1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S | FileCheck %s 3target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" 4 5 6; Function Attrs: nofree norecurse nounwind 7define void @a(i8* readnone %b) { 8; CHECK-LABEL: @a( 9; CHECK-NEXT: entry: 10; CHECK-NEXT: [[B1:%.*]] = ptrtoint i8* [[B:%.*]] to i64 11; CHECK-NEXT: [[CMP_NOT4:%.*]] = icmp eq i8* [[B]], null 12; CHECK-NEXT: br i1 [[CMP_NOT4]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY_PREHEADER:%.*]] 13; CHECK: for.body.preheader: 14; CHECK-NEXT: [[TMP0:%.*]] = sub i64 0, [[B1]] 15; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 4 16; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 17; CHECK: vector.ph: 18; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 4 19; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]] 20; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[N_VEC]], -1 21; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, i8* null, i64 [[TMP1]] 22; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 23; CHECK: vector.body: 24; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi i8* [ null, [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[PRED_STORE_CONTINUE7:%.*]] ] 25; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE7]] ] 26; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, i8* [[POINTER_PHI]], <4 x i64> <i64 0, i64 -1, i64 -2, i64 -3> 27; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, <4 x i8*> [[TMP2]], i64 -1 28; CHECK-NEXT: [[TMP4:%.*]] = extractelement <4 x i8*> [[TMP3]], i32 0 29; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, i8* [[TMP4]], i32 0 30; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, i8* [[TMP5]], i32 -3 31; CHECK-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to <4 x i8>* 32; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, <4 x i8>* [[TMP7]], align 1 33; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 34; CHECK-NEXT: [[TMP8:%.*]] = icmp eq <4 x i8> [[REVERSE]], zeroinitializer 35; CHECK-NEXT: [[TMP9:%.*]] = xor <4 x i1> [[TMP8]], <i1 true, i1 true, i1 true, i1 true> 36; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x i1> [[TMP9]], i32 0 37; CHECK-NEXT: br i1 [[TMP10]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]] 38; CHECK: pred.store.if: 39; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x i8*> [[TMP3]], i32 0 40; CHECK-NEXT: store i8 95, i8* [[TMP11]], align 1 41; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]] 42; CHECK: pred.store.continue: 43; CHECK-NEXT: [[TMP12:%.*]] = extractelement <4 x i1> [[TMP9]], i32 1 44; CHECK-NEXT: br i1 [[TMP12]], label [[PRED_STORE_IF2:%.*]], label [[PRED_STORE_CONTINUE3:%.*]] 45; CHECK: pred.store.if2: 46; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x i8*> [[TMP3]], i32 1 47; CHECK-NEXT: store i8 95, i8* [[TMP13]], align 1 48; CHECK-NEXT: br label [[PRED_STORE_CONTINUE3]] 49; CHECK: pred.store.continue3: 50; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x i1> [[TMP9]], i32 2 51; CHECK-NEXT: br i1 [[TMP14]], label [[PRED_STORE_IF4:%.*]], label [[PRED_STORE_CONTINUE5:%.*]] 52; CHECK: pred.store.if4: 53; CHECK-NEXT: [[TMP15:%.*]] = extractelement <4 x i8*> [[TMP3]], i32 2 54; CHECK-NEXT: store i8 95, i8* [[TMP15]], align 1 55; CHECK-NEXT: br label [[PRED_STORE_CONTINUE5]] 56; CHECK: pred.store.continue5: 57; CHECK-NEXT: [[TMP16:%.*]] = extractelement <4 x i1> [[TMP9]], i32 3 58; CHECK-NEXT: br i1 [[TMP16]], label [[PRED_STORE_IF6:%.*]], label [[PRED_STORE_CONTINUE7]] 59; CHECK: pred.store.if6: 60; CHECK-NEXT: [[TMP17:%.*]] = extractelement <4 x i8*> [[TMP3]], i32 3 61; CHECK-NEXT: store i8 95, i8* [[TMP17]], align 1 62; CHECK-NEXT: br label [[PRED_STORE_CONTINUE7]] 63; CHECK: pred.store.continue7: 64; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 65; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 66; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, i8* [[POINTER_PHI]], i64 -4 67; CHECK-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 68; CHECK: middle.block: 69; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] 70; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]] 71; CHECK: scalar.ph: 72; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8* [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ null, [[FOR_BODY_PREHEADER]] ] 73; CHECK-NEXT: br label [[FOR_BODY:%.*]] 74; CHECK: for.cond.cleanup.loopexit: 75; CHECK-NEXT: br label [[FOR_COND_CLEANUP]] 76; CHECK: for.cond.cleanup: 77; CHECK-NEXT: ret void 78; CHECK: for.body: 79; CHECK-NEXT: [[C_05:%.*]] = phi i8* [ [[INCDEC_PTR:%.*]], [[IF_END:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 80; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds i8, i8* [[C_05]], i64 -1 81; CHECK-NEXT: [[TMP19:%.*]] = load i8, i8* [[INCDEC_PTR]], align 1 82; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i8 [[TMP19]], 0 83; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label [[IF_END]], label [[IF_THEN:%.*]] 84; CHECK: if.then: 85; CHECK-NEXT: store i8 95, i8* [[INCDEC_PTR]], align 1 86; CHECK-NEXT: br label [[IF_END]] 87; CHECK: if.end: 88; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i8* [[INCDEC_PTR]], [[B]] 89; CHECK-NEXT: br i1 [[CMP_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]] 90; 91 92entry: 93 %cmp.not4 = icmp eq i8* %b, null 94 br i1 %cmp.not4, label %for.cond.cleanup, label %for.body.preheader 95 96for.body.preheader: ; preds = %entry 97 br label %for.body 98 99for.cond.cleanup.loopexit: ; preds = %if.end 100 br label %for.cond.cleanup 101 102for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry 103 ret void 104 105for.body: ; preds = %for.body.preheader, %if.end 106 %c.05 = phi i8* [ %incdec.ptr, %if.end ], [ null, %for.body.preheader ] 107 %incdec.ptr = getelementptr inbounds i8, i8* %c.05, i64 -1 108 %0 = load i8, i8* %incdec.ptr, align 1 109 %tobool.not = icmp eq i8 %0, 0 110 br i1 %tobool.not, label %if.end, label %if.then 111 112if.then: ; preds = %for.body 113 store i8 95, i8* %incdec.ptr, align 1 114 br label %if.end 115 116if.end: ; preds = %for.body, %if.then 117 %cmp.not = icmp eq i8* %incdec.ptr, %b 118 br i1 %cmp.not, label %for.cond.cleanup.loopexit, label %for.body 119} 120 121; In the test below the pointer phi %ptr.iv.2 is used as 122; 1. As a uniform address for the load, and 123; 2. Non-uniform use by the getelementptr which is stored. This requires the 124; vector value. 125define void @pointer_induction_used_as_vector(i8** noalias %start.1, i8* noalias %start.2, i64 %N) { 126; CHECK-LABEL: @pointer_induction_used_as_vector( 127; CHECK-NEXT: entry: 128; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4 129; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 130; CHECK: vector.ph: 131; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 132; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 133; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8*, i8** [[START_1:%.*]], i64 [[N_VEC]] 134; CHECK-NEXT: [[IND_END3:%.*]] = getelementptr i8, i8* [[START_2:%.*]], i64 [[N_VEC]] 135; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 136; CHECK: vector.body: 137; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 138; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 139; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 0 140; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8*, i8** [[START_1]], i64 [[TMP1]] 141; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 0 142; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, i8* [[START_2]], i64 [[TMP2]] 143; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 1 144; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, i8* [[START_2]], i64 [[TMP3]] 145; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 2 146; CHECK-NEXT: [[NEXT_GEP6:%.*]] = getelementptr i8, i8* [[START_2]], i64 [[TMP4]] 147; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 3 148; CHECK-NEXT: [[NEXT_GEP7:%.*]] = getelementptr i8, i8* [[START_2]], i64 [[TMP5]] 149; CHECK-NEXT: [[TMP6:%.*]] = insertelement <4 x i8*> poison, i8* [[NEXT_GEP4]], i32 0 150; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x i8*> [[TMP6]], i8* [[NEXT_GEP5]], i32 1 151; CHECK-NEXT: [[TMP8:%.*]] = insertelement <4 x i8*> [[TMP7]], i8* [[NEXT_GEP6]], i32 2 152; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x i8*> [[TMP8]], i8* [[NEXT_GEP7]], i32 3 153; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, <4 x i8*> [[TMP9]], i64 1 154; CHECK-NEXT: [[TMP11:%.*]] = getelementptr i8*, i8** [[NEXT_GEP]], i32 0 155; CHECK-NEXT: [[TMP12:%.*]] = bitcast i8** [[TMP11]] to <4 x i8*>* 156; CHECK-NEXT: store <4 x i8*> [[TMP10]], <4 x i8*>* [[TMP12]], align 8 157; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, i8* [[NEXT_GEP4]], i32 0 158; CHECK-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to <4 x i8>* 159; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, <4 x i8>* [[TMP14]], align 1 160; CHECK-NEXT: [[TMP15:%.*]] = add <4 x i8> [[WIDE_LOAD]], <i8 1, i8 1, i8 1, i8 1> 161; CHECK-NEXT: [[TMP16:%.*]] = bitcast i8* [[TMP13]] to <4 x i8>* 162; CHECK-NEXT: store <4 x i8> [[TMP15]], <4 x i8>* [[TMP16]], align 1 163; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 164; CHECK-NEXT: [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 165; CHECK-NEXT: br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] 166; CHECK: middle.block: 167; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 168; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 169; CHECK: scalar.ph: 170; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 171; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i8** [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[START_1]], [[ENTRY]] ] 172; CHECK-NEXT: [[BC_RESUME_VAL2:%.*]] = phi i8* [ [[IND_END3]], [[MIDDLE_BLOCK]] ], [ [[START_2]], [[ENTRY]] ] 173; CHECK-NEXT: br label [[LOOP_BODY:%.*]] 174; CHECK: loop.body: 175; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_BODY]] ] 176; CHECK-NEXT: [[PTR_IV_1:%.*]] = phi i8** [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[PTR_IV_1_NEXT:%.*]], [[LOOP_BODY]] ] 177; CHECK-NEXT: [[PTR_IV_2:%.*]] = phi i8* [ [[BC_RESUME_VAL2]], [[SCALAR_PH]] ], [ [[PTR_IV_2_NEXT:%.*]], [[LOOP_BODY]] ] 178; CHECK-NEXT: [[PTR_IV_1_NEXT]] = getelementptr inbounds i8*, i8** [[PTR_IV_1]], i64 1 179; CHECK-NEXT: [[PTR_IV_2_NEXT]] = getelementptr inbounds i8, i8* [[PTR_IV_2]], i64 1 180; CHECK-NEXT: store i8* [[PTR_IV_2_NEXT]], i8** [[PTR_IV_1]], align 8 181; CHECK-NEXT: [[LV:%.*]] = load i8, i8* [[PTR_IV_2]], align 1 182; CHECK-NEXT: [[ADD:%.*]] = add i8 [[LV]], 1 183; CHECK-NEXT: store i8 [[ADD]], i8* [[PTR_IV_2]], align 1 184; CHECK-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1 185; CHECK-NEXT: [[C:%.*]] = icmp ne i64 [[IV_NEXT]], [[N]] 186; CHECK-NEXT: br i1 [[C]], label [[LOOP_BODY]], label [[EXIT]], !llvm.loop [[LOOP5:![0-9]+]] 187; CHECK: exit: 188; CHECK-NEXT: ret void 189; 190 191entry: 192 br label %loop.body 193 194loop.body: ; preds = %loop.body, %entry 195 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.body ] 196 %ptr.iv.1 = phi i8** [ %start.1, %entry ], [ %ptr.iv.1.next, %loop.body ] 197 %ptr.iv.2 = phi i8* [ %start.2, %entry ], [ %ptr.iv.2.next, %loop.body ] 198 %ptr.iv.1.next = getelementptr inbounds i8*, i8** %ptr.iv.1, i64 1 199 %ptr.iv.2.next = getelementptr inbounds i8, i8* %ptr.iv.2, i64 1 200 store i8* %ptr.iv.2.next, i8** %ptr.iv.1, align 8 201 %lv = load i8, i8* %ptr.iv.2, align 1 202 %add = add i8 %lv, 1 203 store i8 %add, i8* %ptr.iv.2, align 1 204 %iv.next = add nuw i64 %iv, 1 205 %c = icmp ne i64 %iv.next, %N 206 br i1 %c, label %loop.body, label %exit 207 208exit: ; preds = %loop.body 209 ret void 210} 211