1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -scoped-noalias-aa -slp-vectorizer -mtriple=arm64-apple-darwin -enable-new-pm=false -S %s | FileCheck %s 3 4define void @needs_versioning_not_profitable(i32* %dst, i32* %src) { 5; CHECK-LABEL: @needs_versioning_not_profitable( 6; CHECK-NEXT: entry: 7; CHECK-NEXT: [[SRC_0:%.*]] = load i32, i32* [[SRC:%.*]], align 4 8; CHECK-NEXT: [[R_0:%.*]] = ashr i32 [[SRC_0]], 16 9; CHECK-NEXT: store i32 [[R_0]], i32* [[DST:%.*]], align 4 10; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 1 11; CHECK-NEXT: [[SRC_1:%.*]] = load i32, i32* [[SRC_GEP_1]], align 4 12; CHECK-NEXT: [[R_1:%.*]] = ashr i32 [[SRC_1]], 16 13; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 1 14; CHECK-NEXT: store i32 [[R_1]], i32* [[DST_GEP_1]], align 4 15; CHECK-NEXT: ret void 16; 17entry: 18 %src.0 = load i32, i32* %src, align 4 19 %r.0 = ashr i32 %src.0, 16 20 store i32 %r.0, i32* %dst, align 4 21 %src.gep.1 = getelementptr inbounds i32, i32* %src, i64 1 22 %src.1 = load i32, i32* %src.gep.1, align 4 23 %r.1 = ashr i32 %src.1, 16 24 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 25 store i32 %r.1, i32* %dst.gep.1, align 4 26 ret void 27} 28 29define void @needs_versioning_profitable(i32* %dst, i32* %src) { 30; CHECK-LABEL: @needs_versioning_profitable( 31; CHECK-NEXT: entry: 32; CHECK-NEXT: [[SRC_0:%.*]] = load i32, i32* [[SRC:%.*]], align 4 33; CHECK-NEXT: [[R_0:%.*]] = ashr i32 [[SRC_0]], 16 34; CHECK-NEXT: store i32 [[R_0]], i32* [[DST:%.*]], align 4 35; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 1 36; CHECK-NEXT: [[SRC_1:%.*]] = load i32, i32* [[SRC_GEP_1]], align 4 37; CHECK-NEXT: [[R_1:%.*]] = ashr i32 [[SRC_1]], 16 38; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 1 39; CHECK-NEXT: store i32 [[R_1]], i32* [[DST_GEP_1]], align 4 40; CHECK-NEXT: [[SRC_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 2 41; CHECK-NEXT: [[SRC_2:%.*]] = load i32, i32* [[SRC_GEP_2]], align 4 42; CHECK-NEXT: [[R_2:%.*]] = ashr i32 [[SRC_2]], 16 43; CHECK-NEXT: [[DST_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 2 44; CHECK-NEXT: store i32 [[R_2]], i32* [[DST_GEP_2]], align 4 45; CHECK-NEXT: [[SRC_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 3 46; CHECK-NEXT: [[SRC_3:%.*]] = load i32, i32* [[SRC_GEP_3]], align 4 47; CHECK-NEXT: [[R_3:%.*]] = ashr i32 [[SRC_3]], 16 48; CHECK-NEXT: [[DST_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 3 49; CHECK-NEXT: store i32 [[R_3]], i32* [[DST_GEP_3]], align 4 50; CHECK-NEXT: ret void 51; 52entry: 53 %src.0 = load i32, i32* %src, align 4 54 %r.0 = ashr i32 %src.0, 16 55 store i32 %r.0, i32* %dst, align 4 56 %src.gep.1 = getelementptr inbounds i32, i32* %src, i64 1 57 %src.1 = load i32, i32* %src.gep.1, align 4 58 %r.1 = ashr i32 %src.1, 16 59 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 60 store i32 %r.1, i32* %dst.gep.1, align 4 61 %src.gep.2 = getelementptr inbounds i32, i32* %src, i64 2 62 %src.2 = load i32, i32* %src.gep.2, align 4 63 %r.2 = ashr i32 %src.2, 16 64 %dst.gep.2 = getelementptr inbounds i32, i32* %dst, i64 2 65 store i32 %r.2, i32* %dst.gep.2, align 4 66 %src.gep.3 = getelementptr inbounds i32, i32* %src, i64 3 67 %src.3 = load i32, i32* %src.gep.3, align 4 68 %r.3 = ashr i32 %src.3, 16 69 %dst.gep.3 = getelementptr inbounds i32, i32* %dst, i64 3 70 store i32 %r.3, i32* %dst.gep.3, align 4 71 72 ret void 73} 74 75define void @needs_versioning_profitable_2_sources(i32* %dst, i32* %A, i32* %B) { 76; CHECK-LABEL: @needs_versioning_profitable_2_sources( 77; CHECK-NEXT: entry: 78; CHECK-NEXT: [[A_0:%.*]] = load i32, i32* [[A:%.*]], align 4 79; CHECK-NEXT: [[B_0:%.*]] = load i32, i32* [[B:%.*]], align 4 80; CHECK-NEXT: [[R_0:%.*]] = add i32 [[A_0]], [[B_0]] 81; CHECK-NEXT: [[MUL_0:%.*]] = mul i32 [[R_0]], 2 82; CHECK-NEXT: store i32 [[MUL_0]], i32* [[DST:%.*]], align 4 83; CHECK-NEXT: [[A_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 1 84; CHECK-NEXT: [[A_1:%.*]] = load i32, i32* [[A_GEP_1]], align 4 85; CHECK-NEXT: [[B_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 1 86; CHECK-NEXT: [[B_1:%.*]] = load i32, i32* [[B_GEP_1]], align 4 87; CHECK-NEXT: [[R_1:%.*]] = add i32 [[A_1]], [[B_1]] 88; CHECK-NEXT: [[MUL_1:%.*]] = mul i32 [[R_1]], 2 89; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 1 90; CHECK-NEXT: store i32 [[MUL_1]], i32* [[DST_GEP_1]], align 4 91; CHECK-NEXT: [[A_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 2 92; CHECK-NEXT: [[A_2:%.*]] = load i32, i32* [[A_GEP_2]], align 4 93; CHECK-NEXT: [[B_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 2 94; CHECK-NEXT: [[B_2:%.*]] = load i32, i32* [[B_GEP_2]], align 4 95; CHECK-NEXT: [[R_2:%.*]] = add i32 [[A_2]], [[B_2]] 96; CHECK-NEXT: [[MUL_2:%.*]] = mul i32 [[R_2]], 2 97; CHECK-NEXT: [[DST_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 2 98; CHECK-NEXT: store i32 [[MUL_2]], i32* [[DST_GEP_2]], align 4 99; CHECK-NEXT: [[A_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 3 100; CHECK-NEXT: [[A_3:%.*]] = load i32, i32* [[A_GEP_3]], align 4 101; CHECK-NEXT: [[B_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 3 102; CHECK-NEXT: [[B_3:%.*]] = load i32, i32* [[B_GEP_3]], align 4 103; CHECK-NEXT: [[R_3:%.*]] = add i32 [[A_3]], [[B_3]] 104; CHECK-NEXT: [[MUL_3:%.*]] = mul i32 [[R_3]], 2 105; CHECK-NEXT: [[DST_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 3 106; CHECK-NEXT: store i32 [[MUL_3]], i32* [[DST_GEP_3]], align 4 107; CHECK-NEXT: ret void 108; 109entry: 110 %A.0 = load i32, i32* %A, align 4 111 %B.0 = load i32, i32* %B, align 4 112 %r.0 = add i32 %A.0, %B.0 113 %mul.0 = mul i32 %r.0, 2 114 store i32 %mul.0, i32* %dst, align 4 115 %A.gep.1 = getelementptr inbounds i32, i32* %A, i64 1 116 %A.1 = load i32, i32* %A.gep.1, align 4 117 %B.gep.1 = getelementptr inbounds i32, i32* %B, i64 1 118 %B.1 = load i32, i32* %B.gep.1, align 4 119 %r.1 = add i32 %A.1, %B.1 120 %mul.1 = mul i32 %r.1, 2 121 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 122 store i32 %mul.1, i32* %dst.gep.1, align 4 123 %A.gep.2 = getelementptr inbounds i32, i32* %A, i64 2 124 %A.2 = load i32, i32* %A.gep.2, align 4 125 %B.gep.2 = getelementptr inbounds i32, i32* %B, i64 2 126 %B.2 = load i32, i32* %B.gep.2, align 4 127 %r.2 = add i32 %A.2, %B.2 128 %mul.2 = mul i32 %r.2, 2 129 %dst.gep.2 = getelementptr inbounds i32, i32* %dst, i64 2 130 store i32 %mul.2, i32* %dst.gep.2, align 4 131 %A.gep.3 = getelementptr inbounds i32, i32* %A, i64 3 132 %A.3 = load i32, i32* %A.gep.3, align 4 133 %B.gep.3 = getelementptr inbounds i32, i32* %B, i64 3 134 %B.3 = load i32, i32* %B.gep.3, align 4 135 %r.3 = add i32 %A.3, %B.3 136 %mul.3 = mul i32 %r.3, 2 137 %dst.gep.3 = getelementptr inbounds i32, i32* %dst, i64 3 138 store i32 %mul.3, i32* %dst.gep.3, align 4 139 140 ret void 141} 142 143declare void @use(i32) 144 145declare void @bar() 146 147define void @needs_versioning_profitable_split_points(i32* %dst, i32* %src) { 148; CHECK-LABEL: @needs_versioning_profitable_split_points( 149; CHECK-NEXT: entry: 150; CHECK-NEXT: call void @bar() 151; CHECK-NEXT: call void @bar() 152; CHECK-NEXT: call void @bar() 153; CHECK-NEXT: [[SRC_0:%.*]] = load i32, i32* [[SRC:%.*]], align 4 154; CHECK-NEXT: [[R_0:%.*]] = ashr i32 [[SRC_0]], 16 155; CHECK-NEXT: store i32 [[R_0]], i32* [[DST:%.*]], align 4 156; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 1 157; CHECK-NEXT: [[SRC_1:%.*]] = load i32, i32* [[SRC_GEP_1]], align 4 158; CHECK-NEXT: [[R_1:%.*]] = ashr i32 [[SRC_1]], 16 159; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 1 160; CHECK-NEXT: store i32 [[R_1]], i32* [[DST_GEP_1]], align 4 161; CHECK-NEXT: [[SRC_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 2 162; CHECK-NEXT: [[SRC_2:%.*]] = load i32, i32* [[SRC_GEP_2]], align 4 163; CHECK-NEXT: [[R_2:%.*]] = ashr i32 [[SRC_2]], 16 164; CHECK-NEXT: [[DST_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 2 165; CHECK-NEXT: store i32 [[R_2]], i32* [[DST_GEP_2]], align 4 166; CHECK-NEXT: [[SRC_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 3 167; CHECK-NEXT: [[SRC_3:%.*]] = load i32, i32* [[SRC_GEP_3]], align 4 168; CHECK-NEXT: [[R_3:%.*]] = ashr i32 [[SRC_3]], 16 169; CHECK-NEXT: [[DST_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 3 170; CHECK-NEXT: store i32 [[R_3]], i32* [[DST_GEP_3]], align 4 171; CHECK-NEXT: call void @bar() 172; CHECK-NEXT: ret void 173; 174entry: 175 call void @bar() 176 call void @bar() 177 call void @bar() 178 179 %src.0 = load i32, i32* %src, align 4 180 %r.0 = ashr i32 %src.0, 16 181 store i32 %r.0, i32* %dst, align 4 182 %src.gep.1 = getelementptr inbounds i32, i32* %src, i64 1 183 %src.1 = load i32, i32* %src.gep.1, align 4 184 %r.1 = ashr i32 %src.1, 16 185 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 186 store i32 %r.1, i32* %dst.gep.1, align 4 187 %src.gep.2 = getelementptr inbounds i32, i32* %src, i64 2 188 %src.2 = load i32, i32* %src.gep.2, align 4 189 %r.2 = ashr i32 %src.2, 16 190 %dst.gep.2 = getelementptr inbounds i32, i32* %dst, i64 2 191 store i32 %r.2, i32* %dst.gep.2, align 4 192 %src.gep.3 = getelementptr inbounds i32, i32* %src, i64 3 193 %src.3 = load i32, i32* %src.gep.3, align 4 194 %r.3 = ashr i32 %src.3, 16 195 %dst.gep.3 = getelementptr inbounds i32, i32* %dst, i64 3 196 store i32 %r.3, i32* %dst.gep.3, align 4 197 198 call void @bar() 199 ret void 200} 201 202define void @needs_versioning_profitable_load_used_outside_region1(i32* %dst, i32* %src, i1 %c) { 203; CHECK-LABEL: @needs_versioning_profitable_load_used_outside_region1( 204; CHECK-NEXT: entry: 205; CHECK-NEXT: br i1 [[C:%.*]], label [[THEN:%.*]], label [[EXIT:%.*]] 206; CHECK: then: 207; CHECK-NEXT: [[SRC_0:%.*]] = load i32, i32* [[SRC:%.*]], align 4 208; CHECK-NEXT: [[R_0:%.*]] = ashr i32 [[SRC_0]], 16 209; CHECK-NEXT: store i32 [[R_0]], i32* [[DST:%.*]], align 4 210; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 1 211; CHECK-NEXT: [[SRC_1:%.*]] = load i32, i32* [[SRC_GEP_1]], align 4 212; CHECK-NEXT: [[R_1:%.*]] = ashr i32 [[SRC_1]], 16 213; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 1 214; CHECK-NEXT: store i32 [[R_1]], i32* [[DST_GEP_1]], align 4 215; CHECK-NEXT: [[SRC_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 2 216; CHECK-NEXT: [[SRC_2:%.*]] = load i32, i32* [[SRC_GEP_2]], align 4 217; CHECK-NEXT: [[R_2:%.*]] = ashr i32 [[SRC_2]], 16 218; CHECK-NEXT: [[DST_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 2 219; CHECK-NEXT: store i32 [[R_2]], i32* [[DST_GEP_2]], align 4 220; CHECK-NEXT: [[SRC_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 3 221; CHECK-NEXT: [[SRC_3:%.*]] = load i32, i32* [[SRC_GEP_3]], align 4 222; CHECK-NEXT: [[R_3:%.*]] = ashr i32 [[SRC_3]], 16 223; CHECK-NEXT: [[DST_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 3 224; CHECK-NEXT: store i32 [[R_3]], i32* [[DST_GEP_3]], align 4 225; CHECK-NEXT: [[SRC_GEP_5:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 5 226; CHECK-NEXT: [[L:%.*]] = load i32, i32* [[SRC_GEP_5]], align 4 227; CHECK-NEXT: call void @use(i32 [[L]]) 228; CHECK-NEXT: br label [[EXIT]] 229; CHECK: exit: 230; CHECK-NEXT: ret void 231; 232entry: 233 br i1 %c, label %then, label %exit 234 235then: 236 %src.0 = load i32, i32* %src, align 4 237 %r.0 = ashr i32 %src.0, 16 238 store i32 %r.0, i32* %dst, align 4 239 %src.gep.1 = getelementptr inbounds i32, i32* %src, i64 1 240 %src.1 = load i32, i32* %src.gep.1, align 4 241 %r.1 = ashr i32 %src.1, 16 242 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 243 store i32 %r.1, i32* %dst.gep.1, align 4 244 %src.gep.2 = getelementptr inbounds i32, i32* %src, i64 2 245 %src.2 = load i32, i32* %src.gep.2, align 4 246 %r.2 = ashr i32 %src.2, 16 247 %dst.gep.2 = getelementptr inbounds i32, i32* %dst, i64 2 248 store i32 %r.2, i32* %dst.gep.2, align 4 249 %src.gep.3 = getelementptr inbounds i32, i32* %src, i64 3 250 %src.3 = load i32, i32* %src.gep.3, align 4 251 %r.3 = ashr i32 %src.3, 16 252 %dst.gep.3 = getelementptr inbounds i32, i32* %dst, i64 3 253 store i32 %r.3, i32* %dst.gep.3, align 4 254 %src.gep.5 = getelementptr inbounds i32, i32* %src, i64 5 255 %l = load i32, i32* %src.gep.5 256 call void @use(i32 %l) 257 br label %exit 258 259exit: 260 ret void 261} 262 263define void @needs_versioning_profitable_load_used_outside_region2(i32* %dst, i32* %src, i1 %c) { 264; CHECK-LABEL: @needs_versioning_profitable_load_used_outside_region2( 265; CHECK-NEXT: entry: 266; CHECK-NEXT: br i1 [[C:%.*]], label [[THEN:%.*]], label [[EXIT:%.*]] 267; CHECK: then: 268; CHECK-NEXT: [[SRC_0:%.*]] = load i32, i32* [[SRC:%.*]], align 4 269; CHECK-NEXT: [[R_0:%.*]] = ashr i32 [[SRC_0]], 16 270; CHECK-NEXT: store i32 [[R_0]], i32* [[DST:%.*]], align 4 271; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 1 272; CHECK-NEXT: [[SRC_1:%.*]] = load i32, i32* [[SRC_GEP_1]], align 4 273; CHECK-NEXT: [[R_1:%.*]] = ashr i32 [[SRC_1]], 16 274; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 1 275; CHECK-NEXT: store i32 [[R_1]], i32* [[DST_GEP_1]], align 4 276; CHECK-NEXT: [[SRC_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 2 277; CHECK-NEXT: [[SRC_2:%.*]] = load i32, i32* [[SRC_GEP_2]], align 4 278; CHECK-NEXT: [[SRC_GEP_5:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 5 279; CHECK-NEXT: [[L:%.*]] = load i32, i32* [[SRC_GEP_5]], align 4 280; CHECK-NEXT: [[R_2:%.*]] = ashr i32 [[SRC_2]], 16 281; CHECK-NEXT: [[DST_GEP_2:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 2 282; CHECK-NEXT: store i32 [[R_2]], i32* [[DST_GEP_2]], align 4 283; CHECK-NEXT: [[SRC_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 3 284; CHECK-NEXT: [[SRC_3:%.*]] = load i32, i32* [[SRC_GEP_3]], align 4 285; CHECK-NEXT: [[R_3:%.*]] = ashr i32 [[SRC_3]], 16 286; CHECK-NEXT: [[DST_GEP_3:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 3 287; CHECK-NEXT: store i32 [[R_3]], i32* [[DST_GEP_3]], align 4 288; CHECK-NEXT: call void @use(i32 [[L]]) 289; CHECK-NEXT: br label [[EXIT]] 290; CHECK: exit: 291; CHECK-NEXT: ret void 292; 293entry: 294 br i1 %c, label %then, label %exit 295 296then: 297 %src.0 = load i32, i32* %src, align 4 298 %r.0 = ashr i32 %src.0, 16 299 store i32 %r.0, i32* %dst, align 4 300 %src.gep.1 = getelementptr inbounds i32, i32* %src, i64 1 301 %src.1 = load i32, i32* %src.gep.1, align 4 302 %r.1 = ashr i32 %src.1, 16 303 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 304 store i32 %r.1, i32* %dst.gep.1, align 4 305 %src.gep.2 = getelementptr inbounds i32, i32* %src, i64 2 306 %src.2 = load i32, i32* %src.gep.2, align 4 307 %src.gep.5 = getelementptr inbounds i32, i32* %src, i64 5 308 %l = load i32, i32* %src.gep.5 309 %r.2 = ashr i32 %src.2, 16 310 %dst.gep.2 = getelementptr inbounds i32, i32* %dst, i64 2 311 store i32 %r.2, i32* %dst.gep.2, align 4 312 %src.gep.3 = getelementptr inbounds i32, i32* %src, i64 3 313 %src.3 = load i32, i32* %src.gep.3, align 4 314 %r.3 = ashr i32 %src.3, 16 315 %dst.gep.3 = getelementptr inbounds i32, i32* %dst, i64 3 316 store i32 %r.3, i32* %dst.gep.3, align 4 317 call void @use(i32 %l) 318 br label %exit 319 320exit: 321 ret void 322} 323 324define void @no_version(i32* nocapture %dst, i32* nocapture readonly %src) { 325; CHECK-LABEL: @no_version( 326; CHECK-NEXT: entry: 327; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[SRC:%.*]], i64 1 328; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[SRC]] to <2 x i32>* 329; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, <2 x i32>* [[TMP0]], align 4 330; CHECK-NEXT: [[TMP2:%.*]] = ashr <2 x i32> [[TMP1]], <i32 16, i32 16> 331; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i64 1 332; CHECK-NEXT: [[TMP3:%.*]] = bitcast i32* [[DST]] to <2 x i32>* 333; CHECK-NEXT: store <2 x i32> [[TMP2]], <2 x i32>* [[TMP3]], align 4 334; CHECK-NEXT: ret void 335; 336entry: 337 %src.0 = load i32, i32* %src, align 4 338 %src.gep.1 = getelementptr inbounds i32, i32* %src, i64 1 339 %src.1 = load i32, i32* %src.gep.1, align 4 340 %r.0 = ashr i32 %src.0, 16 341 %r.1 = ashr i32 %src.1, 16 342 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 343 store i32 %r.0, i32* %dst, align 4 344 store i32 %r.1, i32* %dst.gep.1, align 4 345 ret void 346} 347 348define void @version_multiple(i32* nocapture %out_block, i32* nocapture readonly %counter) { 349; CHECK-LABEL: @version_multiple( 350; CHECK-NEXT: entry: 351; CHECK-NEXT: [[TMP0:%.*]] = load i32, i32* [[COUNTER:%.*]], align 4 352; CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* [[OUT_BLOCK:%.*]], align 4 353; CHECK-NEXT: [[XOR:%.*]] = xor i32 [[TMP1]], [[TMP0]] 354; CHECK-NEXT: store i32 [[XOR]], i32* [[OUT_BLOCK]], align 4 355; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[COUNTER]], i64 1 356; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX_1]], align 4 357; CHECK-NEXT: [[ARRAYIDX2_1:%.*]] = getelementptr inbounds i32, i32* [[OUT_BLOCK]], i64 1 358; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* [[ARRAYIDX2_1]], align 4 359; CHECK-NEXT: [[XOR_1:%.*]] = xor i32 [[TMP3]], [[TMP2]] 360; CHECK-NEXT: store i32 [[XOR_1]], i32* [[ARRAYIDX2_1]], align 4 361; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[COUNTER]], i64 2 362; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32* [[ARRAYIDX_2]], align 4 363; CHECK-NEXT: [[ARRAYIDX2_2:%.*]] = getelementptr inbounds i32, i32* [[OUT_BLOCK]], i64 2 364; CHECK-NEXT: [[TMP5:%.*]] = load i32, i32* [[ARRAYIDX2_2]], align 4 365; CHECK-NEXT: [[XOR_2:%.*]] = xor i32 [[TMP5]], [[TMP4]] 366; CHECK-NEXT: store i32 [[XOR_2]], i32* [[ARRAYIDX2_2]], align 4 367; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[COUNTER]], i64 3 368; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32* [[ARRAYIDX_3]], align 4 369; CHECK-NEXT: [[ARRAYIDX2_3:%.*]] = getelementptr inbounds i32, i32* [[OUT_BLOCK]], i64 3 370; CHECK-NEXT: [[TMP7:%.*]] = load i32, i32* [[ARRAYIDX2_3]], align 4 371; CHECK-NEXT: [[XOR_3:%.*]] = xor i32 [[TMP7]], [[TMP6]] 372; CHECK-NEXT: store i32 [[XOR_3]], i32* [[ARRAYIDX2_3]], align 4 373; CHECK-NEXT: ret void 374; 375entry: 376 %0 = load i32, i32* %counter, align 4 377 %1 = load i32, i32* %out_block, align 4 378 %xor = xor i32 %1, %0 379 store i32 %xor, i32* %out_block, align 4 380 %arrayidx.1 = getelementptr inbounds i32, i32* %counter, i64 1 381 %2 = load i32, i32* %arrayidx.1, align 4 382 %arrayidx2.1 = getelementptr inbounds i32, i32* %out_block, i64 1 383 %3 = load i32, i32* %arrayidx2.1, align 4 384 %xor.1 = xor i32 %3, %2 385 store i32 %xor.1, i32* %arrayidx2.1, align 4 386 %arrayidx.2 = getelementptr inbounds i32, i32* %counter, i64 2 387 %4 = load i32, i32* %arrayidx.2, align 4 388 %arrayidx2.2 = getelementptr inbounds i32, i32* %out_block, i64 2 389 %5 = load i32, i32* %arrayidx2.2, align 4 390 %xor.2 = xor i32 %5, %4 391 store i32 %xor.2, i32* %arrayidx2.2, align 4 392 %arrayidx.3 = getelementptr inbounds i32, i32* %counter, i64 3 393 %6 = load i32, i32* %arrayidx.3, align 4 394 %arrayidx2.3 = getelementptr inbounds i32, i32* %out_block, i64 3 395 %7 = load i32, i32* %arrayidx2.3, align 4 396 %xor.3 = xor i32 %7, %6 397 store i32 %xor.3, i32* %arrayidx2.3, align 4 398 ret void 399} 400 401define i32 @use_outside_version_bb(i32* %dst, i32* %src, i1 %c.1) { 402; CHECK-LABEL: @use_outside_version_bb( 403; CHECK-NEXT: entry: 404; CHECK-NEXT: [[SRC_0:%.*]] = load i32, i32* [[SRC:%.*]], align 4 405; CHECK-NEXT: [[R_0:%.*]] = ashr i32 [[SRC_0]], 16 406; CHECK-NEXT: store i32 [[R_0]], i32* [[DST:%.*]], align 4 407; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 1 408; CHECK-NEXT: [[SRC_1:%.*]] = load i32, i32* [[SRC_GEP_1]], align 4 409; CHECK-NEXT: [[R_1:%.*]] = ashr i32 [[SRC_1]], 16 410; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 1 411; CHECK-NEXT: store i32 [[R_1]], i32* [[DST_GEP_1]], align 4 412; CHECK-NEXT: br label [[EXIT:%.*]] 413; CHECK: exit: 414; CHECK-NEXT: ret i32 [[R_0]] 415; 416entry: 417 %src.0 = load i32, i32* %src, align 4 418 %r.0 = ashr i32 %src.0, 16 419 store i32 %r.0, i32* %dst, align 4 420 %src.gep.1 = getelementptr inbounds i32, i32* %src, i64 1 421 %src.1 = load i32, i32* %src.gep.1, align 4 422 %r.1 = ashr i32 %src.1, 16 423 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 424 store i32 %r.1, i32* %dst.gep.1, align 4 425 br label %exit 426 427exit: 428 ret i32 %r.0 429} 430 431define i32 @value_used_in_return(i32* %dst, i32* %src, i32 %x) { 432; CHECK-LABEL: @value_used_in_return( 433; CHECK-NEXT: entry: 434; CHECK-NEXT: [[SRC_0:%.*]] = load i32, i32* [[SRC:%.*]], align 4 435; CHECK-NEXT: [[R_0:%.*]] = ashr i32 [[SRC_0]], 16 436; CHECK-NEXT: store i32 [[R_0]], i32* [[DST:%.*]], align 4 437; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 1 438; CHECK-NEXT: [[SRC_1:%.*]] = load i32, i32* [[SRC_GEP_1]], align 4 439; CHECK-NEXT: [[R_1:%.*]] = ashr i32 [[SRC_1]], 16 440; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 1 441; CHECK-NEXT: store i32 [[R_1]], i32* [[DST_GEP_1]], align 4 442; CHECK-NEXT: [[ADD:%.*]] = add i32 [[X:%.*]], 20 443; CHECK-NEXT: ret i32 [[ADD]] 444; 445entry: 446 %src.0 = load i32, i32* %src, align 4 447 %r.0 = ashr i32 %src.0, 16 448 store i32 %r.0, i32* %dst, align 4 449 %src.gep.1 = getelementptr inbounds i32, i32* %src, i64 1 450 %src.1 = load i32, i32* %src.gep.1, align 4 451 %r.1 = ashr i32 %src.1, 16 452 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 453 store i32 %r.1, i32* %dst.gep.1, align 4 454 %add = add i32 %x, 20 455 ret i32 %add 456} 457define i32 @needs_versioning2_cond_br(i32* %dst, i32* %src, i1 %c.1) { 458; CHECK-LABEL: @needs_versioning2_cond_br( 459; CHECK-NEXT: entry: 460; CHECK-NEXT: br i1 [[C_1:%.*]], label [[THEN:%.*]], label [[ELSE:%.*]] 461; CHECK: then: 462; CHECK-NEXT: [[SRC_0:%.*]] = load i32, i32* [[SRC:%.*]], align 4 463; CHECK-NEXT: [[R_0:%.*]] = ashr i32 [[SRC_0]], 16 464; CHECK-NEXT: store i32 [[R_0]], i32* [[DST:%.*]], align 4 465; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 1 466; CHECK-NEXT: [[SRC_1:%.*]] = load i32, i32* [[SRC_GEP_1]], align 4 467; CHECK-NEXT: [[R_1:%.*]] = ashr i32 [[SRC_1]], 16 468; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 1 469; CHECK-NEXT: store i32 [[R_1]], i32* [[DST_GEP_1]], align 4 470; CHECK-NEXT: ret i32 10 471; CHECK: else: 472; CHECK-NEXT: ret i32 0 473; 474entry: 475 br i1 %c.1, label %then, label %else 476 477then: 478 %src.0 = load i32, i32* %src, align 4 479 %r.0 = ashr i32 %src.0, 16 480 store i32 %r.0, i32* %dst, align 4 481 %src.gep.1 = getelementptr inbounds i32, i32* %src, i64 1 482 %src.1 = load i32, i32* %src.gep.1, align 4 483 %r.1 = ashr i32 %src.1, 16 484 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 485 store i32 %r.1, i32* %dst.gep.1, align 4 486 ret i32 10 487 488 489else: 490 ret i32 0 491} 492 493define void @pointer_defined_in_bb(i32* %dst, i32** %src.p) { 494; CHECK-LABEL: @pointer_defined_in_bb( 495; CHECK-NEXT: entry: 496; CHECK-NEXT: [[SRC:%.*]] = load i32*, i32** [[SRC_P:%.*]], align 8 497; CHECK-NEXT: [[SRC_0:%.*]] = load i32, i32* [[SRC]], align 4 498; CHECK-NEXT: [[R_0:%.*]] = ashr i32 [[SRC_0]], 16 499; CHECK-NEXT: store i32 [[R_0]], i32* [[DST:%.*]], align 4 500; CHECK-NEXT: [[SRC_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[SRC]], i64 1 501; CHECK-NEXT: [[SRC_1:%.*]] = load i32, i32* [[SRC_GEP_1]], align 4 502; CHECK-NEXT: [[R_1:%.*]] = ashr i32 [[SRC_1]], 16 503; CHECK-NEXT: [[DST_GEP_1:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 1 504; CHECK-NEXT: store i32 [[R_1]], i32* [[DST_GEP_1]], align 4 505; CHECK-NEXT: ret void 506; 507entry: 508 %src = load i32*, i32** %src.p 509 %src.0 = load i32, i32* %src, align 4 510 %r.0 = ashr i32 %src.0, 16 511 store i32 %r.0, i32* %dst, align 4 512 %src.gep.1 = getelementptr inbounds i32, i32* %src, i64 1 513 %src.1 = load i32, i32* %src.gep.1, align 4 514 %r.1 = ashr i32 %src.1, 16 515 %dst.gep.1 = getelementptr inbounds i32, i32* %dst, i64 1 516 store i32 %r.1, i32* %dst.gep.1, align 4 517 ret void 518} 519 520define void @clobber_same_underlying_object(i32* %this) { 521; CHECK-LABEL: @clobber_same_underlying_object( 522; CHECK-NEXT: entry: 523; CHECK-NEXT: [[P_3:%.*]] = getelementptr inbounds i32, i32* [[THIS:%.*]], i32 3 524; CHECK-NEXT: store i32 10, i32* [[P_3]], align 8 525; CHECK-NEXT: tail call void @clobber() 526; CHECK-NEXT: [[P_4:%.*]] = getelementptr inbounds i32, i32* [[THIS]], i32 4 527; CHECK-NEXT: [[L2:%.*]] = load i32, i32* [[P_4]], align 8 528; CHECK-NEXT: store i32 20, i32* [[P_4]], align 8 529; CHECK-NEXT: ret void 530; 531entry: 532 %p.3 = getelementptr inbounds i32, i32* %this, i32 3 533 store i32 10, i32* %p.3, align 8 534 tail call void @clobber() 535 %p.4 = getelementptr inbounds i32, i32* %this, i32 4 536 %l2 = load i32, i32* %p.4, align 8 537 store i32 20, i32* %p.4, align 8 538 ret void 539} 540 541declare void @clobber() 542 543define void @slp_not_beneficial(i32* %A, i32* %B) { 544; CHECK-LABEL: @slp_not_beneficial( 545; CHECK-NEXT: bb: 546; CHECK-NEXT: [[TMP:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 4 547; CHECK-NEXT: store i32 0, i32* [[TMP]], align 8 548; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 5 549; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i32 4 550; CHECK-NEXT: [[TMP5:%.*]] = load i32, i32* [[TMP4]], align 8 551; CHECK-NEXT: store i32 [[TMP5]], i32* [[TMP3]], align 8 552; CHECK-NEXT: ret void 553; 554bb: 555 %tmp = getelementptr inbounds i32, i32* %A, i32 4 556 store i32 0, i32* %tmp, align 8 557 %tmp3 = getelementptr inbounds i32, i32* %A, i32 5 558 %tmp4 = getelementptr inbounds i32, i32* %B, i32 4 559 %tmp5 = load i32, i32* %tmp4, align 8 560 store i32 %tmp5, i32* %tmp3, align 8 561 ret void 562} 563 564define void @widget(double* %ptr, double* %ptr.2) { 565; CHECK-LABEL: @widget( 566; CHECK-NEXT: bb1: 567; CHECK-NEXT: [[TMP3:%.*]] = load double, double* null, align 8 568; CHECK-NEXT: [[TMP4:%.*]] = fmul double undef, [[TMP3]] 569; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds double, double* [[PTR:%.*]], i32 0 570; CHECK-NEXT: [[TMP6:%.*]] = load double, double* [[TMP5]], align 8 571; CHECK-NEXT: [[TMP7:%.*]] = fadd double [[TMP6]], [[TMP4]] 572; CHECK-NEXT: store double [[TMP7]], double* [[TMP5]], align 8 573; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds double, double* [[PTR_2:%.*]], i64 0 574; CHECK-NEXT: [[TMP9:%.*]] = load double, double* [[TMP8]], align 8 575; CHECK-NEXT: [[TMP10:%.*]] = fmul double undef, [[TMP9]] 576; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds double, double* [[PTR]], i32 1 577; CHECK-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8 578; CHECK-NEXT: [[TMP13:%.*]] = fadd double [[TMP12]], [[TMP10]] 579; CHECK-NEXT: store double [[TMP13]], double* [[TMP11]], align 8 580; CHECK-NEXT: br label [[BB15:%.*]] 581; CHECK: bb15: 582; CHECK-NEXT: br label [[BB15]] 583; 584bb1: ; preds = %bb 585 %tmp3 = load double, double* null, align 8 586 %tmp4 = fmul double undef, %tmp3 587 %tmp5 = getelementptr inbounds double, double* %ptr, i32 0 588 %tmp6 = load double, double* %tmp5, align 8 589 %tmp7 = fadd double %tmp6, %tmp4 590 store double %tmp7, double* %tmp5, align 8 591 %tmp8 = getelementptr inbounds double, double* %ptr.2, i64 0 592 %tmp9 = load double, double* %tmp8, align 8 593 %tmp10 = fmul double undef, %tmp9 594 %tmp11 = getelementptr inbounds double, double* %ptr, i32 1 595 %tmp12 = load double, double* %tmp11, align 8 596 %tmp13 = fadd double %tmp12, %tmp10 597 store double %tmp13, double* %tmp11, align 8 598 br label %bb15 599 600bb15: ; preds = %bb15, %bb14 601 br label %bb15 602} 603 604%struct = type { i32, i32, float, float } 605 606; Some points we collected as candidates for runtime checks have been removed 607; before generating runtime checks. Make sure versioning is skipped. 608define void @test_bounds_removed_before_runtime_checks(%struct * %A, i32** %B, i1 %c) { 609; CHECK-LABEL: @test_bounds_removed_before_runtime_checks( 610; CHECK-NEXT: entry: 611; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT:%.*]], %struct* [[A:%.*]], i64 0, i32 0 612; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT]], %struct* [[A]], i64 0, i32 1 613; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[TMP11]] to <2 x i32>* 614; CHECK-NEXT: store <2 x i32> <i32 10, i32 300>, <2 x i32>* [[TMP0]], align 8 615; CHECK-NEXT: [[TMP13:%.*]] = load i32*, i32** [[B:%.*]], align 8 616; CHECK-NEXT: br i1 [[C:%.*]], label [[BB23:%.*]], label [[BB14:%.*]] 617; CHECK: bb14: 618; CHECK-NEXT: [[TMP15:%.*]] = sext i32 10 to i64 619; CHECK-NEXT: [[TMP16:%.*]] = add nsw i64 2, [[TMP15]] 620; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i64 [[TMP16]] 621; CHECK-NEXT: [[TMP18:%.*]] = bitcast i32* [[TMP17]] to i8* 622; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i8, i8* [[TMP18]], i64 3 623; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[STRUCT]], %struct* [[A]], i64 0, i32 2 624; CHECK-NEXT: store float 0.000000e+00, float* [[TMP20]], align 8 625; CHECK-NEXT: [[TMP21:%.*]] = load i8, i8* [[TMP19]], align 1 626; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT]], %struct* [[A]], i64 0, i32 3 627; CHECK-NEXT: store float 0.000000e+00, float* [[TMP22]], align 4 628; CHECK-NEXT: br label [[BB23]] 629; CHECK: bb23: 630; CHECK-NEXT: ret void 631; 632entry: 633 %tmp1 = fmul float 10.0, 20.0 634 %tmp2 = fptosi float %tmp1 to i32 635 %tmp3 = fmul float 30.0, 20.0 636 %tmp4 = fptosi float %tmp3 to i32 637 %tmp5 = icmp sgt i32 100, %tmp2 638 %tmp6 = select i1 %tmp5, i32 %tmp2, i32 10 639 %tmp7 = select i1 false, i32 0, i32 %tmp6 640 %tmp8 = icmp sgt i32 200, %tmp4 641 %tmp9 = select i1 %tmp8, i32 %tmp4, i32 300 642 %tmp10 = select i1 false, i32 0, i32 %tmp9 643 %tmp11 = getelementptr inbounds %struct, %struct* %A, i64 0, i32 0 644 store i32 %tmp7, i32* %tmp11, align 8 645 %tmp12 = getelementptr inbounds %struct, %struct* %A, i64 0, i32 1 646 store i32 %tmp10, i32* %tmp12, align 4 647 %tmp13 = load i32*, i32** %B, align 8 648 br i1 %c, label %bb23, label %bb14 649 650bb14: 651 %tmp15 = sext i32 %tmp7 to i64 652 %tmp16 = add nsw i64 2, %tmp15 653 %tmp17 = getelementptr inbounds i32, i32* %tmp13, i64 %tmp16 654 %tmp18 = bitcast i32* %tmp17 to i8* 655 %tmp19 = getelementptr inbounds i8, i8* %tmp18, i64 3 656 %tmp20 = getelementptr inbounds %struct, %struct* %A, i64 0, i32 2 657 store float 0.0, float* %tmp20, align 8 658 %tmp21 = load i8, i8* %tmp19, align 1 659 %tmp22 = getelementptr inbounds %struct, %struct* %A, i64 0, i32 3 660 store float 0.0, float* %tmp22, align 4 661 br label %bb23 662 663bb23: 664 ret void 665} 666 667; In this test there's a single bound, do not generate runtime checks. 668define void @single_membound(double* %arg, double* %arg1, double %x) { 669; CHECK-LABEL: @single_membound( 670; CHECK-NEXT: entry: 671; CHECK-NEXT: [[TMP:%.*]] = fsub double [[X:%.*]], 9.900000e+01 672; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds double, double* [[ARG:%.*]], i64 1 673; CHECK-NEXT: store double [[TMP]], double* [[TMP9]], align 8 674; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds double, double* [[ARG1:%.*]], i64 0 675; CHECK-NEXT: [[TMP12:%.*]] = load double, double* [[TMP10]], align 8 676; CHECK-NEXT: [[TMP13:%.*]] = fsub double 1.000000e+00, [[TMP12]] 677; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds double, double* [[ARG]], i64 2 678; CHECK-NEXT: br label [[BB15:%.*]] 679; CHECK: bb15: 680; CHECK-NEXT: [[TMP16:%.*]] = fmul double [[TMP]], 2.000000e+01 681; CHECK-NEXT: store double [[TMP16]], double* [[TMP9]], align 8 682; CHECK-NEXT: [[TMP17:%.*]] = fmul double [[TMP13]], 3.000000e+01 683; CHECK-NEXT: store double [[TMP17]], double* [[TMP14]], align 8 684; CHECK-NEXT: ret void 685; 686entry: 687 %tmp = fsub double %x, 99.0 688 %tmp9 = getelementptr inbounds double, double* %arg, i64 1 689 store double %tmp, double* %tmp9, align 8 690 %tmp10 = getelementptr inbounds double, double* %arg1, i64 0 691 %tmp12 = load double, double* %tmp10, align 8 692 %tmp13 = fsub double 1.0, %tmp12 693 %tmp14 = getelementptr inbounds double, double* %arg, i64 2 694 br label %bb15 695 696bb15: 697 %tmp16 = fmul double %tmp, 20.0 698 store double %tmp16, double* %tmp9, align 8 699 %tmp17 = fmul double %tmp13, 30.0 700 store double %tmp17, double* %tmp14, align 8 701 ret void 702} 703 704%struct.2 = type { [4 x float] } 705 706; Make sure we do not crash when we encounter a SCEVCouldNotCompute. 707define void @no_lcssa_phi(%struct.2* %A, float* %B, i1 %c) { 708; CHECK-LABEL: @no_lcssa_phi( 709; CHECK-NEXT: bb: 710; CHECK-NEXT: br label [[LOOP:%.*]] 711; CHECK: loop: 712; CHECK-NEXT: [[PTR_PHI:%.*]] = phi %struct.2* [ [[A:%.*]], [[BB:%.*]] ], [ null, [[LOOP]] ] 713; CHECK-NEXT: br i1 [[C:%.*]], label [[EXIT:%.*]], label [[LOOP]] 714; CHECK: exit: 715; CHECK-NEXT: [[B_GEP_0:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 0 716; CHECK-NEXT: [[L_0:%.*]] = load float, float* [[B_GEP_0]], align 8 717; CHECK-NEXT: [[ADD_0:%.*]] = fadd float [[L_0]], 1.000000e+01 718; CHECK-NEXT: [[MUL_0:%.*]] = fmul float [[ADD_0]], 3.000000e+01 719; CHECK-NEXT: [[A_GEP_0:%.*]] = getelementptr inbounds [[STRUCT_2:%.*]], %struct.2* [[PTR_PHI]], i64 0, i32 0, i32 0 720; CHECK-NEXT: store float [[MUL_0]], float* [[A_GEP_0]], align 8 721; CHECK-NEXT: [[B_GEP_1:%.*]] = getelementptr inbounds float, float* [[B]], i64 1 722; CHECK-NEXT: [[L_1:%.*]] = load float, float* [[B_GEP_1]], align 8 723; CHECK-NEXT: [[ADD_1:%.*]] = fadd float [[L_1]], 1.000000e+01 724; CHECK-NEXT: [[MUL_1:%.*]] = fmul float [[ADD_1]], 3.000000e+01 725; CHECK-NEXT: [[A_GEP_1:%.*]] = getelementptr inbounds [[STRUCT_2]], %struct.2* [[PTR_PHI]], i64 0, i32 0, i32 1 726; CHECK-NEXT: store float [[MUL_1]], float* [[A_GEP_1]], align 8 727; CHECK-NEXT: [[B_GEP_2:%.*]] = getelementptr inbounds float, float* [[B]], i64 2 728; CHECK-NEXT: [[L_2:%.*]] = load float, float* [[B_GEP_2]], align 8 729; CHECK-NEXT: [[ADD_2:%.*]] = fadd float [[L_2]], 1.000000e+01 730; CHECK-NEXT: [[MUL_2:%.*]] = fmul float [[ADD_2]], 3.000000e+01 731; CHECK-NEXT: [[A_GEP_2:%.*]] = getelementptr inbounds [[STRUCT_2]], %struct.2* [[PTR_PHI]], i64 0, i32 0, i32 2 732; CHECK-NEXT: store float [[MUL_2]], float* [[A_GEP_2]], align 8 733; CHECK-NEXT: [[B_GEP_3:%.*]] = getelementptr inbounds float, float* [[B]], i64 3 734; CHECK-NEXT: [[L_3:%.*]] = load float, float* [[B_GEP_3]], align 8 735; CHECK-NEXT: [[ADD_3:%.*]] = fadd float [[L_3]], 1.000000e+01 736; CHECK-NEXT: [[MUL_3:%.*]] = fmul float [[ADD_3]], 3.000000e+01 737; CHECK-NEXT: [[A_GEP_3:%.*]] = getelementptr inbounds [[STRUCT_2]], %struct.2* [[PTR_PHI]], i64 0, i32 0, i32 3 738; CHECK-NEXT: store float [[MUL_3]], float* [[A_GEP_3]], align 8 739; CHECK-NEXT: ret void 740; 741bb: 742 br label %loop 743 744loop: 745 %ptr.phi = phi %struct.2* [ %A, %bb ], [ null, %loop ] 746 br i1 %c, label %exit, label %loop 747 748exit: 749 %B.gep.0 = getelementptr inbounds float, float* %B, i64 0 750 %l.0 = load float, float* %B.gep.0, align 8 751 %add.0 = fadd float %l.0, 10.0 752 %mul.0 = fmul float %add.0, 30.0 753 %A.gep.0 = getelementptr inbounds %struct.2, %struct.2* %ptr.phi, i64 0, i32 0, i32 0 754 store float %mul.0, float* %A.gep.0, align 8 755 %B.gep.1 = getelementptr inbounds float, float* %B, i64 1 756 %l.1 = load float, float* %B.gep.1, align 8 757 %add.1 = fadd float %l.1, 10.0 758 %mul.1 = fmul float %add.1, 30.0 759 %A.gep.1 = getelementptr inbounds %struct.2, %struct.2* %ptr.phi, i64 0, i32 0, i32 1 760 store float %mul.1, float* %A.gep.1, align 8 761 %B.gep.2 = getelementptr inbounds float, float* %B, i64 2 762 %l.2 = load float, float* %B.gep.2, align 8 763 %add.2 = fadd float %l.2, 10.0 764 %mul.2 = fmul float %add.2, 30.0 765 %A.gep.2 = getelementptr inbounds %struct.2, %struct.2* %ptr.phi, i64 0, i32 0, i32 2 766 store float %mul.2, float* %A.gep.2, align 8 767 %B.gep.3 = getelementptr inbounds float, float* %B, i64 3 768 %l.3 = load float, float* %B.gep.3, align 8 769 %add.3 = fadd float %l.3, 10.0 770 %mul.3 = fmul float %add.3, 30.0 771 %A.gep.3 = getelementptr inbounds %struct.2, %struct.2* %ptr.phi, i64 0, i32 0, i32 3 772 store float %mul.3, float* %A.gep.3, align 8 773 ret void 774} 775 776; Make sure lcssa phis as pointer bases are handled properly. 777define void @lcssa_phi(%struct.2* %A, float* %B, i1 %c) { 778; CHECK-LABEL: @lcssa_phi( 779; CHECK-NEXT: bb: 780; CHECK-NEXT: br label [[LOOP:%.*]] 781; CHECK: loop: 782; CHECK-NEXT: [[PTR_PHI:%.*]] = phi %struct.2* [ [[A:%.*]], [[BB:%.*]] ], [ null, [[LOOP]] ] 783; CHECK-NEXT: br i1 [[C:%.*]], label [[EXIT:%.*]], label [[LOOP]] 784; CHECK: exit: 785; CHECK-NEXT: [[PTR_PHI_LCSSA:%.*]] = phi %struct.2* [ [[PTR_PHI]], [[LOOP]] ] 786; CHECK-NEXT: [[B_GEP_0:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 0 787; CHECK-NEXT: [[L_0:%.*]] = load float, float* [[B_GEP_0]], align 8 788; CHECK-NEXT: [[ADD_0:%.*]] = fadd float [[L_0]], 1.000000e+01 789; CHECK-NEXT: [[MUL_0:%.*]] = fmul float [[ADD_0]], 3.000000e+01 790; CHECK-NEXT: [[A_GEP_0:%.*]] = getelementptr inbounds [[STRUCT_2:%.*]], %struct.2* [[PTR_PHI_LCSSA]], i64 0, i32 0, i32 0 791; CHECK-NEXT: store float [[MUL_0]], float* [[A_GEP_0]], align 8 792; CHECK-NEXT: [[B_GEP_1:%.*]] = getelementptr inbounds float, float* [[B]], i64 1 793; CHECK-NEXT: [[L_1:%.*]] = load float, float* [[B_GEP_1]], align 8 794; CHECK-NEXT: [[ADD_1:%.*]] = fadd float [[L_1]], 1.000000e+01 795; CHECK-NEXT: [[MUL_1:%.*]] = fmul float [[ADD_1]], 3.000000e+01 796; CHECK-NEXT: [[A_GEP_1:%.*]] = getelementptr inbounds [[STRUCT_2]], %struct.2* [[PTR_PHI_LCSSA]], i64 0, i32 0, i32 1 797; CHECK-NEXT: store float [[MUL_1]], float* [[A_GEP_1]], align 8 798; CHECK-NEXT: [[B_GEP_2:%.*]] = getelementptr inbounds float, float* [[B]], i64 2 799; CHECK-NEXT: [[L_2:%.*]] = load float, float* [[B_GEP_2]], align 8 800; CHECK-NEXT: [[ADD_2:%.*]] = fadd float [[L_2]], 1.000000e+01 801; CHECK-NEXT: [[MUL_2:%.*]] = fmul float [[ADD_2]], 3.000000e+01 802; CHECK-NEXT: [[A_GEP_2:%.*]] = getelementptr inbounds [[STRUCT_2]], %struct.2* [[PTR_PHI_LCSSA]], i64 0, i32 0, i32 2 803; CHECK-NEXT: store float [[MUL_2]], float* [[A_GEP_2]], align 8 804; CHECK-NEXT: [[B_GEP_3:%.*]] = getelementptr inbounds float, float* [[B]], i64 3 805; CHECK-NEXT: [[L_3:%.*]] = load float, float* [[B_GEP_3]], align 8 806; CHECK-NEXT: [[ADD_3:%.*]] = fadd float [[L_3]], 1.000000e+01 807; CHECK-NEXT: [[MUL_3:%.*]] = fmul float [[ADD_3]], 3.000000e+01 808; CHECK-NEXT: [[A_GEP_3:%.*]] = getelementptr inbounds [[STRUCT_2]], %struct.2* [[PTR_PHI_LCSSA]], i64 0, i32 0, i32 3 809; CHECK-NEXT: store float [[MUL_3]], float* [[A_GEP_3]], align 8 810; CHECK-NEXT: ret void 811; 812bb: 813 br label %loop 814 815loop: 816 %ptr.phi = phi %struct.2* [ %A, %bb ], [ null, %loop ] 817 br i1 %c, label %exit, label %loop 818 819exit: 820 %ptr.phi.lcssa = phi %struct.2* [ %ptr.phi, %loop ] 821 %B.gep.0 = getelementptr inbounds float, float* %B, i64 0 822 %l.0 = load float, float* %B.gep.0, align 8 823 %add.0 = fadd float %l.0, 10.0 824 %mul.0 = fmul float %add.0, 30.0 825 %A.gep.0 = getelementptr inbounds %struct.2, %struct.2* %ptr.phi.lcssa, i64 0, i32 0, i32 0 826 store float %mul.0, float* %A.gep.0, align 8 827 %B.gep.1 = getelementptr inbounds float, float* %B, i64 1 828 %l.1 = load float, float* %B.gep.1, align 8 829 %add.1 = fadd float %l.1, 10.0 830 %mul.1 = fmul float %add.1, 30.0 831 %A.gep.1 = getelementptr inbounds %struct.2, %struct.2* %ptr.phi.lcssa, i64 0, i32 0, i32 1 832 store float %mul.1, float* %A.gep.1, align 8 833 %B.gep.2 = getelementptr inbounds float, float* %B, i64 2 834 %l.2 = load float, float* %B.gep.2, align 8 835 %add.2 = fadd float %l.2, 10.0 836 %mul.2 = fmul float %add.2, 30.0 837 %A.gep.2 = getelementptr inbounds %struct.2, %struct.2* %ptr.phi.lcssa, i64 0, i32 0, i32 2 838 store float %mul.2, float* %A.gep.2, align 8 839 %B.gep.3 = getelementptr inbounds float, float* %B, i64 3 840 %l.3 = load float, float* %B.gep.3, align 8 841 %add.3 = fadd float %l.3, 10.0 842 %mul.3 = fmul float %add.3, 30.0 843 %A.gep.3 = getelementptr inbounds %struct.2, %struct.2* %ptr.phi.lcssa, i64 0, i32 0, i32 3 844 store float %mul.3, float* %A.gep.3, align 8 845 ret void 846} 847 848%struct.spam = type { [60 x i32], i32, [12 x i8] } 849 850declare void @foo(i8*) 851 852; Test case with a basic block where parts can be vectorized without versioning. 853define i32 @block_partly_vectorized_without_versioning(%struct.spam* readonly %arg, i8* nocapture readonly %arg1, i8* nocapture %arg2, i8* nocapture readonly %arg3, i8* %A, i8* %B) { 854; CHECK-LABEL: @block_partly_vectorized_without_versioning( 855; CHECK-NEXT: bb: 856; CHECK-NEXT: [[T:%.*]] = alloca <16 x i8>, align 16 857; CHECK-NEXT: [[T4:%.*]] = getelementptr inbounds <16 x i8>, <16 x i8>* [[T]], i64 0, i64 0 858; CHECK-NEXT: [[T5:%.*]] = getelementptr inbounds i8, i8* [[ARG3:%.*]], i64 1 859; CHECK-NEXT: [[T6:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 2 860; CHECK-NEXT: [[T7:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 3 861; CHECK-NEXT: [[T8:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 4 862; CHECK-NEXT: [[T9:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 5 863; CHECK-NEXT: [[T10:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 6 864; CHECK-NEXT: [[T11:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 7 865; CHECK-NEXT: [[T12:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 8 866; CHECK-NEXT: [[T13:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 9 867; CHECK-NEXT: [[T14:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 10 868; CHECK-NEXT: [[T15:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 11 869; CHECK-NEXT: [[T16:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 12 870; CHECK-NEXT: [[T17:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 13 871; CHECK-NEXT: [[T18:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 14 872; CHECK-NEXT: [[T19:%.*]] = bitcast i8* [[ARG1:%.*]] to <16 x i8>* 873; CHECK-NEXT: [[A_GEP_0:%.*]] = getelementptr i8, i8* [[A:%.*]], i64 0 874; CHECK-NEXT: [[B_GEP_0:%.*]] = getelementptr i8, i8* [[B:%.*]], i64 0 875; CHECK-NEXT: [[A_GEP_1:%.*]] = getelementptr i8, i8* [[A]], i64 1 876; CHECK-NEXT: [[B_GEP_1:%.*]] = getelementptr i8, i8* [[B]], i64 1 877; CHECK-NEXT: [[A_GEP_2:%.*]] = getelementptr i8, i8* [[A]], i64 2 878; CHECK-NEXT: [[B_GEP_2:%.*]] = getelementptr i8, i8* [[B]], i64 2 879; CHECK-NEXT: [[A_GEP_3:%.*]] = getelementptr i8, i8* [[A]], i64 3 880; CHECK-NEXT: [[B_GEP_3:%.*]] = getelementptr i8, i8* [[B]], i64 3 881; CHECK-NEXT: [[A_GEP_4:%.*]] = getelementptr i8, i8* [[A]], i64 4 882; CHECK-NEXT: [[B_GEP_4:%.*]] = getelementptr i8, i8* [[B]], i64 4 883; CHECK-NEXT: [[A_GEP_5:%.*]] = getelementptr i8, i8* [[A]], i64 5 884; CHECK-NEXT: [[B_GEP_5:%.*]] = getelementptr i8, i8* [[B]], i64 5 885; CHECK-NEXT: [[A_GEP_6:%.*]] = getelementptr i8, i8* [[A]], i64 6 886; CHECK-NEXT: [[B_GEP_6:%.*]] = getelementptr i8, i8* [[B]], i64 6 887; CHECK-NEXT: [[A_GEP_7:%.*]] = getelementptr i8, i8* [[A]], i64 7 888; CHECK-NEXT: [[B_GEP_7:%.*]] = getelementptr i8, i8* [[B]], i64 7 889; CHECK-NEXT: [[A_GEP_8:%.*]] = getelementptr i8, i8* [[A]], i64 8 890; CHECK-NEXT: [[B_GEP_8:%.*]] = getelementptr i8, i8* [[B]], i64 8 891; CHECK-NEXT: [[A_GEP_9:%.*]] = getelementptr i8, i8* [[A]], i64 9 892; CHECK-NEXT: [[B_GEP_9:%.*]] = getelementptr i8, i8* [[B]], i64 9 893; CHECK-NEXT: [[A_GEP_10:%.*]] = getelementptr i8, i8* [[A]], i64 10 894; CHECK-NEXT: [[B_GEP_10:%.*]] = getelementptr i8, i8* [[B]], i64 10 895; CHECK-NEXT: [[A_GEP_11:%.*]] = getelementptr i8, i8* [[A]], i64 11 896; CHECK-NEXT: [[B_GEP_11:%.*]] = getelementptr i8, i8* [[B]], i64 11 897; CHECK-NEXT: [[A_GEP_12:%.*]] = getelementptr i8, i8* [[A]], i64 12 898; CHECK-NEXT: [[B_GEP_12:%.*]] = getelementptr i8, i8* [[B]], i64 12 899; CHECK-NEXT: [[A_GEP_13:%.*]] = getelementptr i8, i8* [[A]], i64 13 900; CHECK-NEXT: [[B_GEP_13:%.*]] = getelementptr i8, i8* [[B]], i64 13 901; CHECK-NEXT: [[A_GEP_14:%.*]] = getelementptr i8, i8* [[A]], i64 14 902; CHECK-NEXT: [[B_GEP_14:%.*]] = getelementptr i8, i8* [[B]], i64 14 903; CHECK-NEXT: [[A_GEP_15:%.*]] = getelementptr i8, i8* [[A]], i64 15 904; CHECK-NEXT: [[TMP0:%.*]] = bitcast i8* [[A_GEP_0]] to <16 x i8>* 905; CHECK-NEXT: [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* [[TMP0]], align 1 906; CHECK-NEXT: [[B_GEP_15:%.*]] = getelementptr i8, i8* [[B]], i64 15 907; CHECK-NEXT: [[TMP2:%.*]] = bitcast i8* [[B_GEP_0]] to <16 x i8>* 908; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i8>, <16 x i8>* [[TMP2]], align 1 909; CHECK-NEXT: [[TMP4:%.*]] = xor <16 x i8> [[TMP1]], [[TMP3]] 910; CHECK-NEXT: [[R_GEP_0:%.*]] = getelementptr i8, i8* [[ARG1]], i64 0 911; CHECK-NEXT: [[R_GEP_1:%.*]] = getelementptr i8, i8* [[ARG1]], i64 1 912; CHECK-NEXT: [[R_GEP_2:%.*]] = getelementptr i8, i8* [[ARG1]], i64 2 913; CHECK-NEXT: [[R_GEP_3:%.*]] = getelementptr i8, i8* [[ARG1]], i64 3 914; CHECK-NEXT: [[R_GEP_4:%.*]] = getelementptr i8, i8* [[ARG1]], i64 4 915; CHECK-NEXT: [[R_GEP_5:%.*]] = getelementptr i8, i8* [[ARG1]], i64 5 916; CHECK-NEXT: [[R_GEP_6:%.*]] = getelementptr i8, i8* [[ARG1]], i64 6 917; CHECK-NEXT: [[R_GEP_7:%.*]] = getelementptr i8, i8* [[ARG1]], i64 7 918; CHECK-NEXT: [[R_GEP_8:%.*]] = getelementptr i8, i8* [[ARG1]], i64 8 919; CHECK-NEXT: [[R_GEP_9:%.*]] = getelementptr i8, i8* [[ARG1]], i64 9 920; CHECK-NEXT: [[R_GEP_10:%.*]] = getelementptr i8, i8* [[ARG1]], i64 10 921; CHECK-NEXT: [[R_GEP_11:%.*]] = getelementptr i8, i8* [[ARG1]], i64 11 922; CHECK-NEXT: [[R_GEP_12:%.*]] = getelementptr i8, i8* [[ARG1]], i64 12 923; CHECK-NEXT: [[R_GEP_13:%.*]] = getelementptr i8, i8* [[ARG1]], i64 13 924; CHECK-NEXT: [[R_GEP_14:%.*]] = getelementptr i8, i8* [[ARG1]], i64 14 925; CHECK-NEXT: [[R_GEP_15:%.*]] = getelementptr i8, i8* [[ARG1]], i64 15 926; CHECK-NEXT: [[TMP5:%.*]] = bitcast i8* [[R_GEP_0]] to <16 x i8>* 927; CHECK-NEXT: store <16 x i8> [[TMP4]], <16 x i8>* [[TMP5]], align 1 928; CHECK-NEXT: [[T21:%.*]] = getelementptr inbounds i8, i8* [[ARG3]], i64 15 929; CHECK-NEXT: [[T22:%.*]] = bitcast i8* [[ARG3]] to <16 x i8>* 930; CHECK-NEXT: call void @foo(i8* nonnull [[T4]]) 931; CHECK-NEXT: [[T26:%.*]] = load i8, i8* [[ARG3]], align 1 932; CHECK-NEXT: [[T27:%.*]] = load i8, i8* [[ARG2:%.*]], align 1 933; CHECK-NEXT: [[T28:%.*]] = xor i8 [[T27]], [[T26]] 934; CHECK-NEXT: store i8 [[T28]], i8* [[ARG2]], align 1 935; CHECK-NEXT: [[T29:%.*]] = load i8, i8* [[T5]], align 1 936; CHECK-NEXT: [[T30:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 1 937; CHECK-NEXT: [[T31:%.*]] = load i8, i8* [[T30]], align 1 938; CHECK-NEXT: [[T32:%.*]] = xor i8 [[T31]], [[T29]] 939; CHECK-NEXT: store i8 [[T32]], i8* [[T30]], align 1 940; CHECK-NEXT: [[T33:%.*]] = load i8, i8* [[T6]], align 1 941; CHECK-NEXT: [[T34:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 2 942; CHECK-NEXT: [[T35:%.*]] = load i8, i8* [[T34]], align 1 943; CHECK-NEXT: [[T36:%.*]] = xor i8 [[T35]], [[T33]] 944; CHECK-NEXT: store i8 [[T36]], i8* [[T34]], align 1 945; CHECK-NEXT: [[T37:%.*]] = load i8, i8* [[T7]], align 1 946; CHECK-NEXT: [[T38:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 3 947; CHECK-NEXT: [[T39:%.*]] = load i8, i8* [[T38]], align 1 948; CHECK-NEXT: [[T40:%.*]] = xor i8 [[T39]], [[T37]] 949; CHECK-NEXT: store i8 [[T40]], i8* [[T38]], align 1 950; CHECK-NEXT: [[T41:%.*]] = load i8, i8* [[T8]], align 1 951; CHECK-NEXT: [[T42:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 4 952; CHECK-NEXT: [[T43:%.*]] = load i8, i8* [[T42]], align 1 953; CHECK-NEXT: [[T44:%.*]] = xor i8 [[T43]], [[T41]] 954; CHECK-NEXT: store i8 [[T44]], i8* [[T42]], align 1 955; CHECK-NEXT: [[T45:%.*]] = load i8, i8* [[T9]], align 1 956; CHECK-NEXT: [[T46:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 5 957; CHECK-NEXT: [[T47:%.*]] = load i8, i8* [[T46]], align 1 958; CHECK-NEXT: [[T48:%.*]] = xor i8 [[T47]], [[T45]] 959; CHECK-NEXT: store i8 [[T48]], i8* [[T46]], align 1 960; CHECK-NEXT: [[T49:%.*]] = load i8, i8* [[T10]], align 1 961; CHECK-NEXT: [[T50:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 6 962; CHECK-NEXT: [[T51:%.*]] = load i8, i8* [[T50]], align 1 963; CHECK-NEXT: [[T52:%.*]] = xor i8 [[T51]], [[T49]] 964; CHECK-NEXT: store i8 [[T52]], i8* [[T50]], align 1 965; CHECK-NEXT: [[T53:%.*]] = load i8, i8* [[T11]], align 1 966; CHECK-NEXT: [[T54:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 7 967; CHECK-NEXT: [[T55:%.*]] = load i8, i8* [[T54]], align 1 968; CHECK-NEXT: [[T56:%.*]] = xor i8 [[T55]], [[T53]] 969; CHECK-NEXT: store i8 [[T56]], i8* [[T54]], align 1 970; CHECK-NEXT: [[T57:%.*]] = load i8, i8* [[T12]], align 1 971; CHECK-NEXT: [[T58:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 8 972; CHECK-NEXT: [[T59:%.*]] = load i8, i8* [[T58]], align 1 973; CHECK-NEXT: [[T60:%.*]] = xor i8 [[T59]], [[T57]] 974; CHECK-NEXT: store i8 [[T60]], i8* [[T58]], align 1 975; CHECK-NEXT: [[T61:%.*]] = load i8, i8* [[T13]], align 1 976; CHECK-NEXT: [[T62:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 9 977; CHECK-NEXT: [[T63:%.*]] = load i8, i8* [[T62]], align 1 978; CHECK-NEXT: [[T64:%.*]] = xor i8 [[T63]], [[T61]] 979; CHECK-NEXT: store i8 [[T64]], i8* [[T62]], align 1 980; CHECK-NEXT: [[T65:%.*]] = load i8, i8* [[T14]], align 1 981; CHECK-NEXT: [[T66:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 10 982; CHECK-NEXT: [[T67:%.*]] = load i8, i8* [[T66]], align 1 983; CHECK-NEXT: [[T68:%.*]] = xor i8 [[T67]], [[T65]] 984; CHECK-NEXT: store i8 [[T68]], i8* [[T66]], align 1 985; CHECK-NEXT: [[T69:%.*]] = load i8, i8* [[T15]], align 1 986; CHECK-NEXT: [[T70:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 11 987; CHECK-NEXT: [[T71:%.*]] = load i8, i8* [[T70]], align 1 988; CHECK-NEXT: [[T72:%.*]] = xor i8 [[T71]], [[T69]] 989; CHECK-NEXT: store i8 [[T72]], i8* [[T70]], align 1 990; CHECK-NEXT: [[T73:%.*]] = load i8, i8* [[T16]], align 1 991; CHECK-NEXT: [[T74:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 12 992; CHECK-NEXT: [[T75:%.*]] = load i8, i8* [[T74]], align 1 993; CHECK-NEXT: [[T76:%.*]] = xor i8 [[T75]], [[T73]] 994; CHECK-NEXT: store i8 [[T76]], i8* [[T74]], align 1 995; CHECK-NEXT: [[T77:%.*]] = load i8, i8* [[T17]], align 1 996; CHECK-NEXT: [[T78:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 13 997; CHECK-NEXT: [[T79:%.*]] = load i8, i8* [[T78]], align 1 998; CHECK-NEXT: [[T80:%.*]] = xor i8 [[T79]], [[T77]] 999; CHECK-NEXT: store i8 [[T80]], i8* [[T78]], align 1 1000; CHECK-NEXT: [[T81:%.*]] = load i8, i8* [[T18]], align 1 1001; CHECK-NEXT: [[T82:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 14 1002; CHECK-NEXT: [[T83:%.*]] = load i8, i8* [[T82]], align 1 1003; CHECK-NEXT: [[T84:%.*]] = xor i8 [[T83]], [[T81]] 1004; CHECK-NEXT: store i8 [[T84]], i8* [[T82]], align 1 1005; CHECK-NEXT: [[T85:%.*]] = load i8, i8* [[T21]], align 1 1006; CHECK-NEXT: [[T86:%.*]] = getelementptr inbounds i8, i8* [[ARG2]], i64 15 1007; CHECK-NEXT: [[T87:%.*]] = load i8, i8* [[T86]], align 1 1008; CHECK-NEXT: [[T88:%.*]] = xor i8 [[T87]], [[T85]] 1009; CHECK-NEXT: store i8 [[T88]], i8* [[T86]], align 1 1010; CHECK-NEXT: ret i32 1 1011; 1012bb: 1013 %t = alloca <16 x i8>, align 16 1014 %t4 = getelementptr inbounds <16 x i8>, <16 x i8>* %t, i64 0, i64 0 1015 %t5 = getelementptr inbounds i8, i8* %arg3, i64 1 1016 %t6 = getelementptr inbounds i8, i8* %arg3, i64 2 1017 %t7 = getelementptr inbounds i8, i8* %arg3, i64 3 1018 %t8 = getelementptr inbounds i8, i8* %arg3, i64 4 1019 %t9 = getelementptr inbounds i8, i8* %arg3, i64 5 1020 %t10 = getelementptr inbounds i8, i8* %arg3, i64 6 1021 %t11 = getelementptr inbounds i8, i8* %arg3, i64 7 1022 %t12 = getelementptr inbounds i8, i8* %arg3, i64 8 1023 %t13 = getelementptr inbounds i8, i8* %arg3, i64 9 1024 %t14 = getelementptr inbounds i8, i8* %arg3, i64 10 1025 %t15 = getelementptr inbounds i8, i8* %arg3, i64 11 1026 %t16 = getelementptr inbounds i8, i8* %arg3, i64 12 1027 %t17 = getelementptr inbounds i8, i8* %arg3, i64 13 1028 %t18 = getelementptr inbounds i8, i8* %arg3, i64 14 1029 %t19 = bitcast i8* %arg1 to <16 x i8>* 1030 %A.gep.0 = getelementptr i8, i8* %A, i64 0 1031 %A.0 = load i8, i8* %A.gep.0 1032 %B.gep.0 = getelementptr i8, i8* %B, i64 0 1033 %B.0 = load i8, i8* %B.gep.0 1034 %xor.0 = xor i8 %A.0, %B.0 1035 %A.gep.1 = getelementptr i8, i8* %A, i64 1 1036 %A.1 = load i8, i8* %A.gep.1 1037 %B.gep.1 = getelementptr i8, i8* %B, i64 1 1038 %B.1 = load i8, i8* %B.gep.1 1039 %xor.1 = xor i8 %A.1, %B.1 1040 %A.gep.2 = getelementptr i8, i8* %A, i64 2 1041 %A.2 = load i8, i8* %A.gep.2 1042 %B.gep.2 = getelementptr i8, i8* %B, i64 2 1043 %B.2 = load i8, i8* %B.gep.2 1044 %xor.2 = xor i8 %A.2, %B.2 1045 %A.gep.3 = getelementptr i8, i8* %A, i64 3 1046 %A.3 = load i8, i8* %A.gep.3 1047 %B.gep.3 = getelementptr i8, i8* %B, i64 3 1048 %B.3 = load i8, i8* %B.gep.3 1049 %xor.3 = xor i8 %A.3, %B.3 1050 %A.gep.4 = getelementptr i8, i8* %A, i64 4 1051 %A.4 = load i8, i8* %A.gep.4 1052 %B.gep.4 = getelementptr i8, i8* %B, i64 4 1053 %B.4 = load i8, i8* %B.gep.4 1054 %xor.4 = xor i8 %A.4, %B.4 1055 %A.gep.5 = getelementptr i8, i8* %A, i64 5 1056 %A.5 = load i8, i8* %A.gep.5 1057 %B.gep.5 = getelementptr i8, i8* %B, i64 5 1058 %B.5 = load i8, i8* %B.gep.5 1059 %xor.5 = xor i8 %A.5, %B.5 1060 %A.gep.6 = getelementptr i8, i8* %A, i64 6 1061 %A.6 = load i8, i8* %A.gep.6 1062 %B.gep.6 = getelementptr i8, i8* %B, i64 6 1063 %B.6 = load i8, i8* %B.gep.6 1064 %xor.6 = xor i8 %A.6, %B.6 1065 %A.gep.7 = getelementptr i8, i8* %A, i64 7 1066 %A.7 = load i8, i8* %A.gep.7 1067 %B.gep.7 = getelementptr i8, i8* %B, i64 7 1068 %B.7 = load i8, i8* %B.gep.7 1069 %xor.7 = xor i8 %A.7, %B.7 1070 %A.gep.8 = getelementptr i8, i8* %A, i64 8 1071 %A.8 = load i8, i8* %A.gep.8 1072 %B.gep.8 = getelementptr i8, i8* %B, i64 8 1073 %B.8 = load i8, i8* %B.gep.8 1074 %xor.8 = xor i8 %A.8, %B.8 1075 %A.gep.9 = getelementptr i8, i8* %A, i64 9 1076 %A.9 = load i8, i8* %A.gep.9 1077 %B.gep.9 = getelementptr i8, i8* %B, i64 9 1078 %B.9 = load i8, i8* %B.gep.9 1079 %xor.9 = xor i8 %A.9, %B.9 1080 %A.gep.10 = getelementptr i8, i8* %A, i64 10 1081 %A.10 = load i8, i8* %A.gep.10 1082 %B.gep.10 = getelementptr i8, i8* %B, i64 10 1083 %B.10 = load i8, i8* %B.gep.10 1084 %xor.10 = xor i8 %A.10, %B.10 1085 %A.gep.11 = getelementptr i8, i8* %A, i64 11 1086 %A.11 = load i8, i8* %A.gep.11 1087 %B.gep.11 = getelementptr i8, i8* %B, i64 11 1088 %B.11 = load i8, i8* %B.gep.11 1089 %xor.11 = xor i8 %A.11, %B.11 1090 %A.gep.12 = getelementptr i8, i8* %A, i64 12 1091 %A.12 = load i8, i8* %A.gep.12 1092 %B.gep.12 = getelementptr i8, i8* %B, i64 12 1093 %B.12 = load i8, i8* %B.gep.12 1094 %xor.12 = xor i8 %A.12, %B.12 1095 %A.gep.13 = getelementptr i8, i8* %A, i64 13 1096 %A.13 = load i8, i8* %A.gep.13 1097 %B.gep.13 = getelementptr i8, i8* %B, i64 13 1098 %B.13 = load i8, i8* %B.gep.13 1099 %xor.13 = xor i8 %A.13, %B.13 1100 %A.gep.14 = getelementptr i8, i8* %A, i64 14 1101 %A.14 = load i8, i8* %A.gep.14 1102 %B.gep.14 = getelementptr i8, i8* %B, i64 14 1103 %B.14 = load i8, i8* %B.gep.14 1104 %xor.14 = xor i8 %A.14, %B.14 1105 %A.gep.15 = getelementptr i8, i8* %A, i64 15 1106 %A.15 = load i8, i8* %A.gep.15 1107 %B.gep.15 = getelementptr i8, i8* %B, i64 15 1108 %B.15 = load i8, i8* %B.gep.15 1109 %xor.15 = xor i8 %A.15, %B.15 1110 %R.gep.0 = getelementptr i8, i8* %arg1, i64 0 1111 store i8 %xor.0, i8* %R.gep.0 1112 %R.gep.1 = getelementptr i8, i8* %arg1, i64 1 1113 store i8 %xor.1, i8* %R.gep.1 1114 %R.gep.2 = getelementptr i8, i8* %arg1, i64 2 1115 store i8 %xor.2, i8* %R.gep.2 1116 %R.gep.3 = getelementptr i8, i8* %arg1, i64 3 1117 store i8 %xor.3, i8* %R.gep.3 1118 %R.gep.4 = getelementptr i8, i8* %arg1, i64 4 1119 store i8 %xor.4, i8* %R.gep.4 1120 %R.gep.5 = getelementptr i8, i8* %arg1, i64 5 1121 store i8 %xor.5, i8* %R.gep.5 1122 %R.gep.6 = getelementptr i8, i8* %arg1, i64 6 1123 store i8 %xor.6, i8* %R.gep.6 1124 %R.gep.7 = getelementptr i8, i8* %arg1, i64 7 1125 store i8 %xor.7, i8* %R.gep.7 1126 %R.gep.8 = getelementptr i8, i8* %arg1, i64 8 1127 store i8 %xor.8, i8* %R.gep.8 1128 %R.gep.9 = getelementptr i8, i8* %arg1, i64 9 1129 store i8 %xor.9, i8* %R.gep.9 1130 %R.gep.10 = getelementptr i8, i8* %arg1, i64 10 1131 store i8 %xor.10, i8* %R.gep.10 1132 %R.gep.11 = getelementptr i8, i8* %arg1, i64 11 1133 store i8 %xor.11, i8* %R.gep.11 1134 %R.gep.12 = getelementptr i8, i8* %arg1, i64 12 1135 store i8 %xor.12, i8* %R.gep.12 1136 %R.gep.13 = getelementptr i8, i8* %arg1, i64 13 1137 store i8 %xor.13, i8* %R.gep.13 1138 %R.gep.14 = getelementptr i8, i8* %arg1, i64 14 1139 store i8 %xor.14, i8* %R.gep.14 1140 %R.gep.15 = getelementptr i8, i8* %arg1, i64 15 1141 store i8 %xor.15, i8* %R.gep.15 1142 1143 1144 %t21 = getelementptr inbounds i8, i8* %arg3, i64 15 1145 %t22 = bitcast i8* %arg3 to <16 x i8>* 1146 1147 call void @foo(i8* nonnull %t4) 1148 %t26 = load i8, i8* %arg3, align 1 1149 %t27 = load i8, i8* %arg2, align 1 1150 %t28 = xor i8 %t27, %t26 1151 store i8 %t28, i8* %arg2, align 1 1152 %t29 = load i8, i8* %t5, align 1 1153 %t30 = getelementptr inbounds i8, i8* %arg2, i64 1 1154 %t31 = load i8, i8* %t30, align 1 1155 %t32 = xor i8 %t31, %t29 1156 store i8 %t32, i8* %t30, align 1 1157 %t33 = load i8, i8* %t6, align 1 1158 %t34 = getelementptr inbounds i8, i8* %arg2, i64 2 1159 %t35 = load i8, i8* %t34, align 1 1160 %t36 = xor i8 %t35, %t33 1161 store i8 %t36, i8* %t34, align 1 1162 %t37 = load i8, i8* %t7, align 1 1163 %t38 = getelementptr inbounds i8, i8* %arg2, i64 3 1164 %t39 = load i8, i8* %t38, align 1 1165 %t40 = xor i8 %t39, %t37 1166 store i8 %t40, i8* %t38, align 1 1167 %t41 = load i8, i8* %t8, align 1 1168 %t42 = getelementptr inbounds i8, i8* %arg2, i64 4 1169 %t43 = load i8, i8* %t42, align 1 1170 %t44 = xor i8 %t43, %t41 1171 store i8 %t44, i8* %t42, align 1 1172 %t45 = load i8, i8* %t9, align 1 1173 %t46 = getelementptr inbounds i8, i8* %arg2, i64 5 1174 %t47 = load i8, i8* %t46, align 1 1175 %t48 = xor i8 %t47, %t45 1176 store i8 %t48, i8* %t46, align 1 1177 %t49 = load i8, i8* %t10, align 1 1178 %t50 = getelementptr inbounds i8, i8* %arg2, i64 6 1179 %t51 = load i8, i8* %t50, align 1 1180 %t52 = xor i8 %t51, %t49 1181 store i8 %t52, i8* %t50, align 1 1182 %t53 = load i8, i8* %t11, align 1 1183 %t54 = getelementptr inbounds i8, i8* %arg2, i64 7 1184 %t55 = load i8, i8* %t54, align 1 1185 %t56 = xor i8 %t55, %t53 1186 store i8 %t56, i8* %t54, align 1 1187 %t57 = load i8, i8* %t12, align 1 1188 %t58 = getelementptr inbounds i8, i8* %arg2, i64 8 1189 %t59 = load i8, i8* %t58, align 1 1190 %t60 = xor i8 %t59, %t57 1191 store i8 %t60, i8* %t58, align 1 1192 %t61 = load i8, i8* %t13, align 1 1193 %t62 = getelementptr inbounds i8, i8* %arg2, i64 9 1194 %t63 = load i8, i8* %t62, align 1 1195 %t64 = xor i8 %t63, %t61 1196 store i8 %t64, i8* %t62, align 1 1197 %t65 = load i8, i8* %t14, align 1 1198 %t66 = getelementptr inbounds i8, i8* %arg2, i64 10 1199 %t67 = load i8, i8* %t66, align 1 1200 %t68 = xor i8 %t67, %t65 1201 store i8 %t68, i8* %t66, align 1 1202 %t69 = load i8, i8* %t15, align 1 1203 %t70 = getelementptr inbounds i8, i8* %arg2, i64 11 1204 %t71 = load i8, i8* %t70, align 1 1205 %t72 = xor i8 %t71, %t69 1206 store i8 %t72, i8* %t70, align 1 1207 %t73 = load i8, i8* %t16, align 1 1208 %t74 = getelementptr inbounds i8, i8* %arg2, i64 12 1209 %t75 = load i8, i8* %t74, align 1 1210 %t76 = xor i8 %t75, %t73 1211 store i8 %t76, i8* %t74, align 1 1212 %t77 = load i8, i8* %t17, align 1 1213 %t78 = getelementptr inbounds i8, i8* %arg2, i64 13 1214 %t79 = load i8, i8* %t78, align 1 1215 %t80 = xor i8 %t79, %t77 1216 store i8 %t80, i8* %t78, align 1 1217 %t81 = load i8, i8* %t18, align 1 1218 %t82 = getelementptr inbounds i8, i8* %arg2, i64 14 1219 %t83 = load i8, i8* %t82, align 1 1220 %t84 = xor i8 %t83, %t81 1221 store i8 %t84, i8* %t82, align 1 1222 %t85 = load i8, i8* %t21, align 1 1223 %t86 = getelementptr inbounds i8, i8* %arg2, i64 15 1224 %t87 = load i8, i8* %t86, align 1 1225 %t88 = xor i8 %t87, %t85 1226 store i8 %t88, i8* %t86, align 1 1227 ret i32 1 1228} 1229 1230; A test case where instructions required to compute the pointer bounds get 1231; vectorized before versioning. Make sure there is no crash. 1232define void @crash_instructions_deleted(float* %t, i32* %a, i32** noalias %ptr) { 1233; CHECK-LABEL: @crash_instructions_deleted( 1234; CHECK-NEXT: bb: 1235; CHECK-NEXT: [[T15:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 2 1236; CHECK-NEXT: [[T16:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 3 1237; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[T15]] to <2 x i32>* 1238; CHECK-NEXT: store <2 x i32> <i32 0, i32 10>, <2 x i32>* [[TMP0]], align 8 1239; CHECK-NEXT: [[T17:%.*]] = load i32*, i32** [[PTR:%.*]], align 8 1240; CHECK-NEXT: br label [[BB18:%.*]] 1241; CHECK: bb18: 1242; CHECK-NEXT: [[T19:%.*]] = sext i32 0 to i64 1243; CHECK-NEXT: [[T20:%.*]] = add nsw i64 1, [[T19]] 1244; CHECK-NEXT: [[T21:%.*]] = getelementptr inbounds i32, i32* [[T17]], i64 [[T20]] 1245; CHECK-NEXT: [[T22:%.*]] = bitcast i32* [[T21]] to i8* 1246; CHECK-NEXT: [[T23:%.*]] = getelementptr inbounds i8, i8* [[T22]], i64 1 1247; CHECK-NEXT: [[T24:%.*]] = getelementptr inbounds i8, i8* [[T22]], i64 2 1248; CHECK-NEXT: [[T25:%.*]] = getelementptr inbounds i8, i8* [[T22]], i64 3 1249; CHECK-NEXT: [[T26:%.*]] = load i8, i8* [[T22]], align 1 1250; CHECK-NEXT: [[T27:%.*]] = uitofp i8 [[T26]] to float 1251; CHECK-NEXT: [[T28:%.*]] = fdiv float [[T27]], 2.550000e+02 1252; CHECK-NEXT: [[T29:%.*]] = getelementptr inbounds float, float* [[T:%.*]], i64 0 1253; CHECK-NEXT: store float [[T28]], float* [[T29]], align 8 1254; CHECK-NEXT: [[T30:%.*]] = load i8, i8* [[T23]], align 1 1255; CHECK-NEXT: [[T31:%.*]] = uitofp i8 [[T30]] to float 1256; CHECK-NEXT: [[T32:%.*]] = fdiv float [[T31]], 2.550000e+02 1257; CHECK-NEXT: [[T33:%.*]] = getelementptr inbounds float, float* [[T]], i64 1 1258; CHECK-NEXT: store float [[T32]], float* [[T33]], align 4 1259; CHECK-NEXT: [[T34:%.*]] = load i8, i8* [[T24]], align 1 1260; CHECK-NEXT: [[T35:%.*]] = uitofp i8 [[T34]] to float 1261; CHECK-NEXT: [[T36:%.*]] = fdiv float [[T35]], 2.550000e+02 1262; CHECK-NEXT: [[T37:%.*]] = getelementptr inbounds float, float* [[T]], i64 2 1263; CHECK-NEXT: store float [[T36]], float* [[T37]], align 8 1264; CHECK-NEXT: [[T38:%.*]] = load i8, i8* [[T25]], align 1 1265; CHECK-NEXT: [[T39:%.*]] = uitofp i8 [[T38]] to float 1266; CHECK-NEXT: [[T40:%.*]] = fdiv float [[T39]], 2.550000e+02 1267; CHECK-NEXT: [[T41:%.*]] = getelementptr inbounds float, float* [[T]], i64 3 1268; CHECK-NEXT: store float [[T40]], float* [[T41]], align 4 1269; CHECK-NEXT: ret void 1270; 1271bb: 1272 %t6 = icmp slt i32 10, 0 1273 %t7 = icmp sgt i32 20, 20 1274 %t9 = select i1 %t7, i32 5, i32 0 1275 %t10 = select i1 %t6, i32 0, i32 %t9 1276 %t11 = icmp slt i32 10, 0 1277 %t12 = icmp sgt i32 20, 20 1278 %t13 = select i1 %t12, i32 5, i32 10 1279 %t14 = select i1 %t11, i32 0, i32 %t13 1280 %t15 = getelementptr inbounds i32, i32* %a, i32 2 1281 store i32 %t10, i32* %t15, align 8 1282 %t16 = getelementptr inbounds i32, i32* %a, i32 3 1283 store i32 %t14, i32* %t16, align 4 1284 %t17 = load i32*, i32** %ptr, align 8 1285 br label %bb18 1286 1287bb18: ; preds = %bb5 1288 %t19 = sext i32 %t10 to i64 1289 %t20 = add nsw i64 1, %t19 1290 %t21 = getelementptr inbounds i32, i32* %t17, i64 %t20 1291 %t22 = bitcast i32* %t21 to i8* 1292 %t23 = getelementptr inbounds i8, i8* %t22, i64 1 1293 %t24 = getelementptr inbounds i8, i8* %t22, i64 2 1294 %t25 = getelementptr inbounds i8, i8* %t22, i64 3 1295 %t26 = load i8, i8* %t22, align 1 1296 %t27 = uitofp i8 %t26 to float 1297 %t28 = fdiv float %t27, 2.550000e+02 1298 %t29 = getelementptr inbounds float, float* %t, i64 0 1299 store float %t28, float* %t29, align 8 1300 %t30 = load i8, i8* %t23, align 1 1301 %t31 = uitofp i8 %t30 to float 1302 %t32 = fdiv float %t31, 2.550000e+02 1303 %t33 = getelementptr inbounds float, float* %t, i64 1 1304 store float %t32, float* %t33, align 4 1305 %t34 = load i8, i8* %t24, align 1 1306 %t35 = uitofp i8 %t34 to float 1307 %t36 = fdiv float %t35, 2.550000e+02 1308 %t37 = getelementptr inbounds float, float* %t, i64 2 1309 store float %t36, float* %t37, align 8 1310 %t38 = load i8, i8* %t25, align 1 1311 %t39 = uitofp i8 %t38 to float 1312 %t40 = fdiv float %t39, 2.550000e+02 1313 %t41 = getelementptr inbounds float, float* %t, i64 3 1314 store float %t40, float* %t41, align 4 1315 ret void 1316} 1317 1318; A test case where there are no instructions accessing a tracked object in a 1319; block for which versioning was requested. 1320define void @crash_no_tracked_instructions(float** %arg, float* %arg.2, float* %arg.3, i1 %c) { 1321; CHECK-LABEL: @crash_no_tracked_instructions( 1322; CHECK-NEXT: entry: 1323; CHECK-NEXT: [[T19:%.*]] = load float*, float** [[ARG:%.*]], align 8 1324; CHECK-NEXT: [[T20:%.*]] = load float, float* [[ARG_3:%.*]], align 4 1325; CHECK-NEXT: [[T21:%.*]] = getelementptr inbounds float, float* [[ARG_2:%.*]], i64 0 1326; CHECK-NEXT: br i1 [[C:%.*]], label [[BB22:%.*]], label [[BB30:%.*]] 1327; CHECK: bb22: 1328; CHECK-NEXT: [[T23:%.*]] = fmul float [[T20]], 9.900000e+01 1329; CHECK-NEXT: [[T24:%.*]] = fmul float [[T23]], 9.900000e+01 1330; CHECK-NEXT: [[T25:%.*]] = getelementptr inbounds float, float* [[T19]], i64 2 1331; CHECK-NEXT: [[T26:%.*]] = fmul float [[T23]], 1.000000e+01 1332; CHECK-NEXT: store float [[T26]], float* [[T25]], align 4 1333; CHECK-NEXT: [[T27:%.*]] = load float, float* [[T21]], align 8 1334; CHECK-NEXT: [[T28:%.*]] = fadd float [[T24]], 2.000000e+01 1335; CHECK-NEXT: [[T29:%.*]] = fadd float [[T26]], 2.000000e+01 1336; CHECK-NEXT: br label [[BB30]] 1337; CHECK: bb30: 1338; CHECK-NEXT: [[T31:%.*]] = phi float [ [[T28]], [[BB22]] ], [ 0.000000e+00, [[ENTRY:%.*]] ] 1339; CHECK-NEXT: [[T32:%.*]] = phi float [ [[T29]], [[BB22]] ], [ [[T20]], [[ENTRY]] ] 1340; CHECK-NEXT: br label [[BB36:%.*]] 1341; CHECK: bb36: 1342; CHECK-NEXT: [[T37:%.*]] = fmul float [[T31]], 3.000000e+00 1343; CHECK-NEXT: [[T38:%.*]] = getelementptr inbounds float, float* [[ARG_3]], i64 0 1344; CHECK-NEXT: store float [[T37]], float* [[T38]], align 4 1345; CHECK-NEXT: [[T39:%.*]] = fmul float [[T32]], 3.000000e+00 1346; CHECK-NEXT: [[T40:%.*]] = getelementptr inbounds float, float* [[ARG_3]], i64 1 1347; CHECK-NEXT: store float [[T39]], float* [[T40]], align 4 1348; CHECK-NEXT: br label [[BB41:%.*]] 1349; CHECK: bb41: 1350; CHECK-NEXT: ret void 1351; 1352entry: 1353 %t19 = load float*, float** %arg 1354 %t20 = load float, float* %arg.3, align 4 1355 %t21 = getelementptr inbounds float, float* %arg.2, i64 0 1356 br i1 %c, label %bb22, label %bb30 1357 1358bb22: 1359 %t23 = fmul float %t20, 99.0 1360 %t24 = fmul float %t23, 99.0 1361 %t25 = getelementptr inbounds float, float* %t19, i64 2 1362 %t26 = fmul float %t23, 10.0 1363 store float %t26, float* %t25, align 4 1364 %t27 = load float, float* %t21, align 8 1365 %t28 = fadd float %t24, 20.0 1366 %t29 = fadd float %t26, 20.0 1367 br label %bb30 1368 1369bb30: 1370 %t31 = phi float [ %t28, %bb22 ], [ 0.0, %entry ] 1371 %t32 = phi float [ %t29, %bb22 ], [ %t20, %entry ] 1372 br label %bb36 1373 1374bb36: 1375 %t37 = fmul float %t31, 3.0 1376 %t38 = getelementptr inbounds float, float* %arg.3, i64 0 1377 store float %t37, float* %t38, align 4 1378 %t39 = fmul float %t32, 3.0 1379 %t40 = getelementptr inbounds float, float* %arg.3, i64 1 1380 store float %t39, float* %t40, align 4 1381 br label %bb41 1382 1383bb41: 1384 ret void 1385} 1386