1; RUN: opt < %s -loop-reroll -S | FileCheck %s 2target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128" 3target triple = "x86_64-unknown-linux-gnu" 4 5; int foo(int a); 6; void bar(int *x) { 7; for (int i = 0; i < 500; i += 3) { 8; foo(i); 9; foo(i+1); 10; foo(i+2); 11; } 12; } 13 14; Function Attrs: nounwind uwtable 15define void @bar(i32* nocapture readnone %x) #0 { 16entry: 17 br label %for.body 18 19for.body: ; preds = %for.body, %entry 20 %i.08 = phi i32 [ 0, %entry ], [ %add3, %for.body ] 21 %call = tail call i32 @foo(i32 %i.08) #1 22 %add = add nsw i32 %i.08, 1 23 %call1 = tail call i32 @foo(i32 %add) #1 24 %add2 = add nsw i32 %i.08, 2 25 %call3 = tail call i32 @foo(i32 %add2) #1 26 %add3 = add nsw i32 %i.08, 3 27 %exitcond = icmp sge i32 %add3, 500 28 br i1 %exitcond, label %for.end, label %for.body 29 30; CHECK-LABEL: @bar 31 32; CHECK: for.body: 33; CHECK: %indvar = phi i32 [ %indvar.next, %for.body ], [ 0, %entry ] 34; CHECK: %call = tail call i32 @foo(i32 %indvar) #1 35; CHECK: %indvar.next = add i32 %indvar, 1 36; CHECK: %exitcond1 = icmp eq i32 %indvar, 500 37; CHECK: br i1 %exitcond1, label %for.end, label %for.body 38 39; CHECK: ret 40 41for.end: ; preds = %for.body 42 ret void 43} 44 45declare i32 @foo(i32) 46 47; void hi1(int *x) { 48; for (int i = 0; i < 1500; i += 3) { 49; x[i] = foo(0); 50; x[i+1] = foo(0); 51; x[i+2] = foo(0); 52; } 53; } 54 55; Function Attrs: nounwind uwtable 56define void @hi1(i32* nocapture %x) #0 { 57entry: 58 br label %for.body 59 60for.body: ; preds = %entry, %for.body 61 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 62 %call = tail call i32 @foo(i32 0) #1 63 %arrayidx = getelementptr inbounds i32, i32* %x, i64 %indvars.iv 64 store i32 %call, i32* %arrayidx, align 4 65 %call1 = tail call i32 @foo(i32 0) #1 66 %0 = add nsw i64 %indvars.iv, 1 67 %arrayidx3 = getelementptr inbounds i32, i32* %x, i64 %0 68 store i32 %call1, i32* %arrayidx3, align 4 69 %call4 = tail call i32 @foo(i32 0) #1 70 %1 = add nsw i64 %indvars.iv, 2 71 %arrayidx7 = getelementptr inbounds i32, i32* %x, i64 %1 72 store i32 %call4, i32* %arrayidx7, align 4 73 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 3 74 %2 = trunc i64 %indvars.iv.next to i32 75 %cmp = icmp slt i32 %2, 1500 76 br i1 %cmp, label %for.body, label %for.end 77 78; CHECK-LABEL: @hi1 79 80; CHECK: for.body: 81; CHECK: %indvar = phi i64 [ %indvar.next, %for.body ], [ 0, %entry ] 82; CHECK: %0 = trunc i64 %indvar to i32 83; CHECK: %call = tail call i32 @foo(i32 0) #1 84; CHECK: %arrayidx = getelementptr inbounds i32, i32* %x, i64 %indvar 85; CHECK: store i32 %call, i32* %arrayidx, align 4 86; CHECK: %indvar.next = add i64 %indvar, 1 87; CHECK: %exitcond = icmp eq i32 %0, 1499 88; CHECK: br i1 %exitcond, label %for.end, label %for.body 89 90; CHECK: ret 91 92for.end: ; preds = %for.body 93 ret void 94} 95 96; void hi2(int *x) { 97; for (int i = 0; i < 500; ++i) { 98; x[3*i] = foo(0); 99; x[3*i+1] = foo(0); 100; x[3*i+2] = foo(0); 101; } 102; } 103 104; Function Attrs: nounwind uwtable 105define void @hi2(i32* nocapture %x) #0 { 106entry: 107 br label %for.body 108 109for.body: ; preds = %for.body, %entry 110 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 111 %call = tail call i32 @foo(i32 0) #1 112 %0 = mul nsw i64 %indvars.iv, 3 113 %arrayidx = getelementptr inbounds i32, i32* %x, i64 %0 114 store i32 %call, i32* %arrayidx, align 4 115 %call1 = tail call i32 @foo(i32 0) #1 116 %1 = add nsw i64 %0, 1 117 %arrayidx4 = getelementptr inbounds i32, i32* %x, i64 %1 118 store i32 %call1, i32* %arrayidx4, align 4 119 %call5 = tail call i32 @foo(i32 0) #1 120 %2 = add nsw i64 %0, 2 121 %arrayidx9 = getelementptr inbounds i32, i32* %x, i64 %2 122 store i32 %call5, i32* %arrayidx9, align 4 123 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 124 %exitcond = icmp eq i64 %indvars.iv.next, 500 125 br i1 %exitcond, label %for.end, label %for.body 126 127; CHECK-LABEL: @hi2 128 129; CHECK: for.body: 130; CHECK: %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 131; CHECK: %call = tail call i32 @foo(i32 0) #1 132; CHECK: %arrayidx = getelementptr inbounds i32, i32* %x, i64 %indvars.iv 133; CHECK: store i32 %call, i32* %arrayidx, align 4 134; CHECK: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 135; CHECK: %exitcond1 = icmp eq i64 %indvars.iv, 1499 136; CHECK: br i1 %exitcond1, label %for.end, label %for.body 137 138; CHECK: ret 139 140for.end: ; preds = %for.body 141 ret void 142} 143 144; void goo(float alpha, float *a, float *b) { 145; for (int i = 0; i < 3200; i += 5) { 146; a[i] += alpha * b[i]; 147; a[i + 1] += alpha * b[i + 1]; 148; a[i + 2] += alpha * b[i + 2]; 149; a[i + 3] += alpha * b[i + 3]; 150; a[i + 4] += alpha * b[i + 4]; 151; } 152; } 153 154; Function Attrs: nounwind uwtable 155define void @goo(float %alpha, float* nocapture %a, float* nocapture readonly %b) #0 { 156entry: 157 br label %for.body 158 159for.body: ; preds = %entry, %for.body 160 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 161 %arrayidx = getelementptr inbounds float, float* %b, i64 %indvars.iv 162 %0 = load float, float* %arrayidx, align 4 163 %mul = fmul float %0, %alpha 164 %arrayidx2 = getelementptr inbounds float, float* %a, i64 %indvars.iv 165 %1 = load float, float* %arrayidx2, align 4 166 %add = fadd float %1, %mul 167 store float %add, float* %arrayidx2, align 4 168 %2 = add nsw i64 %indvars.iv, 1 169 %arrayidx5 = getelementptr inbounds float, float* %b, i64 %2 170 %3 = load float, float* %arrayidx5, align 4 171 %mul6 = fmul float %3, %alpha 172 %arrayidx9 = getelementptr inbounds float, float* %a, i64 %2 173 %4 = load float, float* %arrayidx9, align 4 174 %add10 = fadd float %4, %mul6 175 store float %add10, float* %arrayidx9, align 4 176 %5 = add nsw i64 %indvars.iv, 2 177 %arrayidx13 = getelementptr inbounds float, float* %b, i64 %5 178 %6 = load float, float* %arrayidx13, align 4 179 %mul14 = fmul float %6, %alpha 180 %arrayidx17 = getelementptr inbounds float, float* %a, i64 %5 181 %7 = load float, float* %arrayidx17, align 4 182 %add18 = fadd float %7, %mul14 183 store float %add18, float* %arrayidx17, align 4 184 %8 = add nsw i64 %indvars.iv, 3 185 %arrayidx21 = getelementptr inbounds float, float* %b, i64 %8 186 %9 = load float, float* %arrayidx21, align 4 187 %mul22 = fmul float %9, %alpha 188 %arrayidx25 = getelementptr inbounds float, float* %a, i64 %8 189 %10 = load float, float* %arrayidx25, align 4 190 %add26 = fadd float %10, %mul22 191 store float %add26, float* %arrayidx25, align 4 192 %11 = add nsw i64 %indvars.iv, 4 193 %arrayidx29 = getelementptr inbounds float, float* %b, i64 %11 194 %12 = load float, float* %arrayidx29, align 4 195 %mul30 = fmul float %12, %alpha 196 %arrayidx33 = getelementptr inbounds float, float* %a, i64 %11 197 %13 = load float, float* %arrayidx33, align 4 198 %add34 = fadd float %13, %mul30 199 store float %add34, float* %arrayidx33, align 4 200 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 5 201 %14 = trunc i64 %indvars.iv.next to i32 202 %cmp = icmp slt i32 %14, 3200 203 br i1 %cmp, label %for.body, label %for.end 204 205; CHECK-LABEL: @goo 206 207; CHECK: for.body: 208; CHECK: %indvar = phi i64 [ %indvar.next, %for.body ], [ 0, %entry ] 209; CHECK: %0 = trunc i64 %indvar to i32 210; CHECK: %arrayidx = getelementptr inbounds float, float* %b, i64 %indvar 211; CHECK: %1 = load float, float* %arrayidx, align 4 212; CHECK: %mul = fmul float %1, %alpha 213; CHECK: %arrayidx2 = getelementptr inbounds float, float* %a, i64 %indvar 214; CHECK: %2 = load float, float* %arrayidx2, align 4 215; CHECK: %add = fadd float %2, %mul 216; CHECK: store float %add, float* %arrayidx2, align 4 217; CHECK: %indvar.next = add i64 %indvar, 1 218; CHECK: %exitcond = icmp eq i32 %0, 3199 219; CHECK: br i1 %exitcond, label %for.end, label %for.body 220 221; CHECK: ret 222 223for.end: ; preds = %for.body 224 ret void 225} 226 227; void hoo(float alpha, float *a, float *b, int *ip) { 228; for (int i = 0; i < 3200; i += 5) { 229; a[i] += alpha * b[ip[i]]; 230; a[i + 1] += alpha * b[ip[i + 1]]; 231; a[i + 2] += alpha * b[ip[i + 2]]; 232; a[i + 3] += alpha * b[ip[i + 3]]; 233; a[i + 4] += alpha * b[ip[i + 4]]; 234; } 235; } 236 237; Function Attrs: nounwind uwtable 238define void @hoo(float %alpha, float* nocapture %a, float* nocapture readonly %b, i32* nocapture readonly %ip) #0 { 239entry: 240 br label %for.body 241 242for.body: ; preds = %entry, %for.body 243 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 244 %arrayidx = getelementptr inbounds i32, i32* %ip, i64 %indvars.iv 245 %0 = load i32, i32* %arrayidx, align 4 246 %idxprom1 = sext i32 %0 to i64 247 %arrayidx2 = getelementptr inbounds float, float* %b, i64 %idxprom1 248 %1 = load float, float* %arrayidx2, align 4 249 %mul = fmul float %1, %alpha 250 %arrayidx4 = getelementptr inbounds float, float* %a, i64 %indvars.iv 251 %2 = load float, float* %arrayidx4, align 4 252 %add = fadd float %2, %mul 253 store float %add, float* %arrayidx4, align 4 254 %3 = add nsw i64 %indvars.iv, 1 255 %arrayidx7 = getelementptr inbounds i32, i32* %ip, i64 %3 256 %4 = load i32, i32* %arrayidx7, align 4 257 %idxprom8 = sext i32 %4 to i64 258 %arrayidx9 = getelementptr inbounds float, float* %b, i64 %idxprom8 259 %5 = load float, float* %arrayidx9, align 4 260 %mul10 = fmul float %5, %alpha 261 %arrayidx13 = getelementptr inbounds float, float* %a, i64 %3 262 %6 = load float, float* %arrayidx13, align 4 263 %add14 = fadd float %6, %mul10 264 store float %add14, float* %arrayidx13, align 4 265 %7 = add nsw i64 %indvars.iv, 2 266 %arrayidx17 = getelementptr inbounds i32, i32* %ip, i64 %7 267 %8 = load i32, i32* %arrayidx17, align 4 268 %idxprom18 = sext i32 %8 to i64 269 %arrayidx19 = getelementptr inbounds float, float* %b, i64 %idxprom18 270 %9 = load float, float* %arrayidx19, align 4 271 %mul20 = fmul float %9, %alpha 272 %arrayidx23 = getelementptr inbounds float, float* %a, i64 %7 273 %10 = load float, float* %arrayidx23, align 4 274 %add24 = fadd float %10, %mul20 275 store float %add24, float* %arrayidx23, align 4 276 %11 = add nsw i64 %indvars.iv, 3 277 %arrayidx27 = getelementptr inbounds i32, i32* %ip, i64 %11 278 %12 = load i32, i32* %arrayidx27, align 4 279 %idxprom28 = sext i32 %12 to i64 280 %arrayidx29 = getelementptr inbounds float, float* %b, i64 %idxprom28 281 %13 = load float, float* %arrayidx29, align 4 282 %mul30 = fmul float %13, %alpha 283 %arrayidx33 = getelementptr inbounds float, float* %a, i64 %11 284 %14 = load float, float* %arrayidx33, align 4 285 %add34 = fadd float %14, %mul30 286 store float %add34, float* %arrayidx33, align 4 287 %15 = add nsw i64 %indvars.iv, 4 288 %arrayidx37 = getelementptr inbounds i32, i32* %ip, i64 %15 289 %16 = load i32, i32* %arrayidx37, align 4 290 %idxprom38 = sext i32 %16 to i64 291 %arrayidx39 = getelementptr inbounds float, float* %b, i64 %idxprom38 292 %17 = load float, float* %arrayidx39, align 4 293 %mul40 = fmul float %17, %alpha 294 %arrayidx43 = getelementptr inbounds float, float* %a, i64 %15 295 %18 = load float, float* %arrayidx43, align 4 296 %add44 = fadd float %18, %mul40 297 store float %add44, float* %arrayidx43, align 4 298 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 5 299 %19 = trunc i64 %indvars.iv.next to i32 300 %cmp = icmp slt i32 %19, 3200 301 br i1 %cmp, label %for.body, label %for.end 302 303; CHECK-LABEL: @hoo 304 305; CHECK: for.body: 306; CHECK: %indvar = phi i64 [ %indvar.next, %for.body ], [ 0, %entry ] 307; CHECK: %0 = trunc i64 %indvar to i32 308; CHECK: %arrayidx = getelementptr inbounds i32, i32* %ip, i64 %indvar 309; CHECK: %1 = load i32, i32* %arrayidx, align 4 310; CHECK: %idxprom1 = sext i32 %1 to i64 311; CHECK: %arrayidx2 = getelementptr inbounds float, float* %b, i64 %idxprom1 312; CHECK: %2 = load float, float* %arrayidx2, align 4 313; CHECK: %mul = fmul float %2, %alpha 314; CHECK: %arrayidx4 = getelementptr inbounds float, float* %a, i64 %indvar 315; CHECK: %3 = load float, float* %arrayidx4, align 4 316; CHECK: %add = fadd float %3, %mul 317; CHECK: store float %add, float* %arrayidx4, align 4 318; CHECK: %indvar.next = add i64 %indvar, 1 319; CHECK: %exitcond = icmp eq i32 %0, 3199 320; CHECK: br i1 %exitcond, label %for.end, label %for.body 321 322; CHECK: ret 323 324for.end: ; preds = %for.body 325 ret void 326} 327 328; void multi1(int *x) { 329; y = foo(0) 330; for (int i = 0; i < 500; ++i) { 331; x[3*i] = y; 332; x[3*i+1] = y; 333; x[3*i+2] = y; 334; x[3*i+6] = y; 335; x[3*i+7] = y; 336; x[3*i+8] = y; 337; } 338; } 339 340; Function Attrs: nounwind uwtable 341define void @multi1(i32* nocapture %x) #0 { 342entry: 343 %call = tail call i32 @foo(i32 0) #1 344 br label %for.body 345 346for.body: ; preds = %for.body, %entry 347 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 348 %0 = mul nsw i64 %indvars.iv, 3 349 %arrayidx = getelementptr inbounds i32, i32* %x, i64 %0 350 store i32 %call, i32* %arrayidx, align 4 351 %1 = add nsw i64 %0, 1 352 %arrayidx4 = getelementptr inbounds i32, i32* %x, i64 %1 353 store i32 %call, i32* %arrayidx4, align 4 354 %2 = add nsw i64 %0, 2 355 %arrayidx9 = getelementptr inbounds i32, i32* %x, i64 %2 356 store i32 %call, i32* %arrayidx9, align 4 357 %3 = add nsw i64 %0, 6 358 %arrayidx6 = getelementptr inbounds i32, i32* %x, i64 %3 359 store i32 %call, i32* %arrayidx6, align 4 360 %4 = add nsw i64 %0, 7 361 %arrayidx7 = getelementptr inbounds i32, i32* %x, i64 %4 362 store i32 %call, i32* %arrayidx7, align 4 363 %5 = add nsw i64 %0, 8 364 %arrayidx8 = getelementptr inbounds i32, i32* %x, i64 %5 365 store i32 %call, i32* %arrayidx8, align 4 366 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 367 %exitcond = icmp eq i64 %indvars.iv.next, 500 368 br i1 %exitcond, label %for.end, label %for.body 369 370; CHECK-LABEL: @multi1 371 372; CHECK:for.body: 373; CHECK: %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 374; CHECK: %0 = add i64 %indvars.iv, 6 375; CHECK: %arrayidx = getelementptr inbounds i32, i32* %x, i64 %indvars.iv 376; CHECK: store i32 %call, i32* %arrayidx, align 4 377; CHECK: %arrayidx6 = getelementptr inbounds i32, i32* %x, i64 %0 378; CHECK: store i32 %call, i32* %arrayidx6, align 4 379; CHECK: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 380; CHECK: %exitcond1 = icmp eq i64 %indvars.iv, 1499 381; CHECK: br i1 %exitcond1, label %for.end, label %for.body 382 383for.end: ; preds = %for.body 384 ret void 385} 386 387; void multi2(int *x) { 388; y = foo(0) 389; for (int i = 0; i < 500; ++i) { 390; x[3*i] = y; 391; x[3*i+1] = y; 392; x[3*i+2] = y; 393; x[3*(i+1)] = y; 394; x[3*(i+1)+1] = y; 395; x[3*(i+1)+2] = y; 396; } 397; } 398 399; Function Attrs: nounwind uwtable 400define void @multi2(i32* nocapture %x) #0 { 401entry: 402 %call = tail call i32 @foo(i32 0) #1 403 br label %for.body 404 405for.body: ; preds = %for.body, %entry 406 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 407 %0 = mul nsw i64 %indvars.iv, 3 408 %add = add nsw i64 %indvars.iv, 1 409 %newmul = mul nsw i64 %add, 3 410 %arrayidx = getelementptr inbounds i32, i32* %x, i64 %0 411 store i32 %call, i32* %arrayidx, align 4 412 %1 = add nsw i64 %0, 1 413 %arrayidx4 = getelementptr inbounds i32, i32* %x, i64 %1 414 store i32 %call, i32* %arrayidx4, align 4 415 %2 = add nsw i64 %0, 2 416 %arrayidx9 = getelementptr inbounds i32, i32* %x, i64 %2 417 store i32 %call, i32* %arrayidx9, align 4 418 %arrayidx6 = getelementptr inbounds i32, i32* %x, i64 %newmul 419 store i32 %call, i32* %arrayidx6, align 4 420 %3 = add nsw i64 %newmul, 1 421 %arrayidx7 = getelementptr inbounds i32, i32* %x, i64 %3 422 store i32 %call, i32* %arrayidx7, align 4 423 %4 = add nsw i64 %newmul, 2 424 %arrayidx8 = getelementptr inbounds i32, i32* %x, i64 %4 425 store i32 %call, i32* %arrayidx8, align 4 426 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 427 %exitcond = icmp eq i64 %indvars.iv.next, 500 428 br i1 %exitcond, label %for.end, label %for.body 429 430; CHECK-LABEL: @multi2 431 432; CHECK:for.body: 433; CHECK: %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 434; CHECK: %0 = add i64 %indvars.iv, 3 435; CHECK: %arrayidx = getelementptr inbounds i32, i32* %x, i64 %indvars.iv 436; CHECK: store i32 %call, i32* %arrayidx, align 4 437; CHECK: %arrayidx6 = getelementptr inbounds i32, i32* %x, i64 %0 438; CHECK: store i32 %call, i32* %arrayidx6, align 4 439; CHECK: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 440; CHECK: %exitcond1 = icmp eq i64 %indvars.iv, 1499 441; CHECK: br i1 %exitcond1, label %for.end, label %for.body 442 443for.end: ; preds = %for.body 444 ret void 445} 446 447; void multi3(int *x) { 448; y = foo(0) 449; for (int i = 0; i < 500; ++i) { 450; // Note: No zero index 451; x[3*i+3] = y; 452; x[3*i+4] = y; 453; x[3*i+5] = y; 454; } 455; } 456 457; Function Attrs: nounwind uwtable 458define void @multi3(i32* nocapture %x) #0 { 459entry: 460 %call = tail call i32 @foo(i32 0) #1 461 br label %for.body 462 463for.body: ; preds = %for.body, %entry 464 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 465 %0 = mul nsw i64 %indvars.iv, 3 466 %x0 = add nsw i64 %0, 3 467 %add = add nsw i64 %indvars.iv, 1 468 %arrayidx = getelementptr inbounds i32, i32* %x, i64 %x0 469 store i32 %call, i32* %arrayidx, align 4 470 %1 = add nsw i64 %0, 4 471 %arrayidx4 = getelementptr inbounds i32, i32* %x, i64 %1 472 store i32 %call, i32* %arrayidx4, align 4 473 %2 = add nsw i64 %0, 5 474 %arrayidx9 = getelementptr inbounds i32, i32* %x, i64 %2 475 store i32 %call, i32* %arrayidx9, align 4 476 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 477 %exitcond = icmp eq i64 %indvars.iv.next, 500 478 br i1 %exitcond, label %for.end, label %for.body 479 480; CHECK-LABEL: @multi3 481; CHECK: for.body: 482; CHECK: %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 483; CHECK: %0 = add i64 %indvars.iv, 3 484; CHECK: %arrayidx = getelementptr inbounds i32, i32* %x, i64 %0 485; CHECK: store i32 %call, i32* %arrayidx, align 4 486; CHECK: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 487; CHECK: %exitcond1 = icmp eq i64 %indvars.iv, 1499 488; CHECK: br i1 %exitcond1, label %for.end, label %for.body 489 490for.end: ; preds = %for.body 491 ret void 492} 493 494; int foo(int a); 495; void bar2(int *x, int y, int z) { 496; for (int i = 0; i < 500; i += 3) { 497; foo(i+y+i*z); // Slightly reordered instruction order 498; foo(i+1+y+(i+1)*z); 499; foo(i+2+y+(i+2)*z); 500; } 501; } 502 503; Function Attrs: nounwind uwtable 504define void @bar2(i32* nocapture readnone %x, i32 %y, i32 %z) #0 { 505entry: 506 br label %for.body 507 508for.body: ; preds = %for.body, %entry 509 %i.08 = phi i32 [ 0, %entry ], [ %add3, %for.body ] 510 511 %tmp1 = add i32 %i.08, %y 512 %tmp2 = mul i32 %i.08, %z 513 %tmp3 = add i32 %tmp2, %tmp1 514 %call = tail call i32 @foo(i32 %tmp3) #1 515 516 %add = add nsw i32 %i.08, 1 517 %tmp2a = mul i32 %add, %z 518 %tmp1a = add i32 %add, %y 519 %tmp3a = add i32 %tmp2a, %tmp1a 520 %calla = tail call i32 @foo(i32 %tmp3a) #1 521 522 %add2 = add nsw i32 %i.08, 2 523 %tmp2b = mul i32 %add2, %z 524 %tmp1b = add i32 %add2, %y 525 %tmp3b = add i32 %tmp2b, %tmp1b 526 %callb = tail call i32 @foo(i32 %tmp3b) #1 527 528 %add3 = add nsw i32 %i.08, 3 529 530 %exitcond = icmp sge i32 %add3, 500 531 br i1 %exitcond, label %for.end, label %for.body 532 533; CHECK-LABEL: @bar2 534 535; CHECK: for.body: 536; CHECK: %indvar = phi i32 [ %indvar.next, %for.body ], [ 0, %entry ] 537; CHECK: %tmp1 = add i32 %indvar, %y 538; CHECK: %tmp2 = mul i32 %indvar, %z 539; CHECK: %tmp3 = add i32 %tmp2, %tmp1 540; CHECK: %call = tail call i32 @foo(i32 %tmp3) #1 541; CHECK: %indvar.next = add i32 %indvar, 1 542; CHECK: %exitcond1 = icmp eq i32 %indvar, 500 543; CHECK: br i1 %exitcond1, label %for.end, label %for.body 544 545; CHECK: ret 546 547for.end: ; preds = %for.body 548 ret void 549} 550 551%struct.s = type { i32, i32 } 552 553; Function Attrs: nounwind uwtable 554define void @gep1(%struct.s* nocapture %x) #0 { 555entry: 556 %call = tail call i32 @foo(i32 0) #1 557 br label %for.body 558 559for.body: ; preds = %for.body, %entry 560 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 561 %0 = mul nsw i64 %indvars.iv, 3 562 %arrayidx = getelementptr inbounds %struct.s, %struct.s* %x, i64 %0, i32 0 563 store i32 %call, i32* %arrayidx, align 4 564 %1 = add nsw i64 %0, 1 565 %arrayidx4 = getelementptr inbounds %struct.s, %struct.s* %x, i64 %1, i32 0 566 store i32 %call, i32* %arrayidx4, align 4 567 %2 = add nsw i64 %0, 2 568 %arrayidx9 = getelementptr inbounds %struct.s, %struct.s* %x, i64 %2, i32 0 569 store i32 %call, i32* %arrayidx9, align 4 570 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 571 %exitcond = icmp eq i64 %indvars.iv.next, 500 572 br i1 %exitcond, label %for.end, label %for.body 573 574; CHECK-LABEL: @gep1 575; This test is a crash test only. 576; CHECK: ret 577for.end: ; preds = %for.body 578 ret void 579} 580 581define void @gep-indexing(i32* nocapture %x) { 582entry: 583 %call = tail call i32 @foo(i32 0) #1 584 br label %for.body 585 586for.body: ; preds = %for.body, %entry 587 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 588 %0 = mul nsw i64 %indvars.iv, 3 589 %arrayidx = getelementptr inbounds i32, i32* %x, i64 %0 590 store i32 %call, i32* %arrayidx, align 4 591 %arrayidx4 = getelementptr inbounds i32, i32* %arrayidx, i64 1 592 store i32 %call, i32* %arrayidx4, align 4 593 %arrayidx9 = getelementptr inbounds i32, i32* %arrayidx, i64 2 594 store i32 %call, i32* %arrayidx9, align 4 595 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 596 %exitcond = icmp eq i64 %indvars.iv.next, 500 597 br i1 %exitcond, label %for.end, label %for.body 598 599; CHECK-LABEL: @gep-indexing 600; CHECK: for.body: 601; CHECK-NEXT: %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 602; CHECK-NEXT: %scevgep = getelementptr i32, i32* %x, i64 %indvars.iv 603; CHECK-NEXT: store i32 %call, i32* %scevgep, align 4 604; CHECK-NEXT: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 605; CHECK-NEXT: %exitcond1 = icmp eq i64 %indvars.iv, 1499 606; CHECK-NEXT: br i1 %exitcond1, label %for.end, label %for.body 607 608for.end: ; preds = %for.body 609 ret void 610} 611 612 613define void @unordered_atomic_ops(i32* noalias %buf_0, i32* noalias %buf_1) { 614; CHECK-LABEL: @unordered_atomic_ops( 615 616; CHECK: for.body: 617; CHECK-NEXT: %indvar = phi i32 [ %indvar.next, %for.body ], [ 0, %entry ] 618; CHECK-NEXT: %buf0_a = getelementptr i32, i32* %buf_0, i32 %indvar 619; CHECK-NEXT: %buf1_a = getelementptr i32, i32* %buf_1, i32 %indvar 620; CHECK-NEXT: %va = load atomic i32, i32* %buf0_a unordered, align 4 621; CHECK-NEXT: store atomic i32 %va, i32* %buf1_a unordered, align 4 622; CHECK-NEXT: %indvar.next = add i32 %indvar, 1 623; CHECK-NEXT: %exitcond = icmp eq i32 %indvar, 3199 624; CHECK-NEXT: br i1 %exitcond, label %for.end, label %for.body 625 626entry: 627 br label %for.body 628 629for.body: 630 %indvars.iv = phi i32 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 631 %indvars.iv.next = add i32 %indvars.iv, 2 632 %indvars.mid = add i32 %indvars.iv, 1 633 %buf0_a = getelementptr i32, i32* %buf_0, i32 %indvars.iv 634 %buf0_b = getelementptr i32, i32* %buf_0, i32 %indvars.mid 635 %buf1_a = getelementptr i32, i32* %buf_1, i32 %indvars.iv 636 %buf1_b = getelementptr i32, i32* %buf_1, i32 %indvars.mid 637 %va = load atomic i32, i32* %buf0_a unordered, align 4 638 %vb = load atomic i32, i32* %buf0_b unordered, align 4 639 store atomic i32 %va, i32* %buf1_a unordered, align 4 640 store atomic i32 %vb, i32* %buf1_b unordered, align 4 641 %cmp = icmp slt i32 %indvars.iv.next, 3200 642 br i1 %cmp, label %for.body, label %for.end 643 644for.end: 645 ret void 646} 647 648define void @unordered_atomic_ops_nomatch(i32* noalias %buf_0, i32* noalias %buf_1) { 649; Negative test 650 651; CHECK-LABEL: @unordered_atomic_ops_nomatch( 652entry: 653 br label %for.body 654 655for.body: 656; CHECK: for.body: 657; CHECK: %indvars.iv.next = add i32 %indvars.iv, 2 658; CHECK: %indvars.mid = add i32 %indvars.iv, 1 659; CHECK: %cmp = icmp slt i32 %indvars.iv.next, 3200 660; CHECK: br i1 %cmp, label %for.body, label %for.end 661 662 %indvars.iv = phi i32 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 663 %indvars.iv.next = add i32 %indvars.iv, 2 664 %indvars.mid = add i32 %indvars.iv, 1 665 %buf0_a = getelementptr i32, i32* %buf_0, i32 %indvars.iv 666 %buf0_b = getelementptr i32, i32* %buf_0, i32 %indvars.mid 667 %buf1_a = getelementptr i32, i32* %buf_1, i32 %indvars.iv 668 %buf1_b = getelementptr i32, i32* %buf_1, i32 %indvars.mid 669 %va = load atomic i32, i32* %buf0_a unordered, align 4 670 %vb = load atomic i32, i32* %buf0_b unordered, align 4 671 store i32 %va, i32* %buf1_a, align 4 ;; Not atomic 672 store atomic i32 %vb, i32* %buf1_b unordered, align 4 673 %cmp = icmp slt i32 %indvars.iv.next, 3200 674 br i1 %cmp, label %for.body, label %for.end 675 676for.end: 677 ret void 678} 679 680define void @ordered_atomic_ops(i32* noalias %buf_0, i32* noalias %buf_1) { 681; Negative test 682 683; CHECK-LABEL: @ordered_atomic_ops( 684entry: 685 br label %for.body 686 687for.body: 688; CHECK: for.body: 689; CHECK: %indvars.iv.next = add i32 %indvars.iv, 2 690; CHECK: %indvars.mid = add i32 %indvars.iv, 1 691; CHECK: %cmp = icmp slt i32 %indvars.iv.next, 3200 692; CHECK: br i1 %cmp, label %for.body, label %for.end 693 694 %indvars.iv = phi i32 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 695 %indvars.iv.next = add i32 %indvars.iv, 2 696 %indvars.mid = add i32 %indvars.iv, 1 697 %buf0_a = getelementptr i32, i32* %buf_0, i32 %indvars.iv 698 %buf0_b = getelementptr i32, i32* %buf_0, i32 %indvars.mid 699 %buf1_a = getelementptr i32, i32* %buf_1, i32 %indvars.iv 700 %buf1_b = getelementptr i32, i32* %buf_1, i32 %indvars.mid 701 %va = load atomic i32, i32* %buf0_a acquire, align 4 702 %vb = load atomic i32, i32* %buf0_b acquire, align 4 703 store atomic i32 %va, i32* %buf1_a release, align 4 704 store atomic i32 %vb, i32* %buf1_b release, align 4 705 %cmp = icmp slt i32 %indvars.iv.next, 3200 706 br i1 %cmp, label %for.body, label %for.end 707 708for.end: 709 ret void 710} 711 712define void @unordered_atomic_ops_with_fence(i32* noalias %buf_0, i32* noalias %buf_1) { 713; CHECK-LABEL: @unordered_atomic_ops_with_fence( 714entry: 715 br label %for.body 716 717for.body: 718; CHECK: for.body: 719; CHECK: %va = load atomic i32, i32* %buf0_a unordered, align 4 720; CHECK-NEXT: %vb = load atomic i32, i32* %buf0_b unordered, align 4 721; CHECK-NEXT: fence seq_cst 722; CHECK-NEXT: store atomic i32 %va, i32* %buf1_a unordered, align 4 723; CHECK-NEXT: store atomic i32 %vb, i32* %buf1_b unordered, align 4 724 725 %indvars.iv = phi i32 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 726 %indvars.iv.next = add i32 %indvars.iv, 2 727 %indvars.mid = add i32 %indvars.iv, 1 728 %buf0_a = getelementptr i32, i32* %buf_0, i32 %indvars.iv 729 %buf0_b = getelementptr i32, i32* %buf_0, i32 %indvars.mid 730 %buf1_a = getelementptr i32, i32* %buf_1, i32 %indvars.iv 731 %buf1_b = getelementptr i32, i32* %buf_1, i32 %indvars.mid 732 %va = load atomic i32, i32* %buf0_a unordered, align 4 733 %vb = load atomic i32, i32* %buf0_b unordered, align 4 734 fence seq_cst 735 store atomic i32 %va, i32* %buf1_a unordered, align 4 736 store atomic i32 %vb, i32* %buf1_b unordered, align 4 737 %cmp = icmp slt i32 %indvars.iv.next, 3200 738 br i1 %cmp, label %for.body, label %for.end 739 740for.end: 741 ret void 742} 743 744define void @pointer_bitcast_baseinst(i16* %arg, i8* %arg1, i64 %arg2) { 745; CHECK-LABEL: @pointer_bitcast_baseinst( 746; CHECK: bb3: 747; CHECK-NEXT: %indvar = phi i64 [ %indvar.next, %bb3 ], [ 0, %bb ] 748; CHECK-NEXT: %4 = shl nuw i64 %indvar, 3 749; CHECK-NEXT: %5 = add i64 %4, 1 750; CHECK-NEXT: %tmp5 = shl nuw i64 %5, 1 751; CHECK-NEXT: %tmp6 = getelementptr i8, i8* %arg1, i64 %tmp5 752; CHECK-NEXT: %tmp7 = bitcast i8* %tmp6 to <8 x i16>* 753; CHECK-NEXT: %tmp8 = load <8 x i16>, <8 x i16>* %tmp7, align 2 754; CHECK-NEXT: %tmp13 = getelementptr i16, i16* %arg, i64 %5 755; CHECK-NEXT: %tmp14 = bitcast i16* %tmp13 to <8 x i16>* 756; CHECK-NEXT: store <8 x i16> %tmp8, <8 x i16>* %tmp14, align 2 757; CHECK-NEXT: %indvar.next = add i64 %indvar, 1 758; CHECK-NEXT: %exitcond = icmp eq i64 %indvar, %3 759; CHECK-NEXT: br i1 %exitcond, label %bb19, label %bb3 760bb: 761 br label %bb3 762 763bb3: ; preds = %bb3, %bb 764 %tmp = phi i64 [ 1, %bb ], [ %tmp17, %bb3 ] 765 %tmp4 = add nuw i64 %tmp, 8 766 %tmp5 = shl nuw i64 %tmp, 1 767 %tmp6 = getelementptr i8, i8* %arg1, i64 %tmp5 768 %tmp7 = bitcast i8* %tmp6 to <8 x i16>* 769 %tmp8 = load <8 x i16>, <8 x i16>* %tmp7, align 2 770 %tmp9 = shl i64 %tmp4, 1 771 %tmp10 = getelementptr i8, i8* %arg1, i64 %tmp9 772 %tmp11 = bitcast i8* %tmp10 to <8 x i16>* 773 %tmp12 = load <8 x i16>, <8 x i16>* %tmp11, align 2 774 %tmp13 = getelementptr i16, i16* %arg, i64 %tmp 775 %tmp14 = bitcast i16* %tmp13 to <8 x i16>* 776 store <8 x i16> %tmp8, <8 x i16>* %tmp14, align 2 777 %tmp15 = getelementptr i16, i16* %arg, i64 %tmp4 778 %tmp16 = bitcast i16* %tmp15 to <8 x i16>* 779 store <8 x i16> %tmp12, <8 x i16>* %tmp16, align 2 780 %tmp17 = add nuw nsw i64 %tmp, 16 781 %tmp18 = icmp eq i64 %tmp17, %arg2 782 br i1 %tmp18, label %bb19, label %bb3 783 784bb19: ; preds = %bb3 785 ret void 786} 787 788define void @bad_step(i32* nocapture readnone %x) #0 { 789entry: 790 br label %for.body 791 792for.body: ; preds = %for.body, %entry 793 %i.08 = phi i32 [ 0, %entry ], [ %add3, %for.body ] 794 %call = tail call i32 @foo(i32 %i.08) #1 795 %add = add nsw i32 %i.08, 2 796 %call1 = tail call i32 @foo(i32 %add) #1 797 %add2 = add nsw i32 %i.08, 3 798 %call3 = tail call i32 @foo(i32 %add2) #1 799 %add3 = add nsw i32 %i.08, 6 800 %exitcond = icmp sge i32 %add3, 500 801 br i1 %exitcond, label %for.end, label %for.body 802 803; CHECK-LABEL: @bad_step 804; CHECK: %add = add nsw i32 %i.08, 2 805; CHECK: %add2 = add nsw i32 %i.08, 3 806; CHECK: %add3 = add nsw i32 %i.08, 6 807 808for.end: ; preds = %for.body 809 ret void 810} 811 812attributes #0 = { nounwind uwtable } 813attributes #1 = { nounwind } 814 815