1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -loop-vectorize -force-vector-width=4 -S < %s | FileCheck %s 3 4; This is the test case from PR26314. 5; When we were retrying dependence checking with memchecks only, 6; the loop-invariant access in the inner loop was incorrectly determined to be wrapping 7; because it was not strided in the inner loop. 8; Improved wrapping detection allows vectorization in the following case. 9 10; #define Z 32 11; typedef struct s { 12; int v1[Z]; 13; int v2[Z]; 14; int v3[Z][Z]; 15; } s; 16; 17; void slow_function (s* const obj, int z) { 18; for (int j=0; j<Z; j++) { 19; for (int k=0; k<z; k++) { 20; int x = obj->v1[k] + obj->v2[j]; 21; obj->v3[j][k] += x; 22; } 23; } 24; } 25 26target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 27 28%struct.s = type { [32 x i32], [32 x i32], [32 x [32 x i32]] } 29 30define void @Test(%struct.s* nocapture %obj, i64 %z) #0 { 31; CHECK-LABEL: @Test( 32; CHECK-NEXT: [[OBJ4:%.*]] = bitcast %struct.s* [[OBJ:%.*]] to i8* 33; CHECK-NEXT: [[SCEVGEP5:%.*]] = getelementptr [[STRUCT_S:%.*]], %struct.s* [[OBJ]], i64 0, i32 0, i64 [[Z:%.*]] 34; CHECK-NEXT: [[SCEVGEP56:%.*]] = bitcast i32* [[SCEVGEP5]] to i8* 35; CHECK-NEXT: br label [[DOTOUTER_PREHEADER:%.*]] 36; CHECK: .outer.preheader: 37; CHECK-NEXT: [[I:%.*]] = phi i64 [ 0, [[TMP0:%.*]] ], [ [[I_NEXT:%.*]], [[DOTOUTER:%.*]] ] 38; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr [[STRUCT_S]], %struct.s* [[OBJ]], i64 0, i32 2, i64 [[I]], i64 0 39; CHECK-NEXT: [[SCEVGEP1:%.*]] = bitcast i32* [[SCEVGEP]] to i8* 40; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr [[STRUCT_S]], %struct.s* [[OBJ]], i64 0, i32 2, i64 [[I]], i64 [[Z]] 41; CHECK-NEXT: [[SCEVGEP23:%.*]] = bitcast i32* [[SCEVGEP2]] to i8* 42; CHECK-NEXT: [[SCEVGEP7:%.*]] = getelementptr [[STRUCT_S]], %struct.s* [[OBJ]], i64 0, i32 1, i64 [[I]] 43; CHECK-NEXT: [[SCEVGEP78:%.*]] = bitcast i32* [[SCEVGEP7]] to i8* 44; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[I]], 1 45; CHECK-NEXT: [[SCEVGEP9:%.*]] = getelementptr [[STRUCT_S]], %struct.s* [[OBJ]], i64 0, i32 1, i64 [[TMP1]] 46; CHECK-NEXT: [[SCEVGEP910:%.*]] = bitcast i32* [[SCEVGEP9]] to i8* 47; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.s* [[OBJ]], i64 0, i32 1, i64 [[I]] 48; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[Z]], 4 49; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] 50; CHECK: vector.memcheck: 51; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult i8* [[SCEVGEP1]], [[SCEVGEP56]] 52; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult i8* [[OBJ4]], [[SCEVGEP23]] 53; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 54; CHECK-NEXT: [[BOUND011:%.*]] = icmp ult i8* [[SCEVGEP1]], [[SCEVGEP910]] 55; CHECK-NEXT: [[BOUND112:%.*]] = icmp ult i8* [[SCEVGEP78]], [[SCEVGEP23]] 56; CHECK-NEXT: [[FOUND_CONFLICT13:%.*]] = and i1 [[BOUND011]], [[BOUND112]] 57; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT13]] 58; CHECK-NEXT: [[MEMCHECK_CONFLICT:%.*]] = and i1 [[CONFLICT_RDX]], true 59; CHECK-NEXT: br i1 [[MEMCHECK_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] 60; CHECK: vector.ph: 61; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[Z]], 4 62; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[Z]], [[N_MOD_VF]] 63; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 64; CHECK: vector.body: 65; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 66; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0 67; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.s* [[OBJ]], i64 0, i32 0, i64 [[TMP3]] 68; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TMP4]], i32 0 69; CHECK-NEXT: [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <4 x i32>* 70; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, <4 x i32>* [[TMP6]], align 4, !alias.scope !0 71; CHECK-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP2]], align 4, !alias.scope !3 72; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TMP7]], i32 0 73; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer 74; CHECK-NEXT: [[TMP8:%.*]] = add nsw <4 x i32> [[BROADCAST_SPLAT]], [[WIDE_LOAD]] 75; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.s* [[OBJ]], i64 0, i32 2, i64 [[I]], i64 [[TMP3]] 76; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TMP9]], i32 0 77; CHECK-NEXT: [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <4 x i32>* 78; CHECK-NEXT: [[WIDE_LOAD14:%.*]] = load <4 x i32>, <4 x i32>* [[TMP11]], align 4, !alias.scope !5, !noalias !7 79; CHECK-NEXT: [[TMP12:%.*]] = add nsw <4 x i32> [[TMP8]], [[WIDE_LOAD14]] 80; CHECK-NEXT: [[TMP13:%.*]] = bitcast i32* [[TMP10]] to <4 x i32>* 81; CHECK-NEXT: store <4 x i32> [[TMP12]], <4 x i32>* [[TMP13]], align 4, !alias.scope !5, !noalias !7 82; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 83; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 84; CHECK-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] 85; CHECK: middle.block: 86; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[Z]], [[N_VEC]] 87; CHECK-NEXT: br i1 [[CMP_N]], label [[DOTOUTER]], label [[SCALAR_PH]] 88; CHECK: scalar.ph: 89; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[DOTOUTER_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ] 90; CHECK-NEXT: br label [[DOTINNER:%.*]] 91; CHECK: .exit: 92; CHECK-NEXT: ret void 93; CHECK: .outer: 94; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 95; CHECK-NEXT: [[EXITCOND_OUTER:%.*]] = icmp eq i64 [[I_NEXT]], 32 96; CHECK-NEXT: br i1 [[EXITCOND_OUTER]], label [[DOTEXIT:%.*]], label [[DOTOUTER_PREHEADER]] 97; CHECK: .inner: 98; CHECK-NEXT: [[J:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[J_NEXT:%.*]], [[DOTINNER]] ] 99; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.s* [[OBJ]], i64 0, i32 0, i64 [[J]] 100; CHECK-NEXT: [[TMP16:%.*]] = load i32, i32* [[TMP15]], align 4 101; CHECK-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP2]], align 4 102; CHECK-NEXT: [[TMP18:%.*]] = add nsw i32 [[TMP17]], [[TMP16]] 103; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.s* [[OBJ]], i64 0, i32 2, i64 [[I]], i64 [[J]] 104; CHECK-NEXT: [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4 105; CHECK-NEXT: [[TMP21:%.*]] = add nsw i32 [[TMP18]], [[TMP20]] 106; CHECK-NEXT: store i32 [[TMP21]], i32* [[TMP19]], align 4 107; CHECK-NEXT: [[J_NEXT]] = add nuw nsw i64 [[J]], 1 108; CHECK-NEXT: [[EXITCOND_INNER:%.*]] = icmp eq i64 [[J_NEXT]], [[Z]] 109; CHECK-NEXT: br i1 [[EXITCOND_INNER]], label [[DOTOUTER]], label [[DOTINNER]], !llvm.loop [[LOOP10:![0-9]+]] 110; 111 br label %.outer.preheader 112 113 114.outer.preheader: 115 %i = phi i64 [ 0, %0 ], [ %i.next, %.outer ] 116 %1 = getelementptr inbounds %struct.s, %struct.s* %obj, i64 0, i32 1, i64 %i 117 br label %.inner 118 119.exit: 120 ret void 121 122.outer: 123 %i.next = add nuw nsw i64 %i, 1 124 %exitcond.outer = icmp eq i64 %i.next, 32 125 br i1 %exitcond.outer, label %.exit, label %.outer.preheader 126 127.inner: 128 %j = phi i64 [ 0, %.outer.preheader ], [ %j.next, %.inner ] 129 %2 = getelementptr inbounds %struct.s, %struct.s* %obj, i64 0, i32 0, i64 %j 130 %3 = load i32, i32* %2 131 %4 = load i32, i32* %1 132 %5 = add nsw i32 %4, %3 133 %6 = getelementptr inbounds %struct.s, %struct.s* %obj, i64 0, i32 2, i64 %i, i64 %j 134 %7 = load i32, i32* %6 135 %8 = add nsw i32 %5, %7 136 store i32 %8, i32* %6 137 %j.next = add nuw nsw i64 %j, 1 138 %exitcond.inner = icmp eq i64 %j.next, %z 139 br i1 %exitcond.inner, label %.outer, label %.inner 140} 141