Home | History | Annotate | Download | only in X86
      1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
      2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512vnni,+avx512vl --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86
      3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vnni,+avx512vl --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64
      4 
      5 declare <8 x i32> @llvm.x86.avx512.mask.vpdpbusd.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
      6 declare <8 x i32> @llvm.x86.avx512.maskz.vpdpbusd.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
      7 
      8 define <8 x i32>@test_int_x86_avx512_mask_vpdpbusd_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32>* %x2p, <8 x i32> %x4, i8 %x3) {
      9 ; X86-LABEL: test_int_x86_avx512_mask_vpdpbusd_256:
     10 ; X86:       # %bb.0:
     11 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
     12 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
     13 ; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
     14 ; X86-NEXT:    vmovaps %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xd8]
     15 ; X86-NEXT:    vpdpbusd (%eax), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x50,0x18]
     16 ; X86-NEXT:    vmovaps %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xe0]
     17 ; X86-NEXT:    vpdpbusd %ymm2, %ymm1, %ymm4 # encoding: [0x62,0xf2,0x75,0x28,0x50,0xe2]
     18 ; X86-NEXT:    vpdpbusd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x50,0xc2]
     19 ; X86-NEXT:    vpaddd %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xfe,0xc0]
     20 ; X86-NEXT:    vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
     21 ; X86-NEXT:    retl # encoding: [0xc3]
     22 ;
     23 ; X64-LABEL: test_int_x86_avx512_mask_vpdpbusd_256:
     24 ; X64:       # %bb.0:
     25 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
     26 ; X64-NEXT:    vmovaps %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xd8]
     27 ; X64-NEXT:    vpdpbusd (%rdi), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x50,0x1f]
     28 ; X64-NEXT:    vmovaps %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xe0]
     29 ; X64-NEXT:    vpdpbusd %ymm2, %ymm1, %ymm4 # encoding: [0x62,0xf2,0x75,0x28,0x50,0xe2]
     30 ; X64-NEXT:    vpdpbusd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x50,0xc2]
     31 ; X64-NEXT:    vpaddd %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xfe,0xc0]
     32 ; X64-NEXT:    vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
     33 ; X64-NEXT:    retq # encoding: [0xc3]
     34   %x2 = load <8 x i32>, <8 x i32>* %x2p
     35   %res = call <8 x i32> @llvm.x86.avx512.mask.vpdpbusd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
     36   %res1 = call <8 x i32> @llvm.x86.avx512.mask.vpdpbusd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4, i8 -1)
     37   %res2 = call <8 x i32> @llvm.x86.avx512.maskz.vpdpbusd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4, i8  %x3)
     38   %res3 = add <8 x i32> %res, %res1
     39   %res4 = add <8 x i32> %res2, %res3
     40   ret <8 x i32> %res4
     41 }
     42 
     43 declare <4 x i32> @llvm.x86.avx512.mask.vpdpbusd.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
     44 declare <4 x i32> @llvm.x86.avx512.maskz.vpdpbusd.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
     45 
     46 define <4 x i32>@test_int_x86_avx512_mask_vpdpbusd_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32>* %x2p, <4 x i32> %x4, i8 %x3) {
     47 ; X86-LABEL: test_int_x86_avx512_mask_vpdpbusd_128:
     48 ; X86:       # %bb.0:
     49 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
     50 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
     51 ; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
     52 ; X86-NEXT:    vmovaps %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xd8]
     53 ; X86-NEXT:    vpdpbusd (%eax), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x50,0x18]
     54 ; X86-NEXT:    vmovaps %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xe0]
     55 ; X86-NEXT:    vpdpbusd %xmm2, %xmm1, %xmm4 # encoding: [0x62,0xf2,0x75,0x08,0x50,0xe2]
     56 ; X86-NEXT:    vpdpbusd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x50,0xc2]
     57 ; X86-NEXT:    vpaddd %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xfe,0xc0]
     58 ; X86-NEXT:    vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
     59 ; X86-NEXT:    retl # encoding: [0xc3]
     60 ;
     61 ; X64-LABEL: test_int_x86_avx512_mask_vpdpbusd_128:
     62 ; X64:       # %bb.0:
     63 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
     64 ; X64-NEXT:    vmovaps %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xd8]
     65 ; X64-NEXT:    vpdpbusd (%rdi), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x50,0x1f]
     66 ; X64-NEXT:    vmovaps %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xe0]
     67 ; X64-NEXT:    vpdpbusd %xmm2, %xmm1, %xmm4 # encoding: [0x62,0xf2,0x75,0x08,0x50,0xe2]
     68 ; X64-NEXT:    vpdpbusd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x50,0xc2]
     69 ; X64-NEXT:    vpaddd %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xfe,0xc0]
     70 ; X64-NEXT:    vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
     71 ; X64-NEXT:    retq # encoding: [0xc3]
     72   %x2 = load <4 x i32>, <4 x i32>* %x2p
     73   %res = call <4 x i32> @llvm.x86.avx512.mask.vpdpbusd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3)
     74   %res1 = call <4 x i32> @llvm.x86.avx512.mask.vpdpbusd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4, i8 -1)
     75   %res2 = call <4 x i32> @llvm.x86.avx512.maskz.vpdpbusd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4, i8  %x3)
     76   %res3 = add <4 x i32> %res, %res1
     77   %res4 = add <4 x i32> %res2, %res3
     78   ret <4 x i32> %res4
     79 }
     80 
     81 declare <8 x i32> @llvm.x86.avx512.mask.vpdpbusds.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
     82 declare <8 x i32> @llvm.x86.avx512.maskz.vpdpbusds.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
     83 
     84 define <8 x i32>@test_int_x86_avx512_mask_vpdpbusds_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32>* %x2p, <8 x i32> %x4, i8 %x3) {
     85 ; X86-LABEL: test_int_x86_avx512_mask_vpdpbusds_256:
     86 ; X86:       # %bb.0:
     87 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
     88 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
     89 ; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
     90 ; X86-NEXT:    vmovaps %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xd8]
     91 ; X86-NEXT:    vpdpbusds (%eax), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x51,0x18]
     92 ; X86-NEXT:    vmovaps %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xe0]
     93 ; X86-NEXT:    vpdpbusds %ymm2, %ymm1, %ymm4 # encoding: [0x62,0xf2,0x75,0x28,0x51,0xe2]
     94 ; X86-NEXT:    vpdpbusds %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x51,0xc2]
     95 ; X86-NEXT:    vpaddd %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xfe,0xc0]
     96 ; X86-NEXT:    vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
     97 ; X86-NEXT:    retl # encoding: [0xc3]
     98 ;
     99 ; X64-LABEL: test_int_x86_avx512_mask_vpdpbusds_256:
    100 ; X64:       # %bb.0:
    101 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
    102 ; X64-NEXT:    vmovaps %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xd8]
    103 ; X64-NEXT:    vpdpbusds (%rdi), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x51,0x1f]
    104 ; X64-NEXT:    vmovaps %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xe0]
    105 ; X64-NEXT:    vpdpbusds %ymm2, %ymm1, %ymm4 # encoding: [0x62,0xf2,0x75,0x28,0x51,0xe2]
    106 ; X64-NEXT:    vpdpbusds %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x51,0xc2]
    107 ; X64-NEXT:    vpaddd %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xfe,0xc0]
    108 ; X64-NEXT:    vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
    109 ; X64-NEXT:    retq # encoding: [0xc3]
    110   %x2 = load <8 x i32>, <8 x i32>* %x2p
    111   %res = call <8 x i32> @llvm.x86.avx512.mask.vpdpbusds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
    112   %res1 = call <8 x i32> @llvm.x86.avx512.mask.vpdpbusds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4, i8 -1)
    113   %res2 = call <8 x i32> @llvm.x86.avx512.maskz.vpdpbusds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4, i8  %x3)
    114   %res3 = add <8 x i32> %res, %res1
    115   %res4 = add <8 x i32> %res2, %res3
    116   ret <8 x i32> %res4
    117 }
    118 
    119 declare <4 x i32> @llvm.x86.avx512.mask.vpdpbusds.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
    120 declare <4 x i32> @llvm.x86.avx512.maskz.vpdpbusds.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
    121 
    122 define <4 x i32>@test_int_x86_avx512_mask_vpdpbusds_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32>* %x2p, <4 x i32> %x4, i8 %x3) {
    123 ; X86-LABEL: test_int_x86_avx512_mask_vpdpbusds_128:
    124 ; X86:       # %bb.0:
    125 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
    126 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
    127 ; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
    128 ; X86-NEXT:    vmovaps %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xd8]
    129 ; X86-NEXT:    vpdpbusds (%eax), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x51,0x18]
    130 ; X86-NEXT:    vmovaps %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xe0]
    131 ; X86-NEXT:    vpdpbusds %xmm2, %xmm1, %xmm4 # encoding: [0x62,0xf2,0x75,0x08,0x51,0xe2]
    132 ; X86-NEXT:    vpdpbusds %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x51,0xc2]
    133 ; X86-NEXT:    vpaddd %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xfe,0xc0]
    134 ; X86-NEXT:    vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
    135 ; X86-NEXT:    retl # encoding: [0xc3]
    136 ;
    137 ; X64-LABEL: test_int_x86_avx512_mask_vpdpbusds_128:
    138 ; X64:       # %bb.0:
    139 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
    140 ; X64-NEXT:    vmovaps %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xd8]
    141 ; X64-NEXT:    vpdpbusds (%rdi), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x51,0x1f]
    142 ; X64-NEXT:    vmovaps %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xe0]
    143 ; X64-NEXT:    vpdpbusds %xmm2, %xmm1, %xmm4 # encoding: [0x62,0xf2,0x75,0x08,0x51,0xe2]
    144 ; X64-NEXT:    vpdpbusds %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x51,0xc2]
    145 ; X64-NEXT:    vpaddd %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xfe,0xc0]
    146 ; X64-NEXT:    vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
    147 ; X64-NEXT:    retq # encoding: [0xc3]
    148   %x2 = load <4 x i32>, <4 x i32>* %x2p
    149   %res = call <4 x i32> @llvm.x86.avx512.mask.vpdpbusds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3)
    150   %res1 = call <4 x i32> @llvm.x86.avx512.mask.vpdpbusds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4, i8 -1)
    151   %res2 = call <4 x i32> @llvm.x86.avx512.maskz.vpdpbusds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4, i8  %x3)
    152   %res3 = add <4 x i32> %res, %res1
    153   %res4 = add <4 x i32> %res2, %res3
    154   ret <4 x i32> %res4
    155 }
    156 
    157 declare <8 x i32> @llvm.x86.avx512.mask.vpdpwssd.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
    158 declare <8 x i32> @llvm.x86.avx512.maskz.vpdpwssd.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
    159 
    160 define <8 x i32>@test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32>* %x2p, <8 x i32> %x4, i8 %x3) {
    161 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssd_256:
    162 ; X86:       # %bb.0:
    163 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
    164 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
    165 ; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
    166 ; X86-NEXT:    vmovaps %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xd8]
    167 ; X86-NEXT:    vpdpwssd (%eax), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x52,0x18]
    168 ; X86-NEXT:    vmovaps %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xe0]
    169 ; X86-NEXT:    vpdpwssd %ymm2, %ymm1, %ymm4 # encoding: [0x62,0xf2,0x75,0x28,0x52,0xe2]
    170 ; X86-NEXT:    vpdpwssd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x52,0xc2]
    171 ; X86-NEXT:    vpaddd %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xfe,0xc0]
    172 ; X86-NEXT:    vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
    173 ; X86-NEXT:    retl # encoding: [0xc3]
    174 ;
    175 ; X64-LABEL: test_int_x86_avx512_mask_vpdpwssd_256:
    176 ; X64:       # %bb.0:
    177 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
    178 ; X64-NEXT:    vmovaps %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xd8]
    179 ; X64-NEXT:    vpdpwssd (%rdi), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x52,0x1f]
    180 ; X64-NEXT:    vmovaps %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xe0]
    181 ; X64-NEXT:    vpdpwssd %ymm2, %ymm1, %ymm4 # encoding: [0x62,0xf2,0x75,0x28,0x52,0xe2]
    182 ; X64-NEXT:    vpdpwssd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x52,0xc2]
    183 ; X64-NEXT:    vpaddd %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xfe,0xc0]
    184 ; X64-NEXT:    vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
    185 ; X64-NEXT:    retq # encoding: [0xc3]
    186   %x2 = load <8 x i32>, <8 x i32>* %x2p
    187   %res = call <8 x i32> @llvm.x86.avx512.mask.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
    188   %res1 = call <8 x i32> @llvm.x86.avx512.mask.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4, i8 -1)
    189   %res2 = call <8 x i32> @llvm.x86.avx512.maskz.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4, i8  %x3)
    190   %res3 = add <8 x i32> %res, %res1
    191   %res4 = add <8 x i32> %res2, %res3
    192   ret <8 x i32> %res4
    193 }
    194 
    195 declare <4 x i32> @llvm.x86.avx512.mask.vpdpwssd.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
    196 declare <4 x i32> @llvm.x86.avx512.maskz.vpdpwssd.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
    197 
    198 define <4 x i32>@test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32>* %x2p, <4 x i32> %x4, i8 %x3) {
    199 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssd_128:
    200 ; X86:       # %bb.0:
    201 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
    202 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
    203 ; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
    204 ; X86-NEXT:    vmovaps %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xd8]
    205 ; X86-NEXT:    vpdpwssd (%eax), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x52,0x18]
    206 ; X86-NEXT:    vmovaps %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xe0]
    207 ; X86-NEXT:    vpdpwssd %xmm2, %xmm1, %xmm4 # encoding: [0x62,0xf2,0x75,0x08,0x52,0xe2]
    208 ; X86-NEXT:    vpdpwssd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x52,0xc2]
    209 ; X86-NEXT:    vpaddd %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xfe,0xc0]
    210 ; X86-NEXT:    vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
    211 ; X86-NEXT:    retl # encoding: [0xc3]
    212 ;
    213 ; X64-LABEL: test_int_x86_avx512_mask_vpdpwssd_128:
    214 ; X64:       # %bb.0:
    215 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
    216 ; X64-NEXT:    vmovaps %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xd8]
    217 ; X64-NEXT:    vpdpwssd (%rdi), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x52,0x1f]
    218 ; X64-NEXT:    vmovaps %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xe0]
    219 ; X64-NEXT:    vpdpwssd %xmm2, %xmm1, %xmm4 # encoding: [0x62,0xf2,0x75,0x08,0x52,0xe2]
    220 ; X64-NEXT:    vpdpwssd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x52,0xc2]
    221 ; X64-NEXT:    vpaddd %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xfe,0xc0]
    222 ; X64-NEXT:    vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
    223 ; X64-NEXT:    retq # encoding: [0xc3]
    224   %x2 = load <4 x i32>, <4 x i32>* %x2p
    225   %res = call <4 x i32> @llvm.x86.avx512.mask.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3)
    226   %res1 = call <4 x i32> @llvm.x86.avx512.mask.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4, i8 -1)
    227   %res2 = call <4 x i32> @llvm.x86.avx512.maskz.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4, i8  %x3)
    228   %res3 = add <4 x i32> %res, %res1
    229   %res4 = add <4 x i32> %res2, %res3
    230   ret <4 x i32> %res4
    231 }
    232 
    233 
    234 declare <8 x i32> @llvm.x86.avx512.mask.vpdpwssds.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
    235 declare <8 x i32> @llvm.x86.avx512.maskz.vpdpwssds.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
    236 
    237 define <8 x i32>@test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32>* %x2p, <8 x i32> %x4, i8 %x3) {
    238 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssds_256:
    239 ; X86:       # %bb.0:
    240 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
    241 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
    242 ; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
    243 ; X86-NEXT:    vmovaps %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xd8]
    244 ; X86-NEXT:    vpdpwssds (%eax), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x53,0x18]
    245 ; X86-NEXT:    vmovaps %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xe0]
    246 ; X86-NEXT:    vpdpwssds %ymm2, %ymm1, %ymm4 # encoding: [0x62,0xf2,0x75,0x28,0x53,0xe2]
    247 ; X86-NEXT:    vpdpwssds %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x53,0xc2]
    248 ; X86-NEXT:    vpaddd %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xfe,0xc0]
    249 ; X86-NEXT:    vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
    250 ; X86-NEXT:    retl # encoding: [0xc3]
    251 ;
    252 ; X64-LABEL: test_int_x86_avx512_mask_vpdpwssds_256:
    253 ; X64:       # %bb.0:
    254 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
    255 ; X64-NEXT:    vmovaps %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xd8]
    256 ; X64-NEXT:    vpdpwssds (%rdi), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x53,0x1f]
    257 ; X64-NEXT:    vmovaps %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xe0]
    258 ; X64-NEXT:    vpdpwssds %ymm2, %ymm1, %ymm4 # encoding: [0x62,0xf2,0x75,0x28,0x53,0xe2]
    259 ; X64-NEXT:    vpdpwssds %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x53,0xc2]
    260 ; X64-NEXT:    vpaddd %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xfe,0xc0]
    261 ; X64-NEXT:    vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
    262 ; X64-NEXT:    retq # encoding: [0xc3]
    263   %x2 = load <8 x i32>, <8 x i32>* %x2p
    264   %res = call <8 x i32> @llvm.x86.avx512.mask.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
    265   %res1 = call <8 x i32> @llvm.x86.avx512.mask.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4, i8 -1)
    266   %res2 = call <8 x i32> @llvm.x86.avx512.maskz.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4, i8  %x3)
    267   %res3 = add <8 x i32> %res, %res1
    268   %res4 = add <8 x i32> %res2, %res3
    269   ret <8 x i32> %res4
    270 }
    271 
    272 declare <4 x i32> @llvm.x86.avx512.mask.vpdpwssds.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
    273 declare <4 x i32> @llvm.x86.avx512.maskz.vpdpwssds.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
    274 
    275 define <4 x i32>@test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32>* %x2p, <4 x i32> %x4, i8 %x3) {
    276 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssds_128:
    277 ; X86:       # %bb.0:
    278 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
    279 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
    280 ; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
    281 ; X86-NEXT:    vmovaps %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xd8]
    282 ; X86-NEXT:    vpdpwssds (%eax), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x53,0x18]
    283 ; X86-NEXT:    vmovaps %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xe0]
    284 ; X86-NEXT:    vpdpwssds %xmm2, %xmm1, %xmm4 # encoding: [0x62,0xf2,0x75,0x08,0x53,0xe2]
    285 ; X86-NEXT:    vpdpwssds %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x53,0xc2]
    286 ; X86-NEXT:    vpaddd %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xfe,0xc0]
    287 ; X86-NEXT:    vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
    288 ; X86-NEXT:    retl # encoding: [0xc3]
    289 ;
    290 ; X64-LABEL: test_int_x86_avx512_mask_vpdpwssds_128:
    291 ; X64:       # %bb.0:
    292 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
    293 ; X64-NEXT:    vmovaps %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xd8]
    294 ; X64-NEXT:    vpdpwssds (%rdi), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x53,0x1f]
    295 ; X64-NEXT:    vmovaps %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xe0]
    296 ; X64-NEXT:    vpdpwssds %xmm2, %xmm1, %xmm4 # encoding: [0x62,0xf2,0x75,0x08,0x53,0xe2]
    297 ; X64-NEXT:    vpdpwssds %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x53,0xc2]
    298 ; X64-NEXT:    vpaddd %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xfe,0xc0]
    299 ; X64-NEXT:    vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
    300 ; X64-NEXT:    retq # encoding: [0xc3]
    301   %x2 = load <4 x i32>, <4 x i32>* %x2p
    302   %res = call <4 x i32> @llvm.x86.avx512.mask.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3)
    303   %res1 = call <4 x i32> @llvm.x86.avx512.mask.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4, i8 -1)
    304   %res2 = call <4 x i32> @llvm.x86.avx512.maskz.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4, i8  %x3)
    305   %res3 = add <4 x i32> %res, %res1
    306   %res4 = add <4 x i32> %res2, %res3
    307   ret <4 x i32> %res4
    308 }
    309 
    310