1; RUN: opt < %s -loop-unroll -unroll-runtime=true -unroll-runtime-epilog=true -unroll-runtime-multi-exit=true -verify-dom-info -verify-loop-info -S | FileCheck %s -check-prefix=EPILOG-NO-IC 2; RUN: opt < %s -loop-unroll -unroll-runtime=true -unroll-runtime-epilog=true -unroll-runtime-multi-exit=true -verify-dom-info -verify-loop-info -instcombine -S | FileCheck %s -check-prefix=EPILOG 3; RUN: opt < %s -loop-unroll -unroll-runtime -unroll-count=2 -unroll-runtime-epilog=true -unroll-runtime-multi-exit=true -verify-dom-info -verify-loop-info -instcombine 4; RUN: opt < %s -loop-unroll -unroll-runtime=true -unroll-runtime-epilog=false -unroll-runtime-multi-exit=true -verify-dom-info -verify-loop-info -instcombine -S | FileCheck %s -check-prefix=PROLOG 5; RUN: opt < %s -loop-unroll -unroll-runtime -unroll-runtime-epilog=false -unroll-count=2 -unroll-runtime-multi-exit=true -verify-dom-info -verify-loop-info -instcombine 6 7; REQUIRES: asserts 8 9; the third and fifth RUNs generate an epilog/prolog remainder block for all the test 10; cases below (it does not generate a loop). 11 12; test with three exiting and three exit blocks. 13; none of the exit blocks have successors 14define void @test1(i64 %trip, i1 %cond) { 15; EPILOG: test1( 16; EPILOG-NEXT: entry: 17; EPILOG-NEXT: [[TMP0:%.*]] = add i64 [[TRIP:%.*]], -1 18; EPILOG-NEXT: [[XTRAITER:%.*]] = and i64 [[TRIP]], 7 19; EPILOG-NEXT: [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 7 20; EPILOG-NEXT: br i1 [[TMP1]], label %exit2.loopexit.unr-lcssa, label [[ENTRY_NEW:%.*]] 21; EPILOG: entry.new: 22; EPILOG-NEXT: [[UNROLL_ITER:%.*]] = sub i64 [[TRIP]], [[XTRAITER]] 23; EPILOG-NEXT: br label [[LOOP_HEADER:%.*]] 24; EPILOG: loop_latch.epil: 25; EPILOG-NEXT: %epil.iter.sub = add i64 %epil.iter, -1 26; EPILOG-NEXT: %epil.iter.cmp = icmp eq i64 %epil.iter.sub, 0 27; EPILOG-NEXT: br i1 %epil.iter.cmp, label %exit2.loopexit.epilog-lcssa, label %loop_header.epil 28; EPILOG: loop_latch.7: 29; EPILOG-NEXT: %niter.nsub.7 = add i64 %niter, -8 30; EPILOG-NEXT: %niter.ncmp.7 = icmp eq i64 %niter.nsub.7, 0 31; EPILOG-NEXT: br i1 %niter.ncmp.7, label %exit2.loopexit.unr-lcssa.loopexit, label %loop_header 32 33; PROLOG: test1( 34; PROLOG-NEXT: entry: 35; PROLOG-NEXT: [[TMP0:%.*]] = add i64 [[TRIP:%.*]], -1 36; PROLOG-NEXT: [[XTRAITER:%.*]] = and i64 [[TRIP]], 7 37; PROLOG-NEXT: [[TMP1:%.*]] = icmp eq i64 [[XTRAITER]], 0 38; PROLOG-NEXT: br i1 [[TMP1]], label %loop_header.prol.loopexit, label %loop_header.prol.preheader 39; PROLOG: loop_header.prol: 40; PROLOG-NEXT: %iv.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %iv_next.prol, %loop_latch.prol ] 41; PROLOG-NEXT: %prol.iter = phi i64 [ [[XTRAITER]], %loop_header.prol.preheader ], [ %prol.iter.sub, %loop_latch.prol ] 42; PROLOG-NEXT: br i1 %cond, label %loop_latch.prol, label %loop_exiting_bb1.prol 43; PROLOG: loop_latch.prol: 44; PROLOG-NEXT: %iv_next.prol = add i64 %iv.prol, 1 45; PROLOG-NEXT: %prol.iter.sub = add i64 %prol.iter, -1 46; PROLOG-NEXT: %prol.iter.cmp = icmp eq i64 %prol.iter.sub, 0 47; PROLOG-NEXT: br i1 %prol.iter.cmp, label %loop_header.prol.loopexit.unr-lcssa, label %loop_header.prol 48; PROLOG: loop_latch.7: 49; PROLOG-NEXT: %iv_next.7 = add i64 %iv, 8 50; PROLOG-NEXT: %cmp.7 = icmp eq i64 %iv_next.7, %trip 51; PROLOG-NEXT: br i1 %cmp.7, label %exit2.loopexit.unr-lcssa, label %loop_header 52entry: 53 br label %loop_header 54 55loop_header: 56 %iv = phi i64 [ 0, %entry ], [ %iv_next, %loop_latch ] 57 br i1 %cond, label %loop_latch, label %loop_exiting_bb1 58 59loop_exiting_bb1: 60 br i1 false, label %loop_exiting_bb2, label %exit1 61 62loop_exiting_bb2: 63 br i1 false, label %loop_latch, label %exit3 64 65exit3: 66 ret void 67 68loop_latch: 69 %iv_next = add i64 %iv, 1 70 %cmp = icmp ne i64 %iv_next, %trip 71 br i1 %cmp, label %loop_header, label %exit2.loopexit 72 73exit1: 74 ret void 75 76exit2.loopexit: 77 ret void 78} 79 80 81; test with three exiting and two exit blocks. 82; The non-latch exit block has 2 unique predecessors. 83; There are 2 values passed to the exit blocks that are calculated at every iteration. 84; %sum.02 and %add. Both of these are incoming values for phi from every exiting 85; unrolled block. 86define i32 @test2(i32* nocapture %a, i64 %n) { 87; EPILOG: test2( 88; EPILOG: for.exit2.loopexit: 89; EPILOG-NEXT: %retval.ph = phi i32 [ 42, %for.exiting_block ], [ %sum.02, %header ], [ %add, %for.body ], [ 42, %for.exiting_block.1 ], [ %add.1, %for.body.1 ], [ 42, %for.exiting_block.2 ], [ %add.2, %for.body.2 ], [ 42, %for.exiting_block.3 ], 90; EPILOG-NEXT: br label %for.exit2 91; EPILOG: for.exit2.loopexit2: 92; EPILOG-NEXT: %retval.ph3 = phi i32 [ 42, %for.exiting_block.epil ], [ %sum.02.epil, %header.epil ] 93; EPILOG-NEXT: br label %for.exit2 94; EPILOG: for.exit2: 95; EPILOG-NEXT: %retval = phi i32 [ %retval.ph, %for.exit2.loopexit ], [ %retval.ph3, %for.exit2.loopexit2 ] 96; EPILOG-NEXT: ret i32 %retval 97; EPILOG: %niter.nsub.7 = add i64 %niter, -8 98 99; PROLOG: test2( 100; PROLOG: for.exit2.loopexit: 101; PROLOG-NEXT: %retval.ph = phi i32 [ 42, %for.exiting_block ], [ %sum.02, %header ], [ %add, %for.body ], [ 42, %for.exiting_block.1 ], [ %add.1, %for.body.1 ], [ 42, %for.exiting_block.2 ], [ %add.2, %for.body.2 ], [ 42, %for.exiting_block.3 ], 102; PROLOG-NEXT: br label %for.exit2 103; PROLOG: for.exit2.loopexit1: 104; PROLOG-NEXT: %retval.ph2 = phi i32 [ 42, %for.exiting_block.prol ], [ %sum.02.prol, %header.prol ] 105; PROLOG-NEXT: br label %for.exit2 106; PROLOG: for.exit2: 107; PROLOG-NEXT: %retval = phi i32 [ %retval.ph, %for.exit2.loopexit ], [ %retval.ph2, %for.exit2.loopexit1 ] 108; PROLOG-NEXT: ret i32 %retval 109; PROLOG: %indvars.iv.next.7 = add i64 %indvars.iv, 8 110 111entry: 112 br label %header 113 114header: 115 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ] 116 %sum.02 = phi i32 [ %add, %for.body ], [ 0, %entry ] 117 br i1 false, label %for.exit2, label %for.exiting_block 118 119for.exiting_block: 120 %cmp = icmp eq i64 %n, 42 121 br i1 %cmp, label %for.exit2, label %for.body 122 123for.body: 124 %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv 125 %0 = load i32, i32* %arrayidx, align 4 126 %add = add nsw i32 %0, %sum.02 127 %indvars.iv.next = add i64 %indvars.iv, 1 128 %exitcond = icmp eq i64 %indvars.iv.next, %n 129 br i1 %exitcond, label %for.end, label %header 130 131for.end: ; preds = %for.body 132 %sum.0.lcssa = phi i32 [ %add, %for.body ] 133 ret i32 %sum.0.lcssa 134 135for.exit2: 136 %retval = phi i32 [ %sum.02, %header ], [ 42, %for.exiting_block ] 137 ret i32 %retval 138} 139 140; test with two exiting and three exit blocks. 141; the non-latch exiting block has a switch. 142define void @test3(i64 %trip, i64 %add) { 143; EPILOG: test3( 144; EPILOG-NEXT: entry: 145; EPILOG-NEXT: [[TMP0:%.*]] = add i64 [[TRIP:%.*]], -1 146; EPILOG-NEXT: [[XTRAITER:%.*]] = and i64 [[TRIP]], 7 147; EPILOG-NEXT: [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 7 148; EPILOG-NEXT: br i1 [[TMP1]], label %exit2.loopexit.unr-lcssa, label [[ENTRY_NEW:%.*]] 149; EPILOG: entry.new: 150; EPILOG-NEXT: %unroll_iter = sub i64 [[TRIP]], [[XTRAITER]] 151; EPILOG-NEXT: br label [[LOOP_HEADER:%.*]] 152; EPILOG: loop_header: 153; EPILOG-NEXT: %sum = phi i64 [ 0, %entry.new ], [ %sum.next.7, %loop_latch.7 ] 154; EPILOG-NEXT: %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %loop_latch.7 ] 155; EPILOG: loop_exiting_bb1.7: 156; EPILOG-NEXT: switch i64 %sum.next.6, label %loop_latch.7 157; EPILOG: loop_latch.7: 158; EPILOG-NEXT: %sum.next.7 = add i64 %sum.next.6, %add 159; EPILOG-NEXT: %niter.nsub.7 = add i64 %niter, -8 160; EPILOG-NEXT: %niter.ncmp.7 = icmp eq i64 %niter.nsub.7, 0 161; EPILOG-NEXT: br i1 %niter.ncmp.7, label %exit2.loopexit.unr-lcssa.loopexit, label %loop_header 162 163; PROLOG: test3( 164; PROLOG-NEXT: entry: 165; PROLOG-NEXT: [[TMP0:%.*]] = add i64 [[TRIP:%.*]], -1 166; PROLOG-NEXT: [[XTRAITER:%.*]] = and i64 [[TRIP]], 7 167; PROLOG-NEXT: [[TMP1:%.*]] = icmp eq i64 [[XTRAITER]], 0 168; PROLOG-NEXT: br i1 [[TMP1]], label %loop_header.prol.loopexit, label %loop_header.prol.preheader 169; PROLOG: loop_header: 170; PROLOG-NEXT: %iv = phi i64 [ %iv.unr, %entry.new ], [ %iv_next.7, %loop_latch.7 ] 171; PROLOG-NEXT: %sum = phi i64 [ %sum.unr, %entry.new ], [ %sum.next.7, %loop_latch.7 ] 172; PROLOG: loop_exiting_bb1.7: 173; PROLOG-NEXT: switch i64 %sum.next.6, label %loop_latch.7 174; PROLOG: loop_latch.7: 175; PROLOG-NEXT: %iv_next.7 = add nsw i64 %iv, 8 176; PROLOG-NEXT: %sum.next.7 = add i64 %sum.next.6, %add 177; PROLOG-NEXT: %cmp.7 = icmp eq i64 %iv_next.7, %trip 178; PROLOG-NEXT: br i1 %cmp.7, label %exit2.loopexit.unr-lcssa, label %loop_header 179entry: 180 br label %loop_header 181 182loop_header: 183 %iv = phi i64 [ 0, %entry ], [ %iv_next, %loop_latch ] 184 %sum = phi i64 [ 0, %entry ], [ %sum.next, %loop_latch ] 185 br i1 undef, label %loop_latch, label %loop_exiting_bb1 186 187loop_exiting_bb1: 188 switch i64 %sum, label %loop_latch [ 189 i64 24, label %exit1 190 i64 42, label %exit3 191 ] 192 193exit3: 194 ret void 195 196loop_latch: 197 %iv_next = add nuw nsw i64 %iv, 1 198 %sum.next = add i64 %sum, %add 199 %cmp = icmp ne i64 %iv_next, %trip 200 br i1 %cmp, label %loop_header, label %exit2.loopexit 201 202exit1: 203 ret void 204 205exit2.loopexit: 206 ret void 207} 208 209; FIXME: Support multiple exiting blocks to the same latch exit block. 210define i32 @test4(i32* nocapture %a, i64 %n, i1 %cond) { 211; EPILOG: test4( 212; EPILOG-NOT: .unr 213; EPILOG-NOT: .epil 214 215; PROLOG: test4( 216; PROLOG-NOT: .unr 217; PROLOG-NOT: .prol 218entry: 219 br label %header 220 221header: 222 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ] 223 %sum.02 = phi i32 [ %add, %for.body ], [ 0, %entry ] 224 br i1 %cond, label %for.end, label %for.exiting_block 225 226for.exiting_block: 227 %cmp = icmp eq i64 %n, 42 228 br i1 %cmp, label %for.exit2, label %for.body 229 230for.body: ; preds = %for.body, %entry 231 %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv 232 %0 = load i32, i32* %arrayidx, align 4 233 %add = add nsw i32 %0, %sum.02 234 %indvars.iv.next = add i64 %indvars.iv, 1 235 %exitcond = icmp eq i64 %indvars.iv.next, %n 236 br i1 %exitcond, label %for.end, label %header 237 238for.end: ; preds = %for.body, %entry 239 %sum.0.lcssa = phi i32 [ 0, %header ], [ %add, %for.body ] 240 ret i32 %sum.0.lcssa 241 242for.exit2: 243 ret i32 42 244} 245 246; FIXME: Support multiple exiting blocks to the unique exit block. 247define void @unique_exit(i32 %arg) { 248; EPILOG: unique_exit( 249; EPILOG-NOT: .unr 250; EPILOG-NOT: .epil 251 252; PROLOG: unique_exit( 253; PROLOG-NOT: .unr 254; PROLOG-NOT: .prol 255entry: 256 %tmp = icmp sgt i32 undef, %arg 257 br i1 %tmp, label %preheader, label %returnblock 258 259preheader: ; preds = %entry 260 br label %header 261 262LoopExit: ; preds = %header, %latch 263 %tmp2.ph = phi i32 [ %tmp4, %header ], [ -1, %latch ] 264 br label %returnblock 265 266returnblock: ; preds = %LoopExit, %entry 267 %tmp2 = phi i32 [ -1, %entry ], [ %tmp2.ph, %LoopExit ] 268 ret void 269 270header: ; preds = %preheader, %latch 271 %tmp4 = phi i32 [ %inc, %latch ], [ %arg, %preheader ] 272 %inc = add nsw i32 %tmp4, 1 273 br i1 true, label %LoopExit, label %latch 274 275latch: ; preds = %header 276 %cmp = icmp slt i32 %inc, undef 277 br i1 %cmp, label %header, label %LoopExit 278} 279 280; two exiting and two exit blocks. 281; the non-latch exiting block has duplicate edges to the non-latch exit block. 282define i64 @test5(i64 %trip, i64 %add, i1 %cond) { 283; EPILOG: test5( 284; EPILOG: exit1.loopexit: 285; EPILOG-NEXT: %result.ph = phi i64 [ %ivy, %loop_exiting ], [ %ivy, %loop_exiting ], [ %ivy.1, %loop_exiting.1 ], [ %ivy.1, %loop_exiting.1 ], [ %ivy.2, %loop_exiting.2 ], 286; EPILOG-NEXT: br label %exit1 287; EPILOG: exit1.loopexit2: 288; EPILOG-NEXT: %ivy.epil = add i64 %iv.epil, %add 289; EPILOG-NEXT: br label %exit1 290; EPILOG: exit1: 291; EPILOG-NEXT: %result = phi i64 [ %result.ph, %exit1.loopexit ], [ %ivy.epil, %exit1.loopexit2 ] 292; EPILOG-NEXT: ret i64 %result 293; EPILOG: loop_latch.7: 294; EPILOG: %niter.nsub.7 = add i64 %niter, -8 295 296; PROLOG: test5( 297; PROLOG: exit1.loopexit: 298; PROLOG-NEXT: %result.ph = phi i64 [ %ivy, %loop_exiting ], [ %ivy, %loop_exiting ], [ %ivy.1, %loop_exiting.1 ], [ %ivy.1, %loop_exiting.1 ], [ %ivy.2, %loop_exiting.2 ], 299; PROLOG-NEXT: br label %exit1 300; PROLOG: exit1.loopexit1: 301; PROLOG-NEXT: %ivy.prol = add i64 %iv.prol, %add 302; PROLOG-NEXT: br label %exit1 303; PROLOG: exit1: 304; PROLOG-NEXT: %result = phi i64 [ %result.ph, %exit1.loopexit ], [ %ivy.prol, %exit1.loopexit1 ] 305; PROLOG-NEXT: ret i64 %result 306; PROLOG: loop_latch.7: 307; PROLOG: %iv_next.7 = add nsw i64 %iv, 8 308entry: 309 br label %loop_header 310 311loop_header: 312 %iv = phi i64 [ 0, %entry ], [ %iv_next, %loop_latch ] 313 %sum = phi i64 [ 0, %entry ], [ %sum.next, %loop_latch ] 314 br i1 %cond, label %loop_latch, label %loop_exiting 315 316loop_exiting: 317 %ivy = add i64 %iv, %add 318 switch i64 %sum, label %loop_latch [ 319 i64 24, label %exit1 320 i64 42, label %exit1 321 ] 322 323loop_latch: 324 %iv_next = add nuw nsw i64 %iv, 1 325 %sum.next = add i64 %sum, %add 326 %cmp = icmp ne i64 %iv_next, %trip 327 br i1 %cmp, label %loop_header, label %latchexit 328 329exit1: 330 %result = phi i64 [ %ivy, %loop_exiting ], [ %ivy, %loop_exiting ] 331 ret i64 %result 332 333latchexit: 334 ret i64 %sum.next 335} 336 337; test when exit blocks have successors. 338define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) { 339; EPILOG: test6( 340; EPILOG: for.exit2.loopexit: 341; EPILOG-NEXT: %retval.ph = phi i32 [ 42, %for.exiting_block ], [ %sum.02, %header ], [ %add, %latch ], [ 42, %for.exiting_block.1 ], [ %add.1, %latch.1 ], [ 42, %for.exiting_block.2 ], [ %add.2, %latch.2 ], 342; EPILOG-NEXT: br label %for.exit2 343; EPILOG: for.exit2.loopexit2: 344; EPILOG-NEXT: %retval.ph3 = phi i32 [ 42, %for.exiting_block.epil ], [ %sum.02.epil, %header.epil ] 345; EPILOG-NEXT: br label %for.exit2 346; EPILOG: for.exit2: 347; EPILOG-NEXT: %retval = phi i32 [ %retval.ph, %for.exit2.loopexit ], [ %retval.ph3, %for.exit2.loopexit2 ] 348; EPILOG-NEXT: br i1 %cond, label %exit_true, label %exit_false 349; EPILOG: latch.7: 350; EPILOG: %niter.nsub.7 = add i64 %niter, -8 351 352; PROLOG: test6( 353; PROLOG: for.exit2.loopexit: 354; PROLOG-NEXT: %retval.ph = phi i32 [ 42, %for.exiting_block ], [ %sum.02, %header ], [ %add, %latch ], [ 42, %for.exiting_block.1 ], [ %add.1, %latch.1 ], [ 42, %for.exiting_block.2 ], [ %add.2, %latch.2 ], 355; PROLOG-NEXT: br label %for.exit2 356; PROLOG: for.exit2.loopexit1: 357; PROLOG-NEXT: %retval.ph2 = phi i32 [ 42, %for.exiting_block.prol ], [ %sum.02.prol, %header.prol ] 358; PROLOG-NEXT: br label %for.exit2 359; PROLOG: for.exit2: 360; PROLOG-NEXT: %retval = phi i32 [ %retval.ph, %for.exit2.loopexit ], [ %retval.ph2, %for.exit2.loopexit1 ] 361; PROLOG-NEXT: br i1 %cond, label %exit_true, label %exit_false 362; PROLOG: latch.7: 363; PROLOG: %indvars.iv.next.7 = add i64 %indvars.iv, 8 364entry: 365 br label %header 366 367header: 368 %indvars.iv = phi i64 [ %indvars.iv.next, %latch ], [ 0, %entry ] 369 %sum.02 = phi i32 [ %add, %latch ], [ 0, %entry ] 370 br i1 false, label %for.exit2, label %for.exiting_block 371 372for.exiting_block: 373 %cmp = icmp eq i64 %n, 42 374 br i1 %cmp, label %for.exit2, label %latch 375 376latch: 377 %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv 378 %load = load i32, i32* %arrayidx, align 4 379 %add = add nsw i32 %load, %sum.02 380 %indvars.iv.next = add i64 %indvars.iv, 1 381 %exitcond = icmp eq i64 %indvars.iv.next, %n 382 br i1 %exitcond, label %latch_exit, label %header 383 384latch_exit: 385 %sum.0.lcssa = phi i32 [ %add, %latch ] 386 ret i32 %sum.0.lcssa 387 388for.exit2: 389 %retval = phi i32 [ %sum.02, %header ], [ 42, %for.exiting_block ] 390 %addx = add i32 %retval, %x 391 br i1 %cond, label %exit_true, label %exit_false 392 393exit_true: 394 ret i32 %retval 395 396exit_false: 397 ret i32 %addx 398} 399 400; test when value in exit block does not have VMap. 401define i32 @test7(i32 %arg, i32 %arg1, i32 %arg2) { 402; EPILOG-NO-IC: test7( 403; EPILOG-NO-IC: loopexit1.loopexit: 404; EPILOG-NO-IC-NEXT: %sext3.ph = phi i32 [ %shft, %header ], [ %shft, %latch ], [ %shft, %latch.1 ], [ %shft, %latch.2 ], [ %shft, %latch.3 ], [ %shft, %latch.4 ], [ %shft, %latch.5 ], [ %shft, %latch.6 ] 405; EPILOG-NO-IC-NEXT: br label %loopexit1 406; EPILOG-NO-IC: loopexit1.loopexit1: 407; EPILOG-NO-IC-NEXT: %sext3.ph2 = phi i32 [ %shft, %header.epil ] 408; EPILOG-NO-IC-NEXT: br label %loopexit1 409; EPILOG-NO-IC: loopexit1: 410; EPILOG-NO-IC-NEXT: %sext3 = phi i32 [ %sext3.ph, %loopexit1.loopexit ], [ %sext3.ph2, %loopexit1.loopexit1 ] 411bb: 412 %tmp = icmp slt i32 undef, 2 413 %sext = sext i32 undef to i64 414 %shft = ashr exact i32 %arg, 16 415 br i1 %tmp, label %loopexit2, label %preheader 416 417preheader: ; preds = %bb2 418 br label %header 419 420header: ; preds = %latch, %preheader 421 %tmp6 = phi i64 [ 1, %preheader ], [ %add, %latch ] 422 br i1 false, label %loopexit1, label %latch 423 424latch: ; preds = %header 425 %add = add nuw nsw i64 %tmp6, 1 426 %tmp9 = icmp slt i64 %add, %sext 427 br i1 %tmp9, label %header, label %latchexit 428 429latchexit: ; preds = %latch 430 unreachable 431 432loopexit2: ; preds = %bb2 433 ret i32 %shft 434 435loopexit1: ; preds = %header 436 %sext3 = phi i32 [ %shft, %header ] 437 ret i32 %sext3 438} 439 440; Nested loop and inner loop is unrolled 441; FIXME: we cannot unroll with epilog remainder currently, because 442; the outer loop does not contain the epilog preheader and epilog exit (while 443; infact it should). This causes us to choke up on LCSSA form being incorrect in 444; outer loop. However, the exit block where LCSSA fails, is infact still within 445; the outer loop. For now, we just bail out in presence of outer loop and epilog 446; loop is generated. 447; The outer loop header is the preheader for the inner loop and the inner header 448; branches back to the outer loop. 449define void @test8() { 450; EPILOG: test8( 451; EPILOG-NOT: niter 452 453; PROLOG: test8( 454; PROLOG: outerloop: 455; PROLOG-NEXT: phi i64 [ 3, %bb ], [ 0, %outerloop.loopexit ] 456; PROLOG: %lcmp.mod = icmp eq i64 457; PROLOG-NEXT: br i1 %lcmp.mod, label %innerH.prol.loopexit, label %innerH.prol.preheader 458; PROLOG: latch.6: 459; PROLOG-NEXT: %tmp4.7 = add nsw i64 %tmp3, 8 460; PROLOG-NEXT: br i1 false, label %outerloop.loopexit.loopexit, label %latch.7 461; PROLOG: latch.7 462; PROLOG-NEXT: %tmp6.7 = icmp ult i64 %tmp4.7, 100 463; PROLOG-NEXT: br i1 %tmp6.7, label %innerH, label %exit.unr-lcssa 464bb: 465 br label %outerloop 466 467outerloop: ; preds = %innerH, %bb 468 %tmp = phi i64 [ 3, %bb ], [ 0, %innerH ] 469 br label %innerH 470 471innerH: ; preds = %latch, %outerloop 472 %tmp3 = phi i64 [ %tmp4, %latch ], [ %tmp, %outerloop ] 473 %tmp4 = add nuw nsw i64 %tmp3, 1 474 br i1 false, label %outerloop, label %latch 475 476latch: ; preds = %innerH 477 %tmp6 = icmp ult i64 %tmp4, 100 478 br i1 %tmp6, label %innerH, label %exit 479 480exit: ; preds = %latch 481 ret void 482} 483 484declare i8 addrspace(1)* @foo(i32) 485; inner loop prolog unrolled 486; a value from outer loop is used in exit block of inner loop. 487; Don't create VMap entries for such values (%trip). 488define i8 addrspace(1)* @test9(i8* nocapture readonly %arg, i32 %n) { 489; PROLOG: test9( 490; PROLOG: header.prol: 491; PROLOG-NEXT: %phi.prol = phi i64 [ 0, %header.prol.preheader ], [ %iv.next.prol, %latch.prol ] 492; PROLOG: latch.prol: 493; PROLOG-NOT: trip 494; PROLOG: br i1 %prol.iter.cmp, label %header.prol.loopexit.unr-lcssa, label %header.prol 495bb: 496 br label %outerloopHdr 497 498outerloopHdr: ; preds = %outerLatch, %bb 499 %trip = add i32 %n, -1 500 %outercnd = icmp slt i32 0, %trip 501 br i1 %outercnd, label %preheader, label %outerLatch 502 503preheader: ; preds = %outerloopHdr 504 %tmp4 = zext i32 0 to i64 505 br label %header 506 507header: ; preds = %latch, %preheader 508 %phi = phi i64 [ %tmp4, %preheader ], [ %iv.next, %latch ] 509 %tmp7 = trunc i64 %phi to i32 510 br i1 true, label %latch, label %innerexit 511 512innerexit: ; preds = %header 513 %tmp9 = call i8 addrspace(1)* @foo(i32 %trip) 514 ret i8 addrspace(1)* %tmp9 515 516latch: ; preds = %header 517 %tmp11 = add nsw i32 %tmp7, 1 518 %innercnd = icmp slt i32 %tmp11, %trip 519 %iv.next = add nuw nsw i64 %phi, 1 520 br i1 %innercnd, label %header, label %outerLatch 521 522outerLatch: ; preds = %latch, %outerloopHdr 523 br label %outerloopHdr 524} 525