1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -loop-vectorize -force-vector-width=2 -force-vector-interleave=2 -S | FileCheck %s 3 4define void @test1(float* noalias nocapture %a, float* noalias nocapture readonly %b) { 5; CHECK-LABEL: @test1( 6; CHECK-NEXT: entry: 7; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 8; CHECK: vector.ph: 9; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 10; CHECK: vector.body: 11; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 12; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 13; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 14; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[TMP0]] 15; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP1]] 16; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, float* [[TMP2]], i32 0 17; CHECK-NEXT: [[TMP5:%.*]] = bitcast float* [[TMP4]] to <2 x float>* 18; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP5]], align 4 19; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, float* [[TMP2]], i32 2 20; CHECK-NEXT: [[TMP7:%.*]] = bitcast float* [[TMP6]] to <2 x float>* 21; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <2 x float>, <2 x float>* [[TMP7]], align 4 22; CHECK-NEXT: [[TMP8:%.*]] = fcmp ogt <2 x float> [[WIDE_LOAD]], <float 1.000000e+02, float 1.000000e+02> 23; CHECK-NEXT: [[TMP9:%.*]] = fcmp ogt <2 x float> [[WIDE_LOAD1]], <float 1.000000e+02, float 1.000000e+02> 24; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i1> [[TMP8]], i32 0 25; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP10]]) 26; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i1> [[TMP8]], i32 1 27; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) 28; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i1> [[TMP9]], i32 0 29; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP12]]) 30; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i1> [[TMP9]], i32 1 31; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP13]]) 32; CHECK-NEXT: [[TMP14:%.*]] = fadd <2 x float> [[WIDE_LOAD]], <float 1.000000e+00, float 1.000000e+00> 33; CHECK-NEXT: [[TMP15:%.*]] = fadd <2 x float> [[WIDE_LOAD1]], <float 1.000000e+00, float 1.000000e+00> 34; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[TMP0]] 35; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP1]] 36; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, float* [[TMP16]], i32 0 37; CHECK-NEXT: [[TMP19:%.*]] = bitcast float* [[TMP18]] to <2 x float>* 38; CHECK-NEXT: store <2 x float> [[TMP14]], <2 x float>* [[TMP19]], align 4 39; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds float, float* [[TMP16]], i32 2 40; CHECK-NEXT: [[TMP21:%.*]] = bitcast float* [[TMP20]] to <2 x float>* 41; CHECK-NEXT: store <2 x float> [[TMP15]], <2 x float>* [[TMP21]], align 4 42; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 43; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1600 44; CHECK-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 45; CHECK: middle.block: 46; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1600, 1600 47; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 48; CHECK: scalar.ph: 49; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1600, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 50; CHECK-NEXT: br label [[FOR_BODY:%.*]] 51; CHECK: for.body: 52; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] 53; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[INDVARS_IV]] 54; CHECK-NEXT: [[TMP23:%.*]] = load float, float* [[ARRAYIDX]], align 4 55; CHECK-NEXT: [[CMP1:%.*]] = fcmp ogt float [[TMP23]], 1.000000e+02 56; CHECK-NEXT: tail call void @llvm.assume(i1 [[CMP1]]) 57; CHECK-NEXT: [[ADD:%.*]] = fadd float [[TMP23]], 1.000000e+00 58; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]] 59; CHECK-NEXT: store float [[ADD]], float* [[ARRAYIDX5]], align 4 60; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 61; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV]], 1599 62; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]] 63; CHECK: for.end: 64; CHECK-NEXT: ret void 65; 66entry: 67 br label %for.body 68 69 70for.body: ; preds = %for.body, %entry 71 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 72 %arrayidx = getelementptr inbounds float, float* %b, i64 %indvars.iv 73 %0 = load float, float* %arrayidx, align 4 74 %cmp1 = fcmp ogt float %0, 1.000000e+02 75 tail call void @llvm.assume(i1 %cmp1) 76 %add = fadd float %0, 1.000000e+00 77 %arrayidx5 = getelementptr inbounds float, float* %a, i64 %indvars.iv 78 store float %add, float* %arrayidx5, align 4 79 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 80 %exitcond = icmp eq i64 %indvars.iv, 1599 81 br i1 %exitcond, label %for.end, label %for.body 82 83for.end: ; preds = %for.body 84 ret void 85} 86 87declare void @llvm.assume(i1) #0 88 89attributes #0 = { nounwind willreturn } 90 91%struct.data = type { float*, float* } 92 93define void @test2(%struct.data* nocapture readonly %d) { 94; CHECK-LABEL: @test2( 95; CHECK-NEXT: entry: 96; CHECK-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_DATA:%.*]], %struct.data* [[D:%.*]], i64 0, i32 1 97; CHECK-NEXT: [[TMP0:%.*]] = load float*, float** [[B]], align 8 98; CHECK-NEXT: [[TMP1:%.*]] = bitcast float* [[TMP0]] to i8* 99; CHECK-NEXT: [[PTRINT:%.*]] = ptrtoint float* [[TMP0]] to i64 100; CHECK-NEXT: [[MASKEDPTR:%.*]] = and i64 [[PTRINT]], 31 101; CHECK-NEXT: [[MASKCOND:%.*]] = icmp eq i64 [[MASKEDPTR]], 0 102; CHECK-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_DATA]], %struct.data* [[D]], i64 0, i32 0 103; CHECK-NEXT: [[TMP2:%.*]] = load float*, float** [[A]], align 8 104; CHECK-NEXT: [[TMP3:%.*]] = bitcast float* [[TMP2]] to i8* 105; CHECK-NEXT: [[PTRINT2:%.*]] = ptrtoint float* [[TMP2]] to i64 106; CHECK-NEXT: [[MASKEDPTR3:%.*]] = and i64 [[PTRINT2]], 31 107; CHECK-NEXT: [[MASKCOND4:%.*]] = icmp eq i64 [[MASKEDPTR3]], 0 108; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] 109; CHECK: vector.memcheck: 110; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr float, float* [[TMP2]], i64 1600 111; CHECK-NEXT: [[SCEVGEP1:%.*]] = bitcast float* [[SCEVGEP]] to i8* 112; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr float, float* [[TMP0]], i64 1600 113; CHECK-NEXT: [[SCEVGEP23:%.*]] = bitcast float* [[SCEVGEP2]] to i8* 114; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult i8* [[TMP3]], [[SCEVGEP23]] 115; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult i8* [[TMP1]], [[SCEVGEP1]] 116; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 117; CHECK-NEXT: [[MEMCHECK_CONFLICT:%.*]] = and i1 [[FOUND_CONFLICT]], true 118; CHECK-NEXT: br i1 [[MEMCHECK_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] 119; CHECK: vector.ph: 120; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 121; CHECK: vector.body: 122; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 123; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0 124; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 2 125; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) 126; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) 127; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) 128; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) 129; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, float* [[TMP0]], i64 [[TMP4]] 130; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, float* [[TMP0]], i64 [[TMP5]] 131; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, float* [[TMP6]], i32 0 132; CHECK-NEXT: [[TMP9:%.*]] = bitcast float* [[TMP8]] to <2 x float>* 133; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP9]], align 4, !alias.scope !4 134; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, float* [[TMP6]], i32 2 135; CHECK-NEXT: [[TMP11:%.*]] = bitcast float* [[TMP10]] to <2 x float>* 136; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x float>, <2 x float>* [[TMP11]], align 4, !alias.scope !4 137; CHECK-NEXT: [[TMP12:%.*]] = fadd <2 x float> [[WIDE_LOAD]], <float 1.000000e+00, float 1.000000e+00> 138; CHECK-NEXT: [[TMP13:%.*]] = fadd <2 x float> [[WIDE_LOAD4]], <float 1.000000e+00, float 1.000000e+00> 139; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) 140; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) 141; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) 142; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) 143; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, float* [[TMP2]], i64 [[TMP4]] 144; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, float* [[TMP2]], i64 [[TMP5]] 145; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds float, float* [[TMP14]], i32 0 146; CHECK-NEXT: [[TMP17:%.*]] = bitcast float* [[TMP16]] to <2 x float>* 147; CHECK-NEXT: store <2 x float> [[TMP12]], <2 x float>* [[TMP17]], align 4, !alias.scope !7, !noalias !4 148; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, float* [[TMP14]], i32 2 149; CHECK-NEXT: [[TMP19:%.*]] = bitcast float* [[TMP18]] to <2 x float>* 150; CHECK-NEXT: store <2 x float> [[TMP13]], <2 x float>* [[TMP19]], align 4, !alias.scope !7, !noalias !4 151; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 152; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1600 153; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] 154; CHECK: middle.block: 155; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1600, 1600 156; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 157; CHECK: scalar.ph: 158; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1600, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ] 159; CHECK-NEXT: br label [[FOR_BODY:%.*]] 160; CHECK: for.body: 161; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] 162; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) 163; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP0]], i64 [[INDVARS_IV]] 164; CHECK-NEXT: [[TMP21:%.*]] = load float, float* [[ARRAYIDX]], align 4 165; CHECK-NEXT: [[ADD:%.*]] = fadd float [[TMP21]], 1.000000e+00 166; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) 167; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[TMP2]], i64 [[INDVARS_IV]] 168; CHECK-NEXT: store float [[ADD]], float* [[ARRAYIDX5]], align 4 169; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 170; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV]], 1599 171; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] 172; CHECK: for.end: 173; CHECK-NEXT: ret void 174; 175entry: 176 %b = getelementptr inbounds %struct.data, %struct.data* %d, i64 0, i32 1 177 %0 = load float*, float** %b, align 8 178 %ptrint = ptrtoint float* %0 to i64 179 %maskedptr = and i64 %ptrint, 31 180 %maskcond = icmp eq i64 %maskedptr, 0 181 %a = getelementptr inbounds %struct.data, %struct.data* %d, i64 0, i32 0 182 %1 = load float*, float** %a, align 8 183 %ptrint2 = ptrtoint float* %1 to i64 184 %maskedptr3 = and i64 %ptrint2, 31 185 %maskcond4 = icmp eq i64 %maskedptr3, 0 186 br label %for.body 187 188 189for.body: ; preds = %for.body, %entry 190 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 191 tail call void @llvm.assume(i1 %maskcond) 192 %arrayidx = getelementptr inbounds float, float* %0, i64 %indvars.iv 193 %2 = load float, float* %arrayidx, align 4 194 %add = fadd float %2, 1.000000e+00 195 tail call void @llvm.assume(i1 %maskcond4) 196 %arrayidx5 = getelementptr inbounds float, float* %1, i64 %indvars.iv 197 store float %add, float* %arrayidx5, align 4 198 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 199 %exitcond = icmp eq i64 %indvars.iv, 1599 200 br i1 %exitcond, label %for.end, label %for.body 201 202for.end: ; preds = %for.body 203 ret void 204} 205 206; Test case for PR43620. Make sure we can vectorize with predication in presence 207; of assume calls. For now, check that we drop all assumes in predicated blocks 208; in the vector body. 209define void @predicated_assume(float* noalias nocapture readonly %a, float* noalias nocapture %b, i32 %n) { 210; Check that the vector.body does not contain any assumes. 211; CHECK-LABEL: @predicated_assume( 212; CHECK-NEXT: entry: 213; CHECK-NEXT: [[CMP15:%.*]] = icmp eq i32 [[N:%.*]], 0 214; CHECK-NEXT: br i1 [[CMP15]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY_PREHEADER:%.*]] 215; CHECK: for.body.preheader: 216; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[N]] to i64 217; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 4 218; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 219; CHECK: vector.ph: 220; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 4 221; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]] 222; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 223; CHECK: vector.body: 224; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 225; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] 226; CHECK-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2> 227; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 0 228; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 1 229; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 2 230; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 3 231; CHECK-NEXT: [[TMP5:%.*]] = icmp ult <2 x i64> [[VEC_IND]], <i64 495616, i64 495616> 232; CHECK-NEXT: [[TMP6:%.*]] = icmp ult <2 x i64> [[STEP_ADD]], <i64 495616, i64 495616> 233; CHECK-NEXT: [[TMP7:%.*]] = icmp ult <2 x i64> [[VEC_IND]], <i64 991232, i64 991232> 234; CHECK-NEXT: [[TMP8:%.*]] = icmp ult <2 x i64> [[STEP_ADD]], <i64 991232, i64 991232> 235; CHECK-NEXT: [[TMP9:%.*]] = xor <2 x i1> [[TMP5]], <i1 true, i1 true> 236; CHECK-NEXT: [[TMP10:%.*]] = xor <2 x i1> [[TMP6]], <i1 true, i1 true> 237; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP5]], <2 x float> <float 2.300000e+01, float 2.300000e+01>, <2 x float> <float 4.200000e+01, float 4.200000e+01> 238; CHECK-NEXT: [[PREDPHI2:%.*]] = select <2 x i1> [[TMP6]], <2 x float> <float 2.300000e+01, float 2.300000e+01>, <2 x float> <float 4.200000e+01, float 4.200000e+01> 239; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[TMP1]] 240; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP3]] 241; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds float, float* [[TMP11]], i32 0 242; CHECK-NEXT: [[TMP14:%.*]] = bitcast float* [[TMP13]] to <2 x float>* 243; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP14]], align 4 244; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, float* [[TMP11]], i32 2 245; CHECK-NEXT: [[TMP16:%.*]] = bitcast float* [[TMP15]] to <2 x float>* 246; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <2 x float>, <2 x float>* [[TMP16]], align 4 247; CHECK-NEXT: [[TMP17:%.*]] = fmul <2 x float> [[PREDPHI]], [[WIDE_LOAD]] 248; CHECK-NEXT: [[TMP18:%.*]] = fmul <2 x float> [[PREDPHI2]], [[WIDE_LOAD3]] 249; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[TMP1]] 250; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP3]] 251; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, float* [[TMP19]], i32 0 252; CHECK-NEXT: [[TMP22:%.*]] = bitcast float* [[TMP21]] to <2 x float>* 253; CHECK-NEXT: store <2 x float> [[TMP17]], <2 x float>* [[TMP22]], align 4 254; CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds float, float* [[TMP19]], i32 2 255; CHECK-NEXT: [[TMP24:%.*]] = bitcast float* [[TMP23]] to <2 x float>* 256; CHECK-NEXT: store <2 x float> [[TMP18]], <2 x float>* [[TMP24]], align 4 257; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 258; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[STEP_ADD]], <i64 2, i64 2> 259; CHECK-NEXT: [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 260; CHECK-NEXT: br i1 [[TMP25]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] 261; CHECK: middle.block: 262; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] 263; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]] 264; CHECK: scalar.ph: 265; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ] 266; CHECK-NEXT: br label [[FOR_BODY:%.*]] 267; CHECK: for.cond.cleanup.loopexit: 268; CHECK-NEXT: br label [[FOR_COND_CLEANUP]] 269; CHECK: for.cond.cleanup: 270; CHECK-NEXT: ret void 271; CHECK: for.body: 272; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[IF_END5:%.*]] ] 273; CHECK-NEXT: [[CMP1:%.*]] = icmp ult i64 [[INDVARS_IV]], 495616 274; CHECK-NEXT: br i1 [[CMP1]], label [[IF_END5]], label [[IF_ELSE:%.*]] 275; CHECK: if.else: 276; CHECK-NEXT: [[CMP2:%.*]] = icmp ult i64 [[INDVARS_IV]], 991232 277; CHECK-NEXT: tail call void @llvm.assume(i1 [[CMP2]]) 278; CHECK-NEXT: br label [[IF_END5]] 279; CHECK: if.end5: 280; CHECK-NEXT: [[X_0:%.*]] = phi float [ 4.200000e+01, [[IF_ELSE]] ], [ 2.300000e+01, [[FOR_BODY]] ] 281; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]] 282; CHECK-NEXT: [[TMP26:%.*]] = load float, float* [[ARRAYIDX]], align 4 283; CHECK-NEXT: [[MUL:%.*]] = fmul float [[X_0]], [[TMP26]] 284; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[INDVARS_IV]] 285; CHECK-NEXT: store float [[MUL]], float* [[ARRAYIDX7]], align 4 286; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 287; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[TMP0]] 288; CHECK-NEXT: br i1 [[CMP]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]] 289; 290entry: 291 %cmp15 = icmp eq i32 %n, 0 292 br i1 %cmp15, label %for.cond.cleanup, label %for.body.preheader 293 294for.body.preheader: ; preds = %entry 295 %0 = zext i32 %n to i64 296 br label %for.body 297 298for.cond.cleanup.loopexit: ; preds = %if.end5 299 br label %for.cond.cleanup 300 301for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry 302 ret void 303 304for.body: ; preds = %for.body.preheader, %if.end5 305 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %if.end5 ] 306 %cmp1 = icmp ult i64 %indvars.iv, 495616 307 br i1 %cmp1, label %if.end5, label %if.else 308 309if.else: ; preds = %for.body 310 %cmp2 = icmp ult i64 %indvars.iv, 991232 311 tail call void @llvm.assume(i1 %cmp2) 312 br label %if.end5 313 314if.end5: ; preds = %for.body, %if.else 315 %x.0 = phi float [ 4.200000e+01, %if.else ], [ 2.300000e+01, %for.body ] 316 %arrayidx = getelementptr inbounds float, float* %a, i64 %indvars.iv 317 %1 = load float, float* %arrayidx, align 4 318 %mul = fmul float %x.0, %1 319 %arrayidx7 = getelementptr inbounds float, float* %b, i64 %indvars.iv 320 store float %mul, float* %arrayidx7, align 4 321 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 322 %cmp = icmp eq i64 %indvars.iv.next, %0 323 br i1 %cmp, label %for.cond.cleanup.loopexit, label %for.body 324} 325