94 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue -mcpu=neoverse-v1 -S %s | FileCheck %s3 4target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"5target triple = "aarch64-unknown-linux-gnu"6 7define i32 @test_phi_iterator_invalidation(ptr %A, ptr noalias %B) {8; CHECK-LABEL: @test_phi_iterator_invalidation(9; CHECK-NEXT: entry:10; CHECK-NEXT: br label [[VECTOR_PH:%.*]]11; CHECK: vector.ph:12; CHECK-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i64(i64 0, i64 1002)13; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]14; CHECK: vector.body:15; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE6:%.*]] ]16; CHECK-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <4 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[VECTOR_PH]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[PRED_LOAD_CONTINUE6]] ]17; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_LOAD_CONTINUE6]] ]18; CHECK-NEXT: [[VECTOR_RECUR:%.*]] = phi <4 x i16> [ <i16 poison, i16 poison, i16 poison, i16 0>, [[VECTOR_PH]] ], [ [[TMP24:%.*]], [[PRED_LOAD_CONTINUE6]] ]19; CHECK-NEXT: [[TMP0:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 1)20; CHECK-NEXT: [[TMP27:%.*]] = extractelement <4 x i64> [[TMP0]], i32 021; CHECK-NEXT: [[TMP1:%.*]] = extractelement <4 x i1> [[ACTIVE_LANE_MASK]], i32 022; CHECK-NEXT: br i1 [[TMP1]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]23; CHECK: pred.load.if:24; CHECK-NEXT: [[TMP2:%.*]] = extractelement <4 x i64> [[TMP0]], i32 025; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i32, ptr [[A:%.*]], i64 [[TMP2]]26; CHECK-NEXT: [[TMP4:%.*]] = load i16, ptr [[TMP3]], align 227; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x i16> poison, i16 [[TMP4]], i32 028; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]]29; CHECK: pred.load.continue:30; CHECK-NEXT: [[TMP6:%.*]] = phi <4 x i16> [ poison, [[VECTOR_BODY]] ], [ [[TMP5]], [[PRED_LOAD_IF]] ]31; CHECK-NEXT: [[TMP7:%.*]] = extractelement <4 x i1> [[ACTIVE_LANE_MASK]], i32 132; CHECK-NEXT: br i1 [[TMP7]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2:%.*]]33; CHECK: pred.load.if1:34; CHECK-NEXT: [[TMP8:%.*]] = extractelement <4 x i64> [[TMP0]], i32 135; CHECK-NEXT: [[TMP9:%.*]] = getelementptr i32, ptr [[A]], i64 [[TMP8]]36; CHECK-NEXT: [[TMP10:%.*]] = load i16, ptr [[TMP9]], align 237; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x i16> [[TMP6]], i16 [[TMP10]], i32 138; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE2]]39; CHECK: pred.load.continue2:40; CHECK-NEXT: [[TMP12:%.*]] = phi <4 x i16> [ [[TMP6]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP11]], [[PRED_LOAD_IF1]] ]41; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x i1> [[ACTIVE_LANE_MASK]], i32 242; CHECK-NEXT: br i1 [[TMP13]], label [[PRED_LOAD_IF3:%.*]], label [[PRED_LOAD_CONTINUE4:%.*]]43; CHECK: pred.load.if3:44; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x i64> [[TMP0]], i32 245; CHECK-NEXT: [[TMP15:%.*]] = getelementptr i32, ptr [[A]], i64 [[TMP14]]46; CHECK-NEXT: [[TMP16:%.*]] = load i16, ptr [[TMP15]], align 247; CHECK-NEXT: [[TMP17:%.*]] = insertelement <4 x i16> [[TMP12]], i16 [[TMP16]], i32 248; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE4]]49; CHECK: pred.load.continue4:50; CHECK-NEXT: [[TMP18:%.*]] = phi <4 x i16> [ [[TMP12]], [[PRED_LOAD_CONTINUE2]] ], [ [[TMP17]], [[PRED_LOAD_IF3]] ]51; CHECK-NEXT: [[TMP19:%.*]] = extractelement <4 x i1> [[ACTIVE_LANE_MASK]], i32 352; CHECK-NEXT: br i1 [[TMP19]], label [[PRED_LOAD_IF5:%.*]], label [[PRED_LOAD_CONTINUE6]]53; CHECK: pred.load.if5:54; CHECK-NEXT: [[TMP20:%.*]] = extractelement <4 x i64> [[TMP0]], i32 355; CHECK-NEXT: [[TMP21:%.*]] = getelementptr i32, ptr [[A]], i64 [[TMP20]]56; CHECK-NEXT: [[TMP22:%.*]] = load i16, ptr [[TMP21]], align 257; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x i16> [[TMP18]], i16 [[TMP22]], i32 358; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE6]]59; CHECK: pred.load.continue6:60; CHECK-NEXT: [[TMP24]] = phi <4 x i16> [ [[TMP18]], [[PRED_LOAD_CONTINUE4]] ], [ [[TMP23]], [[PRED_LOAD_IF5]] ]61; CHECK-NEXT: [[TMP25:%.*]] = shufflevector <4 x i16> [[VECTOR_RECUR]], <4 x i16> [[TMP24]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>62; CHECK-NEXT: [[TMP26:%.*]] = sext <4 x i16> [[TMP25]] to <4 x i32>63; CHECK-NEXT: [[TMP28:%.*]] = getelementptr i32, ptr [[B:%.*]], i64 [[TMP27]]64; CHECK-NEXT: call void @llvm.masked.store.v4i32.p0(<4 x i32> [[TMP26]], ptr align 4 [[TMP28]], <4 x i1> [[ACTIVE_LANE_MASK]])65; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 466; CHECK-NEXT: [[ACTIVE_LANE_MASK_NEXT]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i64(i64 [[INDEX_NEXT]], i64 1002)67; CHECK-NEXT: [[TMP29:%.*]] = extractelement <4 x i1> [[ACTIVE_LANE_MASK_NEXT]], i32 068; CHECK-NEXT: [[TMP30:%.*]] = xor i1 [[TMP29]], true69; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4)70; CHECK-NEXT: br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]71; CHECK: middle.block:72; CHECK-NEXT: br label [[LOOP:%.*]]73; CHECK: exit:74; CHECK-NEXT: ret i32 075;76entry:77 br label %loop78 79loop:80 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]81 %for = phi i16 [ 0, %entry ], [ %for.next, %loop ]82 %sext = sext i16 %for to i3283 %iv.next = add i64 %iv, 184 %gep.A = getelementptr i32, ptr %A, i64 %iv.next85 %for.next = load i16, ptr %gep.A, align 286 %gep.B = getelementptr i32, ptr %B, i64 %iv.next87 store i32 %sext, ptr %gep.B88 %ec = icmp eq i64 %iv, 100189 br i1 %ec, label %exit, label %loop90 91exit:92 ret i32 093}94