1027 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 52; RUN: opt < %s -mattr=+avx512vnni,+avx512vl -passes=msan -S | FileCheck %s3 4; Forked from llvm/test/CodeGen/X86/avx512vl_vnni-intrinsics.ll5;6; Handled strictly: (none)7;8; Handled heuristically: (none)9 10target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"11target triple = "x86_64-unknown-linux-gnu"12 13declare <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32>, <32 x i8>, <32 x i8>)14 15define <8 x i32>@test_int_x86_avx512_vpdpbusd_256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) sanitize_memory {16; CHECK-LABEL: define <8 x i32> @test_int_x86_avx512_vpdpbusd_256(17; CHECK-SAME: <8 x i32> [[X0:%.*]], <32 x i8> [[X1:%.*]], <32 x i8> [[X2:%.*]]) #[[ATTR1:[0-9]+]] {18; CHECK-NEXT: [[TMP24:%.*]] = load <32 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 819; CHECK-NEXT: [[TMP5:%.*]] = load <32 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 820; CHECK-NEXT: [[TMP23:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 821; CHECK-NEXT: call void @llvm.donothing()22; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <32 x i8> [[TMP24]], zeroinitializer23; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <32 x i8> [[TMP5]], zeroinitializer24; CHECK-NEXT: [[TMP6:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer25; CHECK-NEXT: [[TMP7:%.*]] = icmp ne <32 x i8> [[X2]], zeroinitializer26; CHECK-NEXT: [[TMP12:%.*]] = and <32 x i1> [[TMP10]], [[TMP11]]27; CHECK-NEXT: [[TMP13:%.*]] = and <32 x i1> [[TMP6]], [[TMP11]]28; CHECK-NEXT: [[TMP14:%.*]] = and <32 x i1> [[TMP10]], [[TMP7]]29; CHECK-NEXT: [[TMP15:%.*]] = or <32 x i1> [[TMP12]], [[TMP13]]30; CHECK-NEXT: [[TMP16:%.*]] = or <32 x i1> [[TMP15]], [[TMP14]]31; CHECK-NEXT: [[TMP17:%.*]] = sext <32 x i1> [[TMP16]] to <32 x i8>32; CHECK-NEXT: [[TMP18:%.*]] = bitcast <32 x i8> [[TMP17]] to <8 x i32>33; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <8 x i32> [[TMP18]], zeroinitializer34; CHECK-NEXT: [[TMP22:%.*]] = sext <8 x i1> [[TMP19]] to <8 x i32>35; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP22]], [[TMP23]]36; CHECK-NEXT: [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X2]])37; CHECK-NEXT: store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 838; CHECK-NEXT: ret <8 x i32> [[TMP4]]39;40 %1 = call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2)41 ret <8 x i32> %142}43 44define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpbusd_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4, i8 %x3) sanitize_memory {45; CHECK-LABEL: define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpbusd_256(46; CHECK-SAME: <8 x i32> [[X0:%.*]], <32 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <32 x i8> [[X4:%.*]], i8 [[X3:%.*]]) #[[ATTR1]] {47; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 848; CHECK-NEXT: [[TMP33:%.*]] = load <32 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 849; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 850; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 104), align 851; CHECK-NEXT: [[TMP40:%.*]] = load <32 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 852; CHECK-NEXT: call void @llvm.donothing()53; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 054; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1:![0-9]+]]55; CHECK: [[BB6]]:56; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4:[0-9]+]]57; CHECK-NEXT: unreachable58; CHECK: [[BB7]]:59; CHECK-NEXT: [[X2:%.*]] = load <32 x i8>, ptr [[X2P]], align 3260; CHECK-NEXT: [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i6461; CHECK-NEXT: [[TMP9:%.*]] = xor i64 [[TMP8]], 8796093022208062; CHECK-NEXT: [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr63; CHECK-NEXT: [[TMP30:%.*]] = load <32 x i8>, ptr [[TMP10]], align 3264; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <32 x i8> [[TMP33]], zeroinitializer65; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <32 x i8> [[TMP30]], zeroinitializer66; CHECK-NEXT: [[TMP31:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer67; CHECK-NEXT: [[TMP32:%.*]] = icmp ne <32 x i8> [[X2]], zeroinitializer68; CHECK-NEXT: [[TMP38:%.*]] = and <32 x i1> [[TMP35]], [[TMP36]]69; CHECK-NEXT: [[TMP58:%.*]] = and <32 x i1> [[TMP31]], [[TMP36]]70; CHECK-NEXT: [[TMP59:%.*]] = and <32 x i1> [[TMP35]], [[TMP32]]71; CHECK-NEXT: [[TMP60:%.*]] = or <32 x i1> [[TMP38]], [[TMP58]]72; CHECK-NEXT: [[TMP61:%.*]] = or <32 x i1> [[TMP60]], [[TMP59]]73; CHECK-NEXT: [[TMP62:%.*]] = sext <32 x i1> [[TMP61]] to <32 x i8>74; CHECK-NEXT: [[TMP53:%.*]] = bitcast <32 x i8> [[TMP62]] to <8 x i32>75; CHECK-NEXT: [[TMP54:%.*]] = icmp ne <8 x i32> [[TMP53]], zeroinitializer76; CHECK-NEXT: [[TMP29:%.*]] = sext <8 x i1> [[TMP54]] to <8 x i32>77; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP29]], [[TMP2]]78; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X2]])79; CHECK-NEXT: [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>80; CHECK-NEXT: [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>81; CHECK-NEXT: [[TMP14:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP2]]82; CHECK-NEXT: [[TMP15:%.*]] = xor <8 x i32> [[TMP11]], [[X0]]83; CHECK-NEXT: [[TMP16:%.*]] = or <8 x i32> [[TMP15]], [[_MSPROP1]]84; CHECK-NEXT: [[TMP17:%.*]] = or <8 x i32> [[TMP16]], [[TMP2]]85; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <8 x i1> [[TMP12]], <8 x i32> [[TMP17]], <8 x i32> [[TMP14]]86; CHECK-NEXT: [[TMP18:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[TMP11]], <8 x i32> [[X0]]87; CHECK-NEXT: [[TMP39:%.*]] = icmp ne <32 x i8> [[TMP33]], zeroinitializer88; CHECK-NEXT: [[TMP46:%.*]] = icmp ne <32 x i8> [[TMP40]], zeroinitializer89; CHECK-NEXT: [[TMP37:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer90; CHECK-NEXT: [[TMP41:%.*]] = icmp ne <32 x i8> [[X4]], zeroinitializer91; CHECK-NEXT: [[TMP47:%.*]] = and <32 x i1> [[TMP39]], [[TMP46]]92; CHECK-NEXT: [[TMP48:%.*]] = and <32 x i1> [[TMP37]], [[TMP46]]93; CHECK-NEXT: [[TMP49:%.*]] = and <32 x i1> [[TMP39]], [[TMP41]]94; CHECK-NEXT: [[TMP50:%.*]] = or <32 x i1> [[TMP47]], [[TMP48]]95; CHECK-NEXT: [[TMP51:%.*]] = or <32 x i1> [[TMP50]], [[TMP49]]96; CHECK-NEXT: [[TMP52:%.*]] = sext <32 x i1> [[TMP51]] to <32 x i8>97; CHECK-NEXT: [[TMP55:%.*]] = bitcast <32 x i8> [[TMP52]] to <8 x i32>98; CHECK-NEXT: [[TMP56:%.*]] = icmp ne <8 x i32> [[TMP55]], zeroinitializer99; CHECK-NEXT: [[TMP57:%.*]] = sext <8 x i1> [[TMP56]] to <8 x i32>100; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[TMP57]], [[TMP2]]101; CHECK-NEXT: [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X4]])102; CHECK-NEXT: [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>103; CHECK-NEXT: [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>104; CHECK-NEXT: [[TMP22:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[_MSPROP3]], <8 x i32> zeroinitializer105; CHECK-NEXT: [[TMP23:%.*]] = xor <8 x i32> [[TMP19]], zeroinitializer106; CHECK-NEXT: [[TMP24:%.*]] = or <8 x i32> [[TMP23]], [[_MSPROP3]]107; CHECK-NEXT: [[TMP25:%.*]] = or <8 x i32> [[TMP24]], zeroinitializer108; CHECK-NEXT: [[_MSPROP_SELECT4:%.*]] = select <8 x i1> [[TMP20]], <8 x i32> [[TMP25]], <8 x i32> [[TMP22]]109; CHECK-NEXT: [[TMP26:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[TMP19]], <8 x i32> zeroinitializer110; CHECK-NEXT: [[TMP27:%.*]] = insertvalue { <8 x i32>, <8 x i32> } { <8 x i32> splat (i32 -1), <8 x i32> splat (i32 -1) }, <8 x i32> [[_MSPROP_SELECT]], 0111; CHECK-NEXT: [[RES1:%.*]] = insertvalue { <8 x i32>, <8 x i32> } poison, <8 x i32> [[TMP18]], 0112; CHECK-NEXT: [[TMP28:%.*]] = insertvalue { <8 x i32>, <8 x i32> } [[TMP27]], <8 x i32> [[_MSPROP_SELECT4]], 1113; CHECK-NEXT: [[RES2:%.*]] = insertvalue { <8 x i32>, <8 x i32> } [[RES1]], <8 x i32> [[TMP26]], 1114; CHECK-NEXT: store { <8 x i32>, <8 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8115; CHECK-NEXT: ret { <8 x i32>, <8 x i32> } [[RES2]]116;117 %x2 = load <32 x i8>, ptr %x2p118 %1 = call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2)119 %2 = bitcast i8 %x3 to <8 x i1>120 %3 = select <8 x i1> %2, <8 x i32> %1, <8 x i32> %x0121 %4 = call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4)122 %5 = bitcast i8 %x3 to <8 x i1>123 %6 = select <8 x i1> %5, <8 x i32> %4, <8 x i32> zeroinitializer124 %res1 = insertvalue { <8 x i32>, <8 x i32> } poison, <8 x i32> %3, 0125 %res2 = insertvalue { <8 x i32>, <8 x i32> } %res1, <8 x i32> %6, 1126 ret { <8 x i32>, <8 x i32> } %res2127}128 129declare <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32>, <16 x i8>, <16 x i8>)130 131define <4 x i32>@test_int_x86_avx512_vpdpbusd_128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) sanitize_memory {132; CHECK-LABEL: define <4 x i32> @test_int_x86_avx512_vpdpbusd_128(133; CHECK-SAME: <4 x i32> [[X0:%.*]], <16 x i8> [[X1:%.*]], <16 x i8> [[X2:%.*]]) #[[ATTR1]] {134; CHECK-NEXT: [[TMP24:%.*]] = load <16 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8135; CHECK-NEXT: [[TMP5:%.*]] = load <16 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8136; CHECK-NEXT: [[TMP23:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8137; CHECK-NEXT: call void @llvm.donothing()138; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <16 x i8> [[TMP24]], zeroinitializer139; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <16 x i8> [[TMP5]], zeroinitializer140; CHECK-NEXT: [[TMP6:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer141; CHECK-NEXT: [[TMP7:%.*]] = icmp ne <16 x i8> [[X2]], zeroinitializer142; CHECK-NEXT: [[TMP12:%.*]] = and <16 x i1> [[TMP10]], [[TMP11]]143; CHECK-NEXT: [[TMP13:%.*]] = and <16 x i1> [[TMP6]], [[TMP11]]144; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP10]], [[TMP7]]145; CHECK-NEXT: [[TMP15:%.*]] = or <16 x i1> [[TMP12]], [[TMP13]]146; CHECK-NEXT: [[TMP16:%.*]] = or <16 x i1> [[TMP15]], [[TMP14]]147; CHECK-NEXT: [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i8>148; CHECK-NEXT: [[TMP18:%.*]] = bitcast <16 x i8> [[TMP17]] to <4 x i32>149; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <4 x i32> [[TMP18]], zeroinitializer150; CHECK-NEXT: [[TMP22:%.*]] = sext <4 x i1> [[TMP19]] to <4 x i32>151; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP22]], [[TMP23]]152; CHECK-NEXT: [[TMP4:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X2]])153; CHECK-NEXT: store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8154; CHECK-NEXT: ret <4 x i32> [[TMP4]]155;156 %1 = call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2)157 ret <4 x i32> %1158}159 160define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpbusd_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4, i8 %x3) sanitize_memory {161; CHECK-LABEL: define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpbusd_128(162; CHECK-SAME: <4 x i32> [[X0:%.*]], <16 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <16 x i8> [[X4:%.*]], i8 [[X3:%.*]]) #[[ATTR1]] {163; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8164; CHECK-NEXT: [[TMP33:%.*]] = load <16 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8165; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8166; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 56), align 8167; CHECK-NEXT: [[TMP40:%.*]] = load <16 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8168; CHECK-NEXT: call void @llvm.donothing()169; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0170; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]171; CHECK: [[BB6]]:172; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]173; CHECK-NEXT: unreachable174; CHECK: [[BB7]]:175; CHECK-NEXT: [[X2:%.*]] = load <16 x i8>, ptr [[X2P]], align 16176; CHECK-NEXT: [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64177; CHECK-NEXT: [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080178; CHECK-NEXT: [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr179; CHECK-NEXT: [[TMP30:%.*]] = load <16 x i8>, ptr [[TMP10]], align 16180; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <16 x i8> [[TMP33]], zeroinitializer181; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <16 x i8> [[TMP30]], zeroinitializer182; CHECK-NEXT: [[TMP31:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer183; CHECK-NEXT: [[TMP32:%.*]] = icmp ne <16 x i8> [[X2]], zeroinitializer184; CHECK-NEXT: [[TMP38:%.*]] = and <16 x i1> [[TMP35]], [[TMP36]]185; CHECK-NEXT: [[TMP58:%.*]] = and <16 x i1> [[TMP31]], [[TMP36]]186; CHECK-NEXT: [[TMP59:%.*]] = and <16 x i1> [[TMP35]], [[TMP32]]187; CHECK-NEXT: [[TMP60:%.*]] = or <16 x i1> [[TMP38]], [[TMP58]]188; CHECK-NEXT: [[TMP61:%.*]] = or <16 x i1> [[TMP60]], [[TMP59]]189; CHECK-NEXT: [[TMP62:%.*]] = sext <16 x i1> [[TMP61]] to <16 x i8>190; CHECK-NEXT: [[TMP53:%.*]] = bitcast <16 x i8> [[TMP62]] to <4 x i32>191; CHECK-NEXT: [[TMP54:%.*]] = icmp ne <4 x i32> [[TMP53]], zeroinitializer192; CHECK-NEXT: [[TMP29:%.*]] = sext <4 x i1> [[TMP54]] to <4 x i32>193; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP29]], [[TMP2]]194; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X2]])195; CHECK-NEXT: [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>196; CHECK-NEXT: [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>197; CHECK-NEXT: [[_MSPROP2:%.*]] = shufflevector <8 x i1> [[TMP12]], <8 x i1> [[TMP12]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>198; CHECK-NEXT: [[EXTRACT:%.*]] = shufflevector <8 x i1> [[TMP13]], <8 x i1> [[TMP13]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>199; CHECK-NEXT: [[TMP14:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[_MSPROP1]], <4 x i32> [[TMP2]]200; CHECK-NEXT: [[TMP15:%.*]] = xor <4 x i32> [[TMP11]], [[X0]]201; CHECK-NEXT: [[TMP16:%.*]] = or <4 x i32> [[TMP15]], [[_MSPROP1]]202; CHECK-NEXT: [[TMP17:%.*]] = or <4 x i32> [[TMP16]], [[TMP2]]203; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <4 x i1> [[_MSPROP2]], <4 x i32> [[TMP17]], <4 x i32> [[TMP14]]204; CHECK-NEXT: [[TMP18:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[TMP11]], <4 x i32> [[X0]]205; CHECK-NEXT: [[TMP39:%.*]] = icmp ne <16 x i8> [[TMP33]], zeroinitializer206; CHECK-NEXT: [[TMP46:%.*]] = icmp ne <16 x i8> [[TMP40]], zeroinitializer207; CHECK-NEXT: [[TMP37:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer208; CHECK-NEXT: [[TMP41:%.*]] = icmp ne <16 x i8> [[X4]], zeroinitializer209; CHECK-NEXT: [[TMP47:%.*]] = and <16 x i1> [[TMP39]], [[TMP46]]210; CHECK-NEXT: [[TMP48:%.*]] = and <16 x i1> [[TMP37]], [[TMP46]]211; CHECK-NEXT: [[TMP49:%.*]] = and <16 x i1> [[TMP39]], [[TMP41]]212; CHECK-NEXT: [[TMP50:%.*]] = or <16 x i1> [[TMP47]], [[TMP48]]213; CHECK-NEXT: [[TMP51:%.*]] = or <16 x i1> [[TMP50]], [[TMP49]]214; CHECK-NEXT: [[TMP52:%.*]] = sext <16 x i1> [[TMP51]] to <16 x i8>215; CHECK-NEXT: [[TMP55:%.*]] = bitcast <16 x i8> [[TMP52]] to <4 x i32>216; CHECK-NEXT: [[TMP56:%.*]] = icmp ne <4 x i32> [[TMP55]], zeroinitializer217; CHECK-NEXT: [[TMP57:%.*]] = sext <4 x i1> [[TMP56]] to <4 x i32>218; CHECK-NEXT: [[_MSPROP4:%.*]] = or <4 x i32> [[TMP57]], [[TMP2]]219; CHECK-NEXT: [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X4]])220; CHECK-NEXT: [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>221; CHECK-NEXT: [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>222; CHECK-NEXT: [[_MSPROP5:%.*]] = shufflevector <8 x i1> [[TMP20]], <8 x i1> [[TMP20]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>223; CHECK-NEXT: [[EXTRACT1:%.*]] = shufflevector <8 x i1> [[TMP21]], <8 x i1> [[TMP21]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>224; CHECK-NEXT: [[TMP22:%.*]] = select <4 x i1> [[EXTRACT1]], <4 x i32> [[_MSPROP4]], <4 x i32> zeroinitializer225; CHECK-NEXT: [[TMP23:%.*]] = xor <4 x i32> [[TMP19]], zeroinitializer226; CHECK-NEXT: [[TMP24:%.*]] = or <4 x i32> [[TMP23]], [[_MSPROP4]]227; CHECK-NEXT: [[TMP25:%.*]] = or <4 x i32> [[TMP24]], zeroinitializer228; CHECK-NEXT: [[_MSPROP_SELECT6:%.*]] = select <4 x i1> [[_MSPROP5]], <4 x i32> [[TMP25]], <4 x i32> [[TMP22]]229; CHECK-NEXT: [[TMP26:%.*]] = select <4 x i1> [[EXTRACT1]], <4 x i32> [[TMP19]], <4 x i32> zeroinitializer230; CHECK-NEXT: [[TMP27:%.*]] = insertvalue { <4 x i32>, <4 x i32> } { <4 x i32> splat (i32 -1), <4 x i32> splat (i32 -1) }, <4 x i32> [[_MSPROP_SELECT]], 0231; CHECK-NEXT: [[RES1:%.*]] = insertvalue { <4 x i32>, <4 x i32> } poison, <4 x i32> [[TMP18]], 0232; CHECK-NEXT: [[TMP28:%.*]] = insertvalue { <4 x i32>, <4 x i32> } [[TMP27]], <4 x i32> [[_MSPROP_SELECT6]], 1233; CHECK-NEXT: [[RES2:%.*]] = insertvalue { <4 x i32>, <4 x i32> } [[RES1]], <4 x i32> [[TMP26]], 1234; CHECK-NEXT: store { <4 x i32>, <4 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8235; CHECK-NEXT: ret { <4 x i32>, <4 x i32> } [[RES2]]236;237 %x2 = load <16 x i8>, ptr %x2p238 %1 = call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2)239 %2 = bitcast i8 %x3 to <8 x i1>240 %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>241 %3 = select <4 x i1> %extract, <4 x i32> %1, <4 x i32> %x0242 %4 = call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4)243 %5 = bitcast i8 %x3 to <8 x i1>244 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>245 %6 = select <4 x i1> %extract1, <4 x i32> %4, <4 x i32> zeroinitializer246 %res1 = insertvalue { <4 x i32>, <4 x i32> } poison, <4 x i32> %3, 0247 %res2 = insertvalue { <4 x i32>, <4 x i32> } %res1, <4 x i32> %6, 1248 ret { <4 x i32>, <4 x i32> } %res2249}250 251declare <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32>, <32 x i8>, <32 x i8>)252 253define <8 x i32>@test_int_x86_avx512_vpdpbusds_256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) sanitize_memory {254; CHECK-LABEL: define <8 x i32> @test_int_x86_avx512_vpdpbusds_256(255; CHECK-SAME: <8 x i32> [[X0:%.*]], <32 x i8> [[X1:%.*]], <32 x i8> [[X2:%.*]]) #[[ATTR1]] {256; CHECK-NEXT: [[TMP24:%.*]] = load <32 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8257; CHECK-NEXT: [[TMP5:%.*]] = load <32 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8258; CHECK-NEXT: [[TMP23:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8259; CHECK-NEXT: call void @llvm.donothing()260; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <32 x i8> [[TMP24]], zeroinitializer261; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <32 x i8> [[TMP5]], zeroinitializer262; CHECK-NEXT: [[TMP6:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer263; CHECK-NEXT: [[TMP7:%.*]] = icmp ne <32 x i8> [[X2]], zeroinitializer264; CHECK-NEXT: [[TMP12:%.*]] = and <32 x i1> [[TMP10]], [[TMP11]]265; CHECK-NEXT: [[TMP13:%.*]] = and <32 x i1> [[TMP6]], [[TMP11]]266; CHECK-NEXT: [[TMP14:%.*]] = and <32 x i1> [[TMP10]], [[TMP7]]267; CHECK-NEXT: [[TMP15:%.*]] = or <32 x i1> [[TMP12]], [[TMP13]]268; CHECK-NEXT: [[TMP16:%.*]] = or <32 x i1> [[TMP15]], [[TMP14]]269; CHECK-NEXT: [[TMP17:%.*]] = sext <32 x i1> [[TMP16]] to <32 x i8>270; CHECK-NEXT: [[TMP18:%.*]] = bitcast <32 x i8> [[TMP17]] to <8 x i32>271; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <8 x i32> [[TMP18]], zeroinitializer272; CHECK-NEXT: [[TMP22:%.*]] = sext <8 x i1> [[TMP19]] to <8 x i32>273; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP22]], [[TMP23]]274; CHECK-NEXT: [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X2]])275; CHECK-NEXT: store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8276; CHECK-NEXT: ret <8 x i32> [[TMP4]]277;278 %1 = call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2)279 ret <8 x i32> %1280}281 282define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpbusds_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4, i8 %x3) sanitize_memory {283; CHECK-LABEL: define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpbusds_256(284; CHECK-SAME: <8 x i32> [[X0:%.*]], <32 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <32 x i8> [[X4:%.*]], i8 [[X3:%.*]]) #[[ATTR1]] {285; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8286; CHECK-NEXT: [[TMP33:%.*]] = load <32 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8287; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8288; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 104), align 8289; CHECK-NEXT: [[TMP40:%.*]] = load <32 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8290; CHECK-NEXT: call void @llvm.donothing()291; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0292; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]293; CHECK: [[BB6]]:294; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]295; CHECK-NEXT: unreachable296; CHECK: [[BB7]]:297; CHECK-NEXT: [[X2:%.*]] = load <32 x i8>, ptr [[X2P]], align 32298; CHECK-NEXT: [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64299; CHECK-NEXT: [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080300; CHECK-NEXT: [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr301; CHECK-NEXT: [[TMP30:%.*]] = load <32 x i8>, ptr [[TMP10]], align 32302; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <32 x i8> [[TMP33]], zeroinitializer303; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <32 x i8> [[TMP30]], zeroinitializer304; CHECK-NEXT: [[TMP31:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer305; CHECK-NEXT: [[TMP32:%.*]] = icmp ne <32 x i8> [[X2]], zeroinitializer306; CHECK-NEXT: [[TMP38:%.*]] = and <32 x i1> [[TMP35]], [[TMP36]]307; CHECK-NEXT: [[TMP58:%.*]] = and <32 x i1> [[TMP31]], [[TMP36]]308; CHECK-NEXT: [[TMP59:%.*]] = and <32 x i1> [[TMP35]], [[TMP32]]309; CHECK-NEXT: [[TMP60:%.*]] = or <32 x i1> [[TMP38]], [[TMP58]]310; CHECK-NEXT: [[TMP61:%.*]] = or <32 x i1> [[TMP60]], [[TMP59]]311; CHECK-NEXT: [[TMP62:%.*]] = sext <32 x i1> [[TMP61]] to <32 x i8>312; CHECK-NEXT: [[TMP53:%.*]] = bitcast <32 x i8> [[TMP62]] to <8 x i32>313; CHECK-NEXT: [[TMP54:%.*]] = icmp ne <8 x i32> [[TMP53]], zeroinitializer314; CHECK-NEXT: [[TMP29:%.*]] = sext <8 x i1> [[TMP54]] to <8 x i32>315; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP29]], [[TMP2]]316; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X2]])317; CHECK-NEXT: [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>318; CHECK-NEXT: [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>319; CHECK-NEXT: [[TMP14:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP2]]320; CHECK-NEXT: [[TMP15:%.*]] = xor <8 x i32> [[TMP11]], [[X0]]321; CHECK-NEXT: [[TMP16:%.*]] = or <8 x i32> [[TMP15]], [[_MSPROP1]]322; CHECK-NEXT: [[TMP17:%.*]] = or <8 x i32> [[TMP16]], [[TMP2]]323; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <8 x i1> [[TMP12]], <8 x i32> [[TMP17]], <8 x i32> [[TMP14]]324; CHECK-NEXT: [[TMP18:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[TMP11]], <8 x i32> [[X0]]325; CHECK-NEXT: [[TMP39:%.*]] = icmp ne <32 x i8> [[TMP33]], zeroinitializer326; CHECK-NEXT: [[TMP46:%.*]] = icmp ne <32 x i8> [[TMP40]], zeroinitializer327; CHECK-NEXT: [[TMP37:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer328; CHECK-NEXT: [[TMP41:%.*]] = icmp ne <32 x i8> [[X4]], zeroinitializer329; CHECK-NEXT: [[TMP47:%.*]] = and <32 x i1> [[TMP39]], [[TMP46]]330; CHECK-NEXT: [[TMP48:%.*]] = and <32 x i1> [[TMP37]], [[TMP46]]331; CHECK-NEXT: [[TMP49:%.*]] = and <32 x i1> [[TMP39]], [[TMP41]]332; CHECK-NEXT: [[TMP50:%.*]] = or <32 x i1> [[TMP47]], [[TMP48]]333; CHECK-NEXT: [[TMP51:%.*]] = or <32 x i1> [[TMP50]], [[TMP49]]334; CHECK-NEXT: [[TMP52:%.*]] = sext <32 x i1> [[TMP51]] to <32 x i8>335; CHECK-NEXT: [[TMP55:%.*]] = bitcast <32 x i8> [[TMP52]] to <8 x i32>336; CHECK-NEXT: [[TMP56:%.*]] = icmp ne <8 x i32> [[TMP55]], zeroinitializer337; CHECK-NEXT: [[TMP57:%.*]] = sext <8 x i1> [[TMP56]] to <8 x i32>338; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[TMP57]], [[TMP2]]339; CHECK-NEXT: [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X4]])340; CHECK-NEXT: [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>341; CHECK-NEXT: [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>342; CHECK-NEXT: [[TMP22:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[_MSPROP3]], <8 x i32> zeroinitializer343; CHECK-NEXT: [[TMP23:%.*]] = xor <8 x i32> [[TMP19]], zeroinitializer344; CHECK-NEXT: [[TMP24:%.*]] = or <8 x i32> [[TMP23]], [[_MSPROP3]]345; CHECK-NEXT: [[TMP25:%.*]] = or <8 x i32> [[TMP24]], zeroinitializer346; CHECK-NEXT: [[_MSPROP_SELECT4:%.*]] = select <8 x i1> [[TMP20]], <8 x i32> [[TMP25]], <8 x i32> [[TMP22]]347; CHECK-NEXT: [[TMP26:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[TMP19]], <8 x i32> zeroinitializer348; CHECK-NEXT: [[TMP27:%.*]] = insertvalue { <8 x i32>, <8 x i32> } { <8 x i32> splat (i32 -1), <8 x i32> splat (i32 -1) }, <8 x i32> [[_MSPROP_SELECT]], 0349; CHECK-NEXT: [[RES1:%.*]] = insertvalue { <8 x i32>, <8 x i32> } poison, <8 x i32> [[TMP18]], 0350; CHECK-NEXT: [[TMP28:%.*]] = insertvalue { <8 x i32>, <8 x i32> } [[TMP27]], <8 x i32> [[_MSPROP_SELECT4]], 1351; CHECK-NEXT: [[RES2:%.*]] = insertvalue { <8 x i32>, <8 x i32> } [[RES1]], <8 x i32> [[TMP26]], 1352; CHECK-NEXT: store { <8 x i32>, <8 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8353; CHECK-NEXT: ret { <8 x i32>, <8 x i32> } [[RES2]]354;355 %x2 = load <32 x i8>, ptr %x2p356 %1 = call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2)357 %2 = bitcast i8 %x3 to <8 x i1>358 %3 = select <8 x i1> %2, <8 x i32> %1, <8 x i32> %x0359 %4 = call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4)360 %5 = bitcast i8 %x3 to <8 x i1>361 %6 = select <8 x i1> %5, <8 x i32> %4, <8 x i32> zeroinitializer362 %res1 = insertvalue { <8 x i32>, <8 x i32> } poison, <8 x i32> %3, 0363 %res2 = insertvalue { <8 x i32>, <8 x i32> } %res1, <8 x i32> %6, 1364 ret { <8 x i32>, <8 x i32> } %res2365}366 367declare <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32>, <16 x i8>, <16 x i8>)368 369define <4 x i32>@test_int_x86_avx512_vpdpbusds_128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) sanitize_memory {370; CHECK-LABEL: define <4 x i32> @test_int_x86_avx512_vpdpbusds_128(371; CHECK-SAME: <4 x i32> [[X0:%.*]], <16 x i8> [[X1:%.*]], <16 x i8> [[X2:%.*]]) #[[ATTR1]] {372; CHECK-NEXT: [[TMP24:%.*]] = load <16 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8373; CHECK-NEXT: [[TMP5:%.*]] = load <16 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8374; CHECK-NEXT: [[TMP23:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8375; CHECK-NEXT: call void @llvm.donothing()376; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <16 x i8> [[TMP24]], zeroinitializer377; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <16 x i8> [[TMP5]], zeroinitializer378; CHECK-NEXT: [[TMP6:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer379; CHECK-NEXT: [[TMP7:%.*]] = icmp ne <16 x i8> [[X2]], zeroinitializer380; CHECK-NEXT: [[TMP12:%.*]] = and <16 x i1> [[TMP10]], [[TMP11]]381; CHECK-NEXT: [[TMP13:%.*]] = and <16 x i1> [[TMP6]], [[TMP11]]382; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP10]], [[TMP7]]383; CHECK-NEXT: [[TMP15:%.*]] = or <16 x i1> [[TMP12]], [[TMP13]]384; CHECK-NEXT: [[TMP16:%.*]] = or <16 x i1> [[TMP15]], [[TMP14]]385; CHECK-NEXT: [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i8>386; CHECK-NEXT: [[TMP18:%.*]] = bitcast <16 x i8> [[TMP17]] to <4 x i32>387; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <4 x i32> [[TMP18]], zeroinitializer388; CHECK-NEXT: [[TMP22:%.*]] = sext <4 x i1> [[TMP19]] to <4 x i32>389; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP22]], [[TMP23]]390; CHECK-NEXT: [[TMP4:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X2]])391; CHECK-NEXT: store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8392; CHECK-NEXT: ret <4 x i32> [[TMP4]]393;394 %1 = call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2)395 ret <4 x i32> %1396}397 398define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpbusds_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4, i8 %x3) sanitize_memory {399; CHECK-LABEL: define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpbusds_128(400; CHECK-SAME: <4 x i32> [[X0:%.*]], <16 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <16 x i8> [[X4:%.*]], i8 [[X3:%.*]]) #[[ATTR1]] {401; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8402; CHECK-NEXT: [[TMP33:%.*]] = load <16 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8403; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8404; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 56), align 8405; CHECK-NEXT: [[TMP40:%.*]] = load <16 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8406; CHECK-NEXT: call void @llvm.donothing()407; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0408; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]409; CHECK: [[BB6]]:410; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]411; CHECK-NEXT: unreachable412; CHECK: [[BB7]]:413; CHECK-NEXT: [[X2:%.*]] = load <16 x i8>, ptr [[X2P]], align 16414; CHECK-NEXT: [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64415; CHECK-NEXT: [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080416; CHECK-NEXT: [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr417; CHECK-NEXT: [[TMP30:%.*]] = load <16 x i8>, ptr [[TMP10]], align 16418; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <16 x i8> [[TMP33]], zeroinitializer419; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <16 x i8> [[TMP30]], zeroinitializer420; CHECK-NEXT: [[TMP31:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer421; CHECK-NEXT: [[TMP32:%.*]] = icmp ne <16 x i8> [[X2]], zeroinitializer422; CHECK-NEXT: [[TMP38:%.*]] = and <16 x i1> [[TMP35]], [[TMP36]]423; CHECK-NEXT: [[TMP58:%.*]] = and <16 x i1> [[TMP31]], [[TMP36]]424; CHECK-NEXT: [[TMP59:%.*]] = and <16 x i1> [[TMP35]], [[TMP32]]425; CHECK-NEXT: [[TMP60:%.*]] = or <16 x i1> [[TMP38]], [[TMP58]]426; CHECK-NEXT: [[TMP61:%.*]] = or <16 x i1> [[TMP60]], [[TMP59]]427; CHECK-NEXT: [[TMP62:%.*]] = sext <16 x i1> [[TMP61]] to <16 x i8>428; CHECK-NEXT: [[TMP53:%.*]] = bitcast <16 x i8> [[TMP62]] to <4 x i32>429; CHECK-NEXT: [[TMP54:%.*]] = icmp ne <4 x i32> [[TMP53]], zeroinitializer430; CHECK-NEXT: [[TMP29:%.*]] = sext <4 x i1> [[TMP54]] to <4 x i32>431; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP29]], [[TMP2]]432; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X2]])433; CHECK-NEXT: [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>434; CHECK-NEXT: [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>435; CHECK-NEXT: [[_MSPROP2:%.*]] = shufflevector <8 x i1> [[TMP12]], <8 x i1> [[TMP12]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>436; CHECK-NEXT: [[EXTRACT:%.*]] = shufflevector <8 x i1> [[TMP13]], <8 x i1> [[TMP13]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>437; CHECK-NEXT: [[TMP14:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[_MSPROP1]], <4 x i32> [[TMP2]]438; CHECK-NEXT: [[TMP15:%.*]] = xor <4 x i32> [[TMP11]], [[X0]]439; CHECK-NEXT: [[TMP16:%.*]] = or <4 x i32> [[TMP15]], [[_MSPROP1]]440; CHECK-NEXT: [[TMP17:%.*]] = or <4 x i32> [[TMP16]], [[TMP2]]441; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <4 x i1> [[_MSPROP2]], <4 x i32> [[TMP17]], <4 x i32> [[TMP14]]442; CHECK-NEXT: [[TMP18:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[TMP11]], <4 x i32> [[X0]]443; CHECK-NEXT: [[TMP39:%.*]] = icmp ne <16 x i8> [[TMP33]], zeroinitializer444; CHECK-NEXT: [[TMP46:%.*]] = icmp ne <16 x i8> [[TMP40]], zeroinitializer445; CHECK-NEXT: [[TMP37:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer446; CHECK-NEXT: [[TMP41:%.*]] = icmp ne <16 x i8> [[X4]], zeroinitializer447; CHECK-NEXT: [[TMP47:%.*]] = and <16 x i1> [[TMP39]], [[TMP46]]448; CHECK-NEXT: [[TMP48:%.*]] = and <16 x i1> [[TMP37]], [[TMP46]]449; CHECK-NEXT: [[TMP49:%.*]] = and <16 x i1> [[TMP39]], [[TMP41]]450; CHECK-NEXT: [[TMP50:%.*]] = or <16 x i1> [[TMP47]], [[TMP48]]451; CHECK-NEXT: [[TMP51:%.*]] = or <16 x i1> [[TMP50]], [[TMP49]]452; CHECK-NEXT: [[TMP52:%.*]] = sext <16 x i1> [[TMP51]] to <16 x i8>453; CHECK-NEXT: [[TMP55:%.*]] = bitcast <16 x i8> [[TMP52]] to <4 x i32>454; CHECK-NEXT: [[TMP56:%.*]] = icmp ne <4 x i32> [[TMP55]], zeroinitializer455; CHECK-NEXT: [[TMP57:%.*]] = sext <4 x i1> [[TMP56]] to <4 x i32>456; CHECK-NEXT: [[_MSPROP4:%.*]] = or <4 x i32> [[TMP57]], [[TMP2]]457; CHECK-NEXT: [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X4]])458; CHECK-NEXT: [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>459; CHECK-NEXT: [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>460; CHECK-NEXT: [[_MSPROP5:%.*]] = shufflevector <8 x i1> [[TMP20]], <8 x i1> [[TMP20]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>461; CHECK-NEXT: [[EXTRACT1:%.*]] = shufflevector <8 x i1> [[TMP21]], <8 x i1> [[TMP21]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>462; CHECK-NEXT: [[TMP22:%.*]] = select <4 x i1> [[EXTRACT1]], <4 x i32> [[_MSPROP4]], <4 x i32> zeroinitializer463; CHECK-NEXT: [[TMP23:%.*]] = xor <4 x i32> [[TMP19]], zeroinitializer464; CHECK-NEXT: [[TMP24:%.*]] = or <4 x i32> [[TMP23]], [[_MSPROP4]]465; CHECK-NEXT: [[TMP25:%.*]] = or <4 x i32> [[TMP24]], zeroinitializer466; CHECK-NEXT: [[_MSPROP_SELECT6:%.*]] = select <4 x i1> [[_MSPROP5]], <4 x i32> [[TMP25]], <4 x i32> [[TMP22]]467; CHECK-NEXT: [[TMP26:%.*]] = select <4 x i1> [[EXTRACT1]], <4 x i32> [[TMP19]], <4 x i32> zeroinitializer468; CHECK-NEXT: [[TMP27:%.*]] = insertvalue { <4 x i32>, <4 x i32> } { <4 x i32> splat (i32 -1), <4 x i32> splat (i32 -1) }, <4 x i32> [[_MSPROP_SELECT]], 0469; CHECK-NEXT: [[RES1:%.*]] = insertvalue { <4 x i32>, <4 x i32> } poison, <4 x i32> [[TMP18]], 0470; CHECK-NEXT: [[TMP28:%.*]] = insertvalue { <4 x i32>, <4 x i32> } [[TMP27]], <4 x i32> [[_MSPROP_SELECT6]], 1471; CHECK-NEXT: [[RES2:%.*]] = insertvalue { <4 x i32>, <4 x i32> } [[RES1]], <4 x i32> [[TMP26]], 1472; CHECK-NEXT: store { <4 x i32>, <4 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8473; CHECK-NEXT: ret { <4 x i32>, <4 x i32> } [[RES2]]474;475 %x2 = load <16 x i8>, ptr %x2p476 %1 = call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2)477 %2 = bitcast i8 %x3 to <8 x i1>478 %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>479 %3 = select <4 x i1> %extract, <4 x i32> %1, <4 x i32> %x0480 %4 = call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4)481 %5 = bitcast i8 %x3 to <8 x i1>482 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>483 %6 = select <4 x i1> %extract1, <4 x i32> %4, <4 x i32> zeroinitializer484 %res1 = insertvalue { <4 x i32>, <4 x i32> } poison, <4 x i32> %3, 0485 %res2 = insertvalue { <4 x i32>, <4 x i32> } %res1, <4 x i32> %6, 1486 ret { <4 x i32>, <4 x i32> } %res2487}488 489declare <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32>, <8 x i32>, <8 x i32>)490 491define <8 x i32>@test_int_x86_avx512_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) sanitize_memory {492; CHECK-LABEL: define <8 x i32> @test_int_x86_avx512_vpdpwssd_256(493; CHECK-SAME: <8 x i32> [[X0:%.*]], <8 x i32> [[X1:%.*]], <8 x i32> [[X2:%.*]]) #[[ATTR1]] {494; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8495; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8496; CHECK-NEXT: [[TMP21:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8497; CHECK-NEXT: call void @llvm.donothing()498; CHECK-NEXT: [[TMP22:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>499; CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>500; CHECK-NEXT: [[TMP6:%.*]] = bitcast <8 x i32> [[TMP2]] to <16 x i16>501; CHECK-NEXT: [[TMP7:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>502; CHECK-NEXT: [[TMP8:%.*]] = icmp ne <16 x i16> [[TMP6]], zeroinitializer503; CHECK-NEXT: [[TMP9:%.*]] = icmp ne <16 x i16> [[TMP7]], zeroinitializer504; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <16 x i16> [[TMP22]], zeroinitializer505; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <16 x i16> [[TMP5]], zeroinitializer506; CHECK-NEXT: [[TMP12:%.*]] = and <16 x i1> [[TMP8]], [[TMP9]]507; CHECK-NEXT: [[TMP13:%.*]] = and <16 x i1> [[TMP10]], [[TMP9]]508; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP8]], [[TMP11]]509; CHECK-NEXT: [[TMP15:%.*]] = or <16 x i1> [[TMP12]], [[TMP13]]510; CHECK-NEXT: [[TMP16:%.*]] = or <16 x i1> [[TMP15]], [[TMP14]]511; CHECK-NEXT: [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i16>512; CHECK-NEXT: [[TMP18:%.*]] = bitcast <16 x i16> [[TMP17]] to <8 x i32>513; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <8 x i32> [[TMP18]], zeroinitializer514; CHECK-NEXT: [[TMP20:%.*]] = sext <8 x i1> [[TMP19]] to <8 x i32>515; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP20]], [[TMP21]]516; CHECK-NEXT: [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])517; CHECK-NEXT: store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8518; CHECK-NEXT: ret <8 x i32> [[TMP4]]519;520 %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)521 ret <8 x i32> %1522}523 524define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4, i8 %x3) sanitize_memory {525; CHECK-LABEL: define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(526; CHECK-SAME: <8 x i32> [[X0:%.*]], <8 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <8 x i32> [[X4:%.*]], i8 [[X3:%.*]]) #[[ATTR1]] {527; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8528; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8529; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8530; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 104), align 8531; CHECK-NEXT: [[TMP5:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8532; CHECK-NEXT: call void @llvm.donothing()533; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0534; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]535; CHECK: [[BB6]]:536; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]537; CHECK-NEXT: unreachable538; CHECK: [[BB7]]:539; CHECK-NEXT: [[X2:%.*]] = load <8 x i32>, ptr [[X2P]], align 32540; CHECK-NEXT: [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64541; CHECK-NEXT: [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080542; CHECK-NEXT: [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr543; CHECK-NEXT: [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP10]], align 32544; CHECK-NEXT: [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>545; CHECK-NEXT: [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>546; CHECK-NEXT: [[TMP31:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>547; CHECK-NEXT: [[TMP32:%.*]] = bitcast <8 x i32> [[_MSLD]] to <16 x i16>548; CHECK-NEXT: [[TMP33:%.*]] = icmp ne <16 x i16> [[TMP31]], zeroinitializer549; CHECK-NEXT: [[TMP34:%.*]] = icmp ne <16 x i16> [[TMP32]], zeroinitializer550; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <16 x i16> [[TMP29]], zeroinitializer551; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <16 x i16> [[TMP30]], zeroinitializer552; CHECK-NEXT: [[TMP54:%.*]] = and <16 x i1> [[TMP33]], [[TMP34]]553; CHECK-NEXT: [[TMP55:%.*]] = and <16 x i1> [[TMP35]], [[TMP34]]554; CHECK-NEXT: [[TMP56:%.*]] = and <16 x i1> [[TMP33]], [[TMP36]]555; CHECK-NEXT: [[TMP57:%.*]] = or <16 x i1> [[TMP54]], [[TMP55]]556; CHECK-NEXT: [[TMP58:%.*]] = or <16 x i1> [[TMP57]], [[TMP56]]557; CHECK-NEXT: [[TMP59:%.*]] = sext <16 x i1> [[TMP58]] to <16 x i16>558; CHECK-NEXT: [[TMP60:%.*]] = bitcast <16 x i16> [[TMP59]] to <8 x i32>559; CHECK-NEXT: [[TMP61:%.*]] = icmp ne <8 x i32> [[TMP60]], zeroinitializer560; CHECK-NEXT: [[TMP62:%.*]] = sext <8 x i1> [[TMP61]] to <8 x i32>561; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP62]], [[TMP2]]562; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])563; CHECK-NEXT: [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>564; CHECK-NEXT: [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>565; CHECK-NEXT: [[TMP14:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP2]]566; CHECK-NEXT: [[TMP15:%.*]] = xor <8 x i32> [[TMP11]], [[X0]]567; CHECK-NEXT: [[TMP16:%.*]] = or <8 x i32> [[TMP15]], [[_MSPROP1]]568; CHECK-NEXT: [[TMP17:%.*]] = or <8 x i32> [[TMP16]], [[TMP2]]569; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <8 x i1> [[TMP12]], <8 x i32> [[TMP17]], <8 x i32> [[TMP14]]570; CHECK-NEXT: [[TMP18:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[TMP11]], <8 x i32> [[X0]]571; CHECK-NEXT: [[TMP37:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>572; CHECK-NEXT: [[TMP38:%.*]] = bitcast <8 x i32> [[X4]] to <16 x i16>573; CHECK-NEXT: [[TMP39:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>574; CHECK-NEXT: [[TMP40:%.*]] = bitcast <8 x i32> [[TMP5]] to <16 x i16>575; CHECK-NEXT: [[TMP41:%.*]] = icmp ne <16 x i16> [[TMP39]], zeroinitializer576; CHECK-NEXT: [[TMP42:%.*]] = icmp ne <16 x i16> [[TMP40]], zeroinitializer577; CHECK-NEXT: [[TMP43:%.*]] = icmp ne <16 x i16> [[TMP37]], zeroinitializer578; CHECK-NEXT: [[TMP44:%.*]] = icmp ne <16 x i16> [[TMP38]], zeroinitializer579; CHECK-NEXT: [[TMP45:%.*]] = and <16 x i1> [[TMP41]], [[TMP42]]580; CHECK-NEXT: [[TMP46:%.*]] = and <16 x i1> [[TMP43]], [[TMP42]]581; CHECK-NEXT: [[TMP47:%.*]] = and <16 x i1> [[TMP41]], [[TMP44]]582; CHECK-NEXT: [[TMP48:%.*]] = or <16 x i1> [[TMP45]], [[TMP46]]583; CHECK-NEXT: [[TMP49:%.*]] = or <16 x i1> [[TMP48]], [[TMP47]]584; CHECK-NEXT: [[TMP50:%.*]] = sext <16 x i1> [[TMP49]] to <16 x i16>585; CHECK-NEXT: [[TMP51:%.*]] = bitcast <16 x i16> [[TMP50]] to <8 x i32>586; CHECK-NEXT: [[TMP52:%.*]] = icmp ne <8 x i32> [[TMP51]], zeroinitializer587; CHECK-NEXT: [[TMP53:%.*]] = sext <8 x i1> [[TMP52]] to <8 x i32>588; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[TMP53]], [[TMP2]]589; CHECK-NEXT: [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]])590; CHECK-NEXT: [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>591; CHECK-NEXT: [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>592; CHECK-NEXT: [[TMP22:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[_MSPROP3]], <8 x i32> zeroinitializer593; CHECK-NEXT: [[TMP23:%.*]] = xor <8 x i32> [[TMP19]], zeroinitializer594; CHECK-NEXT: [[TMP24:%.*]] = or <8 x i32> [[TMP23]], [[_MSPROP3]]595; CHECK-NEXT: [[TMP25:%.*]] = or <8 x i32> [[TMP24]], zeroinitializer596; CHECK-NEXT: [[_MSPROP_SELECT4:%.*]] = select <8 x i1> [[TMP20]], <8 x i32> [[TMP25]], <8 x i32> [[TMP22]]597; CHECK-NEXT: [[TMP26:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[TMP19]], <8 x i32> zeroinitializer598; CHECK-NEXT: [[TMP27:%.*]] = insertvalue { <8 x i32>, <8 x i32> } { <8 x i32> splat (i32 -1), <8 x i32> splat (i32 -1) }, <8 x i32> [[_MSPROP_SELECT]], 0599; CHECK-NEXT: [[RES1:%.*]] = insertvalue { <8 x i32>, <8 x i32> } poison, <8 x i32> [[TMP18]], 0600; CHECK-NEXT: [[TMP28:%.*]] = insertvalue { <8 x i32>, <8 x i32> } [[TMP27]], <8 x i32> [[_MSPROP_SELECT4]], 1601; CHECK-NEXT: [[RES2:%.*]] = insertvalue { <8 x i32>, <8 x i32> } [[RES1]], <8 x i32> [[TMP26]], 1602; CHECK-NEXT: store { <8 x i32>, <8 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8603; CHECK-NEXT: ret { <8 x i32>, <8 x i32> } [[RES2]]604;605 %x2 = load <8 x i32>, ptr %x2p606 %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)607 %2 = bitcast i8 %x3 to <8 x i1>608 %3 = select <8 x i1> %2, <8 x i32> %1, <8 x i32> %x0609 %4 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4)610 %5 = bitcast i8 %x3 to <8 x i1>611 %6 = select <8 x i1> %5, <8 x i32> %4, <8 x i32> zeroinitializer612 %res1 = insertvalue { <8 x i32>, <8 x i32> } poison, <8 x i32> %3, 0613 %res2 = insertvalue { <8 x i32>, <8 x i32> } %res1, <8 x i32> %6, 1614 ret { <8 x i32>, <8 x i32> } %res2615}616 617declare <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32>, <4 x i32>, <4 x i32>)618 619define <4 x i32>@test_int_x86_avx512_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) sanitize_memory {620; CHECK-LABEL: define <4 x i32> @test_int_x86_avx512_vpdpwssd_128(621; CHECK-SAME: <4 x i32> [[X0:%.*]], <4 x i32> [[X1:%.*]], <4 x i32> [[X2:%.*]]) #[[ATTR1]] {622; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8623; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8624; CHECK-NEXT: [[TMP21:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8625; CHECK-NEXT: call void @llvm.donothing()626; CHECK-NEXT: [[TMP22:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>627; CHECK-NEXT: [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>628; CHECK-NEXT: [[TMP6:%.*]] = bitcast <4 x i32> [[TMP2]] to <8 x i16>629; CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>630; CHECK-NEXT: [[TMP8:%.*]] = icmp ne <8 x i16> [[TMP6]], zeroinitializer631; CHECK-NEXT: [[TMP9:%.*]] = icmp ne <8 x i16> [[TMP7]], zeroinitializer632; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <8 x i16> [[TMP22]], zeroinitializer633; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <8 x i16> [[TMP5]], zeroinitializer634; CHECK-NEXT: [[TMP12:%.*]] = and <8 x i1> [[TMP8]], [[TMP9]]635; CHECK-NEXT: [[TMP13:%.*]] = and <8 x i1> [[TMP10]], [[TMP9]]636; CHECK-NEXT: [[TMP14:%.*]] = and <8 x i1> [[TMP8]], [[TMP11]]637; CHECK-NEXT: [[TMP15:%.*]] = or <8 x i1> [[TMP12]], [[TMP13]]638; CHECK-NEXT: [[TMP16:%.*]] = or <8 x i1> [[TMP15]], [[TMP14]]639; CHECK-NEXT: [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i16>640; CHECK-NEXT: [[TMP18:%.*]] = bitcast <8 x i16> [[TMP17]] to <4 x i32>641; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <4 x i32> [[TMP18]], zeroinitializer642; CHECK-NEXT: [[TMP20:%.*]] = sext <4 x i1> [[TMP19]] to <4 x i32>643; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP20]], [[TMP21]]644; CHECK-NEXT: [[TMP4:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])645; CHECK-NEXT: store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8646; CHECK-NEXT: ret <4 x i32> [[TMP4]]647;648 %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)649 ret <4 x i32> %1650}651 652define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4, i8 %x3) sanitize_memory {653; CHECK-LABEL: define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(654; CHECK-SAME: <4 x i32> [[X0:%.*]], <4 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <4 x i32> [[X4:%.*]], i8 [[X3:%.*]]) #[[ATTR1]] {655; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8656; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8657; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8658; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 56), align 8659; CHECK-NEXT: [[TMP5:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8660; CHECK-NEXT: call void @llvm.donothing()661; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0662; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]663; CHECK: [[BB6]]:664; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]665; CHECK-NEXT: unreachable666; CHECK: [[BB7]]:667; CHECK-NEXT: [[X2:%.*]] = load <4 x i32>, ptr [[X2P]], align 16668; CHECK-NEXT: [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64669; CHECK-NEXT: [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080670; CHECK-NEXT: [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr671; CHECK-NEXT: [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP10]], align 16672; CHECK-NEXT: [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>673; CHECK-NEXT: [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>674; CHECK-NEXT: [[TMP31:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>675; CHECK-NEXT: [[TMP32:%.*]] = bitcast <4 x i32> [[_MSLD]] to <8 x i16>676; CHECK-NEXT: [[TMP33:%.*]] = icmp ne <8 x i16> [[TMP31]], zeroinitializer677; CHECK-NEXT: [[TMP34:%.*]] = icmp ne <8 x i16> [[TMP32]], zeroinitializer678; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <8 x i16> [[TMP29]], zeroinitializer679; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <8 x i16> [[TMP30]], zeroinitializer680; CHECK-NEXT: [[TMP54:%.*]] = and <8 x i1> [[TMP33]], [[TMP34]]681; CHECK-NEXT: [[TMP55:%.*]] = and <8 x i1> [[TMP35]], [[TMP34]]682; CHECK-NEXT: [[TMP56:%.*]] = and <8 x i1> [[TMP33]], [[TMP36]]683; CHECK-NEXT: [[TMP57:%.*]] = or <8 x i1> [[TMP54]], [[TMP55]]684; CHECK-NEXT: [[TMP58:%.*]] = or <8 x i1> [[TMP57]], [[TMP56]]685; CHECK-NEXT: [[TMP59:%.*]] = sext <8 x i1> [[TMP58]] to <8 x i16>686; CHECK-NEXT: [[TMP60:%.*]] = bitcast <8 x i16> [[TMP59]] to <4 x i32>687; CHECK-NEXT: [[TMP61:%.*]] = icmp ne <4 x i32> [[TMP60]], zeroinitializer688; CHECK-NEXT: [[TMP62:%.*]] = sext <4 x i1> [[TMP61]] to <4 x i32>689; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP62]], [[TMP2]]690; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])691; CHECK-NEXT: [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>692; CHECK-NEXT: [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>693; CHECK-NEXT: [[_MSPROP2:%.*]] = shufflevector <8 x i1> [[TMP12]], <8 x i1> [[TMP12]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>694; CHECK-NEXT: [[EXTRACT:%.*]] = shufflevector <8 x i1> [[TMP13]], <8 x i1> [[TMP13]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>695; CHECK-NEXT: [[TMP14:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[_MSPROP1]], <4 x i32> [[TMP2]]696; CHECK-NEXT: [[TMP15:%.*]] = xor <4 x i32> [[TMP11]], [[X0]]697; CHECK-NEXT: [[TMP16:%.*]] = or <4 x i32> [[TMP15]], [[_MSPROP1]]698; CHECK-NEXT: [[TMP17:%.*]] = or <4 x i32> [[TMP16]], [[TMP2]]699; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <4 x i1> [[_MSPROP2]], <4 x i32> [[TMP17]], <4 x i32> [[TMP14]]700; CHECK-NEXT: [[TMP18:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[TMP11]], <4 x i32> [[X0]]701; CHECK-NEXT: [[TMP37:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>702; CHECK-NEXT: [[TMP38:%.*]] = bitcast <4 x i32> [[X4]] to <8 x i16>703; CHECK-NEXT: [[TMP39:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>704; CHECK-NEXT: [[TMP40:%.*]] = bitcast <4 x i32> [[TMP5]] to <8 x i16>705; CHECK-NEXT: [[TMP41:%.*]] = icmp ne <8 x i16> [[TMP39]], zeroinitializer706; CHECK-NEXT: [[TMP42:%.*]] = icmp ne <8 x i16> [[TMP40]], zeroinitializer707; CHECK-NEXT: [[TMP43:%.*]] = icmp ne <8 x i16> [[TMP37]], zeroinitializer708; CHECK-NEXT: [[TMP44:%.*]] = icmp ne <8 x i16> [[TMP38]], zeroinitializer709; CHECK-NEXT: [[TMP45:%.*]] = and <8 x i1> [[TMP41]], [[TMP42]]710; CHECK-NEXT: [[TMP46:%.*]] = and <8 x i1> [[TMP43]], [[TMP42]]711; CHECK-NEXT: [[TMP47:%.*]] = and <8 x i1> [[TMP41]], [[TMP44]]712; CHECK-NEXT: [[TMP48:%.*]] = or <8 x i1> [[TMP45]], [[TMP46]]713; CHECK-NEXT: [[TMP49:%.*]] = or <8 x i1> [[TMP48]], [[TMP47]]714; CHECK-NEXT: [[TMP50:%.*]] = sext <8 x i1> [[TMP49]] to <8 x i16>715; CHECK-NEXT: [[TMP51:%.*]] = bitcast <8 x i16> [[TMP50]] to <4 x i32>716; CHECK-NEXT: [[TMP52:%.*]] = icmp ne <4 x i32> [[TMP51]], zeroinitializer717; CHECK-NEXT: [[TMP53:%.*]] = sext <4 x i1> [[TMP52]] to <4 x i32>718; CHECK-NEXT: [[_MSPROP4:%.*]] = or <4 x i32> [[TMP53]], [[TMP2]]719; CHECK-NEXT: [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]])720; CHECK-NEXT: [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>721; CHECK-NEXT: [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>722; CHECK-NEXT: [[_MSPROP5:%.*]] = shufflevector <8 x i1> [[TMP20]], <8 x i1> [[TMP20]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>723; CHECK-NEXT: [[EXTRACT1:%.*]] = shufflevector <8 x i1> [[TMP21]], <8 x i1> [[TMP21]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>724; CHECK-NEXT: [[TMP22:%.*]] = select <4 x i1> [[EXTRACT1]], <4 x i32> [[_MSPROP4]], <4 x i32> zeroinitializer725; CHECK-NEXT: [[TMP23:%.*]] = xor <4 x i32> [[TMP19]], zeroinitializer726; CHECK-NEXT: [[TMP24:%.*]] = or <4 x i32> [[TMP23]], [[_MSPROP4]]727; CHECK-NEXT: [[TMP25:%.*]] = or <4 x i32> [[TMP24]], zeroinitializer728; CHECK-NEXT: [[_MSPROP_SELECT6:%.*]] = select <4 x i1> [[_MSPROP5]], <4 x i32> [[TMP25]], <4 x i32> [[TMP22]]729; CHECK-NEXT: [[TMP26:%.*]] = select <4 x i1> [[EXTRACT1]], <4 x i32> [[TMP19]], <4 x i32> zeroinitializer730; CHECK-NEXT: [[TMP27:%.*]] = insertvalue { <4 x i32>, <4 x i32> } { <4 x i32> splat (i32 -1), <4 x i32> splat (i32 -1) }, <4 x i32> [[_MSPROP_SELECT]], 0731; CHECK-NEXT: [[RES1:%.*]] = insertvalue { <4 x i32>, <4 x i32> } poison, <4 x i32> [[TMP18]], 0732; CHECK-NEXT: [[TMP28:%.*]] = insertvalue { <4 x i32>, <4 x i32> } [[TMP27]], <4 x i32> [[_MSPROP_SELECT6]], 1733; CHECK-NEXT: [[RES2:%.*]] = insertvalue { <4 x i32>, <4 x i32> } [[RES1]], <4 x i32> [[TMP26]], 1734; CHECK-NEXT: store { <4 x i32>, <4 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8735; CHECK-NEXT: ret { <4 x i32>, <4 x i32> } [[RES2]]736;737 %x2 = load <4 x i32>, ptr %x2p738 %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)739 %2 = bitcast i8 %x3 to <8 x i1>740 %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>741 %3 = select <4 x i1> %extract, <4 x i32> %1, <4 x i32> %x0742 %4 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4)743 %5 = bitcast i8 %x3 to <8 x i1>744 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>745 %6 = select <4 x i1> %extract1, <4 x i32> %4, <4 x i32> zeroinitializer746 %res1 = insertvalue { <4 x i32>, <4 x i32> } poison, <4 x i32> %3, 0747 %res2 = insertvalue { <4 x i32>, <4 x i32> } %res1, <4 x i32> %6, 1748 ret { <4 x i32>, <4 x i32> } %res2749}750 751declare <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32>, <8 x i32>, <8 x i32>)752 753define <8 x i32>@test_int_x86_avx512_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) sanitize_memory {754; CHECK-LABEL: define <8 x i32> @test_int_x86_avx512_vpdpwssds_256(755; CHECK-SAME: <8 x i32> [[X0:%.*]], <8 x i32> [[X1:%.*]], <8 x i32> [[X2:%.*]]) #[[ATTR1]] {756; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8757; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8758; CHECK-NEXT: [[TMP21:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8759; CHECK-NEXT: call void @llvm.donothing()760; CHECK-NEXT: [[TMP22:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>761; CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>762; CHECK-NEXT: [[TMP6:%.*]] = bitcast <8 x i32> [[TMP2]] to <16 x i16>763; CHECK-NEXT: [[TMP7:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>764; CHECK-NEXT: [[TMP8:%.*]] = icmp ne <16 x i16> [[TMP6]], zeroinitializer765; CHECK-NEXT: [[TMP9:%.*]] = icmp ne <16 x i16> [[TMP7]], zeroinitializer766; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <16 x i16> [[TMP22]], zeroinitializer767; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <16 x i16> [[TMP5]], zeroinitializer768; CHECK-NEXT: [[TMP12:%.*]] = and <16 x i1> [[TMP8]], [[TMP9]]769; CHECK-NEXT: [[TMP13:%.*]] = and <16 x i1> [[TMP10]], [[TMP9]]770; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP8]], [[TMP11]]771; CHECK-NEXT: [[TMP15:%.*]] = or <16 x i1> [[TMP12]], [[TMP13]]772; CHECK-NEXT: [[TMP16:%.*]] = or <16 x i1> [[TMP15]], [[TMP14]]773; CHECK-NEXT: [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i16>774; CHECK-NEXT: [[TMP18:%.*]] = bitcast <16 x i16> [[TMP17]] to <8 x i32>775; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <8 x i32> [[TMP18]], zeroinitializer776; CHECK-NEXT: [[TMP20:%.*]] = sext <8 x i1> [[TMP19]] to <8 x i32>777; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP20]], [[TMP21]]778; CHECK-NEXT: [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])779; CHECK-NEXT: store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8780; CHECK-NEXT: ret <8 x i32> [[TMP4]]781;782 %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)783 ret <8 x i32> %1784}785 786define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4, i8 %x3) sanitize_memory {787; CHECK-LABEL: define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(788; CHECK-SAME: <8 x i32> [[X0:%.*]], <8 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <8 x i32> [[X4:%.*]], i8 [[X3:%.*]]) #[[ATTR1]] {789; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8790; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8791; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8792; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 104), align 8793; CHECK-NEXT: [[TMP5:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8794; CHECK-NEXT: call void @llvm.donothing()795; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0796; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]797; CHECK: [[BB6]]:798; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]799; CHECK-NEXT: unreachable800; CHECK: [[BB7]]:801; CHECK-NEXT: [[X2:%.*]] = load <8 x i32>, ptr [[X2P]], align 32802; CHECK-NEXT: [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64803; CHECK-NEXT: [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080804; CHECK-NEXT: [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr805; CHECK-NEXT: [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP10]], align 32806; CHECK-NEXT: [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>807; CHECK-NEXT: [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>808; CHECK-NEXT: [[TMP31:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>809; CHECK-NEXT: [[TMP32:%.*]] = bitcast <8 x i32> [[_MSLD]] to <16 x i16>810; CHECK-NEXT: [[TMP33:%.*]] = icmp ne <16 x i16> [[TMP31]], zeroinitializer811; CHECK-NEXT: [[TMP34:%.*]] = icmp ne <16 x i16> [[TMP32]], zeroinitializer812; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <16 x i16> [[TMP29]], zeroinitializer813; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <16 x i16> [[TMP30]], zeroinitializer814; CHECK-NEXT: [[TMP54:%.*]] = and <16 x i1> [[TMP33]], [[TMP34]]815; CHECK-NEXT: [[TMP55:%.*]] = and <16 x i1> [[TMP35]], [[TMP34]]816; CHECK-NEXT: [[TMP56:%.*]] = and <16 x i1> [[TMP33]], [[TMP36]]817; CHECK-NEXT: [[TMP57:%.*]] = or <16 x i1> [[TMP54]], [[TMP55]]818; CHECK-NEXT: [[TMP58:%.*]] = or <16 x i1> [[TMP57]], [[TMP56]]819; CHECK-NEXT: [[TMP59:%.*]] = sext <16 x i1> [[TMP58]] to <16 x i16>820; CHECK-NEXT: [[TMP60:%.*]] = bitcast <16 x i16> [[TMP59]] to <8 x i32>821; CHECK-NEXT: [[TMP61:%.*]] = icmp ne <8 x i32> [[TMP60]], zeroinitializer822; CHECK-NEXT: [[TMP62:%.*]] = sext <8 x i1> [[TMP61]] to <8 x i32>823; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP62]], [[TMP2]]824; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])825; CHECK-NEXT: [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>826; CHECK-NEXT: [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>827; CHECK-NEXT: [[TMP14:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP2]]828; CHECK-NEXT: [[TMP15:%.*]] = xor <8 x i32> [[TMP11]], [[X0]]829; CHECK-NEXT: [[TMP16:%.*]] = or <8 x i32> [[TMP15]], [[_MSPROP1]]830; CHECK-NEXT: [[TMP17:%.*]] = or <8 x i32> [[TMP16]], [[TMP2]]831; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <8 x i1> [[TMP12]], <8 x i32> [[TMP17]], <8 x i32> [[TMP14]]832; CHECK-NEXT: [[TMP18:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[TMP11]], <8 x i32> [[X0]]833; CHECK-NEXT: [[TMP37:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>834; CHECK-NEXT: [[TMP38:%.*]] = bitcast <8 x i32> [[X4]] to <16 x i16>835; CHECK-NEXT: [[TMP39:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>836; CHECK-NEXT: [[TMP40:%.*]] = bitcast <8 x i32> [[TMP5]] to <16 x i16>837; CHECK-NEXT: [[TMP41:%.*]] = icmp ne <16 x i16> [[TMP39]], zeroinitializer838; CHECK-NEXT: [[TMP42:%.*]] = icmp ne <16 x i16> [[TMP40]], zeroinitializer839; CHECK-NEXT: [[TMP43:%.*]] = icmp ne <16 x i16> [[TMP37]], zeroinitializer840; CHECK-NEXT: [[TMP44:%.*]] = icmp ne <16 x i16> [[TMP38]], zeroinitializer841; CHECK-NEXT: [[TMP45:%.*]] = and <16 x i1> [[TMP41]], [[TMP42]]842; CHECK-NEXT: [[TMP46:%.*]] = and <16 x i1> [[TMP43]], [[TMP42]]843; CHECK-NEXT: [[TMP47:%.*]] = and <16 x i1> [[TMP41]], [[TMP44]]844; CHECK-NEXT: [[TMP48:%.*]] = or <16 x i1> [[TMP45]], [[TMP46]]845; CHECK-NEXT: [[TMP49:%.*]] = or <16 x i1> [[TMP48]], [[TMP47]]846; CHECK-NEXT: [[TMP50:%.*]] = sext <16 x i1> [[TMP49]] to <16 x i16>847; CHECK-NEXT: [[TMP51:%.*]] = bitcast <16 x i16> [[TMP50]] to <8 x i32>848; CHECK-NEXT: [[TMP52:%.*]] = icmp ne <8 x i32> [[TMP51]], zeroinitializer849; CHECK-NEXT: [[TMP53:%.*]] = sext <8 x i1> [[TMP52]] to <8 x i32>850; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[TMP53]], [[TMP2]]851; CHECK-NEXT: [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]])852; CHECK-NEXT: [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>853; CHECK-NEXT: [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>854; CHECK-NEXT: [[TMP22:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[_MSPROP3]], <8 x i32> zeroinitializer855; CHECK-NEXT: [[TMP23:%.*]] = xor <8 x i32> [[TMP19]], zeroinitializer856; CHECK-NEXT: [[TMP24:%.*]] = or <8 x i32> [[TMP23]], [[_MSPROP3]]857; CHECK-NEXT: [[TMP25:%.*]] = or <8 x i32> [[TMP24]], zeroinitializer858; CHECK-NEXT: [[_MSPROP_SELECT4:%.*]] = select <8 x i1> [[TMP20]], <8 x i32> [[TMP25]], <8 x i32> [[TMP22]]859; CHECK-NEXT: [[TMP26:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[TMP19]], <8 x i32> zeroinitializer860; CHECK-NEXT: [[TMP27:%.*]] = insertvalue { <8 x i32>, <8 x i32> } { <8 x i32> splat (i32 -1), <8 x i32> splat (i32 -1) }, <8 x i32> [[_MSPROP_SELECT]], 0861; CHECK-NEXT: [[RES1:%.*]] = insertvalue { <8 x i32>, <8 x i32> } poison, <8 x i32> [[TMP18]], 0862; CHECK-NEXT: [[TMP28:%.*]] = insertvalue { <8 x i32>, <8 x i32> } [[TMP27]], <8 x i32> [[_MSPROP_SELECT4]], 1863; CHECK-NEXT: [[RES2:%.*]] = insertvalue { <8 x i32>, <8 x i32> } [[RES1]], <8 x i32> [[TMP26]], 1864; CHECK-NEXT: store { <8 x i32>, <8 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8865; CHECK-NEXT: ret { <8 x i32>, <8 x i32> } [[RES2]]866;867 %x2 = load <8 x i32>, ptr %x2p868 %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)869 %2 = bitcast i8 %x3 to <8 x i1>870 %3 = select <8 x i1> %2, <8 x i32> %1, <8 x i32> %x0871 %4 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4)872 %5 = bitcast i8 %x3 to <8 x i1>873 %6 = select <8 x i1> %5, <8 x i32> %4, <8 x i32> zeroinitializer874 %res1 = insertvalue { <8 x i32>, <8 x i32> } poison, <8 x i32> %3, 0875 %res2 = insertvalue { <8 x i32>, <8 x i32> } %res1, <8 x i32> %6, 1876 ret { <8 x i32>, <8 x i32> } %res2877}878 879declare <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32>, <4 x i32>, <4 x i32>)880 881define <4 x i32>@test_int_x86_avx512_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p) sanitize_memory {882; CHECK-LABEL: define <4 x i32> @test_int_x86_avx512_vpdpwssds_128(883; CHECK-SAME: <4 x i32> [[X0:%.*]], <4 x i32> [[X1:%.*]], ptr [[X2P:%.*]]) #[[ATTR1]] {884; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8885; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8886; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8887; CHECK-NEXT: call void @llvm.donothing()888; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0889; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB4:.*]], label %[[BB5:.*]], !prof [[PROF1]]890; CHECK: [[BB4]]:891; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]892; CHECK-NEXT: unreachable893; CHECK: [[BB5]]:894; CHECK-NEXT: [[X2:%.*]] = load <4 x i32>, ptr [[X2P]], align 16895; CHECK-NEXT: [[TMP6:%.*]] = ptrtoint ptr [[X2P]] to i64896; CHECK-NEXT: [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080897; CHECK-NEXT: [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr898; CHECK-NEXT: [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP8]], align 16899; CHECK-NEXT: [[TMP26:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>900; CHECK-NEXT: [[TMP10:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>901; CHECK-NEXT: [[TMP11:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>902; CHECK-NEXT: [[TMP12:%.*]] = bitcast <4 x i32> [[_MSLD]] to <8 x i16>903; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <8 x i16> [[TMP11]], zeroinitializer904; CHECK-NEXT: [[TMP14:%.*]] = icmp ne <8 x i16> [[TMP12]], zeroinitializer905; CHECK-NEXT: [[TMP15:%.*]] = icmp ne <8 x i16> [[TMP26]], zeroinitializer906; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <8 x i16> [[TMP10]], zeroinitializer907; CHECK-NEXT: [[TMP17:%.*]] = and <8 x i1> [[TMP13]], [[TMP14]]908; CHECK-NEXT: [[TMP18:%.*]] = and <8 x i1> [[TMP15]], [[TMP14]]909; CHECK-NEXT: [[TMP19:%.*]] = and <8 x i1> [[TMP13]], [[TMP16]]910; CHECK-NEXT: [[TMP20:%.*]] = or <8 x i1> [[TMP17]], [[TMP18]]911; CHECK-NEXT: [[TMP21:%.*]] = or <8 x i1> [[TMP20]], [[TMP19]]912; CHECK-NEXT: [[TMP22:%.*]] = sext <8 x i1> [[TMP21]] to <8 x i16>913; CHECK-NEXT: [[TMP23:%.*]] = bitcast <8 x i16> [[TMP22]] to <4 x i32>914; CHECK-NEXT: [[TMP24:%.*]] = icmp ne <4 x i32> [[TMP23]], zeroinitializer915; CHECK-NEXT: [[TMP25:%.*]] = sext <4 x i1> [[TMP24]] to <4 x i32>916; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP25]], [[TMP4]]917; CHECK-NEXT: [[TMP9:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])918; CHECK-NEXT: store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8919; CHECK-NEXT: ret <4 x i32> [[TMP9]]920;921 %x2 = load <4 x i32>, ptr %x2p922 %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)923 ret <4 x i32> %1924}925 926define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4, i8 %x3) sanitize_memory {927; CHECK-LABEL: define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(928; CHECK-SAME: <4 x i32> [[X0:%.*]], <4 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <4 x i32> [[X4:%.*]], i8 [[X3:%.*]]) #[[ATTR1]] {929; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8930; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8931; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8932; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 56), align 8933; CHECK-NEXT: [[TMP5:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8934; CHECK-NEXT: call void @llvm.donothing()935; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0936; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]937; CHECK: [[BB6]]:938; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]939; CHECK-NEXT: unreachable940; CHECK: [[BB7]]:941; CHECK-NEXT: [[X2:%.*]] = load <4 x i32>, ptr [[X2P]], align 16942; CHECK-NEXT: [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64943; CHECK-NEXT: [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080944; CHECK-NEXT: [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr945; CHECK-NEXT: [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP10]], align 16946; CHECK-NEXT: [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>947; CHECK-NEXT: [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>948; CHECK-NEXT: [[TMP31:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>949; CHECK-NEXT: [[TMP32:%.*]] = bitcast <4 x i32> [[_MSLD]] to <8 x i16>950; CHECK-NEXT: [[TMP33:%.*]] = icmp ne <8 x i16> [[TMP31]], zeroinitializer951; CHECK-NEXT: [[TMP34:%.*]] = icmp ne <8 x i16> [[TMP32]], zeroinitializer952; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <8 x i16> [[TMP29]], zeroinitializer953; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <8 x i16> [[TMP30]], zeroinitializer954; CHECK-NEXT: [[TMP54:%.*]] = and <8 x i1> [[TMP33]], [[TMP34]]955; CHECK-NEXT: [[TMP55:%.*]] = and <8 x i1> [[TMP35]], [[TMP34]]956; CHECK-NEXT: [[TMP56:%.*]] = and <8 x i1> [[TMP33]], [[TMP36]]957; CHECK-NEXT: [[TMP57:%.*]] = or <8 x i1> [[TMP54]], [[TMP55]]958; CHECK-NEXT: [[TMP58:%.*]] = or <8 x i1> [[TMP57]], [[TMP56]]959; CHECK-NEXT: [[TMP59:%.*]] = sext <8 x i1> [[TMP58]] to <8 x i16>960; CHECK-NEXT: [[TMP60:%.*]] = bitcast <8 x i16> [[TMP59]] to <4 x i32>961; CHECK-NEXT: [[TMP61:%.*]] = icmp ne <4 x i32> [[TMP60]], zeroinitializer962; CHECK-NEXT: [[TMP62:%.*]] = sext <4 x i1> [[TMP61]] to <4 x i32>963; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP62]], [[TMP2]]964; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])965; CHECK-NEXT: [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>966; CHECK-NEXT: [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>967; CHECK-NEXT: [[_MSPROP2:%.*]] = shufflevector <8 x i1> [[TMP12]], <8 x i1> [[TMP12]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>968; CHECK-NEXT: [[EXTRACT:%.*]] = shufflevector <8 x i1> [[TMP13]], <8 x i1> [[TMP13]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>969; CHECK-NEXT: [[TMP14:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[_MSPROP1]], <4 x i32> [[TMP2]]970; CHECK-NEXT: [[TMP15:%.*]] = xor <4 x i32> [[TMP11]], [[X0]]971; CHECK-NEXT: [[TMP16:%.*]] = or <4 x i32> [[TMP15]], [[_MSPROP1]]972; CHECK-NEXT: [[TMP17:%.*]] = or <4 x i32> [[TMP16]], [[TMP2]]973; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <4 x i1> [[_MSPROP2]], <4 x i32> [[TMP17]], <4 x i32> [[TMP14]]974; CHECK-NEXT: [[TMP18:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[TMP11]], <4 x i32> [[X0]]975; CHECK-NEXT: [[TMP37:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>976; CHECK-NEXT: [[TMP38:%.*]] = bitcast <4 x i32> [[X4]] to <8 x i16>977; CHECK-NEXT: [[TMP39:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>978; CHECK-NEXT: [[TMP40:%.*]] = bitcast <4 x i32> [[TMP5]] to <8 x i16>979; CHECK-NEXT: [[TMP41:%.*]] = icmp ne <8 x i16> [[TMP39]], zeroinitializer980; CHECK-NEXT: [[TMP42:%.*]] = icmp ne <8 x i16> [[TMP40]], zeroinitializer981; CHECK-NEXT: [[TMP43:%.*]] = icmp ne <8 x i16> [[TMP37]], zeroinitializer982; CHECK-NEXT: [[TMP44:%.*]] = icmp ne <8 x i16> [[TMP38]], zeroinitializer983; CHECK-NEXT: [[TMP45:%.*]] = and <8 x i1> [[TMP41]], [[TMP42]]984; CHECK-NEXT: [[TMP46:%.*]] = and <8 x i1> [[TMP43]], [[TMP42]]985; CHECK-NEXT: [[TMP47:%.*]] = and <8 x i1> [[TMP41]], [[TMP44]]986; CHECK-NEXT: [[TMP48:%.*]] = or <8 x i1> [[TMP45]], [[TMP46]]987; CHECK-NEXT: [[TMP49:%.*]] = or <8 x i1> [[TMP48]], [[TMP47]]988; CHECK-NEXT: [[TMP50:%.*]] = sext <8 x i1> [[TMP49]] to <8 x i16>989; CHECK-NEXT: [[TMP51:%.*]] = bitcast <8 x i16> [[TMP50]] to <4 x i32>990; CHECK-NEXT: [[TMP52:%.*]] = icmp ne <4 x i32> [[TMP51]], zeroinitializer991; CHECK-NEXT: [[TMP53:%.*]] = sext <4 x i1> [[TMP52]] to <4 x i32>992; CHECK-NEXT: [[_MSPROP4:%.*]] = or <4 x i32> [[TMP53]], [[TMP2]]993; CHECK-NEXT: [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]])994; CHECK-NEXT: [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>995; CHECK-NEXT: [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>996; CHECK-NEXT: [[_MSPROP5:%.*]] = shufflevector <8 x i1> [[TMP20]], <8 x i1> [[TMP20]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>997; CHECK-NEXT: [[EXTRACT1:%.*]] = shufflevector <8 x i1> [[TMP21]], <8 x i1> [[TMP21]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>998; CHECK-NEXT: [[TMP22:%.*]] = select <4 x i1> [[EXTRACT1]], <4 x i32> [[_MSPROP4]], <4 x i32> zeroinitializer999; CHECK-NEXT: [[TMP23:%.*]] = xor <4 x i32> [[TMP19]], zeroinitializer1000; CHECK-NEXT: [[TMP24:%.*]] = or <4 x i32> [[TMP23]], [[_MSPROP4]]1001; CHECK-NEXT: [[TMP25:%.*]] = or <4 x i32> [[TMP24]], zeroinitializer1002; CHECK-NEXT: [[_MSPROP_SELECT6:%.*]] = select <4 x i1> [[_MSPROP5]], <4 x i32> [[TMP25]], <4 x i32> [[TMP22]]1003; CHECK-NEXT: [[TMP26:%.*]] = select <4 x i1> [[EXTRACT1]], <4 x i32> [[TMP19]], <4 x i32> zeroinitializer1004; CHECK-NEXT: [[TMP27:%.*]] = insertvalue { <4 x i32>, <4 x i32> } { <4 x i32> splat (i32 -1), <4 x i32> splat (i32 -1) }, <4 x i32> [[_MSPROP_SELECT]], 01005; CHECK-NEXT: [[RES1:%.*]] = insertvalue { <4 x i32>, <4 x i32> } poison, <4 x i32> [[TMP18]], 01006; CHECK-NEXT: [[TMP28:%.*]] = insertvalue { <4 x i32>, <4 x i32> } [[TMP27]], <4 x i32> [[_MSPROP_SELECT6]], 11007; CHECK-NEXT: [[RES2:%.*]] = insertvalue { <4 x i32>, <4 x i32> } [[RES1]], <4 x i32> [[TMP26]], 11008; CHECK-NEXT: store { <4 x i32>, <4 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 81009; CHECK-NEXT: ret { <4 x i32>, <4 x i32> } [[RES2]]1010;1011 %x2 = load <4 x i32>, ptr %x2p1012 %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)1013 %2 = bitcast i8 %x3 to <8 x i1>1014 %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>1015 %3 = select <4 x i1> %extract, <4 x i32> %1, <4 x i32> %x01016 %4 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4)1017 %5 = bitcast i8 %x3 to <8 x i1>1018 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>1019 %6 = select <4 x i1> %extract1, <4 x i32> %4, <4 x i32> zeroinitializer1020 %res1 = insertvalue { <4 x i32>, <4 x i32> } poison, <4 x i32> %3, 01021 %res2 = insertvalue { <4 x i32>, <4 x i32> } %res1, <4 x i32> %6, 11022 ret { <4 x i32>, <4 x i32> } %res21023}1024;.1025; CHECK: [[PROF1]] = !{!"branch_weights", i32 1, i32 1048575}1026;.1027