340 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 62; RUN: opt -S -passes=msan -mattr=+avx512bf16 < %s | FileCheck %s3;4; Forked from llvm/test/CodeGen/X86/avx512bf16-intrinsics.ll5;6; Strictly handled:7; - llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float> %A, <16 x float> %B)8; - llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> %A)9;10; Heuristically handled: (none)11 12target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"13target triple = "x86_64-unknown-linux-gnu"14 15declare <32 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float>, <16 x float>) #316 17define <8 x i64> @test_mm512_cvtne2ps2bf16_512(<16 x float> %A, <16 x float> %B) local_unnamed_addr #2 sanitize_memory {18; CHECK-LABEL: define <8 x i64> @test_mm512_cvtne2ps2bf16_512(19; CHECK-SAME: <16 x float> [[A:%.*]], <16 x float> [[B:%.*]]) local_unnamed_addr #[[ATTR1:[0-9]+]] {20; CHECK-NEXT: [[ENTRY:.*:]]21; CHECK-NEXT: [[TMP0:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 822; CHECK-NEXT: [[TMP1:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 823; CHECK-NEXT: call void @llvm.donothing()24; CHECK-NEXT: [[TMP2:%.*]] = bitcast <16 x i32> [[TMP0]] to i51225; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i512 [[TMP2]], 026; CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x i32> [[TMP1]] to i51227; CHECK-NEXT: [[_MSCMP1:%.*]] = icmp ne i512 [[TMP3]], 028; CHECK-NEXT: [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]29; CHECK-NEXT: br i1 [[_MSOR]], label %[[BB4:.*]], label %[[BB5:.*]], !prof [[PROF1:![0-9]+]]30; CHECK: [[BB4]]:31; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4:[0-9]+]]32; CHECK-NEXT: unreachable33; CHECK: [[BB5]]:34; CHECK-NEXT: [[TMP6:%.*]] = tail call <32 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float> [[A]], <16 x float> [[B]])35; CHECK-NEXT: [[TMP7:%.*]] = bitcast <32 x bfloat> [[TMP6]] to <8 x i64>36; CHECK-NEXT: store <8 x i64> zeroinitializer, ptr @__msan_retval_tls, align 837; CHECK-NEXT: ret <8 x i64> [[TMP7]]38;39entry:40 %0 = tail call <32 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float> %A, <16 x float> %B) #441 %1 = bitcast <32 x bfloat> %0 to <8 x i64>42 ret <8 x i64> %143}44 45define <8 x i64> @test_mm512_maskz_cvtne2ps2bf16_512(<16 x float> %A, <16 x float> %B, i32 %U) local_unnamed_addr #2 sanitize_memory {46; CHECK-LABEL: define <8 x i64> @test_mm512_maskz_cvtne2ps2bf16_512(47; CHECK-SAME: <16 x float> [[A:%.*]], <16 x float> [[B:%.*]], i32 [[U:%.*]]) local_unnamed_addr #[[ATTR1]] {48; CHECK-NEXT: [[ENTRY:.*:]]49; CHECK-NEXT: [[TMP0:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 850; CHECK-NEXT: [[TMP1:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 851; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 852; CHECK-NEXT: call void @llvm.donothing()53; CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x i32> [[TMP0]] to i51254; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 055; CHECK-NEXT: [[TMP4:%.*]] = bitcast <16 x i32> [[TMP1]] to i51256; CHECK-NEXT: [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 057; CHECK-NEXT: [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]58; CHECK-NEXT: br i1 [[_MSOR]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]]59; CHECK: [[BB5]]:60; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]61; CHECK-NEXT: unreachable62; CHECK: [[BB6]]:63; CHECK-NEXT: [[TMP7:%.*]] = tail call <32 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float> [[A]], <16 x float> [[B]])64; CHECK-NEXT: [[TMP8:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>65; CHECK-NEXT: [[TMP9:%.*]] = bitcast i32 [[U]] to <32 x i1>66; CHECK-NEXT: [[TMP10:%.*]] = select <32 x i1> [[TMP9]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer67; CHECK-NEXT: [[TMP11:%.*]] = bitcast <32 x bfloat> [[TMP7]] to <32 x i16>68; CHECK-NEXT: [[TMP12:%.*]] = xor <32 x i16> [[TMP11]], zeroinitializer69; CHECK-NEXT: [[TMP13:%.*]] = or <32 x i16> [[TMP12]], zeroinitializer70; CHECK-NEXT: [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer71; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP8]], <32 x i16> [[TMP14]], <32 x i16> [[TMP10]]72; CHECK-NEXT: [[TMP15:%.*]] = select <32 x i1> [[TMP9]], <32 x bfloat> [[TMP7]], <32 x bfloat> zeroinitializer73; CHECK-NEXT: [[TMP16:%.*]] = bitcast <32 x i16> [[_MSPROP_SELECT]] to <8 x i64>74; CHECK-NEXT: [[TMP17:%.*]] = bitcast <32 x bfloat> [[TMP15]] to <8 x i64>75; CHECK-NEXT: store <8 x i64> [[TMP16]], ptr @__msan_retval_tls, align 876; CHECK-NEXT: ret <8 x i64> [[TMP17]]77;78entry:79 %0 = tail call <32 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float> %A, <16 x float> %B) #480 %1 = bitcast i32 %U to <32 x i1>81 %2 = select <32 x i1> %1, <32 x bfloat> %0, <32 x bfloat> zeroinitializer82 %3 = bitcast <32 x bfloat> %2 to <8 x i64>83 ret <8 x i64> %384}85 86define <8 x i64> @test_mm512_mask_cvtne2ps2bf16_512(<8 x i64> %C, i32 %U, <16 x float> %A, <16 x float> %B) local_unnamed_addr #2 sanitize_memory {87; CHECK-LABEL: define <8 x i64> @test_mm512_mask_cvtne2ps2bf16_512(88; CHECK-SAME: <8 x i64> [[C:%.*]], i32 [[U:%.*]], <16 x float> [[A:%.*]], <16 x float> [[B:%.*]]) local_unnamed_addr #[[ATTR1]] {89; CHECK-NEXT: [[ENTRY:.*:]]90; CHECK-NEXT: [[TMP0:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 891; CHECK-NEXT: [[TMP1:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 892; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i64>, ptr @__msan_param_tls, align 893; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 894; CHECK-NEXT: call void @llvm.donothing()95; CHECK-NEXT: [[TMP4:%.*]] = bitcast <16 x i32> [[TMP0]] to i51296; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 097; CHECK-NEXT: [[TMP5:%.*]] = bitcast <16 x i32> [[TMP1]] to i51298; CHECK-NEXT: [[_MSCMP1:%.*]] = icmp ne i512 [[TMP5]], 099; CHECK-NEXT: [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]100; CHECK-NEXT: br i1 [[_MSOR]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]101; CHECK: [[BB6]]:102; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]103; CHECK-NEXT: unreachable104; CHECK: [[BB7]]:105; CHECK-NEXT: [[TMP8:%.*]] = tail call <32 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float> [[A]], <16 x float> [[B]])106; CHECK-NEXT: [[TMP9:%.*]] = bitcast <8 x i64> [[TMP2]] to <32 x i16>107; CHECK-NEXT: [[TMP10:%.*]] = bitcast <8 x i64> [[C]] to <32 x bfloat>108; CHECK-NEXT: [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>109; CHECK-NEXT: [[TMP12:%.*]] = bitcast i32 [[U]] to <32 x i1>110; CHECK-NEXT: [[TMP13:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> zeroinitializer, <32 x i16> [[TMP9]]111; CHECK-NEXT: [[TMP14:%.*]] = bitcast <32 x bfloat> [[TMP8]] to <32 x i16>112; CHECK-NEXT: [[TMP15:%.*]] = bitcast <32 x bfloat> [[TMP10]] to <32 x i16>113; CHECK-NEXT: [[TMP16:%.*]] = xor <32 x i16> [[TMP14]], [[TMP15]]114; CHECK-NEXT: [[TMP17:%.*]] = or <32 x i16> [[TMP16]], zeroinitializer115; CHECK-NEXT: [[TMP18:%.*]] = or <32 x i16> [[TMP17]], [[TMP9]]116; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP18]], <32 x i16> [[TMP13]]117; CHECK-NEXT: [[TMP19:%.*]] = select <32 x i1> [[TMP12]], <32 x bfloat> [[TMP8]], <32 x bfloat> [[TMP10]]118; CHECK-NEXT: [[TMP20:%.*]] = bitcast <32 x i16> [[_MSPROP_SELECT]] to <8 x i64>119; CHECK-NEXT: [[TMP21:%.*]] = bitcast <32 x bfloat> [[TMP19]] to <8 x i64>120; CHECK-NEXT: store <8 x i64> [[TMP20]], ptr @__msan_retval_tls, align 8121; CHECK-NEXT: ret <8 x i64> [[TMP21]]122;123entry:124 %0 = tail call <32 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float> %A, <16 x float> %B) #4125 %1 = bitcast <8 x i64> %C to <32 x bfloat>126 %2 = bitcast i32 %U to <32 x i1>127 %3 = select <32 x i1> %2, <32 x bfloat> %0, <32 x bfloat> %1128 %4 = bitcast <32 x bfloat> %3 to <8 x i64>129 ret <8 x i64> %4130}131 132declare <16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float>) #3133 134define <4 x i64> @test_mm512_cvtneps2bf16_512(<16 x float> %A) local_unnamed_addr #2 sanitize_memory {135; CHECK-LABEL: define <4 x i64> @test_mm512_cvtneps2bf16_512(136; CHECK-SAME: <16 x float> [[A:%.*]]) local_unnamed_addr #[[ATTR1]] {137; CHECK-NEXT: [[ENTRY:.*:]]138; CHECK-NEXT: [[TMP0:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8139; CHECK-NEXT: call void @llvm.donothing()140; CHECK-NEXT: [[TMP1:%.*]] = bitcast <16 x i32> [[TMP0]] to i512141; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i512 [[TMP1]], 0142; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB2:.*]], label %[[BB3:.*]], !prof [[PROF1]]143; CHECK: [[BB2]]:144; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]145; CHECK-NEXT: unreachable146; CHECK: [[BB3]]:147; CHECK-NEXT: [[TMP4:%.*]] = tail call <16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> [[A]])148; CHECK-NEXT: [[TMP5:%.*]] = bitcast <16 x bfloat> [[TMP4]] to <4 x i64>149; CHECK-NEXT: store <4 x i64> zeroinitializer, ptr @__msan_retval_tls, align 8150; CHECK-NEXT: ret <4 x i64> [[TMP5]]151;152entry:153 %0 = tail call <16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> %A) #4154 %1 = bitcast <16 x bfloat> %0 to <4 x i64>155 ret <4 x i64> %1156}157 158define <4 x i64> @test_mm512_maskz_cvtneps2bf16_512(<16 x float> %A, i16 %U) local_unnamed_addr #2 sanitize_memory {159; CHECK-LABEL: define <4 x i64> @test_mm512_maskz_cvtneps2bf16_512(160; CHECK-SAME: <16 x float> [[A:%.*]], i16 [[U:%.*]]) local_unnamed_addr #[[ATTR1]] {161; CHECK-NEXT: [[ENTRY:.*:]]162; CHECK-NEXT: [[TMP0:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8163; CHECK-NEXT: [[TMP1:%.*]] = load i16, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8164; CHECK-NEXT: call void @llvm.donothing()165; CHECK-NEXT: [[TMP2:%.*]] = bitcast <16 x i32> [[TMP0]] to i512166; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i512 [[TMP2]], 0167; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB3:.*]], label %[[BB4:.*]], !prof [[PROF1]]168; CHECK: [[BB3]]:169; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]170; CHECK-NEXT: unreachable171; CHECK: [[BB4]]:172; CHECK-NEXT: [[TMP5:%.*]] = tail call <16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> [[A]])173; CHECK-NEXT: [[TMP6:%.*]] = bitcast i16 [[TMP1]] to <16 x i1>174; CHECK-NEXT: [[TMP7:%.*]] = bitcast i16 [[U]] to <16 x i1>175; CHECK-NEXT: [[TMP8:%.*]] = select <16 x i1> [[TMP7]], <16 x i16> zeroinitializer, <16 x i16> zeroinitializer176; CHECK-NEXT: [[TMP9:%.*]] = bitcast <16 x bfloat> [[TMP5]] to <16 x i16>177; CHECK-NEXT: [[TMP10:%.*]] = xor <16 x i16> [[TMP9]], zeroinitializer178; CHECK-NEXT: [[TMP11:%.*]] = or <16 x i16> [[TMP10]], zeroinitializer179; CHECK-NEXT: [[TMP12:%.*]] = or <16 x i16> [[TMP11]], zeroinitializer180; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP6]], <16 x i16> [[TMP12]], <16 x i16> [[TMP8]]181; CHECK-NEXT: [[TMP13:%.*]] = select <16 x i1> [[TMP7]], <16 x bfloat> [[TMP5]], <16 x bfloat> zeroinitializer182; CHECK-NEXT: [[TMP14:%.*]] = bitcast <16 x i16> [[_MSPROP_SELECT]] to <4 x i64>183; CHECK-NEXT: [[TMP15:%.*]] = bitcast <16 x bfloat> [[TMP13]] to <4 x i64>184; CHECK-NEXT: store <4 x i64> [[TMP14]], ptr @__msan_retval_tls, align 8185; CHECK-NEXT: ret <4 x i64> [[TMP15]]186;187entry:188 %0 = tail call <16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> %A) #4189 %1 = bitcast i16 %U to <16 x i1>190 %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> zeroinitializer191 %3 = bitcast <16 x bfloat> %2 to <4 x i64>192 ret <4 x i64> %3193}194 195define <4 x i64> @test_mm512_mask_cvtneps2bf16_512(<4 x i64> %C, i16 %U, <16 x float> %A) local_unnamed_addr #2 sanitize_memory {196; CHECK-LABEL: define <4 x i64> @test_mm512_mask_cvtneps2bf16_512(197; CHECK-SAME: <4 x i64> [[C:%.*]], i16 [[U:%.*]], <16 x float> [[A:%.*]]) local_unnamed_addr #[[ATTR1]] {198; CHECK-NEXT: [[ENTRY:.*:]]199; CHECK-NEXT: [[TMP0:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8200; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i64>, ptr @__msan_param_tls, align 8201; CHECK-NEXT: [[TMP2:%.*]] = load i16, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8202; CHECK-NEXT: call void @llvm.donothing()203; CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x i32> [[TMP0]] to i512204; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0205; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB4:.*]], label %[[BB5:.*]], !prof [[PROF1]]206; CHECK: [[BB4]]:207; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]]208; CHECK-NEXT: unreachable209; CHECK: [[BB5]]:210; CHECK-NEXT: [[TMP6:%.*]] = tail call <16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> [[A]])211; CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x i64> [[TMP1]] to <16 x i16>212; CHECK-NEXT: [[TMP8:%.*]] = bitcast <4 x i64> [[C]] to <16 x bfloat>213; CHECK-NEXT: [[TMP9:%.*]] = bitcast i16 [[TMP2]] to <16 x i1>214; CHECK-NEXT: [[TMP10:%.*]] = bitcast i16 [[U]] to <16 x i1>215; CHECK-NEXT: [[TMP11:%.*]] = select <16 x i1> [[TMP10]], <16 x i16> zeroinitializer, <16 x i16> [[TMP7]]216; CHECK-NEXT: [[TMP12:%.*]] = bitcast <16 x bfloat> [[TMP6]] to <16 x i16>217; CHECK-NEXT: [[TMP13:%.*]] = bitcast <16 x bfloat> [[TMP8]] to <16 x i16>218; CHECK-NEXT: [[TMP14:%.*]] = xor <16 x i16> [[TMP12]], [[TMP13]]219; CHECK-NEXT: [[TMP15:%.*]] = or <16 x i16> [[TMP14]], zeroinitializer220; CHECK-NEXT: [[TMP16:%.*]] = or <16 x i16> [[TMP15]], [[TMP7]]221; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP9]], <16 x i16> [[TMP16]], <16 x i16> [[TMP11]]222; CHECK-NEXT: [[TMP17:%.*]] = select <16 x i1> [[TMP10]], <16 x bfloat> [[TMP6]], <16 x bfloat> [[TMP8]]223; CHECK-NEXT: [[TMP18:%.*]] = bitcast <16 x i16> [[_MSPROP_SELECT]] to <4 x i64>224; CHECK-NEXT: [[TMP19:%.*]] = bitcast <16 x bfloat> [[TMP17]] to <4 x i64>225; CHECK-NEXT: store <4 x i64> [[TMP18]], ptr @__msan_retval_tls, align 8226; CHECK-NEXT: ret <4 x i64> [[TMP19]]227;228entry:229 %0 = tail call <16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> %A) #4230 %1 = bitcast <4 x i64> %C to <16 x bfloat>231 %2 = bitcast i16 %U to <16 x i1>232 %3 = select <16 x i1> %2, <16 x bfloat> %0, <16 x bfloat> %1233 %4 = bitcast <16 x bfloat> %3 to <4 x i64>234 ret <4 x i64> %4235}236 237declare <16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float>, <32 x bfloat>, <32 x bfloat>) #3238 239define <16 x float> @test_mm512_dpbf16ps_512(<16 x float> %E, <32 x bfloat> %A, <32 x bfloat> %B) local_unnamed_addr #2 sanitize_memory {240; CHECK-LABEL: define <16 x float> @test_mm512_dpbf16ps_512(241; CHECK-SAME: <16 x float> [[E:%.*]], <32 x bfloat> [[A:%.*]], <32 x bfloat> [[B:%.*]]) local_unnamed_addr #[[ATTR1]] {242; CHECK-NEXT: [[ENTRY:.*:]]243; CHECK-NEXT: [[TMP1:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8244; CHECK-NEXT: [[TMP2:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8245; CHECK-NEXT: [[TMP11:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8246; CHECK-NEXT: call void @llvm.donothing()247; CHECK-NEXT: [[TMP3:%.*]] = icmp ne <32 x i16> [[TMP1]], zeroinitializer248; CHECK-NEXT: [[TMP4:%.*]] = icmp ne <32 x i16> [[TMP2]], zeroinitializer249; CHECK-NEXT: [[TMP5:%.*]] = or <32 x i1> [[TMP3]], [[TMP4]]250; CHECK-NEXT: [[TMP6:%.*]] = sext <32 x i1> [[TMP5]] to <32 x i16>251; CHECK-NEXT: [[TMP7:%.*]] = bitcast <32 x i16> [[TMP6]] to <16 x i32>252; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <16 x i32> [[TMP7]], zeroinitializer253; CHECK-NEXT: [[TMP9:%.*]] = sext <16 x i1> [[TMP12]] to <16 x i32>254; CHECK-NEXT: [[TMP10:%.*]] = or <16 x i32> [[TMP9]], [[TMP11]]255; CHECK-NEXT: [[TMP8:%.*]] = tail call <16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float> [[E]], <32 x bfloat> [[A]], <32 x bfloat> [[B]])256; CHECK-NEXT: store <16 x i32> [[TMP10]], ptr @__msan_retval_tls, align 8257; CHECK-NEXT: ret <16 x float> [[TMP8]]258;259entry:260 %0 = tail call <16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float> %E, <32 x bfloat> %A, <32 x bfloat> %B) #4261 ret <16 x float> %0262}263 264define <16 x float> @test_mm512_maskz_dpbf16ps_512(<16 x float> %E, <32 x bfloat> %A, <32 x bfloat> %B, i16 zeroext %U) local_unnamed_addr #2 sanitize_memory {265; CHECK-LABEL: define <16 x float> @test_mm512_maskz_dpbf16ps_512(266; CHECK-SAME: <16 x float> [[E:%.*]], <32 x bfloat> [[A:%.*]], <32 x bfloat> [[B:%.*]], i16 zeroext [[U:%.*]]) local_unnamed_addr #[[ATTR1]] {267; CHECK-NEXT: [[ENTRY:.*:]]268; CHECK-NEXT: [[TMP1:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8269; CHECK-NEXT: [[TMP2:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8270; CHECK-NEXT: [[TMP18:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8271; CHECK-NEXT: [[TMP3:%.*]] = load i16, ptr getelementptr (i8, ptr @__msan_param_tls, i64 192), align 8272; CHECK-NEXT: call void @llvm.donothing()273; CHECK-NEXT: [[TMP4:%.*]] = icmp ne <32 x i16> [[TMP1]], zeroinitializer274; CHECK-NEXT: [[TMP5:%.*]] = icmp ne <32 x i16> [[TMP2]], zeroinitializer275; CHECK-NEXT: [[TMP6:%.*]] = or <32 x i1> [[TMP4]], [[TMP5]]276; CHECK-NEXT: [[TMP7:%.*]] = sext <32 x i1> [[TMP6]] to <32 x i16>277; CHECK-NEXT: [[TMP8:%.*]] = bitcast <32 x i16> [[TMP7]] to <16 x i32>278; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <16 x i32> [[TMP8]], zeroinitializer279; CHECK-NEXT: [[TMP20:%.*]] = sext <16 x i1> [[TMP19]] to <16 x i32>280; CHECK-NEXT: [[TMP21:%.*]] = or <16 x i32> [[TMP20]], [[TMP18]]281; CHECK-NEXT: [[TMP9:%.*]] = tail call <16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float> [[E]], <32 x bfloat> [[A]], <32 x bfloat> [[B]])282; CHECK-NEXT: [[TMP10:%.*]] = bitcast i16 [[TMP3]] to <16 x i1>283; CHECK-NEXT: [[TMP11:%.*]] = bitcast i16 [[U]] to <16 x i1>284; CHECK-NEXT: [[TMP12:%.*]] = select <16 x i1> [[TMP11]], <16 x i32> [[TMP21]], <16 x i32> zeroinitializer285; CHECK-NEXT: [[TMP13:%.*]] = bitcast <16 x float> [[TMP9]] to <16 x i32>286; CHECK-NEXT: [[TMP14:%.*]] = xor <16 x i32> [[TMP13]], zeroinitializer287; CHECK-NEXT: [[TMP15:%.*]] = or <16 x i32> [[TMP14]], [[TMP21]]288; CHECK-NEXT: [[TMP16:%.*]] = or <16 x i32> [[TMP15]], zeroinitializer289; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP10]], <16 x i32> [[TMP16]], <16 x i32> [[TMP12]]290; CHECK-NEXT: [[TMP17:%.*]] = select <16 x i1> [[TMP11]], <16 x float> [[TMP9]], <16 x float> zeroinitializer291; CHECK-NEXT: store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8292; CHECK-NEXT: ret <16 x float> [[TMP17]]293;294entry:295 %0 = tail call <16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float> %E, <32 x bfloat> %A, <32 x bfloat> %B) #4296 %1 = bitcast i16 %U to <16 x i1>297 %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer298 ret <16 x float> %2299}300define <16 x float> @test_mm512_mask_dpbf16ps_512(i16 zeroext %U, <16 x float> %E, <32 x bfloat> %A, <32 x bfloat> %B) local_unnamed_addr #2 sanitize_memory {301; CHECK-LABEL: define <16 x float> @test_mm512_mask_dpbf16ps_512(302; CHECK-SAME: i16 zeroext [[U:%.*]], <16 x float> [[E:%.*]], <32 x bfloat> [[A:%.*]], <32 x bfloat> [[B:%.*]]) local_unnamed_addr #[[ATTR1]] {303; CHECK-NEXT: [[ENTRY:.*:]]304; CHECK-NEXT: [[TMP1:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8305; CHECK-NEXT: [[TMP2:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8306; CHECK-NEXT: [[TMP0:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8307; CHECK-NEXT: [[TMP3:%.*]] = load i16, ptr @__msan_param_tls, align 8308; CHECK-NEXT: call void @llvm.donothing()309; CHECK-NEXT: [[TMP4:%.*]] = icmp ne <32 x i16> [[TMP1]], zeroinitializer310; CHECK-NEXT: [[TMP5:%.*]] = icmp ne <32 x i16> [[TMP2]], zeroinitializer311; CHECK-NEXT: [[TMP6:%.*]] = or <32 x i1> [[TMP4]], [[TMP5]]312; CHECK-NEXT: [[TMP7:%.*]] = sext <32 x i1> [[TMP6]] to <32 x i16>313; CHECK-NEXT: [[TMP8:%.*]] = bitcast <32 x i16> [[TMP7]] to <16 x i32>314; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <16 x i32> [[TMP8]], zeroinitializer315; CHECK-NEXT: [[TMP20:%.*]] = sext <16 x i1> [[TMP19]] to <16 x i32>316; CHECK-NEXT: [[TMP21:%.*]] = or <16 x i32> [[TMP20]], [[TMP0]]317; CHECK-NEXT: [[TMP9:%.*]] = tail call <16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float> [[E]], <32 x bfloat> [[A]], <32 x bfloat> [[B]])318; CHECK-NEXT: [[TMP10:%.*]] = bitcast i16 [[TMP3]] to <16 x i1>319; CHECK-NEXT: [[TMP11:%.*]] = bitcast i16 [[U]] to <16 x i1>320; CHECK-NEXT: [[TMP12:%.*]] = select <16 x i1> [[TMP11]], <16 x i32> [[TMP21]], <16 x i32> [[TMP0]]321; CHECK-NEXT: [[TMP13:%.*]] = bitcast <16 x float> [[TMP9]] to <16 x i32>322; CHECK-NEXT: [[TMP14:%.*]] = bitcast <16 x float> [[E]] to <16 x i32>323; CHECK-NEXT: [[TMP15:%.*]] = xor <16 x i32> [[TMP13]], [[TMP14]]324; CHECK-NEXT: [[TMP16:%.*]] = or <16 x i32> [[TMP15]], [[TMP21]]325; CHECK-NEXT: [[TMP17:%.*]] = or <16 x i32> [[TMP16]], [[TMP0]]326; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP10]], <16 x i32> [[TMP17]], <16 x i32> [[TMP12]]327; CHECK-NEXT: [[TMP18:%.*]] = select <16 x i1> [[TMP11]], <16 x float> [[TMP9]], <16 x float> [[E]]328; CHECK-NEXT: store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8329; CHECK-NEXT: ret <16 x float> [[TMP18]]330;331entry:332 %0 = tail call <16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float> %E, <32 x bfloat> %A, <32 x bfloat> %B) #4333 %1 = bitcast i16 %U to <16 x i1>334 %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %E335 ret <16 x float> %2336}337;.338; CHECK: [[PROF1]] = !{!"branch_weights", i32 1, i32 1048575}339;.340