1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -loop-simplify -loop-fusion < %s | FileCheck %s 3 4@B = common global [1024 x i32] zeroinitializer, align 16 5 6define void @dep_free(i32* noalias %arg) { 7; CHECK-LABEL: @dep_free( 8; CHECK-NEXT: bb: 9; CHECK-NEXT: br label [[BB7:%.*]] 10; CHECK: bb7: 11; CHECK-NEXT: [[DOT014:%.*]] = phi i32 [ 0, [[BB:%.*]] ], [ [[TMP15:%.*]], [[BB27:%.*]] ] 12; CHECK-NEXT: [[INDVARS_IV23:%.*]] = phi i64 [ 0, [[BB]] ], [ [[INDVARS_IV_NEXT3:%.*]], [[BB27]] ] 13; CHECK-NEXT: [[DOT02:%.*]] = phi i32 [ 0, [[BB]] ], [ [[TMP28:%.*]], [[BB27]] ] 14; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ 0, [[BB]] ], [ [[INDVARS_IV_NEXT:%.*]], [[BB27]] ] 15; CHECK-NEXT: [[TMP:%.*]] = add nsw i32 [[DOT014]], -3 16; CHECK-NEXT: [[TMP8:%.*]] = add nuw nsw i64 [[INDVARS_IV23]], 3 17; CHECK-NEXT: [[TMP9:%.*]] = trunc i64 [[TMP8]] to i32 18; CHECK-NEXT: [[TMP10:%.*]] = mul nsw i32 [[TMP]], [[TMP9]] 19; CHECK-NEXT: [[TMP11:%.*]] = trunc i64 [[INDVARS_IV23]] to i32 20; CHECK-NEXT: [[TMP12:%.*]] = srem i32 [[TMP10]], [[TMP11]] 21; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[INDVARS_IV23]] 22; CHECK-NEXT: store i32 [[TMP12]], i32* [[TMP13]], align 4 23; CHECK-NEXT: br label [[BB14:%.*]] 24; CHECK: bb14: 25; CHECK-NEXT: [[TMP20:%.*]] = add nsw i32 [[DOT02]], -3 26; CHECK-NEXT: [[TMP21:%.*]] = add nuw nsw i64 [[INDVARS_IV1]], 3 27; CHECK-NEXT: [[TMP22:%.*]] = trunc i64 [[TMP21]] to i32 28; CHECK-NEXT: [[TMP23:%.*]] = mul nsw i32 [[TMP20]], [[TMP22]] 29; CHECK-NEXT: [[TMP24:%.*]] = trunc i64 [[INDVARS_IV1]] to i32 30; CHECK-NEXT: [[TMP25:%.*]] = srem i32 [[TMP23]], [[TMP24]] 31; CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 [[INDVARS_IV1]] 32; CHECK-NEXT: store i32 [[TMP25]], i32* [[TMP26]], align 4 33; CHECK-NEXT: br label [[BB27]] 34; CHECK: bb27: 35; CHECK-NEXT: [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV23]], 1 36; CHECK-NEXT: [[TMP15]] = add nuw nsw i32 [[DOT014]], 1 37; CHECK-NEXT: [[EXITCOND4:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT3]], 100 38; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV1]], 1 39; CHECK-NEXT: [[TMP28]] = add nuw nsw i32 [[DOT02]], 1 40; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], 100 41; CHECK-NEXT: br i1 [[EXITCOND]], label [[BB7]], label [[BB18:%.*]] 42; CHECK: bb18: 43; CHECK-NEXT: br label [[BB29:%.*]] 44; CHECK: bb29: 45; CHECK-NEXT: ret void 46; 47bb: 48 br label %bb7 49 50bb7: ; preds = %bb, %bb14 51 %.014 = phi i32 [ 0, %bb ], [ %tmp15, %bb14 ] 52 %indvars.iv23 = phi i64 [ 0, %bb ], [ %indvars.iv.next3, %bb14 ] 53 %tmp = add nsw i32 %.014, -3 54 %tmp8 = add nuw nsw i64 %indvars.iv23, 3 55 %tmp9 = trunc i64 %tmp8 to i32 56 %tmp10 = mul nsw i32 %tmp, %tmp9 57 %tmp11 = trunc i64 %indvars.iv23 to i32 58 %tmp12 = srem i32 %tmp10, %tmp11 59 %tmp13 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv23 60 store i32 %tmp12, i32* %tmp13, align 4 61 br label %bb14 62 63bb14: ; preds = %bb7 64 %indvars.iv.next3 = add nuw nsw i64 %indvars.iv23, 1 65 %tmp15 = add nuw nsw i32 %.014, 1 66 %exitcond4 = icmp ne i64 %indvars.iv.next3, 100 67 br i1 %exitcond4, label %bb7, label %bb17.preheader 68 69bb17.preheader: ; preds = %bb14 70 br label %bb19 71 72bb19: ; preds = %bb17.preheader, %bb27 73 %.02 = phi i32 [ 0, %bb17.preheader ], [ %tmp28, %bb27 ] 74 %indvars.iv1 = phi i64 [ 0, %bb17.preheader ], [ %indvars.iv.next, %bb27 ] 75 %tmp20 = add nsw i32 %.02, -3 76 %tmp21 = add nuw nsw i64 %indvars.iv1, 3 77 %tmp22 = trunc i64 %tmp21 to i32 78 %tmp23 = mul nsw i32 %tmp20, %tmp22 79 %tmp24 = trunc i64 %indvars.iv1 to i32 80 %tmp25 = srem i32 %tmp23, %tmp24 81 %tmp26 = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 %indvars.iv1 82 store i32 %tmp25, i32* %tmp26, align 4 83 br label %bb27 84 85bb27: ; preds = %bb19 86 %indvars.iv.next = add nuw nsw i64 %indvars.iv1, 1 87 %tmp28 = add nuw nsw i32 %.02, 1 88 %exitcond = icmp ne i64 %indvars.iv.next, 100 89 br i1 %exitcond, label %bb19, label %bb18 90 91bb18: ; preds = %bb27 92 br label %bb29 93 94bb29: ; preds = %bb18 95 ret void 96} 97 98define void @dep_free_parametric(i32* noalias %arg, i64 %arg2) { 99; CHECK-LABEL: @dep_free_parametric( 100; CHECK-NEXT: bb: 101; CHECK-NEXT: [[TMP3:%.*]] = icmp slt i64 0, [[ARG2:%.*]] 102; CHECK-NEXT: [[TMP161:%.*]] = icmp slt i64 0, [[ARG2]] 103; CHECK-NEXT: br i1 [[TMP3]], label [[BB5_PREHEADER:%.*]], label [[BB27:%.*]] 104; CHECK: bb5.preheader: 105; CHECK-NEXT: br label [[BB5:%.*]] 106; CHECK: bb5: 107; CHECK-NEXT: [[DOT014:%.*]] = phi i64 [ [[TMP13:%.*]], [[BB25:%.*]] ], [ 0, [[BB5_PREHEADER]] ] 108; CHECK-NEXT: [[DOT02:%.*]] = phi i64 [ [[TMP26:%.*]], [[BB25]] ], [ 0, [[BB5_PREHEADER]] ] 109; CHECK-NEXT: [[TMP6:%.*]] = add nsw i64 [[DOT014]], -3 110; CHECK-NEXT: [[TMP7:%.*]] = add nuw nsw i64 [[DOT014]], 3 111; CHECK-NEXT: [[TMP8:%.*]] = mul nsw i64 [[TMP6]], [[TMP7]] 112; CHECK-NEXT: [[TMP9:%.*]] = srem i64 [[TMP8]], [[DOT014]] 113; CHECK-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32 114; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[DOT014]] 115; CHECK-NEXT: store i32 [[TMP10]], i32* [[TMP11]], align 4 116; CHECK-NEXT: br label [[BB12:%.*]] 117; CHECK: bb12: 118; CHECK-NEXT: [[TMP19:%.*]] = add nsw i64 [[DOT02]], -3 119; CHECK-NEXT: [[TMP20:%.*]] = add nuw nsw i64 [[DOT02]], 3 120; CHECK-NEXT: [[TMP21:%.*]] = mul nsw i64 [[TMP19]], [[TMP20]] 121; CHECK-NEXT: [[TMP22:%.*]] = srem i64 [[TMP21]], [[DOT02]] 122; CHECK-NEXT: [[TMP23:%.*]] = trunc i64 [[TMP22]] to i32 123; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 [[DOT02]] 124; CHECK-NEXT: store i32 [[TMP23]], i32* [[TMP24]], align 4 125; CHECK-NEXT: br label [[BB25]] 126; CHECK: bb25: 127; CHECK-NEXT: [[TMP13]] = add nuw nsw i64 [[DOT014]], 1 128; CHECK-NEXT: [[TMP:%.*]] = icmp slt i64 [[TMP13]], [[ARG2]] 129; CHECK-NEXT: [[TMP26]] = add nuw nsw i64 [[DOT02]], 1 130; CHECK-NEXT: [[TMP16:%.*]] = icmp slt i64 [[TMP26]], [[ARG2]] 131; CHECK-NEXT: br i1 [[TMP16]], label [[BB5]], label [[BB27_LOOPEXIT:%.*]] 132; CHECK: bb27.loopexit: 133; CHECK-NEXT: br label [[BB27]] 134; CHECK: bb27: 135; CHECK-NEXT: ret void 136; 137bb: 138 %tmp3 = icmp slt i64 0, %arg2 139 br i1 %tmp3, label %bb5, label %bb15.preheader 140 141bb5: ; preds = %bb5, %bb12 142 %.014 = phi i64 [ 0, %bb ], [ %tmp13, %bb12 ] 143 %tmp6 = add nsw i64 %.014, -3 144 %tmp7 = add nuw nsw i64 %.014, 3 145 %tmp8 = mul nsw i64 %tmp6, %tmp7 146 %tmp9 = srem i64 %tmp8, %.014 147 %tmp10 = trunc i64 %tmp9 to i32 148 %tmp11 = getelementptr inbounds i32, i32* %arg, i64 %.014 149 store i32 %tmp10, i32* %tmp11, align 4 150 br label %bb12 151 152bb12: ; preds = %bb5 153 %tmp13 = add nuw nsw i64 %.014, 1 154 %tmp = icmp slt i64 %tmp13, %arg2 155 br i1 %tmp, label %bb5, label %bb15.preheader 156 157bb15.preheader: ; preds = %bb12, %bb 158 %tmp161 = icmp slt i64 0, %arg2 159 br i1 %tmp161, label %bb18, label %bb27 160 161bb18: ; preds = %bb15.preheader, %bb25 162 %.02 = phi i64 [ 0, %bb15.preheader ], [ %tmp26, %bb25 ] 163 %tmp19 = add nsw i64 %.02, -3 164 %tmp20 = add nuw nsw i64 %.02, 3 165 %tmp21 = mul nsw i64 %tmp19, %tmp20 166 %tmp22 = srem i64 %tmp21, %.02 167 %tmp23 = trunc i64 %tmp22 to i32 168 %tmp24 = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 %.02 169 store i32 %tmp23, i32* %tmp24, align 4 170 br label %bb25 171 172bb25: ; preds = %bb18 173 %tmp26 = add nuw nsw i64 %.02, 1 174 %tmp16 = icmp slt i64 %tmp26, %arg2 175 br i1 %tmp16, label %bb18, label %bb27 176 177bb27: ; preds = %bb17 178 ret void 179} 180 181define void @raw_only(i32* noalias %arg) { 182; CHECK-LABEL: @raw_only( 183; CHECK-NEXT: bb: 184; CHECK-NEXT: br label [[BB7:%.*]] 185; CHECK: bb7: 186; CHECK-NEXT: [[INDVARS_IV22:%.*]] = phi i64 [ 0, [[BB:%.*]] ], [ [[INDVARS_IV_NEXT3:%.*]], [[BB18:%.*]] ] 187; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ 0, [[BB]] ], [ [[INDVARS_IV_NEXT:%.*]], [[BB18]] ] 188; CHECK-NEXT: [[TMP:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[INDVARS_IV22]] 189; CHECK-NEXT: [[TMP8:%.*]] = trunc i64 [[INDVARS_IV22]] to i32 190; CHECK-NEXT: store i32 [[TMP8]], i32* [[TMP]], align 4 191; CHECK-NEXT: br label [[BB9:%.*]] 192; CHECK: bb9: 193; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[INDVARS_IV1]] 194; CHECK-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP14]], align 4 195; CHECK-NEXT: [[TMP16:%.*]] = shl nsw i32 [[TMP15]], 1 196; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 [[INDVARS_IV1]] 197; CHECK-NEXT: store i32 [[TMP16]], i32* [[TMP17]], align 4 198; CHECK-NEXT: br label [[BB18]] 199; CHECK: bb18: 200; CHECK-NEXT: [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV22]], 1 201; CHECK-NEXT: [[EXITCOND4:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT3]], 100 202; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV1]], 1 203; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], 100 204; CHECK-NEXT: br i1 [[EXITCOND]], label [[BB7]], label [[BB19:%.*]] 205; CHECK: bb19: 206; CHECK-NEXT: ret void 207; 208bb: 209 br label %bb7 210 211bb11.preheader: ; preds = %bb9 212 br label %bb13 213 214bb7: ; preds = %bb, %bb9 215 %indvars.iv22 = phi i64 [ 0, %bb ], [ %indvars.iv.next3, %bb9 ] 216 %tmp = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv22 217 %tmp8 = trunc i64 %indvars.iv22 to i32 218 store i32 %tmp8, i32* %tmp, align 4 219 br label %bb9 220 221bb9: ; preds = %bb7 222 %indvars.iv.next3 = add nuw nsw i64 %indvars.iv22, 1 223 %exitcond4 = icmp ne i64 %indvars.iv.next3, 100 224 br i1 %exitcond4, label %bb7, label %bb11.preheader 225 226bb13: ; preds = %bb11.preheader, %bb18 227 %indvars.iv1 = phi i64 [ 0, %bb11.preheader ], [ %indvars.iv.next, %bb18 ] 228 %tmp14 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv1 229 %tmp15 = load i32, i32* %tmp14, align 4 230 %tmp16 = shl nsw i32 %tmp15, 1 231 %tmp17 = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 %indvars.iv1 232 store i32 %tmp16, i32* %tmp17, align 4 233 br label %bb18 234 235bb18: ; preds = %bb13 236 %indvars.iv.next = add nuw nsw i64 %indvars.iv1, 1 237 %exitcond = icmp ne i64 %indvars.iv.next, 100 br i1 %exitcond, label %bb13, label %bb19 238 239bb19: ; preds = %bb18 240 ret void 241} 242 243define void @raw_only_parametric(i32* noalias %arg, i32 %arg4) { 244; CHECK-LABEL: @raw_only_parametric( 245; CHECK-NEXT: bb: 246; CHECK-NEXT: [[TMP:%.*]] = sext i32 [[ARG4:%.*]] to i64 247; CHECK-NEXT: [[TMP64:%.*]] = icmp sgt i32 [[ARG4]], 0 248; CHECK-NEXT: br i1 [[TMP64]], label [[BB8_PREHEADER:%.*]], label [[BB23:%.*]] 249; CHECK: bb8.preheader: 250; CHECK-NEXT: br label [[BB8:%.*]] 251; CHECK: bb8: 252; CHECK-NEXT: [[INDVARS_IV25:%.*]] = phi i64 [ [[INDVARS_IV_NEXT3:%.*]], [[BB8]] ], [ 0, [[BB8_PREHEADER]] ] 253; CHECK-NEXT: [[INDVARS_IV3:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[BB8]] ], [ 0, [[BB8_PREHEADER]] ] 254; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[INDVARS_IV25]] 255; CHECK-NEXT: [[TMP10:%.*]] = trunc i64 [[INDVARS_IV25]] to i32 256; CHECK-NEXT: store i32 [[TMP10]], i32* [[TMP9]], align 4 257; CHECK-NEXT: [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV25]], 1 258; CHECK-NEXT: [[TMP6:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT3]], [[TMP]] 259; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[INDVARS_IV3]] 260; CHECK-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4 261; CHECK-NEXT: [[TMP20:%.*]] = shl nsw i32 [[TMP19]], 1 262; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 [[INDVARS_IV3]] 263; CHECK-NEXT: store i32 [[TMP20]], i32* [[TMP21]], align 4 264; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV3]], 1 265; CHECK-NEXT: [[TMP15:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT]], [[TMP]] 266; CHECK-NEXT: br i1 [[TMP15]], label [[BB8]], label [[BB23_LOOPEXIT:%.*]] 267; CHECK: bb23.loopexit: 268; CHECK-NEXT: br label [[BB23]] 269; CHECK: bb23: 270; CHECK-NEXT: ret void 271; 272bb: 273 %tmp = sext i32 %arg4 to i64 274 %tmp64 = icmp sgt i32 %arg4, 0 275 br i1 %tmp64, label %bb8, label %bb23 276 277bb8: ; preds = %bb, %bb8 278 %indvars.iv25 = phi i64 [ %indvars.iv.next3, %bb8 ], [ 0, %bb ] 279 %tmp9 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv25 280 %tmp10 = trunc i64 %indvars.iv25 to i32 281 store i32 %tmp10, i32* %tmp9, align 4 282 %indvars.iv.next3 = add nuw nsw i64 %indvars.iv25, 1 283 %tmp6 = icmp slt i64 %indvars.iv.next3, %tmp 284 br i1 %tmp6, label %bb8, label %bb17 285 286bb17: ; preds = %bb8, %bb17 287 %indvars.iv3 = phi i64 [ %indvars.iv.next, %bb17 ], [ 0, %bb8 ] 288 %tmp18 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv3 289 %tmp19 = load i32, i32* %tmp18, align 4 290 %tmp20 = shl nsw i32 %tmp19, 1 291 %tmp21 = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 %indvars.iv3 292 store i32 %tmp20, i32* %tmp21, align 4 293 %indvars.iv.next = add nuw nsw i64 %indvars.iv3, 1 294 %tmp15 = icmp slt i64 %indvars.iv.next, %tmp 295 br i1 %tmp15, label %bb17, label %bb23 296 297bb23: ; preds = %bb17, %bb 298 ret void 299} 300 301define void @forward_dep(i32* noalias %arg) { 302; CHECK-LABEL: @forward_dep( 303; CHECK-NEXT: bb: 304; CHECK-NEXT: br label [[BB7:%.*]] 305; CHECK: bb7: 306; CHECK-NEXT: [[DOT013:%.*]] = phi i32 [ 0, [[BB:%.*]] ], [ [[TMP15:%.*]], [[BB14:%.*]] ] 307; CHECK-NEXT: [[INDVARS_IV22:%.*]] = phi i64 [ 0, [[BB]] ], [ [[INDVARS_IV_NEXT3:%.*]], [[BB14]] ] 308; CHECK-NEXT: [[TMP:%.*]] = add nsw i32 [[DOT013]], -3 309; CHECK-NEXT: [[TMP8:%.*]] = add nuw nsw i64 [[INDVARS_IV22]], 3 310; CHECK-NEXT: [[TMP9:%.*]] = trunc i64 [[TMP8]] to i32 311; CHECK-NEXT: [[TMP10:%.*]] = mul nsw i32 [[TMP]], [[TMP9]] 312; CHECK-NEXT: [[TMP11:%.*]] = trunc i64 [[INDVARS_IV22]] to i32 313; CHECK-NEXT: [[TMP12:%.*]] = srem i32 [[TMP10]], [[TMP11]] 314; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[INDVARS_IV22]] 315; CHECK-NEXT: store i32 [[TMP12]], i32* [[TMP13]], align 4 316; CHECK-NEXT: br label [[BB14]] 317; CHECK: bb14: 318; CHECK-NEXT: [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV22]], 1 319; CHECK-NEXT: [[TMP15]] = add nuw nsw i32 [[DOT013]], 1 320; CHECK-NEXT: [[EXITCOND4:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT3]], 100 321; CHECK-NEXT: br i1 [[EXITCOND4]], label [[BB7]], label [[BB19_PREHEADER:%.*]] 322; CHECK: bb19.preheader: 323; CHECK-NEXT: br label [[BB19:%.*]] 324; CHECK: bb19: 325; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[BB25:%.*]] ], [ 0, [[BB19_PREHEADER]] ] 326; CHECK-NEXT: [[TMP20:%.*]] = add nsw i64 [[INDVARS_IV1]], -3 327; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[TMP20]] 328; CHECK-NEXT: [[TMP22:%.*]] = load i32, i32* [[TMP21]], align 4 329; CHECK-NEXT: [[TMP23:%.*]] = mul nsw i32 [[TMP22]], 3 330; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[INDVARS_IV1]] 331; CHECK-NEXT: store i32 [[TMP23]], i32* [[TMP24]], align 4 332; CHECK-NEXT: br label [[BB25]] 333; CHECK: bb25: 334; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV1]], 1 335; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], 100 336; CHECK-NEXT: br i1 [[EXITCOND]], label [[BB19]], label [[BB26:%.*]] 337; CHECK: bb26: 338; CHECK-NEXT: ret void 339; 340bb: 341 br label %bb7 342 343bb7: ; preds = %bb, %bb14 344 %.013 = phi i32 [ 0, %bb ], [ %tmp15, %bb14 ] 345 %indvars.iv22 = phi i64 [ 0, %bb ], [ %indvars.iv.next3, %bb14 ] 346 %tmp = add nsw i32 %.013, -3 347 %tmp8 = add nuw nsw i64 %indvars.iv22, 3 348 %tmp9 = trunc i64 %tmp8 to i32 349 %tmp10 = mul nsw i32 %tmp, %tmp9 350 %tmp11 = trunc i64 %indvars.iv22 to i32 351 %tmp12 = srem i32 %tmp10, %tmp11 352 %tmp13 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv22 353 store i32 %tmp12, i32* %tmp13, align 4 354 br label %bb14 355 356bb14: ; preds = %bb7 357 %indvars.iv.next3 = add nuw nsw i64 %indvars.iv22, 1 358 %tmp15 = add nuw nsw i32 %.013, 1 359 %exitcond4 = icmp ne i64 %indvars.iv.next3, 100 360 br i1 %exitcond4, label %bb7, label %bb19 361 362bb19: ; preds = %bb14, %bb25 363 %indvars.iv1 = phi i64 [ 0, %bb14 ], [ %indvars.iv.next, %bb25 ] 364 %tmp20 = add nsw i64 %indvars.iv1, -3 365 %tmp21 = getelementptr inbounds i32, i32* %arg, i64 %tmp20 366 %tmp22 = load i32, i32* %tmp21, align 4 367 %tmp23 = mul nsw i32 %tmp22, 3 368 %tmp24 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv1 369 store i32 %tmp23, i32* %tmp24, align 4 370 br label %bb25 371 372bb25: ; preds = %bb19 373 %indvars.iv.next = add nuw nsw i64 %indvars.iv1, 1 374 %exitcond = icmp ne i64 %indvars.iv.next, 100 375 br i1 %exitcond, label %bb19, label %bb26 376 377bb26: ; preds = %bb25 378 ret void 379} 380 381; Test that instructions in loop 1 latch are moved to the beginning of loop 2 382; latch iff it is proven safe. %inc.first and %cmp.first are moved, but 383; `store i32 0, i32* %Ai.first` is not. 384 385define void @flow_dep(i32* noalias %A, i32* noalias %B) { 386; CHECK-LABEL: @flow_dep( 387; CHECK-NEXT: entry: 388; CHECK-NEXT: br label [[FOR_FIRST:%.*]] 389; CHECK: for.first: 390; CHECK-NEXT: [[I_FIRST:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INC_FIRST:%.*]], [[FOR_SECOND_LATCH:%.*]] ] 391; CHECK-NEXT: [[I_SECOND:%.*]] = phi i64 [ [[INC_SECOND:%.*]], [[FOR_SECOND_LATCH]] ], [ 0, [[ENTRY]] ] 392; CHECK-NEXT: [[AI_FIRST:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[I_FIRST]] 393; CHECK-NEXT: store i32 0, i32* [[AI_FIRST]], align 4 394; CHECK-NEXT: [[AI_SECOND:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I_SECOND]] 395; CHECK-NEXT: [[TMP0:%.*]] = load i32, i32* [[AI_SECOND]], align 4 396; CHECK-NEXT: [[BI:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[I_SECOND]] 397; CHECK-NEXT: store i32 [[TMP0]], i32* [[BI]], align 4 398; CHECK-NEXT: br label [[FOR_SECOND_LATCH]] 399; CHECK: for.second.latch: 400; CHECK-NEXT: [[INC_FIRST]] = add nsw i64 [[I_FIRST]], 1 401; CHECK-NEXT: [[CMP_FIRST:%.*]] = icmp slt i64 [[INC_FIRST]], 100 402; CHECK-NEXT: [[INC_SECOND]] = add nsw i64 [[I_SECOND]], 1 403; CHECK-NEXT: [[CMP_SECOND:%.*]] = icmp slt i64 [[INC_SECOND]], 100 404; CHECK-NEXT: br i1 [[CMP_SECOND]], label [[FOR_FIRST]], label [[FOR_END:%.*]] 405; CHECK: for.end: 406; CHECK-NEXT: ret void 407; 408entry: 409 br label %for.first 410 411for.first: 412 %i.first = phi i64 [ 0, %entry ], [ %inc.first, %for.first ] 413 %Ai.first = getelementptr inbounds i32, i32* %A, i64 %i.first 414 store i32 0, i32* %Ai.first, align 4 415 %inc.first = add nsw i64 %i.first, 1 416 %cmp.first = icmp slt i64 %inc.first, 100 417 br i1 %cmp.first, label %for.first, label %for.second.preheader 418 419for.second.preheader: 420 br label %for.second 421 422for.second: 423 %i.second = phi i64 [ %inc.second, %for.second.latch ], [ 0, %for.second.preheader ] 424 %Ai.second = getelementptr inbounds i32, i32* %A, i64 %i.second 425 %0 = load i32, i32* %Ai.second, align 4 426 %Bi = getelementptr inbounds i32, i32* %B, i64 %i.second 427 store i32 %0, i32* %Bi, align 4 428 br label %for.second.latch 429 430for.second.latch: 431 %inc.second = add nsw i64 %i.second, 1 432 %cmp.second = icmp slt i64 %inc.second, 100 433 br i1 %cmp.second, label %for.second, label %for.end 434 435for.end: 436 ret void 437} 438 439; Test that `%add` is moved in basic block entry, and the two loops for.first 440; and for.second are fused. 441 442define i32 @moveinsts_preheader(i32* %A, i32 %x) { 443; CHECK-LABEL: @moveinsts_preheader( 444; CHECK-NEXT: entry: 445; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[X:%.*]], 1 446; CHECK-NEXT: br label [[FOR_FIRST:%.*]] 447; CHECK: for.first: 448; CHECK-NEXT: [[I:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INC_I:%.*]], [[FOR_FIRST]] ] 449; CHECK-NEXT: [[J:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[INC_J:%.*]], [[FOR_FIRST]] ] 450; CHECK-NEXT: [[AI:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[I]] 451; CHECK-NEXT: store i32 0, i32* [[AI]], align 4 452; CHECK-NEXT: [[INC_I]] = add nsw i64 [[I]], 1 453; CHECK-NEXT: [[CMP_I:%.*]] = icmp slt i64 [[INC_I]], 100 454; CHECK-NEXT: [[AJ:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[J]] 455; CHECK-NEXT: store i32 2, i32* [[AJ]], align 4 456; CHECK-NEXT: [[INC_J]] = add nsw i64 [[J]], 1 457; CHECK-NEXT: [[CMP_J:%.*]] = icmp slt i64 [[INC_J]], 100 458; CHECK-NEXT: br i1 [[CMP_J]], label [[FOR_FIRST]], label [[FOR_SECOND_EXIT:%.*]] 459; CHECK: for.second.exit: 460; CHECK-NEXT: ret i32 [[ADD]] 461; 462entry: 463 br label %for.first 464 465for.first: 466 %i = phi i64 [ 0, %entry ], [ %inc.i, %for.first ] 467 %Ai = getelementptr inbounds i32, i32* %A, i64 %i 468 store i32 0, i32* %Ai, align 4 469 %inc.i = add nsw i64 %i, 1 470 %cmp.i = icmp slt i64 %inc.i, 100 471 br i1 %cmp.i, label %for.first, label %for.first.exit 472 473for.first.exit: 474 %add = add nsw i32 %x, 1 475 br label %for.second 476 477for.second: 478 %j = phi i64 [ 0, %for.first.exit ], [ %inc.j, %for.second ] 479 %Aj = getelementptr inbounds i32, i32* %A, i64 %j 480 store i32 2, i32* %Aj, align 4 481 %inc.j = add nsw i64 %j, 1 482 %cmp.j = icmp slt i64 %inc.j, 100 483 br i1 %cmp.j, label %for.second, label %for.second.exit 484 485for.second.exit: 486 ret i32 %add 487} 488 489; Test that `%add` cannot be moved to basic block entry, as it uses %i, which 490; defined after basic block entry. And the two loops for.first and for.second 491; are not fused. 492 493define i64 @unsafe_preheader(i32* %A, i64 %x) { 494; CHECK-LABEL: @unsafe_preheader( 495; CHECK-NEXT: entry: 496; CHECK-NEXT: br label [[FOR_FIRST:%.*]] 497; CHECK: for.first: 498; CHECK-NEXT: [[I:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INC_I:%.*]], [[FOR_FIRST]] ] 499; CHECK-NEXT: [[AI:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[I]] 500; CHECK-NEXT: store i32 0, i32* [[AI]], align 4 501; CHECK-NEXT: [[INC_I]] = add nsw i64 [[I]], 1 502; CHECK-NEXT: [[CMP_I:%.*]] = icmp slt i64 [[INC_I]], 100 503; CHECK-NEXT: br i1 [[CMP_I]], label [[FOR_FIRST]], label [[FOR_FIRST_EXIT:%.*]] 504; CHECK: for.first.exit: 505; CHECK-NEXT: [[ADD:%.*]] = add nsw i64 [[X:%.*]], [[I]] 506; CHECK-NEXT: br label [[FOR_SECOND:%.*]] 507; CHECK: for.second: 508; CHECK-NEXT: [[J:%.*]] = phi i64 [ 0, [[FOR_FIRST_EXIT]] ], [ [[INC_J:%.*]], [[FOR_SECOND]] ] 509; CHECK-NEXT: [[AJ:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[J]] 510; CHECK-NEXT: store i32 2, i32* [[AJ]], align 4 511; CHECK-NEXT: [[INC_J]] = add nsw i64 [[J]], 1 512; CHECK-NEXT: [[CMP_J:%.*]] = icmp slt i64 [[INC_J]], 100 513; CHECK-NEXT: br i1 [[CMP_J]], label [[FOR_SECOND]], label [[FOR_SECOND_EXIT:%.*]] 514; CHECK: for.second.exit: 515; CHECK-NEXT: ret i64 [[ADD]] 516; 517entry: 518 br label %for.first 519 520for.first: 521 %i = phi i64 [ 0, %entry ], [ %inc.i, %for.first ] 522 %Ai = getelementptr inbounds i32, i32* %A, i64 %i 523 store i32 0, i32* %Ai, align 4 524 %inc.i = add nsw i64 %i, 1 525 %cmp.i = icmp slt i64 %inc.i, 100 526 br i1 %cmp.i, label %for.first, label %for.first.exit 527 528for.first.exit: 529 %add = add nsw i64 %x, %i 530 br label %for.second 531 532for.second: 533 %j = phi i64 [ 0, %for.first.exit ], [ %inc.j, %for.second ] 534 %Aj = getelementptr inbounds i32, i32* %A, i64 %j 535 store i32 2, i32* %Aj, align 4 536 %inc.j = add nsw i64 %j, 1 537 %cmp.j = icmp slt i64 %inc.j, 100 538 br i1 %cmp.j, label %for.second, label %for.second.exit 539 540for.second.exit: 541 ret i64 %add 542} 543