87 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize -mtriple=s390x-ibm-linux -mcpu=z13 -force-vector-width=2 -S %s | FileCheck %s3 4; Test case from PR44020.5 6; In func_21, %rec forms a first order recurrence and we predicate to avoid7; scalar iteration overhead for a low trip count loop. Make sure we pick8; the correct insertion point when fixing first order recurrences.9 10@A = external dso_local global [5 x i32], align 411@B = external dso_local global [5 x i32], align 412 13define void @func_21() {14; CHECK-LABEL: @func_21(15; CHECK-NEXT: entry:16; CHECK-NEXT: br label [[VECTOR_PH:%.*]]17; CHECK: vector.ph:18; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]19; CHECK: vector.body:20; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE4:%.*]] ]21; CHECK-NEXT: [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[TMP12:%.*]], [[PRED_STORE_CONTINUE4]] ]22; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i8> [ <i8 0, i8 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE4]] ]23; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 024; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 125; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <2 x i8> [[VEC_IND]], splat (i8 4)26; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i1> [[TMP2]], i32 027; CHECK-NEXT: br i1 [[TMP3]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]28; CHECK: pred.load.if:29; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds [5 x i32], ptr @A, i64 0, i64 [[TMP0]]30; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 431; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i32> poison, i32 [[TMP5]], i32 032; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]]33; CHECK: pred.load.continue:34; CHECK-NEXT: [[TMP7:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP6]], [[PRED_LOAD_IF]] ]35; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x i1> [[TMP2]], i32 136; CHECK-NEXT: br i1 [[TMP8]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2:%.*]]37; CHECK: pred.load.if1:38; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds [5 x i32], ptr @A, i64 0, i64 [[TMP1]]39; CHECK-NEXT: [[TMP10:%.*]] = load i32, ptr [[TMP9]], align 440; CHECK-NEXT: [[TMP11:%.*]] = insertelement <2 x i32> [[TMP7]], i32 [[TMP10]], i32 141; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE2]]42; CHECK: pred.load.continue2:43; CHECK-NEXT: [[TMP12]] = phi <2 x i32> [ [[TMP7]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP11]], [[PRED_LOAD_IF1]] ]44; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[TMP12]], <2 x i32> <i32 1, i32 2>45; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i1> [[TMP2]], i32 046; CHECK-NEXT: br i1 [[TMP14]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]47; CHECK: pred.store.if:48; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds [5 x i32], ptr @B, i64 0, i64 [[TMP0]]49; CHECK-NEXT: [[TMP16:%.*]] = extractelement <2 x i32> [[TMP13]], i32 050; CHECK-NEXT: store i32 [[TMP16]], ptr [[TMP15]], align 451; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]]52; CHECK: pred.store.continue:53; CHECK-NEXT: [[TMP17:%.*]] = extractelement <2 x i1> [[TMP2]], i32 154; CHECK-NEXT: br i1 [[TMP17]], label [[PRED_STORE_IF3:%.*]], label [[PRED_STORE_CONTINUE4]]55; CHECK: pred.store.if3:56; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds [5 x i32], ptr @B, i64 0, i64 [[TMP1]]57; CHECK-NEXT: [[TMP19:%.*]] = extractelement <2 x i32> [[TMP13]], i32 158; CHECK-NEXT: store i32 [[TMP19]], ptr [[TMP18]], align 459; CHECK-NEXT: br label [[PRED_STORE_CONTINUE4]]60; CHECK: pred.store.continue4:61; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 262; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i8> [[VEC_IND]], splat (i8 2)63; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], 664; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]65; CHECK: middle.block:66; CHECK-NEXT: br label [[LOOP:%.*]]67; CHECK: exit:68; CHECK-NEXT: ret void69;70entry:71 br label %loop72 73loop: ; preds = %loop, %entry74 %rec = phi i32 [ 0, %entry], [ %lv, %loop ]75 %indvars.iv = phi i64 [ 0, %entry], [ %indvars.iv.next, %loop ]76 %A.ptr= getelementptr inbounds [5 x i32], ptr @A, i64 0, i64 %indvars.iv77 %lv = load i32, ptr %A.ptr, align 478 %B.ptr = getelementptr inbounds [5 x i32], ptr @B, i64 0, i64 %indvars.iv79 store i32 %rec, ptr %B.ptr, align 480 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 181 %exitcond = icmp eq i64 %indvars.iv.next, 582 br i1 %exitcond, label %exit, label %loop83 84exit: ; preds = %loop85 ret void86}87