1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -loop-vectorize -S -mattr=avx512f -instcombine < %s | FileCheck %s 3 4target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 5target triple = "x86_64-unknown-linux-gnu" 6 7; first test checks that loop with a reduction and a uniform store gets 8; vectorized. 9 10define i32 @inv_val_store_to_inv_address_with_reduction(i32* %a, i64 %n, i32* %b) { 11; CHECK-LABEL: @inv_val_store_to_inv_address_with_reduction( 12; CHECK-NEXT: iter.check: 13; CHECK-NEXT: [[NTRUNC:%.*]] = trunc i64 [[N:%.*]] to i32 14; CHECK-NEXT: [[SMAX6:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 15; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX6]], 8 16; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] 17; CHECK: vector.memcheck: 18; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i32, i32* [[A:%.*]], i64 1 19; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 20; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr i32, i32* [[B:%.*]], i64 [[SMAX]] 21; CHECK-NEXT: [[BOUND0:%.*]] = icmp ugt i32* [[SCEVGEP4]], [[A]] 22; CHECK-NEXT: [[BOUND1:%.*]] = icmp ugt i32* [[SCEVGEP]], [[B]] 23; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 24; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]] 25; CHECK: vector.main.loop.iter.check: 26; CHECK-NEXT: [[MIN_ITERS_CHECK7:%.*]] = icmp ult i64 [[SMAX6]], 64 27; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK7]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 28; CHECK: vector.ph: 29; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX6]], 9223372036854775744 30; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 31; CHECK: vector.body: 32; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 33; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP8:%.*]], [[VECTOR_BODY]] ] 34; CHECK-NEXT: [[VEC_PHI8:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ] 35; CHECK-NEXT: [[VEC_PHI9:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP10:%.*]], [[VECTOR_BODY]] ] 36; CHECK-NEXT: [[VEC_PHI10:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP11:%.*]], [[VECTOR_BODY]] ] 37; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDEX]] 38; CHECK-NEXT: [[TMP1:%.*]] = bitcast i32* [[TMP0]] to <16 x i32>* 39; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i32>, <16 x i32>* [[TMP1]], align 8, !alias.scope !0 40; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP0]], i64 16 41; CHECK-NEXT: [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <16 x i32>* 42; CHECK-NEXT: [[WIDE_LOAD11:%.*]] = load <16 x i32>, <16 x i32>* [[TMP3]], align 8, !alias.scope !0 43; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP0]], i64 32 44; CHECK-NEXT: [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <16 x i32>* 45; CHECK-NEXT: [[WIDE_LOAD12:%.*]] = load <16 x i32>, <16 x i32>* [[TMP5]], align 8, !alias.scope !0 46; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[TMP0]], i64 48 47; CHECK-NEXT: [[TMP7:%.*]] = bitcast i32* [[TMP6]] to <16 x i32>* 48; CHECK-NEXT: [[WIDE_LOAD13:%.*]] = load <16 x i32>, <16 x i32>* [[TMP7]], align 8, !alias.scope !0 49; CHECK-NEXT: [[TMP8]] = add <16 x i32> [[VEC_PHI]], [[WIDE_LOAD]] 50; CHECK-NEXT: [[TMP9]] = add <16 x i32> [[VEC_PHI8]], [[WIDE_LOAD11]] 51; CHECK-NEXT: [[TMP10]] = add <16 x i32> [[VEC_PHI9]], [[WIDE_LOAD12]] 52; CHECK-NEXT: [[TMP11]] = add <16 x i32> [[VEC_PHI10]], [[WIDE_LOAD13]] 53; CHECK-NEXT: store i32 [[NTRUNC]], i32* [[A]], align 4, !alias.scope !3, !noalias !0 54; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 64 55; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 56; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 57; CHECK: middle.block: 58; CHECK-NEXT: [[BIN_RDX:%.*]] = add <16 x i32> [[TMP9]], [[TMP8]] 59; CHECK-NEXT: [[BIN_RDX14:%.*]] = add <16 x i32> [[TMP10]], [[BIN_RDX]] 60; CHECK-NEXT: [[BIN_RDX15:%.*]] = add <16 x i32> [[TMP11]], [[BIN_RDX14]] 61; CHECK-NEXT: [[TMP13:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[BIN_RDX15]]) 62; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX6]], [[N_VEC]] 63; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 64; CHECK: vec.epilog.iter.check: 65; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = and i64 [[SMAX6]], 56 66; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0 67; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 68; CHECK: vec.epilog.ph: 69; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ], [ [[TMP13]], [[VEC_EPILOG_ITER_CHECK]] ] 70; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ] 71; CHECK-NEXT: [[SMAX16:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 72; CHECK-NEXT: [[N_VEC18:%.*]] = and i64 [[SMAX16]], 9223372036854775800 73; CHECK-NEXT: [[TMP14:%.*]] = insertelement <8 x i32> <i32 poison, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>, i32 [[BC_MERGE_RDX]], i64 0 74; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 75; CHECK: vec.epilog.vector.body: 76; CHECK-NEXT: [[INDEX20:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT23:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 77; CHECK-NEXT: [[VEC_PHI21:%.*]] = phi <8 x i32> [ [[TMP14]], [[VEC_EPILOG_PH]] ], [ [[TMP17:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 78; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDEX20]] 79; CHECK-NEXT: [[TMP16:%.*]] = bitcast i32* [[TMP15]] to <8 x i32>* 80; CHECK-NEXT: [[WIDE_LOAD22:%.*]] = load <8 x i32>, <8 x i32>* [[TMP16]], align 8 81; CHECK-NEXT: [[TMP17]] = add <8 x i32> [[VEC_PHI21]], [[WIDE_LOAD22]] 82; CHECK-NEXT: store i32 [[NTRUNC]], i32* [[A]], align 4 83; CHECK-NEXT: [[INDEX_NEXT23]] = add nuw i64 [[INDEX20]], 8 84; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT23]], [[N_VEC18]] 85; CHECK-NEXT: br i1 [[TMP18]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] 86; CHECK: vec.epilog.middle.block: 87; CHECK-NEXT: [[TMP19:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP17]]) 88; CHECK-NEXT: [[CMP_N19:%.*]] = icmp eq i64 [[SMAX16]], [[N_VEC18]] 89; CHECK-NEXT: br i1 [[CMP_N19]], label [[FOR_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_SCALAR_PH]] 90; CHECK: vec.epilog.scalar.ph: 91; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC18]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ITER_CHECK:%.*]] ] 92; CHECK-NEXT: [[BC_MERGE_RDX24:%.*]] = phi i32 [ [[TMP19]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[TMP13]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ITER_CHECK]] ] 93; CHECK-NEXT: br label [[FOR_BODY:%.*]] 94; CHECK: for.body: 95; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ] 96; CHECK-NEXT: [[T0:%.*]] = phi i32 [ [[T3:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX24]], [[VEC_EPILOG_SCALAR_PH]] ] 97; CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[I]] 98; CHECK-NEXT: [[T2:%.*]] = load i32, i32* [[T1]], align 8 99; CHECK-NEXT: [[T3]] = add i32 [[T0]], [[T2]] 100; CHECK-NEXT: store i32 [[NTRUNC]], i32* [[A]], align 4 101; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 102; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] 103; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]], !llvm.loop [[LOOP9:![0-9]+]] 104; CHECK: for.end.loopexit: 105; CHECK-NEXT: [[T3_LCSSA:%.*]] = phi i32 [ [[T3]], [[FOR_BODY]] ], [ [[TMP19]], [[VEC_EPILOG_MIDDLE_BLOCK]] ] 106; CHECK-NEXT: br label [[FOR_END]] 107; CHECK: for.end: 108; CHECK-NEXT: [[T4:%.*]] = phi i32 [ [[TMP13]], [[MIDDLE_BLOCK]] ], [ [[T3_LCSSA]], [[FOR_END_LOOPEXIT]] ] 109; CHECK-NEXT: ret i32 [[T4]] 110; 111entry: 112 %ntrunc = trunc i64 %n to i32 113 br label %for.body 114 115for.body: ; preds = %for.body, %entry 116 %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] 117 %t0 = phi i32 [ %t3, %for.body ], [ 0, %entry ] 118 %t1 = getelementptr inbounds i32, i32* %b, i64 %i 119 %t2 = load i32, i32* %t1, align 8 120 %t3 = add i32 %t0, %t2 121 store i32 %ntrunc, i32* %a 122 %i.next = add nuw nsw i64 %i, 1 123 %cond = icmp slt i64 %i.next, %n 124 br i1 %cond, label %for.body, label %for.end 125 126for.end: ; preds = %for.body 127 %t4 = phi i32 [ %t3, %for.body ] 128 ret i32 %t4 129} 130 131; Conditional store 132; if (b[i] == k) a = ntrunc 133define void @inv_val_store_to_inv_address_conditional(i32* %a, i64 %n, i32* %b, i32 %k) { 134; CHECK-LABEL: @inv_val_store_to_inv_address_conditional( 135; CHECK-NEXT: iter.check: 136; CHECK-NEXT: [[NTRUNC:%.*]] = trunc i64 [[N:%.*]] to i32 137; CHECK-NEXT: [[SMAX6:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 138; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX6]], 8 139; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] 140; CHECK: vector.memcheck: 141; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 142; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i32, i32* [[B:%.*]], i64 [[SMAX]] 143; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr i32, i32* [[A:%.*]], i64 1 144; CHECK-NEXT: [[BOUND0:%.*]] = icmp ugt i32* [[SCEVGEP4]], [[B]] 145; CHECK-NEXT: [[BOUND1:%.*]] = icmp ugt i32* [[SCEVGEP]], [[A]] 146; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 147; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]] 148; CHECK: vector.main.loop.iter.check: 149; CHECK-NEXT: [[MIN_ITERS_CHECK7:%.*]] = icmp ult i64 [[SMAX6]], 16 150; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK7]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 151; CHECK: vector.ph: 152; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX6]], 9223372036854775792 153; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i32> poison, i32 [[K:%.*]], i64 0 154; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT]], <16 x i32> poison, <16 x i32> zeroinitializer 155; CHECK-NEXT: [[BROADCAST_SPLATINSERT8:%.*]] = insertelement <16 x i32> poison, i32 [[NTRUNC]], i64 0 156; CHECK-NEXT: [[BROADCAST_SPLAT9:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT8]], <16 x i32> poison, <16 x i32> zeroinitializer 157; CHECK-NEXT: [[BROADCAST_SPLATINSERT10:%.*]] = insertelement <16 x i32*> poison, i32* [[A]], i64 0 158; CHECK-NEXT: [[BROADCAST_SPLAT11:%.*]] = shufflevector <16 x i32*> [[BROADCAST_SPLATINSERT10]], <16 x i32*> poison, <16 x i32> zeroinitializer 159; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 160; CHECK: vector.body: 161; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 162; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDEX]] 163; CHECK-NEXT: [[TMP1:%.*]] = bitcast i32* [[TMP0]] to <16 x i32>* 164; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i32>, <16 x i32>* [[TMP1]], align 8, !alias.scope !10, !noalias !13 165; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <16 x i32> [[WIDE_LOAD]], [[BROADCAST_SPLAT]] 166; CHECK-NEXT: [[TMP3:%.*]] = bitcast i32* [[TMP0]] to <16 x i32>* 167; CHECK-NEXT: store <16 x i32> [[BROADCAST_SPLAT9]], <16 x i32>* [[TMP3]], align 4, !alias.scope !10, !noalias !13 168; CHECK-NEXT: call void @llvm.masked.scatter.v16i32.v16p0i32(<16 x i32> [[BROADCAST_SPLAT9]], <16 x i32*> [[BROADCAST_SPLAT11]], i32 4, <16 x i1> [[TMP2]]), !alias.scope !13 169; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 170; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 171; CHECK-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]] 172; CHECK: middle.block: 173; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX6]], [[N_VEC]] 174; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 175; CHECK: vec.epilog.iter.check: 176; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = and i64 [[SMAX6]], 8 177; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK_NOT_NOT:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0 178; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK_NOT_NOT]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 179; CHECK: vec.epilog.ph: 180; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 181; CHECK-NEXT: [[SMAX12:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 182; CHECK-NEXT: [[N_VEC14:%.*]] = and i64 [[SMAX12]], 9223372036854775800 183; CHECK-NEXT: [[BROADCAST_SPLATINSERT18:%.*]] = insertelement <8 x i32> poison, i32 [[K]], i64 0 184; CHECK-NEXT: [[BROADCAST_SPLAT19:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT18]], <8 x i32> poison, <8 x i32> zeroinitializer 185; CHECK-NEXT: [[BROADCAST_SPLATINSERT20:%.*]] = insertelement <8 x i32> poison, i32 [[NTRUNC]], i64 0 186; CHECK-NEXT: [[BROADCAST_SPLAT21:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT20]], <8 x i32> poison, <8 x i32> zeroinitializer 187; CHECK-NEXT: [[BROADCAST_SPLATINSERT22:%.*]] = insertelement <8 x i32*> poison, i32* [[A]], i64 0 188; CHECK-NEXT: [[BROADCAST_SPLAT23:%.*]] = shufflevector <8 x i32*> [[BROADCAST_SPLATINSERT22]], <8 x i32*> poison, <8 x i32> zeroinitializer 189; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 190; CHECK: vec.epilog.vector.body: 191; CHECK-NEXT: [[INDEX16:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT24:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 192; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDEX16]] 193; CHECK-NEXT: [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <8 x i32>* 194; CHECK-NEXT: [[WIDE_LOAD17:%.*]] = load <8 x i32>, <8 x i32>* [[TMP6]], align 8 195; CHECK-NEXT: [[TMP7:%.*]] = icmp eq <8 x i32> [[WIDE_LOAD17]], [[BROADCAST_SPLAT19]] 196; CHECK-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP5]] to <8 x i32>* 197; CHECK-NEXT: store <8 x i32> [[BROADCAST_SPLAT21]], <8 x i32>* [[TMP8]], align 4 198; CHECK-NEXT: call void @llvm.masked.scatter.v8i32.v8p0i32(<8 x i32> [[BROADCAST_SPLAT21]], <8 x i32*> [[BROADCAST_SPLAT23]], i32 4, <8 x i1> [[TMP7]]) 199; CHECK-NEXT: [[INDEX_NEXT24]] = add nuw i64 [[INDEX16]], 8 200; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT24]], [[N_VEC14]] 201; CHECK-NEXT: br i1 [[TMP9]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]] 202; CHECK: vec.epilog.middle.block: 203; CHECK-NEXT: [[CMP_N15:%.*]] = icmp eq i64 [[SMAX12]], [[N_VEC14]] 204; CHECK-NEXT: br i1 [[CMP_N15]], label [[FOR_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_SCALAR_PH]] 205; CHECK: vec.epilog.scalar.ph: 206; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC14]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ITER_CHECK:%.*]] ] 207; CHECK-NEXT: br label [[FOR_BODY:%.*]] 208; CHECK: for.body: 209; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[LATCH:%.*]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ] 210; CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[I]] 211; CHECK-NEXT: [[T2:%.*]] = load i32, i32* [[T1]], align 8 212; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[T2]], [[K]] 213; CHECK-NEXT: store i32 [[NTRUNC]], i32* [[T1]], align 4 214; CHECK-NEXT: br i1 [[CMP]], label [[COND_STORE:%.*]], label [[LATCH]] 215; CHECK: cond_store: 216; CHECK-NEXT: store i32 [[NTRUNC]], i32* [[A]], align 4 217; CHECK-NEXT: br label [[LATCH]] 218; CHECK: latch: 219; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 220; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] 221; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]], !llvm.loop [[LOOP17:![0-9]+]] 222; CHECK: for.end.loopexit: 223; CHECK-NEXT: br label [[FOR_END]] 224; CHECK: for.end: 225; CHECK-NEXT: ret void 226; 227entry: 228 %ntrunc = trunc i64 %n to i32 229 br label %for.body 230 231for.body: ; preds = %for.body, %entry 232 %i = phi i64 [ %i.next, %latch ], [ 0, %entry ] 233 %t1 = getelementptr inbounds i32, i32* %b, i64 %i 234 %t2 = load i32, i32* %t1, align 8 235 %cmp = icmp eq i32 %t2, %k 236 store i32 %ntrunc, i32* %t1 237 br i1 %cmp, label %cond_store, label %latch 238 239cond_store: 240 store i32 %ntrunc, i32* %a 241 br label %latch 242 243latch: 244 %i.next = add nuw nsw i64 %i, 1 245 %cond = icmp slt i64 %i.next, %n 246 br i1 %cond, label %for.body, label %for.end 247 248for.end: ; preds = %for.body 249 ret void 250} 251 252define void @variant_val_store_to_inv_address_conditional(i32* %a, i64 %n, i32* %b, i32* %c, i32 %k) { 253; CHECK-LABEL: @variant_val_store_to_inv_address_conditional( 254; CHECK-NEXT: iter.check: 255; CHECK-NEXT: [[NTRUNC:%.*]] = trunc i64 [[N:%.*]] to i32 256; CHECK-NEXT: [[SMAX16:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 257; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX16]], 8 258; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] 259; CHECK: vector.memcheck: 260; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 261; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i32, i32* [[B:%.*]], i64 [[SMAX]] 262; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr i32, i32* [[A:%.*]], i64 1 263; CHECK-NEXT: [[SCEVGEP7:%.*]] = getelementptr i32, i32* [[C:%.*]], i64 [[SMAX]] 264; CHECK-NEXT: [[BOUND0:%.*]] = icmp ugt i32* [[SCEVGEP4]], [[B]] 265; CHECK-NEXT: [[BOUND1:%.*]] = icmp ugt i32* [[SCEVGEP]], [[A]] 266; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 267; CHECK-NEXT: [[BOUND09:%.*]] = icmp ugt i32* [[SCEVGEP7]], [[B]] 268; CHECK-NEXT: [[BOUND110:%.*]] = icmp ugt i32* [[SCEVGEP]], [[C]] 269; CHECK-NEXT: [[FOUND_CONFLICT11:%.*]] = and i1 [[BOUND09]], [[BOUND110]] 270; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT11]] 271; CHECK-NEXT: [[BOUND012:%.*]] = icmp ugt i32* [[SCEVGEP7]], [[A]] 272; CHECK-NEXT: [[BOUND113:%.*]] = icmp ugt i32* [[SCEVGEP4]], [[C]] 273; CHECK-NEXT: [[FOUND_CONFLICT14:%.*]] = and i1 [[BOUND012]], [[BOUND113]] 274; CHECK-NEXT: [[CONFLICT_RDX15:%.*]] = or i1 [[CONFLICT_RDX]], [[FOUND_CONFLICT14]] 275; CHECK-NEXT: br i1 [[CONFLICT_RDX15]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]] 276; CHECK: vector.main.loop.iter.check: 277; CHECK-NEXT: [[MIN_ITERS_CHECK17:%.*]] = icmp ult i64 [[SMAX16]], 16 278; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK17]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 279; CHECK: vector.ph: 280; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX16]], 9223372036854775792 281; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i32> poison, i32 [[K:%.*]], i64 0 282; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT]], <16 x i32> poison, <16 x i32> zeroinitializer 283; CHECK-NEXT: [[BROADCAST_SPLATINSERT18:%.*]] = insertelement <16 x i32> poison, i32 [[NTRUNC]], i64 0 284; CHECK-NEXT: [[BROADCAST_SPLAT19:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT18]], <16 x i32> poison, <16 x i32> zeroinitializer 285; CHECK-NEXT: [[BROADCAST_SPLATINSERT20:%.*]] = insertelement <16 x i32*> poison, i32* [[A]], i64 0 286; CHECK-NEXT: [[BROADCAST_SPLAT21:%.*]] = shufflevector <16 x i32*> [[BROADCAST_SPLATINSERT20]], <16 x i32*> poison, <16 x i32> zeroinitializer 287; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 288; CHECK: vector.body: 289; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 290; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDEX]] 291; CHECK-NEXT: [[TMP1:%.*]] = bitcast i32* [[TMP0]] to <16 x i32>* 292; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i32>, <16 x i32>* [[TMP1]], align 8, !alias.scope !18, !noalias !21 293; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <16 x i32> [[WIDE_LOAD]], [[BROADCAST_SPLAT]] 294; CHECK-NEXT: [[TMP3:%.*]] = bitcast i32* [[TMP0]] to <16 x i32>* 295; CHECK-NEXT: store <16 x i32> [[BROADCAST_SPLAT19]], <16 x i32>* [[TMP3]], align 4, !alias.scope !18, !noalias !21 296; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i32, i32* [[C]], i64 [[INDEX]] 297; CHECK-NEXT: [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <16 x i32>* 298; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <16 x i32> @llvm.masked.load.v16i32.p0v16i32(<16 x i32>* [[TMP5]], i32 8, <16 x i1> [[TMP2]], <16 x i32> poison), !alias.scope !24 299; CHECK-NEXT: call void @llvm.masked.scatter.v16i32.v16p0i32(<16 x i32> [[WIDE_MASKED_LOAD]], <16 x i32*> [[BROADCAST_SPLAT21]], i32 4, <16 x i1> [[TMP2]]), !alias.scope !25, !noalias !24 300; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 301; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 302; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]] 303; CHECK: middle.block: 304; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX16]], [[N_VEC]] 305; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 306; CHECK: vec.epilog.iter.check: 307; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = and i64 [[SMAX16]], 8 308; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK_NOT_NOT:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0 309; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK_NOT_NOT]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 310; CHECK: vec.epilog.ph: 311; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 312; CHECK-NEXT: [[SMAX22:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 1) 313; CHECK-NEXT: [[N_VEC24:%.*]] = and i64 [[SMAX22]], 9223372036854775800 314; CHECK-NEXT: [[BROADCAST_SPLATINSERT28:%.*]] = insertelement <8 x i32> poison, i32 [[K]], i64 0 315; CHECK-NEXT: [[BROADCAST_SPLAT29:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT28]], <8 x i32> poison, <8 x i32> zeroinitializer 316; CHECK-NEXT: [[BROADCAST_SPLATINSERT30:%.*]] = insertelement <8 x i32> poison, i32 [[NTRUNC]], i64 0 317; CHECK-NEXT: [[BROADCAST_SPLAT31:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT30]], <8 x i32> poison, <8 x i32> zeroinitializer 318; CHECK-NEXT: [[BROADCAST_SPLATINSERT33:%.*]] = insertelement <8 x i32*> poison, i32* [[A]], i64 0 319; CHECK-NEXT: [[BROADCAST_SPLAT34:%.*]] = shufflevector <8 x i32*> [[BROADCAST_SPLATINSERT33]], <8 x i32*> poison, <8 x i32> zeroinitializer 320; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 321; CHECK: vec.epilog.vector.body: 322; CHECK-NEXT: [[INDEX26:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT35:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 323; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDEX26]] 324; CHECK-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to <8 x i32>* 325; CHECK-NEXT: [[WIDE_LOAD27:%.*]] = load <8 x i32>, <8 x i32>* [[TMP8]], align 8 326; CHECK-NEXT: [[TMP9:%.*]] = icmp eq <8 x i32> [[WIDE_LOAD27]], [[BROADCAST_SPLAT29]] 327; CHECK-NEXT: [[TMP10:%.*]] = bitcast i32* [[TMP7]] to <8 x i32>* 328; CHECK-NEXT: store <8 x i32> [[BROADCAST_SPLAT31]], <8 x i32>* [[TMP10]], align 4 329; CHECK-NEXT: [[TMP11:%.*]] = getelementptr i32, i32* [[C]], i64 [[INDEX26]] 330; CHECK-NEXT: [[TMP12:%.*]] = bitcast i32* [[TMP11]] to <8 x i32>* 331; CHECK-NEXT: [[WIDE_MASKED_LOAD32:%.*]] = call <8 x i32> @llvm.masked.load.v8i32.p0v8i32(<8 x i32>* [[TMP12]], i32 8, <8 x i1> [[TMP9]], <8 x i32> poison) 332; CHECK-NEXT: call void @llvm.masked.scatter.v8i32.v8p0i32(<8 x i32> [[WIDE_MASKED_LOAD32]], <8 x i32*> [[BROADCAST_SPLAT34]], i32 4, <8 x i1> [[TMP9]]) 333; CHECK-NEXT: [[INDEX_NEXT35]] = add nuw i64 [[INDEX26]], 8 334; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT35]], [[N_VEC24]] 335; CHECK-NEXT: br i1 [[TMP13]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]] 336; CHECK: vec.epilog.middle.block: 337; CHECK-NEXT: [[CMP_N25:%.*]] = icmp eq i64 [[SMAX22]], [[N_VEC24]] 338; CHECK-NEXT: br i1 [[CMP_N25]], label [[FOR_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_SCALAR_PH]] 339; CHECK: vec.epilog.scalar.ph: 340; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC24]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ITER_CHECK:%.*]] ] 341; CHECK-NEXT: br label [[FOR_BODY:%.*]] 342; CHECK: for.body: 343; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[LATCH:%.*]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ] 344; CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[I]] 345; CHECK-NEXT: [[T2:%.*]] = load i32, i32* [[T1]], align 8 346; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[T2]], [[K]] 347; CHECK-NEXT: store i32 [[NTRUNC]], i32* [[T1]], align 4 348; CHECK-NEXT: br i1 [[CMP]], label [[COND_STORE:%.*]], label [[LATCH]] 349; CHECK: cond_store: 350; CHECK-NEXT: [[T3:%.*]] = getelementptr inbounds i32, i32* [[C]], i64 [[I]] 351; CHECK-NEXT: [[T4:%.*]] = load i32, i32* [[T3]], align 8 352; CHECK-NEXT: store i32 [[T4]], i32* [[A]], align 4 353; CHECK-NEXT: br label [[LATCH]] 354; CHECK: latch: 355; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 356; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] 357; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]], !llvm.loop [[LOOP28:![0-9]+]] 358; CHECK: for.end.loopexit: 359; CHECK-NEXT: br label [[FOR_END]] 360; CHECK: for.end: 361; CHECK-NEXT: ret void 362; 363entry: 364 %ntrunc = trunc i64 %n to i32 365 br label %for.body 366 367for.body: ; preds = %for.body, %entry 368 %i = phi i64 [ %i.next, %latch ], [ 0, %entry ] 369 %t1 = getelementptr inbounds i32, i32* %b, i64 %i 370 %t2 = load i32, i32* %t1, align 8 371 %cmp = icmp eq i32 %t2, %k 372 store i32 %ntrunc, i32* %t1 373 br i1 %cmp, label %cond_store, label %latch 374 375cond_store: 376 %t3 = getelementptr inbounds i32, i32* %c, i64 %i 377 %t4 = load i32, i32* %t3, align 8 378 store i32 %t4, i32* %a 379 br label %latch 380 381latch: 382 %i.next = add nuw nsw i64 %i, 1 383 %cond = icmp slt i64 %i.next, %n 384 br i1 %cond, label %for.body, label %for.end 385 386for.end: ; preds = %for.body 387 ret void 388} 389