1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -loop-vectorize -force-vector-interleave=4 -pass-remarks='loop-vectorize' -disable-output -S 2>&1 | FileCheck %s --check-prefix=CHECK-REMARKS 3; RUN: opt < %s -loop-vectorize -force-vector-interleave=4 -S | FileCheck %s 4 5; These tests are to check that fold-tail procedure produces correct scalar code when 6; loop-vectorization is only unrolling but not vectorizing. 7 8; CHECK-REMARKS: remark: {{.*}} interleaved loop (interleaved count: 4) 9; CHECK-REMARKS-NEXT: remark: {{.*}} interleaved loop (interleaved count: 4) 10; CHECK-REMARKS-NOT: remark: {{.*}} vectorized loop 11 12define void @VF1-VPlanExe(i32* %dst) { 13; CHECK-LABEL: @VF1-VPlanExe( 14; CHECK-NEXT: entry: 15; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 16; CHECK: vector.ph: 17; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 18; CHECK: vector.body: 19; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE9:%.*]] ] 20; CHECK-NEXT: [[INDUCTION:%.*]] = add i64 [[INDEX]], 0 21; CHECK-NEXT: [[INDUCTION1:%.*]] = add i64 [[INDEX]], 1 22; CHECK-NEXT: [[INDUCTION2:%.*]] = add i64 [[INDEX]], 2 23; CHECK-NEXT: [[INDUCTION3:%.*]] = add i64 [[INDEX]], 3 24; CHECK-NEXT: [[TMP0:%.*]] = icmp ule i64 [[INDUCTION]], 14 25; CHECK-NEXT: [[TMP1:%.*]] = icmp ule i64 [[INDUCTION1]], 14 26; CHECK-NEXT: [[TMP2:%.*]] = icmp ule i64 [[INDUCTION2]], 14 27; CHECK-NEXT: [[TMP3:%.*]] = icmp ule i64 [[INDUCTION3]], 14 28; CHECK-NEXT: br i1 [[TMP0]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]] 29; CHECK: pred.store.if: 30; CHECK-NEXT: [[SUNK_IND0:%.*]] = add i64 [[INDEX]], 0 31; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i64 [[SUNK_IND0]] 32; CHECK-NEXT: store i32 0, i32* [[TMP4]], align 4 33; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]] 34; CHECK: pred.store.continue: 35; CHECK-NEXT: br i1 [[TMP1]], label [[PRED_STORE_IF4:%.*]], label [[PRED_STORE_CONTINUE5:%.*]] 36; CHECK: pred.store.if7: 37; CHECK-NEXT: [[SUNK_IND1:%.*]] = add i64 [[INDEX]], 1 38; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 [[SUNK_IND1]] 39; CHECK-NEXT: store i32 0, i32* [[TMP5]], align 4 40; CHECK-NEXT: br label [[PRED_STORE_CONTINUE5]] 41; CHECK: pred.store.continue8: 42; CHECK-NEXT: br i1 [[TMP2]], label [[PRED_STORE_IF6:%.*]], label [[PRED_STORE_CONTINUE7:%.*]] 43; CHECK: pred.store.if9: 44; CHECK-NEXT: [[SUNK_IND2:%.*]] = add i64 [[INDEX]], 2 45; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 [[SUNK_IND2]] 46; CHECK-NEXT: store i32 0, i32* [[TMP6]], align 4 47; CHECK-NEXT: br label [[PRED_STORE_CONTINUE7]] 48; CHECK: pred.store.continue10: 49; CHECK-NEXT: br i1 [[TMP3]], label [[PRED_STORE_IF8:%.*]], label [[PRED_STORE_CONTINUE9]] 50; CHECK: pred.store.if11: 51; CHECK-NEXT: [[SUNK_IND3:%.*]] = add i64 [[INDEX]], 3 52; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 [[SUNK_IND3]] 53; CHECK-NEXT: store i32 0, i32* [[TMP7]], align 4 54; CHECK-NEXT: br label [[PRED_STORE_CONTINUE9]] 55; CHECK: pred.store.continue12: 56; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 57; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16 58; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 59; CHECK: middle.block: 60; CHECK-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]] 61; CHECK: scalar.ph: 62; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 16, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 63; CHECK-NEXT: br label [[FOR_BODY:%.*]] 64; CHECK: for.cond.cleanup: 65; CHECK-NEXT: ret void 66; CHECK: for.body: 67; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] 68; CHECK-NEXT: [[DST_PTR:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 [[INDVARS_IV]] 69; CHECK-NEXT: store i32 0, i32* [[DST_PTR]], align 4 70; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 71; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 15 72; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]] 73; 74entry: 75 br label %for.body 76 77for.cond.cleanup: 78 ret void 79 80for.body: 81 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 82 %dst.ptr = getelementptr inbounds i32, i32* %dst, i64 %indvars.iv 83 store i32 0, i32* %dst.ptr 84 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 85 %exitcond = icmp eq i64 %indvars.iv.next, 15 86 br i1 %exitcond, label %for.cond.cleanup, label %for.body 87} 88 89define void @VF1-VPWidenCanonicalIVRecipeExe(double* %ptr1) { 90; CHECK-LABEL: @VF1-VPWidenCanonicalIVRecipeExe( 91; CHECK-NEXT: entry: 92; CHECK-NEXT: [[PTR2:%.*]] = getelementptr inbounds double, double* [[PTR1:%.*]], i64 15 93; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 94; CHECK: vector.ph: 95; CHECK-NEXT: [[IND_END:%.*]] = getelementptr double, double* [[PTR1]], i64 16 96; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 97; CHECK: vector.body: 98; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE12:%.*]] ] 99; CHECK-NEXT: [[VEC_IV:%.*]] = add i64 [[INDEX]], 0 100; CHECK-NEXT: [[VEC_IV4:%.*]] = add i64 [[INDEX]], 1 101; CHECK-NEXT: [[VEC_IV5:%.*]] = add i64 [[INDEX]], 2 102; CHECK-NEXT: [[VEC_IV6:%.*]] = add i64 [[INDEX]], 3 103; CHECK-NEXT: [[TMP0:%.*]] = icmp ule i64 [[VEC_IV]], 14 104; CHECK-NEXT: [[TMP1:%.*]] = icmp ule i64 [[VEC_IV4]], 14 105; CHECK-NEXT: [[TMP2:%.*]] = icmp ule i64 [[VEC_IV5]], 14 106; CHECK-NEXT: [[TMP3:%.*]] = icmp ule i64 [[VEC_IV6]], 14 107; CHECK-NEXT: br i1 [[TMP0]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]] 108; CHECK: pred.store.if: 109; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0 110; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr double, double* [[PTR1]], i64 [[TMP4]] 111; CHECK-NEXT: store double 0.000000e+00, double* [[NEXT_GEP]], align 8 112; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]] 113; CHECK: pred.store.continue: 114; CHECK-NEXT: br i1 [[TMP1]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]] 115; CHECK: pred.store.if7: 116; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 1 117; CHECK-NEXT: [[NEXT_GEP1:%.*]] = getelementptr double, double* [[PTR1]], i64 [[TMP5]] 118; CHECK-NEXT: store double 0.000000e+00, double* [[NEXT_GEP1]], align 8 119; CHECK-NEXT: br label [[PRED_STORE_CONTINUE8]] 120; CHECK: pred.store.continue8: 121; CHECK-NEXT: br i1 [[TMP2]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10:%.*]] 122; CHECK: pred.store.if9: 123; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 2 124; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr double, double* [[PTR1]], i64 [[TMP6]] 125; CHECK-NEXT: store double 0.000000e+00, double* [[NEXT_GEP2]], align 8 126; CHECK-NEXT: br label [[PRED_STORE_CONTINUE10]] 127; CHECK: pred.store.continue10: 128; CHECK-NEXT: br i1 [[TMP3]], label [[PRED_STORE_IF11:%.*]], label [[PRED_STORE_CONTINUE12]] 129; CHECK: pred.store.if11: 130; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[INDEX]], 3 131; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr double, double* [[PTR1]], i64 [[TMP7]] 132; CHECK-NEXT: store double 0.000000e+00, double* [[NEXT_GEP3]], align 8 133; CHECK-NEXT: br label [[PRED_STORE_CONTINUE12]] 134; CHECK: pred.store.continue12: 135; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 136; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16 137; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 138; CHECK: middle.block: 139; CHECK-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]] 140; CHECK: scalar.ph: 141; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi double* [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[PTR1]], [[ENTRY:%.*]] ] 142; CHECK-NEXT: br label [[FOR_BODY:%.*]] 143; CHECK: for.cond.cleanup: 144; CHECK-NEXT: ret void 145; CHECK: for.body: 146; CHECK-NEXT: [[ADDR:%.*]] = phi double* [ [[PTR:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 147; CHECK-NEXT: store double 0.000000e+00, double* [[ADDR]], align 8 148; CHECK-NEXT: [[PTR]] = getelementptr inbounds double, double* [[ADDR]], i64 1 149; CHECK-NEXT: [[COND:%.*]] = icmp eq double* [[PTR]], [[PTR2]] 150; CHECK-NEXT: br i1 [[COND]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] 151; 152entry: 153 %ptr2 = getelementptr inbounds double, double* %ptr1, i64 15 154 br label %for.body 155 156for.cond.cleanup: 157 ret void 158 159for.body: 160 %addr = phi double* [ %ptr, %for.body ], [ %ptr1, %entry ] 161 store double 0.0, double* %addr 162 %ptr = getelementptr inbounds double, double* %addr, i64 1 163 %cond = icmp eq double* %ptr, %ptr2 164 br i1 %cond, label %for.cond.cleanup, label %for.body 165} 166