQ-FCP-NEXT:    vpshuflw {{.*#+}} xmm8 = xmm11[1,1,0,0,4,5,6,7]
@@ -4577,8 +4566,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-FCP-NEXT:    vpor %xmm0, %xmm8, %xmm0
 ; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm8 = xmm15[0],xmm13[0],xmm15[1],xmm13[1],xmm15[2],xmm13[2],xmm15[3],xmm13[3],xmm15[4],xmm13[4],xmm15[5],xmm13[5],xmm15[6],xmm13[6],xmm15[7],xmm13[7]
 ; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512DQ-FCP-NEXT:    vinserti32x4 $2, %xmm0, %zmm8, %zmm0
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} zmm8 = zmm0[0,1,0,1,4,5,4,5]
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm8[0,1,0,1],zmm0[0,1,0,1]
 ; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm8 = zmm17 ^ (mem & (zmm8 ^ zmm17))
 ; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm8 = zmm8 ^ (mem & (zmm8 ^ zmm16))
 ; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm4[14,u,u],zero,zero,zero,zero,ymm4[15,u,u],zero,zero,zero,zero,ymm4[16,u,u],zero,zero,zero,zero,ymm4[17,u,u],zero,zero,zero,zero,ymm4[18]
@@ -4642,74 +4630,73 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512BW-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm4
-; AVX512BW-NEXT:    vmovdqa (%rsi), %ymm2
-; AVX512BW-NEXT:    vmovdqa (%rdx), %ymm1
-; AVX512BW-NEXT:    vmovdqa (%rcx), %ymm3
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[0,1,0,1,14],zero,ymm1[14,15,0,1,14,15],zero,ymm1[13,14,15,16,17,16],zero,ymm1[30,31,30,31,16,17],zero,ymm1[31,28,29,30,31]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm5 = zero,zero,zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero
-; AVX512BW-NEXT:    vpor %ymm0, %ymm5, %ymm0
-; AVX512BW-NEXT:    vmovdqa (%rdi), %xmm9
-; AVX512BW-NEXT:    vmovdqa (%rsi), %xmm10
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm10[8],xmm9[8],xmm10[9],xmm9[9],xmm10[10],xmm9[10],xmm10[11],xmm9[11],xmm10[12],xmm9[12],xmm10[13],xmm9[13],xmm10[14],xmm9[14],xmm10[15],xmm9[15]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512BW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm5, %zmm5
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[0,1,14],zero,ymm4[12,13,0,1,14,15],zero,ymm4[3,12,13,2,3,16],zero,ymm4[30,31,28,29,16,17],zero,ymm4[31,18,19,28,29,18],zero
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,ymm2[14],zero,zero,zero,zero,zero,zero,ymm2[15],zero,zero,zero,zero,zero,zero,ymm2[16],zero,zero,zero,zero,zero,zero,ymm2[17],zero,zero,zero,zero,zero,zero,ymm2[18]
-; AVX512BW-NEXT:    vpor %ymm0, %ymm6, %ymm0
-; AVX512BW-NEXT:    vmovdqa (%rdx), %xmm12
-; AVX512BW-NEXT:    vmovdqa (%rcx), %xmm14
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm14[8],xmm12[8],xmm14[9],xmm12[9],xmm14[10],xmm12[10],xmm14[11],xmm12[11],xmm14[12],xmm12[12],xmm14[13],xmm12[13],xmm14[14],xmm12[14],xmm14[15],xmm12[15]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512BW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm6, %zmm0
+; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm13
+; AVX512BW-NEXT:    vmovdqa (%rsi), %ymm7
+; AVX512BW-NEXT:    vmovdqa (%rdx), %ymm6
+; AVX512BW-NEXT:    vmovdqa (%rcx), %ymm9
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[0,1,0,1,14],zero,ymm6[14,15,0,1,14,15],zero,ymm6[13,14,15,16,17,16],zero,ymm6[30,31,30,31,16,17],zero,ymm6[31,28,29,30,31]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,ymm9[14],zero,zero,zero,zero,zero,zero,ymm9[15],zero,zero,zero,zero,zero,zero,ymm9[16],zero,zero,zero,zero,zero,zero,ymm9[17],zero,zero,zero,zero,zero
+; AVX512BW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512BW-NEXT:    vmovdqa (%rdi), %xmm3
+; AVX512BW-NEXT:    vmovdqa (%rsi), %xmm4
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm4[8],xmm3[8],xmm4[9],xmm3[9],xmm4[10],xmm3[10],xmm4[11],xmm3[11],xmm4[12],xmm3[12],xmm4[13],xmm3[13],xmm4[14],xmm3[14],xmm4[15],xmm3[15]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512BW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
+; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm13[0,1,14],zero,ymm13[12,13,0,1,14,15],zero,ymm13[3,12,13,2,3,16],zero,ymm13[30,31,28,29,16,17],zero,ymm13[31,18,19,28,29,18],zero
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,ymm7[14],zero,zero,zero,zero,zero,zero,ymm7[15],zero,zero,zero,zero,zero,zero,ymm7[16],zero,zero,zero,zero,zero,zero,ymm7[17],zero,zero,zero,zero,zero,zero,ymm7[18]
+; AVX512BW-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512BW-NEXT:    vmovdqa (%rdx), %xmm10
+; AVX512BW-NEXT:    vmovdqa (%rcx), %xmm12
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm12[8],xmm10[8],xmm12[9],xmm10[9],xmm12[10],xmm10[10],xmm12[11],xmm10[11],xmm12[12],xmm10[12],xmm12[13],xmm10[13],xmm12[14],xmm10[14],xmm12[15],xmm10[15]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512BW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
+; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
 ; AVX512BW-NEXT:    movabsq $435749858791416001, %rcx # imm = 0x60C1830183060C1
 ; AVX512BW-NEXT:    kmovq %rcx, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm0 {%k1}
+; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm1 = mem[0,1,2,3,0,1,2,3]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm1[13],zero,zero,zero,zero,zero,zero,ymm1[14],zero,zero,zero,zero,zero,zero,ymm1[15],zero,zero,zero,zero,zero,zero,ymm1[16],zero,zero,zero,zero,zero,zero,ymm1[17],zero,zero,zero
+; AVX512BW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm2 = mem[0,1,2,3,0,1,2,3]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm8 = zero,zero,zero,zero,zero,zero,ymm2[14],zero,zero,zero,zero,zero,zero,ymm2[15],zero,zero,zero,zero,zero,zero,ymm2[16],zero,zero,zero,zero,zero,zero,ymm2[17],zero,zero,zero,zero
+; AVX512BW-NEXT:    vpor %ymm5, %ymm8, %ymm5
+; AVX512BW-NEXT:    vmovdqa (%r9), %xmm8
+; AVX512BW-NEXT:    vmovdqa (%r8), %xmm11
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} xmm14 = xmm11[8],xmm8[8],xmm11[9],xmm8[9],xmm11[10],xmm8[10],xmm11[11],xmm8[11],xmm11[12],xmm8[12],xmm11[13],xmm8[13],xmm11[14],xmm8[14],xmm11[15],xmm8[15]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm14 = xmm14[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX512BW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,0,1]
+; AVX512BW-NEXT:    vinserti64x4 $1, %ymm5, %zmm14, %zmm14
 ; AVX512BW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm5 = mem[0,1,2,3,0,1,2,3]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm5[13],zero,zero,zero,zero,zero,zero,ymm5[14],zero,zero,zero,zero,zero,zero,ymm5[15],zero,zero,zero,zero,zero,zero,ymm5[16],zero,zero,zero,zero,zero,zero,ymm5[17],zero,zero,zero
-; AVX512BW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm6 = mem[0,1,2,3,0,1,2,3]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm8 = zero,zero,zero,zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero
-; AVX512BW-NEXT:    vpor %ymm7, %ymm8, %ymm7
-; AVX512BW-NEXT:    vmovdqa (%r9), %xmm11
-; AVX512BW-NEXT:    vmovdqa (%r8), %xmm13
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm13[8],xmm11[8],xmm13[9],xmm11[9],xmm13[10],xmm11[10],xmm13[11],xmm11[11],xmm13[12],xmm11[12],xmm13[13],xmm11[13],xmm13[14],xmm11[14],xmm13[15],xmm11[15]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX512BW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm7, %zmm8, %zmm8
-; AVX512BW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm7 = mem[0,1,2,3,0,1,2,3]
 ; AVX512BW-NEXT:    vpmovsxbw {{.*#+}} zmm15 = [4,5,4,5,5,6,5,6,4,5,4,5,5,6,5,6,6,1,6,0,7,1,0,7,14,9,14,8,15,9,8,15]
-; AVX512BW-NEXT:    vpermw %zmm7, %zmm15, %zmm15
+; AVX512BW-NEXT:    vpermw %zmm5, %zmm15, %zmm15
 ; AVX512BW-NEXT:    movabsq $2323999253380730912, %rcx # imm = 0x2040810204081020
 ; AVX512BW-NEXT:    kmovq %rcx, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm15, %zmm8 {%k1}
+; AVX512BW-NEXT:    vmovdqu8 %zmm15, %zmm14 {%k1}
 ; AVX512BW-NEXT:    movabsq $4066998693416279096, %rcx # imm = 0x3870E1C3870E1C38
 ; AVX512BW-NEXT:    kmovq %rcx, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm8, %zmm0 {%k1}
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm8
-; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm8 = zmm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zmm8[18,19,20,21],zero,zmm8[19],zero,zmm8[25,26,27,22],zero,zmm8[20],zero,zmm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,57],zero,zmm8[55],zero,zero,zero,zero,zmm8[58],zero,zmm8[56],zero,zero,zero,zero,zmm8[59],zero
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm15
-; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm15 = zmm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm15[18],zero,zero,zero,zero,zmm15[21],zero,zmm15[19],zero,zero,zero,zero,zmm15[22],zero,zmm15[20,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm15[55],zero,zero,zero,zero,zmm15[58],zero,zmm15[56],zero,zero,zero,zero,zmm15[59],zero,zmm15[57]
-; AVX512BW-NEXT:    vporq %zmm8, %zmm15, %zmm8
-; AVX512BW-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[2,3,2,3,6,7,6,7]
-; AVX512BW-NEXT:    vpshuflw {{.*#+}} ymm15 = ymm4[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512BW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[0,0,1,1,4,4,5,5]
+; AVX512BW-NEXT:    vmovdqu8 %zmm14, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm13[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512BW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[0,0,1,1,4,4,5,5]
 ; AVX512BW-NEXT:    movl $676341840, %ecx # imm = 0x28502850
 ; AVX512BW-NEXT:    kmovd %ecx, %k1
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm15 {%k1} = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,21,u,19,u,u,u,u,22,u,20,u,u]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm1[23],zero,ymm1[21,22,23,26],zero,ymm1[24],zero,ymm1[28,29,26,27]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm17 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero,zero,zero
-; AVX512BW-NEXT:    vporq %ymm16, %ymm17, %ymm16
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm16, %zmm15, %zmm15
-; AVX512BW-NEXT:    vpermq {{.*#+}} zmm15 = zmm15[2,3,2,3,6,7,6,7]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm14 {%k1} = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,21,u,19,u,u,u,u,22,u,20,u,u]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm15 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm6[23],zero,ymm6[21,22,23,26],zero,ymm6[24],zero,ymm6[28,29,26,27]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm9[25],zero,ymm9[23],zero,zero,zero,zero,ymm9[26],zero,ymm9[24],zero,zero,zero,zero
+; AVX512BW-NEXT:    vporq %ymm15, %ymm16, %ymm15
+; AVX512BW-NEXT:    vshufi64x2 {{.*#+}} zmm15 = zmm14[2,3,2,3],zmm15[2,3,2,3]
+; AVX512BW-NEXT:    vinserti64x4 $1, %ymm7, %zmm6, %zmm14
+; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm14 = zmm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zmm14[18,19,20,21],zero,zmm14[19],zero,zmm14[25,26,27,22],zero,zmm14[20],zero,zmm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,57],zero,zmm14[55],zero,zero,zero,zero,zmm14[58],zero,zmm14[56],zero,zero,zero,zero,zmm14[59],zero
+; AVX512BW-NEXT:    vinserti64x4 $1, %ymm13, %zmm9, %zmm16
+; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm16 = zmm16[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm16[18],zero,zero,zero,zero,zmm16[21],zero,zmm16[19],zero,zero,zero,zero,zmm16[22],zero,zmm16[20,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm16[55],zero,zero,zero,zero,zmm16[58],zero,zmm16[56],zero,zero,zero,zero,zmm16[59],zero,zmm16[57]
+; AVX512BW-NEXT:    vporq %zmm14, %zmm16, %zmm14
+; AVX512BW-NEXT:    vpermq {{.*#+}} zmm14 = zmm14[2,3,2,3,6,7,6,7]
 ; AVX512BW-NEXT:    movabsq $-9005497107459067808, %rcx # imm = 0x83060C180C183060
 ; AVX512BW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-NEXT:    vmovdqu8 %zmm15, %zmm8 {%k2}
+; AVX512BW-NEXT:    vmovdqu8 %zmm15, %zmm14 {%k2}
 ; AVX512BW-NEXT:    vpmovsxbw {{.*#+}} zmm15 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10,27,29,28,27,28,29,29,28,27,29,28,27,28,29,29,28]
-; AVX512BW-NEXT:    vpermw %zmm7, %zmm15, %zmm15
-; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm16 = zmm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,zmm6[18],zero,zmm6[20,21,20,21],zero,zmm6[19],zero,zmm6[19,20,21,22],zero,zmm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57,56,57],zero,zmm6[55],zero,zmm6[55,56,57,58],zero,zmm6[56],zero,zmm6[62,63]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm17 = zmm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm5[20],zero,zmm5[18],zero,zero,zero,zero,zmm5[21],zero,zmm5[19],zero,zero,zero,zero,zmm5[22,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm5[57],zero,zmm5[55],zero,zero,zero,zero,zmm5[58],zero,zmm5[56],zero,zero
+; AVX512BW-NEXT:    vpermw %zmm5, %zmm15, %zmm15
+; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm16 = zmm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,zmm2[18],zero,zmm2[20,21,20,21],zero,zmm2[19],zero,zmm2[19,20,21,22],zero,zmm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57,56,57],zero,zmm2[55],zero,zmm2[55,56,57,58],zero,zmm2[56],zero,zmm2[62,63]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm17 = zmm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm1[20],zero,zmm1[18],zero,zero,zero,zero,zmm1[21],zero,zmm1[19],zero,zero,zero,zero,zmm1[22,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm1[57],zero,zmm1[55],zero,zero,zero,zero,zmm1[58],zero,zmm1[56],zero,zero
 ; AVX512BW-NEXT:    vporq %zmm16, %zmm17, %zmm16
 ; AVX512BW-NEXT:    vpermq {{.*#+}} zmm16 = zmm16[2,3,2,3,6,7,6,7]
 ; AVX512BW-NEXT:    movabsq $1161999626690365456, %rcx # imm = 0x1020408102040810
@@ -4717,69 +4704,66 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512BW-NEXT:    vmovdqu8 %zmm15, %zmm16 {%k2}
 ; AVX512BW-NEXT:    movabsq $2033499346708139548, %rcx # imm = 0x1C3870E1C3870E1C
 ; AVX512BW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-NEXT:    vmovdqu8 %zmm16, %zmm8 {%k2}
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm15 = xmm14[u,u,u],zero,xmm14[7],zero,xmm14[5,u,u,u],zero,xmm14[8],zero,xmm14[6,u,u]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm16 = xmm12[u,u,u,7],zero,xmm12[5],zero,xmm12[u,u,u,8],zero,xmm12[6],zero,xmm12[u,u]
-; AVX512BW-NEXT:    vporq %xmm15, %xmm16, %xmm15
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm12[0],xmm14[0],xmm12[1],xmm14[1],xmm12[2],xmm14[2],xmm12[3],xmm14[3],xmm12[4],xmm14[4],xmm12[5],xmm14[5],xmm12[6],xmm14[6],xmm12[7],xmm14[7]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512BW-NEXT:    vinserti32x4 $2, %xmm15, %zmm12, %zmm12
-; AVX512BW-NEXT:    vpermq {{.*#+}} zmm12 = zmm12[0,1,0,1,4,5,4,5]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm14 = xmm10[u],zero,xmm10[7],zero,xmm10[5,u,u,u],zero,xmm10[8],zero,xmm10[6,u,u,u],zero
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm15 = xmm9[u,7],zero,xmm9[5],zero,xmm9[u,u,u,8],zero,xmm9[6],zero,xmm9[u,u,u,9]
-; AVX512BW-NEXT:    vpor %xmm14, %xmm15, %xmm14
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3],xmm9[4],xmm10[4],xmm9[5],xmm10[5],xmm9[6],xmm10[6],xmm9[7],xmm10[7]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm9[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512BW-NEXT:    vinserti32x4 $2, %xmm14, %zmm9, %zmm9
-; AVX512BW-NEXT:    vpermq {{.*#+}} zmm9 = zmm9[0,1,0,1,4,5,4,5]
-; AVX512BW-NEXT:    movabsq $871499720017774092, %rcx # imm = 0xC183060C183060C
-; AVX512BW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-NEXT:    vmovdqu8 %zmm12, %zmm9 {%k2}
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm10 = zero,xmm11[4,u,u,u],zero,xmm11[7],zero,xmm11[5,u,u,u],zero,xmm11[8],zero,xmm11[6]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm12 = xmm13[4],zero,xmm13[u,u,u,7],zero,xmm13[5],zero,xmm13[u,u,u,8],zero,xmm13[6],zero
-; AVX512BW-NEXT:    vpor %xmm10, %xmm12, %xmm10
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm13[0],xmm11[0],xmm13[1],xmm11[1],xmm13[2],xmm11[2],xmm13[3],xmm11[3],xmm13[4],xmm11[4],xmm13[5],xmm11[5],xmm13[6],xmm11[6],xmm13[7],xmm11[7]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512BW-NEXT:    vinserti32x4 $2, %xmm10, %zmm11, %zmm10
-; AVX512BW-NEXT:    vpermq {{.*#+}} zmm10 = zmm10[0,1,0,1,4,5,4,5]
-; AVX512BW-NEXT:    vpmovsxbw {{.*#+}} zmm11 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
-; AVX512BW-NEXT:    vpermw %zmm7, %zmm11, %zmm11
-; AVX512BW-NEXT:    movabsq $4647998506761461824, %rcx # imm = 0x4081020408102040
-; AVX512BW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-NEXT:    vmovdqu8 %zmm11, %zmm10 {%k2}
-; AVX512BW-NEXT:    movabsq $8133997386832558192, %rcx # imm = 0x70E1C3870E1C3870
-; AVX512BW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-NEXT:    vmovdqu8 %zmm10, %zmm9 {%k2}
-; AVX512BW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm4[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX512BW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,2,3,3,6,6,7,7]
+; AVX512BW-NEXT:    vmovdqu8 %zmm16, %zmm14 {%k2}
+; AVX512BW-NEXT:    vpshufhw {{.*#+}} ymm13 = ymm13[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
+; AVX512BW-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[2,2,3,3,6,6,7,7]
 ; AVX512BW-NEXT:    movl $338170920, %ecx # imm = 0x14281428
 ; AVX512BW-NEXT:    kmovd %ecx, %k2
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm4 {%k2} = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u,u,u]
-; AVX512BW-NEXT:    vpermq {{.*#+}} ymm2 = ymm4[2,3,2,3]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
-; AVX512BW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512BW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,2,3,3,4,6,7,7]
-; AVX512BW-NEXT:    vmovdqu8 %ymm1, %ymm3 {%k1}
-; AVX512BW-NEXT:    vpermq {{.*#+}} ymm1 = ymm3[2,3,2,3]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm13 {%k2} = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u,u,u]
+; AVX512BW-NEXT:    vpermq {{.*#+}} ymm7 = ymm13[2,3,2,3]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm9 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
+; AVX512BW-NEXT:    vpshufhw {{.*#+}} ymm6 = ymm6[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
+; AVX512BW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[0,2,3,3,4,6,7,7]
+; AVX512BW-NEXT:    vmovdqu8 %ymm6, %ymm9 {%k1}
+; AVX512BW-NEXT:    vpermq {{.*#+}} ymm6 = ymm9[2,3,2,3]
 ; AVX512BW-NEXT:    movl $101455920, %ecx # imm = 0x60C1830
 ; AVX512BW-NEXT:    kmovd %ecx, %k1
-; AVX512BW-NEXT:    vmovdqu8 %ymm2, %ymm1 {%k1}
-; AVX512BW-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15]
-; AVX512BW-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512BW-NEXT:    vpermw %ymm7, %ymm2, %ymm2
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm5[27],zero,zero,zero,zero,ymm5[30],zero,ymm5[28],zero,zero,zero,zero,ymm5[31],zero
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm6[27],zero,zero,zero,zero,ymm6[30],zero,ymm6[28],zero,zero,zero,zero,ymm6[31],zero,ymm6[29]
-; AVX512BW-NEXT:    vpor %ymm3, %ymm4, %ymm3
-; AVX512BW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX512BW-NEXT:    vmovdqu8 %ymm7, %ymm6 {%k1}
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm12[u,u,u],zero,xmm12[7],zero,xmm12[5,u,u,u],zero,xmm12[8],zero,xmm12[6,u,u]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm10[u,u,u,7],zero,xmm10[5],zero,xmm10[u,u,u,8],zero,xmm10[6],zero,xmm10[u,u]
+; AVX512BW-NEXT:    vpor %xmm7, %xmm9, %xmm7
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm10[0],xmm12[0],xmm10[1],xmm12[1],xmm10[2],xmm12[2],xmm10[3],xmm12[3],xmm10[4],xmm12[4],xmm10[5],xmm12[5],xmm10[6],xmm12[6],xmm10[7],xmm12[7]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm9[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX512BW-NEXT:    vshufi64x2 {{.*#+}} zmm7 = zmm9[0,1,0,1],zmm7[0,1,0,1]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm4[u],zero,xmm4[7],zero,xmm4[5,u,u,u],zero,xmm4[8],zero,xmm4[6,u,u,u],zero
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm10 = xmm3[u,7],zero,xmm3[5],zero,xmm3[u,u,u,8],zero,xmm3[6],zero,xmm3[u,u,u,9]
+; AVX512BW-NEXT:    vpor %xmm9, %xmm10, %xmm9
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
+; AVX512BW-NEXT:    vshufi64x2 {{.*#+}} zmm3 = zmm3[0,1,0,1],zmm9[0,1,0,1]
+; AVX512BW-NEXT:    movabsq $871499720017774092, %rcx # imm = 0xC183060C183060C
+; AVX512BW-NEXT:    kmovq %rcx, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm7, %zmm3 {%k1}
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm4 = zero,xmm8[4,u,u,u],zero,xmm8[7],zero,xmm8[5,u,u,u],zero,xmm8[8],zero,xmm8[6]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm11[4],zero,xmm11[u,u,u,7],zero,xmm11[5],zero,xmm11[u,u,u,8],zero,xmm11[6],zero
+; AVX512BW-NEXT:    vpor %xmm4, %xmm7, %xmm4
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} xmm7 = xmm11[0],xmm8[0],xmm11[1],xmm8[1],xmm11[2],xmm8[2],xmm11[3],xmm8[3],xmm11[4],xmm8[4],xmm11[5],xmm8[5],xmm11[6],xmm8[6],xmm11[7],xmm8[7]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
+; AVX512BW-NEXT:    vshufi64x2 {{.*#+}} zmm4 = zmm7[0,1,0,1],zmm4[0,1,0,1]
+; AVX512BW-NEXT:    vpmovsxbw {{.*#+}} zmm7 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
+; AVX512BW-NEXT:    vpermw %zmm5, %zmm7, %zmm7
+; AVX512BW-NEXT:    movabsq $4647998506761461824, %rcx # imm = 0x4081020408102040
+; AVX512BW-NEXT:    kmovq %rcx, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm7, %zmm4 {%k1}
+; AVX512BW-NEXT:    movabsq $8133997386832558192, %rcx # imm = 0x70E1C3870E1C3870
+; AVX512BW-NEXT:    kmovq %rcx, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k1}
+; AVX512BW-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15]
+; AVX512BW-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512BW-NEXT:    vpermw %ymm5, %ymm4, %ymm4
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm1[27],zero,zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero
+; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29]
+; AVX512BW-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512BW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX512BW-NEXT:    movl $-2130574328, %ecx # imm = 0x81020408
 ; AVX512BW-NEXT:    kmovd %ecx, %k1
-; AVX512BW-NEXT:    vmovdqu8 %ymm2, %ymm3 {%k1}
+; AVX512BW-NEXT:    vmovdqu8 %ymm4, %ymm1 {%k1}
+; AVX512BW-NEXT:    vmovdqa64 %zmm3, (%rax)
 ; AVX512BW-NEXT:    movl $-507279602, %ecx # imm = 0xE1C3870E
 ; AVX512BW-NEXT:    kmovd %ecx, %k1
-; AVX512BW-NEXT:    vmovdqu8 %ymm3, %ymm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %ymm1, 192(%rax)
-; AVX512BW-NEXT:    vmovdqa64 %zmm9, (%rax)
-; AVX512BW-NEXT:    vmovdqa64 %zmm8, 128(%rax)
+; AVX512BW-NEXT:    vmovdqu8 %ymm1, %ymm6 {%k1}
+; AVX512BW-NEXT:    vmovdqa %ymm6, 192(%rax)
+; AVX512BW-NEXT:    vmovdqa64 %zmm14, 128(%rax)
 ; AVX512BW-NEXT:    vmovdqa64 %zmm0, 64(%rax)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
@@ -4865,15 +4849,13 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512BW-FCP-NEXT:    vporq %xmm15, %xmm16, %xmm15
 ; AVX512BW-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm12[0],xmm14[0],xmm12[1],xmm14[1],xmm12[2],xmm14[2],xmm12[3],xmm14[3],xmm12[4],xmm14[4],xmm12[5],xmm14[5],xmm12[6],xmm14[6],xmm12[7],xmm14[7]
 ; AVX512BW-FCP-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512BW-FCP-NEXT:    vinserti32x4 $2, %xmm15, %zmm12, %zmm12
-; AVX512BW-FCP-NEXT:    vpermq {{.*#+}} zmm12 = zmm12[0,1,0,1,4,5,4,5]
+; AVX512BW-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm12 = zmm12[0,1,0,1],zmm15[0,1,0,1]
 ; AVX512BW-FCP-NEXT:    vpshufb {{.*#+}} xmm14 = xmm9[u],zero,xmm9[7],zero,xmm9[5,u,u,u],zero,xmm9[8],zero,xmm9[6,u,u,u],zero
 ; AVX512BW-FCP-NEXT:    vpshufb {{.*#+}} xmm15 = xmm8[u,7],zero,xmm8[5],zero,xmm8[u,u,u,8],zero,xmm8[6],zero,xmm8[u,u,u,9]
 ; AVX512BW-FCP-NEXT:    vpor %xmm14, %xmm15, %xmm14
 ; AVX512BW-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm8 = xmm8[0],xmm9[0],xmm8[1],xmm9[1],xmm8[2],xmm9[2],xmm8[3],xmm9[3],xmm8[4],xmm9[4],xmm8[5],xmm9[5],xmm8[6],xmm9[6],xmm8[7],xmm9[7]
 ; AVX512BW-FCP-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512BW-FCP-NEXT:    vinserti32x4 $2, %xmm14, %zmm8, %zmm8
-; AVX512BW-FCP-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512BW-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm8[0,1,0,1],zmm14[0,1,0,1]
 ; AVX512BW-FCP-NEXT:    movabsq $871499720017774092, %rcx # imm = 0xC183060C183060C
 ; AVX512BW-FCP-NEXT:    kmovq %rcx, %k1
 ; AVX512BW-FCP-NEXT:    vmovdqu8 %zmm12, %zmm8 {%k1}
@@ -4882,8 +4864,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512BW-FCP-NEXT:    vpor %xmm9, %xmm12, %xmm9
 ; AVX512BW-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm13[0],xmm11[0],xmm13[1],xmm11[1],xmm13[2],xmm11[2],xmm13[3],xmm11[3],xmm13[4],xmm11[4],xmm13[5],xmm11[5],xmm13[6],xmm11[6],xmm13[7],xmm11[7]
 ; AVX512BW-FCP-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512BW-FCP-NEXT:    vinserti32x4 $2, %xmm9, %zmm11, %zmm9
-; AVX512BW-FCP-NEXT:    vpermq {{.*#+}} zmm9 = zmm9[0,1,0,1,4,5,4,5]
+; AVX512BW-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm9 = zmm11[0,1,0,1],zmm9[0,1,0,1]
 ; AVX512BW-FCP-NEXT:    vpmovsxbw {{.*#+}} zmm11 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
 ; AVX512BW-FCP-NEXT:    vpermw %zmm7, %zmm11, %zmm11
 ; AVX512BW-FCP-NEXT:    movabsq $4647998506761461824, %rcx # imm = 0x4081020408102040
@@ -4913,11 +4894,11 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512BW-FCP-NEXT:    movl $-2130574328, %ecx # imm = 0x81020408
 ; AVX512BW-FCP-NEXT:    kmovd %ecx, %k1
 ; AVX512BW-FCP-NEXT:    vmovdqu8 %ymm2, %ymm3 {%k1}
+; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm8, (%rax)
 ; AVX512BW-FCP-NEXT:    movl $-507279602, %ecx # imm = 0xE1C3870E
 ; AVX512BW-FCP-NEXT:    kmovd %ecx, %k1
 ; AVX512BW-FCP-NEXT:    vmovdqu8 %ymm3, %ymm1 {%k1}
 ; AVX512BW-FCP-NEXT:    vmovdqa %ymm1, 192(%rax)
-; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm8, (%rax)
 ; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm10, 128(%rax)
 ; AVX512BW-FCP-NEXT:    vmovdqa64 %zmm0, 64(%rax)
 ; AVX512BW-FCP-NEXT:    vzeroupper
@@ -4927,74 +4908,73 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-BW:       # %bb.0:
 ; AVX512DQ-BW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512DQ-BW-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX512DQ-BW-NEXT:    vmovdqa (%rdi), %ymm4
-; AVX512DQ-BW-NEXT:    vmovdqa (%rsi), %ymm2
-; AVX512DQ-BW-NEXT:    vmovdqa (%rdx), %ymm1
-; AVX512DQ-BW-NEXT:    vmovdqa (%rcx), %ymm3
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[0,1,0,1,14],zero,ymm1[14,15,0,1,14,15],zero,ymm1[13,14,15,16,17,16],zero,ymm1[30,31,30,31,16,17],zero,ymm1[31,28,29,30,31]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm5 = zero,zero,zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero
-; AVX512DQ-BW-NEXT:    vpor %ymm0, %ymm5, %ymm0
-; AVX512DQ-BW-NEXT:    vmovdqa (%rdi), %xmm9
-; AVX512DQ-BW-NEXT:    vmovdqa (%rsi), %xmm10
-; AVX512DQ-BW-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm10[8],xmm9[8],xmm10[9],xmm9[9],xmm10[10],xmm9[10],xmm10[11],xmm9[11],xmm10[12],xmm9[12],xmm10[13],xmm9[13],xmm10[14],xmm9[14],xmm10[15],xmm9[15]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
-; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm5, %zmm5
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[0,1,14],zero,ymm4[12,13,0,1,14,15],zero,ymm4[3,12,13,2,3,16],zero,ymm4[30,31,28,29,16,17],zero,ymm4[31,18,19,28,29,18],zero
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,ymm2[14],zero,zero,zero,zero,zero,zero,ymm2[15],zero,zero,zero,zero,zero,zero,ymm2[16],zero,zero,zero,zero,zero,zero,ymm2[17],zero,zero,zero,zero,zero,zero,ymm2[18]
-; AVX512DQ-BW-NEXT:    vpor %ymm0, %ymm6, %ymm0
-; AVX512DQ-BW-NEXT:    vmovdqa (%rdx), %xmm12
-; AVX512DQ-BW-NEXT:    vmovdqa (%rcx), %xmm14
-; AVX512DQ-BW-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm14[8],xmm12[8],xmm14[9],xmm12[9],xmm14[10],xmm12[10],xmm14[11],xmm12[11],xmm14[12],xmm12[12],xmm14[13],xmm12[13],xmm14[14],xmm12[14],xmm14[15],xmm12[15]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
-; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm6, %zmm0
+; AVX512DQ-BW-NEXT:    vmovdqa (%rdi), %ymm13
+; AVX512DQ-BW-NEXT:    vmovdqa (%rsi), %ymm7
+; AVX512DQ-BW-NEXT:    vmovdqa (%rdx), %ymm6
+; AVX512DQ-BW-NEXT:    vmovdqa (%rcx), %ymm9
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[0,1,0,1,14],zero,ymm6[14,15,0,1,14,15],zero,ymm6[13,14,15,16,17,16],zero,ymm6[30,31,30,31,16,17],zero,ymm6[31,28,29,30,31]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,ymm9[14],zero,zero,zero,zero,zero,zero,ymm9[15],zero,zero,zero,zero,zero,zero,ymm9[16],zero,zero,zero,zero,zero,zero,ymm9[17],zero,zero,zero,zero,zero
+; AVX512DQ-BW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-BW-NEXT:    vmovdqa (%rdi), %xmm3
+; AVX512DQ-BW-NEXT:    vmovdqa (%rsi), %xmm4
+; AVX512DQ-BW-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm4[8],xmm3[8],xmm4[9],xmm3[9],xmm4[10],xmm3[10],xmm4[11],xmm3[11],xmm4[12],xmm3[12],xmm4[13],xmm3[13],xmm4[14],xmm3[14],xmm4[15],xmm3[15]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
+; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm1
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm13[0,1,14],zero,ymm13[12,13,0,1,14,15],zero,ymm13[3,12,13,2,3,16],zero,ymm13[30,31,28,29,16,17],zero,ymm13[31,18,19,28,29,18],zero
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,ymm7[14],zero,zero,zero,zero,zero,zero,ymm7[15],zero,zero,zero,zero,zero,zero,ymm7[16],zero,zero,zero,zero,zero,zero,ymm7[17],zero,zero,zero,zero,zero,zero,ymm7[18]
+; AVX512DQ-BW-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512DQ-BW-NEXT:    vmovdqa (%rdx), %xmm10
+; AVX512DQ-BW-NEXT:    vmovdqa (%rcx), %xmm12
+; AVX512DQ-BW-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm12[8],xmm10[8],xmm12[9],xmm10[9],xmm12[10],xmm10[10],xmm12[11],xmm10[11],xmm12[12],xmm10[12],xmm12[13],xmm10[13],xmm12[14],xmm10[14],xmm12[15],xmm10[15]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
+; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
 ; AVX512DQ-BW-NEXT:    movabsq $435749858791416001, %rcx # imm = 0x60C1830183060C1
 ; AVX512DQ-BW-NEXT:    kmovq %rcx, %k1
-; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm5, %zmm0 {%k1}
+; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512DQ-BW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm1 = mem[0,1,2,3,0,1,2,3]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm1[13],zero,zero,zero,zero,zero,zero,ymm1[14],zero,zero,zero,zero,zero,zero,ymm1[15],zero,zero,zero,zero,zero,zero,ymm1[16],zero,zero,zero,zero,zero,zero,ymm1[17],zero,zero,zero
+; AVX512DQ-BW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm2 = mem[0,1,2,3,0,1,2,3]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm8 = zero,zero,zero,zero,zero,zero,ymm2[14],zero,zero,zero,zero,zero,zero,ymm2[15],zero,zero,zero,zero,zero,zero,ymm2[16],zero,zero,zero,zero,zero,zero,ymm2[17],zero,zero,zero,zero
+; AVX512DQ-BW-NEXT:    vpor %ymm5, %ymm8, %ymm5
+; AVX512DQ-BW-NEXT:    vmovdqa (%r9), %xmm8
+; AVX512DQ-BW-NEXT:    vmovdqa (%r8), %xmm11
+; AVX512DQ-BW-NEXT:    vpunpckhbw {{.*#+}} xmm14 = xmm11[8],xmm8[8],xmm11[9],xmm8[9],xmm11[10],xmm8[10],xmm11[11],xmm8[11],xmm11[12],xmm8[12],xmm11[13],xmm8[13],xmm11[14],xmm8[14],xmm11[15],xmm8[15]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm14 = xmm14[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,0,1]
+; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm5, %zmm14, %zmm14
 ; AVX512DQ-BW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm5 = mem[0,1,2,3,0,1,2,3]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm5[13],zero,zero,zero,zero,zero,zero,ymm5[14],zero,zero,zero,zero,zero,zero,ymm5[15],zero,zero,zero,zero,zero,zero,ymm5[16],zero,zero,zero,zero,zero,zero,ymm5[17],zero,zero,zero
-; AVX512DQ-BW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm6 = mem[0,1,2,3,0,1,2,3]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm8 = zero,zero,zero,zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero
-; AVX512DQ-BW-NEXT:    vpor %ymm7, %ymm8, %ymm7
-; AVX512DQ-BW-NEXT:    vmovdqa (%r9), %xmm11
-; AVX512DQ-BW-NEXT:    vmovdqa (%r8), %xmm13
-; AVX512DQ-BW-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm13[8],xmm11[8],xmm13[9],xmm11[9],xmm13[10],xmm11[10],xmm13[11],xmm11[11],xmm13[12],xmm11[12],xmm13[13],xmm11[13],xmm13[14],xmm11[14],xmm13[15],xmm11[15]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
-; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm7, %zmm8, %zmm8
-; AVX512DQ-BW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm7 = mem[0,1,2,3,0,1,2,3]
 ; AVX512DQ-BW-NEXT:    vpmovsxbw {{.*#+}} zmm15 = [4,5,4,5,5,6,5,6,4,5,4,5,5,6,5,6,6,1,6,0,7,1,0,7,14,9,14,8,15,9,8,15]
-; AVX512DQ-BW-NEXT:    vpermw %zmm7, %zmm15, %zmm15
+; AVX512DQ-BW-NEXT:    vpermw %zmm5, %zmm15, %zmm15
 ; AVX512DQ-BW-NEXT:    movabsq $2323999253380730912, %rcx # imm = 0x2040810204081020
 ; AVX512DQ-BW-NEXT:    kmovq %rcx, %k1
-; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm15, %zmm8 {%k1}
+; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm15, %zmm14 {%k1}
 ; AVX512DQ-BW-NEXT:    movabsq $4066998693416279096, %rcx # imm = 0x3870E1C3870E1C38
 ; AVX512DQ-BW-NEXT:    kmovq %rcx, %k1
-; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm8, %zmm0 {%k1}
-; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm8
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} zmm8 = zmm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zmm8[18,19,20,21],zero,zmm8[19],zero,zmm8[25,26,27,22],zero,zmm8[20],zero,zmm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,57],zero,zmm8[55],zero,zero,zero,zero,zmm8[58],zero,zmm8[56],zero,zero,zero,zero,zmm8[59],zero
-; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm15
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} zmm15 = zmm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm15[18],zero,zero,zero,zero,zmm15[21],zero,zmm15[19],zero,zero,zero,zero,zmm15[22],zero,zmm15[20,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm15[55],zero,zero,zero,zero,zmm15[58],zero,zmm15[56],zero,zero,zero,zero,zmm15[59],zero,zmm15[57]
-; AVX512DQ-BW-NEXT:    vporq %zmm8, %zmm15, %zmm8
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[2,3,2,3,6,7,6,7]
-; AVX512DQ-BW-NEXT:    vpshuflw {{.*#+}} ymm15 = ymm4[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512DQ-BW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[0,0,1,1,4,4,5,5]
+; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm14, %zmm0 {%k1}
+; AVX512DQ-BW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm13[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512DQ-BW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[0,0,1,1,4,4,5,5]
 ; AVX512DQ-BW-NEXT:    movl $676341840, %ecx # imm = 0x28502850
 ; AVX512DQ-BW-NEXT:    kmovd %ecx, %k1
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm15 {%k1} = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,21,u,19,u,u,u,u,22,u,20,u,u]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm1[23],zero,ymm1[21,22,23,26],zero,ymm1[24],zero,ymm1[28,29,26,27]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm17 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero,zero,zero
-; AVX512DQ-BW-NEXT:    vporq %ymm16, %ymm17, %ymm16
-; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm16, %zmm15, %zmm15
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} zmm15 = zmm15[2,3,2,3,6,7,6,7]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm14 {%k1} = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,21,u,19,u,u,u,u,22,u,20,u,u]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm15 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm6[23],zero,ymm6[21,22,23,26],zero,ymm6[24],zero,ymm6[28,29,26,27]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm9[25],zero,ymm9[23],zero,zero,zero,zero,ymm9[26],zero,ymm9[24],zero,zero,zero,zero
+; AVX512DQ-BW-NEXT:    vporq %ymm15, %ymm16, %ymm15
+; AVX512DQ-BW-NEXT:    vshufi64x2 {{.*#+}} zmm15 = zmm14[2,3,2,3],zmm15[2,3,2,3]
+; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm7, %zmm6, %zmm14
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} zmm14 = zmm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zmm14[18,19,20,21],zero,zmm14[19],zero,zmm14[25,26,27,22],zero,zmm14[20],zero,zmm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,57],zero,zmm14[55],zero,zero,zero,zero,zmm14[58],zero,zmm14[56],zero,zero,zero,zero,zmm14[59],zero
+; AVX512DQ-BW-NEXT:    vinserti64x4 $1, %ymm13, %zmm9, %zmm16
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} zmm16 = zmm16[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm16[18],zero,zero,zero,zero,zmm16[21],zero,zmm16[19],zero,zero,zero,zero,zmm16[22],zero,zmm16[20,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm16[55],zero,zero,zero,zero,zmm16[58],zero,zmm16[56],zero,zero,zero,zero,zmm16[59],zero,zmm16[57]
+; AVX512DQ-BW-NEXT:    vporq %zmm14, %zmm16, %zmm14
+; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} zmm14 = zmm14[2,3,2,3,6,7,6,7]
 ; AVX512DQ-BW-NEXT:    movabsq $-9005497107459067808, %rcx # imm = 0x83060C180C183060
 ; AVX512DQ-BW-NEXT:    kmovq %rcx, %k2
-; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm15, %zmm8 {%k2}
+; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm15, %zmm14 {%k2}
 ; AVX512DQ-BW-NEXT:    vpmovsxbw {{.*#+}} zmm15 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10,27,29,28,27,28,29,29,28,27,29,28,27,28,29,29,28]
-; AVX512DQ-BW-NEXT:    vpermw %zmm7, %zmm15, %zmm15
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} zmm16 = zmm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,zmm6[18],zero,zmm6[20,21,20,21],zero,zmm6[19],zero,zmm6[19,20,21,22],zero,zmm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57,56,57],zero,zmm6[55],zero,zmm6[55,56,57,58],zero,zmm6[56],zero,zmm6[62,63]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} zmm17 = zmm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm5[20],zero,zmm5[18],zero,zero,zero,zero,zmm5[21],zero,zmm5[19],zero,zero,zero,zero,zmm5[22,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm5[57],zero,zmm5[55],zero,zero,zero,zero,zmm5[58],zero,zmm5[56],zero,zero
+; AVX512DQ-BW-NEXT:    vpermw %zmm5, %zmm15, %zmm15
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} zmm16 = zmm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,zmm2[18],zero,zmm2[20,21,20,21],zero,zmm2[19],zero,zmm2[19,20,21,22],zero,zmm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57,56,57],zero,zmm2[55],zero,zmm2[55,56,57,58],zero,zmm2[56],zero,zmm2[62,63]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} zmm17 = zmm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm1[20],zero,zmm1[18],zero,zero,zero,zero,zmm1[21],zero,zmm1[19],zero,zero,zero,zero,zmm1[22,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm1[57],zero,zmm1[55],zero,zero,zero,zero,zmm1[58],zero,zmm1[56],zero,zero
 ; AVX512DQ-BW-NEXT:    vporq %zmm16, %zmm17, %zmm16
 ; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} zmm16 = zmm16[2,3,2,3,6,7,6,7]
 ; AVX512DQ-BW-NEXT:    movabsq $1161999626690365456, %rcx # imm = 0x1020408102040810
@@ -5002,69 +4982,66 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm15, %zmm16 {%k2}
 ; AVX512DQ-BW-NEXT:    movabsq $2033499346708139548, %rcx # imm = 0x1C3870E1C3870E1C
 ; AVX512DQ-BW-NEXT:    kmovq %rcx, %k2
-; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm16, %zmm8 {%k2}
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm15 = xmm14[u,u,u],zero,xmm14[7],zero,xmm14[5,u,u,u],zero,xmm14[8],zero,xmm14[6,u,u]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm16 = xmm12[u,u,u,7],zero,xmm12[5],zero,xmm12[u,u,u,8],zero,xmm12[6],zero,xmm12[u,u]
-; AVX512DQ-BW-NEXT:    vporq %xmm15, %xmm16, %xmm15
-; AVX512DQ-BW-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm12[0],xmm14[0],xmm12[1],xmm14[1],xmm12[2],xmm14[2],xmm12[3],xmm14[3],xmm12[4],xmm14[4],xmm12[5],xmm14[5],xmm12[6],xmm14[6],xmm12[7],xmm14[7]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512DQ-BW-NEXT:    vinserti32x4 $2, %xmm15, %zmm12, %zmm12
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} zmm12 = zmm12[0,1,0,1,4,5,4,5]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm14 = xmm10[u],zero,xmm10[7],zero,xmm10[5,u,u,u],zero,xmm10[8],zero,xmm10[6,u,u,u],zero
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm15 = xmm9[u,7],zero,xmm9[5],zero,xmm9[u,u,u,8],zero,xmm9[6],zero,xmm9[u,u,u,9]
-; AVX512DQ-BW-NEXT:    vpor %xmm14, %xmm15, %xmm14
-; AVX512DQ-BW-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3],xmm9[4],xmm10[4],xmm9[5],xmm10[5],xmm9[6],xmm10[6],xmm9[7],xmm10[7]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm9[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512DQ-BW-NEXT:    vinserti32x4 $2, %xmm14, %zmm9, %zmm9
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} zmm9 = zmm9[0,1,0,1,4,5,4,5]
-; AVX512DQ-BW-NEXT:    movabsq $871499720017774092, %rcx # imm = 0xC183060C183060C
-; AVX512DQ-BW-NEXT:    kmovq %rcx, %k2
-; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm12, %zmm9 {%k2}
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm10 = zero,xmm11[4,u,u,u],zero,xmm11[7],zero,xmm11[5,u,u,u],zero,xmm11[8],zero,xmm11[6]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm12 = xmm13[4],zero,xmm13[u,u,u,7],zero,xmm13[5],zero,xmm13[u,u,u,8],zero,xmm13[6],zero
-; AVX512DQ-BW-NEXT:    vpor %xmm10, %xmm12, %xmm10
-; AVX512DQ-BW-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm13[0],xmm11[0],xmm13[1],xmm11[1],xmm13[2],xmm11[2],xmm13[3],xmm11[3],xmm13[4],xmm11[4],xmm13[5],xmm11[5],xmm13[6],xmm11[6],xmm13[7],xmm11[7]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512DQ-BW-NEXT:    vinserti32x4 $2, %xmm10, %zmm11, %zmm10
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} zmm10 = zmm10[0,1,0,1,4,5,4,5]
-; AVX512DQ-BW-NEXT:    vpmovsxbw {{.*#+}} zmm11 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
-; AVX512DQ-BW-NEXT:    vpermw %zmm7, %zmm11, %zmm11
-; AVX512DQ-BW-NEXT:    movabsq $4647998506761461824, %rcx # imm = 0x4081020408102040
-; AVX512DQ-BW-NEXT:    kmovq %rcx, %k2
-; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm11, %zmm10 {%k2}
-; AVX512DQ-BW-NEXT:    movabsq $8133997386832558192, %rcx # imm = 0x70E1C3870E1C3870
-; AVX512DQ-BW-NEXT:    kmovq %rcx, %k2
-; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm10, %zmm9 {%k2}
-; AVX512DQ-BW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm4[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX512DQ-BW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,2,3,3,6,6,7,7]
+; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm16, %zmm14 {%k2}
+; AVX512DQ-BW-NEXT:    vpshufhw {{.*#+}} ymm13 = ymm13[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
+; AVX512DQ-BW-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[2,2,3,3,6,6,7,7]
 ; AVX512DQ-BW-NEXT:    movl $338170920, %ecx # imm = 0x14281428
 ; AVX512DQ-BW-NEXT:    kmovd %ecx, %k2
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm4 {%k2} = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u,u,u]
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm2 = ymm4[2,3,2,3]
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
-; AVX512DQ-BW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512DQ-BW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,2,3,3,4,6,7,7]
-; AVX512DQ-BW-NEXT:    vmovdqu8 %ymm1, %ymm3 {%k1}
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm1 = ymm3[2,3,2,3]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm13 {%k2} = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u,u,u]
+; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm7 = ymm13[2,3,2,3]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm9 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
+; AVX512DQ-BW-NEXT:    vpshufhw {{.*#+}} ymm6 = ymm6[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
+; AVX512DQ-BW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[0,2,3,3,4,6,7,7]
+; AVX512DQ-BW-NEXT:    vmovdqu8 %ymm6, %ymm9 {%k1}
+; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm6 = ymm9[2,3,2,3]
 ; AVX512DQ-BW-NEXT:    movl $101455920, %ecx # imm = 0x60C1830
 ; AVX512DQ-BW-NEXT:    kmovd %ecx, %k1
-; AVX512DQ-BW-NEXT:    vmovdqu8 %ymm2, %ymm1 {%k1}
-; AVX512DQ-BW-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15]
-; AVX512DQ-BW-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512DQ-BW-NEXT:    vpermw %ymm7, %ymm2, %ymm2
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm5[27],zero,zero,zero,zero,ymm5[30],zero,ymm5[28],zero,zero,zero,zero,ymm5[31],zero
-; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm6[27],zero,zero,zero,zero,ymm6[30],zero,ymm6[28],zero,zero,zero,zero,ymm6[31],zero,ymm6[29]
-; AVX512DQ-BW-NEXT:    vpor %ymm3, %ymm4, %ymm3
-; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX512DQ-BW-NEXT:    vmovdqu8 %ymm7, %ymm6 {%k1}
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm12[u,u,u],zero,xmm12[7],zero,xmm12[5,u,u,u],zero,xmm12[8],zero,xmm12[6,u,u]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm10[u,u,u,7],zero,xmm10[5],zero,xmm10[u,u,u,8],zero,xmm10[6],zero,xmm10[u,u]
+; AVX512DQ-BW-NEXT:    vpor %xmm7, %xmm9, %xmm7
+; AVX512DQ-BW-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm10[0],xmm12[0],xmm10[1],xmm12[1],xmm10[2],xmm12[2],xmm10[3],xmm12[3],xmm10[4],xmm12[4],xmm10[5],xmm12[5],xmm10[6],xmm12[6],xmm10[7],xmm12[7]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm9[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX512DQ-BW-NEXT:    vshufi64x2 {{.*#+}} zmm7 = zmm9[0,1,0,1],zmm7[0,1,0,1]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm4[u],zero,xmm4[7],zero,xmm4[5,u,u,u],zero,xmm4[8],zero,xmm4[6,u,u,u],zero
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm10 = xmm3[u,7],zero,xmm3[5],zero,xmm3[u,u,u,8],zero,xmm3[6],zero,xmm3[u,u,u,9]
+; AVX512DQ-BW-NEXT:    vpor %xmm9, %xmm10, %xmm9
+; AVX512DQ-BW-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
+; AVX512DQ-BW-NEXT:    vshufi64x2 {{.*#+}} zmm3 = zmm3[0,1,0,1],zmm9[0,1,0,1]
+; AVX512DQ-BW-NEXT:    movabsq $871499720017774092, %rcx # imm = 0xC183060C183060C
+; AVX512DQ-BW-NEXT:    kmovq %rcx, %k1
+; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm7, %zmm3 {%k1}
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm4 = zero,xmm8[4,u,u,u],zero,xmm8[7],zero,xmm8[5,u,u,u],zero,xmm8[8],zero,xmm8[6]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm11[4],zero,xmm11[u,u,u,7],zero,xmm11[5],zero,xmm11[u,u,u,8],zero,xmm11[6],zero
+; AVX512DQ-BW-NEXT:    vpor %xmm4, %xmm7, %xmm4
+; AVX512DQ-BW-NEXT:    vpunpcklbw {{.*#+}} xmm7 = xmm11[0],xmm8[0],xmm11[1],xmm8[1],xmm11[2],xmm8[2],xmm11[3],xmm8[3],xmm11[4],xmm8[4],xmm11[5],xmm8[5],xmm11[6],xmm8[6],xmm11[7],xmm8[7]
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
+; AVX512DQ-BW-NEXT:    vshufi64x2 {{.*#+}} zmm4 = zmm7[0,1,0,1],zmm4[0,1,0,1]
+; AVX512DQ-BW-NEXT:    vpmovsxbw {{.*#+}} zmm7 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
+; AVX512DQ-BW-NEXT:    vpermw %zmm5, %zmm7, %zmm7
+; AVX512DQ-BW-NEXT:    movabsq $4647998506761461824, %rcx # imm = 0x4081020408102040
+; AVX512DQ-BW-NEXT:    kmovq %rcx, %k1
+; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm7, %zmm4 {%k1}
+; AVX512DQ-BW-NEXT:    movabsq $8133997386832558192, %rcx # imm = 0x70E1C3870E1C3870
+; AVX512DQ-BW-NEXT:    kmovq %rcx, %k1
+; AVX512DQ-BW-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k1}
+; AVX512DQ-BW-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15]
+; AVX512DQ-BW-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512DQ-BW-NEXT:    vpermw %ymm5, %ymm4, %ymm4
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm1[27],zero,zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero
+; AVX512DQ-BW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29]
+; AVX512DQ-BW-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512DQ-BW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX512DQ-BW-NEXT:    movl $-2130574328, %ecx # imm = 0x81020408
 ; AVX512DQ-BW-NEXT:    kmovd %ecx, %k1
-; AVX512DQ-BW-NEXT:    vmovdqu8 %ymm2, %ymm3 {%k1}
+; AVX512DQ-BW-NEXT:    vmovdqu8 %ymm4, %ymm1 {%k1}
+; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm3, (%rax)
 ; AVX512DQ-BW-NEXT:    movl $-507279602, %ecx # imm = 0xE1C3870E
 ; AVX512DQ-BW-NEXT:    kmovd %ecx, %k1
-; AVX512DQ-BW-NEXT:    vmovdqu8 %ymm3, %ymm1 {%k1}
-; AVX512DQ-BW-NEXT:    vmovdqa %ymm1, 192(%rax)
-; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm9, (%rax)
-; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm8, 128(%rax)
+; AVX512DQ-BW-NEXT:    vmovdqu8 %ymm1, %ymm6 {%k1}
+; AVX512DQ-BW-NEXT:    vmovdqa %ymm6, 192(%rax)
+; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm14, 128(%rax)
 ; AVX512DQ-BW-NEXT:    vmovdqa64 %zmm0, 64(%rax)
 ; AVX512DQ-BW-NEXT:    vzeroupper
 ; AVX512DQ-BW-NEXT:    retq
@@ -5150,15 +5127,13 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-BW-FCP-NEXT:    vporq %xmm15, %xmm16, %xmm15
 ; AVX512DQ-BW-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm12[0],xmm14[0],xmm12[1],xmm14[1],xmm12[2],xmm14[2],xmm12[3],xmm14[3],xmm12[4],xmm14[4],xmm12[5],xmm14[5],xmm12[6],xmm14[6],xmm12[7],xmm14[7]
 ; AVX512DQ-BW-FCP-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512DQ-BW-FCP-NEXT:    vinserti32x4 $2, %xmm15, %zmm12, %zmm12
-; AVX512DQ-BW-FCP-NEXT:    vpermq {{.*#+}} zmm12 = zmm12[0,1,0,1,4,5,4,5]
+; AVX512DQ-BW-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm12 = zmm12[0,1,0,1],zmm15[0,1,0,1]
 ; AVX512DQ-BW-FCP-NEXT:    vpshufb {{.*#+}} xmm14 = xmm9[u],zero,xmm9[7],zero,xmm9[5,u,u,u],zero,xmm9[8],zero,xmm9[6,u,u,u],zero
 ; AVX512DQ-BW-FCP-NEXT:    vpshufb {{.*#+}} xmm15 = xmm8[u,7],zero,xmm8[5],zero,xmm8[u,u,u,8],zero,xmm8[6],zero,xmm8[u,u,u,9]
 ; AVX512DQ-BW-FCP-NEXT:    vpor %xmm14, %xmm15, %xmm14
 ; AVX512DQ-BW-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm8 = xmm8[0],xmm9[0],xmm8[1],xmm9[1],xmm8[2],xmm9[2],xmm8[3],xmm9[3],xmm8[4],xmm9[4],xmm8[5],xmm9[5],xmm8[6],xmm9[6],xmm8[7],xmm9[7]
 ; AVX512DQ-BW-FCP-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512DQ-BW-FCP-NEXT:    vinserti32x4 $2, %xmm14, %zmm8, %zmm8
-; AVX512DQ-BW-FCP-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512DQ-BW-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm8[0,1,0,1],zmm14[0,1,0,1]
 ; AVX512DQ-BW-FCP-NEXT:    movabsq $871499720017774092, %rcx # imm = 0xC183060C183060C
 ; AVX512DQ-BW-FCP-NEXT:    kmovq %rcx, %k1
 ; AVX512DQ-BW-FCP-NEXT:    vmovdqu8 %zmm12, %zmm8 {%k1}
@@ -5167,8 +5142,7 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-BW-FCP-NEXT:    vpor %xmm9, %xmm12, %xmm9
 ; AVX512DQ-BW-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm13[0],xmm11[0],xmm13[1],xmm11[1],xmm13[2],xmm11[2],xmm13[3],xmm11[3],xmm13[4],xmm11[4],xmm13[5],xmm11[5],xmm13[6],xmm11[6],xmm13[7],xmm11[7]
 ; AVX512DQ-BW-FCP-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512DQ-BW-FCP-NEXT:    vinserti32x4 $2, %xmm9, %zmm11, %zmm9
-; AVX512DQ-BW-FCP-NEXT:    vpermq {{.*#+}} zmm9 = zmm9[0,1,0,1,4,5,4,5]
+; AVX512DQ-BW-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm9 = zmm11[0,1,0,1],zmm9[0,1,0,1]
 ; AVX512DQ-BW-FCP-NEXT:    vpmovsxbw {{.*#+}} zmm11 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
 ; AVX512DQ-BW-FCP-NEXT:    vpermw %zmm7, %zmm11, %zmm11
 ; AVX512DQ-BW-FCP-NEXT:    movabsq $4647998506761461824, %rcx # imm = 0x4081020408102040
@@ -5198,11 +5172,11 @@ define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX512DQ-BW-FCP-NEXT:    movl $-2130574328, %ecx # imm = 0x81020408
 ; AVX512DQ-BW-FCP-NEXT:    kmovd %ecx, %k1
 ; AVX512DQ-BW-FCP-NEXT:    vmovdqu8 %ymm2, %ymm3 {%k1}
+; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm8, (%rax)
 ; AVX512DQ-BW-FCP-NEXT:    movl $-507279602, %ecx # imm = 0xE1C3870E
 ; AVX512DQ-BW-FCP-NEXT:    kmovd %ecx, %k1
 ; AVX512DQ-BW-FCP-NEXT:    vmovdqu8 %ymm3, %ymm1 {%k1}
 ; AVX512DQ-BW-FCP-NEXT:    vmovdqa %ymm1, 192(%rax)
-; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm8, (%rax)
 ; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm10, 128(%rax)
 ; AVX512DQ-BW-FCP-NEXT:    vmovdqa64 %zmm0, 64(%rax)
 ; AVX512DQ-BW-FCP-NEXT:    vzeroupper
@@ -8425,1577 +8399,1535 @@ define void @store_i8_stride7_vf64(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ;
 ; AVX512-LABEL: store_i8_stride7_vf64:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    subq $1560, %rsp # imm = 0x618
-; AVX512-NEXT:    vmovdqa (%rsi), %ymm6
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero,zero,zero,ymm6[18]
-; AVX512-NEXT:    vmovdqa (%rdi), %ymm11
+; AVX512-NEXT:    subq $1496, %rsp # imm = 0x5D8
+; AVX512-NEXT:    vmovdqa (%rsi), %ymm3
+; AVX512-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
+; AVX512-NEXT:    vpshufb %ymm1, %ymm3, %ymm0
+; AVX512-NEXT:    vmovdqa64 %ymm1, %ymm23
+; AVX512-NEXT:    vmovdqa (%rdi), %ymm10
 ; AVX512-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,14,128,12,13,0,1,14,15,128,3,12,13,2,3,16,128,30,31,28,29,16,17,128,31,18,19,28,29,18,128]
-; AVX512-NEXT:    vpshufb %ymm2, %ymm11, %ymm1
+; AVX512-NEXT:    vpshufb %ymm2, %ymm10, %ymm1
 ; AVX512-NEXT:    vmovdqa64 %ymm2, %ymm16
 ; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-NEXT:    vmovdqa (%rcx), %ymm13
+; AVX512-NEXT:    vmovdqa (%rcx), %ymm8
 ; AVX512-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
-; AVX512-NEXT:    vpshufb %ymm1, %ymm13, %ymm0
-; AVX512-NEXT:    vmovdqa64 %ymm1, %ymm24
-; AVX512-NEXT:    vmovdqa (%rdx), %ymm2
-; AVX512-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
-; AVX512-NEXT:    vpshufb %ymm3, %ymm2, %ymm1
-; AVX512-NEXT:    vmovdqa64 %ymm3, %ymm26
-; AVX512-NEXT:    vmovdqa %ymm2, %ymm3
+; AVX512-NEXT:    vpshufb %ymm1, %ymm8, %ymm0
+; AVX512-NEXT:    vmovdqa64 %ymm1, %ymm18
+; AVX512-NEXT:    vmovdqa (%rdx), %ymm7
+; AVX512-NEXT:    vmovdqa {{.*#+}} ymm11 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
+; AVX512-NEXT:    vpshufb %ymm11, %ymm7, %ymm1
 ; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-NEXT:    vmovdqa (%r8), %ymm4
-; AVX512-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
-; AVX512-NEXT:    vpshufb %ymm2, %ymm4, %ymm0
-; AVX512-NEXT:    vmovdqa64 %ymm2, %ymm21
-; AVX512-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-NEXT:    vmovdqa (%r9), %ymm8
-; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [13,0,0,0,128,16,128,14,0,0,0,128,17,128,15,0,13,0,0,0,128,16,128,14,0,0,0,128,17,128,15,0]
-; AVX512-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm5, %ymm8, %ymm1
-; AVX512-NEXT:    vmovdqa64 %ymm5, %ymm19
+; AVX512-NEXT:    vmovdqa (%r8), %ymm6
+; AVX512-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
+; AVX512-NEXT:    vpshufb %ymm1, %ymm6, %ymm0
+; AVX512-NEXT:    vmovdqa64 %ymm1, %ymm24
+; AVX512-NEXT:    vmovdqa (%r9), %ymm5
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [13,0,0,0,128,16,128,14,0,0,0,128,17,128,15,0,13,0,0,0,128,16,128,14,0,0,0,128,17,128,15,0]
+; AVX512-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX512-NEXT:    vpshufb %ymm2, %ymm5, %ymm1
+; AVX512-NEXT:    vmovdqa64 %ymm2, %ymm25
 ; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128]
-; AVX512-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm1, %ymm6, %ymm0
-; AVX512-NEXT:    vmovdqa64 %ymm6, %ymm20
-; AVX512-NEXT:    vmovdqa {{.*#+}} ymm5 = [12,13,14,128,12,128,14,15,14,15,128,13,128,15,12,13,28,29,30,128,28,128,30,31,30,31,128,29,128,31,28,29]
-; AVX512-NEXT:    vpshufb %ymm5, %ymm11, %ymm2
-; AVX512-NEXT:    vpor %ymm0, %ymm2, %ymm0
-; AVX512-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128]
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm14 = [128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128]
+; AVX512-NEXT:    # ymm14 = mem[0,1,0,1]
+; AVX512-NEXT:    vpshufb %ymm14, %ymm3, %ymm1
+; AVX512-NEXT:    vmovdqa64 %ymm3, %ymm17
+; AVX512-NEXT:    vmovdqa {{.*#+}} ymm0 = [12,13,14,128,12,128,14,15,14,15,128,13,128,15,12,13,28,29,30,128,28,128,30,31,30,31,128,29,128,31,28,29]
+; AVX512-NEXT:    vpshufb %ymm0, %ymm10, %ymm2
+; AVX512-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm15 = [128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128]
+; AVX512-NEXT:    # ymm15 = mem[0,1,0,1]
+; AVX512-NEXT:    vpshufb %ymm15, %ymm7, %ymm1
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0]
 ; AVX512-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm2, %ymm3, %ymm0
-; AVX512-NEXT:    vmovdqa64 %ymm3, %ymm22
-; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0]
-; AVX512-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm7, %ymm13, %ymm3
-; AVX512-NEXT:    vpor %ymm0, %ymm3, %ymm0
-; AVX512-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vpshufb %ymm2, %ymm8, %ymm3
+; AVX512-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [128,27,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,27,128,128,128,128,30,128,28,128,128,128,128,31,128,29]
 ; AVX512-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm3, %ymm4, %ymm0
-; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128]
-; AVX512-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm4, %ymm8, %ymm6
-; AVX512-NEXT:    vmovdqa64 %ymm8, %ymm17
-; AVX512-NEXT:    vpor %ymm0, %ymm6, %ymm0
+; AVX512-NEXT:    vpshufb %ymm3, %ymm6, %ymm1
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128]
+; AVX512-NEXT:    # ymm9 = mem[0,1,0,1]
+; AVX512-NEXT:    vpshufb %ymm9, %ymm5, %ymm13
+; AVX512-NEXT:    vpor %ymm1, %ymm13, %ymm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vmovdqa 32(%rsi), %ymm13
+; AVX512-NEXT:    vpshufb %ymm14, %ymm13, %ymm1
+; AVX512-NEXT:    vmovdqa 32(%rdi), %ymm14
+; AVX512-NEXT:    vpshufb %ymm0, %ymm14, %ymm0
+; AVX512-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa 32(%rsi), %ymm10
-; AVX512-NEXT:    vpshufb %ymm1, %ymm10, %ymm6
-; AVX512-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX512-NEXT:    vpshufb %ymm5, %ymm1, %ymm5
-; AVX512-NEXT:    vpor %ymm6, %ymm5, %ymm5
-; AVX512-NEXT:    vmovdqu64 %zmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25]
-; AVX512-NEXT:    # ymm6 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm6, %ymm1, %ymm5
-; AVX512-NEXT:    vmovdqa64 %ymm6, %ymm29
-; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128,25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128]
-; AVX512-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm8, %ymm10, %ymm6
-; AVX512-NEXT:    vmovdqa64 %ymm8, %ymm25
-; AVX512-NEXT:    vpor %ymm5, %ymm6, %ymm5
-; AVX512-NEXT:    vmovdqu64 %zmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa 32(%rdx), %ymm5
-; AVX512-NEXT:    vpshufb %ymm2, %ymm5, %ymm2
-; AVX512-NEXT:    vmovdqa 32(%rcx), %ymm6
-; AVX512-NEXT:    vpshufb %ymm7, %ymm6, %ymm7
-; AVX512-NEXT:    vpor %ymm2, %ymm7, %ymm2
+; AVX512-NEXT:    vpshufb {{.*#+}} ymm0 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm14[23],zero,zero,zero,zero,ymm14[26],zero,ymm14[24],zero,zero,zero,zero,ymm14[27],zero,ymm14[25]
+; AVX512-NEXT:    vpshufb {{.*#+}} ymm1 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm13[23,u,u,u],zero,ymm13[26],zero,ymm13[24,u,u,u],zero,ymm13[27],zero
+; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vmovdqa 32(%rdx), %ymm1
+; AVX512-NEXT:    vpshufb %ymm15, %ymm1, %ymm15
+; AVX512-NEXT:    vmovdqa 32(%rcx), %ymm0
+; AVX512-NEXT:    vpshufb %ymm2, %ymm0, %ymm2
+; AVX512-NEXT:    vpor %ymm2, %ymm15, %ymm2
 ; AVX512-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm2 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero,zero,zero
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm7 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm5[23],zero,ymm5[21,22,23,26],zero,ymm5[24],zero,ymm5[28,29,26,27]
-; AVX512-NEXT:    vpor %ymm2, %ymm7, %ymm2
+; AVX512-NEXT:    vpshufb {{.*#+}} ymm2 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm0[25],zero,ymm0[23],zero,zero,zero,zero,ymm0[26],zero,ymm0[24],zero,zero,zero,zero
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm12 = [24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27,24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27]
+; AVX512-NEXT:    # ymm12 = mem[0,1,0,1]
+; AVX512-NEXT:    vpshufb %ymm12, %ymm1, %ymm15
+; AVX512-NEXT:    vmovdqa64 %ymm12, %ymm22
+; AVX512-NEXT:    vpor %ymm2, %ymm15, %ymm2
 ; AVX512-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa 32(%r8), %ymm2
-; AVX512-NEXT:    vpshufb %ymm3, %ymm2, %ymm7
-; AVX512-NEXT:    vmovdqa 32(%r9), %ymm3
-; AVX512-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512-NEXT:    vpor %ymm7, %ymm4, %ymm4
-; AVX512-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm4 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm7 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm2[23],zero,ymm2[23,24,25,26],zero,ymm2[24],zero,ymm2[30,31]
-; AVX512-NEXT:    vpor %ymm4, %ymm7, %ymm4
-; AVX512-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vmovdqa 32(%r8), %ymm15
+; AVX512-NEXT:    vpshufb %ymm3, %ymm15, %ymm3
+; AVX512-NEXT:    vmovdqa 32(%r9), %ymm2
+; AVX512-NEXT:    vpshufb %ymm9, %ymm2, %ymm9
+; AVX512-NEXT:    vpor %ymm3, %ymm9, %ymm3
+; AVX512-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128]
+; AVX512-NEXT:    # ymm9 = mem[0,1,0,1]
+; AVX512-NEXT:    vpshufb %ymm9, %ymm2, %ymm3
+; AVX512-NEXT:    vmovdqa64 %ymm9, %ymm21
+; AVX512-NEXT:    vpshufb {{.*#+}} ymm9 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm15[23],zero,ymm15[23,24,25,26],zero,ymm15[24],zero,ymm15[30,31]
+; AVX512-NEXT:    vpor %ymm3, %ymm9, %ymm3
+; AVX512-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512-NEXT:    vmovdqa 32(%rax), %ymm7
-; AVX512-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm4 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm7 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512-NEXT:    vinserti64x4 $1, %ymm4, %zmm7, %zmm4
-; AVX512-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa 32(%rdx), %xmm7
-; AVX512-NEXT:    vmovdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-NEXT:    vmovdqa 32(%rcx), %xmm4
-; AVX512-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm15 = [u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u]
-; AVX512-NEXT:    vpshufb %xmm15, %xmm4, %xmm4
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm14 = [u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u]
-; AVX512-NEXT:    vpshufb %xmm14, %xmm7, %xmm7
-; AVX512-NEXT:    vmovdqa64 %xmm14, %xmm23
-; AVX512-NEXT:    vpor %xmm4, %xmm7, %xmm4
-; AVX512-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa 32(%rdi), %xmm9
-; AVX512-NEXT:    vmovdqa 32(%rsi), %xmm0
-; AVX512-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm7 = [u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128]
-; AVX512-NEXT:    vpshufb %xmm7, %xmm0, %xmm4
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm14 = [u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9]
-; AVX512-NEXT:    vpshufb %xmm14, %xmm9, %xmm8
-; AVX512-NEXT:    vmovdqa64 %xmm9, %xmm31
-; AVX512-NEXT:    vpor %xmm4, %xmm8, %xmm4
-; AVX512-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm28 = [0,0,0,0,2,3,0,1,0,18,0,19,18,0,19,0]
-; AVX512-NEXT:    vmovdqa 32(%rax), %xmm8
-; AVX512-NEXT:    vmovdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-NEXT:    vpshufhw {{.*#+}} xmm4 = xmm8[0,1,2,3,4,5,5,6]
-; AVX512-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
-; AVX512-NEXT:    vpermi2d %zmm4, %zmm8, %zmm28
-; AVX512-NEXT:    vmovdqa 32(%r9), %xmm4
-; AVX512-NEXT:    vmovdqa 32(%r8), %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm12 = [128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6]
-; AVX512-NEXT:    vpshufb %xmm12, %xmm4, %xmm8
-; AVX512-NEXT:    vmovdqa64 %xmm4, %xmm30
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128]
-; AVX512-NEXT:    vpshufb %xmm4, %xmm0, %xmm9
-; AVX512-NEXT:    vmovdqa64 %xmm4, %xmm18
-; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm27
-; AVX512-NEXT:    vpor %xmm8, %xmm9, %xmm0
-; AVX512-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa64 %ymm24, %ymm4
-; AVX512-NEXT:    vpshufb %ymm4, %ymm6, %ymm8
-; AVX512-NEXT:    vmovdqa64 %ymm26, %ymm4
-; AVX512-NEXT:    vpshufb %ymm4, %ymm5, %ymm9
-; AVX512-NEXT:    vpor %ymm8, %ymm9, %ymm0
-; AVX512-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20]
-; AVX512-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm8, %ymm6, %ymm6
+; AVX512-NEXT:    vmovdqa 32(%rax), %ymm9
+; AVX512-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512-NEXT:    vpshufb {{.*#+}} ymm3 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512-NEXT:    vpshufb %ymm4, %ymm9, %ymm12
+; AVX512-NEXT:    vmovdqa64 %ymm4, %ymm20
+; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm29 = zmm12[2,3,2,3],zmm3[2,3,2,3]
+; AVX512-NEXT:    vmovdqa64 %ymm18, %ymm3
+; AVX512-NEXT:    vpshufb %ymm3, %ymm0, %ymm3
+; AVX512-NEXT:    vpshufb %ymm11, %ymm1, %ymm11
+; AVX512-NEXT:    vpor %ymm3, %ymm11, %ymm3
+; AVX512-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20]
+; AVX512-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512-NEXT:    vpshufb %ymm4, %ymm0, %ymm0
 ; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128,18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128]
 ; AVX512-NEXT:    # ymm9 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm9, %ymm5, %ymm5
-; AVX512-NEXT:    vpor %ymm6, %ymm5, %ymm0
+; AVX512-NEXT:    vpshufb %ymm9, %ymm1, %ymm1
+; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm5 = zero,zero,zero,ymm10[14],zero,zero,zero,zero,zero,zero,ymm10[15],zero,zero,zero,zero,zero,zero,ymm10[16],zero,zero,zero,zero,zero,zero,ymm10[17],zero,zero,zero,zero,zero,zero,ymm10[18]
-; AVX512-NEXT:    vmovdqa64 %ymm16, %ymm0
-; AVX512-NEXT:    vpshufb %ymm0, %ymm1, %ymm6
-; AVX512-NEXT:    vpor %ymm5, %ymm6, %ymm0
+; AVX512-NEXT:    vmovdqa64 %ymm23, %ymm0
+; AVX512-NEXT:    vpshufb %ymm0, %ymm13, %ymm0
+; AVX512-NEXT:    vmovdqa64 %ymm16, %ymm1
+; AVX512-NEXT:    vpshufb %ymm1, %ymm14, %ymm1
+; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128]
-; AVX512-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm0, %ymm10, %ymm5
-; AVX512-NEXT:    vmovdqa64 %ymm0, %ymm16
-; AVX512-NEXT:    vmovdqa {{.*#+}} ymm0 = [2,3,4,5,128,3,128,5,4,5,6,128,4,128,6,7,18,19,20,21,128,19,128,21,20,21,22,128,20,128,22,23]
-; AVX512-NEXT:    vpshufb %ymm0, %ymm1, %ymm1
-; AVX512-NEXT:    vpor %ymm5, %ymm1, %ymm1
-; AVX512-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128]
+; AVX512-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512-NEXT:    vpshufb %ymm1, %ymm13, %ymm0
+; AVX512-NEXT:    vmovdqa64 %ymm1, %ymm26
+; AVX512-NEXT:    vmovdqa {{.*#+}} ymm13 = [2,3,4,5,128,3,128,5,4,5,6,128,4,128,6,7,18,19,20,21,128,19,128,21,20,21,22,128,20,128,22,23]
+; AVX512-NEXT:    vpshufb %ymm13, %ymm14, %ymm1
+; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22]
 ; AVX512-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm1, %ymm3, %ymm5
-; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128,20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128]
-; AVX512-NEXT:    # ymm10 = mem[0,1,0,1]
-; AVX512-NEXT:    vpshufb %ymm10, %ymm2, %ymm6
-; AVX512-NEXT:    vpor %ymm5, %ymm6, %ymm4
-; AVX512-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa64 %ymm21, %ymm4
-; AVX512-NEXT:    vpshufb %ymm4, %ymm2, %ymm2
-; AVX512-NEXT:    vmovdqa64 %ymm19, %ymm4
-; AVX512-NEXT:    vpshufb %ymm4, %ymm3, %ymm3
-; AVX512-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa (%rsi), %xmm3
-; AVX512-NEXT:    vpshufb %xmm7, %xmm3, %xmm2
-; AVX512-NEXT:    vmovdqa64 %xmm3, %xmm24
-; AVX512-NEXT:    vmovdqa (%rdi), %xmm4
-; AVX512-NEXT:    vpshufb %xmm14, %xmm4, %xmm3
-; AVX512-NEXT:    vmovdqa64 %xmm4, %xmm21
-; AVX512-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX512-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa (%rcx), %xmm3
-; AVX512-NEXT:    vpshufb %xmm15, %xmm3, %xmm2
-; AVX512-NEXT:    vmovdqa64 %xmm3, %xmm19
-; AVX512-NEXT:    vmovdqa (%rdx), %xmm15
-; AVX512-NEXT:    vmovdqa64 %xmm23, %xmm3
-; AVX512-NEXT:    vpshufb %xmm3, %xmm15, %xmm3
-; AVX512-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX512-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa (%r9), %xmm14
-; AVX512-NEXT:    vpshufb %xmm12, %xmm14, %xmm2
-; AVX512-NEXT:    vmovdqa (%r8), %xmm12
-; AVX512-NEXT:    vmovdqa64 %xmm18, %xmm3
-; AVX512-NEXT:    vpshufb %xmm3, %xmm12, %xmm3
-; AVX512-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX512-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-NEXT:    vmovdqa64 %ymm29, %ymm2
-; AVX512-NEXT:    vpshufb %ymm2, %ymm11, %ymm2
-; AVX512-NEXT:    vmovdqa64 %ymm25, %ymm3
-; AVX512-NEXT:    vmovdqa64 %ymm20, %ymm4
-; AVX512-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
-; AVX512-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa64 %ymm16, %ymm2
-; AVX512-NEXT:    vpshufb %ymm2, %ymm4, %ymm2
-; AVX512-NEXT:    vpshufb %ymm0, %ymm11, %ymm0
-; AVX512-NEXT:    vporq %ymm2, %ymm0, %ymm29
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm0 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm13[25],zero,ymm13[23],zero,zero,zero,zero,ymm13[26],zero,ymm13[24],zero,zero,zero,zero
-; AVX512-NEXT:    vmovdqa64 %ymm22, %ymm3
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm2 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm3[23],zero,ymm3[21,22,23,26],zero,ymm3[24],zero,ymm3[28,29,26,27]
-; AVX512-NEXT:    vporq %ymm0, %ymm2, %ymm25
-; AVX512-NEXT:    vpshufb %ymm8, %ymm13, %ymm0
-; AVX512-NEXT:    vpshufb %ymm9, %ymm3, %ymm2
-; AVX512-NEXT:    vporq %ymm0, %ymm2, %ymm22
-; AVX512-NEXT:    vmovdqa64 %ymm17, %ymm2
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm2[25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero
+; AVX512-NEXT:    vpshufb %ymm1, %ymm2, %ymm0
+; AVX512-NEXT:    vmovdqa64 %ymm1, %ymm18
+; AVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128,20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128]
+; AVX512-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX512-NEXT:    vpshufb %ymm3, %ymm15, %ymm1
+; AVX512-NEXT:    vmovdqa64 %ymm3, %ymm19
+; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vmovdqa64 %ymm24, %ymm0
+; AVX512-NEXT:    vpshufb %ymm0, %ymm15, %ymm0
+; AVX512-NEXT:    vmovdqa64 %ymm25, %ymm1
 ; AVX512-NEXT:    vpshufb %ymm1, %ymm2, %ymm1
-; AVX512-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm1[23],zero,ymm1[23,24,25,26],zero,ymm1[24],zero,ymm1[30,31]
-; AVX512-NEXT:    vpshufb %ymm10, %ymm1, %ymm1
-; AVX512-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512-NEXT:    vmovdqa (%rdi), %xmm2
+; AVX512-NEXT:    vmovdqa (%rsi), %xmm1
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm12 = [u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128]
+; AVX512-NEXT:    vpshufb %xmm12, %xmm1, %xmm0
+; AVX512-NEXT:    vmovdqa64 %xmm1, %xmm16
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm14 = [u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9]
+; AVX512-NEXT:    vpshufb %xmm14, %xmm2, %xmm1
+; AVX512-NEXT:    vmovdqa64 %xmm2, %xmm28
+; AVX512-NEXT:    vpor %xmm0, %xmm1, %xmm0
 ; AVX512-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-NEXT:    vmovdqa (%rax), %ymm5
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm1 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm13 = [0,5,4,0,5,0,4,0,20,21,0,23,0,21,0,23]
-; AVX512-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm5[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512-NEXT:    vpermi2d %zmm1, %zmm2, %zmm13
-; AVX512-NEXT:    vmovdqa (%rax), %xmm11
-; AVX512-NEXT:    vpshufhw {{.*#+}} xmm2 = xmm11[0,1,2,3,4,5,5,6]
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX512-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
-; AVX512-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
-; AVX512-NEXT:    vmovdqa {{.*#+}} ymm10 = [128,13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128]
-; AVX512-NEXT:    vpshufb %ymm10, %ymm5, %ymm3
-; AVX512-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm16
-; AVX512-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; AVX512-NEXT:    vmovdqa64 %xmm31, %xmm7
-; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm7[0],xmm8[0],xmm7[1],xmm8[1],xmm7[2],xmm8[2],xmm7[3],xmm8[3],xmm7[4],xmm8[4],xmm7[5],xmm8[5],xmm7[6],xmm8[6],xmm7[7],xmm8[7]
+; AVX512-NEXT:    vmovdqa (%rdx), %xmm3
+; AVX512-NEXT:    vmovdqa (%rcx), %xmm0
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm11 = [u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u]
+; AVX512-NEXT:    vpshufb %xmm11, %xmm0, %xmm1
+; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm31
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm0 = [u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u]
+; AVX512-NEXT:    vpshufb %xmm0, %xmm3, %xmm2
+; AVX512-NEXT:    vmovdqa64 %xmm3, %xmm30
+; AVX512-NEXT:    vpor %xmm1, %xmm2, %xmm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vmovdqa (%r9), %xmm1
+; AVX512-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512-NEXT:    vmovdqa (%r8), %xmm3
+; AVX512-NEXT:    vpshufb {{.*#+}} xmm1 = zero,xmm1[4,u,u,u],zero,xmm1[7],zero,xmm1[5,u,u,u],zero,xmm1[8],zero,xmm1[6]
+; AVX512-NEXT:    vpshufb {{.*#+}} xmm2 = xmm3[4],zero,xmm3[u,u,u,7],zero,xmm3[5],zero,xmm3[u,u,u,8],zero,xmm3[6],zero
+; AVX512-NEXT:    vmovdqa64 %xmm3, %xmm23
+; AVX512-NEXT:    vpor %xmm1, %xmm2, %xmm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vpshufb {{.*#+}} ymm1 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm10[23],zero,zero,zero,zero,ymm10[26],zero,ymm10[24],zero,zero,zero,zero,ymm10[27],zero,ymm10[25]
+; AVX512-NEXT:    vmovdqa64 %ymm17, %ymm3
+; AVX512-NEXT:    vpshufb {{.*#+}} ymm2 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm3[23,u,u,u],zero,ymm3[26],zero,ymm3[24,u,u,u],zero,ymm3[27],zero
+; AVX512-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vmovdqa64 %ymm26, %ymm1
+; AVX512-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512-NEXT:    vpshufb %ymm13, %ymm10, %ymm2
+; AVX512-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vpshufb {{.*#+}} ymm1 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm8[25],zero,ymm8[23],zero,zero,zero,zero,ymm8[26],zero,ymm8[24],zero,zero,zero,zero
+; AVX512-NEXT:    vmovdqa64 %ymm22, %ymm2
+; AVX512-NEXT:    vpshufb %ymm2, %ymm7, %ymm2
+; AVX512-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vpshufb %ymm4, %ymm8, %ymm1
+; AVX512-NEXT:    vpshufb %ymm9, %ymm7, %ymm2
+; AVX512-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vmovdqa64 %ymm21, %ymm1
+; AVX512-NEXT:    vpshufb %ymm1, %ymm5, %ymm1
+; AVX512-NEXT:    vmovdqa64 %ymm18, %ymm2
+; AVX512-NEXT:    vpshufb %ymm2, %ymm5, %ymm2
+; AVX512-NEXT:    vinserti64x4 $1, %ymm1, %zmm2, %zmm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vpshufb {{.*#+}} ymm1 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm6[23],zero,ymm6[23,24,25,26],zero,ymm6[24],zero,ymm6[30,31]
+; AVX512-NEXT:    vmovdqa64 %ymm19, %ymm2
+; AVX512-NEXT:    vpshufb %ymm2, %ymm6, %ymm2
+; AVX512-NEXT:    vinserti64x4 $1, %ymm1, %zmm2, %zmm1
+; AVX512-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-NEXT:    vmovdqa (%rax), %ymm4
+; AVX512-NEXT:    vmovdqa64 %ymm20, %ymm1
+; AVX512-NEXT:    vpshufb %ymm1, %ymm4, %ymm2
+; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [0,5,4,0,5,0,4,0,20,21,0,23,0,21,0,23]
+; AVX512-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm4[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512-NEXT:    vpermi2d %zmm2, %zmm3, %zmm18
+; AVX512-NEXT:    vmovdqa 32(%rcx), %xmm13
+; AVX512-NEXT:    vpshufb %xmm11, %xmm13, %xmm2
+; AVX512-NEXT:    vmovdqa 32(%rdx), %xmm10
+; AVX512-NEXT:    vpshufb %xmm0, %xmm10, %xmm0
+; AVX512-NEXT:    vporq %xmm2, %xmm0, %xmm26
+; AVX512-NEXT:    vmovdqa 32(%rsi), %xmm11
+; AVX512-NEXT:    vpshufb %xmm12, %xmm11, %xmm0
+; AVX512-NEXT:    vmovdqa 32(%rdi), %xmm8
+; AVX512-NEXT:    vpshufb %xmm14, %xmm8, %xmm2
+; AVX512-NEXT:    vporq %xmm0, %xmm2, %xmm24
+; AVX512-NEXT:    vmovdqa 32(%rax), %xmm1
+; AVX512-NEXT:    vpmovsxbd {{.*#+}} zmm20 = [0,0,0,0,2,3,0,1,0,18,0,19,18,0,19,0]
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
+; AVX512-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,5,5,6]
+; AVX512-NEXT:    vpshufb %xmm3, %xmm1, %xmm2
+; AVX512-NEXT:    vmovdqa64 %xmm3, %xmm27
+; AVX512-NEXT:    vmovdqa64 %xmm1, %xmm25
+; AVX512-NEXT:    vpermi2d %zmm0, %zmm2, %zmm20
+; AVX512-NEXT:    vmovdqa 32(%r9), %xmm15
+; AVX512-NEXT:    vpshufb {{.*#+}} xmm0 = zero,xmm15[4,u,u,u],zero,xmm15[7],zero,xmm15[5,u,u,u],zero,xmm15[8],zero,xmm15[6]
+; AVX512-NEXT:    vmovdqa 32(%r8), %xmm5
+; AVX512-NEXT:    vpshufb {{.*#+}} xmm3 = xmm5[4],zero,xmm5[u,u,u,7],zero,xmm5[5],zero,xmm5[u,u,u,8],zero,xmm5[6],zero
+; AVX512-NEXT:    vporq %xmm0, %xmm3, %xmm22
+; AVX512-NEXT:    vmovdqa (%rax), %xmm2
+; AVX512-NEXT:    vpshufhw {{.*#+}} xmm3 = xmm2[0,1,2,3,4,5,5,6]
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX512-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,0,1]
+; AVX512-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
+; AVX512-NEXT:    vmovdqa {{.*#+}} ymm9 = [128,13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128]
+; AVX512-NEXT:    vpshufb %ymm9, %ymm4, %ymm6
+; AVX512-NEXT:    vinserti64x4 $1, %ymm6, %zmm3, %zmm3
+; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm6 = xmm8[0],xmm11[0],xmm8[1],xmm11[1],xmm8[2],xmm11[2],xmm8[3],xmm11[3],xmm8[4],xmm11[4],xmm8[5],xmm11[5],xmm8[6],xmm11[6],xmm8[7],xmm11[7]
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512-NEXT:    vpshufb %xmm0, %xmm3, %xmm3
-; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm26
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm9 = zmm0[2,3,2,3],zmm3[0,1,0,1]
-; AVX512-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; AVX512-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm0 = [4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512-NEXT:    vpshufb %xmm0, %xmm3, %xmm3
-; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm23
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm18 = zmm0[2,3,2,3],zmm3[0,1,0,1]
-; AVX512-NEXT:    vmovdqa64 %xmm30, %xmm4
-; AVX512-NEXT:    vmovdqa64 %xmm27, %xmm3
-; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm6 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm0 = [u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
 ; AVX512-NEXT:    vpshufb %xmm0, %xmm6, %xmm6
-; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm20
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm30 = zmm0[2,3,2,3],zmm6[0,1,0,1]
-; AVX512-NEXT:    vpshufb {{.*#+}} ymm0 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512-NEXT:    vmovdqa64 %ymm0, %ymm17
-; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm1[8],xmm2[8],xmm1[9],xmm2[9],xmm1[10],xmm2[10],xmm1[11],xmm2[11],xmm1[12],xmm2[12],xmm1[13],xmm2[13],xmm1[14],xmm2[14],xmm1[15],xmm2[15]
-; AVX512-NEXT:    vmovdqa64 %xmm19, %xmm2
-; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm2[8],xmm15[8],xmm2[9],xmm15[9],xmm2[10],xmm15[10],xmm2[11],xmm15[11],xmm2[12],xmm15[12],xmm2[13],xmm15[13],xmm2[14],xmm15[14],xmm2[15],xmm15[15]
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512-NEXT:    vpshufb %xmm1, %xmm6, %xmm0
-; AVX512-NEXT:    vmovdqa64 %ymm0, %ymm19
-; AVX512-NEXT:    vpshufb %xmm1, %xmm5, %xmm1
+; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm21
 ; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm31 = zmm0[0,1,0,1],zmm1[0,1,0,1]
-; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm8[8],xmm7[8],xmm8[9],xmm7[9],xmm8[10],xmm7[10],xmm8[11],xmm7[11],xmm8[12],xmm7[12],xmm8[13],xmm7[13],xmm8[14],xmm7[14],xmm8[15],xmm7[15]
-; AVX512-NEXT:    vmovdqa64 %xmm24, %xmm7
-; AVX512-NEXT:    vmovdqa64 %xmm21, %xmm8
-; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm7[8],xmm8[8],xmm7[9],xmm8[9],xmm7[10],xmm8[10],xmm7[11],xmm8[11],xmm7[12],xmm8[12],xmm7[13],xmm8[13],xmm7[14],xmm8[14],xmm7[15],xmm8[15]
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512-NEXT:    vpshufb %xmm1, %xmm6, %xmm6
-; AVX512-NEXT:    vpshufb %xmm1, %xmm5, %xmm1
+; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm6 = zmm0[2,3,2,3],zmm6[0,1,0,1]
+; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm7 = xmm10[0],xmm13[0],xmm10[1],xmm13[1],xmm10[2],xmm13[2],xmm10[3],xmm13[3],xmm10[4],xmm13[4],xmm10[5],xmm13[5],xmm10[6],xmm13[6],xmm10[7],xmm13[7]
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm14 = [4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX512-NEXT:    vpshufb %xmm14, %xmm7, %xmm7
 ; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm21 = zmm0[0,1,0,1],zmm1[0,1,0,1]
-; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm3[8],xmm4[8],xmm3[9],xmm4[9],xmm3[10],xmm4[10],xmm3[11],xmm4[11],xmm3[12],xmm4[12],xmm3[13],xmm4[13],xmm3[14],xmm4[14],xmm3[15],xmm4[15]
-; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm12[8],xmm14[8],xmm12[9],xmm14[9],xmm12[10],xmm14[10],xmm12[11],xmm14[11],xmm12[12],xmm14[12],xmm12[13],xmm14[13],xmm12[14],xmm14[14],xmm12[15],xmm14[15]
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX512-NEXT:    vpshufb %xmm1, %xmm5, %xmm5
+; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm7 = zmm0[2,3,2,3],zmm7[0,1,0,1]
+; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm5[0],xmm15[0],xmm5[1],xmm15[1],xmm5[2],xmm15[2],xmm5[3],xmm15[3],xmm5[4],xmm15[4],xmm5[5],xmm15[5],xmm5[6],xmm15[6],xmm5[7],xmm15[7]
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
 ; AVX512-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm4
+; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm19 # 64-byte Reload
+; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm19 = zmm19[2,3,2,3],zmm0[0,1,0,1]
+; AVX512-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512-NEXT:    vmovdqa64 %ymm0, %ymm17
 ; AVX512-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512-NEXT:    vpshufb %ymm10, %ymm0, %ymm1
-; AVX512-NEXT:    vpshuflw {{.*#+}} ymm10 = ymm0[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[0,1,1,3,4,5,5,7]
-; AVX512-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,2,3,2]
-; AVX512-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm10, %ymm10
-; AVX512-NEXT:    vinserti64x4 $1, %ymm10, %zmm1, %zmm10
-; AVX512-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm18 = zmm9 ^ (zmm1 & (zmm18 ^ zmm9))
-; AVX512-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX512-NEXT:    # ymm9 = mem[2,3,2,3]
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512-NEXT:    vinserti64x4 $1, %ymm9, %zmm0, %zmm9
-; AVX512-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm27 # 32-byte Folded Reload
-; AVX512-NEXT:    # ymm27 = mem[2,3,2,3]
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512-NEXT:    vinserti64x4 $1, %ymm27, %zmm0, %zmm27
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm27 = zmm9 ^ (zmm1 & (zmm27 ^ zmm9))
-; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm8[0],xmm7[0],xmm8[1],xmm7[1],xmm8[2],xmm7[2],xmm8[3],xmm7[3],xmm8[4],xmm7[4],xmm8[5],xmm7[5],xmm8[6],xmm7[6],xmm8[7],xmm7[7]
-; AVX512-NEXT:    vmovdqa64 %xmm26, %xmm0
-; AVX512-NEXT:    vpshufb %xmm0, %xmm1, %xmm1
-; AVX512-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm1 # 64-byte Folded Reload
-; AVX512-NEXT:    # zmm1 = zmm1[0,1,0,1],mem[0,1,0,1]
-; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm15[0],xmm2[0],xmm15[1],xmm2[1],xmm15[2],xmm2[2],xmm15[3],xmm2[3],xmm15[4],xmm2[4],xmm15[5],xmm2[5],xmm15[6],xmm2[6],xmm15[7],xmm2[7]
-; AVX512-NEXT:    vmovdqa64 %xmm23, %xmm0
-; AVX512-NEXT:    vpshufb %xmm0, %xmm9, %xmm9
-; AVX512-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9, %zmm9 # 64-byte Folded Reload
-; AVX512-NEXT:    # zmm9 = zmm9[0,1,0,1],mem[0,1,0,1]
-; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm12[0],xmm14[0],xmm12[1],xmm14[1],xmm12[2],xmm14[2],xmm12[3],xmm14[3],xmm12[4],xmm14[4],xmm12[5],xmm14[5],xmm12[6],xmm14[6],xmm12[7],xmm14[7]
-; AVX512-NEXT:    vmovdqa64 %xmm20, %xmm0
-; AVX512-NEXT:    vpshufb %xmm0, %xmm12, %xmm12
-; AVX512-NEXT:    vinserti32x4 $2, {{[-0-9]+}}(%r{{[sb]}}p), %zmm12, %zmm12 # 16-byte Folded Reload
-; AVX512-NEXT:    vpshufb {{.*#+}} xmm14 = xmm11[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
-; AVX512-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm11[1,1,0,0,4,5,6,7]
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm7 = xmm7[0,1,2,0]
-; AVX512-NEXT:    vinserti64x4 $1, %ymm14, %zmm7, %zmm7
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm14 # 64-byte Reload
-; AVX512-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm14, %zmm14 # 64-byte Folded Reload
-; AVX512-NEXT:    # zmm14 = zmm14[2,3,2,3],mem[2,3,2,3]
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm15 # 64-byte Reload
-; AVX512-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm15, %zmm15 # 64-byte Folded Reload
-; AVX512-NEXT:    # zmm15 = zmm15[2,3,2,3],mem[2,3,2,3]
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm15 = zmm15 ^ (mem & (zmm15 ^ zmm14))
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm14 # 64-byte Reload
-; AVX512-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm14, %zmm14 # 64-byte Folded Reload
-; AVX512-NEXT:    # zmm14 = zmm14[2,3,2,3],mem[2,3,2,3]
-; AVX512-NEXT:    vmovdqa64 {{.*#+}} zmm23 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm14 = zmm14 ^ (zmm23 & (zmm14 ^ zmm15))
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm21 = zmm31 ^ (zmm23 & (zmm21 ^ zmm31))
-; AVX512-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
-; AVX512-NEXT:    vpermq {{.*#+}} ymm11 = ymm19[0,1,0,1]
-; AVX512-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
-; AVX512-NEXT:    vpermq {{.*#+}} ymm0 = ymm17[2,3,2,3]
-; AVX512-NEXT:    vpshuflw $5, {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
-; AVX512-NEXT:    # xmm15 = mem[1,1,0,0,4,5,6,7]
+; AVX512-NEXT:    vpshufb %ymm9, %ymm0, %ymm9
+; AVX512-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm0[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,1,1,3,4,5,5,7]
+; AVX512-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,3,2]
+; AVX512-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX512-NEXT:    vinserti64x4 $1, %ymm0, %zmm9, %zmm9
+; AVX512-NEXT:    vmovdqa64 %xmm28, %xmm4
+; AVX512-NEXT:    vmovdqa64 %xmm16, %xmm0
+; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm12 = xmm0[8],xmm4[8],xmm0[9],xmm4[9],xmm0[10],xmm4[10],xmm0[11],xmm4[11],xmm0[12],xmm4[12],xmm0[13],xmm4[13],xmm0[14],xmm4[14],xmm0[15],xmm4[15]
+; AVX512-NEXT:    vmovdqa64 %xmm12, %xmm16
+; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3],xmm4[4],xmm0[4],xmm4[5],xmm0[5],xmm4[6],xmm0[6],xmm4[7],xmm0[7]
+; AVX512-NEXT:    vmovdqa64 %xmm21, %xmm0
+; AVX512-NEXT:    vpshufb %xmm0, %xmm4, %xmm4
+; AVX512-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm28 # 64-byte Folded Reload
+; AVX512-NEXT:    # zmm28 = zmm4[0,1,0,1],mem[0,1,0,1]
+; AVX512-NEXT:    vmovdqa64 %xmm30, %xmm0
+; AVX512-NEXT:    vmovdqa64 %xmm31, %xmm12
+; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm12[8],xmm0[8],xmm12[9],xmm0[9],xmm12[10],xmm0[10],xmm12[11],xmm0[11],xmm12[12],xmm0[12],xmm12[13],xmm0[13],xmm12[14],xmm0[14],xmm12[15],xmm0[15]
+; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm0[0],xmm12[0],xmm0[1],xmm12[1],xmm0[2],xmm12[2],xmm0[3],xmm12[3],xmm0[4],xmm12[4],xmm0[5],xmm12[5],xmm0[6],xmm12[6],xmm0[7],xmm12[7]
+; AVX512-NEXT:    vpshufb %xmm14, %xmm12, %xmm12
+; AVX512-NEXT:    vmovdqa64 {{.*#+}} zmm14 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm7 = zmm6 ^ (zmm14 & (zmm7 ^ zmm6))
+; AVX512-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
+; AVX512-NEXT:    # ymm6 = mem[2,3,2,3]
+; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm31 # 64-byte Reload
+; AVX512-NEXT:    vinserti64x4 $1, %ymm6, %zmm31, %zmm6
+; AVX512-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm31 # 32-byte Folded Reload
+; AVX512-NEXT:    # ymm31 = mem[2,3,2,3]
+; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm30 # 64-byte Reload
+; AVX512-NEXT:    vinserti64x4 $1, %ymm31, %zmm30, %zmm31
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm31 = zmm6 ^ (zmm14 & (zmm31 ^ zmm6))
+; AVX512-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm12, %zmm6 # 64-byte Folded Reload
+; AVX512-NEXT:    # zmm6 = zmm12[0,1,0,1],mem[0,1,0,1]
+; AVX512-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX512-NEXT:    vmovdqa64 %xmm23, %xmm14
+; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm12 = xmm14[8],xmm0[8],xmm14[9],xmm0[9],xmm14[10],xmm0[10],xmm14[11],xmm0[11],xmm14[12],xmm0[12],xmm14[13],xmm0[13],xmm14[14],xmm0[14],xmm14[15],xmm0[15]
+; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm14 = xmm14[0],xmm0[0],xmm14[1],xmm0[1],xmm14[2],xmm0[2],xmm14[3],xmm0[3],xmm14[4],xmm0[4],xmm14[5],xmm0[5],xmm14[6],xmm0[6],xmm14[7],xmm0[7]
+; AVX512-NEXT:    vpshufb %xmm1, %xmm14, %xmm14
+; AVX512-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm14, %zmm23 # 64-byte Folded Reload
+; AVX512-NEXT:    # zmm23 = zmm14[0,1,0,1],mem[0,1,0,1]
+; AVX512-NEXT:    vmovdqa64 %xmm27, %xmm0
+; AVX512-NEXT:    vpshufb %xmm0, %xmm2, %xmm14
+; AVX512-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm2[1,1,0,0,4,5,6,7]
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,2,0]
+; AVX512-NEXT:    vinserti64x4 $1, %ymm14, %zmm1, %zmm1
+; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm10 = xmm13[8],xmm10[8],xmm13[9],xmm10[9],xmm13[10],xmm10[10],xmm13[11],xmm10[11],xmm13[12],xmm10[12],xmm13[13],xmm10[13],xmm13[14],xmm10[14],xmm13[15],xmm10[15]
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm13 = [6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512-NEXT:    vpshufb %xmm13, %xmm4, %xmm4
+; AVX512-NEXT:    vpshufb %xmm13, %xmm10, %xmm10
+; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm10 = zmm26[0,1,0,1],zmm10[0,1,0,1]
+; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm11[8],xmm8[8],xmm11[9],xmm8[9],xmm11[10],xmm8[10],xmm11[11],xmm8[11],xmm11[12],xmm8[12],xmm11[13],xmm8[13],xmm11[14],xmm8[14],xmm11[15],xmm8[15]
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm11 = [2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512-NEXT:    vmovdqa64 %xmm16, %xmm0
+; AVX512-NEXT:    vpshufb %xmm11, %xmm0, %xmm0
+; AVX512-NEXT:    vpshufb %xmm11, %xmm8, %xmm8
+; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm24[0,1,0,1],zmm8[0,1,0,1]
+; AVX512-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm5[8],xmm15[8],xmm5[9],xmm15[9],xmm5[10],xmm15[10],xmm5[11],xmm15[11],xmm5[12],xmm15[12],xmm5[13],xmm15[13],xmm5[14],xmm15[14],xmm5[15],xmm15[15]
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm5 = [u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX512-NEXT:    vpshufb %xmm5, %xmm12, %xmm11
+; AVX512-NEXT:    vpshufb %xmm5, %xmm2, %xmm2
+; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm2 = zmm22[0,1,0,1],zmm2[0,1,0,1]
+; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
+; AVX512-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5, %zmm5 # 64-byte Folded Reload
+; AVX512-NEXT:    # zmm5 = zmm5[2,3,2,3],mem[2,3,2,3]
+; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm12 # 64-byte Reload
+; AVX512-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm12, %zmm12 # 64-byte Folded Reload
+; AVX512-NEXT:    # zmm12 = zmm12[2,3,2,3],mem[2,3,2,3]
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm12 = zmm12 ^ (mem & (zmm12 ^ zmm5))
+; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
+; AVX512-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5, %zmm5 # 64-byte Folded Reload
+; AVX512-NEXT:    # zmm5 = zmm5[2,3,2,3],mem[2,3,2,3]
+; AVX512-NEXT:    vmovdqa64 {{.*#+}} zmm13 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm5 = zmm5 ^ (zmm13 & (zmm5 ^ zmm12))
+; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm12 # 64-byte Reload
+; AVX512-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm12, %zmm12 # 64-byte Folded Reload
+; AVX512-NEXT:    # zmm12 = zmm12[2,3,2,3],mem[2,3,2,3]
+; AVX512-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX512-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,0,1]
+; AVX512-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,0,1]
+; AVX512-NEXT:    vpermq {{.*#+}} ymm14 = ymm17[2,3,2,3]
+; AVX512-NEXT:    vmovdqa64 %xmm25, %xmm15
+; AVX512-NEXT:    vpshuflw {{.*#+}} xmm15 = xmm15[1,1,0,0,4,5,6,7]
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm15 = xmm15[0,1,2,0]
-; AVX512-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm29, %zmm20 # 64-byte Folded Reload
-; AVX512-NEXT:    # zmm20 = zmm29[2,3,2,3],mem[2,3,2,3]
-; AVX512-NEXT:    vshufi64x2 {{.*#+}} zmm17 = zmm22[2,3,2,3],zmm25[2,3,2,3]
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm17 = zmm20 ^ (zmm23 & (zmm17 ^ zmm20))
-; AVX512-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm6, %zmm6 # 32-byte Folded Reload
-; AVX512-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm11, %zmm11 # 32-byte Folded Reload
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm11 = zmm11 ^ (mem & (zmm11 ^ zmm6))
-; AVX512-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5, %zmm5 # 32-byte Folded Reload
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 | (zmm5 & mem)
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 ^ (mem & (zmm16 ^ zmm11))
-; AVX512-NEXT:    vpermq {{.*#+}} ymm5 = ymm15[0,0,1,0]
-; AVX512-NEXT:    vinserti64x4 $1, %ymm5, %zmm0, %zmm5
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm5 = zmm5 ^ (mem & (zmm5 ^ zmm30))
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm5 = zmm5 ^ (mem & (zmm5 ^ zmm18))
-; AVX512-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Folded Reload
-; AVX512-NEXT:    # zmm3 = mem[2,3,2,3,6,7,6,7]
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm3 ^ (mem & (zmm3 ^ zmm14))
-; AVX512-NEXT:    vpermq {{.*#+}} zmm0 = zmm4[0,1,0,1,4,5,4,5]
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm28 ^ (mem & (zmm0 ^ zmm28))
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm21))
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
-; AVX512-NEXT:    vshufi64x2 $84, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm2 # 64-byte Folded Reload
-; AVX512-NEXT:    # zmm2 = zmm2[0,1,2,3],mem[2,3,2,3]
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm10 = zmm10 | (zmm2 & mem)
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm10 = zmm10 ^ (mem & (zmm10 ^ zmm27))
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm9 ^ (mem & (zmm9 ^ zmm1))
-; AVX512-NEXT:    vpermq {{.*#+}} zmm1 = zmm12[0,1,0,1,4,5,4,5]
-; AVX512-NEXT:    vpermq {{.*#+}} zmm2 = zmm7[0,0,1,0,4,4,5,4]
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm2 = zmm2 ^ (mem & (zmm2 ^ zmm1))
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm2 = zmm2 ^ (mem & (zmm2 ^ zmm9))
-; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512-NEXT:    vporq {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm1 # 64-byte Folded Reload
-; AVX512-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[2,3,2,3,6,7,6,7]
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm13 = zmm13 ^ (mem & (zmm13 ^ zmm1))
-; AVX512-NEXT:    vpternlogq {{.*#+}} zmm13 = zmm13 ^ (mem & (zmm13 ^ zmm17))
+; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm17 # 64-byte Reload
+; AVX512-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm17, %zmm17 # 64-byte Folded Reload
+; AVX512-NEXT:    # zmm17 = zmm17[2,3,2,3],mem[2,3,2,3]
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm17 = zmm12 ^ (zmm13 & (zmm17 ^ zmm12))
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm8 = zmm10 ^ (zmm13 & (zmm8 ^ zmm10))
+; AVX512-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 32-byte Folded Reload
+; AVX512-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm4 # 32-byte Folded Reload
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm4 = zmm4 ^ (mem & (zmm4 ^ zmm0))
+; AVX512-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm11, %zmm0 # 32-byte Folded Reload
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm3 | (zmm0 & mem)
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm3 ^ (mem & (zmm3 ^ zmm4))
+; AVX512-NEXT:    vpermq {{.*#+}} ymm0 = ymm15[0,0,1,0]
+; AVX512-NEXT:    vinserti64x4 $1, %ymm0, %zmm14, %zmm0
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm19))
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm7))
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm29 = zmm29 ^ (mem & (zmm29 ^ zmm5))
+; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512-NEXT:    vshufi64x2 $84, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm4 # 64-byte Folded Reload
+; AVX512-NEXT:    # zmm4 = zmm4[0,1,2,3],mem[2,3,2,3]
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm9 | (zmm4 & mem)
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm9 ^ (mem & (zmm9 ^ zmm31))
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm6 = zmm6 ^ (mem & (zmm6 ^ zmm28))
+; AVX512-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[0,0,1,0,4,4,5,4]
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm1 = zmm1 ^ (mem & (zmm1 ^ zmm23))
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm1 = zmm1 ^ (mem & (zmm1 ^ zmm6))
+; AVX512-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512-NEXT:    vporq {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm4 # 64-byte Folded Reload
+; AVX512-NEXT:    vpermq {{.*#+}} zmm4 = zmm4[2,3,2,3,6,7,6,7]
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm18 = zmm18 ^ (mem & (zmm18 ^ zmm4))
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm18 = zmm18 ^ (mem & (zmm18 ^ zmm17))
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm2 = zmm20 ^ (mem & (zmm2 ^ zmm20))
+; AVX512-NEXT:    vpternlogq {{.*#+}} zmm2 = zmm2 ^ (mem & (zmm2 ^ zmm8))
 ; AVX512-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512-NEXT:    vmovdqa64 %zmm13, 128(%rax)
-; AVX512-NEXT:    vmovdqa64 %zmm2, (%rax)
-; AVX512-NEXT:    vmovdqa64 %zmm10, 320(%rax)
-; AVX512-NEXT:    vmovdqa64 %zmm0, 256(%rax)
-; AVX512-NEXT:    vmovdqa64 %zmm3, 384(%rax)
-; AVX512-NEXT:    vmovdqa64 %zmm5, 192(%rax)
-; AVX512-NEXT:    vmovdqa64 %zmm16, 64(%rax)
-; AVX512-NEXT:    addq $1560, %rsp # imm = 0x618
+; AVX512-NEXT:    vmovdqa64 %zmm2, 256(%rax)
+; AVX512-NEXT:    vmovdqa64 %zmm18, 128(%rax)
+; AVX512-NEXT:    vmovdqa64 %zmm1, (%rax)
+; AVX512-NEXT:    vmovdqa64 %zmm9, 320(%rax)
+; AVX512-NEXT:    vmovdqa64 %zmm29, 384(%rax)
+; AVX512-NEXT:    vmovdqa64 %zmm0, 192(%rax)
+; AVX512-NEXT:    vmovdqa64 %zmm3, 64(%rax)
+; AVX512-NEXT:    addq $1496, %rsp # imm = 0x5D8
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
 ;
 ; AVX512-FCP-LABEL: store_i8_stride7_vf64:
 ; AVX512-FCP:       # %bb.0:
-; AVX512-FCP-NEXT:    subq $1640, %rsp # imm = 0x668
-; AVX512-FCP-NEXT:    vmovdqa (%rsi), %ymm7
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm7[14],zero,zero,zero,zero,zero,zero,ymm7[15],zero,zero,zero,zero,zero,zero,ymm7[16],zero,zero,zero,zero,zero,zero,ymm7[17],zero,zero,zero,zero,zero,zero,ymm7[18]
-; AVX512-FCP-NEXT:    vmovdqa (%rdi), %ymm14
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm14[0,1,14],zero,ymm14[12,13,0,1,14,15],zero,ymm14[3,12,13,2,3,16],zero,ymm14[30,31,28,29,16,17],zero,ymm14[31,18,19,28,29,18],zero
+; AVX512-FCP-NEXT:    subq $1416, %rsp # imm = 0x588
+; AVX512-FCP-NEXT:    vmovdqa (%rsi), %ymm12
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm12[14],zero,zero,zero,zero,zero,zero,ymm12[15],zero,zero,zero,zero,zero,zero,ymm12[16],zero,zero,zero,zero,zero,zero,ymm12[17],zero,zero,zero,zero,zero,zero,ymm12[18]
+; AVX512-FCP-NEXT:    vmovdqa (%rdi), %ymm9
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm9[0,1,14],zero,ymm9[12,13,0,1,14,15],zero,ymm9[3,12,13,2,3,16],zero,ymm9[30,31,28,29,16,17],zero,ymm9[31,18,19,28,29,18],zero
 ; AVX512-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa (%rcx), %ymm2
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,ymm2[14],zero,zero,zero,zero,zero,zero,ymm2[15],zero,zero,zero,zero,zero,zero,ymm2[16],zero,zero,zero,zero,zero,zero,ymm2[17],zero,zero,zero,zero,zero
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm2, %ymm25
-; AVX512-FCP-NEXT:    vmovdqa (%rdx), %ymm3
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[0,1,0,1,14],zero,ymm3[14,15,0,1,14,15],zero,ymm3[13,14,15,16,17,16],zero,ymm3[30,31,30,31,16,17],zero,ymm3[31,28,29,30,31]
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm3, %ymm26
-; AVX512-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa (%r8), %ymm1
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,ymm1[14],zero,zero,zero,zero,zero,zero,ymm1[15],zero,zero,zero,zero,zero,zero,ymm1[16],zero,zero,zero,zero,zero,zero,ymm1[17],zero,zero,zero,zero
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm1, %ymm21
-; AVX512-FCP-NEXT:    vmovdqa (%r9), %ymm2
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[13,u,u,u,u,u],zero,ymm2[14,u,u,u,u,u],zero,ymm2[15,u,u,u,u,u],zero,ymm2[16,u,u,u,u,u],zero,ymm2[17,u,u,u]
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm2, %ymm28
-; AVX512-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-FCP-NEXT:    vmovdqa (%rcx), %ymm6
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero,zero
+; AVX512-FCP-NEXT:    vmovdqa (%rdx), %ymm4
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm4[0,1,0,1,14],zero,ymm4[14,15,0,1,14,15],zero,ymm4[13,14,15,16,17,16],zero,ymm4[30,31,30,31,16,17],zero,ymm4[31,28,29,30,31]
+; AVX512-FCP-NEXT:    vpor %ymm1, %ymm3, %ymm0
+; AVX512-FCP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX512-FCP-NEXT:    vmovdqa (%r8), %ymm8
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = zero,zero,zero,zero,zero,zero,ymm8[14],zero,zero,zero,zero,zero,zero,ymm8[15],zero,zero,zero,zero,zero,zero,ymm8[16],zero,zero,zero,zero,zero,zero,ymm8[17],zero,zero,zero,zero
+; AVX512-FCP-NEXT:    vmovdqa (%r9), %ymm11
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[13,u,u,u,u,u],zero,ymm11[14,u,u,u,u,u],zero,ymm11[15,u,u,u,u,u],zero,ymm11[16,u,u,u,u,u],zero,ymm11[17,u,u,u]
+; AVX512-FCP-NEXT:    vpor %ymm3, %ymm5, %ymm0
 ; AVX512-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa 32(%rsi), %ymm2
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero,zero,zero
-; AVX512-FCP-NEXT:    vmovdqa 32(%rdi), %ymm3
-; AVX512-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,30],zero,ymm3[28],zero,ymm3[30,31,30,31],zero,ymm3[29],zero,ymm3[31,28,29]
-; AVX512-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-FCP-NEXT:    vmovdqa 32(%rsi), %ymm0
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128]
+; AVX512-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm0, %ymm3
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm1, %ymm18
+; AVX512-FCP-NEXT:    vmovdqa %ymm0, %ymm1
+; AVX512-FCP-NEXT:    vmovdqa 32(%rdi), %ymm15
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [28,29,30,128,28,128,30,31,30,31,128,29,128,31,28,29,28,29,30,128,28,128,30,31,30,31,128,29,128,31,28,29]
+; AVX512-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm15, %ymm5
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm0, %ymm22
+; AVX512-FCP-NEXT:    vpor %ymm3, %ymm5, %ymm0
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm15 = [128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25]
-; AVX512-FCP-NEXT:    # ymm15 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm15, %ymm3, %ymm0
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128,25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128]
-; AVX512-FCP-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm3, %ymm2, %ymm1
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm3, %ymm16
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm2, %ymm18
-; AVX512-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25]
+; AVX512-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm2, %ymm15, %ymm3
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm2, %ymm20
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128,25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128]
+; AVX512-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm2, %ymm1, %ymm5
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm2, %ymm25
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm1, %ymm19
+; AVX512-FCP-NEXT:    vpor %ymm3, %ymm5, %ymm0
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa 32(%rdx), %ymm2
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero,zero
-; AVX512-FCP-NEXT:    vmovdqa 32(%rcx), %ymm11
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u],zero,ymm11[30],zero,ymm11[28,u,u,u],zero,ymm11[31],zero,ymm11[29,u]
-; AVX512-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-FCP-NEXT:    vmovdqa 32(%rdx), %ymm13
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128]
+; AVX512-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm13, %ymm5
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm0, %ymm24
+; AVX512-FCP-NEXT:    vmovdqa 32(%rcx), %ymm3
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0]
+; AVX512-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm3, %ymm7
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm0, %ymm17
+; AVX512-FCP-NEXT:    vpor %ymm5, %ymm7, %ymm0
+; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128]
+; AVX512-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm2, %ymm3, %ymm5
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm2, %ymm26
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27,24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27]
+; AVX512-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm2, %ymm13, %ymm7
+; AVX512-FCP-NEXT:    vpor %ymm5, %ymm7, %ymm0
+; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-FCP-NEXT:    vmovdqa 32(%r8), %ymm14
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [128,27,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,27,128,128,128,128,30,128,28,128,128,128,128,31,128,29]
+; AVX512-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm14, %ymm5
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm0, %ymm21
+; AVX512-FCP-NEXT:    vmovdqa 32(%r9), %ymm10
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128]
+; AVX512-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm10, %ymm7
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm0, %ymm23
+; AVX512-FCP-NEXT:    vpor %ymm5, %ymm7, %ymm0
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128]
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128]
+; AVX512-FCP-NEXT:    # ymm5 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm5, %ymm10, %ymm0
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [24,25,24,25,128,23,128,23,24,25,26,128,24,128,30,31,24,25,24,25,128,23,128,23,24,25,26,128,24,128,30,31]
 ; AVX512-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm11, %ymm0
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm1, %ymm30
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27,24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27]
-; AVX512-FCP-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm3, %ymm2, %ymm1
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm3, %ymm23
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm2, %ymm27
-; AVX512-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm14, %ymm7
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm1, %ymm27
+; AVX512-FCP-NEXT:    vpor %ymm0, %ymm7, %ymm0
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa 32(%r8), %ymm2
-; AVX512-FCP-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29]
-; AVX512-FCP-NEXT:    vmovdqa 32(%r9), %ymm3
-; AVX512-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm3[27,u,u,u],zero,ymm3[30],zero,ymm3[28,u,u,u],zero,ymm3[31],zero
-; AVX512-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX512-FCP-NEXT:    vmovdqa 32(%rax), %ymm0
+; AVX512-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512-FCP-NEXT:    # ymm7 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm7, %ymm0, %ymm1
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm7, %ymm16
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm7 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm7[2,3,2,3],zmm1[2,3,2,3]
+; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm20, %ymm0
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm9, %ymm0
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm25, %ymm1
+; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm12, %ymm7
+; AVX512-FCP-NEXT:    vpor %ymm0, %ymm7, %ymm0
+; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm12[21],zero,ymm12[19],zero,zero,zero,zero,ymm12[22],zero,ymm12[20],zero,zero
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm7 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21],zero,ymm9[19],zero,ymm9[21,20,21,22],zero,ymm9[20],zero,ymm9[22,23]
+; AVX512-FCP-NEXT:    vpor %ymm0, %ymm7, %ymm0
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128]
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm26, %ymm0
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm6, %ymm0
+; AVX512-FCP-NEXT:    vpshufb %ymm2, %ymm4, %ymm7
+; AVX512-FCP-NEXT:    vpor %ymm0, %ymm7, %ymm0
+; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20]
+; AVX512-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm6, %ymm0
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm1, %ymm26
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128,18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128]
 ; AVX512-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm3, %ymm0
+; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm4, %ymm7
 ; AVX512-FCP-NEXT:    vmovdqa64 %ymm1, %ymm20
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm2[23],zero,ymm2[23,24,25,26],zero,ymm2[24],zero,ymm2[30,31]
-; AVX512-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-FCP-NEXT:    vpor %ymm0, %ymm7, %ymm0
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512-FCP-NEXT:    vmovdqa 32(%rax), %ymm1
-; AVX512-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512-FCP-NEXT:    vpshufb %ymm5, %ymm11, %ymm0
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm11[20],zero,ymm11[18],zero,zero,zero,zero,ymm11[21],zero,ymm11[19],zero,zero,zero,zero,ymm11[22]
+; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm0, %zmm5, %zmm0
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa 32(%rdx), %xmm2
-; AVX512-FCP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa 32(%rcx), %xmm0
-; AVX512-FCP-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u]
-; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm4 = [u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u]
-; AVX512-FCP-NEXT:    vpshufb %xmm4, %xmm2, %xmm5
-; AVX512-FCP-NEXT:    vpor %xmm0, %xmm5, %xmm0
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm27, %ymm0
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm8, %ymm0
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm5 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,ymm8[18],zero,ymm8[20,21,20,21],zero,ymm8[19],zero,ymm8[19,20,21,22],zero
+; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm0, %zmm5, %zmm0
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa 32(%rdi), %xmm2
-; AVX512-FCP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa 32(%rsi), %xmm0
-; AVX512-FCP-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm5 = [u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128]
-; AVX512-FCP-NEXT:    vpshufb %xmm5, %xmm0, %xmm0
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm10 = [u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9]
-; AVX512-FCP-NEXT:    vpshufb %xmm10, %xmm2, %xmm6
-; AVX512-FCP-NEXT:    vpor %xmm0, %xmm6, %xmm0
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm18, %ymm0
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm12, %ymm0
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm22, %ymm1
+; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm9, %ymm5
+; AVX512-FCP-NEXT:    vporq %ymm0, %ymm5, %ymm30
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm24, %ymm0
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm17, %ymm1
+; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm6, %ymm2
+; AVX512-FCP-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm2 = [0,0,0,0,2,3,0,1,0,18,0,19,18,0,19,0]
-; AVX512-FCP-NEXT:    vmovdqa 32(%rax), %xmm0
-; AVX512-FCP-NEXT:    vpshufhw {{.*#+}} xmm6 = xmm0[0,1,2,3,4,5,5,6]
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm9 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
-; AVX512-FCP-NEXT:    vpshufb %xmm9, %xmm0, %xmm8
-; AVX512-FCP-NEXT:    vpermi2d %zmm6, %zmm8, %zmm2
-; AVX512-FCP-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa 32(%r9), %xmm2
-; AVX512-FCP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa 32(%r8), %xmm3
-; AVX512-FCP-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512-FCP-NEXT:    vmovdqa (%rax), %ymm4
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm16, %ymm0
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm29 = [4,0,6,0,4,0,6,7,0,17,0,17,0,16,16,0]
+; AVX512-FCP-NEXT:    vmovdqa 32(%rax), %xmm2
+; AVX512-FCP-NEXT:    vpshuflw {{.*#+}} xmm5 = xmm2[1,1,0,0,4,5,6,7]
+; AVX512-FCP-NEXT:    vpermi2d %zmm5, %zmm0, %zmm29
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm21, %ymm0
+; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm8, %ymm0
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm23, %ymm1
+; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm11, %ymm1
+; AVX512-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-FCP-NEXT:    vmovdqa 32(%rdx), %xmm5
+; AVX512-FCP-NEXT:    vmovdqa 32(%rcx), %xmm1
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm8 = [u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u]
+; AVX512-FCP-NEXT:    vpshufb %xmm8, %xmm1, %xmm0
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm1, %xmm24
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} xmm1 = xmm5[u,u,u,7],zero,xmm5[5],zero,xmm5[u,u,u,8],zero,xmm5[6],zero,xmm5[u,u]
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm5, %xmm23
+; AVX512-FCP-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-FCP-NEXT:    vmovdqa 32(%rdi), %xmm7
+; AVX512-FCP-NEXT:    vmovdqa 32(%rsi), %xmm12
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128]
+; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm12, %xmm5
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm0 = [u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9]
+; AVX512-FCP-NEXT:    vpshufb %xmm0, %xmm7, %xmm6
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm7, %xmm21
+; AVX512-FCP-NEXT:    vpor %xmm5, %xmm6, %xmm5
+; AVX512-FCP-NEXT:    vmovdqu64 %zmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512-FCP-NEXT:    vpshufhw {{.*#+}} xmm5 = xmm2[0,1,2,3,4,5,5,6]
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm6 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
+; AVX512-FCP-NEXT:    vpshufb %xmm6, %xmm2, %xmm2
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm6, %xmm16
+; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm25 = [0,0,0,0,2,3,0,1,0,18,0,19,18,0,19,0]
+; AVX512-FCP-NEXT:    vpermi2d %zmm5, %zmm2, %zmm25
+; AVX512-FCP-NEXT:    vmovdqa 32(%r9), %xmm11
+; AVX512-FCP-NEXT:    vmovdqa 32(%r8), %xmm5
 ; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm6 = [128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6]
-; AVX512-FCP-NEXT:    vpshufb %xmm6, %xmm2, %xmm12
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm8 = [4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128]
-; AVX512-FCP-NEXT:    vpshufb %xmm8, %xmm3, %xmm13
-; AVX512-FCP-NEXT:    vpor %xmm12, %xmm13, %xmm2
+; AVX512-FCP-NEXT:    vpshufb %xmm6, %xmm11, %xmm2
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm6, %xmm17
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm7 = [4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128]
+; AVX512-FCP-NEXT:    vpshufb %xmm7, %xmm5, %xmm6
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm7, %xmm18
+; AVX512-FCP-NEXT:    vporq %xmm2, %xmm6, %xmm31
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm6 = ymm13[0,1,0,1,14],zero,ymm13[14,15,0,1,14,15],zero,ymm13[13,14,15,16,17,16],zero,ymm13[30,31,30,31,16,17],zero,ymm13[31,28,29,30,31]
+; AVX512-FCP-NEXT:    vpor %ymm2, %ymm6, %ymm2
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa (%rsi), %xmm2
-; AVX512-FCP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-FCP-NEXT:    vpshufb %xmm5, %xmm2, %xmm5
-; AVX512-FCP-NEXT:    vmovdqa (%rdi), %xmm2
-; AVX512-FCP-NEXT:    vpshufb %xmm10, %xmm2, %xmm10
-; AVX512-FCP-NEXT:    vmovdqa64 %xmm2, %xmm22
-; AVX512-FCP-NEXT:    vpor %xmm5, %xmm10, %xmm2
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm26, %ymm2
+; AVX512-FCP-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm20, %ymm3
+; AVX512-FCP-NEXT:    vpshufb %ymm3, %ymm13, %ymm3
+; AVX512-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm2
+; AVX512-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm19, %ymm6
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero,zero,zero,ymm6[18]
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm15[0,1,14],zero,ymm15[12,13,0,1,14,15],zero,ymm15[3,12,13,2,3,16],zero,ymm15[30,31,28,29,16,17],zero,ymm15[31,18,19,28,29,18],zero
+; AVX512-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa (%rcx), %xmm2
-; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
-; AVX512-FCP-NEXT:    vmovdqa64 %xmm2, %xmm17
-; AVX512-FCP-NEXT:    vmovdqa (%rdx), %xmm12
-; AVX512-FCP-NEXT:    vpshufb %xmm4, %xmm12, %xmm4
-; AVX512-FCP-NEXT:    vpor %xmm1, %xmm4, %xmm1
-; AVX512-FCP-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa (%rax), %xmm5
-; AVX512-FCP-NEXT:    vpshufb %xmm9, %xmm5, %xmm4
-; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,1,0,1,0,0,0,0,16,0,16,0,18,19,0,17]
-; AVX512-FCP-NEXT:    vpshuflw {{.*#+}} xmm9 = xmm5[1,1,0,0,4,5,6,7]
-; AVX512-FCP-NEXT:    vpermi2d %zmm4, %zmm9, %zmm1
-; AVX512-FCP-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa (%r9), %xmm1
-; AVX512-FCP-NEXT:    vpshufb %xmm6, %xmm1, %xmm4
-; AVX512-FCP-NEXT:    vmovdqa64 %xmm1, %xmm31
-; AVX512-FCP-NEXT:    vmovdqa (%r8), %xmm1
-; AVX512-FCP-NEXT:    vpshufb %xmm8, %xmm1, %xmm8
-; AVX512-FCP-NEXT:    vmovdqa64 %xmm1, %xmm29
-; AVX512-FCP-NEXT:    vpor %xmm4, %xmm8, %xmm1
-; AVX512-FCP-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa %ymm14, %ymm3
-; AVX512-FCP-NEXT:    vpshufb %ymm15, %ymm14, %ymm4
-; AVX512-FCP-NEXT:    vmovdqa %ymm7, %ymm10
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm16, %ymm1
-; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm7, %ymm8
-; AVX512-FCP-NEXT:    vpor %ymm4, %ymm8, %ymm1
-; AVX512-FCP-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128]
-; AVX512-FCP-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm7, %ymm10, %ymm4
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [18,19,20,21,128,19,128,21,20,21,22,128,20,128,22,23,18,19,20,21,128,19,128,21,20,21,22,128,20,128,22,23]
-; AVX512-FCP-NEXT:    # ymm13 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm13, %ymm14, %ymm8
-; AVX512-FCP-NEXT:    vpor %ymm4, %ymm8, %ymm1
-; AVX512-FCP-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm25, %ymm1
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm30, %ymm2
-; AVX512-FCP-NEXT:    vpshufb %ymm2, %ymm1, %ymm4
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm26, %ymm6
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm23, %ymm2
-; AVX512-FCP-NEXT:    vpshufb %ymm2, %ymm6, %ymm8
-; AVX512-FCP-NEXT:    vpor %ymm4, %ymm8, %ymm2
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm6[21],zero,ymm6[19],zero,zero,zero,zero,ymm6[22],zero,ymm6[20],zero,zero
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21],zero,ymm15[19],zero,ymm15[21,20,21,22],zero,ymm15[20],zero,ymm15[22,23]
+; AVX512-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm2
+; AVX512-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm10[20],zero,ymm10[18],zero,zero,zero,zero,ymm10[21],zero,ymm10[19],zero,zero,zero,zero,ymm10[22]
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,ymm14[18],zero,ymm14[20,21,20,21],zero,ymm14[19],zero,ymm14[19,20,21,22],zero
+; AVX512-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20]
-; AVX512-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm2, %ymm1, %ymm4
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm2, %ymm30
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm25, %ymm2
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128,18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128]
-; AVX512-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm6, %ymm8
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm1, %ymm16
-; AVX512-FCP-NEXT:    vpor %ymm4, %ymm8, %ymm1
-; AVX512-FCP-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm28, %ymm1
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm20, %ymm4
-; AVX512-FCP-NEXT:    vpshufb %ymm4, %ymm1, %ymm4
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22]
-; AVX512-FCP-NEXT:    # ymm9 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm9, %ymm1, %ymm8
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm9, %ymm19
-; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm4, %zmm8, %zmm4
-; AVX512-FCP-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vpshuflw {{.*#+}} xmm8 = xmm0[1,1,0,0,4,5,6,7]
-; AVX512-FCP-NEXT:    vmovdqa (%rax), %ymm15
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm25 = [4,0,6,0,4,0,6,7,0,17,0,17,0,16,16,0]
-; AVX512-FCP-NEXT:    vpermi2d %zmm8, %zmm9, %zmm25
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm21, %ymm0
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm0[23],zero,ymm0[23,24,25,26],zero,ymm0[24],zero,ymm0[30,31]
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128,20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128]
-; AVX512-FCP-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpshufb %ymm8, %ymm0, %ymm14
-; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm9, %zmm14, %zmm4
-; AVX512-FCP-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm10[30],zero,ymm10[28],zero,zero,zero,zero,ymm10[31],zero,ymm10[29],zero,zero,zero
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm14 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,30],zero,ymm3[28],zero,ymm3[30,31,30,31],zero,ymm3[29],zero,ymm3[31,28,29]
-; AVX512-FCP-NEXT:    vporq %ymm9, %ymm14, %ymm26
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm6[30],zero,ymm6[28],zero,zero,zero,zero,ymm6[31],zero,ymm6[29],zero,zero
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm14 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u],zero,ymm2[30],zero,ymm2[28,u,u,u],zero,ymm2[31],zero,ymm2[29,u]
-; AVX512-FCP-NEXT:    vporq %ymm9, %ymm14, %ymm24
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[27],zero,zero,zero,zero,ymm0[30],zero,ymm0[28],zero,zero,zero,zero,ymm0[31],zero,ymm0[29]
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm14 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm1[27,u,u,u],zero,ymm1[30],zero,ymm1[28,u,u,u],zero,ymm1[31],zero
-; AVX512-FCP-NEXT:    vporq %ymm9, %ymm14, %ymm23
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = zero,zero,zero,zero,zero,ymm11[14],zero,zero,zero,zero,zero,zero,ymm11[15],zero,zero,zero,zero,zero,zero,ymm11[16],zero,zero,zero,zero,zero,zero,ymm11[17],zero,zero,zero,zero,zero
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm27, %ymm1
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm14 = ymm1[0,1,0,1,14],zero,ymm1[14,15,0,1,14,15],zero,ymm1[13,14,15,16,17,16],zero,ymm1[30,31,30,31,16,17],zero,ymm1[31,28,29,30,31]
-; AVX512-FCP-NEXT:    vpor %ymm9, %ymm14, %ymm0
-; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm30, %ymm0
-; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm11, %ymm2
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm16, %ymm0
-; AVX512-FCP-NEXT:    vpshufb %ymm0, %ymm1, %ymm3
-; AVX512-FCP-NEXT:    vporq %ymm2, %ymm3, %ymm30
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm18, %ymm4
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,ymm4[14],zero,zero,zero,zero,zero,zero,ymm4[15],zero,zero,zero,zero,zero,zero,ymm4[16],zero,zero,zero,zero,zero,zero,ymm4[17],zero,zero,zero,zero,zero,zero,ymm4[18]
-; AVX512-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm1[0,1,14],zero,ymm1[12,13,0,1,14,15],zero,ymm1[3,12,13,2,3,16],zero,ymm1[30,31,28,29,16,17],zero,ymm1[31,18,19,28,29,18],zero
-; AVX512-FCP-NEXT:    vporq %ymm2, %ymm3, %ymm28
-; AVX512-FCP-NEXT:    vpshufb %ymm7, %ymm4, %ymm2
-; AVX512-FCP-NEXT:    vpshufb %ymm13, %ymm1, %ymm3
-; AVX512-FCP-NEXT:    vporq %ymm2, %ymm3, %ymm27
-; AVX512-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm19, %ymm1
-; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm0, %ymm2
-; AVX512-FCP-NEXT:    vmovdqu (%rsp), %ymm1 # 32-byte Reload
-; AVX512-FCP-NEXT:    vpshufb %ymm8, %ymm1, %ymm3
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,ymm14[14],zero,zero,zero,zero,zero,zero,ymm14[15],zero,zero,zero,zero,zero,zero,ymm14[16],zero,zero,zero,zero,zero,zero,ymm14[17],zero,zero,zero,zero
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm10[13,u,u,u,u,u],zero,ymm10[14,u,u,u,u,u],zero,ymm10[15,u,u,u,u,u],zero,ymm10[16,u,u,u,u,u],zero,ymm10[17,u,u,u]
 ; AVX512-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512-FCP-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,ymm1[14],zero,zero,zero,zero,zero,zero,ymm1[15],zero,zero,zero,zero,zero,zero,ymm1[16],zero,zero,zero,zero,zero,zero,ymm1[17],zero,zero,zero,zero
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm0[13,u,u,u,u,u],zero,ymm0[14,u,u,u,u,u],zero,ymm0[15,u,u,u,u,u],zero,ymm0[16,u,u,u,u,u],zero,ymm0[17,u,u,u]
-; AVX512-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm0
-; AVX512-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512-FCP-NEXT:    vpshufhw {{.*#+}} xmm1 = xmm5[0,1,2,3,4,5,5,6]
-; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [2,2,3,3,2,2,3,3]
-; AVX512-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX512-FCP-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,ymm15[13],zero,zero,zero,zero,zero,zero,ymm15[14],zero,zero,zero,zero,zero,zero,ymm15[15],zero,zero,zero,zero,zero,zero,ymm15[16],zero,zero,zero,zero,zero,zero,ymm15[17],zero,zero
-; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm16
-; AVX512-FCP-NEXT:    vmovdqa64 %xmm17, %xmm8
-; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm8[8],xmm12[8],xmm8[9],xmm12[9],xmm8[10],xmm12[10],xmm8[11],xmm12[11],xmm8[12],xmm12[12],xmm8[13],xmm12[13],xmm8[14],xmm12[14],xmm8[15],xmm12[15]
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm3 = [6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512-FCP-NEXT:    vpshufb %xmm3, %xmm2, %xmm0
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm0, %ymm19
-; AVX512-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
-; AVX512-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm13[8],xmm9[8],xmm13[9],xmm9[9],xmm13[10],xmm9[10],xmm13[11],xmm9[11],xmm13[12],xmm9[12],xmm13[13],xmm9[13],xmm13[14],xmm9[14],xmm13[15],xmm9[15]
-; AVX512-FCP-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm20 = zmm0[0,1,0,1],zmm2[0,1,0,1]
-; AVX512-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; AVX512-FCP-NEXT:    vmovdqa64 %xmm22, %xmm1
-; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm11[8],xmm1[8],xmm11[9],xmm1[9],xmm11[10],xmm1[10],xmm11[11],xmm1[11],xmm11[12],xmm1[12],xmm11[13],xmm1[13],xmm11[14],xmm1[14],xmm11[15],xmm1[15]
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm0 = [2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512-FCP-NEXT:    vpshufb %xmm0, %xmm2, %xmm2
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm2, %ymm18
-; AVX512-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; AVX512-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm7[8],xmm5[8],xmm7[9],xmm5[9],xmm7[10],xmm5[10],xmm7[11],xmm5[11],xmm7[12],xmm5[12],xmm7[13],xmm5[13],xmm7[14],xmm5[14],xmm7[15],xmm5[15]
-; AVX512-FCP-NEXT:    vpshufb %xmm0, %xmm2, %xmm2
-; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm22 = zmm0[0,1,0,1],zmm2[0,1,0,1]
-; AVX512-FCP-NEXT:    vmovdqa64 %xmm31, %xmm3
-; AVX512-FCP-NEXT:    vmovdqa64 %xmm29, %xmm2
-; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm3[8],xmm2[9],xmm3[9],xmm2[10],xmm3[10],xmm2[11],xmm3[11],xmm2[12],xmm3[12],xmm2[13],xmm3[13],xmm2[14],xmm3[14],xmm2[15],xmm3[15]
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm0 = [u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX512-FCP-NEXT:    vpshufb %xmm0, %xmm4, %xmm4
-; AVX512-FCP-NEXT:    vmovdqa64 %ymm4, %ymm17
-; AVX512-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; AVX512-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm14 = xmm4[8],xmm6[8],xmm4[9],xmm6[9],xmm4[10],xmm6[10],xmm4[11],xmm6[11],xmm4[12],xmm6[12],xmm4[13],xmm6[13],xmm4[14],xmm6[14],xmm4[15],xmm6[15]
-; AVX512-FCP-NEXT:    vpshufb %xmm0, %xmm14, %xmm0
-; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm10 # 64-byte Reload
-; AVX512-FCP-NEXT:    vinserti32x4 $2, %xmm0, %zmm10, %zmm29
-; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm11[0],xmm1[1],xmm11[1],xmm1[2],xmm11[2],xmm1[3],xmm11[3],xmm1[4],xmm11[4],xmm1[5],xmm11[5],xmm1[6],xmm11[6],xmm1[7],xmm11[7]
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm14 = [0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512-FCP-NEXT:    vpshufb %xmm14, %xmm0, %xmm0
-; AVX512-FCP-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm21 # 64-byte Folded Reload
-; AVX512-FCP-NEXT:    # zmm21 = zmm0[0,1,0,1],mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm12[0],xmm8[0],xmm12[1],xmm8[1],xmm12[2],xmm8[2],xmm12[3],xmm8[3],xmm12[4],xmm8[4],xmm12[5],xmm8[5],xmm12[6],xmm8[6],xmm12[7],xmm8[7]
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm8 = [4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512-FCP-NEXT:    vpshufb %xmm8, %xmm0, %xmm0
-; AVX512-FCP-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm11 # 64-byte Folded Reload
-; AVX512-FCP-NEXT:    # zmm11 = zmm0[0,1,0,1],mem[0,1,0,1]
-; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3],xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX512-FCP-NEXT:    vinserti32x4 $2, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm31 # 16-byte Folded Reload
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm10 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512-FCP-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm15[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm2 = [4,5,4,5,5,7,4,5,20,21,22,23,20,21,22,23]
-; AVX512-FCP-NEXT:    vpermt2d %zmm10, %zmm2, %zmm0
-; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm10 = xmm5[0],xmm7[0],xmm5[1],xmm7[1],xmm5[2],xmm7[2],xmm5[3],xmm7[3],xmm5[4],xmm7[4],xmm5[5],xmm7[5],xmm5[6],xmm7[6],xmm5[7],xmm7[7]
-; AVX512-FCP-NEXT:    vpshufb %xmm14, %xmm10, %xmm10
-; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm10 = zmm26[2,3,2,3],zmm10[0,1,0,1]
-; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm14 = xmm9[0],xmm13[0],xmm9[1],xmm13[1],xmm9[2],xmm13[2],xmm9[3],xmm13[3],xmm9[4],xmm13[4],xmm9[5],xmm13[5],xmm9[6],xmm13[6],xmm9[7],xmm13[7]
-; AVX512-FCP-NEXT:    vpshufb %xmm8, %xmm14, %xmm14
-; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm14 = zmm24[2,3,2,3],zmm14[0,1,0,1]
-; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm15 = xmm4[0],xmm6[0],xmm4[1],xmm6[1],xmm4[2],xmm6[2],xmm4[3],xmm6[3],xmm4[4],xmm6[4],xmm4[5],xmm6[5],xmm4[6],xmm6[6],xmm4[7],xmm6[7]
-; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm15, %xmm12
-; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm12 = zmm23[2,3,2,3],zmm12[0,1,0,1]
-; AVX512-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = zero,ymm15[13],zero,zero,zero,zero,zero,zero,ymm15[14],zero,zero,zero,zero,zero,zero,ymm15[15],zero,zero,zero,zero,zero,zero,ymm15[16],zero,zero,zero,zero,zero,zero,ymm15[17],zero,zero
-; AVX512-FCP-NEXT:    vpshuflw {{.*#+}} ymm15 = ymm15[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512-FCP-NEXT:    vpermd %ymm15, %ymm2, %ymm4
-; AVX512-FCP-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm4, %ymm4
-; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm4, %zmm9, %zmm4
-; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm9 # 64-byte Folded Reload
-; AVX512-FCP-NEXT:    # zmm9 = zmm1[2,3,2,3],mem[2,3,2,3]
-; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm15 # 64-byte Folded Reload
-; AVX512-FCP-NEXT:    # zmm15 = zmm1[2,3,2,3],mem[2,3,2,3]
-; AVX512-FCP-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm15 = zmm9 ^ (zmm1 & (zmm15 ^ zmm9))
+; AVX512-FCP-NEXT:    vmovdqa (%rsi), %xmm10
+; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm10, %xmm1
+; AVX512-FCP-NEXT:    vmovdqa (%rdi), %xmm9
+; AVX512-FCP-NEXT:    vpshufb %xmm0, %xmm9, %xmm0
+; AVX512-FCP-NEXT:    vporq %xmm1, %xmm0, %xmm20
+; AVX512-FCP-NEXT:    vmovdqa (%rcx), %xmm13
+; AVX512-FCP-NEXT:    vpshufb %xmm8, %xmm13, %xmm0
+; AVX512-FCP-NEXT:    vmovdqa (%rdx), %xmm8
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} xmm1 = xmm8[u,u,u,7],zero,xmm8[5],zero,xmm8[u,u,u,8],zero,xmm8[6],zero,xmm8[u,u]
+; AVX512-FCP-NEXT:    vporq %xmm0, %xmm1, %xmm19
+; AVX512-FCP-NEXT:    vmovdqa (%rax), %xmm0
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm16, %xmm1
+; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
+; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm16 = [0,1,0,1,0,0,0,0,16,0,16,0,18,19,0,17]
+; AVX512-FCP-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm0[1,1,0,0,4,5,6,7]
+; AVX512-FCP-NEXT:    vpermi2d %zmm1, %zmm2, %zmm16
+; AVX512-FCP-NEXT:    vmovdqa (%r9), %xmm7
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm17, %xmm1
+; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm7, %xmm3
+; AVX512-FCP-NEXT:    vmovdqa (%r8), %xmm6
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm18, %xmm1
+; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm6, %xmm15
+; AVX512-FCP-NEXT:    vporq %xmm3, %xmm15, %xmm18
+; AVX512-FCP-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,5,6]
+; AVX512-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [2,2,3,3,2,2,3,3]
+; AVX512-FCP-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX512-FCP-NEXT:    vpermd %ymm0, %ymm3, %ymm0
+; AVX512-FCP-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128]
+; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm4, %ymm3
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm1, %ymm28
+; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm14
+; AVX512-FCP-NEXT:    vpshufb {{.*#+}} ymm15 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512-FCP-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm4[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm17 = [4,5,4,5,5,7,4,5,20,21,22,23,20,21,22,23]
+; AVX512-FCP-NEXT:    vpermt2d %zmm15, %zmm17, %zmm0
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm21, %xmm3
+; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm3[0],xmm12[0],xmm3[1],xmm12[1],xmm3[2],xmm12[2],xmm3[3],xmm12[3],xmm3[4],xmm12[4],xmm3[5],xmm12[5],xmm3[6],xmm12[6],xmm3[7],xmm12[7]
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
+; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm4, %xmm4
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm1, %xmm27
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm30 = zmm30[2,3,2,3],zmm4[0,1,0,1]
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm23, %xmm1
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm24, %xmm15
+; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm1[0],xmm15[0],xmm1[1],xmm15[1],xmm1[2],xmm15[2],xmm1[3],xmm15[3],xmm1[4],xmm15[4],xmm1[5],xmm15[5],xmm1[6],xmm15[6],xmm1[7],xmm15[7]
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX512-FCP-NEXT:    vpshufb %xmm2, %xmm4, %xmm4
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm2, %xmm22
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm23 = zmm2[2,3,2,3],zmm4[0,1,0,1]
+; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm5[0],xmm11[0],xmm5[1],xmm11[1],xmm5[2],xmm11[2],xmm5[3],xmm11[3],xmm5[4],xmm11[4],xmm5[5],xmm11[5],xmm5[6],xmm11[6],xmm5[7],xmm11[7]
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm2 = [u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
+; AVX512-FCP-NEXT:    vpshufb %xmm2, %xmm4, %xmm4
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm2, %xmm26
 ; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
-; AVX512-FCP-NEXT:    vporq {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm9 # 64-byte Folded Reload
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} zmm9 = zmm9[2,3,2,3,6,7,6,7]
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm4 = zmm2[2,3,2,3],zmm4[0,1,0,1]
+; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm15[8],xmm1[8],xmm15[9],xmm1[9],xmm15[10],xmm1[10],xmm15[11],xmm1[11],xmm15[12],xmm1[12],xmm15[13],xmm1[13],xmm15[14],xmm1[14],xmm15[15],xmm1[15]
+; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm15 = xmm13[8],xmm8[8],xmm13[9],xmm8[9],xmm13[10],xmm8[10],xmm13[11],xmm8[11],xmm13[12],xmm8[12],xmm13[13],xmm8[13],xmm13[14],xmm8[14],xmm13[15],xmm8[15]
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm2 = [6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512-FCP-NEXT:    vpshufb %xmm2, %xmm15, %xmm15
+; AVX512-FCP-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm24 = zmm2[0,1,0,1],zmm1[0,1,0,1]
+; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm12[8],xmm3[8],xmm12[9],xmm3[9],xmm12[10],xmm3[10],xmm12[11],xmm3[11],xmm12[12],xmm3[12],xmm12[13],xmm3[13],xmm12[14],xmm3[14],xmm12[15],xmm3[15]
+; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm12 = xmm10[8],xmm9[8],xmm10[9],xmm9[9],xmm10[10],xmm9[10],xmm10[11],xmm9[11],xmm10[12],xmm9[12],xmm10[13],xmm9[13],xmm10[14],xmm9[14],xmm10[15],xmm9[15]
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm2 = [2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512-FCP-NEXT:    vpshufb %xmm2, %xmm12, %xmm12
+; AVX512-FCP-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm21 = zmm2[0,1,0,1],zmm1[0,1,0,1]
+; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm5[8],xmm11[8],xmm5[9],xmm11[9],xmm5[10],xmm11[10],xmm5[11],xmm11[11],xmm5[12],xmm11[12],xmm5[13],xmm11[13],xmm5[14],xmm11[14],xmm5[15],xmm11[15]
+; AVX512-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm6[8],xmm7[8],xmm6[9],xmm7[9],xmm6[10],xmm7[10],xmm6[11],xmm7[11],xmm6[12],xmm7[12],xmm6[13],xmm7[13],xmm6[14],xmm7[14],xmm6[15],xmm7[15]
+; AVX512-FCP-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm5, %xmm11
+; AVX512-FCP-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm5 = zmm31[0,1,0,1],zmm1[0,1,0,1]
+; AVX512-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX512-FCP-NEXT:    vmovdqa64 %ymm28, %ymm1
+; AVX512-FCP-NEXT:    vpshufb %ymm1, %ymm2, %ymm1
+; AVX512-FCP-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm2[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512-FCP-NEXT:    vpermd %ymm2, %ymm17, %ymm2
+; AVX512-FCP-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3],xmm9[4],xmm10[4],xmm9[5],xmm10[5],xmm9[6],xmm10[6],xmm9[7],xmm10[7]
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm27, %xmm3
+; AVX512-FCP-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm2 = zmm2[0,1,0,1],zmm20[0,1,0,1]
+; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm8 = xmm8[0],xmm13[0],xmm8[1],xmm13[1],xmm8[2],xmm13[2],xmm8[3],xmm13[3],xmm8[4],xmm13[4],xmm8[5],xmm13[5],xmm8[6],xmm13[6],xmm8[7],xmm13[7]
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm22, %xmm3
+; AVX512-FCP-NEXT:    vpshufb %xmm3, %xmm8, %xmm8
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm8[0,1,0,1],zmm19[0,1,0,1]
+; AVX512-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1],xmm6[2],xmm7[2],xmm6[3],xmm7[3],xmm6[4],xmm7[4],xmm6[5],xmm7[5],xmm6[6],xmm7[6],xmm6[7],xmm7[7]
+; AVX512-FCP-NEXT:    vmovdqa64 %xmm26, %xmm3
+; AVX512-FCP-NEXT:    vpshufb %xmm3, %xmm6, %xmm6
+; AVX512-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm6 = zmm6[0,1,0,1],zmm18[0,1,0,1]
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
+; AVX512-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm7, %zmm7 # 64-byte Folded Reload
+; AVX512-FCP-NEXT:    # zmm7 = zmm7[2,3,2,3],mem[2,3,2,3]
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Reload
+; AVX512-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9, %zmm9 # 64-byte Folded Reload
+; AVX512-FCP-NEXT:    # zmm9 = zmm9[2,3,2,3],mem[2,3,2,3]
+; AVX512-FCP-NEXT:    vmovdqa64 {{.*#+}} zmm10 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm7 ^ (zmm10 & (zmm9 ^ zmm7))
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
+; AVX512-FCP-NEXT:    vporq {{[-0-9]+}}(%r{{[sb]}}p), %zmm7, %zmm7 # 64-byte Folded Reload
+; AVX512-FCP-NEXT:    vpermq {{.*#+}} zmm7 = zmm7[2,3,2,3,6,7,6,7]
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm7))
 ; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm9))
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm15))
 ; AVX512-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512-FCP-NEXT:    vmovdqa64 %zmm0, 128(%rax)
 ; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
 ; AVX512-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 64-byte Folded Reload
 ; AVX512-FCP-NEXT:    # zmm0 = zmm0[2,3,2,3],mem[2,3,2,3]
-; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Reload
-; AVX512-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9, %zmm9 # 64-byte Folded Reload
-; AVX512-FCP-NEXT:    # zmm9 = zmm9[2,3,2,3],mem[2,3,2,3]
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm18[0,1,0,1]
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} ymm7 = ymm19[0,1,0,1]
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm17[0,1,0,1]
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm9 ^ (mem & (zmm9 ^ zmm0))
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
+; AVX512-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm7, %zmm7 # 64-byte Folded Reload
+; AVX512-FCP-NEXT:    # zmm7 = zmm7[2,3,2,3],mem[2,3,2,3]
+; AVX512-FCP-NEXT:    vpermq {{.*#+}} ymm9 = ymm12[0,1,0,1]
+; AVX512-FCP-NEXT:    vpermq {{.*#+}} ymm12 = ymm15[0,1,0,1]
+; AVX512-FCP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,0,1]
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm7 = zmm7 ^ (mem & (zmm7 ^ zmm0))
 ; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
 ; AVX512-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 64-byte Folded Reload
 ; AVX512-FCP-NEXT:    # zmm0 = zmm0[2,3,2,3],mem[2,3,2,3]
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (zmm1 & (zmm0 ^ zmm9))
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm22 = zmm20 ^ (zmm1 & (zmm22 ^ zmm20))
-; AVX512-FCP-NEXT:    vmovdqa64 {{.*#+}} zmm8 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm14 = zmm10 ^ (zmm8 & (zmm14 ^ zmm10))
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} ymm9 = ymm30[2,3,2,3]
-; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm9, %zmm1, %zmm9
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} ymm10 = ymm27[2,3,2,3]
-; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm10, %zmm28, %zmm10
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm10 = zmm9 ^ (zmm8 & (zmm10 ^ zmm9))
-; AVX512-FCP-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3, %zmm3 # 32-byte Folded Reload
-; AVX512-FCP-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm7, %zmm7 # 32-byte Folded Reload
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm7 = zmm7 ^ (mem & (zmm7 ^ zmm3))
-; AVX512-FCP-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm2 # 32-byte Folded Reload
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 | (zmm2 & mem)
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 ^ (mem & (zmm16 ^ zmm7))
-; AVX512-FCP-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Folded Reload
-; AVX512-FCP-NEXT:    # zmm2 = mem[2,3,2,3,6,7,6,7]
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm2 = zmm2 ^ (mem & (zmm2 ^ zmm0))
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} zmm0 = zmm29[0,1,0,1,4,5,4,5]
-; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm1 ^ (mem & (zmm0 ^ zmm1))
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm22))
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm11 = zmm11 ^ (mem & (zmm11 ^ zmm21))
-; AVX512-FCP-NEXT:    vpermq {{.*#+}} zmm3 = zmm31[0,1,0,1,4,5,4,5]
-; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm1 ^ (mem & (zmm3 ^ zmm1))
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm3 ^ (mem & (zmm3 ^ zmm11))
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm12 = zmm25 ^ (mem & (zmm12 ^ zmm25))
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm12 = zmm12 ^ (mem & (zmm12 ^ zmm14))
-; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512-FCP-NEXT:    vshufi64x2 $84, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm5 # 64-byte Folded Reload
-; AVX512-FCP-NEXT:    # zmm5 = zmm1[0,1,2,3],mem[2,3,2,3]
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm4 = zmm4 | (zmm5 & mem)
-; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm4 = zmm4 ^ (mem & (zmm4 ^ zmm10))
-; AVX512-FCP-NEXT:    vmovdqa64 %zmm4, 320(%rax)
-; AVX512-FCP-NEXT:    vmovdqa64 %zmm12, 192(%rax)
-; AVX512-FCP-NEXT:    vmovdqa64 %zmm3, (%rax)
-; AVX512-FCP-NEXT:    vmovdqa64 %zmm0, 256(%rax)
-; AVX512-FCP-NEXT:    vmovdqa64 %zmm2, 384(%rax)
-; AVX512-FCP-NEXT:    vmovdqa64 %zmm16, 64(%rax)
-; AVX512-FCP-NEXT:    addq $1640, %rsp # imm = 0x668
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (zmm10 & (zmm0 ^ zmm7))
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm21 = zmm24 ^ (zmm10 & (zmm21 ^ zmm24))
+; AVX512-FCP-NEXT:    vmovdqa64 {{.*#+}} zmm7 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm23 = zmm30 ^ (zmm7 & (zmm23 ^ zmm30))
+; AVX512-FCP-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
+; AVX512-FCP-NEXT:    # ymm10 = mem[2,3,2,3]
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm13 # 64-byte Reload
+; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm10, %zmm13, %zmm10
+; AVX512-FCP-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
+; AVX512-FCP-NEXT:    # ymm13 = mem[2,3,2,3]
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
+; AVX512-FCP-NEXT:    vinserti64x4 $1, %ymm13, %zmm3, %zmm13
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm13 = zmm10 ^ (zmm7 & (zmm13 ^ zmm10))
+; AVX512-FCP-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9, %zmm7 # 32-byte Folded Reload
+; AVX512-FCP-NEXT:    vinserti64x4 $1, (%rsp), %zmm12, %zmm9 # 32-byte Folded Reload
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm9 ^ (mem & (zmm9 ^ zmm7))
+; AVX512-FCP-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm11, %zmm7 # 32-byte Folded Reload
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm14 = zmm14 | (zmm7 & mem)
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm14 = zmm14 ^ (mem & (zmm14 ^ zmm9))
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm3 ^ (mem & (zmm3 ^ zmm0))
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm4 = zmm29 ^ (mem & (zmm4 ^ zmm29))
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm4 = zmm4 ^ (mem & (zmm4 ^ zmm23))
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm5 = zmm25 ^ (mem & (zmm5 ^ zmm25))
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm5 = zmm5 ^ (mem & (zmm5 ^ zmm21))
+; AVX512-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
+; AVX512-FCP-NEXT:    vshufi64x2 $84, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 64-byte Folded Reload
+; AVX512-FCP-NEXT:    # zmm0 = zmm0[0,1,2,3],mem[2,3,2,3]
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm1 = zmm1 | (zmm0 & mem)
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm1 = zmm1 ^ (mem & (zmm1 ^ zmm13))
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm8 = zmm8 ^ (mem & (zmm8 ^ zmm2))
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm6 = zmm16 ^ (mem & (zmm6 ^ zmm16))
+; AVX512-FCP-NEXT:    vpternlogq {{.*#+}} zmm6 = zmm6 ^ (mem & (zmm6 ^ zmm8))
+; AVX512-FCP-NEXT:    vmovdqa64 %zmm6, (%rax)
+; AVX512-FCP-NEXT:    vmovdqa64 %zmm1, 320(%rax)
+; AVX512-FCP-NEXT:    vmovdqa64 %zmm5, 256(%rax)
+; AVX512-FCP-NEXT:    vmovdqa64 %zmm4, 192(%rax)
+; AVX512-FCP-NEXT:    vmovdqa64 %zmm3, 384(%rax)
+; AVX512-FCP-NEXT:    vmovdqa64 %zmm14, 64(%rax)
+; AVX512-FCP-NEXT:    addq $1416, %rsp # imm = 0x588
 ; AVX512-FCP-NEXT:    vzeroupper
 ; AVX512-FCP-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: store_i8_stride7_vf64:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    subq $1560, %rsp # imm = 0x618
-; AVX512DQ-NEXT:    vmovdqa (%rsi), %ymm6
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero,zero,zero,ymm6[18]
-; AVX512DQ-NEXT:    vmovdqa (%rdi), %ymm11
+; AVX512DQ-NEXT:    subq $1496, %rsp # imm = 0x5D8
+; AVX512DQ-NEXT:    vmovdqa (%rsi), %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm3, %ymm0
+; AVX512DQ-NEXT:    vmovdqa64 %ymm1, %ymm23
+; AVX512DQ-NEXT:    vmovdqa (%rdi), %ymm10
 ; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,14,128,12,13,0,1,14,15,128,3,12,13,2,3,16,128,30,31,28,29,16,17,128,31,18,19,28,29,18,128]
-; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm11, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm10, %ymm1
 ; AVX512DQ-NEXT:    vmovdqa64 %ymm2, %ymm16
 ; AVX512DQ-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512DQ-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-NEXT:    vmovdqa (%rcx), %ymm13
+; AVX512DQ-NEXT:    vmovdqa (%rcx), %ymm8
 ; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
-; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm13, %ymm0
-; AVX512DQ-NEXT:    vmovdqa64 %ymm1, %ymm24
-; AVX512DQ-NEXT:    vmovdqa (%rdx), %ymm2
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm2, %ymm1
-; AVX512DQ-NEXT:    vmovdqa64 %ymm3, %ymm26
-; AVX512DQ-NEXT:    vmovdqa %ymm2, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm8, %ymm0
+; AVX512DQ-NEXT:    vmovdqa64 %ymm1, %ymm18
+; AVX512DQ-NEXT:    vmovdqa (%rdx), %ymm7
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm11 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
+; AVX512DQ-NEXT:    vpshufb %ymm11, %ymm7, %ymm1
 ; AVX512DQ-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512DQ-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-NEXT:    vmovdqa (%r8), %ymm4
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
-; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm4, %ymm0
-; AVX512DQ-NEXT:    vmovdqa64 %ymm2, %ymm21
-; AVX512DQ-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-NEXT:    vmovdqa (%r9), %ymm8
-; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [13,0,0,0,128,16,128,14,0,0,0,128,17,128,15,0,13,0,0,0,128,16,128,14,0,0,0,128,17,128,15,0]
-; AVX512DQ-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm8, %ymm1
-; AVX512DQ-NEXT:    vmovdqa64 %ymm5, %ymm19
+; AVX512DQ-NEXT:    vmovdqa (%r8), %ymm6
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm6, %ymm0
+; AVX512DQ-NEXT:    vmovdqa64 %ymm1, %ymm24
+; AVX512DQ-NEXT:    vmovdqa (%r9), %ymm5
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [13,0,0,0,128,16,128,14,0,0,0,128,17,128,15,0,13,0,0,0,128,16,128,14,0,0,0,128,17,128,15,0]
+; AVX512DQ-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm5, %ymm1
+; AVX512DQ-NEXT:    vmovdqa64 %ymm2, %ymm25
 ; AVX512DQ-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512DQ-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128]
-; AVX512DQ-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm6, %ymm0
-; AVX512DQ-NEXT:    vmovdqa64 %ymm6, %ymm20
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm5 = [12,13,14,128,12,128,14,15,14,15,128,13,128,15,12,13,28,29,30,128,28,128,30,31,30,31,128,29,128,31,28,29]
-; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm11, %ymm2
-; AVX512DQ-NEXT:    vpor %ymm0, %ymm2, %ymm0
-; AVX512DQ-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128]
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm14 = [128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128]
+; AVX512DQ-NEXT:    # ymm14 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb %ymm14, %ymm3, %ymm1
+; AVX512DQ-NEXT:    vmovdqa64 %ymm3, %ymm17
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm0 = [12,13,14,128,12,128,14,15,14,15,128,13,128,15,12,13,28,29,30,128,28,128,30,31,30,31,128,29,128,31,28,29]
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm10, %ymm2
+; AVX512DQ-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm15 = [128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128]
+; AVX512DQ-NEXT:    # ymm15 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb %ymm15, %ymm7, %ymm1
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0]
 ; AVX512DQ-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm0
-; AVX512DQ-NEXT:    vmovdqa64 %ymm3, %ymm22
-; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0]
-; AVX512DQ-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm7, %ymm13, %ymm3
-; AVX512DQ-NEXT:    vpor %ymm0, %ymm3, %ymm0
-; AVX512DQ-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm8, %ymm3
+; AVX512DQ-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [128,27,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,27,128,128,128,128,30,128,28,128,128,128,128,31,128,29]
 ; AVX512DQ-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm0
-; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128]
-; AVX512DQ-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm8, %ymm6
-; AVX512DQ-NEXT:    vmovdqa64 %ymm8, %ymm17
-; AVX512DQ-NEXT:    vpor %ymm0, %ymm6, %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm6, %ymm1
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128]
+; AVX512DQ-NEXT:    # ymm9 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb %ymm9, %ymm5, %ymm13
+; AVX512DQ-NEXT:    vpor %ymm1, %ymm13, %ymm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vmovdqa 32(%rsi), %ymm13
+; AVX512DQ-NEXT:    vpshufb %ymm14, %ymm13, %ymm1
+; AVX512DQ-NEXT:    vmovdqa 32(%rdi), %ymm14
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm14, %ymm0
+; AVX512DQ-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa 32(%rsi), %ymm10
-; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm10, %ymm6
-; AVX512DQ-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm1, %ymm5
-; AVX512DQ-NEXT:    vpor %ymm6, %ymm5, %ymm5
-; AVX512DQ-NEXT:    vmovdqu64 %zmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25]
-; AVX512DQ-NEXT:    # ymm6 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm6, %ymm1, %ymm5
-; AVX512DQ-NEXT:    vmovdqa64 %ymm6, %ymm29
-; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128,25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128]
-; AVX512DQ-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm8, %ymm10, %ymm6
-; AVX512DQ-NEXT:    vmovdqa64 %ymm8, %ymm25
-; AVX512DQ-NEXT:    vpor %ymm5, %ymm6, %ymm5
-; AVX512DQ-NEXT:    vmovdqu64 %zmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa 32(%rdx), %ymm5
-; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm5, %ymm2
-; AVX512DQ-NEXT:    vmovdqa 32(%rcx), %ymm6
-; AVX512DQ-NEXT:    vpshufb %ymm7, %ymm6, %ymm7
-; AVX512DQ-NEXT:    vpor %ymm2, %ymm7, %ymm2
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm0 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm14[23],zero,zero,zero,zero,ymm14[26],zero,ymm14[24],zero,zero,zero,zero,ymm14[27],zero,ymm14[25]
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm1 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm13[23,u,u,u],zero,ymm13[26],zero,ymm13[24,u,u,u],zero,ymm13[27],zero
+; AVX512DQ-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vmovdqa 32(%rdx), %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm15, %ymm1, %ymm15
+; AVX512DQ-NEXT:    vmovdqa 32(%rcx), %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm0, %ymm2
+; AVX512DQ-NEXT:    vpor %ymm2, %ymm15, %ymm2
 ; AVX512DQ-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm2 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero,zero,zero
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm7 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm5[23],zero,ymm5[21,22,23,26],zero,ymm5[24],zero,ymm5[28,29,26,27]
-; AVX512DQ-NEXT:    vpor %ymm2, %ymm7, %ymm2
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm2 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm0[25],zero,ymm0[23],zero,zero,zero,zero,ymm0[26],zero,ymm0[24],zero,zero,zero,zero
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm12 = [24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27,24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27]
+; AVX512DQ-NEXT:    # ymm12 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb %ymm12, %ymm1, %ymm15
+; AVX512DQ-NEXT:    vmovdqa64 %ymm12, %ymm22
+; AVX512DQ-NEXT:    vpor %ymm2, %ymm15, %ymm2
 ; AVX512DQ-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa 32(%r8), %ymm2
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm2, %ymm7
-; AVX512DQ-NEXT:    vmovdqa 32(%r9), %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512DQ-NEXT:    vpor %ymm7, %ymm4, %ymm4
-; AVX512DQ-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm4 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm7 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm2[23],zero,ymm2[23,24,25,26],zero,ymm2[24],zero,ymm2[30,31]
-; AVX512DQ-NEXT:    vpor %ymm4, %ymm7, %ymm4
-; AVX512DQ-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vmovdqa 32(%r8), %ymm15
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm15, %ymm3
+; AVX512DQ-NEXT:    vmovdqa 32(%r9), %ymm2
+; AVX512DQ-NEXT:    vpshufb %ymm9, %ymm2, %ymm9
+; AVX512DQ-NEXT:    vpor %ymm3, %ymm9, %ymm3
+; AVX512DQ-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128]
+; AVX512DQ-NEXT:    # ymm9 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb %ymm9, %ymm2, %ymm3
+; AVX512DQ-NEXT:    vmovdqa64 %ymm9, %ymm21
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm9 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm15[23],zero,ymm15[23,24,25,26],zero,ymm15[24],zero,ymm15[30,31]
+; AVX512DQ-NEXT:    vpor %ymm3, %ymm9, %ymm3
+; AVX512DQ-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512DQ-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512DQ-NEXT:    vmovdqa 32(%rax), %ymm7
-; AVX512DQ-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm4 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm7 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm4, %zmm7, %zmm4
-; AVX512DQ-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa 32(%rdx), %xmm7
-; AVX512DQ-NEXT:    vmovdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-NEXT:    vmovdqa 32(%rcx), %xmm4
-; AVX512DQ-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm15 = [u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u]
-; AVX512DQ-NEXT:    vpshufb %xmm15, %xmm4, %xmm4
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm14 = [u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u]
-; AVX512DQ-NEXT:    vpshufb %xmm14, %xmm7, %xmm7
-; AVX512DQ-NEXT:    vmovdqa64 %xmm14, %xmm23
-; AVX512DQ-NEXT:    vpor %xmm4, %xmm7, %xmm4
-; AVX512DQ-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa 32(%rdi), %xmm9
-; AVX512DQ-NEXT:    vmovdqa 32(%rsi), %xmm0
-; AVX512DQ-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm7 = [u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128]
-; AVX512DQ-NEXT:    vpshufb %xmm7, %xmm0, %xmm4
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm14 = [u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9]
-; AVX512DQ-NEXT:    vpshufb %xmm14, %xmm9, %xmm8
-; AVX512DQ-NEXT:    vmovdqa64 %xmm9, %xmm31
-; AVX512DQ-NEXT:    vpor %xmm4, %xmm8, %xmm4
-; AVX512DQ-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm28 = [0,0,0,0,2,3,0,1,0,18,0,19,18,0,19,0]
-; AVX512DQ-NEXT:    vmovdqa 32(%rax), %xmm8
-; AVX512DQ-NEXT:    vmovdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-NEXT:    vpshufhw {{.*#+}} xmm4 = xmm8[0,1,2,3,4,5,5,6]
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
-; AVX512DQ-NEXT:    vpermi2d %zmm4, %zmm8, %zmm28
-; AVX512DQ-NEXT:    vmovdqa 32(%r9), %xmm4
-; AVX512DQ-NEXT:    vmovdqa 32(%r8), %xmm0
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm12 = [128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6]
-; AVX512DQ-NEXT:    vpshufb %xmm12, %xmm4, %xmm8
-; AVX512DQ-NEXT:    vmovdqa64 %xmm4, %xmm30
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128]
-; AVX512DQ-NEXT:    vpshufb %xmm4, %xmm0, %xmm9
-; AVX512DQ-NEXT:    vmovdqa64 %xmm4, %xmm18
-; AVX512DQ-NEXT:    vmovdqa64 %xmm0, %xmm27
-; AVX512DQ-NEXT:    vpor %xmm8, %xmm9, %xmm0
-; AVX512DQ-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa64 %ymm24, %ymm4
-; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm6, %ymm8
-; AVX512DQ-NEXT:    vmovdqa64 %ymm26, %ymm4
-; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm5, %ymm9
-; AVX512DQ-NEXT:    vpor %ymm8, %ymm9, %ymm0
-; AVX512DQ-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20]
-; AVX512DQ-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm8, %ymm6, %ymm6
+; AVX512DQ-NEXT:    vmovdqa 32(%rax), %ymm9
+; AVX512DQ-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm3 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512DQ-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm9, %ymm12
+; AVX512DQ-NEXT:    vmovdqa64 %ymm4, %ymm20
+; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm29 = zmm12[2,3,2,3],zmm3[2,3,2,3]
+; AVX512DQ-NEXT:    vmovdqa64 %ymm18, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm0, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm11, %ymm1, %ymm11
+; AVX512DQ-NEXT:    vpor %ymm3, %ymm11, %ymm3
+; AVX512DQ-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20]
+; AVX512DQ-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128,18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128]
 ; AVX512DQ-NEXT:    # ymm9 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm9, %ymm5, %ymm5
-; AVX512DQ-NEXT:    vpor %ymm6, %ymm5, %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm9, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512DQ-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm5 = zero,zero,zero,ymm10[14],zero,zero,zero,zero,zero,zero,ymm10[15],zero,zero,zero,zero,zero,zero,ymm10[16],zero,zero,zero,zero,zero,zero,ymm10[17],zero,zero,zero,zero,zero,zero,ymm10[18]
-; AVX512DQ-NEXT:    vmovdqa64 %ymm16, %ymm0
-; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm6
-; AVX512DQ-NEXT:    vpor %ymm5, %ymm6, %ymm0
+; AVX512DQ-NEXT:    vmovdqa64 %ymm23, %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm13, %ymm0
+; AVX512DQ-NEXT:    vmovdqa64 %ymm16, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm14, %ymm1
+; AVX512DQ-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512DQ-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128]
-; AVX512DQ-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm10, %ymm5
-; AVX512DQ-NEXT:    vmovdqa64 %ymm0, %ymm16
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm0 = [2,3,4,5,128,3,128,5,4,5,6,128,4,128,6,7,18,19,20,21,128,19,128,21,20,21,22,128,20,128,22,23]
-; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpor %ymm5, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128]
+; AVX512DQ-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm13, %ymm0
+; AVX512DQ-NEXT:    vmovdqa64 %ymm1, %ymm26
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm13 = [2,3,4,5,128,3,128,5,4,5,6,128,4,128,6,7,18,19,20,21,128,19,128,21,20,21,22,128,20,128,22,23]
+; AVX512DQ-NEXT:    vpshufb %ymm13, %ymm14, %ymm1
+; AVX512DQ-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22]
 ; AVX512DQ-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm3, %ymm5
-; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128,20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128]
-; AVX512DQ-NEXT:    # ymm10 = mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb %ymm10, %ymm2, %ymm6
-; AVX512DQ-NEXT:    vpor %ymm5, %ymm6, %ymm4
-; AVX512DQ-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa64 %ymm21, %ymm4
-; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vmovdqa64 %ymm19, %ymm4
-; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512DQ-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa (%rsi), %xmm3
-; AVX512DQ-NEXT:    vpshufb %xmm7, %xmm3, %xmm2
-; AVX512DQ-NEXT:    vmovdqa64 %xmm3, %xmm24
-; AVX512DQ-NEXT:    vmovdqa (%rdi), %xmm4
-; AVX512DQ-NEXT:    vpshufb %xmm14, %xmm4, %xmm3
-; AVX512DQ-NEXT:    vmovdqa64 %xmm4, %xmm21
-; AVX512DQ-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX512DQ-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa (%rcx), %xmm3
-; AVX512DQ-NEXT:    vpshufb %xmm15, %xmm3, %xmm2
-; AVX512DQ-NEXT:    vmovdqa64 %xmm3, %xmm19
-; AVX512DQ-NEXT:    vmovdqa (%rdx), %xmm15
-; AVX512DQ-NEXT:    vmovdqa64 %xmm23, %xmm3
-; AVX512DQ-NEXT:    vpshufb %xmm3, %xmm15, %xmm3
-; AVX512DQ-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX512DQ-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa (%r9), %xmm14
-; AVX512DQ-NEXT:    vpshufb %xmm12, %xmm14, %xmm2
-; AVX512DQ-NEXT:    vmovdqa (%r8), %xmm12
-; AVX512DQ-NEXT:    vmovdqa64 %xmm18, %xmm3
-; AVX512DQ-NEXT:    vpshufb %xmm3, %xmm12, %xmm3
-; AVX512DQ-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX512DQ-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-NEXT:    vmovdqa64 %ymm29, %ymm2
-; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm11, %ymm2
-; AVX512DQ-NEXT:    vmovdqa64 %ymm25, %ymm3
-; AVX512DQ-NEXT:    vmovdqa64 %ymm20, %ymm4
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
-; AVX512DQ-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512DQ-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa64 %ymm16, %ymm2
-; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm4, %ymm2
-; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm11, %ymm0
-; AVX512DQ-NEXT:    vporq %ymm2, %ymm0, %ymm29
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm0 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm13[25],zero,ymm13[23],zero,zero,zero,zero,ymm13[26],zero,ymm13[24],zero,zero,zero,zero
-; AVX512DQ-NEXT:    vmovdqa64 %ymm22, %ymm3
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm2 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm3[23],zero,ymm3[21,22,23,26],zero,ymm3[24],zero,ymm3[28,29,26,27]
-; AVX512DQ-NEXT:    vporq %ymm0, %ymm2, %ymm25
-; AVX512DQ-NEXT:    vpshufb %ymm8, %ymm13, %ymm0
-; AVX512DQ-NEXT:    vpshufb %ymm9, %ymm3, %ymm2
-; AVX512DQ-NEXT:    vporq %ymm0, %ymm2, %ymm22
-; AVX512DQ-NEXT:    vmovdqa64 %ymm17, %ymm2
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm2[25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm2, %ymm0
+; AVX512DQ-NEXT:    vmovdqa64 %ymm1, %ymm18
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128,20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128]
+; AVX512DQ-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm15, %ymm1
+; AVX512DQ-NEXT:    vmovdqa64 %ymm3, %ymm19
+; AVX512DQ-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vmovdqa64 %ymm24, %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm15, %ymm0
+; AVX512DQ-NEXT:    vmovdqa64 %ymm25, %ymm1
 ; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm2, %ymm1
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512DQ-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512DQ-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm1[23],zero,ymm1[23,24,25,26],zero,ymm1[24],zero,ymm1[30,31]
-; AVX512DQ-NEXT:    vpshufb %ymm10, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512DQ-NEXT:    vmovdqa (%rdi), %xmm2
+; AVX512DQ-NEXT:    vmovdqa (%rsi), %xmm1
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm12 = [u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128]
+; AVX512DQ-NEXT:    vpshufb %xmm12, %xmm1, %xmm0
+; AVX512DQ-NEXT:    vmovdqa64 %xmm1, %xmm16
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm14 = [u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9]
+; AVX512DQ-NEXT:    vpshufb %xmm14, %xmm2, %xmm1
+; AVX512DQ-NEXT:    vmovdqa64 %xmm2, %xmm28
+; AVX512DQ-NEXT:    vpor %xmm0, %xmm1, %xmm0
 ; AVX512DQ-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-NEXT:    vmovdqa (%rax), %ymm5
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm1 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm13 = [0,5,4,0,5,0,4,0,20,21,0,23,0,21,0,23]
-; AVX512DQ-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm5[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512DQ-NEXT:    vpermi2d %zmm1, %zmm2, %zmm13
-; AVX512DQ-NEXT:    vmovdqa (%rax), %xmm11
-; AVX512DQ-NEXT:    vpshufhw {{.*#+}} xmm2 = xmm11[0,1,2,3,4,5,5,6]
-; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
-; AVX512DQ-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm10 = [128,13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128]
-; AVX512DQ-NEXT:    vpshufb %ymm10, %ymm5, %ymm3
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm16
-; AVX512DQ-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; AVX512DQ-NEXT:    vmovdqa64 %xmm31, %xmm7
-; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm7[0],xmm8[0],xmm7[1],xmm8[1],xmm7[2],xmm8[2],xmm7[3],xmm8[3],xmm7[4],xmm8[4],xmm7[5],xmm8[5],xmm7[6],xmm8[6],xmm7[7],xmm8[7]
+; AVX512DQ-NEXT:    vmovdqa (%rdx), %xmm3
+; AVX512DQ-NEXT:    vmovdqa (%rcx), %xmm0
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm11 = [u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u]
+; AVX512DQ-NEXT:    vpshufb %xmm11, %xmm0, %xmm1
+; AVX512DQ-NEXT:    vmovdqa64 %xmm0, %xmm31
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm0 = [u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u]
+; AVX512DQ-NEXT:    vpshufb %xmm0, %xmm3, %xmm2
+; AVX512DQ-NEXT:    vmovdqa64 %xmm3, %xmm30
+; AVX512DQ-NEXT:    vpor %xmm1, %xmm2, %xmm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vmovdqa (%r9), %xmm1
+; AVX512DQ-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512DQ-NEXT:    vmovdqa (%r8), %xmm3
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm1 = zero,xmm1[4,u,u,u],zero,xmm1[7],zero,xmm1[5,u,u,u],zero,xmm1[8],zero,xmm1[6]
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm2 = xmm3[4],zero,xmm3[u,u,u,7],zero,xmm3[5],zero,xmm3[u,u,u,8],zero,xmm3[6],zero
+; AVX512DQ-NEXT:    vmovdqa64 %xmm3, %xmm23
+; AVX512DQ-NEXT:    vpor %xmm1, %xmm2, %xmm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm1 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm10[23],zero,zero,zero,zero,ymm10[26],zero,ymm10[24],zero,zero,zero,zero,ymm10[27],zero,ymm10[25]
+; AVX512DQ-NEXT:    vmovdqa64 %ymm17, %ymm3
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm2 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm3[23,u,u,u],zero,ymm3[26],zero,ymm3[24,u,u,u],zero,ymm3[27],zero
+; AVX512DQ-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vmovdqa64 %ymm26, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm13, %ymm10, %ymm2
+; AVX512DQ-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm1 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm8[25],zero,ymm8[23],zero,zero,zero,zero,ymm8[26],zero,ymm8[24],zero,zero,zero,zero
+; AVX512DQ-NEXT:    vmovdqa64 %ymm22, %ymm2
+; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm7, %ymm2
+; AVX512DQ-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm8, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm9, %ymm7, %ymm2
+; AVX512DQ-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vmovdqa64 %ymm21, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm5, %ymm1
+; AVX512DQ-NEXT:    vmovdqa64 %ymm18, %ymm2
+; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm5, %ymm2
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm1, %zmm2, %zmm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm1 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm6[23],zero,ymm6[23,24,25,26],zero,ymm6[24],zero,ymm6[30,31]
+; AVX512DQ-NEXT:    vmovdqa64 %ymm19, %ymm2
+; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm6, %ymm2
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm1, %zmm2, %zmm1
+; AVX512DQ-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-NEXT:    vmovdqa (%rax), %ymm4
+; AVX512DQ-NEXT:    vmovdqa64 %ymm20, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm4, %ymm2
+; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm18 = [0,5,4,0,5,0,4,0,20,21,0,23,0,21,0,23]
+; AVX512DQ-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm4[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512DQ-NEXT:    vpermi2d %zmm2, %zmm3, %zmm18
+; AVX512DQ-NEXT:    vmovdqa 32(%rcx), %xmm13
+; AVX512DQ-NEXT:    vpshufb %xmm11, %xmm13, %xmm2
+; AVX512DQ-NEXT:    vmovdqa 32(%rdx), %xmm10
+; AVX512DQ-NEXT:    vpshufb %xmm0, %xmm10, %xmm0
+; AVX512DQ-NEXT:    vporq %xmm2, %xmm0, %xmm26
+; AVX512DQ-NEXT:    vmovdqa 32(%rsi), %xmm11
+; AVX512DQ-NEXT:    vpshufb %xmm12, %xmm11, %xmm0
+; AVX512DQ-NEXT:    vmovdqa 32(%rdi), %xmm8
+; AVX512DQ-NEXT:    vpshufb %xmm14, %xmm8, %xmm2
+; AVX512DQ-NEXT:    vporq %xmm0, %xmm2, %xmm24
+; AVX512DQ-NEXT:    vmovdqa 32(%rax), %xmm1
+; AVX512DQ-NEXT:    vpmovsxbd {{.*#+}} zmm20 = [0,0,0,0,2,3,0,1,0,18,0,19,18,0,19,0]
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
+; AVX512DQ-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,5,5,6]
+; AVX512DQ-NEXT:    vpshufb %xmm3, %xmm1, %xmm2
+; AVX512DQ-NEXT:    vmovdqa64 %xmm3, %xmm27
+; AVX512DQ-NEXT:    vmovdqa64 %xmm1, %xmm25
+; AVX512DQ-NEXT:    vpermi2d %zmm0, %zmm2, %zmm20
+; AVX512DQ-NEXT:    vmovdqa 32(%r9), %xmm15
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm0 = zero,xmm15[4,u,u,u],zero,xmm15[7],zero,xmm15[5,u,u,u],zero,xmm15[8],zero,xmm15[6]
+; AVX512DQ-NEXT:    vmovdqa 32(%r8), %xmm5
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm3 = xmm5[4],zero,xmm5[u,u,u,7],zero,xmm5[5],zero,xmm5[u,u,u,8],zero,xmm5[6],zero
+; AVX512DQ-NEXT:    vporq %xmm0, %xmm3, %xmm22
+; AVX512DQ-NEXT:    vmovdqa (%rax), %xmm2
+; AVX512DQ-NEXT:    vpshufhw {{.*#+}} xmm3 = xmm2[0,1,2,3,4,5,5,6]
+; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,0,1]
+; AVX512DQ-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm9 = [128,13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128]
+; AVX512DQ-NEXT:    vpshufb %ymm9, %ymm4, %ymm6
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm6, %zmm3, %zmm3
+; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm6 = xmm8[0],xmm11[0],xmm8[1],xmm11[1],xmm8[2],xmm11[2],xmm8[3],xmm11[3],xmm8[4],xmm11[4],xmm8[5],xmm11[5],xmm8[6],xmm11[6],xmm8[7],xmm11[7]
 ; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512DQ-NEXT:    vpshufb %xmm0, %xmm3, %xmm3
-; AVX512DQ-NEXT:    vmovdqa64 %xmm0, %xmm26
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm9 = zmm0[2,3,2,3],zmm3[0,1,0,1]
-; AVX512DQ-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; AVX512DQ-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm0 = [4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512DQ-NEXT:    vpshufb %xmm0, %xmm3, %xmm3
-; AVX512DQ-NEXT:    vmovdqa64 %xmm0, %xmm23
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm18 = zmm0[2,3,2,3],zmm3[0,1,0,1]
-; AVX512DQ-NEXT:    vmovdqa64 %xmm30, %xmm4
-; AVX512DQ-NEXT:    vmovdqa64 %xmm27, %xmm3
-; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm6 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm0 = [u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
 ; AVX512DQ-NEXT:    vpshufb %xmm0, %xmm6, %xmm6
-; AVX512DQ-NEXT:    vmovdqa64 %xmm0, %xmm20
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm30 = zmm0[2,3,2,3],zmm6[0,1,0,1]
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm0 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512DQ-NEXT:    vmovdqa64 %ymm0, %ymm17
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm1[8],xmm2[8],xmm1[9],xmm2[9],xmm1[10],xmm2[10],xmm1[11],xmm2[11],xmm1[12],xmm2[12],xmm1[13],xmm2[13],xmm1[14],xmm2[14],xmm1[15],xmm2[15]
-; AVX512DQ-NEXT:    vmovdqa64 %xmm19, %xmm2
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm2[8],xmm15[8],xmm2[9],xmm15[9],xmm2[10],xmm15[10],xmm2[11],xmm15[11],xmm2[12],xmm15[12],xmm2[13],xmm15[13],xmm2[14],xmm15[14],xmm2[15],xmm15[15]
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512DQ-NEXT:    vpshufb %xmm1, %xmm6, %xmm0
-; AVX512DQ-NEXT:    vmovdqa64 %ymm0, %ymm19
-; AVX512DQ-NEXT:    vpshufb %xmm1, %xmm5, %xmm1
+; AVX512DQ-NEXT:    vmovdqa64 %xmm0, %xmm21
 ; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm31 = zmm0[0,1,0,1],zmm1[0,1,0,1]
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm8[8],xmm7[8],xmm8[9],xmm7[9],xmm8[10],xmm7[10],xmm8[11],xmm7[11],xmm8[12],xmm7[12],xmm8[13],xmm7[13],xmm8[14],xmm7[14],xmm8[15],xmm7[15]
-; AVX512DQ-NEXT:    vmovdqa64 %xmm24, %xmm7
-; AVX512DQ-NEXT:    vmovdqa64 %xmm21, %xmm8
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm7[8],xmm8[8],xmm7[9],xmm8[9],xmm7[10],xmm8[10],xmm7[11],xmm8[11],xmm7[12],xmm8[12],xmm7[13],xmm8[13],xmm7[14],xmm8[14],xmm7[15],xmm8[15]
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512DQ-NEXT:    vpshufb %xmm1, %xmm6, %xmm6
-; AVX512DQ-NEXT:    vpshufb %xmm1, %xmm5, %xmm1
+; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm6 = zmm0[2,3,2,3],zmm6[0,1,0,1]
+; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm7 = xmm10[0],xmm13[0],xmm10[1],xmm13[1],xmm10[2],xmm13[2],xmm10[3],xmm13[3],xmm10[4],xmm13[4],xmm10[5],xmm13[5],xmm10[6],xmm13[6],xmm10[7],xmm13[7]
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm14 = [4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX512DQ-NEXT:    vpshufb %xmm14, %xmm7, %xmm7
 ; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm21 = zmm0[0,1,0,1],zmm1[0,1,0,1]
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm3[8],xmm4[8],xmm3[9],xmm4[9],xmm3[10],xmm4[10],xmm3[11],xmm4[11],xmm3[12],xmm4[12],xmm3[13],xmm4[13],xmm3[14],xmm4[14],xmm3[15],xmm4[15]
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm12[8],xmm14[8],xmm12[9],xmm14[9],xmm12[10],xmm14[10],xmm12[11],xmm14[11],xmm12[12],xmm14[12],xmm12[13],xmm14[13],xmm12[14],xmm14[14],xmm12[15],xmm14[15]
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX512DQ-NEXT:    vpshufb %xmm1, %xmm5, %xmm5
+; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm7 = zmm0[2,3,2,3],zmm7[0,1,0,1]
+; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm5[0],xmm15[0],xmm5[1],xmm15[1],xmm5[2],xmm15[2],xmm5[3],xmm15[3],xmm5[4],xmm15[4],xmm5[5],xmm15[5],xmm5[6],xmm15[6],xmm5[7],xmm15[7]
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
 ; AVX512DQ-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512DQ-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm4
+; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm19 # 64-byte Reload
+; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm19 = zmm19[2,3,2,3],zmm0[0,1,0,1]
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512DQ-NEXT:    vmovdqa64 %ymm0, %ymm17
 ; AVX512DQ-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512DQ-NEXT:    vpshufb %ymm10, %ymm0, %ymm1
-; AVX512DQ-NEXT:    vpshuflw {{.*#+}} ymm10 = ymm0[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512DQ-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[0,1,1,3,4,5,5,7]
-; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,2,3,2]
-; AVX512DQ-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm10, %ymm10
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm10, %zmm1, %zmm10
-; AVX512DQ-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm18 = zmm9 ^ (zmm1 & (zmm18 ^ zmm9))
-; AVX512DQ-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX512DQ-NEXT:    # ymm9 = mem[2,3,2,3]
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm9, %zmm0, %zmm9
-; AVX512DQ-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm27 # 32-byte Folded Reload
-; AVX512DQ-NEXT:    # ymm27 = mem[2,3,2,3]
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm27, %zmm0, %zmm27
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm27 = zmm9 ^ (zmm1 & (zmm27 ^ zmm9))
-; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm8[0],xmm7[0],xmm8[1],xmm7[1],xmm8[2],xmm7[2],xmm8[3],xmm7[3],xmm8[4],xmm7[4],xmm8[5],xmm7[5],xmm8[6],xmm7[6],xmm8[7],xmm7[7]
-; AVX512DQ-NEXT:    vmovdqa64 %xmm26, %xmm0
-; AVX512DQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm1
-; AVX512DQ-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm1 # 64-byte Folded Reload
-; AVX512DQ-NEXT:    # zmm1 = zmm1[0,1,0,1],mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm15[0],xmm2[0],xmm15[1],xmm2[1],xmm15[2],xmm2[2],xmm15[3],xmm2[3],xmm15[4],xmm2[4],xmm15[5],xmm2[5],xmm15[6],xmm2[6],xmm15[7],xmm2[7]
-; AVX512DQ-NEXT:    vmovdqa64 %xmm23, %xmm0
-; AVX512DQ-NEXT:    vpshufb %xmm0, %xmm9, %xmm9
-; AVX512DQ-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9, %zmm9 # 64-byte Folded Reload
-; AVX512DQ-NEXT:    # zmm9 = zmm9[0,1,0,1],mem[0,1,0,1]
-; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm12[0],xmm14[0],xmm12[1],xmm14[1],xmm12[2],xmm14[2],xmm12[3],xmm14[3],xmm12[4],xmm14[4],xmm12[5],xmm14[5],xmm12[6],xmm14[6],xmm12[7],xmm14[7]
-; AVX512DQ-NEXT:    vmovdqa64 %xmm20, %xmm0
-; AVX512DQ-NEXT:    vpshufb %xmm0, %xmm12, %xmm12
-; AVX512DQ-NEXT:    vinserti32x4 $2, {{[-0-9]+}}(%r{{[sb]}}p), %zmm12, %zmm12 # 16-byte Folded Reload
-; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm14 = xmm11[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
-; AVX512DQ-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm11[1,1,0,0,4,5,6,7]
-; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm7 = xmm7[0,1,2,0]
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm14, %zmm7, %zmm7
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm14 # 64-byte Reload
-; AVX512DQ-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm14, %zmm14 # 64-byte Folded Reload
-; AVX512DQ-NEXT:    # zmm14 = zmm14[2,3,2,3],mem[2,3,2,3]
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm15 # 64-byte Reload
-; AVX512DQ-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm15, %zmm15 # 64-byte Folded Reload
-; AVX512DQ-NEXT:    # zmm15 = zmm15[2,3,2,3],mem[2,3,2,3]
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm15 = zmm15 ^ (mem & (zmm15 ^ zmm14))
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm14 # 64-byte Reload
-; AVX512DQ-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm14, %zmm14 # 64-byte Folded Reload
-; AVX512DQ-NEXT:    # zmm14 = zmm14[2,3,2,3],mem[2,3,2,3]
-; AVX512DQ-NEXT:    vmovdqa64 {{.*#+}} zmm23 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm14 = zmm14 ^ (zmm23 & (zmm14 ^ zmm15))
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm21 = zmm31 ^ (zmm23 & (zmm21 ^ zmm31))
-; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
-; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm11 = ymm19[0,1,0,1]
-; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
-; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm0 = ymm17[2,3,2,3]
-; AVX512DQ-NEXT:    vpshuflw $5, {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
-; AVX512DQ-NEXT:    # xmm15 = mem[1,1,0,0,4,5,6,7]
+; AVX512DQ-NEXT:    vpshufb %ymm9, %ymm0, %ymm9
+; AVX512DQ-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm0[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512DQ-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,1,1,3,4,5,5,7]
+; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,3,2]
+; AVX512DQ-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm9, %zmm9
+; AVX512DQ-NEXT:    vmovdqa64 %xmm28, %xmm4
+; AVX512DQ-NEXT:    vmovdqa64 %xmm16, %xmm0
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm12 = xmm0[8],xmm4[8],xmm0[9],xmm4[9],xmm0[10],xmm4[10],xmm0[11],xmm4[11],xmm0[12],xmm4[12],xmm0[13],xmm4[13],xmm0[14],xmm4[14],xmm0[15],xmm4[15]
+; AVX512DQ-NEXT:    vmovdqa64 %xmm12, %xmm16
+; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3],xmm4[4],xmm0[4],xmm4[5],xmm0[5],xmm4[6],xmm0[6],xmm4[7],xmm0[7]
+; AVX512DQ-NEXT:    vmovdqa64 %xmm21, %xmm0
+; AVX512DQ-NEXT:    vpshufb %xmm0, %xmm4, %xmm4
+; AVX512DQ-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm28 # 64-byte Folded Reload
+; AVX512DQ-NEXT:    # zmm28 = zmm4[0,1,0,1],mem[0,1,0,1]
+; AVX512DQ-NEXT:    vmovdqa64 %xmm30, %xmm0
+; AVX512DQ-NEXT:    vmovdqa64 %xmm31, %xmm12
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm12[8],xmm0[8],xmm12[9],xmm0[9],xmm12[10],xmm0[10],xmm12[11],xmm0[11],xmm12[12],xmm0[12],xmm12[13],xmm0[13],xmm12[14],xmm0[14],xmm12[15],xmm0[15]
+; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm0[0],xmm12[0],xmm0[1],xmm12[1],xmm0[2],xmm12[2],xmm0[3],xmm12[3],xmm0[4],xmm12[4],xmm0[5],xmm12[5],xmm0[6],xmm12[6],xmm0[7],xmm12[7]
+; AVX512DQ-NEXT:    vpshufb %xmm14, %xmm12, %xmm12
+; AVX512DQ-NEXT:    vmovdqa64 {{.*#+}} zmm14 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm7 = zmm6 ^ (zmm14 & (zmm7 ^ zmm6))
+; AVX512DQ-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
+; AVX512DQ-NEXT:    # ymm6 = mem[2,3,2,3]
+; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm31 # 64-byte Reload
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm6, %zmm31, %zmm6
+; AVX512DQ-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm31 # 32-byte Folded Reload
+; AVX512DQ-NEXT:    # ymm31 = mem[2,3,2,3]
+; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm30 # 64-byte Reload
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm31, %zmm30, %zmm31
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm31 = zmm6 ^ (zmm14 & (zmm31 ^ zmm6))
+; AVX512DQ-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm12, %zmm6 # 64-byte Folded Reload
+; AVX512DQ-NEXT:    # zmm6 = zmm12[0,1,0,1],mem[0,1,0,1]
+; AVX512DQ-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX512DQ-NEXT:    vmovdqa64 %xmm23, %xmm14
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm12 = xmm14[8],xmm0[8],xmm14[9],xmm0[9],xmm14[10],xmm0[10],xmm14[11],xmm0[11],xmm14[12],xmm0[12],xmm14[13],xmm0[13],xmm14[14],xmm0[14],xmm14[15],xmm0[15]
+; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm14 = xmm14[0],xmm0[0],xmm14[1],xmm0[1],xmm14[2],xmm0[2],xmm14[3],xmm0[3],xmm14[4],xmm0[4],xmm14[5],xmm0[5],xmm14[6],xmm0[6],xmm14[7],xmm0[7]
+; AVX512DQ-NEXT:    vpshufb %xmm1, %xmm14, %xmm14
+; AVX512DQ-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm14, %zmm23 # 64-byte Folded Reload
+; AVX512DQ-NEXT:    # zmm23 = zmm14[0,1,0,1],mem[0,1,0,1]
+; AVX512DQ-NEXT:    vmovdqa64 %xmm27, %xmm0
+; AVX512DQ-NEXT:    vpshufb %xmm0, %xmm2, %xmm14
+; AVX512DQ-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm2[1,1,0,0,4,5,6,7]
+; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,2,0]
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm14, %zmm1, %zmm1
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm10 = xmm13[8],xmm10[8],xmm13[9],xmm10[9],xmm13[10],xmm10[10],xmm13[11],xmm10[11],xmm13[12],xmm10[12],xmm13[13],xmm10[13],xmm13[14],xmm10[14],xmm13[15],xmm10[15]
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm13 = [6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512DQ-NEXT:    vpshufb %xmm13, %xmm4, %xmm4
+; AVX512DQ-NEXT:    vpshufb %xmm13, %xmm10, %xmm10
+; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm10 = zmm26[0,1,0,1],zmm10[0,1,0,1]
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm11[8],xmm8[8],xmm11[9],xmm8[9],xmm11[10],xmm8[10],xmm11[11],xmm8[11],xmm11[12],xmm8[12],xmm11[13],xmm8[13],xmm11[14],xmm8[14],xmm11[15],xmm8[15]
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm11 = [2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512DQ-NEXT:    vmovdqa64 %xmm16, %xmm0
+; AVX512DQ-NEXT:    vpshufb %xmm11, %xmm0, %xmm0
+; AVX512DQ-NEXT:    vpshufb %xmm11, %xmm8, %xmm8
+; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm24[0,1,0,1],zmm8[0,1,0,1]
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm5[8],xmm15[8],xmm5[9],xmm15[9],xmm5[10],xmm15[10],xmm5[11],xmm15[11],xmm5[12],xmm15[12],xmm5[13],xmm15[13],xmm5[14],xmm15[14],xmm5[15],xmm15[15]
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} xmm5 = [u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX512DQ-NEXT:    vpshufb %xmm5, %xmm12, %xmm11
+; AVX512DQ-NEXT:    vpshufb %xmm5, %xmm2, %xmm2
+; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm2 = zmm22[0,1,0,1],zmm2[0,1,0,1]
+; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
+; AVX512DQ-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5, %zmm5 # 64-byte Folded Reload
+; AVX512DQ-NEXT:    # zmm5 = zmm5[2,3,2,3],mem[2,3,2,3]
+; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm12 # 64-byte Reload
+; AVX512DQ-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm12, %zmm12 # 64-byte Folded Reload
+; AVX512DQ-NEXT:    # zmm12 = zmm12[2,3,2,3],mem[2,3,2,3]
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm12 = zmm12 ^ (mem & (zmm12 ^ zmm5))
+; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
+; AVX512DQ-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5, %zmm5 # 64-byte Folded Reload
+; AVX512DQ-NEXT:    # zmm5 = zmm5[2,3,2,3],mem[2,3,2,3]
+; AVX512DQ-NEXT:    vmovdqa64 {{.*#+}} zmm13 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm5 = zmm5 ^ (zmm13 & (zmm5 ^ zmm12))
+; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm12 # 64-byte Reload
+; AVX512DQ-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm12, %zmm12 # 64-byte Folded Reload
+; AVX512DQ-NEXT:    # zmm12 = zmm12[2,3,2,3],mem[2,3,2,3]
+; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,0,1]
+; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,0,1]
+; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm14 = ymm17[2,3,2,3]
+; AVX512DQ-NEXT:    vmovdqa64 %xmm25, %xmm15
+; AVX512DQ-NEXT:    vpshuflw {{.*#+}} xmm15 = xmm15[1,1,0,0,4,5,6,7]
 ; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm15 = xmm15[0,1,2,0]
-; AVX512DQ-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm29, %zmm20 # 64-byte Folded Reload
-; AVX512DQ-NEXT:    # zmm20 = zmm29[2,3,2,3],mem[2,3,2,3]
-; AVX512DQ-NEXT:    vshufi64x2 {{.*#+}} zmm17 = zmm22[2,3,2,3],zmm25[2,3,2,3]
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm17 = zmm20 ^ (zmm23 & (zmm17 ^ zmm20))
-; AVX512DQ-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm6, %zmm6 # 32-byte Folded Reload
-; AVX512DQ-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm11, %zmm11 # 32-byte Folded Reload
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm11 = zmm11 ^ (mem & (zmm11 ^ zmm6))
-; AVX512DQ-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5, %zmm5 # 32-byte Folded Reload
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 | (zmm5 & mem)
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 ^ (mem & (zmm16 ^ zmm11))
-; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm5 = ymm15[0,0,1,0]
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm5, %zmm0, %zmm5
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm5 = zmm5 ^ (mem & (zmm5 ^ zmm30))
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm5 = zmm5 ^ (mem & (zmm5 ^ zmm18))
-; AVX512DQ-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Folded Reload
-; AVX512DQ-NEXT:    # zmm3 = mem[2,3,2,3,6,7,6,7]
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm3 ^ (mem & (zmm3 ^ zmm14))
-; AVX512DQ-NEXT:    vpermq {{.*#+}} zmm0 = zmm4[0,1,0,1,4,5,4,5]
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm28 ^ (mem & (zmm0 ^ zmm28))
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm21))
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
-; AVX512DQ-NEXT:    vshufi64x2 $84, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm2 # 64-byte Folded Reload
-; AVX512DQ-NEXT:    # zmm2 = zmm2[0,1,2,3],mem[2,3,2,3]
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm10 = zmm10 | (zmm2 & mem)
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm10 = zmm10 ^ (mem & (zmm10 ^ zmm27))
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm9 ^ (mem & (zmm9 ^ zmm1))
-; AVX512DQ-NEXT:    vpermq {{.*#+}} zmm1 = zmm12[0,1,0,1,4,5,4,5]
-; AVX512DQ-NEXT:    vpermq {{.*#+}} zmm2 = zmm7[0,0,1,0,4,4,5,4]
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm2 = zmm2 ^ (mem & (zmm2 ^ zmm1))
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm2 = zmm2 ^ (mem & (zmm2 ^ zmm9))
-; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512DQ-NEXT:    vporq {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm1 # 64-byte Folded Reload
-; AVX512DQ-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[2,3,2,3,6,7,6,7]
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm13 = zmm13 ^ (mem & (zmm13 ^ zmm1))
-; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm13 = zmm13 ^ (mem & (zmm13 ^ zmm17))
+; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm17 # 64-byte Reload
+; AVX512DQ-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm17, %zmm17 # 64-byte Folded Reload
+; AVX512DQ-NEXT:    # zmm17 = zmm17[2,3,2,3],mem[2,3,2,3]
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm17 = zmm12 ^ (zmm13 & (zmm17 ^ zmm12))
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm8 = zmm10 ^ (zmm13 & (zmm8 ^ zmm10))
+; AVX512DQ-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 32-byte Folded Reload
+; AVX512DQ-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm4 # 32-byte Folded Reload
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm4 = zmm4 ^ (mem & (zmm4 ^ zmm0))
+; AVX512DQ-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm11, %zmm0 # 32-byte Folded Reload
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm3 | (zmm0 & mem)
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm3 ^ (mem & (zmm3 ^ zmm4))
+; AVX512DQ-NEXT:    vpermq {{.*#+}} ymm0 = ymm15[0,0,1,0]
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm14, %zmm0
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm19))
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm7))
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm29 = zmm29 ^ (mem & (zmm29 ^ zmm5))
+; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512DQ-NEXT:    vshufi64x2 $84, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm4 # 64-byte Folded Reload
+; AVX512DQ-NEXT:    # zmm4 = zmm4[0,1,2,3],mem[2,3,2,3]
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm9 | (zmm4 & mem)
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm9 ^ (mem & (zmm9 ^ zmm31))
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm6 = zmm6 ^ (mem & (zmm6 ^ zmm28))
+; AVX512DQ-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[0,0,1,0,4,4,5,4]
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm1 = zmm1 ^ (mem & (zmm1 ^ zmm23))
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm1 = zmm1 ^ (mem & (zmm1 ^ zmm6))
+; AVX512DQ-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512DQ-NEXT:    vporq {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm4 # 64-byte Folded Reload
+; AVX512DQ-NEXT:    vpermq {{.*#+}} zmm4 = zmm4[2,3,2,3,6,7,6,7]
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm18 = zmm18 ^ (mem & (zmm18 ^ zmm4))
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm18 = zmm18 ^ (mem & (zmm18 ^ zmm17))
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm2 = zmm20 ^ (mem & (zmm2 ^ zmm20))
+; AVX512DQ-NEXT:    vpternlogq {{.*#+}} zmm2 = zmm2 ^ (mem & (zmm2 ^ zmm8))
 ; AVX512DQ-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512DQ-NEXT:    vmovdqa64 %zmm13, 128(%rax)
-; AVX512DQ-NEXT:    vmovdqa64 %zmm2, (%rax)
-; AVX512DQ-NEXT:    vmovdqa64 %zmm10, 320(%rax)
-; AVX512DQ-NEXT:    vmovdqa64 %zmm0, 256(%rax)
-; AVX512DQ-NEXT:    vmovdqa64 %zmm3, 384(%rax)
-; AVX512DQ-NEXT:    vmovdqa64 %zmm5, 192(%rax)
-; AVX512DQ-NEXT:    vmovdqa64 %zmm16, 64(%rax)
-; AVX512DQ-NEXT:    addq $1560, %rsp # imm = 0x618
+; AVX512DQ-NEXT:    vmovdqa64 %zmm2, 256(%rax)
+; AVX512DQ-NEXT:    vmovdqa64 %zmm18, 128(%rax)
+; AVX512DQ-NEXT:    vmovdqa64 %zmm1, (%rax)
+; AVX512DQ-NEXT:    vmovdqa64 %zmm9, 320(%rax)
+; AVX512DQ-NEXT:    vmovdqa64 %zmm29, 384(%rax)
+; AVX512DQ-NEXT:    vmovdqa64 %zmm0, 192(%rax)
+; AVX512DQ-NEXT:    vmovdqa64 %zmm3, 64(%rax)
+; AVX512DQ-NEXT:    addq $1496, %rsp # imm = 0x5D8
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512DQ-FCP-LABEL: store_i8_stride7_vf64:
 ; AVX512DQ-FCP:       # %bb.0:
-; AVX512DQ-FCP-NEXT:    subq $1640, %rsp # imm = 0x668
-; AVX512DQ-FCP-NEXT:    vmovdqa (%rsi), %ymm7
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm7[14],zero,zero,zero,zero,zero,zero,ymm7[15],zero,zero,zero,zero,zero,zero,ymm7[16],zero,zero,zero,zero,zero,zero,ymm7[17],zero,zero,zero,zero,zero,zero,ymm7[18]
-; AVX512DQ-FCP-NEXT:    vmovdqa (%rdi), %ymm14
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm14[0,1,14],zero,ymm14[12,13,0,1,14,15],zero,ymm14[3,12,13,2,3,16],zero,ymm14[30,31,28,29,16,17],zero,ymm14[31,18,19,28,29,18],zero
+; AVX512DQ-FCP-NEXT:    subq $1416, %rsp # imm = 0x588
+; AVX512DQ-FCP-NEXT:    vmovdqa (%rsi), %ymm12
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm12[14],zero,zero,zero,zero,zero,zero,ymm12[15],zero,zero,zero,zero,zero,zero,ymm12[16],zero,zero,zero,zero,zero,zero,ymm12[17],zero,zero,zero,zero,zero,zero,ymm12[18]
+; AVX512DQ-FCP-NEXT:    vmovdqa (%rdi), %ymm9
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm9[0,1,14],zero,ymm9[12,13,0,1,14,15],zero,ymm9[3,12,13,2,3,16],zero,ymm9[30,31,28,29,16,17],zero,ymm9[31,18,19,28,29,18],zero
 ; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa (%rcx), %ymm2
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,ymm2[14],zero,zero,zero,zero,zero,zero,ymm2[15],zero,zero,zero,zero,zero,zero,ymm2[16],zero,zero,zero,zero,zero,zero,ymm2[17],zero,zero,zero,zero,zero
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm2, %ymm25
-; AVX512DQ-FCP-NEXT:    vmovdqa (%rdx), %ymm3
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[0,1,0,1,14],zero,ymm3[14,15,0,1,14,15],zero,ymm3[13,14,15,16,17,16],zero,ymm3[30,31,30,31,16,17],zero,ymm3[31,28,29,30,31]
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm3, %ymm26
-; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512DQ-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa (%r8), %ymm1
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,ymm1[14],zero,zero,zero,zero,zero,zero,ymm1[15],zero,zero,zero,zero,zero,zero,ymm1[16],zero,zero,zero,zero,zero,zero,ymm1[17],zero,zero,zero,zero
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm1, %ymm21
-; AVX512DQ-FCP-NEXT:    vmovdqa (%r9), %ymm2
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[13,u,u,u,u,u],zero,ymm2[14,u,u,u,u,u],zero,ymm2[15,u,u,u,u,u],zero,ymm2[16,u,u,u,u,u],zero,ymm2[17,u,u,u]
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm2, %ymm28
-; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqa (%rcx), %ymm6
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero,zero
+; AVX512DQ-FCP-NEXT:    vmovdqa (%rdx), %ymm4
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm4[0,1,0,1,14],zero,ymm4[14,15,0,1,14,15],zero,ymm4[13,14,15,16,17,16],zero,ymm4[30,31,30,31,16,17],zero,ymm4[31,28,29,30,31]
+; AVX512DQ-FCP-NEXT:    vpor %ymm1, %ymm3, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX512DQ-FCP-NEXT:    vmovdqa (%r8), %ymm8
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = zero,zero,zero,zero,zero,zero,ymm8[14],zero,zero,zero,zero,zero,zero,ymm8[15],zero,zero,zero,zero,zero,zero,ymm8[16],zero,zero,zero,zero,zero,zero,ymm8[17],zero,zero,zero,zero
+; AVX512DQ-FCP-NEXT:    vmovdqa (%r9), %ymm11
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[13,u,u,u,u,u],zero,ymm11[14,u,u,u,u,u],zero,ymm11[15,u,u,u,u,u],zero,ymm11[16,u,u,u,u,u],zero,ymm11[17,u,u,u]
+; AVX512DQ-FCP-NEXT:    vpor %ymm3, %ymm5, %ymm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rsi), %ymm2
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero,zero,zero
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rdi), %ymm3
-; AVX512DQ-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,30],zero,ymm3[28],zero,ymm3[30,31,30,31],zero,ymm3[29],zero,ymm3[31,28,29]
-; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rsi), %ymm0
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128]
+; AVX512DQ-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm0, %ymm3
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm1, %ymm18
+; AVX512DQ-FCP-NEXT:    vmovdqa %ymm0, %ymm1
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rdi), %ymm15
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [28,29,30,128,28,128,30,31,30,31,128,29,128,31,28,29,28,29,30,128,28,128,30,31,30,31,128,29,128,31,28,29]
+; AVX512DQ-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm15, %ymm5
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm0, %ymm22
+; AVX512DQ-FCP-NEXT:    vpor %ymm3, %ymm5, %ymm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm15 = [128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25]
-; AVX512DQ-FCP-NEXT:    # ymm15 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm15, %ymm3, %ymm0
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128,25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128]
-; AVX512DQ-FCP-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm3, %ymm2, %ymm1
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm3, %ymm16
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm2, %ymm18
-; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25]
+; AVX512DQ-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm2, %ymm15, %ymm3
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm2, %ymm20
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128,25,128,23,0,0,0,128,26,128,24,0,0,0,128,27,128]
+; AVX512DQ-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm2, %ymm1, %ymm5
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm2, %ymm25
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm1, %ymm19
+; AVX512DQ-FCP-NEXT:    vpor %ymm3, %ymm5, %ymm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rdx), %ymm2
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero,zero
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rcx), %ymm11
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u],zero,ymm11[30],zero,ymm11[28,u,u,u],zero,ymm11[31],zero,ymm11[29,u]
-; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rdx), %ymm13
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,128]
+; AVX512DQ-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm13, %ymm5
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm0, %ymm24
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rcx), %ymm3
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,0]
+; AVX512DQ-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm3, %ymm7
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm0, %ymm17
+; AVX512DQ-FCP-NEXT:    vpor %ymm5, %ymm7, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128]
+; AVX512DQ-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm2, %ymm3, %ymm5
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm2, %ymm26
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27,24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27]
+; AVX512DQ-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm2, %ymm13, %ymm7
+; AVX512DQ-FCP-NEXT:    vpor %ymm5, %ymm7, %ymm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128]
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%r8), %ymm14
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [128,27,128,128,128,128,30,128,28,128,128,128,128,31,128,29,128,27,128,128,128,128,30,128,28,128,128,128,128,31,128,29]
+; AVX512DQ-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm14, %ymm5
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm0, %ymm21
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%r9), %ymm10
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = [29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128,29,128,27,0,0,0,128,30,128,28,0,0,0,128,31,128]
+; AVX512DQ-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm10, %ymm7
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm0, %ymm23
+; AVX512DQ-FCP-NEXT:    vpor %ymm5, %ymm7, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128]
+; AVX512DQ-FCP-NEXT:    # ymm5 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm5, %ymm10, %ymm0
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [24,25,24,25,128,23,128,23,24,25,26,128,24,128,30,31,24,25,24,25,128,23,128,23,24,25,26,128,24,128,30,31]
 ; AVX512DQ-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm11, %ymm0
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm1, %ymm30
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27,24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27]
-; AVX512DQ-FCP-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm3, %ymm2, %ymm1
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm3, %ymm23
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm2, %ymm27
-; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm14, %ymm7
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm1, %ymm27
+; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm7, %ymm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%r8), %ymm2
-; AVX512DQ-FCP-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29]
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%r9), %ymm3
-; AVX512DQ-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm3[27,u,u,u],zero,ymm3[30],zero,ymm3[28,u,u,u],zero,ymm3[31],zero
-; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rax), %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512DQ-FCP-NEXT:    # ymm7 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm7, %ymm0, %ymm1
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm7, %ymm16
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm7 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm7[2,3,2,3],zmm1[2,3,2,3]
+; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm20, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm9, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm25, %ymm1
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm12, %ymm7
+; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm7, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm12[21],zero,ymm12[19],zero,zero,zero,zero,ymm12[22],zero,ymm12[20],zero,zero
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm7 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21],zero,ymm9[19],zero,ymm9[21,20,21,22],zero,ymm9[20],zero,ymm9[22,23]
+; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm7, %ymm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128]
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm26, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm6, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm2, %ymm4, %ymm7
+; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm7, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20]
+; AVX512DQ-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm6, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm1, %ymm26
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128,18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128]
 ; AVX512DQ-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm3, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm4, %ymm7
 ; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm1, %ymm20
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm2[23],zero,ymm2[23,24,25,26],zero,ymm2[24],zero,ymm2[30,31]
-; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm7, %ymm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rax), %ymm1
-; AVX512DQ-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm5, %ymm11, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm11[20],zero,ymm11[18],zero,zero,zero,zero,ymm11[21],zero,ymm11[19],zero,zero,zero,zero,ymm11[22]
+; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm0, %zmm5, %zmm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rdx), %xmm2
-; AVX512DQ-FCP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rcx), %xmm0
-; AVX512DQ-FCP-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm4 = [u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm4, %xmm2, %xmm5
-; AVX512DQ-FCP-NEXT:    vpor %xmm0, %xmm5, %xmm0
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm27, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm8, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm5 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,ymm8[18],zero,ymm8[20,21,20,21],zero,ymm8[19],zero,ymm8[19,20,21,22],zero
+; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm0, %zmm5, %zmm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rdi), %xmm2
-; AVX512DQ-FCP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rsi), %xmm0
-; AVX512DQ-FCP-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm5 = [u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm5, %xmm0, %xmm0
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm10 = [u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm10, %xmm2, %xmm6
-; AVX512DQ-FCP-NEXT:    vpor %xmm0, %xmm6, %xmm0
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm18, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm12, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm22, %ymm1
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm9, %ymm5
+; AVX512DQ-FCP-NEXT:    vporq %ymm0, %ymm5, %ymm30
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm24, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm17, %ymm1
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm6, %ymm2
+; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm2 = [0,0,0,0,2,3,0,1,0,18,0,19,18,0,19,0]
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rax), %xmm0
-; AVX512DQ-FCP-NEXT:    vpshufhw {{.*#+}} xmm6 = xmm0[0,1,2,3,4,5,5,6]
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm9 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm9, %xmm0, %xmm8
-; AVX512DQ-FCP-NEXT:    vpermi2d %zmm6, %zmm8, %zmm2
-; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%r9), %xmm2
-; AVX512DQ-FCP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa 32(%r8), %xmm3
-; AVX512DQ-FCP-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512DQ-FCP-NEXT:    vmovdqa (%rax), %ymm4
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm16, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm29 = [4,0,6,0,4,0,6,7,0,17,0,17,0,16,16,0]
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rax), %xmm2
+; AVX512DQ-FCP-NEXT:    vpshuflw {{.*#+}} xmm5 = xmm2[1,1,0,0,4,5,6,7]
+; AVX512DQ-FCP-NEXT:    vpermi2d %zmm5, %zmm0, %zmm29
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm21, %ymm0
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm8, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm23, %ymm1
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm11, %ymm1
+; AVX512DQ-FCP-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rdx), %xmm5
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rcx), %xmm1
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm8 = [u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm8, %xmm1, %xmm0
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm1, %xmm24
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} xmm1 = xmm5[u,u,u,7],zero,xmm5[5],zero,xmm5[u,u,u,8],zero,xmm5[6],zero,xmm5[u,u]
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm5, %xmm23
+; AVX512DQ-FCP-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rdi), %xmm7
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%rsi), %xmm12
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm12, %xmm5
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm0 = [u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm0, %xmm7, %xmm6
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm7, %xmm21
+; AVX512DQ-FCP-NEXT:    vpor %xmm5, %xmm6, %xmm5
+; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512DQ-FCP-NEXT:    vpshufhw {{.*#+}} xmm5 = xmm2[0,1,2,3,4,5,5,6]
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm6 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm6, %xmm2, %xmm2
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm6, %xmm16
+; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm25 = [0,0,0,0,2,3,0,1,0,18,0,19,18,0,19,0]
+; AVX512DQ-FCP-NEXT:    vpermi2d %zmm5, %zmm2, %zmm25
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%r9), %xmm11
+; AVX512DQ-FCP-NEXT:    vmovdqa 32(%r8), %xmm5
 ; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm6 = [128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm6, %xmm2, %xmm12
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm8 = [4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm8, %xmm3, %xmm13
-; AVX512DQ-FCP-NEXT:    vpor %xmm12, %xmm13, %xmm2
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm6, %xmm11, %xmm2
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm6, %xmm17
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm7 = [4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm7, %xmm5, %xmm6
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm7, %xmm18
+; AVX512DQ-FCP-NEXT:    vporq %xmm2, %xmm6, %xmm31
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm6 = ymm13[0,1,0,1,14],zero,ymm13[14,15,0,1,14,15],zero,ymm13[13,14,15,16,17,16],zero,ymm13[30,31,30,31,16,17],zero,ymm13[31,28,29,30,31]
+; AVX512DQ-FCP-NEXT:    vpor %ymm2, %ymm6, %ymm2
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa (%rsi), %xmm2
-; AVX512DQ-FCP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm5, %xmm2, %xmm5
-; AVX512DQ-FCP-NEXT:    vmovdqa (%rdi), %xmm2
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm10, %xmm2, %xmm10
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm2, %xmm22
-; AVX512DQ-FCP-NEXT:    vpor %xmm5, %xmm10, %xmm2
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm26, %ymm2
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm20, %ymm3
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm3, %ymm13, %ymm3
+; AVX512DQ-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm2
+; AVX512DQ-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm19, %ymm6
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero,zero,zero,ymm6[18]
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm15[0,1,14],zero,ymm15[12,13,0,1,14,15],zero,ymm15[3,12,13,2,3,16],zero,ymm15[30,31,28,29,16,17],zero,ymm15[31,18,19,28,29,18],zero
+; AVX512DQ-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa (%rcx), %xmm2
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm2, %xmm17
-; AVX512DQ-FCP-NEXT:    vmovdqa (%rdx), %xmm12
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm4, %xmm12, %xmm4
-; AVX512DQ-FCP-NEXT:    vpor %xmm1, %xmm4, %xmm1
-; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa (%rax), %xmm5
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm9, %xmm5, %xmm4
-; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,1,0,1,0,0,0,0,16,0,16,0,18,19,0,17]
-; AVX512DQ-FCP-NEXT:    vpshuflw {{.*#+}} xmm9 = xmm5[1,1,0,0,4,5,6,7]
-; AVX512DQ-FCP-NEXT:    vpermi2d %zmm4, %zmm9, %zmm1
-; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa (%r9), %xmm1
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm6, %xmm1, %xmm4
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm1, %xmm31
-; AVX512DQ-FCP-NEXT:    vmovdqa (%r8), %xmm1
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm8, %xmm1, %xmm8
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm1, %xmm29
-; AVX512DQ-FCP-NEXT:    vpor %xmm4, %xmm8, %xmm1
-; AVX512DQ-FCP-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa %ymm14, %ymm3
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm15, %ymm14, %ymm4
-; AVX512DQ-FCP-NEXT:    vmovdqa %ymm7, %ymm10
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm16, %ymm1
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm7, %ymm8
-; AVX512DQ-FCP-NEXT:    vpor %ymm4, %ymm8, %ymm1
-; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128]
-; AVX512DQ-FCP-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm7, %ymm10, %ymm4
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [18,19,20,21,128,19,128,21,20,21,22,128,20,128,22,23,18,19,20,21,128,19,128,21,20,21,22,128,20,128,22,23]
-; AVX512DQ-FCP-NEXT:    # ymm13 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm13, %ymm14, %ymm8
-; AVX512DQ-FCP-NEXT:    vpor %ymm4, %ymm8, %ymm1
-; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm25, %ymm1
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm30, %ymm2
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm2, %ymm1, %ymm4
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm26, %ymm6
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm23, %ymm2
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm2, %ymm6, %ymm8
-; AVX512DQ-FCP-NEXT:    vpor %ymm4, %ymm8, %ymm2
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm6[21],zero,ymm6[19],zero,zero,zero,zero,ymm6[22],zero,ymm6[20],zero,zero
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21],zero,ymm15[19],zero,ymm15[21,20,21,22],zero,ymm15[20],zero,ymm15[22,23]
+; AVX512DQ-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm2
+; AVX512DQ-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm10[20],zero,ymm10[18],zero,zero,zero,zero,ymm10[21],zero,ymm10[19],zero,zero,zero,zero,ymm10[22]
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,ymm14[18],zero,ymm14[20,21,20,21],zero,ymm14[19],zero,ymm14[19,20,21,22],zero
+; AVX512DQ-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20]
-; AVX512DQ-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm2, %ymm1, %ymm4
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm2, %ymm30
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm25, %ymm2
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128,18,128,18,19,20,21,128,19,128,25,26,27,22,128,20,128]
-; AVX512DQ-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm6, %ymm8
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm1, %ymm16
-; AVX512DQ-FCP-NEXT:    vpor %ymm4, %ymm8, %ymm1
-; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm28, %ymm1
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm20, %ymm4
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm4, %ymm1, %ymm4
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22]
-; AVX512DQ-FCP-NEXT:    # ymm9 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm9, %ymm1, %ymm8
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm9, %ymm19
-; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm4, %zmm8, %zmm4
-; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vpshuflw {{.*#+}} xmm8 = xmm0[1,1,0,0,4,5,6,7]
-; AVX512DQ-FCP-NEXT:    vmovdqa (%rax), %ymm15
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm25 = [4,0,6,0,4,0,6,7,0,17,0,17,0,16,16,0]
-; AVX512DQ-FCP-NEXT:    vpermi2d %zmm8, %zmm9, %zmm25
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm21, %ymm0
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm0[23],zero,ymm0[23,24,25,26],zero,ymm0[24],zero,ymm0[30,31]
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128,20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128]
-; AVX512DQ-FCP-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm8, %ymm0, %ymm14
-; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm9, %zmm14, %zmm4
-; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm10[30],zero,ymm10[28],zero,zero,zero,zero,ymm10[31],zero,ymm10[29],zero,zero,zero
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm14 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,30],zero,ymm3[28],zero,ymm3[30,31,30,31],zero,ymm3[29],zero,ymm3[31,28,29]
-; AVX512DQ-FCP-NEXT:    vporq %ymm9, %ymm14, %ymm26
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm6[30],zero,ymm6[28],zero,zero,zero,zero,ymm6[31],zero,ymm6[29],zero,zero
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm14 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u],zero,ymm2[30],zero,ymm2[28,u,u,u],zero,ymm2[31],zero,ymm2[29,u]
-; AVX512DQ-FCP-NEXT:    vporq %ymm9, %ymm14, %ymm24
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[27],zero,zero,zero,zero,ymm0[30],zero,ymm0[28],zero,zero,zero,zero,ymm0[31],zero,ymm0[29]
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm14 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm1[27,u,u,u],zero,ymm1[30],zero,ymm1[28,u,u,u],zero,ymm1[31],zero
-; AVX512DQ-FCP-NEXT:    vporq %ymm9, %ymm14, %ymm23
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = zero,zero,zero,zero,zero,ymm11[14],zero,zero,zero,zero,zero,zero,ymm11[15],zero,zero,zero,zero,zero,zero,ymm11[16],zero,zero,zero,zero,zero,zero,ymm11[17],zero,zero,zero,zero,zero
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm27, %ymm1
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm14 = ymm1[0,1,0,1,14],zero,ymm1[14,15,0,1,14,15],zero,ymm1[13,14,15,16,17,16],zero,ymm1[30,31,30,31,16,17],zero,ymm1[31,28,29,30,31]
-; AVX512DQ-FCP-NEXT:    vpor %ymm9, %ymm14, %ymm0
-; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm30, %ymm0
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm11, %ymm2
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm16, %ymm0
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm0, %ymm1, %ymm3
-; AVX512DQ-FCP-NEXT:    vporq %ymm2, %ymm3, %ymm30
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm18, %ymm4
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,ymm4[14],zero,zero,zero,zero,zero,zero,ymm4[15],zero,zero,zero,zero,zero,zero,ymm4[16],zero,zero,zero,zero,zero,zero,ymm4[17],zero,zero,zero,zero,zero,zero,ymm4[18]
-; AVX512DQ-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm1[0,1,14],zero,ymm1[12,13,0,1,14,15],zero,ymm1[3,12,13,2,3,16],zero,ymm1[30,31,28,29,16,17],zero,ymm1[31,18,19,28,29,18],zero
-; AVX512DQ-FCP-NEXT:    vporq %ymm2, %ymm3, %ymm28
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm7, %ymm4, %ymm2
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm13, %ymm1, %ymm3
-; AVX512DQ-FCP-NEXT:    vporq %ymm2, %ymm3, %ymm27
-; AVX512DQ-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm19, %ymm1
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm0, %ymm2
-; AVX512DQ-FCP-NEXT:    vmovdqu (%rsp), %ymm1 # 32-byte Reload
-; AVX512DQ-FCP-NEXT:    vpshufb %ymm8, %ymm1, %ymm3
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,ymm14[14],zero,zero,zero,zero,zero,zero,ymm14[15],zero,zero,zero,zero,zero,zero,ymm14[16],zero,zero,zero,zero,zero,zero,ymm14[17],zero,zero,zero,zero
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm10[13,u,u,u,u,u],zero,ymm10[14,u,u,u,u,u],zero,ymm10[15,u,u,u,u,u],zero,ymm10[16,u,u,u,u,u],zero,ymm10[17,u,u,u]
 ; AVX512DQ-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,ymm1[14],zero,zero,zero,zero,zero,zero,ymm1[15],zero,zero,zero,zero,zero,zero,ymm1[16],zero,zero,zero,zero,zero,zero,ymm1[17],zero,zero,zero,zero
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm3 = ymm0[13,u,u,u,u,u],zero,ymm0[14,u,u,u,u,u],zero,ymm0[15,u,u,u,u,u],zero,ymm0[16,u,u,u,u,u],zero,ymm0[17,u,u,u]
-; AVX512DQ-FCP-NEXT:    vpor %ymm2, %ymm3, %ymm0
-; AVX512DQ-FCP-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512DQ-FCP-NEXT:    vpshufhw {{.*#+}} xmm1 = xmm5[0,1,2,3,4,5,5,6]
-; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [2,2,3,3,2,2,3,3]
-; AVX512DQ-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX512DQ-FCP-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm2 = zero,ymm15[13],zero,zero,zero,zero,zero,zero,ymm15[14],zero,zero,zero,zero,zero,zero,ymm15[15],zero,zero,zero,zero,zero,zero,ymm15[16],zero,zero,zero,zero,zero,zero,ymm15[17],zero,zero
-; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm16
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm17, %xmm8
-; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm8[8],xmm12[8],xmm8[9],xmm12[9],xmm8[10],xmm12[10],xmm8[11],xmm12[11],xmm8[12],xmm12[12],xmm8[13],xmm12[13],xmm8[14],xmm12[14],xmm8[15],xmm12[15]
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm3 = [6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm3, %xmm2, %xmm0
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm0, %ymm19
-; AVX512DQ-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
-; AVX512DQ-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm13[8],xmm9[8],xmm13[9],xmm9[9],xmm13[10],xmm9[10],xmm13[11],xmm9[11],xmm13[12],xmm9[12],xmm13[13],xmm9[13],xmm13[14],xmm9[14],xmm13[15],xmm9[15]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm20 = zmm0[0,1,0,1],zmm2[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm22, %xmm1
-; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm11[8],xmm1[8],xmm11[9],xmm1[9],xmm11[10],xmm1[10],xmm11[11],xmm1[11],xmm11[12],xmm1[12],xmm11[13],xmm1[13],xmm11[14],xmm1[14],xmm11[15],xmm1[15]
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm0 = [2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm0, %xmm2, %xmm2
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm2, %ymm18
-; AVX512DQ-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; AVX512DQ-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm7[8],xmm5[8],xmm7[9],xmm5[9],xmm7[10],xmm5[10],xmm7[11],xmm5[11],xmm7[12],xmm5[12],xmm7[13],xmm5[13],xmm7[14],xmm5[14],xmm7[15],xmm5[15]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm0, %xmm2, %xmm2
-; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm22 = zmm0[0,1,0,1],zmm2[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm31, %xmm3
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm29, %xmm2
-; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm3[8],xmm2[9],xmm3[9],xmm2[10],xmm3[10],xmm2[11],xmm3[11],xmm2[12],xmm3[12],xmm2[13],xmm3[13],xmm2[14],xmm3[14],xmm2[15],xmm3[15]
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm0 = [u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm0, %xmm4, %xmm4
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm4, %ymm17
-; AVX512DQ-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; AVX512DQ-FCP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm14 = xmm4[8],xmm6[8],xmm4[9],xmm6[9],xmm4[10],xmm6[10],xmm4[11],xmm6[11],xmm4[12],xmm6[12],xmm4[13],xmm6[13],xmm4[14],xmm6[14],xmm4[15],xmm6[15]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm0, %xmm14, %xmm0
-; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm10 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vinserti32x4 $2, %xmm0, %zmm10, %zmm29
-; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm11[0],xmm1[1],xmm11[1],xmm1[2],xmm11[2],xmm1[3],xmm11[3],xmm1[4],xmm11[4],xmm1[5],xmm11[5],xmm1[6],xmm11[6],xmm1[7],xmm11[7]
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm14 = [0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm14, %xmm0, %xmm0
-; AVX512DQ-FCP-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm21 # 64-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    # zmm21 = zmm0[0,1,0,1],mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm12[0],xmm8[0],xmm12[1],xmm8[1],xmm12[2],xmm8[2],xmm12[3],xmm8[3],xmm12[4],xmm8[4],xmm12[5],xmm8[5],xmm12[6],xmm8[6],xmm12[7],xmm8[7]
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm8 = [4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm8, %xmm0, %xmm0
-; AVX512DQ-FCP-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm11 # 64-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    # zmm11 = zmm0[0,1,0,1],mem[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3],xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX512DQ-FCP-NEXT:    vinserti32x4 $2, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm31 # 16-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm10 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512DQ-FCP-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm15[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm2 = [4,5,4,5,5,7,4,5,20,21,22,23,20,21,22,23]
-; AVX512DQ-FCP-NEXT:    vpermt2d %zmm10, %zmm2, %zmm0
-; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm10 = xmm5[0],xmm7[0],xmm5[1],xmm7[1],xmm5[2],xmm7[2],xmm5[3],xmm7[3],xmm5[4],xmm7[4],xmm5[5],xmm7[5],xmm5[6],xmm7[6],xmm5[7],xmm7[7]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm14, %xmm10, %xmm10
-; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm10 = zmm26[2,3,2,3],zmm10[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm14 = xmm9[0],xmm13[0],xmm9[1],xmm13[1],xmm9[2],xmm13[2],xmm9[3],xmm13[3],xmm9[4],xmm13[4],xmm9[5],xmm13[5],xmm9[6],xmm13[6],xmm9[7],xmm13[7]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm8, %xmm14, %xmm14
-; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm14 = zmm24[2,3,2,3],zmm14[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm15 = xmm4[0],xmm6[0],xmm4[1],xmm6[1],xmm4[2],xmm6[2],xmm4[3],xmm6[3],xmm4[4],xmm6[4],xmm4[5],xmm6[5],xmm4[6],xmm6[6],xmm4[7],xmm6[7]
-; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm15, %xmm12
-; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm12 = zmm23[2,3,2,3],zmm12[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm9 = zero,ymm15[13],zero,zero,zero,zero,zero,zero,ymm15[14],zero,zero,zero,zero,zero,zero,ymm15[15],zero,zero,zero,zero,zero,zero,ymm15[16],zero,zero,zero,zero,zero,zero,ymm15[17],zero,zero
-; AVX512DQ-FCP-NEXT:    vpshuflw {{.*#+}} ymm15 = ymm15[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512DQ-FCP-NEXT:    vpermd %ymm15, %ymm2, %ymm4
-; AVX512DQ-FCP-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm4, %ymm4
-; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm4, %zmm9, %zmm4
-; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm9 # 64-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    # zmm9 = zmm1[2,3,2,3],mem[2,3,2,3]
-; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm15 # 64-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    # zmm15 = zmm1[2,3,2,3],mem[2,3,2,3]
-; AVX512DQ-FCP-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm15 = zmm9 ^ (zmm1 & (zmm15 ^ zmm9))
+; AVX512DQ-FCP-NEXT:    vmovdqa (%rsi), %xmm10
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm10, %xmm1
+; AVX512DQ-FCP-NEXT:    vmovdqa (%rdi), %xmm9
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm0, %xmm9, %xmm0
+; AVX512DQ-FCP-NEXT:    vporq %xmm1, %xmm0, %xmm20
+; AVX512DQ-FCP-NEXT:    vmovdqa (%rcx), %xmm13
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm8, %xmm13, %xmm0
+; AVX512DQ-FCP-NEXT:    vmovdqa (%rdx), %xmm8
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} xmm1 = xmm8[u,u,u,7],zero,xmm8[5],zero,xmm8[u,u,u,8],zero,xmm8[6],zero,xmm8[u,u]
+; AVX512DQ-FCP-NEXT:    vporq %xmm0, %xmm1, %xmm19
+; AVX512DQ-FCP-NEXT:    vmovdqa (%rax), %xmm0
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm16, %xmm1
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
+; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm16 = [0,1,0,1,0,0,0,0,16,0,16,0,18,19,0,17]
+; AVX512DQ-FCP-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm0[1,1,0,0,4,5,6,7]
+; AVX512DQ-FCP-NEXT:    vpermi2d %zmm1, %zmm2, %zmm16
+; AVX512DQ-FCP-NEXT:    vmovdqa (%r9), %xmm7
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm17, %xmm1
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm7, %xmm3
+; AVX512DQ-FCP-NEXT:    vmovdqa (%r8), %xmm6
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm18, %xmm1
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm6, %xmm15
+; AVX512DQ-FCP-NEXT:    vporq %xmm3, %xmm15, %xmm18
+; AVX512DQ-FCP-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,5,6]
+; AVX512DQ-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [2,2,3,3,2,2,3,3]
+; AVX512DQ-FCP-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpermd %ymm0, %ymm3, %ymm0
+; AVX512DQ-FCP-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128]
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm4, %ymm3
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm1, %ymm28
+; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm14
+; AVX512DQ-FCP-NEXT:    vpshufb {{.*#+}} ymm15 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512DQ-FCP-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm4[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512DQ-FCP-NEXT:    vpmovsxbd {{.*#+}} zmm17 = [4,5,4,5,5,7,4,5,20,21,22,23,20,21,22,23]
+; AVX512DQ-FCP-NEXT:    vpermt2d %zmm15, %zmm17, %zmm0
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm21, %xmm3
+; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm3[0],xmm12[0],xmm3[1],xmm12[1],xmm3[2],xmm12[2],xmm3[3],xmm12[3],xmm3[4],xmm12[4],xmm3[5],xmm12[5],xmm3[6],xmm12[6],xmm3[7],xmm12[7]
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm4, %xmm4
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm1, %xmm27
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm30 = zmm30[2,3,2,3],zmm4[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm23, %xmm1
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm24, %xmm15
+; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm1[0],xmm15[0],xmm1[1],xmm15[1],xmm1[2],xmm15[2],xmm1[3],xmm15[3],xmm1[4],xmm15[4],xmm1[5],xmm15[5],xmm1[6],xmm15[6],xmm1[7],xmm15[7]
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm2, %xmm4, %xmm4
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm2, %xmm22
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm23 = zmm2[2,3,2,3],zmm4[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm5[0],xmm11[0],xmm5[1],xmm11[1],xmm5[2],xmm11[2],xmm5[3],xmm11[3],xmm5[4],xmm11[4],xmm5[5],xmm11[5],xmm5[6],xmm11[6],xmm5[7],xmm11[7]
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm2 = [u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm2, %xmm4, %xmm4
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm2, %xmm26
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vporq {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm9 # 64-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} zmm9 = zmm9[2,3,2,3,6,7,6,7]
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm4 = zmm2[2,3,2,3],zmm4[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm15[8],xmm1[8],xmm15[9],xmm1[9],xmm15[10],xmm1[10],xmm15[11],xmm1[11],xmm15[12],xmm1[12],xmm15[13],xmm1[13],xmm15[14],xmm1[14],xmm15[15],xmm1[15]
+; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm15 = xmm13[8],xmm8[8],xmm13[9],xmm8[9],xmm13[10],xmm8[10],xmm13[11],xmm8[11],xmm13[12],xmm8[12],xmm13[13],xmm8[13],xmm13[14],xmm8[14],xmm13[15],xmm8[15]
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm2 = [6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm2, %xmm15, %xmm15
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm24 = zmm2[0,1,0,1],zmm1[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm12[8],xmm3[8],xmm12[9],xmm3[9],xmm12[10],xmm3[10],xmm12[11],xmm3[11],xmm12[12],xmm3[12],xmm12[13],xmm3[13],xmm12[14],xmm3[14],xmm12[15],xmm3[15]
+; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm12 = xmm10[8],xmm9[8],xmm10[9],xmm9[9],xmm10[10],xmm9[10],xmm10[11],xmm9[11],xmm10[12],xmm9[12],xmm10[13],xmm9[13],xmm10[14],xmm9[14],xmm10[15],xmm9[15]
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm2 = [2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm2, %xmm12, %xmm12
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm21 = zmm2[0,1,0,1],zmm1[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm5[8],xmm11[8],xmm5[9],xmm11[9],xmm5[10],xmm11[10],xmm5[11],xmm11[11],xmm5[12],xmm11[12],xmm5[13],xmm11[13],xmm5[14],xmm11[14],xmm5[15],xmm11[15]
+; AVX512DQ-FCP-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm6[8],xmm7[8],xmm6[9],xmm7[9],xmm6[10],xmm7[10],xmm6[11],xmm7[11],xmm6[12],xmm7[12],xmm6[13],xmm7[13],xmm6[14],xmm7[14],xmm6[15],xmm7[15]
+; AVX512DQ-FCP-NEXT:    vmovdqa {{.*#+}} xmm1 = [u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm5, %xmm11
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm5 = zmm31[0,1,0,1],zmm1[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %ymm28, %ymm1
+; AVX512DQ-FCP-NEXT:    vpshufb %ymm1, %ymm2, %ymm1
+; AVX512DQ-FCP-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm2[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512DQ-FCP-NEXT:    vpermd %ymm2, %ymm17, %ymm2
+; AVX512DQ-FCP-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3],xmm9[4],xmm10[4],xmm9[5],xmm10[5],xmm9[6],xmm10[6],xmm9[7],xmm10[7]
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm27, %xmm3
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm2 = zmm2[0,1,0,1],zmm20[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm8 = xmm8[0],xmm13[0],xmm8[1],xmm13[1],xmm8[2],xmm13[2],xmm8[3],xmm13[3],xmm8[4],xmm13[4],xmm8[5],xmm13[5],xmm8[6],xmm13[6],xmm8[7],xmm13[7]
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm22, %xmm3
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm3, %xmm8, %xmm8
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm8 = zmm8[0,1,0,1],zmm19[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpunpcklbw {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1],xmm6[2],xmm7[2],xmm6[3],xmm7[3],xmm6[4],xmm7[4],xmm6[5],xmm7[5],xmm6[6],xmm7[6],xmm6[7],xmm7[7]
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %xmm26, %xmm3
+; AVX512DQ-FCP-NEXT:    vpshufb %xmm3, %xmm6, %xmm6
+; AVX512DQ-FCP-NEXT:    vshufi64x2 {{.*#+}} zmm6 = zmm6[0,1,0,1],zmm18[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm7, %zmm7 # 64-byte Folded Reload
+; AVX512DQ-FCP-NEXT:    # zmm7 = zmm7[2,3,2,3],mem[2,3,2,3]
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9, %zmm9 # 64-byte Folded Reload
+; AVX512DQ-FCP-NEXT:    # zmm9 = zmm9[2,3,2,3],mem[2,3,2,3]
+; AVX512DQ-FCP-NEXT:    vmovdqa64 {{.*#+}} zmm10 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm7 ^ (zmm10 & (zmm9 ^ zmm7))
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vporq {{[-0-9]+}}(%r{{[sb]}}p), %zmm7, %zmm7 # 64-byte Folded Reload
+; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} zmm7 = zmm7[2,3,2,3,6,7,6,7]
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm7))
 ; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm9))
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm15))
 ; AVX512DQ-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm0, 128(%rax)
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
 ; AVX512DQ-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 64-byte Folded Reload
 ; AVX512DQ-FCP-NEXT:    # zmm0 = zmm0[2,3,2,3],mem[2,3,2,3]
-; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9, %zmm9 # 64-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    # zmm9 = zmm9[2,3,2,3],mem[2,3,2,3]
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm18[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} ymm7 = ymm19[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm17[0,1,0,1]
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm9 ^ (mem & (zmm9 ^ zmm0))
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm7, %zmm7 # 64-byte Folded Reload
+; AVX512DQ-FCP-NEXT:    # zmm7 = zmm7[2,3,2,3],mem[2,3,2,3]
+; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} ymm9 = ymm12[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} ymm12 = ymm15[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,0,1]
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm7 = zmm7 ^ (mem & (zmm7 ^ zmm0))
 ; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
 ; AVX512DQ-FCP-NEXT:    vshufi64x2 $85, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 64-byte Folded Reload
 ; AVX512DQ-FCP-NEXT:    # zmm0 = zmm0[2,3,2,3],mem[2,3,2,3]
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (zmm1 & (zmm0 ^ zmm9))
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm22 = zmm20 ^ (zmm1 & (zmm22 ^ zmm20))
-; AVX512DQ-FCP-NEXT:    vmovdqa64 {{.*#+}} zmm8 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm14 = zmm10 ^ (zmm8 & (zmm14 ^ zmm10))
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} ymm9 = ymm30[2,3,2,3]
-; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm9, %zmm1, %zmm9
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} ymm10 = ymm27[2,3,2,3]
-; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm10, %zmm28, %zmm10
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm10 = zmm9 ^ (zmm8 & (zmm10 ^ zmm9))
-; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3, %zmm3 # 32-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm7, %zmm7 # 32-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm7 = zmm7 ^ (mem & (zmm7 ^ zmm3))
-; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm2 # 32-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 | (zmm2 & mem)
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm16 = zmm16 ^ (mem & (zmm16 ^ zmm7))
-; AVX512DQ-FCP-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    # zmm2 = mem[2,3,2,3,6,7,6,7]
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm2 = zmm2 ^ (mem & (zmm2 ^ zmm0))
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} zmm0 = zmm29[0,1,0,1,4,5,4,5]
-; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm1 ^ (mem & (zmm0 ^ zmm1))
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm22))
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm11 = zmm11 ^ (mem & (zmm11 ^ zmm21))
-; AVX512DQ-FCP-NEXT:    vpermq {{.*#+}} zmm3 = zmm31[0,1,0,1,4,5,4,5]
-; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm1 ^ (mem & (zmm3 ^ zmm1))
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm3 ^ (mem & (zmm3 ^ zmm11))
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm12 = zmm25 ^ (mem & (zmm12 ^ zmm25))
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm12 = zmm12 ^ (mem & (zmm12 ^ zmm14))
-; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512DQ-FCP-NEXT:    vshufi64x2 $84, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm5 # 64-byte Folded Reload
-; AVX512DQ-FCP-NEXT:    # zmm5 = zmm1[0,1,2,3],mem[2,3,2,3]
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm4 = zmm4 | (zmm5 & mem)
-; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm4 = zmm4 ^ (mem & (zmm4 ^ zmm10))
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm4, 320(%rax)
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm12, 192(%rax)
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm3, (%rax)
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm0, 256(%rax)
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm2, 384(%rax)
-; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm16, 64(%rax)
-; AVX512DQ-FCP-NEXT:    addq $1640, %rsp # imm = 0x668
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm0 = zmm0 ^ (zmm10 & (zmm0 ^ zmm7))
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm21 = zmm24 ^ (zmm10 & (zmm21 ^ zmm24))
+; AVX512DQ-FCP-NEXT:    vmovdqa64 {{.*#+}} zmm7 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm23 = zmm30 ^ (zmm7 & (zmm23 ^ zmm30))
+; AVX512DQ-FCP-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
+; AVX512DQ-FCP-NEXT:    # ymm10 = mem[2,3,2,3]
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm13 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm10, %zmm13, %zmm10
+; AVX512DQ-FCP-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
+; AVX512DQ-FCP-NEXT:    # ymm13 = mem[2,3,2,3]
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, %ymm13, %zmm3, %zmm13
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm13 = zmm10 ^ (zmm7 & (zmm13 ^ zmm10))
+; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9, %zmm7 # 32-byte Folded Reload
+; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, (%rsp), %zmm12, %zmm9 # 32-byte Folded Reload
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm9 = zmm9 ^ (mem & (zmm9 ^ zmm7))
+; AVX512DQ-FCP-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm11, %zmm7 # 32-byte Folded Reload
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm14 = zmm14 | (zmm7 & mem)
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm14 = zmm14 ^ (mem & (zmm14 ^ zmm9))
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm3 = zmm3 ^ (mem & (zmm3 ^ zmm0))
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm4 = zmm29 ^ (mem & (zmm4 ^ zmm29))
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm4 = zmm4 ^ (mem & (zmm4 ^ zmm23))
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm5 = zmm25 ^ (mem & (zmm5 ^ zmm25))
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm5 = zmm5 ^ (mem & (zmm5 ^ zmm21))
+; AVX512DQ-FCP-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
+; AVX512DQ-FCP-NEXT:    vshufi64x2 $84, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 64-byte Folded Reload
+; AVX512DQ-FCP-NEXT:    # zmm0 = zmm0[0,1,2,3],mem[2,3,2,3]
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm1 = zmm1 | (zmm0 & mem)
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm1 = zmm1 ^ (mem & (zmm1 ^ zmm13))
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm8 = zmm8 ^ (mem & (zmm8 ^ zmm2))
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm6 = zmm16 ^ (mem & (zmm6 ^ zmm16))
+; AVX512DQ-FCP-NEXT:    vpternlogq {{.*#+}} zmm6 = zmm6 ^ (mem & (zmm6 ^ zmm8))
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm6, (%rax)
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm1, 320(%rax)
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm5, 256(%rax)
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm4, 192(%rax)
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm3, 384(%rax)
+; AVX512DQ-FCP-NEXT:    vmovdqa64 %zmm14, 64(%rax)
+; AVX512DQ-FCP-NEXT:    addq $1416, %rsp # imm = 0x588
 ; AVX512DQ-FCP-NEXT:    vzeroupper
 ; AVX512DQ-FCP-NEXT:    retq
 ;


        


More information about the llvm-commits mailing list