1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512vnni --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86 3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vnni --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64 4 5 declare <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32>, <16 x i32>, <16 x i32>) 6 7 define <16 x i32>@test_int_x86_avx512_mask_vpdpbusd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, <16 x i32> %x4, i16 %x3) { 8 ; X86-LABEL: test_int_x86_avx512_mask_vpdpbusd_512: 9 ; X86: # %bb.0: 10 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] 11 ; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08] 12 ; X86-NEXT: vmovaps %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xd8] 13 ; X86-NEXT: vpdpbusd (%eax), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x50,0x18] 14 ; X86-NEXT: vmovaps %zmm0, %zmm4 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xe0] 15 ; X86-NEXT: vpdpbusd %zmm2, %zmm1, %zmm4 # encoding: [0x62,0xf2,0x75,0x48,0x50,0xe2] 16 ; X86-NEXT: vpdpbusd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x50,0xc2] 17 ; X86-NEXT: vpaddd %zmm0, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfe,0xc0] 18 ; X86-NEXT: vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0] 19 ; X86-NEXT: retl # encoding: [0xc3] 20 ; 21 ; X64-LABEL: test_int_x86_avx512_mask_vpdpbusd_512: 22 ; X64: # %bb.0: 23 ; X64-NEXT: kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce] 24 ; X64-NEXT: vmovaps %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xd8] 25 ; X64-NEXT: vpdpbusd (%rdi), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x50,0x1f] 26 ; X64-NEXT: vmovaps %zmm0, %zmm4 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xe0] 27 ; X64-NEXT: vpdpbusd %zmm2, %zmm1, %zmm4 # encoding: [0x62,0xf2,0x75,0x48,0x50,0xe2] 28 ; X64-NEXT: vpdpbusd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x50,0xc2] 29 ; X64-NEXT: vpaddd %zmm0, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfe,0xc0] 30 ; X64-NEXT: vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0] 31 ; X64-NEXT: retq # encoding: [0xc3] 32 %x2 = load <16 x i32>, <16 x i32>* %x2p 33 %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) 34 %2 = bitcast i16 %x3 to <16 x i1> 35 %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0 36 %4 = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4) 37 %5 = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4) 38 %6 = bitcast i16 %x3 to <16 x i1> 39 %7 = select <16 x i1> %6, <16 x i32> %5, <16 x i32> zeroinitializer 40 %res3 = add <16 x i32> %3, %4 41 %res4 = add <16 x i32> %7, %res3 42 ret <16 x i32> %res4 43 } 44 45 declare <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32>, <16 x i32>, <16 x i32>) 46 47 define <16 x i32>@test_int_x86_avx512_mask_vpdpbusds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, <16 x i32> %x4, i16 %x3) { 48 ; X86-LABEL: test_int_x86_avx512_mask_vpdpbusds_512: 49 ; X86: # %bb.0: 50 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] 51 ; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08] 52 ; X86-NEXT: vmovaps %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xd8] 53 ; X86-NEXT: vpdpbusds (%eax), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x51,0x18] 54 ; X86-NEXT: vmovaps %zmm0, %zmm4 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xe0] 55 ; X86-NEXT: vpdpbusds %zmm2, %zmm1, %zmm4 # encoding: [0x62,0xf2,0x75,0x48,0x51,0xe2] 56 ; X86-NEXT: vpdpbusds %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x51,0xc2] 57 ; X86-NEXT: vpaddd %zmm0, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfe,0xc0] 58 ; X86-NEXT: vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0] 59 ; X86-NEXT: retl # encoding: [0xc3] 60 ; 61 ; X64-LABEL: test_int_x86_avx512_mask_vpdpbusds_512: 62 ; X64: # %bb.0: 63 ; X64-NEXT: kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce] 64 ; X64-NEXT: vmovaps %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xd8] 65 ; X64-NEXT: vpdpbusds (%rdi), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x51,0x1f] 66 ; X64-NEXT: vmovaps %zmm0, %zmm4 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xe0] 67 ; X64-NEXT: vpdpbusds %zmm2, %zmm1, %zmm4 # encoding: [0x62,0xf2,0x75,0x48,0x51,0xe2] 68 ; X64-NEXT: vpdpbusds %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x51,0xc2] 69 ; X64-NEXT: vpaddd %zmm0, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfe,0xc0] 70 ; X64-NEXT: vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0] 71 ; X64-NEXT: retq # encoding: [0xc3] 72 %x2 = load <16 x i32>, <16 x i32>* %x2p 73 %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) 74 %2 = bitcast i16 %x3 to <16 x i1> 75 %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0 76 %4 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4) 77 %5 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4) 78 %6 = bitcast i16 %x3 to <16 x i1> 79 %7 = select <16 x i1> %6, <16 x i32> %5, <16 x i32> zeroinitializer 80 %res3 = add <16 x i32> %3, %4 81 %res4 = add <16 x i32> %7, %res3 82 ret <16 x i32> %res4 83 } 84 85 declare <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32>, <16 x i32>, <16 x i32>) 86 87 define <16 x i32>@test_int_x86_avx512_mask_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, <16 x i32> %x4, i16 %x3) { 88 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssd_512: 89 ; X86: # %bb.0: 90 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] 91 ; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08] 92 ; X86-NEXT: vmovaps %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xd8] 93 ; X86-NEXT: vpdpwssd (%eax), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x52,0x18] 94 ; X86-NEXT: vmovaps %zmm0, %zmm4 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xe0] 95 ; X86-NEXT: vpdpwssd %zmm2, %zmm1, %zmm4 # encoding: [0x62,0xf2,0x75,0x48,0x52,0xe2] 96 ; X86-NEXT: vpdpwssd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x52,0xc2] 97 ; X86-NEXT: vpaddd %zmm0, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfe,0xc0] 98 ; X86-NEXT: vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0] 99 ; X86-NEXT: retl # encoding: [0xc3] 100 ; 101 ; X64-LABEL: test_int_x86_avx512_mask_vpdpwssd_512: 102 ; X64: # %bb.0: 103 ; X64-NEXT: kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce] 104 ; X64-NEXT: vmovaps %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xd8] 105 ; X64-NEXT: vpdpwssd (%rdi), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x52,0x1f] 106 ; X64-NEXT: vmovaps %zmm0, %zmm4 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xe0] 107 ; X64-NEXT: vpdpwssd %zmm2, %zmm1, %zmm4 # encoding: [0x62,0xf2,0x75,0x48,0x52,0xe2] 108 ; X64-NEXT: vpdpwssd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x52,0xc2] 109 ; X64-NEXT: vpaddd %zmm0, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfe,0xc0] 110 ; X64-NEXT: vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0] 111 ; X64-NEXT: retq # encoding: [0xc3] 112 %x2 = load <16 x i32>, <16 x i32>* %x2p 113 %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) 114 %2 = bitcast i16 %x3 to <16 x i1> 115 %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0 116 %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4) 117 %5 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4) 118 %6 = bitcast i16 %x3 to <16 x i1> 119 %7 = select <16 x i1> %6, <16 x i32> %5, <16 x i32> zeroinitializer 120 %res3 = add <16 x i32> %3, %4 121 %res4 = add <16 x i32> %7, %res3 122 ret <16 x i32> %res4 123 } 124 125 declare <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>) 126 127 define <16 x i32>@test_int_x86_avx512_mask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, <16 x i32> %x4, i16 %x3) { 128 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssds_512: 129 ; X86: # %bb.0: 130 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] 131 ; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08] 132 ; X86-NEXT: vmovaps %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xd8] 133 ; X86-NEXT: vpdpwssds (%eax), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x53,0x18] 134 ; X86-NEXT: vmovaps %zmm0, %zmm4 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xe0] 135 ; X86-NEXT: vpdpwssds %zmm2, %zmm1, %zmm4 # encoding: [0x62,0xf2,0x75,0x48,0x53,0xe2] 136 ; X86-NEXT: vpdpwssds %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x53,0xc2] 137 ; X86-NEXT: vpaddd %zmm0, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfe,0xc0] 138 ; X86-NEXT: vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0] 139 ; X86-NEXT: retl # encoding: [0xc3] 140 ; 141 ; X64-LABEL: test_int_x86_avx512_mask_vpdpwssds_512: 142 ; X64: # %bb.0: 143 ; X64-NEXT: kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce] 144 ; X64-NEXT: vmovaps %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xd8] 145 ; X64-NEXT: vpdpwssds (%rdi), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x53,0x1f] 146 ; X64-NEXT: vmovaps %zmm0, %zmm4 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xe0] 147 ; X64-NEXT: vpdpwssds %zmm2, %zmm1, %zmm4 # encoding: [0x62,0xf2,0x75,0x48,0x53,0xe2] 148 ; X64-NEXT: vpdpwssds %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x53,0xc2] 149 ; X64-NEXT: vpaddd %zmm0, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfe,0xc0] 150 ; X64-NEXT: vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0] 151 ; X64-NEXT: retq # encoding: [0xc3] 152 %x2 = load <16 x i32>, <16 x i32>* %x2p 153 %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) 154 %2 = bitcast i16 %x3 to <16 x i1> 155 %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0 156 %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4) 157 %5 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4) 158 %6 = bitcast i16 %x3 to <16 x i1> 159 %7 = select <16 x i1> %6, <16 x i32> %5, <16 x i32> zeroinitializer 160 %res3 = add <16 x i32> %3, %4 161 %res4 = add <16 x i32> %7, %res3 162 ret <16 x i32> %res4 163 } 164 165
