# RUN: llvm-mc --disassemble %s -triple=i386 | FileCheck %s --check-prefixes=ATT
# RUN: llvm-mc --disassemble %s -triple=i386 -x86-asm-syntax=intel --output-asm-variant=1 | FileCheck %s --check-prefixes=INTEL

# VNNI FP16

# ATT:   vdpphps %xmm4, %xmm3, %xmm2
# INTEL: vdpphps xmm2, xmm3, xmm4
0x62,0xf2,0x64,0x08,0x52,0xd4

# ATT:   vdpphps %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vdpphps xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x64,0x0f,0x52,0xd4

# ATT:   vdpphps %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vdpphps xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x64,0x8f,0x52,0xd4

# ATT:   vdpphps %ymm4, %ymm3, %ymm2
# INTEL: vdpphps ymm2, ymm3, ymm4
0x62,0xf2,0x64,0x28,0x52,0xd4

# ATT:   vdpphps %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vdpphps ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x64,0x2f,0x52,0xd4

# ATT:   vdpphps %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vdpphps ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x64,0xaf,0x52,0xd4

# ATT:   vdpphps %zmm4, %zmm3, %zmm2
# INTEL: vdpphps zmm2, zmm3, zmm4
0x62,0xf2,0x64,0x48,0x52,0xd4

# ATT:   vdpphps %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vdpphps zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x64,0x4f,0x52,0xd4

# ATT:   vdpphps %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vdpphps zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x64,0xcf,0x52,0xd4

# ATT:   vdpphps  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vdpphps xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x64,0x08,0x52,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vdpphps  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vdpphps xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x0f,0x52,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vdpphps  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vdpphps xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x64,0x18,0x52,0x10

# ATT:   vdpphps  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vdpphps xmm2, xmm3, xmmword ptr [2*ebp - 512]
0x62,0xf2,0x64,0x08,0x52,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vdpphps  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vdpphps xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x64,0x8f,0x52,0x51,0x7f

# ATT:   vdpphps  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vdpphps xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x64,0x9f,0x52,0x52,0x80

# ATT:   vdpphps  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vdpphps ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x64,0x28,0x52,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vdpphps  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vdpphps ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x2f,0x52,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vdpphps  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vdpphps ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x64,0x38,0x52,0x10

# ATT:   vdpphps  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vdpphps ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0x62,0xf2,0x64,0x28,0x52,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vdpphps  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vdpphps ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x64,0xaf,0x52,0x51,0x7f

# ATT:   vdpphps  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vdpphps ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x64,0xbf,0x52,0x52,0x80

# ATT:   vdpphps  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vdpphps zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x64,0x48,0x52,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vdpphps  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vdpphps zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x4f,0x52,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vdpphps  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vdpphps zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x64,0x58,0x52,0x10

# ATT:   vdpphps  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vdpphps zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x64,0x48,0x52,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vdpphps  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vdpphps zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x64,0xcf,0x52,0x51,0x7f

# ATT:   vdpphps  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vdpphps zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x64,0xdf,0x52,0x52,0x80

# VNNI INT8

# ATT:   vpdpbssd %xmm4, %xmm3, %xmm2
# INTEL: vpdpbssd xmm2, xmm3, xmm4
0xc4,0xe2,0x63,0x50,0xd4

# ATT:   vpdpbssd %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpbssd xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x67,0x0f,0x50,0xd4

# ATT:   vpdpbssd %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbssd xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x67,0x8f,0x50,0xd4

# ATT:   vpdpbssd %ymm4, %ymm3, %ymm2
# INTEL: vpdpbssd ymm2, ymm3, ymm4
0xc4,0xe2,0x67,0x50,0xd4

# ATT:   vpdpbssd %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpbssd ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x67,0x2f,0x50,0xd4

# ATT:   vpdpbssd %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbssd ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x67,0xaf,0x50,0xd4

# ATT:   vpdpbssd %zmm4, %zmm3, %zmm2
# INTEL: vpdpbssd zmm2, zmm3, zmm4
0x62,0xf2,0x67,0x48,0x50,0xd4

# ATT:   vpdpbssd %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpbssd zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x67,0x4f,0x50,0xd4

# ATT:   vpdpbssd %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbssd zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x67,0xcf,0x50,0xd4

# ATT:   vpdpbssd  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpbssd xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x63,0x50,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbssd  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpbssd xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x67,0x0f,0x50,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbssd  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpbssd xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x67,0x18,0x50,0x10

# ATT:   vpdpbssd  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpbssd xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x63,0x50,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpbssd  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbssd xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x67,0x8f,0x50,0x51,0x7f

# ATT:   vpdpbssd  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbssd xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x67,0x9f,0x50,0x52,0x80

# ATT:   vpdpbssd  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpbssd ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x67,0x50,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbssd  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpbssd ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x67,0x2f,0x50,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbssd  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpbssd ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x67,0x38,0x50,0x10

# ATT:   vpdpbssd  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpbssd ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x67,0x50,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpbssd  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbssd ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x67,0xaf,0x50,0x51,0x7f

# ATT:   vpdpbssd  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbssd ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x67,0xbf,0x50,0x52,0x80

# ATT:   vpdpbssd  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpbssd zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x67,0x48,0x50,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbssd  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpbssd zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x67,0x4f,0x50,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbssd  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpbssd zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x67,0x58,0x50,0x10

# ATT:   vpdpbssd  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpbssd zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x67,0x48,0x50,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpbssd  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbssd zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x67,0xcf,0x50,0x51,0x7f

# ATT:   vpdpbssd  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbssd zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x67,0xdf,0x50,0x52,0x80

# ATT:   vpdpbssds %xmm4, %xmm3, %xmm2
# INTEL: vpdpbssds xmm2, xmm3, xmm4
0xc4,0xe2,0x63,0x51,0xd4

# ATT:   vpdpbssds %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpbssds xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x67,0x0f,0x51,0xd4

# ATT:   vpdpbssds %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbssds xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x67,0x8f,0x51,0xd4

# ATT:   vpdpbssds %ymm4, %ymm3, %ymm2
# INTEL: vpdpbssds ymm2, ymm3, ymm4
0xc4,0xe2,0x67,0x51,0xd4

# ATT:   vpdpbssds %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpbssds ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x67,0x2f,0x51,0xd4

# ATT:   vpdpbssds %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbssds ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x67,0xaf,0x51,0xd4

# ATT:   vpdpbssds %zmm4, %zmm3, %zmm2
# INTEL: vpdpbssds zmm2, zmm3, zmm4
0x62,0xf2,0x67,0x48,0x51,0xd4

# ATT:   vpdpbssds %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpbssds zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x67,0x4f,0x51,0xd4

# ATT:   vpdpbssds %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbssds zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x67,0xcf,0x51,0xd4

# ATT:   vpdpbssds  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpbssds xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x63,0x51,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbssds  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpbssds xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x67,0x0f,0x51,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbssds  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpbssds xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x67,0x18,0x51,0x10

# ATT:   vpdpbssds  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpbssds xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x63,0x51,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpbssds  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbssds xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x67,0x8f,0x51,0x51,0x7f

# ATT:   vpdpbssds  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbssds xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x67,0x9f,0x51,0x52,0x80

# ATT:   vpdpbssds  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpbssds ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x67,0x51,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbssds  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpbssds ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x67,0x2f,0x51,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbssds  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpbssds ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x67,0x38,0x51,0x10

# ATT:   vpdpbssds  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpbssds ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x67,0x51,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpbssds  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbssds ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x67,0xaf,0x51,0x51,0x7f

# ATT:   vpdpbssds  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbssds ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x67,0xbf,0x51,0x52,0x80

# ATT:   vpdpbssds  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpbssds zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x67,0x48,0x51,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbssds  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpbssds zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x67,0x4f,0x51,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbssds  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpbssds zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x67,0x58,0x51,0x10

# ATT:   vpdpbssds  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpbssds zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x67,0x48,0x51,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpbssds  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbssds zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x67,0xcf,0x51,0x51,0x7f

# ATT:   vpdpbssds  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbssds zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x67,0xdf,0x51,0x52,0x80

# ATT:   vpdpbsud %xmm4, %xmm3, %xmm2
# INTEL: vpdpbsud xmm2, xmm3, xmm4
0xc4,0xe2,0x62,0x50,0xd4

# ATT:   vpdpbsud %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpbsud xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x66,0x0f,0x50,0xd4

# ATT:   vpdpbsud %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbsud xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x66,0x8f,0x50,0xd4

# ATT:   vpdpbsud %ymm4, %ymm3, %ymm2
# INTEL: vpdpbsud ymm2, ymm3, ymm4
0xc4,0xe2,0x66,0x50,0xd4

# ATT:   vpdpbsud %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpbsud ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x66,0x2f,0x50,0xd4

# ATT:   vpdpbsud %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbsud ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x66,0xaf,0x50,0xd4

# ATT:   vpdpbsud %zmm4, %zmm3, %zmm2
# INTEL: vpdpbsud zmm2, zmm3, zmm4
0x62,0xf2,0x66,0x48,0x50,0xd4

# ATT:   vpdpbsud %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpbsud zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x66,0x4f,0x50,0xd4

# ATT:   vpdpbsud %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbsud zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x66,0xcf,0x50,0xd4

# ATT:   vpdpbsud  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpbsud xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x62,0x50,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbsud  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpbsud xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x0f,0x50,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbsud  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpbsud xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x66,0x18,0x50,0x10

# ATT:   vpdpbsud  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpbsud xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x62,0x50,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpbsud  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbsud xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x66,0x8f,0x50,0x51,0x7f

# ATT:   vpdpbsud  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbsud xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x66,0x9f,0x50,0x52,0x80

# ATT:   vpdpbsud  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpbsud ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x66,0x50,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbsud  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpbsud ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x2f,0x50,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbsud  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpbsud ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x66,0x38,0x50,0x10

# ATT:   vpdpbsud  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpbsud ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x66,0x50,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpbsud  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbsud ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x66,0xaf,0x50,0x51,0x7f

# ATT:   vpdpbsud  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbsud ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x66,0xbf,0x50,0x52,0x80

# ATT:   vpdpbsud  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpbsud zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x66,0x48,0x50,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbsud  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpbsud zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x4f,0x50,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbsud  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpbsud zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x66,0x58,0x50,0x10

# ATT:   vpdpbsud  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpbsud zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x66,0x48,0x50,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpbsud  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbsud zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x66,0xcf,0x50,0x51,0x7f

# ATT:   vpdpbsud  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbsud zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x66,0xdf,0x50,0x52,0x80

# ATT:   vpdpbsuds %xmm4, %xmm3, %xmm2
# INTEL: vpdpbsuds xmm2, xmm3, xmm4
0xc4,0xe2,0x62,0x51,0xd4

# ATT:   vpdpbsuds %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpbsuds xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x66,0x0f,0x51,0xd4

# ATT:   vpdpbsuds %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbsuds xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x66,0x8f,0x51,0xd4

# ATT:   vpdpbsuds %ymm4, %ymm3, %ymm2
# INTEL: vpdpbsuds ymm2, ymm3, ymm4
0xc4,0xe2,0x66,0x51,0xd4

# ATT:   vpdpbsuds %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpbsuds ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x66,0x2f,0x51,0xd4

# ATT:   vpdpbsuds %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbsuds ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x66,0xaf,0x51,0xd4

# ATT:   vpdpbsuds %zmm4, %zmm3, %zmm2
# INTEL: vpdpbsuds zmm2, zmm3, zmm4
0x62,0xf2,0x66,0x48,0x51,0xd4

# ATT:   vpdpbsuds %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpbsuds zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x66,0x4f,0x51,0xd4

# ATT:   vpdpbsuds %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbsuds zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x66,0xcf,0x51,0xd4

# ATT:   vpdpbsuds  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpbsuds xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x62,0x51,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbsuds  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpbsuds xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x0f,0x51,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbsuds  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpbsuds xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x66,0x18,0x51,0x10

# ATT:   vpdpbsuds  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpbsuds xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x62,0x51,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpbsuds  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbsuds xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x66,0x8f,0x51,0x51,0x7f

# ATT:   vpdpbsuds  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbsuds xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x66,0x9f,0x51,0x52,0x80

# ATT:   vpdpbsuds  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpbsuds ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x66,0x51,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbsuds  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpbsuds ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x2f,0x51,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbsuds  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpbsuds ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x66,0x38,0x51,0x10

# ATT:   vpdpbsuds  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpbsuds ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x66,0x51,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpbsuds  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbsuds ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x66,0xaf,0x51,0x51,0x7f

# ATT:   vpdpbsuds  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbsuds ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x66,0xbf,0x51,0x52,0x80

# ATT:   vpdpbsuds  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpbsuds zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x66,0x48,0x51,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbsuds  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpbsuds zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x4f,0x51,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbsuds  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpbsuds zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x66,0x58,0x51,0x10

# ATT:   vpdpbsuds  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpbsuds zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x66,0x48,0x51,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpbsuds  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbsuds zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x66,0xcf,0x51,0x51,0x7f

# ATT:   vpdpbsuds  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbsuds zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x66,0xdf,0x51,0x52,0x80

# ATT:   vpdpbuud %xmm4, %xmm3, %xmm2
# INTEL: vpdpbuud xmm2, xmm3, xmm4
0xc4,0xe2,0x60,0x50,0xd4

# ATT:   vpdpbuud %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpbuud xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x64,0x0f,0x50,0xd4

# ATT:   vpdpbuud %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbuud xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x64,0x8f,0x50,0xd4

# ATT:   vpdpbuud %ymm4, %ymm3, %ymm2
# INTEL: vpdpbuud ymm2, ymm3, ymm4
0xc4,0xe2,0x64,0x50,0xd4

# ATT:   vpdpbuud %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpbuud ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x64,0x2f,0x50,0xd4

# ATT:   vpdpbuud %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbuud ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x64,0xaf,0x50,0xd4

# ATT:   vpdpbuud %zmm4, %zmm3, %zmm2
# INTEL: vpdpbuud zmm2, zmm3, zmm4
0x62,0xf2,0x64,0x48,0x50,0xd4

# ATT:   vpdpbuud %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpbuud zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x64,0x4f,0x50,0xd4

# ATT:   vpdpbuud %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbuud zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x64,0xcf,0x50,0xd4

# ATT:   vpdpbuud  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpbuud xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x60,0x50,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbuud  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpbuud xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x0f,0x50,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbuud  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpbuud xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x64,0x18,0x50,0x10

# ATT:   vpdpbuud  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpbuud xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x60,0x50,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpbuud  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbuud xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x64,0x8f,0x50,0x51,0x7f

# ATT:   vpdpbuud  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbuud xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x64,0x9f,0x50,0x52,0x80

# ATT:   vpdpbuud  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpbuud ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x64,0x50,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbuud  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpbuud ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x2f,0x50,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbuud  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpbuud ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x64,0x38,0x50,0x10

# ATT:   vpdpbuud  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpbuud ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x64,0x50,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpbuud  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbuud ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x64,0xaf,0x50,0x51,0x7f

# ATT:   vpdpbuud  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbuud ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x64,0xbf,0x50,0x52,0x80

# ATT:   vpdpbuud  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpbuud zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x64,0x48,0x50,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbuud  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpbuud zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x4f,0x50,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbuud  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpbuud zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x64,0x58,0x50,0x10

# ATT:   vpdpbuud  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpbuud zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x64,0x48,0x50,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpbuud  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbuud zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x64,0xcf,0x50,0x51,0x7f

# ATT:   vpdpbuud  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbuud zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x64,0xdf,0x50,0x52,0x80

# ATT:   vpdpbuuds %xmm4, %xmm3, %xmm2
# INTEL: vpdpbuuds xmm2, xmm3, xmm4
0xc4,0xe2,0x60,0x51,0xd4

# ATT:   vpdpbuuds %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpbuuds xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x64,0x0f,0x51,0xd4

# ATT:   vpdpbuuds %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbuuds xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x64,0x8f,0x51,0xd4

# ATT:   vpdpbuuds %ymm4, %ymm3, %ymm2
# INTEL: vpdpbuuds ymm2, ymm3, ymm4
0xc4,0xe2,0x64,0x51,0xd4

# ATT:   vpdpbuuds %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpbuuds ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x64,0x2f,0x51,0xd4

# ATT:   vpdpbuuds %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbuuds ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x64,0xaf,0x51,0xd4

# ATT:   vpdpbuuds %zmm4, %zmm3, %zmm2
# INTEL: vpdpbuuds zmm2, zmm3, zmm4
0x62,0xf2,0x64,0x48,0x51,0xd4

# ATT:   vpdpbuuds %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpbuuds zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x64,0x4f,0x51,0xd4

# ATT:   vpdpbuuds %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbuuds zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x64,0xcf,0x51,0xd4

# ATT:   vpdpbuuds  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpbuuds xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x60,0x51,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbuuds  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpbuuds xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x0f,0x51,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbuuds  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpbuuds xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x64,0x18,0x51,0x10

# ATT:   vpdpbuuds  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpbuuds xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x60,0x51,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpbuuds  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbuuds xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x64,0x8f,0x51,0x51,0x7f

# ATT:   vpdpbuuds  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpbuuds xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x64,0x9f,0x51,0x52,0x80

# ATT:   vpdpbuuds  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpbuuds ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x64,0x51,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbuuds  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpbuuds ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x2f,0x51,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbuuds  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpbuuds ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x64,0x38,0x51,0x10

# ATT:   vpdpbuuds  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpbuuds ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x64,0x51,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpbuuds  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbuuds ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x64,0xaf,0x51,0x51,0x7f

# ATT:   vpdpbuuds  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpbuuds ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x64,0xbf,0x51,0x52,0x80

# ATT:   vpdpbuuds  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpbuuds zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x64,0x48,0x51,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpbuuds  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpbuuds zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x4f,0x51,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpbuuds  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpbuuds zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x64,0x58,0x51,0x10

# ATT:   vpdpbuuds  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpbuuds zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x64,0x48,0x51,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpbuuds  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbuuds zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x64,0xcf,0x51,0x51,0x7f

# ATT:   vpdpbuuds  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpbuuds zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x64,0xdf,0x51,0x52,0x80

# VNNI INT16

# ATT:   vpdpwsud %xmm4, %xmm3, %xmm2
# INTEL: vpdpwsud xmm2, xmm3, xmm4
0xc4,0xe2,0x62,0xd2,0xd4

# ATT:   vpdpwsud %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpwsud xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x66,0x0f,0xd2,0xd4

# ATT:   vpdpwsud %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwsud xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x66,0x8f,0xd2,0xd4

# ATT:   vpdpwsud %ymm4, %ymm3, %ymm2
# INTEL: vpdpwsud ymm2, ymm3, ymm4
0xc4,0xe2,0x66,0xd2,0xd4

# ATT:   vpdpwsud %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpwsud ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x66,0x2f,0xd2,0xd4

# ATT:   vpdpwsud %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwsud ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x66,0xaf,0xd2,0xd4

# ATT:   vpdpwsud %zmm4, %zmm3, %zmm2
# INTEL: vpdpwsud zmm2, zmm3, zmm4
0x62,0xf2,0x66,0x48,0xd2,0xd4

# ATT:   vpdpwsud %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpwsud zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x66,0x4f,0xd2,0xd4

# ATT:   vpdpwsud %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwsud zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x66,0xcf,0xd2,0xd4

# ATT:   vpdpwsud  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpwsud xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x62,0xd2,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwsud  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpwsud xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x0f,0xd2,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwsud  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpwsud xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x66,0x18,0xd2,0x10

# ATT:   vpdpwsud  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpwsud xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x62,0xd2,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpwsud  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwsud xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x66,0x8f,0xd2,0x51,0x7f

# ATT:   vpdpwsud  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwsud xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x66,0x9f,0xd2,0x52,0x80

# ATT:   vpdpwsud  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpwsud ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x66,0xd2,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwsud  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpwsud ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x2f,0xd2,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwsud  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpwsud ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x66,0x38,0xd2,0x10

# ATT:   vpdpwsud  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpwsud ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x66,0xd2,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpwsud  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwsud ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x66,0xaf,0xd2,0x51,0x7f

# ATT:   vpdpwsud  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwsud ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x66,0xbf,0xd2,0x52,0x80

# ATT:   vpdpwsud  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpwsud zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x66,0x48,0xd2,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwsud  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpwsud zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x4f,0xd2,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwsud  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpwsud zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x66,0x58,0xd2,0x10

# ATT:   vpdpwsud  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpwsud zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x66,0x48,0xd2,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpwsud  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwsud zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x66,0xcf,0xd2,0x51,0x7f

# ATT:   vpdpwsud  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwsud zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x66,0xdf,0xd2,0x52,0x80

# ATT:   vpdpwsuds %xmm4, %xmm3, %xmm2
# INTEL: vpdpwsuds xmm2, xmm3, xmm4
0xc4,0xe2,0x62,0xd3,0xd4

# ATT:   vpdpwsuds %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpwsuds xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x66,0x0f,0xd3,0xd4

# ATT:   vpdpwsuds %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwsuds xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x66,0x8f,0xd3,0xd4

# ATT:   vpdpwsuds %ymm4, %ymm3, %ymm2
# INTEL: vpdpwsuds ymm2, ymm3, ymm4
0xc4,0xe2,0x66,0xd3,0xd4

# ATT:   vpdpwsuds %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpwsuds ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x66,0x2f,0xd3,0xd4

# ATT:   vpdpwsuds %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwsuds ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x66,0xaf,0xd3,0xd4

# ATT:   vpdpwsuds %zmm4, %zmm3, %zmm2
# INTEL: vpdpwsuds zmm2, zmm3, zmm4
0x62,0xf2,0x66,0x48,0xd3,0xd4

# ATT:   vpdpwsuds %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpwsuds zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x66,0x4f,0xd3,0xd4

# ATT:   vpdpwsuds %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwsuds zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x66,0xcf,0xd3,0xd4

# ATT:   vpdpwsuds  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpwsuds xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x62,0xd3,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwsuds  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpwsuds xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x0f,0xd3,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwsuds  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpwsuds xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x66,0x18,0xd3,0x10

# ATT:   vpdpwsuds  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpwsuds xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x62,0xd3,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpwsuds  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwsuds xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x66,0x8f,0xd3,0x51,0x7f

# ATT:   vpdpwsuds  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwsuds xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x66,0x9f,0xd3,0x52,0x80

# ATT:   vpdpwsuds  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpwsuds ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x66,0xd3,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwsuds  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpwsuds ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x2f,0xd3,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwsuds  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpwsuds ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x66,0x38,0xd3,0x10

# ATT:   vpdpwsuds  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpwsuds ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x66,0xd3,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpwsuds  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwsuds ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x66,0xaf,0xd3,0x51,0x7f

# ATT:   vpdpwsuds  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwsuds ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x66,0xbf,0xd3,0x52,0x80

# ATT:   vpdpwsuds  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpwsuds zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x66,0x48,0xd3,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwsuds  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpwsuds zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x66,0x4f,0xd3,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwsuds  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpwsuds zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x66,0x58,0xd3,0x10

# ATT:   vpdpwsuds  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpwsuds zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x66,0x48,0xd3,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpwsuds  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwsuds zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x66,0xcf,0xd3,0x51,0x7f

# ATT:   vpdpwsuds  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwsuds zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x66,0xdf,0xd3,0x52,0x80

# ATT:   vpdpwusd %xmm4, %xmm3, %xmm2
# INTEL: vpdpwusd xmm2, xmm3, xmm4
0xc4,0xe2,0x61,0xd2,0xd4

# ATT:   vpdpwusd %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpwusd xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x65,0x0f,0xd2,0xd4

# ATT:   vpdpwusd %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwusd xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x65,0x8f,0xd2,0xd4

# ATT:   vpdpwusd %ymm4, %ymm3, %ymm2
# INTEL: vpdpwusd ymm2, ymm3, ymm4
0xc4,0xe2,0x65,0xd2,0xd4

# ATT:   vpdpwusd %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpwusd ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x65,0x2f,0xd2,0xd4

# ATT:   vpdpwusd %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwusd ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x65,0xaf,0xd2,0xd4

# ATT:   vpdpwusd %zmm4, %zmm3, %zmm2
# INTEL: vpdpwusd zmm2, zmm3, zmm4
0x62,0xf2,0x65,0x48,0xd2,0xd4

# ATT:   vpdpwusd %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpwusd zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x65,0x4f,0xd2,0xd4

# ATT:   vpdpwusd %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwusd zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x65,0xcf,0xd2,0xd4

# ATT:   vpdpwusd  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpwusd xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x61,0xd2,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwusd  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpwusd xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x65,0x0f,0xd2,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwusd  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpwusd xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x65,0x18,0xd2,0x10

# ATT:   vpdpwusd  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpwusd xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x61,0xd2,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpwusd  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwusd xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x65,0x8f,0xd2,0x51,0x7f

# ATT:   vpdpwusd  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwusd xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x65,0x9f,0xd2,0x52,0x80

# ATT:   vpdpwusd  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpwusd ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x65,0xd2,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwusd  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpwusd ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x65,0x2f,0xd2,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwusd  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpwusd ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x65,0x38,0xd2,0x10

# ATT:   vpdpwusd  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpwusd ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x65,0xd2,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpwusd  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwusd ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x65,0xaf,0xd2,0x51,0x7f

# ATT:   vpdpwusd  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwusd ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x65,0xbf,0xd2,0x52,0x80

# ATT:   vpdpwusd  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpwusd zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x65,0x48,0xd2,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwusd  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpwusd zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x65,0x4f,0xd2,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwusd  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpwusd zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x65,0x58,0xd2,0x10

# ATT:   vpdpwusd  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpwusd zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x65,0x48,0xd2,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpwusd  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwusd zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x65,0xcf,0xd2,0x51,0x7f

# ATT:   vpdpwusd  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwusd zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x65,0xdf,0xd2,0x52,0x80

# ATT:   vpdpwusds %xmm4, %xmm3, %xmm2
# INTEL: vpdpwusds xmm2, xmm3, xmm4
0xc4,0xe2,0x61,0xd3,0xd4

# ATT:   vpdpwusds %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpwusds xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x65,0x0f,0xd3,0xd4

# ATT:   vpdpwusds %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwusds xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x65,0x8f,0xd3,0xd4

# ATT:   vpdpwusds %ymm4, %ymm3, %ymm2
# INTEL: vpdpwusds ymm2, ymm3, ymm4
0xc4,0xe2,0x65,0xd3,0xd4

# ATT:   vpdpwusds %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpwusds ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x65,0x2f,0xd3,0xd4

# ATT:   vpdpwusds %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwusds ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x65,0xaf,0xd3,0xd4

# ATT:   vpdpwusds %zmm4, %zmm3, %zmm2
# INTEL: vpdpwusds zmm2, zmm3, zmm4
0x62,0xf2,0x65,0x48,0xd3,0xd4

# ATT:   vpdpwusds %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpwusds zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x65,0x4f,0xd3,0xd4

# ATT:   vpdpwusds %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwusds zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x65,0xcf,0xd3,0xd4

# ATT:   vpdpwusds  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpwusds xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x61,0xd3,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwusds  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpwusds xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x65,0x0f,0xd3,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwusds  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpwusds xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x65,0x18,0xd3,0x10

# ATT:   vpdpwusds  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpwusds xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x61,0xd3,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpwusds  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwusds xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x65,0x8f,0xd3,0x51,0x7f

# ATT:   vpdpwusds  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwusds xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x65,0x9f,0xd3,0x52,0x80

# ATT:   vpdpwusds  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpwusds ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x65,0xd3,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwusds  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpwusds ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x65,0x2f,0xd3,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwusds  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpwusds ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x65,0x38,0xd3,0x10

# ATT:   vpdpwusds  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpwusds ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x65,0xd3,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpwusds  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwusds ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x65,0xaf,0xd3,0x51,0x7f

# ATT:   vpdpwusds  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwusds ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x65,0xbf,0xd3,0x52,0x80

# ATT:   vpdpwusds  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpwusds zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x65,0x48,0xd3,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwusds  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpwusds zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x65,0x4f,0xd3,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwusds  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpwusds zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x65,0x58,0xd3,0x10

# ATT:   vpdpwusds  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpwusds zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x65,0x48,0xd3,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpwusds  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwusds zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x65,0xcf,0xd3,0x51,0x7f

# ATT:   vpdpwusds  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwusds zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x65,0xdf,0xd3,0x52,0x80

# ATT:   vpdpwuud %xmm4, %xmm3, %xmm2
# INTEL: vpdpwuud xmm2, xmm3, xmm4
0xc4,0xe2,0x60,0xd2,0xd4

# ATT:   vpdpwuud %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpwuud xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x64,0x0f,0xd2,0xd4

# ATT:   vpdpwuud %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwuud xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x64,0x8f,0xd2,0xd4

# ATT:   vpdpwuud %ymm4, %ymm3, %ymm2
# INTEL: vpdpwuud ymm2, ymm3, ymm4
0xc4,0xe2,0x64,0xd2,0xd4

# ATT:   vpdpwuud %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpwuud ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x64,0x2f,0xd2,0xd4

# ATT:   vpdpwuud %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwuud ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x64,0xaf,0xd2,0xd4

# ATT:   vpdpwuud %zmm4, %zmm3, %zmm2
# INTEL: vpdpwuud zmm2, zmm3, zmm4
0x62,0xf2,0x64,0x48,0xd2,0xd4

# ATT:   vpdpwuud %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpwuud zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x64,0x4f,0xd2,0xd4

# ATT:   vpdpwuud %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwuud zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x64,0xcf,0xd2,0xd4

# ATT:   vpdpwuud  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpwuud xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x60,0xd2,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwuud  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpwuud xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x0f,0xd2,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwuud  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpwuud xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x64,0x18,0xd2,0x10

# ATT:   vpdpwuud  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpwuud xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x60,0xd2,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpwuud  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwuud xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x64,0x8f,0xd2,0x51,0x7f

# ATT:   vpdpwuud  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwuud xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x64,0x9f,0xd2,0x52,0x80

# ATT:   vpdpwuud  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpwuud ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x64,0xd2,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwuud  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpwuud ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x2f,0xd2,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwuud  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpwuud ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x64,0x38,0xd2,0x10

# ATT:   vpdpwuud  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpwuud ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x64,0xd2,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpwuud  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwuud ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x64,0xaf,0xd2,0x51,0x7f

# ATT:   vpdpwuud  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwuud ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x64,0xbf,0xd2,0x52,0x80

# ATT:   vpdpwuud  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpwuud zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x64,0x48,0xd2,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwuud  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpwuud zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x4f,0xd2,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwuud  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpwuud zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x64,0x58,0xd2,0x10

# ATT:   vpdpwuud  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpwuud zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x64,0x48,0xd2,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpwuud  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwuud zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x64,0xcf,0xd2,0x51,0x7f

# ATT:   vpdpwuud  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwuud zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x64,0xdf,0xd2,0x52,0x80

# ATT:   vpdpwuuds %xmm4, %xmm3, %xmm2
# INTEL: vpdpwuuds xmm2, xmm3, xmm4
0xc4,0xe2,0x60,0xd3,0xd4

# ATT:   vpdpwuuds %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vpdpwuuds xmm2 {k7}, xmm3, xmm4
0x62,0xf2,0x64,0x0f,0xd3,0xd4

# ATT:   vpdpwuuds %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwuuds xmm2 {k7} {z}, xmm3, xmm4
0x62,0xf2,0x64,0x8f,0xd3,0xd4

# ATT:   vpdpwuuds %ymm4, %ymm3, %ymm2
# INTEL: vpdpwuuds ymm2, ymm3, ymm4
0xc4,0xe2,0x64,0xd3,0xd4

# ATT:   vpdpwuuds %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vpdpwuuds ymm2 {k7}, ymm3, ymm4
0x62,0xf2,0x64,0x2f,0xd3,0xd4

# ATT:   vpdpwuuds %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwuuds ymm2 {k7} {z}, ymm3, ymm4
0x62,0xf2,0x64,0xaf,0xd3,0xd4

# ATT:   vpdpwuuds %zmm4, %zmm3, %zmm2
# INTEL: vpdpwuuds zmm2, zmm3, zmm4
0x62,0xf2,0x64,0x48,0xd3,0xd4

# ATT:   vpdpwuuds %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vpdpwuuds zmm2 {k7}, zmm3, zmm4
0x62,0xf2,0x64,0x4f,0xd3,0xd4

# ATT:   vpdpwuuds %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwuuds zmm2 {k7} {z}, zmm3, zmm4
0x62,0xf2,0x64,0xcf,0xd3,0xd4

# ATT:   vpdpwuuds  268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vpdpwuuds xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x60,0xd3,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwuuds  291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vpdpwuuds xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x0f,0xd3,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwuuds  (%eax){1to4}, %xmm3, %xmm2
# INTEL: vpdpwuuds xmm2, xmm3, dword ptr [eax]{1to4}
0x62,0xf2,0x64,0x18,0xd3,0x10

# ATT:   vpdpwuuds  -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vpdpwuuds xmm2, xmm3, xmmword ptr [2*ebp - 512]
0xc4,0xe2,0x60,0xd3,0x14,0x6d,0x00,0xfe,0xff,0xff

# ATT:   vpdpwuuds  2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwuuds xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032]
0x62,0xf2,0x64,0x8f,0xd3,0x51,0x7f

# ATT:   vpdpwuuds  -512(%edx){1to4}, %xmm3, %xmm2 {%k7} {z}
# INTEL: vpdpwuuds xmm2 {k7} {z}, xmm3, dword ptr [edx - 512]{1to4}
0x62,0xf2,0x64,0x9f,0xd3,0x52,0x80

# ATT:   vpdpwuuds  268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vpdpwuuds ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456]
0xc4,0xe2,0x64,0xd3,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwuuds  291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vpdpwuuds ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x2f,0xd3,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwuuds  (%eax){1to8}, %ymm3, %ymm2
# INTEL: vpdpwuuds ymm2, ymm3, dword ptr [eax]{1to8}
0x62,0xf2,0x64,0x38,0xd3,0x10

# ATT:   vpdpwuuds  -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vpdpwuuds ymm2, ymm3, ymmword ptr [2*ebp - 1024]
0xc4,0xe2,0x64,0xd3,0x14,0x6d,0x00,0xfc,0xff,0xff

# ATT:   vpdpwuuds  4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwuuds ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064]
0x62,0xf2,0x64,0xaf,0xd3,0x51,0x7f

# ATT:   vpdpwuuds  -512(%edx){1to8}, %ymm3, %ymm2 {%k7} {z}
# INTEL: vpdpwuuds ymm2 {k7} {z}, ymm3, dword ptr [edx - 512]{1to8}
0x62,0xf2,0x64,0xbf,0xd3,0x52,0x80

# ATT:   vpdpwuuds  268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vpdpwuuds zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456]
0x62,0xf2,0x64,0x48,0xd3,0x94,0xf4,0x00,0x00,0x00,0x10

# ATT:   vpdpwuuds  291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vpdpwuuds zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291]
0x62,0xf2,0x64,0x4f,0xd3,0x94,0x87,0x23,0x01,0x00,0x00

# ATT:   vpdpwuuds  (%eax){1to16}, %zmm3, %zmm2
# INTEL: vpdpwuuds zmm2, zmm3, dword ptr [eax]{1to16}
0x62,0xf2,0x64,0x58,0xd3,0x10

# ATT:   vpdpwuuds  -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vpdpwuuds zmm2, zmm3, zmmword ptr [2*ebp - 2048]
0x62,0xf2,0x64,0x48,0xd3,0x14,0x6d,0x00,0xf8,0xff,0xff

# ATT:   vpdpwuuds  8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwuuds zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128]
0x62,0xf2,0x64,0xcf,0xd3,0x51,0x7f

# ATT:   vpdpwuuds  -512(%edx){1to16}, %zmm3, %zmm2 {%k7} {z}
# INTEL: vpdpwuuds zmm2 {k7} {z}, zmm3, dword ptr [edx - 512]{1to16}
0x62,0xf2,0x64,0xdf,0xd3,0x52,0x80

# VMPSADBW

# ATT:   vmpsadbw $123, %xmm4, %xmm3, %xmm2
# INTEL: vmpsadbw xmm2, xmm3, xmm4, 123
0xc4,0xe3,0x61,0x42,0xd4,0x7b

# ATT:   vmpsadbw $123, %xmm4, %xmm3, %xmm2 {%k7}
# INTEL: vmpsadbw xmm2 {k7}, xmm3, xmm4, 123
0x62,0xf3,0x66,0x0f,0x42,0xd4,0x7b

# ATT:   vmpsadbw $123, %xmm4, %xmm3, %xmm2 {%k7} {z}
# INTEL: vmpsadbw xmm2 {k7} {z}, xmm3, xmm4, 123
0x62,0xf3,0x66,0x8f,0x42,0xd4,0x7b

# ATT:   vmpsadbw $123, %ymm4, %ymm3, %ymm2
# INTEL: vmpsadbw ymm2, ymm3, ymm4, 123
0xc4,0xe3,0x65,0x42,0xd4,0x7b

# ATT:   vmpsadbw $123, %ymm4, %ymm3, %ymm2 {%k7}
# INTEL: vmpsadbw ymm2 {k7}, ymm3, ymm4, 123
0x62,0xf3,0x66,0x2f,0x42,0xd4,0x7b

# ATT:   vmpsadbw $123, %ymm4, %ymm3, %ymm2 {%k7} {z}
# INTEL: vmpsadbw ymm2 {k7} {z}, ymm3, ymm4, 123
0x62,0xf3,0x66,0xaf,0x42,0xd4,0x7b

# ATT:   vmpsadbw $123, %zmm4, %zmm3, %zmm2
# INTEL: vmpsadbw zmm2, zmm3, zmm4, 123
0x62,0xf3,0x66,0x48,0x42,0xd4,0x7b

# ATT:   vmpsadbw $123, %zmm4, %zmm3, %zmm2 {%k7}
# INTEL: vmpsadbw zmm2 {k7}, zmm3, zmm4, 123
0x62,0xf3,0x66,0x4f,0x42,0xd4,0x7b

# ATT:   vmpsadbw $123, %zmm4, %zmm3, %zmm2 {%k7} {z}
# INTEL: vmpsadbw zmm2 {k7} {z}, zmm3, zmm4, 123
0x62,0xf3,0x66,0xcf,0x42,0xd4,0x7b

# ATT:   vmpsadbw  $123, 268435456(%esp,%esi,8), %xmm3, %xmm2
# INTEL: vmpsadbw xmm2, xmm3, xmmword ptr [esp + 8*esi + 268435456], 123
0xc4,0xe3,0x61,0x42,0x94,0xf4,0x00,0x00,0x00,0x10,0x7b

# ATT:   vmpsadbw  $123, 291(%edi,%eax,4), %xmm3, %xmm2 {%k7}
# INTEL: vmpsadbw xmm2 {k7}, xmm3, xmmword ptr [edi + 4*eax + 291], 123
0x62,0xf3,0x66,0x0f,0x42,0x94,0x87,0x23,0x01,0x00,0x00,0x7b

# ATT:   vmpsadbw  $123, (%eax), %xmm3, %xmm2
# INTEL: vmpsadbw xmm2, xmm3, xmmword ptr [eax], 123
0xc4,0xe3,0x61,0x42,0x10,0x7b

# ATT:   vmpsadbw  $123, -512(,%ebp,2), %xmm3, %xmm2
# INTEL: vmpsadbw xmm2, xmm3, xmmword ptr [2*ebp - 512], 123
0xc4,0xe3,0x61,0x42,0x14,0x6d,0x00,0xfe,0xff,0xff,0x7b

# ATT:   vmpsadbw  $123, 2032(%ecx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vmpsadbw xmm2 {k7} {z}, xmm3, xmmword ptr [ecx + 2032], 123
0x62,0xf3,0x66,0x8f,0x42,0x51,0x7f,0x7b

# ATT:   vmpsadbw  $123, -2048(%edx), %xmm3, %xmm2 {%k7} {z}
# INTEL: vmpsadbw xmm2 {k7} {z}, xmm3, xmmword ptr [edx - 2048], 123
0x62,0xf3,0x66,0x8f,0x42,0x52,0x80,0x7b

# ATT:   vmpsadbw  $123, 268435456(%esp,%esi,8), %ymm3, %ymm2
# INTEL: vmpsadbw ymm2, ymm3, ymmword ptr [esp + 8*esi + 268435456], 123
0xc4,0xe3,0x65,0x42,0x94,0xf4,0x00,0x00,0x00,0x10,0x7b

# ATT:   vmpsadbw  $123, 291(%edi,%eax,4), %ymm3, %ymm2 {%k7}
# INTEL: vmpsadbw ymm2 {k7}, ymm3, ymmword ptr [edi + 4*eax + 291], 123
0x62,0xf3,0x66,0x2f,0x42,0x94,0x87,0x23,0x01,0x00,0x00,0x7b

# ATT:   vmpsadbw  $123, (%eax), %ymm3, %ymm2
# INTEL: vmpsadbw ymm2, ymm3, ymmword ptr [eax], 123
0xc4,0xe3,0x65,0x42,0x10,0x7b

# ATT:   vmpsadbw  $123, -1024(,%ebp,2), %ymm3, %ymm2
# INTEL: vmpsadbw ymm2, ymm3, ymmword ptr [2*ebp - 1024], 123
0xc4,0xe3,0x65,0x42,0x14,0x6d,0x00,0xfc,0xff,0xff,0x7b

# ATT:   vmpsadbw  $123, 4064(%ecx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vmpsadbw ymm2 {k7} {z}, ymm3, ymmword ptr [ecx + 4064], 123
0x62,0xf3,0x66,0xaf,0x42,0x51,0x7f,0x7b

# ATT:   vmpsadbw  $123, -4096(%edx), %ymm3, %ymm2 {%k7} {z}
# INTEL: vmpsadbw ymm2 {k7} {z}, ymm3, ymmword ptr [edx - 4096], 123
0x62,0xf3,0x66,0xaf,0x42,0x52,0x80,0x7b

# ATT:   vmpsadbw  $123, 268435456(%esp,%esi,8), %zmm3, %zmm2
# INTEL: vmpsadbw zmm2, zmm3, zmmword ptr [esp + 8*esi + 268435456], 123
0x62,0xf3,0x66,0x48,0x42,0x94,0xf4,0x00,0x00,0x00,0x10,0x7b

# ATT:   vmpsadbw  $123, 291(%edi,%eax,4), %zmm3, %zmm2 {%k7}
# INTEL: vmpsadbw zmm2 {k7}, zmm3, zmmword ptr [edi + 4*eax + 291], 123
0x62,0xf3,0x66,0x4f,0x42,0x94,0x87,0x23,0x01,0x00,0x00,0x7b

# ATT:   vmpsadbw  $123, (%eax), %zmm3, %zmm2
# INTEL: vmpsadbw zmm2, zmm3, zmmword ptr [eax], 123
0x62,0xf3,0x66,0x48,0x42,0x10,0x7b

# ATT:   vmpsadbw  $123, -2048(,%ebp,2), %zmm3, %zmm2
# INTEL: vmpsadbw zmm2, zmm3, zmmword ptr [2*ebp - 2048], 123
0x62,0xf3,0x66,0x48,0x42,0x14,0x6d,0x00,0xf8,0xff,0xff,0x7b

# ATT:   vmpsadbw  $123, 8128(%ecx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vmpsadbw zmm2 {k7} {z}, zmm3, zmmword ptr [ecx + 8128], 123
0x62,0xf3,0x66,0xcf,0x42,0x51,0x7f,0x7b

# ATT:   vmpsadbw  $123, -8192(%edx), %zmm3, %zmm2 {%k7} {z}
# INTEL: vmpsadbw zmm2 {k7} {z}, zmm3, zmmword ptr [edx - 8192], 123
0x62,0xf3,0x66,0xcf,0x42,0x52,0x80,0x7b