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