1 /* 2 * Copyright (c) 1998, 2012, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "ci/ciMethodData.hpp" 27 #include "compiler/compileLog.hpp" 28 #include "libadt/vectset.hpp" 29 #include "memory/allocation.inline.hpp" 30 #include "opto/addnode.hpp" 31 #include "opto/callnode.hpp" 32 #include "opto/connode.hpp" 33 #include "opto/divnode.hpp" 34 #include "opto/idealGraphPrinter.hpp" 35 #include "opto/loopnode.hpp" 36 #include "opto/mulnode.hpp" 37 #include "opto/rootnode.hpp" 38 #include "opto/superword.hpp" 39 40 //============================================================================= 41 //------------------------------is_loop_iv------------------------------------- 42 // Determine if a node is Counted loop induction variable. 43 // The method is declared in node.hpp. 44 const Node* Node::is_loop_iv() const { 45 if (this->is_Phi() && !this->as_Phi()->is_copy() && 46 this->as_Phi()->region()->is_CountedLoop() && 47 this->as_Phi()->region()->as_CountedLoop()->phi() == this) { 48 return this; 49 } else { 50 return NULL; 51 } 52 } 53 54 //============================================================================= 55 //------------------------------dump_spec-------------------------------------- 56 // Dump special per-node info 57 #ifndef PRODUCT 58 void LoopNode::dump_spec(outputStream *st) const { 59 if (is_inner_loop()) st->print( "inner " ); 60 if (is_partial_peel_loop()) st->print( "partial_peel " ); 61 if (partial_peel_has_failed()) st->print( "partial_peel_failed " ); 62 } 63 #endif 64 65 //------------------------------is_valid_counted_loop------------------------- 66 bool LoopNode::is_valid_counted_loop() const { 67 if (is_CountedLoop()) { 68 CountedLoopNode* l = as_CountedLoop(); 69 CountedLoopEndNode* le = l->loopexit(); 70 if (le != NULL && 71 le->proj_out(1 /* true */) == l->in(LoopNode::LoopBackControl)) { 72 Node* phi = l->phi(); 73 Node* exit = le->proj_out(0 /* false */); 74 if (exit != NULL && exit->Opcode() == Op_IfFalse && 75 phi != NULL && phi->is_Phi() && 76 phi->in(LoopNode::LoopBackControl) == l->incr() && 77 le->loopnode() == l && le->stride_is_con()) { 78 return true; 79 } 80 } 81 } 82 return false; 83 } 84 85 //------------------------------get_early_ctrl--------------------------------- 86 // Compute earliest legal control 87 Node *PhaseIdealLoop::get_early_ctrl( Node *n ) { 88 assert( !n->is_Phi() && !n->is_CFG(), "this code only handles data nodes" ); 89 uint i; 90 Node *early; 91 if( n->in(0) ) { 92 early = n->in(0); 93 if( !early->is_CFG() ) // Might be a non-CFG multi-def 94 early = get_ctrl(early); // So treat input as a straight data input 95 i = 1; 96 } else { 97 early = get_ctrl(n->in(1)); 98 i = 2; 99 } 100 uint e_d = dom_depth(early); 101 assert( early, "" ); 102 for( ; i < n->req(); i++ ) { 103 Node *cin = get_ctrl(n->in(i)); 104 assert( cin, "" ); 105 // Keep deepest dominator depth 106 uint c_d = dom_depth(cin); 107 if( c_d > e_d ) { // Deeper guy? 108 early = cin; // Keep deepest found so far 109 e_d = c_d; 110 } else if( c_d == e_d && // Same depth? 111 early != cin ) { // If not equal, must use slower algorithm 112 // If same depth but not equal, one _must_ dominate the other 113 // and we want the deeper (i.e., dominated) guy. 114 Node *n1 = early; 115 Node *n2 = cin; 116 while( 1 ) { 117 n1 = idom(n1); // Walk up until break cycle 118 n2 = idom(n2); 119 if( n1 == cin || // Walked early up to cin 120 dom_depth(n2) < c_d ) 121 break; // early is deeper; keep him 122 if( n2 == early || // Walked cin up to early 123 dom_depth(n1) < c_d ) { 124 early = cin; // cin is deeper; keep him 125 break; 126 } 127 } 128 e_d = dom_depth(early); // Reset depth register cache 129 } 130 } 131 132 // Return earliest legal location 133 assert(early == find_non_split_ctrl(early), "unexpected early control"); 134 135 return early; 136 } 137 138 //------------------------------set_early_ctrl--------------------------------- 139 // Set earliest legal control 140 void PhaseIdealLoop::set_early_ctrl( Node *n ) { 141 Node *early = get_early_ctrl(n); 142 143 // Record earliest legal location 144 set_ctrl(n, early); 145 } 146 147 //------------------------------set_subtree_ctrl------------------------------- 148 // set missing _ctrl entries on new nodes 149 void PhaseIdealLoop::set_subtree_ctrl( Node *n ) { 150 // Already set? Get out. 151 if( _nodes[n->_idx] ) return; 152 // Recursively set _nodes array to indicate where the Node goes 153 uint i; 154 for( i = 0; i < n->req(); ++i ) { 155 Node *m = n->in(i); 156 if( m && m != C->root() ) 157 set_subtree_ctrl( m ); 158 } 159 160 // Fixup self 161 set_early_ctrl( n ); 162 } 163 164 //------------------------------is_counted_loop-------------------------------- 165 bool PhaseIdealLoop::is_counted_loop( Node *x, IdealLoopTree *loop ) { 166 PhaseGVN *gvn = &_igvn; 167 168 // Counted loop head must be a good RegionNode with only 3 not NULL 169 // control input edges: Self, Entry, LoopBack. 170 if (x->in(LoopNode::Self) == NULL || x->req() != 3) 171 return false; 172 173 Node *init_control = x->in(LoopNode::EntryControl); 174 Node *back_control = x->in(LoopNode::LoopBackControl); 175 if (init_control == NULL || back_control == NULL) // Partially dead 176 return false; 177 // Must also check for TOP when looking for a dead loop 178 if (init_control->is_top() || back_control->is_top()) 179 return false; 180 181 // Allow funny placement of Safepoint 182 if (back_control->Opcode() == Op_SafePoint) 183 back_control = back_control->in(TypeFunc::Control); 184 185 // Controlling test for loop 186 Node *iftrue = back_control; 187 uint iftrue_op = iftrue->Opcode(); 188 if (iftrue_op != Op_IfTrue && 189 iftrue_op != Op_IfFalse) 190 // I have a weird back-control. Probably the loop-exit test is in 191 // the middle of the loop and I am looking at some trailing control-flow 192 // merge point. To fix this I would have to partially peel the loop. 193 return false; // Obscure back-control 194 195 // Get boolean guarding loop-back test 196 Node *iff = iftrue->in(0); 197 if (get_loop(iff) != loop || !iff->in(1)->is_Bool()) 198 return false; 199 BoolNode *test = iff->in(1)->as_Bool(); 200 BoolTest::mask bt = test->_test._test; 201 float cl_prob = iff->as_If()->_prob; 202 if (iftrue_op == Op_IfFalse) { 203 bt = BoolTest(bt).negate(); 204 cl_prob = 1.0 - cl_prob; 205 } 206 // Get backedge compare 207 Node *cmp = test->in(1); 208 int cmp_op = cmp->Opcode(); 209 if (cmp_op != Op_CmpI) 210 return false; // Avoid pointer & float compares 211 212 // Find the trip-counter increment & limit. Limit must be loop invariant. 213 Node *incr = cmp->in(1); 214 Node *limit = cmp->in(2); 215 216 // --------- 217 // need 'loop()' test to tell if limit is loop invariant 218 // --------- 219 220 if (!is_member(loop, get_ctrl(incr))) { // Swapped trip counter and limit? 221 Node *tmp = incr; // Then reverse order into the CmpI 222 incr = limit; 223 limit = tmp; 224 bt = BoolTest(bt).commute(); // And commute the exit test 225 } 226 if (is_member(loop, get_ctrl(limit))) // Limit must be loop-invariant 227 return false; 228 if (!is_member(loop, get_ctrl(incr))) // Trip counter must be loop-variant 229 return false; 230 231 Node* phi_incr = NULL; 232 // Trip-counter increment must be commutative & associative. 233 if (incr->is_Phi()) { 234 if (incr->as_Phi()->region() != x || incr->req() != 3) 235 return false; // Not simple trip counter expression 236 phi_incr = incr; 237 incr = phi_incr->in(LoopNode::LoopBackControl); // Assume incr is on backedge of Phi 238 if (!is_member(loop, get_ctrl(incr))) // Trip counter must be loop-variant 239 return false; 240 } 241 242 Node* trunc1 = NULL; 243 Node* trunc2 = NULL; 244 const TypeInt* iv_trunc_t = NULL; 245 if (!(incr = CountedLoopNode::match_incr_with_optional_truncation(incr, &trunc1, &trunc2, &iv_trunc_t))) { 246 return false; // Funny increment opcode 247 } 248 assert(incr->Opcode() == Op_AddI, "wrong increment code"); 249 250 // Get merge point 251 Node *xphi = incr->in(1); 252 Node *stride = incr->in(2); 253 if (!stride->is_Con()) { // Oops, swap these 254 if (!xphi->is_Con()) // Is the other guy a constant? 255 return false; // Nope, unknown stride, bail out 256 Node *tmp = xphi; // 'incr' is commutative, so ok to swap 257 xphi = stride; 258 stride = tmp; 259 } 260 // Stride must be constant 261 int stride_con = stride->get_int(); 262 if (stride_con == 0) 263 return false; // missed some peephole opt 264 265 if (!xphi->is_Phi()) 266 return false; // Too much math on the trip counter 267 if (phi_incr != NULL && phi_incr != xphi) 268 return false; 269 PhiNode *phi = xphi->as_Phi(); 270 271 // Phi must be of loop header; backedge must wrap to increment 272 if (phi->region() != x) 273 return false; 274 if (trunc1 == NULL && phi->in(LoopNode::LoopBackControl) != incr || 275 trunc1 != NULL && phi->in(LoopNode::LoopBackControl) != trunc1) { 276 return false; 277 } 278 Node *init_trip = phi->in(LoopNode::EntryControl); 279 280 // If iv trunc type is smaller than int, check for possible wrap. 281 if (!TypeInt::INT->higher_equal(iv_trunc_t)) { 282 assert(trunc1 != NULL, "must have found some truncation"); 283 284 // Get a better type for the phi (filtered thru if's) 285 const TypeInt* phi_ft = filtered_type(phi); 286 287 // Can iv take on a value that will wrap? 288 // 289 // Ensure iv's limit is not within "stride" of the wrap value. 290 // 291 // Example for "short" type 292 // Truncation ensures value is in the range -32768..32767 (iv_trunc_t) 293 // If the stride is +10, then the last value of the induction 294 // variable before the increment (phi_ft->_hi) must be 295 // <= 32767 - 10 and (phi_ft->_lo) must be >= -32768 to 296 // ensure no truncation occurs after the increment. 297 298 if (stride_con > 0) { 299 if (iv_trunc_t->_hi - phi_ft->_hi < stride_con || 300 iv_trunc_t->_lo > phi_ft->_lo) { 301 return false; // truncation may occur 302 } 303 } else if (stride_con < 0) { 304 if (iv_trunc_t->_lo - phi_ft->_lo > stride_con || 305 iv_trunc_t->_hi < phi_ft->_hi) { 306 return false; // truncation may occur 307 } 308 } 309 // No possibility of wrap so truncation can be discarded 310 // Promote iv type to Int 311 } else { 312 assert(trunc1 == NULL && trunc2 == NULL, "no truncation for int"); 313 } 314 315 // If the condition is inverted and we will be rolling 316 // through MININT to MAXINT, then bail out. 317 if (bt == BoolTest::eq || // Bail out, but this loop trips at most twice! 318 // Odd stride 319 bt == BoolTest::ne && stride_con != 1 && stride_con != -1 || 320 // Count down loop rolls through MAXINT 321 (bt == BoolTest::le || bt == BoolTest::lt) && stride_con < 0 || 322 // Count up loop rolls through MININT 323 (bt == BoolTest::ge || bt == BoolTest::gt) && stride_con > 0) { 324 return false; // Bail out 325 } 326 327 const TypeInt* init_t = gvn->type(init_trip)->is_int(); 328 const TypeInt* limit_t = gvn->type(limit)->is_int(); 329 330 if (stride_con > 0) { 331 long init_p = (long)init_t->_lo + stride_con; 332 if (init_p > (long)max_jint || init_p > (long)limit_t->_hi) 333 return false; // cyclic loop or this loop trips only once 334 } else { 335 long init_p = (long)init_t->_hi + stride_con; 336 if (init_p < (long)min_jint || init_p < (long)limit_t->_lo) 337 return false; // cyclic loop or this loop trips only once 338 } 339 340 // ================================================= 341 // ---- SUCCESS! Found A Trip-Counted Loop! ----- 342 // 343 assert(x->Opcode() == Op_Loop, "regular loops only"); 344 C->print_method("Before CountedLoop", 3); 345 346 Node *hook = new (C, 6) Node(6); 347 348 if (LoopLimitCheck) { 349 350 // =================================================== 351 // Generate loop limit check to avoid integer overflow 352 // in cases like next (cyclic loops): 353 // 354 // for (i=0; i <= max_jint; i++) {} 355 // for (i=0; i < max_jint; i+=2) {} 356 // 357 // 358 // Limit check predicate depends on the loop test: 359 // 360 // for(;i != limit; i++) --> limit <= (max_jint) 361 // for(;i < limit; i+=stride) --> limit <= (max_jint - stride + 1) 362 // for(;i <= limit; i+=stride) --> limit <= (max_jint - stride ) 363 // 364 365 // Check if limit is excluded to do more precise int overflow check. 366 bool incl_limit = (bt == BoolTest::le || bt == BoolTest::ge); 367 int stride_m = stride_con - (incl_limit ? 0 : (stride_con > 0 ? 1 : -1)); 368 369 // If compare points directly to the phi we need to adjust 370 // the compare so that it points to the incr. Limit have 371 // to be adjusted to keep trip count the same and the 372 // adjusted limit should be checked for int overflow. 373 if (phi_incr != NULL) { 374 stride_m += stride_con; 375 } 376 377 if (limit->is_Con()) { 378 int limit_con = limit->get_int(); 379 if ((stride_con > 0 && limit_con > (max_jint - stride_m)) || 380 (stride_con < 0 && limit_con < (min_jint - stride_m))) { 381 // Bailout: it could be integer overflow. 382 return false; 383 } 384 } else if ((stride_con > 0 && limit_t->_hi <= (max_jint - stride_m)) || 385 (stride_con < 0 && limit_t->_lo >= (min_jint - stride_m))) { 386 // Limit's type may satisfy the condition, for example, 387 // when it is an array length. 388 } else { 389 // Generate loop's limit check. 390 // Loop limit check predicate should be near the loop. 391 ProjNode *limit_check_proj = find_predicate_insertion_point(init_control, Deoptimization::Reason_loop_limit_check); 392 if (!limit_check_proj) { 393 // The limit check predicate is not generated if this method trapped here before. 394 #ifdef ASSERT 395 if (TraceLoopLimitCheck) { 396 tty->print("missing loop limit check:"); 397 loop->dump_head(); 398 x->dump(1); 399 } 400 #endif 401 return false; 402 } 403 404 IfNode* check_iff = limit_check_proj->in(0)->as_If(); 405 Node* cmp_limit; 406 Node* bol; 407 408 if (stride_con > 0) { 409 cmp_limit = new (C, 3) CmpINode(limit, _igvn.intcon(max_jint - stride_m)); 410 bol = new (C, 2) BoolNode(cmp_limit, BoolTest::le); 411 } else { 412 cmp_limit = new (C, 3) CmpINode(limit, _igvn.intcon(min_jint - stride_m)); 413 bol = new (C, 2) BoolNode(cmp_limit, BoolTest::ge); 414 } 415 cmp_limit = _igvn.register_new_node_with_optimizer(cmp_limit); 416 bol = _igvn.register_new_node_with_optimizer(bol); 417 set_subtree_ctrl(bol); 418 419 // Replace condition in original predicate but preserve Opaque node 420 // so that previous predicates could be found. 421 assert(check_iff->in(1)->Opcode() == Op_Conv2B && 422 check_iff->in(1)->in(1)->Opcode() == Op_Opaque1, ""); 423 Node* opq = check_iff->in(1)->in(1); 424 _igvn.hash_delete(opq); 425 opq->set_req(1, bol); 426 // Update ctrl. 427 set_ctrl(opq, check_iff->in(0)); 428 set_ctrl(check_iff->in(1), check_iff->in(0)); 429 430 #ifndef PRODUCT 431 // report that the loop predication has been actually performed 432 // for this loop 433 if (TraceLoopLimitCheck) { 434 tty->print_cr("Counted Loop Limit Check generated:"); 435 debug_only( bol->dump(2); ) 436 } 437 #endif 438 } 439 440 if (phi_incr != NULL) { 441 // If compare points directly to the phi we need to adjust 442 // the compare so that it points to the incr. Limit have 443 // to be adjusted to keep trip count the same and we 444 // should avoid int overflow. 445 // 446 // i = init; do {} while(i++ < limit); 447 // is converted to 448 // i = init; do {} while(++i < limit+1); 449 // 450 limit = gvn->transform(new (C, 3) AddINode(limit, stride)); 451 } 452 453 // Now we need to canonicalize loop condition. 454 if (bt == BoolTest::ne) { 455 assert(stride_con == 1 || stride_con == -1, "simple increment only"); 456 // 'ne' can be replaced with 'lt' only when init < limit. 457 if (stride_con > 0 && init_t->_hi < limit_t->_lo) 458 bt = BoolTest::lt; 459 // 'ne' can be replaced with 'gt' only when init > limit. 460 if (stride_con < 0 && init_t->_lo > limit_t->_hi) 461 bt = BoolTest::gt; 462 } 463 464 if (incl_limit) { 465 // The limit check guaranties that 'limit <= (max_jint - stride)' so 466 // we can convert 'i <= limit' to 'i < limit+1' since stride != 0. 467 // 468 Node* one = (stride_con > 0) ? gvn->intcon( 1) : gvn->intcon(-1); 469 limit = gvn->transform(new (C, 3) AddINode(limit, one)); 470 if (bt == BoolTest::le) 471 bt = BoolTest::lt; 472 else if (bt == BoolTest::ge) 473 bt = BoolTest::gt; 474 else 475 ShouldNotReachHere(); 476 } 477 set_subtree_ctrl( limit ); 478 479 } else { // LoopLimitCheck 480 481 // If compare points to incr, we are ok. Otherwise the compare 482 // can directly point to the phi; in this case adjust the compare so that 483 // it points to the incr by adjusting the limit. 484 if (cmp->in(1) == phi || cmp->in(2) == phi) 485 limit = gvn->transform(new (C, 3) AddINode(limit,stride)); 486 487 // trip-count for +-tive stride should be: (limit - init_trip + stride - 1)/stride. 488 // Final value for iterator should be: trip_count * stride + init_trip. 489 Node *one_p = gvn->intcon( 1); 490 Node *one_m = gvn->intcon(-1); 491 492 Node *trip_count = NULL; 493 switch( bt ) { 494 case BoolTest::eq: 495 ShouldNotReachHere(); 496 case BoolTest::ne: // Ahh, the case we desire 497 if (stride_con == 1) 498 trip_count = gvn->transform(new (C, 3) SubINode(limit,init_trip)); 499 else if (stride_con == -1) 500 trip_count = gvn->transform(new (C, 3) SubINode(init_trip,limit)); 501 else 502 ShouldNotReachHere(); 503 set_subtree_ctrl(trip_count); 504 //_loop.map(trip_count->_idx,loop(limit)); 505 break; 506 case BoolTest::le: // Maybe convert to '<' case 507 limit = gvn->transform(new (C, 3) AddINode(limit,one_p)); 508 set_subtree_ctrl( limit ); 509 hook->init_req(4, limit); 510 511 bt = BoolTest::lt; 512 // Make the new limit be in the same loop nest as the old limit 513 //_loop.map(limit->_idx,limit_loop); 514 // Fall into next case 515 case BoolTest::lt: { // Maybe convert to '!=' case 516 if (stride_con < 0) // Count down loop rolls through MAXINT 517 ShouldNotReachHere(); 518 Node *range = gvn->transform(new (C, 3) SubINode(limit,init_trip)); 519 set_subtree_ctrl( range ); 520 hook->init_req(0, range); 521 522 Node *bias = gvn->transform(new (C, 3) AddINode(range,stride)); 523 set_subtree_ctrl( bias ); 524 hook->init_req(1, bias); 525 526 Node *bias1 = gvn->transform(new (C, 3) AddINode(bias,one_m)); 527 set_subtree_ctrl( bias1 ); 528 hook->init_req(2, bias1); 529 530 trip_count = gvn->transform(new (C, 3) DivINode(0,bias1,stride)); 531 set_subtree_ctrl( trip_count ); 532 hook->init_req(3, trip_count); 533 break; 534 } 535 536 case BoolTest::ge: // Maybe convert to '>' case 537 limit = gvn->transform(new (C, 3) AddINode(limit,one_m)); 538 set_subtree_ctrl( limit ); 539 hook->init_req(4 ,limit); 540 541 bt = BoolTest::gt; 542 // Make the new limit be in the same loop nest as the old limit 543 //_loop.map(limit->_idx,limit_loop); 544 // Fall into next case 545 case BoolTest::gt: { // Maybe convert to '!=' case 546 if (stride_con > 0) // count up loop rolls through MININT 547 ShouldNotReachHere(); 548 Node *range = gvn->transform(new (C, 3) SubINode(limit,init_trip)); 549 set_subtree_ctrl( range ); 550 hook->init_req(0, range); 551 552 Node *bias = gvn->transform(new (C, 3) AddINode(range,stride)); 553 set_subtree_ctrl( bias ); 554 hook->init_req(1, bias); 555 556 Node *bias1 = gvn->transform(new (C, 3) AddINode(bias,one_p)); 557 set_subtree_ctrl( bias1 ); 558 hook->init_req(2, bias1); 559 560 trip_count = gvn->transform(new (C, 3) DivINode(0,bias1,stride)); 561 set_subtree_ctrl( trip_count ); 562 hook->init_req(3, trip_count); 563 break; 564 } 565 } // switch( bt ) 566 567 Node *span = gvn->transform(new (C, 3) MulINode(trip_count,stride)); 568 set_subtree_ctrl( span ); 569 hook->init_req(5, span); 570 571 limit = gvn->transform(new (C, 3) AddINode(span,init_trip)); 572 set_subtree_ctrl( limit ); 573 574 } // LoopLimitCheck 575 576 // Check for SafePoint on backedge and remove 577 Node *sfpt = x->in(LoopNode::LoopBackControl); 578 if (sfpt->Opcode() == Op_SafePoint && is_deleteable_safept(sfpt)) { 579 lazy_replace( sfpt, iftrue ); 580 loop->_tail = iftrue; 581 } 582 583 // Build a canonical trip test. 584 // Clone code, as old values may be in use. 585 incr = incr->clone(); 586 incr->set_req(1,phi); 587 incr->set_req(2,stride); 588 incr = _igvn.register_new_node_with_optimizer(incr); 589 set_early_ctrl( incr ); 590 _igvn.hash_delete(phi); 591 phi->set_req_X( LoopNode::LoopBackControl, incr, &_igvn ); 592 593 // If phi type is more restrictive than Int, raise to 594 // Int to prevent (almost) infinite recursion in igvn 595 // which can only handle integer types for constants or minint..maxint. 596 if (!TypeInt::INT->higher_equal(phi->bottom_type())) { 597 Node* nphi = PhiNode::make(phi->in(0), phi->in(LoopNode::EntryControl), TypeInt::INT); 598 nphi->set_req(LoopNode::LoopBackControl, phi->in(LoopNode::LoopBackControl)); 599 nphi = _igvn.register_new_node_with_optimizer(nphi); 600 set_ctrl(nphi, get_ctrl(phi)); 601 _igvn.replace_node(phi, nphi); 602 phi = nphi->as_Phi(); 603 } 604 cmp = cmp->clone(); 605 cmp->set_req(1,incr); 606 cmp->set_req(2,limit); 607 cmp = _igvn.register_new_node_with_optimizer(cmp); 608 set_ctrl(cmp, iff->in(0)); 609 610 test = test->clone()->as_Bool(); 611 (*(BoolTest*)&test->_test)._test = bt; 612 test->set_req(1,cmp); 613 _igvn.register_new_node_with_optimizer(test); 614 set_ctrl(test, iff->in(0)); 615 616 // Replace the old IfNode with a new LoopEndNode 617 Node *lex = _igvn.register_new_node_with_optimizer(new (C, 2) CountedLoopEndNode( iff->in(0), test, cl_prob, iff->as_If()->_fcnt )); 618 IfNode *le = lex->as_If(); 619 uint dd = dom_depth(iff); 620 set_idom(le, le->in(0), dd); // Update dominance for loop exit 621 set_loop(le, loop); 622 623 // Get the loop-exit control 624 Node *iffalse = iff->as_If()->proj_out(!(iftrue_op == Op_IfTrue)); 625 626 // Need to swap loop-exit and loop-back control? 627 if (iftrue_op == Op_IfFalse) { 628 Node *ift2=_igvn.register_new_node_with_optimizer(new (C, 1) IfTrueNode (le)); 629 Node *iff2=_igvn.register_new_node_with_optimizer(new (C, 1) IfFalseNode(le)); 630 631 loop->_tail = back_control = ift2; 632 set_loop(ift2, loop); 633 set_loop(iff2, get_loop(iffalse)); 634 635 // Lazy update of 'get_ctrl' mechanism. 636 lazy_replace_proj( iffalse, iff2 ); 637 lazy_replace_proj( iftrue, ift2 ); 638 639 // Swap names 640 iffalse = iff2; 641 iftrue = ift2; 642 } else { 643 _igvn.hash_delete(iffalse); 644 _igvn.hash_delete(iftrue); 645 iffalse->set_req_X( 0, le, &_igvn ); 646 iftrue ->set_req_X( 0, le, &_igvn ); 647 } 648 649 set_idom(iftrue, le, dd+1); 650 set_idom(iffalse, le, dd+1); 651 assert(iff->outcnt() == 0, "should be dead now"); 652 lazy_replace( iff, le ); // fix 'get_ctrl' 653 654 // Now setup a new CountedLoopNode to replace the existing LoopNode 655 CountedLoopNode *l = new (C, 3) CountedLoopNode(init_control, back_control); 656 l->set_unswitch_count(x->as_Loop()->unswitch_count()); // Preserve 657 // The following assert is approximately true, and defines the intention 658 // of can_be_counted_loop. It fails, however, because phase->type 659 // is not yet initialized for this loop and its parts. 660 //assert(l->can_be_counted_loop(this), "sanity"); 661 _igvn.register_new_node_with_optimizer(l); 662 set_loop(l, loop); 663 loop->_head = l; 664 // Fix all data nodes placed at the old loop head. 665 // Uses the lazy-update mechanism of 'get_ctrl'. 666 lazy_replace( x, l ); 667 set_idom(l, init_control, dom_depth(x)); 668 669 // Check for immediately preceding SafePoint and remove 670 Node *sfpt2 = le->in(0); 671 if (sfpt2->Opcode() == Op_SafePoint && is_deleteable_safept(sfpt2)) 672 lazy_replace( sfpt2, sfpt2->in(TypeFunc::Control)); 673 674 // Free up intermediate goo 675 _igvn.remove_dead_node(hook); 676 677 #ifdef ASSERT 678 assert(l->is_valid_counted_loop(), "counted loop shape is messed up"); 679 assert(l == loop->_head && l->phi() == phi && l->loopexit() == lex, "" ); 680 #endif 681 #ifndef PRODUCT 682 if (TraceLoopOpts) { 683 tty->print("Counted "); 684 loop->dump_head(); 685 } 686 #endif 687 688 C->print_method("After CountedLoop", 3); 689 690 return true; 691 } 692 693 //----------------------exact_limit------------------------------------------- 694 Node* PhaseIdealLoop::exact_limit( IdealLoopTree *loop ) { 695 assert(loop->_head->is_CountedLoop(), ""); 696 CountedLoopNode *cl = loop->_head->as_CountedLoop(); 697 assert(cl->is_valid_counted_loop(), ""); 698 699 if (!LoopLimitCheck || ABS(cl->stride_con()) == 1 || 700 cl->limit()->Opcode() == Op_LoopLimit) { 701 // Old code has exact limit (it could be incorrect in case of int overflow). 702 // Loop limit is exact with stride == 1. And loop may already have exact limit. 703 return cl->limit(); 704 } 705 Node *limit = NULL; 706 #ifdef ASSERT 707 BoolTest::mask bt = cl->loopexit()->test_trip(); 708 assert(bt == BoolTest::lt || bt == BoolTest::gt, "canonical test is expected"); 709 #endif 710 if (cl->has_exact_trip_count()) { 711 // Simple case: loop has constant boundaries. 712 // Use longs to avoid integer overflow. 713 int stride_con = cl->stride_con(); 714 long init_con = cl->init_trip()->get_int(); 715 long limit_con = cl->limit()->get_int(); 716 julong trip_cnt = cl->trip_count(); 717 long final_con = init_con + trip_cnt*stride_con; 718 int final_int = (int)final_con; 719 // The final value should be in integer range since the loop 720 // is counted and the limit was checked for overflow. 721 assert(final_con == (long)final_int, "final value should be integer"); 722 limit = _igvn.intcon(final_int); 723 } else { 724 // Create new LoopLimit node to get exact limit (final iv value). 725 limit = new (C, 4) LoopLimitNode(C, cl->init_trip(), cl->limit(), cl->stride()); 726 register_new_node(limit, cl->in(LoopNode::EntryControl)); 727 } 728 assert(limit != NULL, "sanity"); 729 return limit; 730 } 731 732 //------------------------------Ideal------------------------------------------ 733 // Return a node which is more "ideal" than the current node. 734 // Attempt to convert into a counted-loop. 735 Node *LoopNode::Ideal(PhaseGVN *phase, bool can_reshape) { 736 if (!can_be_counted_loop(phase)) { 737 phase->C->set_major_progress(); 738 } 739 return RegionNode::Ideal(phase, can_reshape); 740 } 741 742 743 //============================================================================= 744 //------------------------------Ideal------------------------------------------ 745 // Return a node which is more "ideal" than the current node. 746 // Attempt to convert into a counted-loop. 747 Node *CountedLoopNode::Ideal(PhaseGVN *phase, bool can_reshape) { 748 return RegionNode::Ideal(phase, can_reshape); 749 } 750 751 //------------------------------dump_spec-------------------------------------- 752 // Dump special per-node info 753 #ifndef PRODUCT 754 void CountedLoopNode::dump_spec(outputStream *st) const { 755 LoopNode::dump_spec(st); 756 if (stride_is_con()) { 757 st->print("stride: %d ",stride_con()); 758 } 759 if (is_pre_loop ()) st->print("pre of N%d" , _main_idx); 760 if (is_main_loop()) st->print("main of N%d", _idx); 761 if (is_post_loop()) st->print("post of N%d", _main_idx); 762 } 763 #endif 764 765 //============================================================================= 766 int CountedLoopEndNode::stride_con() const { 767 return stride()->bottom_type()->is_int()->get_con(); 768 } 769 770 //============================================================================= 771 //------------------------------Value----------------------------------------- 772 const Type *LoopLimitNode::Value( PhaseTransform *phase ) const { 773 const Type* init_t = phase->type(in(Init)); 774 const Type* limit_t = phase->type(in(Limit)); 775 const Type* stride_t = phase->type(in(Stride)); 776 // Either input is TOP ==> the result is TOP 777 if (init_t == Type::TOP) return Type::TOP; 778 if (limit_t == Type::TOP) return Type::TOP; 779 if (stride_t == Type::TOP) return Type::TOP; 780 781 int stride_con = stride_t->is_int()->get_con(); 782 if (stride_con == 1) 783 return NULL; // Identity 784 785 if (init_t->is_int()->is_con() && limit_t->is_int()->is_con()) { 786 // Use longs to avoid integer overflow. 787 long init_con = init_t->is_int()->get_con(); 788 long limit_con = limit_t->is_int()->get_con(); 789 int stride_m = stride_con - (stride_con > 0 ? 1 : -1); 790 long trip_count = (limit_con - init_con + stride_m)/stride_con; 791 long final_con = init_con + stride_con*trip_count; 792 int final_int = (int)final_con; 793 // The final value should be in integer range since the loop 794 // is counted and the limit was checked for overflow. 795 assert(final_con == (long)final_int, "final value should be integer"); 796 return TypeInt::make(final_int); 797 } 798 799 return bottom_type(); // TypeInt::INT 800 } 801 802 //------------------------------Ideal------------------------------------------ 803 // Return a node which is more "ideal" than the current node. 804 Node *LoopLimitNode::Ideal(PhaseGVN *phase, bool can_reshape) { 805 if (phase->type(in(Init)) == Type::TOP || 806 phase->type(in(Limit)) == Type::TOP || 807 phase->type(in(Stride)) == Type::TOP) 808 return NULL; // Dead 809 810 int stride_con = phase->type(in(Stride))->is_int()->get_con(); 811 if (stride_con == 1) 812 return NULL; // Identity 813 814 if (in(Init)->is_Con() && in(Limit)->is_Con()) 815 return NULL; // Value 816 817 // Delay following optimizations until all loop optimizations 818 // done to keep Ideal graph simple. 819 if (!can_reshape || phase->C->major_progress()) 820 return NULL; 821 822 const TypeInt* init_t = phase->type(in(Init) )->is_int(); 823 const TypeInt* limit_t = phase->type(in(Limit))->is_int(); 824 int stride_p; 825 long lim, ini; 826 julong max; 827 if (stride_con > 0) { 828 stride_p = stride_con; 829 lim = limit_t->_hi; 830 ini = init_t->_lo; 831 max = (julong)max_jint; 832 } else { 833 stride_p = -stride_con; 834 lim = init_t->_hi; 835 ini = limit_t->_lo; 836 max = (julong)min_jint; 837 } 838 julong range = lim - ini + stride_p; 839 if (range <= max) { 840 // Convert to integer expression if it is not overflow. 841 Node* stride_m = phase->intcon(stride_con - (stride_con > 0 ? 1 : -1)); 842 Node *range = phase->transform(new (phase->C, 3) SubINode(in(Limit), in(Init))); 843 Node *bias = phase->transform(new (phase->C, 3) AddINode(range, stride_m)); 844 Node *trip = phase->transform(new (phase->C, 3) DivINode(0, bias, in(Stride))); 845 Node *span = phase->transform(new (phase->C, 3) MulINode(trip, in(Stride))); 846 return new (phase->C, 3) AddINode(span, in(Init)); // exact limit 847 } 848 849 if (is_power_of_2(stride_p) || // divisor is 2^n 850 !Matcher::has_match_rule(Op_LoopLimit)) { // or no specialized Mach node? 851 // Convert to long expression to avoid integer overflow 852 // and let igvn optimizer convert this division. 853 // 854 Node* init = phase->transform( new (phase->C, 2) ConvI2LNode(in(Init))); 855 Node* limit = phase->transform( new (phase->C, 2) ConvI2LNode(in(Limit))); 856 Node* stride = phase->longcon(stride_con); 857 Node* stride_m = phase->longcon(stride_con - (stride_con > 0 ? 1 : -1)); 858 859 Node *range = phase->transform(new (phase->C, 3) SubLNode(limit, init)); 860 Node *bias = phase->transform(new (phase->C, 3) AddLNode(range, stride_m)); 861 Node *span; 862 if (stride_con > 0 && is_power_of_2(stride_p)) { 863 // bias >= 0 if stride >0, so if stride is 2^n we can use &(-stride) 864 // and avoid generating rounding for division. Zero trip guard should 865 // guarantee that init < limit but sometimes the guard is missing and 866 // we can get situation when init > limit. Note, for the empty loop 867 // optimization zero trip guard is generated explicitly which leaves 868 // only RCE predicate where exact limit is used and the predicate 869 // will simply fail forcing recompilation. 870 Node* neg_stride = phase->longcon(-stride_con); 871 span = phase->transform(new (phase->C, 3) AndLNode(bias, neg_stride)); 872 } else { 873 Node *trip = phase->transform(new (phase->C, 3) DivLNode(0, bias, stride)); 874 span = phase->transform(new (phase->C, 3) MulLNode(trip, stride)); 875 } 876 // Convert back to int 877 Node *span_int = phase->transform(new (phase->C, 2) ConvL2INode(span)); 878 return new (phase->C, 3) AddINode(span_int, in(Init)); // exact limit 879 } 880 881 return NULL; // No progress 882 } 883 884 //------------------------------Identity--------------------------------------- 885 // If stride == 1 return limit node. 886 Node *LoopLimitNode::Identity( PhaseTransform *phase ) { 887 int stride_con = phase->type(in(Stride))->is_int()->get_con(); 888 if (stride_con == 1 || stride_con == -1) 889 return in(Limit); 890 return this; 891 } 892 893 //============================================================================= 894 //----------------------match_incr_with_optional_truncation-------------------- 895 // Match increment with optional truncation: 896 // CHAR: (i+1)&0x7fff, BYTE: ((i+1)<<8)>>8, or SHORT: ((i+1)<<16)>>16 897 // Return NULL for failure. Success returns the increment node. 898 Node* CountedLoopNode::match_incr_with_optional_truncation( 899 Node* expr, Node** trunc1, Node** trunc2, const TypeInt** trunc_type) { 900 // Quick cutouts: 901 if (expr == NULL || expr->req() != 3) return NULL; 902 903 Node *t1 = NULL; 904 Node *t2 = NULL; 905 const TypeInt* trunc_t = TypeInt::INT; 906 Node* n1 = expr; 907 int n1op = n1->Opcode(); 908 909 // Try to strip (n1 & M) or (n1 << N >> N) from n1. 910 if (n1op == Op_AndI && 911 n1->in(2)->is_Con() && 912 n1->in(2)->bottom_type()->is_int()->get_con() == 0x7fff) { 913 // %%% This check should match any mask of 2**K-1. 914 t1 = n1; 915 n1 = t1->in(1); 916 n1op = n1->Opcode(); 917 trunc_t = TypeInt::CHAR; 918 } else if (n1op == Op_RShiftI && 919 n1->in(1) != NULL && 920 n1->in(1)->Opcode() == Op_LShiftI && 921 n1->in(2) == n1->in(1)->in(2) && 922 n1->in(2)->is_Con()) { 923 jint shift = n1->in(2)->bottom_type()->is_int()->get_con(); 924 // %%% This check should match any shift in [1..31]. 925 if (shift == 16 || shift == 8) { 926 t1 = n1; 927 t2 = t1->in(1); 928 n1 = t2->in(1); 929 n1op = n1->Opcode(); 930 if (shift == 16) { 931 trunc_t = TypeInt::SHORT; 932 } else if (shift == 8) { 933 trunc_t = TypeInt::BYTE; 934 } 935 } 936 } 937 938 // If (maybe after stripping) it is an AddI, we won: 939 if (n1op == Op_AddI) { 940 *trunc1 = t1; 941 *trunc2 = t2; 942 *trunc_type = trunc_t; 943 return n1; 944 } 945 946 // failed 947 return NULL; 948 } 949 950 951 //------------------------------filtered_type-------------------------------- 952 // Return a type based on condition control flow 953 // A successful return will be a type that is restricted due 954 // to a series of dominating if-tests, such as: 955 // if (i < 10) { 956 // if (i > 0) { 957 // here: "i" type is [1..10) 958 // } 959 // } 960 // or a control flow merge 961 // if (i < 10) { 962 // do { 963 // phi( , ) -- at top of loop type is [min_int..10) 964 // i = ? 965 // } while ( i < 10) 966 // 967 const TypeInt* PhaseIdealLoop::filtered_type( Node *n, Node* n_ctrl) { 968 assert(n && n->bottom_type()->is_int(), "must be int"); 969 const TypeInt* filtered_t = NULL; 970 if (!n->is_Phi()) { 971 assert(n_ctrl != NULL || n_ctrl == C->top(), "valid control"); 972 filtered_t = filtered_type_from_dominators(n, n_ctrl); 973 974 } else { 975 Node* phi = n->as_Phi(); 976 Node* region = phi->in(0); 977 assert(n_ctrl == NULL || n_ctrl == region, "ctrl parameter must be region"); 978 if (region && region != C->top()) { 979 for (uint i = 1; i < phi->req(); i++) { 980 Node* val = phi->in(i); 981 Node* use_c = region->in(i); 982 const TypeInt* val_t = filtered_type_from_dominators(val, use_c); 983 if (val_t != NULL) { 984 if (filtered_t == NULL) { 985 filtered_t = val_t; 986 } else { 987 filtered_t = filtered_t->meet(val_t)->is_int(); 988 } 989 } 990 } 991 } 992 } 993 const TypeInt* n_t = _igvn.type(n)->is_int(); 994 if (filtered_t != NULL) { 995 n_t = n_t->join(filtered_t)->is_int(); 996 } 997 return n_t; 998 } 999 1000 1001 //------------------------------filtered_type_from_dominators-------------------------------- 1002 // Return a possibly more restrictive type for val based on condition control flow of dominators 1003 const TypeInt* PhaseIdealLoop::filtered_type_from_dominators( Node* val, Node *use_ctrl) { 1004 if (val->is_Con()) { 1005 return val->bottom_type()->is_int(); 1006 } 1007 uint if_limit = 10; // Max number of dominating if's visited 1008 const TypeInt* rtn_t = NULL; 1009 1010 if (use_ctrl && use_ctrl != C->top()) { 1011 Node* val_ctrl = get_ctrl(val); 1012 uint val_dom_depth = dom_depth(val_ctrl); 1013 Node* pred = use_ctrl; 1014 uint if_cnt = 0; 1015 while (if_cnt < if_limit) { 1016 if ((pred->Opcode() == Op_IfTrue || pred->Opcode() == Op_IfFalse)) { 1017 if_cnt++; 1018 const TypeInt* if_t = IfNode::filtered_int_type(&_igvn, val, pred); 1019 if (if_t != NULL) { 1020 if (rtn_t == NULL) { 1021 rtn_t = if_t; 1022 } else { 1023 rtn_t = rtn_t->join(if_t)->is_int(); 1024 } 1025 } 1026 } 1027 pred = idom(pred); 1028 if (pred == NULL || pred == C->top()) { 1029 break; 1030 } 1031 // Stop if going beyond definition block of val 1032 if (dom_depth(pred) < val_dom_depth) { 1033 break; 1034 } 1035 } 1036 } 1037 return rtn_t; 1038 } 1039 1040 1041 //------------------------------dump_spec-------------------------------------- 1042 // Dump special per-node info 1043 #ifndef PRODUCT 1044 void CountedLoopEndNode::dump_spec(outputStream *st) const { 1045 if( in(TestValue)->is_Bool() ) { 1046 BoolTest bt( test_trip()); // Added this for g++. 1047 1048 st->print("["); 1049 bt.dump_on(st); 1050 st->print("]"); 1051 } 1052 st->print(" "); 1053 IfNode::dump_spec(st); 1054 } 1055 #endif 1056 1057 //============================================================================= 1058 //------------------------------is_member-------------------------------------- 1059 // Is 'l' a member of 'this'? 1060 int IdealLoopTree::is_member( const IdealLoopTree *l ) const { 1061 while( l->_nest > _nest ) l = l->_parent; 1062 return l == this; 1063 } 1064 1065 //------------------------------set_nest--------------------------------------- 1066 // Set loop tree nesting depth. Accumulate _has_call bits. 1067 int IdealLoopTree::set_nest( uint depth ) { 1068 _nest = depth; 1069 int bits = _has_call; 1070 if( _child ) bits |= _child->set_nest(depth+1); 1071 if( bits ) _has_call = 1; 1072 if( _next ) bits |= _next ->set_nest(depth ); 1073 return bits; 1074 } 1075 1076 //------------------------------split_fall_in---------------------------------- 1077 // Split out multiple fall-in edges from the loop header. Move them to a 1078 // private RegionNode before the loop. This becomes the loop landing pad. 1079 void IdealLoopTree::split_fall_in( PhaseIdealLoop *phase, int fall_in_cnt ) { 1080 PhaseIterGVN &igvn = phase->_igvn; 1081 uint i; 1082 1083 // Make a new RegionNode to be the landing pad. 1084 Node *landing_pad = new (phase->C, fall_in_cnt+1) RegionNode( fall_in_cnt+1 ); 1085 phase->set_loop(landing_pad,_parent); 1086 // Gather all the fall-in control paths into the landing pad 1087 uint icnt = fall_in_cnt; 1088 uint oreq = _head->req(); 1089 for( i = oreq-1; i>0; i-- ) 1090 if( !phase->is_member( this, _head->in(i) ) ) 1091 landing_pad->set_req(icnt--,_head->in(i)); 1092 1093 // Peel off PhiNode edges as well 1094 for (DUIterator_Fast jmax, j = _head->fast_outs(jmax); j < jmax; j++) { 1095 Node *oj = _head->fast_out(j); 1096 if( oj->is_Phi() ) { 1097 PhiNode* old_phi = oj->as_Phi(); 1098 assert( old_phi->region() == _head, "" ); 1099 igvn.hash_delete(old_phi); // Yank from hash before hacking edges 1100 Node *p = PhiNode::make_blank(landing_pad, old_phi); 1101 uint icnt = fall_in_cnt; 1102 for( i = oreq-1; i>0; i-- ) { 1103 if( !phase->is_member( this, _head->in(i) ) ) { 1104 p->init_req(icnt--, old_phi->in(i)); 1105 // Go ahead and clean out old edges from old phi 1106 old_phi->del_req(i); 1107 } 1108 } 1109 // Search for CSE's here, because ZKM.jar does a lot of 1110 // loop hackery and we need to be a little incremental 1111 // with the CSE to avoid O(N^2) node blow-up. 1112 Node *p2 = igvn.hash_find_insert(p); // Look for a CSE 1113 if( p2 ) { // Found CSE 1114 p->destruct(); // Recover useless new node 1115 p = p2; // Use old node 1116 } else { 1117 igvn.register_new_node_with_optimizer(p, old_phi); 1118 } 1119 // Make old Phi refer to new Phi. 1120 old_phi->add_req(p); 1121 // Check for the special case of making the old phi useless and 1122 // disappear it. In JavaGrande I have a case where this useless 1123 // Phi is the loop limit and prevents recognizing a CountedLoop 1124 // which in turn prevents removing an empty loop. 1125 Node *id_old_phi = old_phi->Identity( &igvn ); 1126 if( id_old_phi != old_phi ) { // Found a simple identity? 1127 // Note that I cannot call 'replace_node' here, because 1128 // that will yank the edge from old_phi to the Region and 1129 // I'm mid-iteration over the Region's uses. 1130 for (DUIterator_Last imin, i = old_phi->last_outs(imin); i >= imin; ) { 1131 Node* use = old_phi->last_out(i); 1132 igvn.hash_delete(use); 1133 igvn._worklist.push(use); 1134 uint uses_found = 0; 1135 for (uint j = 0; j < use->len(); j++) { 1136 if (use->in(j) == old_phi) { 1137 if (j < use->req()) use->set_req (j, id_old_phi); 1138 else use->set_prec(j, id_old_phi); 1139 uses_found++; 1140 } 1141 } 1142 i -= uses_found; // we deleted 1 or more copies of this edge 1143 } 1144 } 1145 igvn._worklist.push(old_phi); 1146 } 1147 } 1148 // Finally clean out the fall-in edges from the RegionNode 1149 for( i = oreq-1; i>0; i-- ) { 1150 if( !phase->is_member( this, _head->in(i) ) ) { 1151 _head->del_req(i); 1152 } 1153 } 1154 // Transform landing pad 1155 igvn.register_new_node_with_optimizer(landing_pad, _head); 1156 // Insert landing pad into the header 1157 _head->add_req(landing_pad); 1158 } 1159 1160 //------------------------------split_outer_loop------------------------------- 1161 // Split out the outermost loop from this shared header. 1162 void IdealLoopTree::split_outer_loop( PhaseIdealLoop *phase ) { 1163 PhaseIterGVN &igvn = phase->_igvn; 1164 1165 // Find index of outermost loop; it should also be my tail. 1166 uint outer_idx = 1; 1167 while( _head->in(outer_idx) != _tail ) outer_idx++; 1168 1169 // Make a LoopNode for the outermost loop. 1170 Node *ctl = _head->in(LoopNode::EntryControl); 1171 Node *outer = new (phase->C, 3) LoopNode( ctl, _head->in(outer_idx) ); 1172 outer = igvn.register_new_node_with_optimizer(outer, _head); 1173 phase->set_created_loop_node(); 1174 1175 // Outermost loop falls into '_head' loop 1176 _head->set_req(LoopNode::EntryControl, outer); 1177 _head->del_req(outer_idx); 1178 // Split all the Phis up between '_head' loop and 'outer' loop. 1179 for (DUIterator_Fast jmax, j = _head->fast_outs(jmax); j < jmax; j++) { 1180 Node *out = _head->fast_out(j); 1181 if( out->is_Phi() ) { 1182 PhiNode *old_phi = out->as_Phi(); 1183 assert( old_phi->region() == _head, "" ); 1184 Node *phi = PhiNode::make_blank(outer, old_phi); 1185 phi->init_req(LoopNode::EntryControl, old_phi->in(LoopNode::EntryControl)); 1186 phi->init_req(LoopNode::LoopBackControl, old_phi->in(outer_idx)); 1187 phi = igvn.register_new_node_with_optimizer(phi, old_phi); 1188 // Make old Phi point to new Phi on the fall-in path 1189 igvn.hash_delete(old_phi); 1190 old_phi->set_req(LoopNode::EntryControl, phi); 1191 old_phi->del_req(outer_idx); 1192 igvn._worklist.push(old_phi); 1193 } 1194 } 1195 1196 // Use the new loop head instead of the old shared one 1197 _head = outer; 1198 phase->set_loop(_head, this); 1199 } 1200 1201 //------------------------------fix_parent------------------------------------- 1202 static void fix_parent( IdealLoopTree *loop, IdealLoopTree *parent ) { 1203 loop->_parent = parent; 1204 if( loop->_child ) fix_parent( loop->_child, loop ); 1205 if( loop->_next ) fix_parent( loop->_next , parent ); 1206 } 1207 1208 //------------------------------estimate_path_freq----------------------------- 1209 static float estimate_path_freq( Node *n ) { 1210 // Try to extract some path frequency info 1211 IfNode *iff; 1212 for( int i = 0; i < 50; i++ ) { // Skip through a bunch of uncommon tests 1213 uint nop = n->Opcode(); 1214 if( nop == Op_SafePoint ) { // Skip any safepoint 1215 n = n->in(0); 1216 continue; 1217 } 1218 if( nop == Op_CatchProj ) { // Get count from a prior call 1219 // Assume call does not always throw exceptions: means the call-site 1220 // count is also the frequency of the fall-through path. 1221 assert( n->is_CatchProj(), "" ); 1222 if( ((CatchProjNode*)n)->_con != CatchProjNode::fall_through_index ) 1223 return 0.0f; // Assume call exception path is rare 1224 Node *call = n->in(0)->in(0)->in(0); 1225 assert( call->is_Call(), "expect a call here" ); 1226 const JVMState *jvms = ((CallNode*)call)->jvms(); 1227 ciMethodData* methodData = jvms->method()->method_data(); 1228 if (!methodData->is_mature()) return 0.0f; // No call-site data 1229 ciProfileData* data = methodData->bci_to_data(jvms->bci()); 1230 if ((data == NULL) || !data->is_CounterData()) { 1231 // no call profile available, try call's control input 1232 n = n->in(0); 1233 continue; 1234 } 1235 return data->as_CounterData()->count()/FreqCountInvocations; 1236 } 1237 // See if there's a gating IF test 1238 Node *n_c = n->in(0); 1239 if( !n_c->is_If() ) break; // No estimate available 1240 iff = n_c->as_If(); 1241 if( iff->_fcnt != COUNT_UNKNOWN ) // Have a valid count? 1242 // Compute how much count comes on this path 1243 return ((nop == Op_IfTrue) ? iff->_prob : 1.0f - iff->_prob) * iff->_fcnt; 1244 // Have no count info. Skip dull uncommon-trap like branches. 1245 if( (nop == Op_IfTrue && iff->_prob < PROB_LIKELY_MAG(5)) || 1246 (nop == Op_IfFalse && iff->_prob > PROB_UNLIKELY_MAG(5)) ) 1247 break; 1248 // Skip through never-taken branch; look for a real loop exit. 1249 n = iff->in(0); 1250 } 1251 return 0.0f; // No estimate available 1252 } 1253 1254 //------------------------------merge_many_backedges--------------------------- 1255 // Merge all the backedges from the shared header into a private Region. 1256 // Feed that region as the one backedge to this loop. 1257 void IdealLoopTree::merge_many_backedges( PhaseIdealLoop *phase ) { 1258 uint i; 1259 1260 // Scan for the top 2 hottest backedges 1261 float hotcnt = 0.0f; 1262 float warmcnt = 0.0f; 1263 uint hot_idx = 0; 1264 // Loop starts at 2 because slot 1 is the fall-in path 1265 for( i = 2; i < _head->req(); i++ ) { 1266 float cnt = estimate_path_freq(_head->in(i)); 1267 if( cnt > hotcnt ) { // Grab hottest path 1268 warmcnt = hotcnt; 1269 hotcnt = cnt; 1270 hot_idx = i; 1271 } else if( cnt > warmcnt ) { // And 2nd hottest path 1272 warmcnt = cnt; 1273 } 1274 } 1275 1276 // See if the hottest backedge is worthy of being an inner loop 1277 // by being much hotter than the next hottest backedge. 1278 if( hotcnt <= 0.0001 || 1279 hotcnt < 2.0*warmcnt ) hot_idx = 0;// No hot backedge 1280 1281 // Peel out the backedges into a private merge point; peel 1282 // them all except optionally hot_idx. 1283 PhaseIterGVN &igvn = phase->_igvn; 1284 1285 Node *hot_tail = NULL; 1286 // Make a Region for the merge point 1287 Node *r = new (phase->C, 1) RegionNode(1); 1288 for( i = 2; i < _head->req(); i++ ) { 1289 if( i != hot_idx ) 1290 r->add_req( _head->in(i) ); 1291 else hot_tail = _head->in(i); 1292 } 1293 igvn.register_new_node_with_optimizer(r, _head); 1294 // Plug region into end of loop _head, followed by hot_tail 1295 while( _head->req() > 3 ) _head->del_req( _head->req()-1 ); 1296 _head->set_req(2, r); 1297 if( hot_idx ) _head->add_req(hot_tail); 1298 1299 // Split all the Phis up between '_head' loop and the Region 'r' 1300 for (DUIterator_Fast jmax, j = _head->fast_outs(jmax); j < jmax; j++) { 1301 Node *out = _head->fast_out(j); 1302 if( out->is_Phi() ) { 1303 PhiNode* n = out->as_Phi(); 1304 igvn.hash_delete(n); // Delete from hash before hacking edges 1305 Node *hot_phi = NULL; 1306 Node *phi = new (phase->C, r->req()) PhiNode(r, n->type(), n->adr_type()); 1307 // Check all inputs for the ones to peel out 1308 uint j = 1; 1309 for( uint i = 2; i < n->req(); i++ ) { 1310 if( i != hot_idx ) 1311 phi->set_req( j++, n->in(i) ); 1312 else hot_phi = n->in(i); 1313 } 1314 // Register the phi but do not transform until whole place transforms 1315 igvn.register_new_node_with_optimizer(phi, n); 1316 // Add the merge phi to the old Phi 1317 while( n->req() > 3 ) n->del_req( n->req()-1 ); 1318 n->set_req(2, phi); 1319 if( hot_idx ) n->add_req(hot_phi); 1320 } 1321 } 1322 1323 1324 // Insert a new IdealLoopTree inserted below me. Turn it into a clone 1325 // of self loop tree. Turn self into a loop headed by _head and with 1326 // tail being the new merge point. 1327 IdealLoopTree *ilt = new IdealLoopTree( phase, _head, _tail ); 1328 phase->set_loop(_tail,ilt); // Adjust tail 1329 _tail = r; // Self's tail is new merge point 1330 phase->set_loop(r,this); 1331 ilt->_child = _child; // New guy has my children 1332 _child = ilt; // Self has new guy as only child 1333 ilt->_parent = this; // new guy has self for parent 1334 ilt->_nest = _nest; // Same nesting depth (for now) 1335 1336 // Starting with 'ilt', look for child loop trees using the same shared 1337 // header. Flatten these out; they will no longer be loops in the end. 1338 IdealLoopTree **pilt = &_child; 1339 while( ilt ) { 1340 if( ilt->_head == _head ) { 1341 uint i; 1342 for( i = 2; i < _head->req(); i++ ) 1343 if( _head->in(i) == ilt->_tail ) 1344 break; // Still a loop 1345 if( i == _head->req() ) { // No longer a loop 1346 // Flatten ilt. Hang ilt's "_next" list from the end of 1347 // ilt's '_child' list. Move the ilt's _child up to replace ilt. 1348 IdealLoopTree **cp = &ilt->_child; 1349 while( *cp ) cp = &(*cp)->_next; // Find end of child list 1350 *cp = ilt->_next; // Hang next list at end of child list 1351 *pilt = ilt->_child; // Move child up to replace ilt 1352 ilt->_head = NULL; // Flag as a loop UNIONED into parent 1353 ilt = ilt->_child; // Repeat using new ilt 1354 continue; // do not advance over ilt->_child 1355 } 1356 assert( ilt->_tail == hot_tail, "expected to only find the hot inner loop here" ); 1357 phase->set_loop(_head,ilt); 1358 } 1359 pilt = &ilt->_child; // Advance to next 1360 ilt = *pilt; 1361 } 1362 1363 if( _child ) fix_parent( _child, this ); 1364 } 1365 1366 //------------------------------beautify_loops--------------------------------- 1367 // Split shared headers and insert loop landing pads. 1368 // Insert a LoopNode to replace the RegionNode. 1369 // Return TRUE if loop tree is structurally changed. 1370 bool IdealLoopTree::beautify_loops( PhaseIdealLoop *phase ) { 1371 bool result = false; 1372 // Cache parts in locals for easy 1373 PhaseIterGVN &igvn = phase->_igvn; 1374 1375 igvn.hash_delete(_head); // Yank from hash before hacking edges 1376 1377 // Check for multiple fall-in paths. Peel off a landing pad if need be. 1378 int fall_in_cnt = 0; 1379 for( uint i = 1; i < _head->req(); i++ ) 1380 if( !phase->is_member( this, _head->in(i) ) ) 1381 fall_in_cnt++; 1382 assert( fall_in_cnt, "at least 1 fall-in path" ); 1383 if( fall_in_cnt > 1 ) // Need a loop landing pad to merge fall-ins 1384 split_fall_in( phase, fall_in_cnt ); 1385 1386 // Swap inputs to the _head and all Phis to move the fall-in edge to 1387 // the left. 1388 fall_in_cnt = 1; 1389 while( phase->is_member( this, _head->in(fall_in_cnt) ) ) 1390 fall_in_cnt++; 1391 if( fall_in_cnt > 1 ) { 1392 // Since I am just swapping inputs I do not need to update def-use info 1393 Node *tmp = _head->in(1); 1394 _head->set_req( 1, _head->in(fall_in_cnt) ); 1395 _head->set_req( fall_in_cnt, tmp ); 1396 // Swap also all Phis 1397 for (DUIterator_Fast imax, i = _head->fast_outs(imax); i < imax; i++) { 1398 Node* phi = _head->fast_out(i); 1399 if( phi->is_Phi() ) { 1400 igvn.hash_delete(phi); // Yank from hash before hacking edges 1401 tmp = phi->in(1); 1402 phi->set_req( 1, phi->in(fall_in_cnt) ); 1403 phi->set_req( fall_in_cnt, tmp ); 1404 } 1405 } 1406 } 1407 assert( !phase->is_member( this, _head->in(1) ), "left edge is fall-in" ); 1408 assert( phase->is_member( this, _head->in(2) ), "right edge is loop" ); 1409 1410 // If I am a shared header (multiple backedges), peel off the many 1411 // backedges into a private merge point and use the merge point as 1412 // the one true backedge. 1413 if( _head->req() > 3 ) { 1414 // Merge the many backedges into a single backedge but leave 1415 // the hottest backedge as separate edge for the following peel. 1416 merge_many_backedges( phase ); 1417 result = true; 1418 } 1419 1420 // If I have one hot backedge, peel off myself loop. 1421 // I better be the outermost loop. 1422 if( _head->req() > 3 ) { 1423 split_outer_loop( phase ); 1424 result = true; 1425 1426 } else if( !_head->is_Loop() && !_irreducible ) { 1427 // Make a new LoopNode to replace the old loop head 1428 Node *l = new (phase->C, 3) LoopNode( _head->in(1), _head->in(2) ); 1429 l = igvn.register_new_node_with_optimizer(l, _head); 1430 phase->set_created_loop_node(); 1431 // Go ahead and replace _head 1432 phase->_igvn.replace_node( _head, l ); 1433 _head = l; 1434 phase->set_loop(_head, this); 1435 } 1436 1437 // Now recursively beautify nested loops 1438 if( _child ) result |= _child->beautify_loops( phase ); 1439 if( _next ) result |= _next ->beautify_loops( phase ); 1440 return result; 1441 } 1442 1443 //------------------------------allpaths_check_safepts---------------------------- 1444 // Allpaths backwards scan from loop tail, terminating each path at first safepoint 1445 // encountered. Helper for check_safepts. 1446 void IdealLoopTree::allpaths_check_safepts(VectorSet &visited, Node_List &stack) { 1447 assert(stack.size() == 0, "empty stack"); 1448 stack.push(_tail); 1449 visited.Clear(); 1450 visited.set(_tail->_idx); 1451 while (stack.size() > 0) { 1452 Node* n = stack.pop(); 1453 if (n->is_Call() && n->as_Call()->guaranteed_safepoint()) { 1454 // Terminate this path 1455 } else if (n->Opcode() == Op_SafePoint) { 1456 if (_phase->get_loop(n) != this) { 1457 if (_required_safept == NULL) _required_safept = new Node_List(); 1458 _required_safept->push(n); // save the one closest to the tail 1459 } 1460 // Terminate this path 1461 } else { 1462 uint start = n->is_Region() ? 1 : 0; 1463 uint end = n->is_Region() && !n->is_Loop() ? n->req() : start + 1; 1464 for (uint i = start; i < end; i++) { 1465 Node* in = n->in(i); 1466 assert(in->is_CFG(), "must be"); 1467 if (!visited.test_set(in->_idx) && is_member(_phase->get_loop(in))) { 1468 stack.push(in); 1469 } 1470 } 1471 } 1472 } 1473 } 1474 1475 //------------------------------check_safepts---------------------------- 1476 // Given dominators, try to find loops with calls that must always be 1477 // executed (call dominates loop tail). These loops do not need non-call 1478 // safepoints (ncsfpt). 1479 // 1480 // A complication is that a safepoint in a inner loop may be needed 1481 // by an outer loop. In the following, the inner loop sees it has a 1482 // call (block 3) on every path from the head (block 2) to the 1483 // backedge (arc 3->2). So it deletes the ncsfpt (non-call safepoint) 1484 // in block 2, _but_ this leaves the outer loop without a safepoint. 1485 // 1486 // entry 0 1487 // | 1488 // v 1489 // outer 1,2 +->1 1490 // | | 1491 // | v 1492 // | 2<---+ ncsfpt in 2 1493 // |_/|\ | 1494 // | v | 1495 // inner 2,3 / 3 | call in 3 1496 // / | | 1497 // v +--+ 1498 // exit 4 1499 // 1500 // 1501 // This method creates a list (_required_safept) of ncsfpt nodes that must 1502 // be protected is created for each loop. When a ncsfpt maybe deleted, it 1503 // is first looked for in the lists for the outer loops of the current loop. 1504 // 1505 // The insights into the problem: 1506 // A) counted loops are okay 1507 // B) innermost loops are okay (only an inner loop can delete 1508 // a ncsfpt needed by an outer loop) 1509 // C) a loop is immune from an inner loop deleting a safepoint 1510 // if the loop has a call on the idom-path 1511 // D) a loop is also immune if it has a ncsfpt (non-call safepoint) on the 1512 // idom-path that is not in a nested loop 1513 // E) otherwise, an ncsfpt on the idom-path that is nested in an inner 1514 // loop needs to be prevented from deletion by an inner loop 1515 // 1516 // There are two analyses: 1517 // 1) The first, and cheaper one, scans the loop body from 1518 // tail to head following the idom (immediate dominator) 1519 // chain, looking for the cases (C,D,E) above. 1520 // Since inner loops are scanned before outer loops, there is summary 1521 // information about inner loops. Inner loops can be skipped over 1522 // when the tail of an inner loop is encountered. 1523 // 1524 // 2) The second, invoked if the first fails to find a call or ncsfpt on 1525 // the idom path (which is rare), scans all predecessor control paths 1526 // from the tail to the head, terminating a path when a call or sfpt 1527 // is encountered, to find the ncsfpt's that are closest to the tail. 1528 // 1529 void IdealLoopTree::check_safepts(VectorSet &visited, Node_List &stack) { 1530 // Bottom up traversal 1531 IdealLoopTree* ch = _child; 1532 while (ch != NULL) { 1533 ch->check_safepts(visited, stack); 1534 ch = ch->_next; 1535 } 1536 1537 if (!_head->is_CountedLoop() && !_has_sfpt && _parent != NULL && !_irreducible) { 1538 bool has_call = false; // call on dom-path 1539 bool has_local_ncsfpt = false; // ncsfpt on dom-path at this loop depth 1540 Node* nonlocal_ncsfpt = NULL; // ncsfpt on dom-path at a deeper depth 1541 // Scan the dom-path nodes from tail to head 1542 for (Node* n = tail(); n != _head; n = _phase->idom(n)) { 1543 if (n->is_Call() && n->as_Call()->guaranteed_safepoint()) { 1544 has_call = true; 1545 _has_sfpt = 1; // Then no need for a safept! 1546 break; 1547 } else if (n->Opcode() == Op_SafePoint) { 1548 if (_phase->get_loop(n) == this) { 1549 has_local_ncsfpt = true; 1550 break; 1551 } 1552 if (nonlocal_ncsfpt == NULL) { 1553 nonlocal_ncsfpt = n; // save the one closest to the tail 1554 } 1555 } else { 1556 IdealLoopTree* nlpt = _phase->get_loop(n); 1557 if (this != nlpt) { 1558 // If at an inner loop tail, see if the inner loop has already 1559 // recorded seeing a call on the dom-path (and stop.) If not, 1560 // jump to the head of the inner loop. 1561 assert(is_member(nlpt), "nested loop"); 1562 Node* tail = nlpt->_tail; 1563 if (tail->in(0)->is_If()) tail = tail->in(0); 1564 if (n == tail) { 1565 // If inner loop has call on dom-path, so does outer loop 1566 if (nlpt->_has_sfpt) { 1567 has_call = true; 1568 _has_sfpt = 1; 1569 break; 1570 } 1571 // Skip to head of inner loop 1572 assert(_phase->is_dominator(_head, nlpt->_head), "inner head dominated by outer head"); 1573 n = nlpt->_head; 1574 } 1575 } 1576 } 1577 } 1578 // Record safept's that this loop needs preserved when an 1579 // inner loop attempts to delete it's safepoints. 1580 if (_child != NULL && !has_call && !has_local_ncsfpt) { 1581 if (nonlocal_ncsfpt != NULL) { 1582 if (_required_safept == NULL) _required_safept = new Node_List(); 1583 _required_safept->push(nonlocal_ncsfpt); 1584 } else { 1585 // Failed to find a suitable safept on the dom-path. Now use 1586 // an all paths walk from tail to head, looking for safepoints to preserve. 1587 allpaths_check_safepts(visited, stack); 1588 } 1589 } 1590 } 1591 } 1592 1593 //---------------------------is_deleteable_safept---------------------------- 1594 // Is safept not required by an outer loop? 1595 bool PhaseIdealLoop::is_deleteable_safept(Node* sfpt) { 1596 assert(sfpt->Opcode() == Op_SafePoint, ""); 1597 IdealLoopTree* lp = get_loop(sfpt)->_parent; 1598 while (lp != NULL) { 1599 Node_List* sfpts = lp->_required_safept; 1600 if (sfpts != NULL) { 1601 for (uint i = 0; i < sfpts->size(); i++) { 1602 if (sfpt == sfpts->at(i)) 1603 return false; 1604 } 1605 } 1606 lp = lp->_parent; 1607 } 1608 return true; 1609 } 1610 1611 //---------------------------replace_parallel_iv------------------------------- 1612 // Replace parallel induction variable (parallel to trip counter) 1613 void PhaseIdealLoop::replace_parallel_iv(IdealLoopTree *loop) { 1614 assert(loop->_head->is_CountedLoop(), ""); 1615 CountedLoopNode *cl = loop->_head->as_CountedLoop(); 1616 if (!cl->is_valid_counted_loop()) 1617 return; // skip malformed counted loop 1618 Node *incr = cl->incr(); 1619 if (incr == NULL) 1620 return; // Dead loop? 1621 Node *init = cl->init_trip(); 1622 Node *phi = cl->phi(); 1623 int stride_con = cl->stride_con(); 1624 1625 // Visit all children, looking for Phis 1626 for (DUIterator i = cl->outs(); cl->has_out(i); i++) { 1627 Node *out = cl->out(i); 1628 // Look for other phis (secondary IVs). Skip dead ones 1629 if (!out->is_Phi() || out == phi || !has_node(out)) 1630 continue; 1631 PhiNode* phi2 = out->as_Phi(); 1632 Node *incr2 = phi2->in( LoopNode::LoopBackControl ); 1633 // Look for induction variables of the form: X += constant 1634 if (phi2->region() != loop->_head || 1635 incr2->req() != 3 || 1636 incr2->in(1) != phi2 || 1637 incr2 == incr || 1638 incr2->Opcode() != Op_AddI || 1639 !incr2->in(2)->is_Con()) 1640 continue; 1641 1642 // Check for parallel induction variable (parallel to trip counter) 1643 // via an affine function. In particular, count-down loops with 1644 // count-up array indices are common. We only RCE references off 1645 // the trip-counter, so we need to convert all these to trip-counter 1646 // expressions. 1647 Node *init2 = phi2->in( LoopNode::EntryControl ); 1648 int stride_con2 = incr2->in(2)->get_int(); 1649 1650 // The general case here gets a little tricky. We want to find the 1651 // GCD of all possible parallel IV's and make a new IV using this 1652 // GCD for the loop. Then all possible IVs are simple multiples of 1653 // the GCD. In practice, this will cover very few extra loops. 1654 // Instead we require 'stride_con2' to be a multiple of 'stride_con', 1655 // where +/-1 is the common case, but other integer multiples are 1656 // also easy to handle. 1657 int ratio_con = stride_con2/stride_con; 1658 1659 if ((ratio_con * stride_con) == stride_con2) { // Check for exact 1660 #ifndef PRODUCT 1661 if (TraceLoopOpts) { 1662 tty->print("Parallel IV: %d ", phi2->_idx); 1663 loop->dump_head(); 1664 } 1665 #endif 1666 // Convert to using the trip counter. The parallel induction 1667 // variable differs from the trip counter by a loop-invariant 1668 // amount, the difference between their respective initial values. 1669 // It is scaled by the 'ratio_con'. 1670 Node* ratio = _igvn.intcon(ratio_con); 1671 set_ctrl(ratio, C->root()); 1672 Node* ratio_init = new (C, 3) MulINode(init, ratio); 1673 _igvn.register_new_node_with_optimizer(ratio_init, init); 1674 set_early_ctrl(ratio_init); 1675 Node* diff = new (C, 3) SubINode(init2, ratio_init); 1676 _igvn.register_new_node_with_optimizer(diff, init2); 1677 set_early_ctrl(diff); 1678 Node* ratio_idx = new (C, 3) MulINode(phi, ratio); 1679 _igvn.register_new_node_with_optimizer(ratio_idx, phi); 1680 set_ctrl(ratio_idx, cl); 1681 Node* add = new (C, 3) AddINode(ratio_idx, diff); 1682 _igvn.register_new_node_with_optimizer(add); 1683 set_ctrl(add, cl); 1684 _igvn.replace_node( phi2, add ); 1685 // Sometimes an induction variable is unused 1686 if (add->outcnt() == 0) { 1687 _igvn.remove_dead_node(add); 1688 } 1689 --i; // deleted this phi; rescan starting with next position 1690 continue; 1691 } 1692 } 1693 } 1694 1695 //------------------------------counted_loop----------------------------------- 1696 // Convert to counted loops where possible 1697 void IdealLoopTree::counted_loop( PhaseIdealLoop *phase ) { 1698 1699 // For grins, set the inner-loop flag here 1700 if (!_child) { 1701 if (_head->is_Loop()) _head->as_Loop()->set_inner_loop(); 1702 } 1703 1704 if (_head->is_CountedLoop() || 1705 phase->is_counted_loop(_head, this)) { 1706 _has_sfpt = 1; // Indicate we do not need a safepoint here 1707 1708 // Look for a safepoint to remove 1709 for (Node* n = tail(); n != _head; n = phase->idom(n)) 1710 if (n->Opcode() == Op_SafePoint && phase->get_loop(n) == this && 1711 phase->is_deleteable_safept(n)) 1712 phase->lazy_replace(n,n->in(TypeFunc::Control)); 1713 1714 // Look for induction variables 1715 phase->replace_parallel_iv(this); 1716 1717 } else if (_parent != NULL && !_irreducible) { 1718 // Not a counted loop. 1719 // Look for a safepoint on the idom-path to remove, preserving the first one 1720 bool found = false; 1721 Node* n = tail(); 1722 for (; n != _head && !found; n = phase->idom(n)) { 1723 if (n->Opcode() == Op_SafePoint && phase->get_loop(n) == this) 1724 found = true; // Found one 1725 } 1726 // Skip past it and delete the others 1727 for (; n != _head; n = phase->idom(n)) { 1728 if (n->Opcode() == Op_SafePoint && phase->get_loop(n) == this && 1729 phase->is_deleteable_safept(n)) 1730 phase->lazy_replace(n,n->in(TypeFunc::Control)); 1731 } 1732 } 1733 1734 // Recursively 1735 if (_child) _child->counted_loop( phase ); 1736 if (_next) _next ->counted_loop( phase ); 1737 } 1738 1739 #ifndef PRODUCT 1740 //------------------------------dump_head-------------------------------------- 1741 // Dump 1 liner for loop header info 1742 void IdealLoopTree::dump_head( ) const { 1743 for (uint i=0; i<_nest; i++) 1744 tty->print(" "); 1745 tty->print("Loop: N%d/N%d ",_head->_idx,_tail->_idx); 1746 if (_irreducible) tty->print(" IRREDUCIBLE"); 1747 Node* entry = _head->in(LoopNode::EntryControl); 1748 if (LoopLimitCheck) { 1749 Node* predicate = PhaseIdealLoop::find_predicate_insertion_point(entry, Deoptimization::Reason_loop_limit_check); 1750 if (predicate != NULL ) { 1751 tty->print(" limit_check"); 1752 entry = entry->in(0)->in(0); 1753 } 1754 } 1755 if (UseLoopPredicate) { 1756 entry = PhaseIdealLoop::find_predicate_insertion_point(entry, Deoptimization::Reason_predicate); 1757 if (entry != NULL) { 1758 tty->print(" predicated"); 1759 } 1760 } 1761 if (_head->is_CountedLoop()) { 1762 CountedLoopNode *cl = _head->as_CountedLoop(); 1763 tty->print(" counted"); 1764 1765 Node* init_n = cl->init_trip(); 1766 if (init_n != NULL && init_n->is_Con()) 1767 tty->print(" [%d,", cl->init_trip()->get_int()); 1768 else 1769 tty->print(" [int,"); 1770 Node* limit_n = cl->limit(); 1771 if (limit_n != NULL && limit_n->is_Con()) 1772 tty->print("%d),", cl->limit()->get_int()); 1773 else 1774 tty->print("int),"); 1775 int stride_con = cl->stride_con(); 1776 if (stride_con > 0) tty->print("+"); 1777 tty->print("%d", stride_con); 1778 1779 if (cl->is_pre_loop ()) tty->print(" pre" ); 1780 if (cl->is_main_loop()) tty->print(" main"); 1781 if (cl->is_post_loop()) tty->print(" post"); 1782 } 1783 tty->cr(); 1784 } 1785 1786 //------------------------------dump------------------------------------------- 1787 // Dump loops by loop tree 1788 void IdealLoopTree::dump( ) const { 1789 dump_head(); 1790 if (_child) _child->dump(); 1791 if (_next) _next ->dump(); 1792 } 1793 1794 #endif 1795 1796 static void log_loop_tree(IdealLoopTree* root, IdealLoopTree* loop, CompileLog* log) { 1797 if (loop == root) { 1798 if (loop->_child != NULL) { 1799 log->begin_head("loop_tree"); 1800 log->end_head(); 1801 if( loop->_child ) log_loop_tree(root, loop->_child, log); 1802 log->tail("loop_tree"); 1803 assert(loop->_next == NULL, "what?"); 1804 } 1805 } else { 1806 Node* head = loop->_head; 1807 log->begin_head("loop"); 1808 log->print(" idx='%d' ", head->_idx); 1809 if (loop->_irreducible) log->print("irreducible='1' "); 1810 if (head->is_Loop()) { 1811 if (head->as_Loop()->is_inner_loop()) log->print("inner_loop='1' "); 1812 if (head->as_Loop()->is_partial_peel_loop()) log->print("partial_peel_loop='1' "); 1813 } 1814 if (head->is_CountedLoop()) { 1815 CountedLoopNode* cl = head->as_CountedLoop(); 1816 if (cl->is_pre_loop()) log->print("pre_loop='%d' ", cl->main_idx()); 1817 if (cl->is_main_loop()) log->print("main_loop='%d' ", cl->_idx); 1818 if (cl->is_post_loop()) log->print("post_loop='%d' ", cl->main_idx()); 1819 } 1820 log->end_head(); 1821 if( loop->_child ) log_loop_tree(root, loop->_child, log); 1822 log->tail("loop"); 1823 if( loop->_next ) log_loop_tree(root, loop->_next, log); 1824 } 1825 } 1826 1827 //---------------------collect_potentially_useful_predicates----------------------- 1828 // Helper function to collect potentially useful predicates to prevent them from 1829 // being eliminated by PhaseIdealLoop::eliminate_useless_predicates 1830 void PhaseIdealLoop::collect_potentially_useful_predicates( 1831 IdealLoopTree * loop, Unique_Node_List &useful_predicates) { 1832 if (loop->_child) { // child 1833 collect_potentially_useful_predicates(loop->_child, useful_predicates); 1834 } 1835 1836 // self (only loops that we can apply loop predication may use their predicates) 1837 if (loop->_head->is_Loop() && 1838 !loop->_irreducible && 1839 !loop->tail()->is_top()) { 1840 LoopNode* lpn = loop->_head->as_Loop(); 1841 Node* entry = lpn->in(LoopNode::EntryControl); 1842 Node* predicate_proj = find_predicate(entry); // loop_limit_check first 1843 if (predicate_proj != NULL ) { // right pattern that can be used by loop predication 1844 assert(entry->in(0)->in(1)->in(1)->Opcode() == Op_Opaque1, "must be"); 1845 useful_predicates.push(entry->in(0)->in(1)->in(1)); // good one 1846 entry = entry->in(0)->in(0); 1847 } 1848 predicate_proj = find_predicate(entry); // Predicate 1849 if (predicate_proj != NULL ) { 1850 useful_predicates.push(entry->in(0)->in(1)->in(1)); // good one 1851 } 1852 } 1853 1854 if (loop->_next) { // sibling 1855 collect_potentially_useful_predicates(loop->_next, useful_predicates); 1856 } 1857 } 1858 1859 //------------------------eliminate_useless_predicates----------------------------- 1860 // Eliminate all inserted predicates if they could not be used by loop predication. 1861 // Note: it will also eliminates loop limits check predicate since it also uses 1862 // Opaque1 node (see Parse::add_predicate()). 1863 void PhaseIdealLoop::eliminate_useless_predicates() { 1864 if (C->predicate_count() == 0) 1865 return; // no predicate left 1866 1867 Unique_Node_List useful_predicates; // to store useful predicates 1868 if (C->has_loops()) { 1869 collect_potentially_useful_predicates(_ltree_root->_child, useful_predicates); 1870 } 1871 1872 for (int i = C->predicate_count(); i > 0; i--) { 1873 Node * n = C->predicate_opaque1_node(i-1); 1874 assert(n->Opcode() == Op_Opaque1, "must be"); 1875 if (!useful_predicates.member(n)) { // not in the useful list 1876 _igvn.replace_node(n, n->in(1)); 1877 } 1878 } 1879 } 1880 1881 //============================================================================= 1882 //----------------------------build_and_optimize------------------------------- 1883 // Create a PhaseLoop. Build the ideal Loop tree. Map each Ideal Node to 1884 // its corresponding LoopNode. If 'optimize' is true, do some loop cleanups. 1885 void PhaseIdealLoop::build_and_optimize(bool do_split_ifs, bool skip_loop_opts) { 1886 ResourceMark rm; 1887 1888 int old_progress = C->major_progress(); 1889 uint orig_worklist_size = _igvn._worklist.size(); 1890 1891 // Reset major-progress flag for the driver's heuristics 1892 C->clear_major_progress(); 1893 1894 #ifndef PRODUCT 1895 // Capture for later assert 1896 uint unique = C->unique(); 1897 _loop_invokes++; 1898 _loop_work += unique; 1899 #endif 1900 1901 // True if the method has at least 1 irreducible loop 1902 _has_irreducible_loops = false; 1903 1904 _created_loop_node = false; 1905 1906 Arena *a = Thread::current()->resource_area(); 1907 VectorSet visited(a); 1908 // Pre-grow the mapping from Nodes to IdealLoopTrees. 1909 _nodes.map(C->unique(), NULL); 1910 memset(_nodes.adr(), 0, wordSize * C->unique()); 1911 1912 // Pre-build the top-level outermost loop tree entry 1913 _ltree_root = new IdealLoopTree( this, C->root(), C->root() ); 1914 // Do not need a safepoint at the top level 1915 _ltree_root->_has_sfpt = 1; 1916 1917 // Initialize Dominators. 1918 // Checked in clone_loop_predicate() during beautify_loops(). 1919 _idom_size = 0; 1920 _idom = NULL; 1921 _dom_depth = NULL; 1922 _dom_stk = NULL; 1923 1924 // Empty pre-order array 1925 allocate_preorders(); 1926 1927 // Build a loop tree on the fly. Build a mapping from CFG nodes to 1928 // IdealLoopTree entries. Data nodes are NOT walked. 1929 build_loop_tree(); 1930 // Check for bailout, and return 1931 if (C->failing()) { 1932 return; 1933 } 1934 1935 // No loops after all 1936 if( !_ltree_root->_child && !_verify_only ) C->set_has_loops(false); 1937 1938 // There should always be an outer loop containing the Root and Return nodes. 1939 // If not, we have a degenerate empty program. Bail out in this case. 1940 if (!has_node(C->root())) { 1941 if (!_verify_only) { 1942 C->clear_major_progress(); 1943 C->record_method_not_compilable("empty program detected during loop optimization"); 1944 } 1945 return; 1946 } 1947 1948 // Nothing to do, so get out 1949 if( !C->has_loops() && !skip_loop_opts && !do_split_ifs && !_verify_me && !_verify_only ) { 1950 _igvn.optimize(); // Cleanup NeverBranches 1951 return; 1952 } 1953 1954 // Set loop nesting depth 1955 _ltree_root->set_nest( 0 ); 1956 1957 // Split shared headers and insert loop landing pads. 1958 // Do not bother doing this on the Root loop of course. 1959 if( !_verify_me && !_verify_only && _ltree_root->_child ) { 1960 C->print_method("Before beautify loops", 3); 1961 if( _ltree_root->_child->beautify_loops( this ) ) { 1962 // Re-build loop tree! 1963 _ltree_root->_child = NULL; 1964 _nodes.clear(); 1965 reallocate_preorders(); 1966 build_loop_tree(); 1967 // Check for bailout, and return 1968 if (C->failing()) { 1969 return; 1970 } 1971 // Reset loop nesting depth 1972 _ltree_root->set_nest( 0 ); 1973 1974 C->print_method("After beautify loops", 3); 1975 } 1976 } 1977 1978 // Build Dominators for elision of NULL checks & loop finding. 1979 // Since nodes do not have a slot for immediate dominator, make 1980 // a persistent side array for that info indexed on node->_idx. 1981 _idom_size = C->unique(); 1982 _idom = NEW_RESOURCE_ARRAY( Node*, _idom_size ); 1983 _dom_depth = NEW_RESOURCE_ARRAY( uint, _idom_size ); 1984 _dom_stk = NULL; // Allocated on demand in recompute_dom_depth 1985 memset( _dom_depth, 0, _idom_size * sizeof(uint) ); 1986 1987 Dominators(); 1988 1989 if (!_verify_only) { 1990 // As a side effect, Dominators removed any unreachable CFG paths 1991 // into RegionNodes. It doesn't do this test against Root, so 1992 // we do it here. 1993 for( uint i = 1; i < C->root()->req(); i++ ) { 1994 if( !_nodes[C->root()->in(i)->_idx] ) { // Dead path into Root? 1995 _igvn.hash_delete(C->root()); 1996 C->root()->del_req(i); 1997 _igvn._worklist.push(C->root()); 1998 i--; // Rerun same iteration on compressed edges 1999 } 2000 } 2001 2002 // Given dominators, try to find inner loops with calls that must 2003 // always be executed (call dominates loop tail). These loops do 2004 // not need a separate safepoint. 2005 Node_List cisstack(a); 2006 _ltree_root->check_safepts(visited, cisstack); 2007 } 2008 2009 // Walk the DATA nodes and place into loops. Find earliest control 2010 // node. For CFG nodes, the _nodes array starts out and remains 2011 // holding the associated IdealLoopTree pointer. For DATA nodes, the 2012 // _nodes array holds the earliest legal controlling CFG node. 2013 2014 // Allocate stack with enough space to avoid frequent realloc 2015 int stack_size = (C->unique() >> 1) + 16; // (unique>>1)+16 from Java2D stats 2016 Node_Stack nstack( a, stack_size ); 2017 2018 visited.Clear(); 2019 Node_List worklist(a); 2020 // Don't need C->root() on worklist since 2021 // it will be processed among C->top() inputs 2022 worklist.push( C->top() ); 2023 visited.set( C->top()->_idx ); // Set C->top() as visited now 2024 build_loop_early( visited, worklist, nstack ); 2025 2026 // Given early legal placement, try finding counted loops. This placement 2027 // is good enough to discover most loop invariants. 2028 if( !_verify_me && !_verify_only ) 2029 _ltree_root->counted_loop( this ); 2030 2031 // Find latest loop placement. Find ideal loop placement. 2032 visited.Clear(); 2033 init_dom_lca_tags(); 2034 // Need C->root() on worklist when processing outs 2035 worklist.push( C->root() ); 2036 NOT_PRODUCT( C->verify_graph_edges(); ) 2037 worklist.push( C->top() ); 2038 build_loop_late( visited, worklist, nstack ); 2039 2040 if (_verify_only) { 2041 // restore major progress flag 2042 for (int i = 0; i < old_progress; i++) 2043 C->set_major_progress(); 2044 assert(C->unique() == unique, "verification mode made Nodes? ? ?"); 2045 assert(_igvn._worklist.size() == orig_worklist_size, "shouldn't push anything"); 2046 return; 2047 } 2048 2049 // Some parser-inserted loop predicates could never be used by loop 2050 // predication or they were moved away from loop during some optimizations. 2051 // For example, peeling. Eliminate them before next loop optimizations. 2052 if (UseLoopPredicate || LoopLimitCheck) { 2053 eliminate_useless_predicates(); 2054 } 2055 2056 // clear out the dead code 2057 while(_deadlist.size()) { 2058 _igvn.remove_globally_dead_node(_deadlist.pop()); 2059 } 2060 2061 #ifndef PRODUCT 2062 C->verify_graph_edges(); 2063 if (_verify_me) { // Nested verify pass? 2064 // Check to see if the verify mode is broken 2065 assert(C->unique() == unique, "non-optimize mode made Nodes? ? ?"); 2066 return; 2067 } 2068 if(VerifyLoopOptimizations) verify(); 2069 if(TraceLoopOpts && C->has_loops()) { 2070 _ltree_root->dump(); 2071 } 2072 #endif 2073 2074 if (skip_loop_opts) { 2075 // Cleanup any modified bits 2076 _igvn.optimize(); 2077 2078 if (C->log() != NULL) { 2079 log_loop_tree(_ltree_root, _ltree_root, C->log()); 2080 } 2081 return; 2082 } 2083 2084 if (ReassociateInvariants) { 2085 // Reassociate invariants and prep for split_thru_phi 2086 for (LoopTreeIterator iter(_ltree_root); !iter.done(); iter.next()) { 2087 IdealLoopTree* lpt = iter.current(); 2088 if (!lpt->is_counted() || !lpt->is_inner()) continue; 2089 2090 lpt->reassociate_invariants(this); 2091 2092 // Because RCE opportunities can be masked by split_thru_phi, 2093 // look for RCE candidates and inhibit split_thru_phi 2094 // on just their loop-phi's for this pass of loop opts 2095 if (SplitIfBlocks && do_split_ifs) { 2096 if (lpt->policy_range_check(this)) { 2097 lpt->_rce_candidate = 1; // = true 2098 } 2099 } 2100 } 2101 } 2102 2103 // Check for aggressive application of split-if and other transforms 2104 // that require basic-block info (like cloning through Phi's) 2105 if( SplitIfBlocks && do_split_ifs ) { 2106 visited.Clear(); 2107 split_if_with_blocks( visited, nstack ); 2108 NOT_PRODUCT( if( VerifyLoopOptimizations ) verify(); ); 2109 } 2110 2111 // Perform loop predication before iteration splitting 2112 if (C->has_loops() && !C->major_progress() && (C->predicate_count() > 0)) { 2113 _ltree_root->_child->loop_predication(this); 2114 } 2115 2116 if (OptimizeFill && UseLoopPredicate && C->has_loops() && !C->major_progress()) { 2117 if (do_intrinsify_fill()) { 2118 C->set_major_progress(); 2119 } 2120 } 2121 2122 // Perform iteration-splitting on inner loops. Split iterations to avoid 2123 // range checks or one-shot null checks. 2124 2125 // If split-if's didn't hack the graph too bad (no CFG changes) 2126 // then do loop opts. 2127 if (C->has_loops() && !C->major_progress()) { 2128 memset( worklist.adr(), 0, worklist.Size()*sizeof(Node*) ); 2129 _ltree_root->_child->iteration_split( this, worklist ); 2130 // No verify after peeling! GCM has hoisted code out of the loop. 2131 // After peeling, the hoisted code could sink inside the peeled area. 2132 // The peeling code does not try to recompute the best location for 2133 // all the code before the peeled area, so the verify pass will always 2134 // complain about it. 2135 } 2136 // Do verify graph edges in any case 2137 NOT_PRODUCT( C->verify_graph_edges(); ); 2138 2139 if (!do_split_ifs) { 2140 // We saw major progress in Split-If to get here. We forced a 2141 // pass with unrolling and not split-if, however more split-if's 2142 // might make progress. If the unrolling didn't make progress 2143 // then the major-progress flag got cleared and we won't try 2144 // another round of Split-If. In particular the ever-common 2145 // instance-of/check-cast pattern requires at least 2 rounds of 2146 // Split-If to clear out. 2147 C->set_major_progress(); 2148 } 2149 2150 // Repeat loop optimizations if new loops were seen 2151 if (created_loop_node()) { 2152 C->set_major_progress(); 2153 } 2154 2155 // Keep loop predicates and perform optimizations with them 2156 // until no more loop optimizations could be done. 2157 // After that switch predicates off and do more loop optimizations. 2158 if (!C->major_progress() && (C->predicate_count() > 0)) { 2159 C->cleanup_loop_predicates(_igvn); 2160 #ifndef PRODUCT 2161 if (TraceLoopOpts) { 2162 tty->print_cr("PredicatesOff"); 2163 } 2164 #endif 2165 C->set_major_progress(); 2166 } 2167 2168 // Convert scalar to superword operations at the end of all loop opts. 2169 if (UseSuperWord && C->has_loops() && !C->major_progress()) { 2170 // SuperWord transform 2171 SuperWord sw(this); 2172 for (LoopTreeIterator iter(_ltree_root); !iter.done(); iter.next()) { 2173 IdealLoopTree* lpt = iter.current(); 2174 if (lpt->is_counted()) { 2175 sw.transform_loop(lpt); 2176 } 2177 } 2178 } 2179 2180 // Cleanup any modified bits 2181 _igvn.optimize(); 2182 2183 // disable assert until issue with split_flow_path is resolved (6742111) 2184 // assert(!_has_irreducible_loops || C->parsed_irreducible_loop() || C->is_osr_compilation(), 2185 // "shouldn't introduce irreducible loops"); 2186 2187 if (C->log() != NULL) { 2188 log_loop_tree(_ltree_root, _ltree_root, C->log()); 2189 } 2190 } 2191 2192 #ifndef PRODUCT 2193 //------------------------------print_statistics------------------------------- 2194 int PhaseIdealLoop::_loop_invokes=0;// Count of PhaseIdealLoop invokes 2195 int PhaseIdealLoop::_loop_work=0; // Sum of PhaseIdealLoop x unique 2196 void PhaseIdealLoop::print_statistics() { 2197 tty->print_cr("PhaseIdealLoop=%d, sum _unique=%d", _loop_invokes, _loop_work); 2198 } 2199 2200 //------------------------------verify----------------------------------------- 2201 // Build a verify-only PhaseIdealLoop, and see that it agrees with me. 2202 static int fail; // debug only, so its multi-thread dont care 2203 void PhaseIdealLoop::verify() const { 2204 int old_progress = C->major_progress(); 2205 ResourceMark rm; 2206 PhaseIdealLoop loop_verify( _igvn, this ); 2207 VectorSet visited(Thread::current()->resource_area()); 2208 2209 fail = 0; 2210 verify_compare( C->root(), &loop_verify, visited ); 2211 assert( fail == 0, "verify loops failed" ); 2212 // Verify loop structure is the same 2213 _ltree_root->verify_tree(loop_verify._ltree_root, NULL); 2214 // Reset major-progress. It was cleared by creating a verify version of 2215 // PhaseIdealLoop. 2216 for( int i=0; i<old_progress; i++ ) 2217 C->set_major_progress(); 2218 } 2219 2220 //------------------------------verify_compare--------------------------------- 2221 // Make sure me and the given PhaseIdealLoop agree on key data structures 2222 void PhaseIdealLoop::verify_compare( Node *n, const PhaseIdealLoop *loop_verify, VectorSet &visited ) const { 2223 if( !n ) return; 2224 if( visited.test_set( n->_idx ) ) return; 2225 if( !_nodes[n->_idx] ) { // Unreachable 2226 assert( !loop_verify->_nodes[n->_idx], "both should be unreachable" ); 2227 return; 2228 } 2229 2230 uint i; 2231 for( i = 0; i < n->req(); i++ ) 2232 verify_compare( n->in(i), loop_verify, visited ); 2233 2234 // Check the '_nodes' block/loop structure 2235 i = n->_idx; 2236 if( has_ctrl(n) ) { // We have control; verify has loop or ctrl 2237 if( _nodes[i] != loop_verify->_nodes[i] && 2238 get_ctrl_no_update(n) != loop_verify->get_ctrl_no_update(n) ) { 2239 tty->print("Mismatched control setting for: "); 2240 n->dump(); 2241 if( fail++ > 10 ) return; 2242 Node *c = get_ctrl_no_update(n); 2243 tty->print("We have it as: "); 2244 if( c->in(0) ) c->dump(); 2245 else tty->print_cr("N%d",c->_idx); 2246 tty->print("Verify thinks: "); 2247 if( loop_verify->has_ctrl(n) ) 2248 loop_verify->get_ctrl_no_update(n)->dump(); 2249 else 2250 loop_verify->get_loop_idx(n)->dump(); 2251 tty->cr(); 2252 } 2253 } else { // We have a loop 2254 IdealLoopTree *us = get_loop_idx(n); 2255 if( loop_verify->has_ctrl(n) ) { 2256 tty->print("Mismatched loop setting for: "); 2257 n->dump(); 2258 if( fail++ > 10 ) return; 2259 tty->print("We have it as: "); 2260 us->dump(); 2261 tty->print("Verify thinks: "); 2262 loop_verify->get_ctrl_no_update(n)->dump(); 2263 tty->cr(); 2264 } else if (!C->major_progress()) { 2265 // Loop selection can be messed up if we did a major progress 2266 // operation, like split-if. Do not verify in that case. 2267 IdealLoopTree *them = loop_verify->get_loop_idx(n); 2268 if( us->_head != them->_head || us->_tail != them->_tail ) { 2269 tty->print("Unequals loops for: "); 2270 n->dump(); 2271 if( fail++ > 10 ) return; 2272 tty->print("We have it as: "); 2273 us->dump(); 2274 tty->print("Verify thinks: "); 2275 them->dump(); 2276 tty->cr(); 2277 } 2278 } 2279 } 2280 2281 // Check for immediate dominators being equal 2282 if( i >= _idom_size ) { 2283 if( !n->is_CFG() ) return; 2284 tty->print("CFG Node with no idom: "); 2285 n->dump(); 2286 return; 2287 } 2288 if( !n->is_CFG() ) return; 2289 if( n == C->root() ) return; // No IDOM here 2290 2291 assert(n->_idx == i, "sanity"); 2292 Node *id = idom_no_update(n); 2293 if( id != loop_verify->idom_no_update(n) ) { 2294 tty->print("Unequals idoms for: "); 2295 n->dump(); 2296 if( fail++ > 10 ) return; 2297 tty->print("We have it as: "); 2298 id->dump(); 2299 tty->print("Verify thinks: "); 2300 loop_verify->idom_no_update(n)->dump(); 2301 tty->cr(); 2302 } 2303 2304 } 2305 2306 //------------------------------verify_tree------------------------------------ 2307 // Verify that tree structures match. Because the CFG can change, siblings 2308 // within the loop tree can be reordered. We attempt to deal with that by 2309 // reordering the verify's loop tree if possible. 2310 void IdealLoopTree::verify_tree(IdealLoopTree *loop, const IdealLoopTree *parent) const { 2311 assert( _parent == parent, "Badly formed loop tree" ); 2312 2313 // Siblings not in same order? Attempt to re-order. 2314 if( _head != loop->_head ) { 2315 // Find _next pointer to update 2316 IdealLoopTree **pp = &loop->_parent->_child; 2317 while( *pp != loop ) 2318 pp = &((*pp)->_next); 2319 // Find proper sibling to be next 2320 IdealLoopTree **nn = &loop->_next; 2321 while( (*nn) && (*nn)->_head != _head ) 2322 nn = &((*nn)->_next); 2323 2324 // Check for no match. 2325 if( !(*nn) ) { 2326 // Annoyingly, irreducible loops can pick different headers 2327 // after a major_progress operation, so the rest of the loop 2328 // tree cannot be matched. 2329 if (_irreducible && Compile::current()->major_progress()) return; 2330 assert( 0, "failed to match loop tree" ); 2331 } 2332 2333 // Move (*nn) to (*pp) 2334 IdealLoopTree *hit = *nn; 2335 *nn = hit->_next; 2336 hit->_next = loop; 2337 *pp = loop; 2338 loop = hit; 2339 // Now try again to verify 2340 } 2341 2342 assert( _head == loop->_head , "mismatched loop head" ); 2343 Node *tail = _tail; // Inline a non-updating version of 2344 while( !tail->in(0) ) // the 'tail()' call. 2345 tail = tail->in(1); 2346 assert( tail == loop->_tail, "mismatched loop tail" ); 2347 2348 // Counted loops that are guarded should be able to find their guards 2349 if( _head->is_CountedLoop() && _head->as_CountedLoop()->is_main_loop() ) { 2350 CountedLoopNode *cl = _head->as_CountedLoop(); 2351 Node *init = cl->init_trip(); 2352 Node *ctrl = cl->in(LoopNode::EntryControl); 2353 assert( ctrl->Opcode() == Op_IfTrue || ctrl->Opcode() == Op_IfFalse, "" ); 2354 Node *iff = ctrl->in(0); 2355 assert( iff->Opcode() == Op_If, "" ); 2356 Node *bol = iff->in(1); 2357 assert( bol->Opcode() == Op_Bool, "" ); 2358 Node *cmp = bol->in(1); 2359 assert( cmp->Opcode() == Op_CmpI, "" ); 2360 Node *add = cmp->in(1); 2361 Node *opaq; 2362 if( add->Opcode() == Op_Opaque1 ) { 2363 opaq = add; 2364 } else { 2365 assert( add->Opcode() == Op_AddI || add->Opcode() == Op_ConI , "" ); 2366 assert( add == init, "" ); 2367 opaq = cmp->in(2); 2368 } 2369 assert( opaq->Opcode() == Op_Opaque1, "" ); 2370 2371 } 2372 2373 if (_child != NULL) _child->verify_tree(loop->_child, this); 2374 if (_next != NULL) _next ->verify_tree(loop->_next, parent); 2375 // Innermost loops need to verify loop bodies, 2376 // but only if no 'major_progress' 2377 int fail = 0; 2378 if (!Compile::current()->major_progress() && _child == NULL) { 2379 for( uint i = 0; i < _body.size(); i++ ) { 2380 Node *n = _body.at(i); 2381 if (n->outcnt() == 0) continue; // Ignore dead 2382 uint j; 2383 for( j = 0; j < loop->_body.size(); j++ ) 2384 if( loop->_body.at(j) == n ) 2385 break; 2386 if( j == loop->_body.size() ) { // Not found in loop body 2387 // Last ditch effort to avoid assertion: Its possible that we 2388 // have some users (so outcnt not zero) but are still dead. 2389 // Try to find from root. 2390 if (Compile::current()->root()->find(n->_idx)) { 2391 fail++; 2392 tty->print("We have that verify does not: "); 2393 n->dump(); 2394 } 2395 } 2396 } 2397 for( uint i2 = 0; i2 < loop->_body.size(); i2++ ) { 2398 Node *n = loop->_body.at(i2); 2399 if (n->outcnt() == 0) continue; // Ignore dead 2400 uint j; 2401 for( j = 0; j < _body.size(); j++ ) 2402 if( _body.at(j) == n ) 2403 break; 2404 if( j == _body.size() ) { // Not found in loop body 2405 // Last ditch effort to avoid assertion: Its possible that we 2406 // have some users (so outcnt not zero) but are still dead. 2407 // Try to find from root. 2408 if (Compile::current()->root()->find(n->_idx)) { 2409 fail++; 2410 tty->print("Verify has that we do not: "); 2411 n->dump(); 2412 } 2413 } 2414 } 2415 assert( !fail, "loop body mismatch" ); 2416 } 2417 } 2418 2419 #endif 2420 2421 //------------------------------set_idom--------------------------------------- 2422 void PhaseIdealLoop::set_idom(Node* d, Node* n, uint dom_depth) { 2423 uint idx = d->_idx; 2424 if (idx >= _idom_size) { 2425 uint newsize = _idom_size<<1; 2426 while( idx >= newsize ) { 2427 newsize <<= 1; 2428 } 2429 _idom = REALLOC_RESOURCE_ARRAY( Node*, _idom,_idom_size,newsize); 2430 _dom_depth = REALLOC_RESOURCE_ARRAY( uint, _dom_depth,_idom_size,newsize); 2431 memset( _dom_depth + _idom_size, 0, (newsize - _idom_size) * sizeof(uint) ); 2432 _idom_size = newsize; 2433 } 2434 _idom[idx] = n; 2435 _dom_depth[idx] = dom_depth; 2436 } 2437 2438 //------------------------------recompute_dom_depth--------------------------------------- 2439 // The dominator tree is constructed with only parent pointers. 2440 // This recomputes the depth in the tree by first tagging all 2441 // nodes as "no depth yet" marker. The next pass then runs up 2442 // the dom tree from each node marked "no depth yet", and computes 2443 // the depth on the way back down. 2444 void PhaseIdealLoop::recompute_dom_depth() { 2445 uint no_depth_marker = C->unique(); 2446 uint i; 2447 // Initialize depth to "no depth yet" 2448 for (i = 0; i < _idom_size; i++) { 2449 if (_dom_depth[i] > 0 && _idom[i] != NULL) { 2450 _dom_depth[i] = no_depth_marker; 2451 } 2452 } 2453 if (_dom_stk == NULL) { 2454 uint init_size = C->unique() / 100; // Guess that 1/100 is a reasonable initial size. 2455 if (init_size < 10) init_size = 10; 2456 _dom_stk = new GrowableArray<uint>(init_size); 2457 } 2458 // Compute new depth for each node. 2459 for (i = 0; i < _idom_size; i++) { 2460 uint j = i; 2461 // Run up the dom tree to find a node with a depth 2462 while (_dom_depth[j] == no_depth_marker) { 2463 _dom_stk->push(j); 2464 j = _idom[j]->_idx; 2465 } 2466 // Compute the depth on the way back down this tree branch 2467 uint dd = _dom_depth[j] + 1; 2468 while (_dom_stk->length() > 0) { 2469 uint j = _dom_stk->pop(); 2470 _dom_depth[j] = dd; 2471 dd++; 2472 } 2473 } 2474 } 2475 2476 //------------------------------sort------------------------------------------- 2477 // Insert 'loop' into the existing loop tree. 'innermost' is a leaf of the 2478 // loop tree, not the root. 2479 IdealLoopTree *PhaseIdealLoop::sort( IdealLoopTree *loop, IdealLoopTree *innermost ) { 2480 if( !innermost ) return loop; // New innermost loop 2481 2482 int loop_preorder = get_preorder(loop->_head); // Cache pre-order number 2483 assert( loop_preorder, "not yet post-walked loop" ); 2484 IdealLoopTree **pp = &innermost; // Pointer to previous next-pointer 2485 IdealLoopTree *l = *pp; // Do I go before or after 'l'? 2486 2487 // Insert at start of list 2488 while( l ) { // Insertion sort based on pre-order 2489 if( l == loop ) return innermost; // Already on list! 2490 int l_preorder = get_preorder(l->_head); // Cache pre-order number 2491 assert( l_preorder, "not yet post-walked l" ); 2492 // Check header pre-order number to figure proper nesting 2493 if( loop_preorder > l_preorder ) 2494 break; // End of insertion 2495 // If headers tie (e.g., shared headers) check tail pre-order numbers. 2496 // Since I split shared headers, you'd think this could not happen. 2497 // BUT: I must first do the preorder numbering before I can discover I 2498 // have shared headers, so the split headers all get the same preorder 2499 // number as the RegionNode they split from. 2500 if( loop_preorder == l_preorder && 2501 get_preorder(loop->_tail) < get_preorder(l->_tail) ) 2502 break; // Also check for shared headers (same pre#) 2503 pp = &l->_parent; // Chain up list 2504 l = *pp; 2505 } 2506 // Link into list 2507 // Point predecessor to me 2508 *pp = loop; 2509 // Point me to successor 2510 IdealLoopTree *p = loop->_parent; 2511 loop->_parent = l; // Point me to successor 2512 if( p ) sort( p, innermost ); // Insert my parents into list as well 2513 return innermost; 2514 } 2515 2516 //------------------------------build_loop_tree-------------------------------- 2517 // I use a modified Vick/Tarjan algorithm. I need pre- and a post- visit 2518 // bits. The _nodes[] array is mapped by Node index and holds a NULL for 2519 // not-yet-pre-walked, pre-order # for pre-but-not-post-walked and holds the 2520 // tightest enclosing IdealLoopTree for post-walked. 2521 // 2522 // During my forward walk I do a short 1-layer lookahead to see if I can find 2523 // a loop backedge with that doesn't have any work on the backedge. This 2524 // helps me construct nested loops with shared headers better. 2525 // 2526 // Once I've done the forward recursion, I do the post-work. For each child 2527 // I check to see if there is a backedge. Backedges define a loop! I 2528 // insert an IdealLoopTree at the target of the backedge. 2529 // 2530 // During the post-work I also check to see if I have several children 2531 // belonging to different loops. If so, then this Node is a decision point 2532 // where control flow can choose to change loop nests. It is at this 2533 // decision point where I can figure out how loops are nested. At this 2534 // time I can properly order the different loop nests from my children. 2535 // Note that there may not be any backedges at the decision point! 2536 // 2537 // Since the decision point can be far removed from the backedges, I can't 2538 // order my loops at the time I discover them. Thus at the decision point 2539 // I need to inspect loop header pre-order numbers to properly nest my 2540 // loops. This means I need to sort my childrens' loops by pre-order. 2541 // The sort is of size number-of-control-children, which generally limits 2542 // it to size 2 (i.e., I just choose between my 2 target loops). 2543 void PhaseIdealLoop::build_loop_tree() { 2544 // Allocate stack of size C->unique()/2 to avoid frequent realloc 2545 GrowableArray <Node *> bltstack(C->unique() >> 1); 2546 Node *n = C->root(); 2547 bltstack.push(n); 2548 int pre_order = 1; 2549 int stack_size; 2550 2551 while ( ( stack_size = bltstack.length() ) != 0 ) { 2552 n = bltstack.top(); // Leave node on stack 2553 if ( !is_visited(n) ) { 2554 // ---- Pre-pass Work ---- 2555 // Pre-walked but not post-walked nodes need a pre_order number. 2556 2557 set_preorder_visited( n, pre_order ); // set as visited 2558 2559 // ---- Scan over children ---- 2560 // Scan first over control projections that lead to loop headers. 2561 // This helps us find inner-to-outer loops with shared headers better. 2562 2563 // Scan children's children for loop headers. 2564 for ( int i = n->outcnt() - 1; i >= 0; --i ) { 2565 Node* m = n->raw_out(i); // Child 2566 if( m->is_CFG() && !is_visited(m) ) { // Only for CFG children 2567 // Scan over children's children to find loop 2568 for (DUIterator_Fast jmax, j = m->fast_outs(jmax); j < jmax; j++) { 2569 Node* l = m->fast_out(j); 2570 if( is_visited(l) && // Been visited? 2571 !is_postvisited(l) && // But not post-visited 2572 get_preorder(l) < pre_order ) { // And smaller pre-order 2573 // Found! Scan the DFS down this path before doing other paths 2574 bltstack.push(m); 2575 break; 2576 } 2577 } 2578 } 2579 } 2580 pre_order++; 2581 } 2582 else if ( !is_postvisited(n) ) { 2583 // Note: build_loop_tree_impl() adds out edges on rare occasions, 2584 // such as com.sun.rsasign.am::a. 2585 // For non-recursive version, first, process current children. 2586 // On next iteration, check if additional children were added. 2587 for ( int k = n->outcnt() - 1; k >= 0; --k ) { 2588 Node* u = n->raw_out(k); 2589 if ( u->is_CFG() && !is_visited(u) ) { 2590 bltstack.push(u); 2591 } 2592 } 2593 if ( bltstack.length() == stack_size ) { 2594 // There were no additional children, post visit node now 2595 (void)bltstack.pop(); // Remove node from stack 2596 pre_order = build_loop_tree_impl( n, pre_order ); 2597 // Check for bailout 2598 if (C->failing()) { 2599 return; 2600 } 2601 // Check to grow _preorders[] array for the case when 2602 // build_loop_tree_impl() adds new nodes. 2603 check_grow_preorders(); 2604 } 2605 } 2606 else { 2607 (void)bltstack.pop(); // Remove post-visited node from stack 2608 } 2609 } 2610 } 2611 2612 //------------------------------build_loop_tree_impl--------------------------- 2613 int PhaseIdealLoop::build_loop_tree_impl( Node *n, int pre_order ) { 2614 // ---- Post-pass Work ---- 2615 // Pre-walked but not post-walked nodes need a pre_order number. 2616 2617 // Tightest enclosing loop for this Node 2618 IdealLoopTree *innermost = NULL; 2619 2620 // For all children, see if any edge is a backedge. If so, make a loop 2621 // for it. Then find the tightest enclosing loop for the self Node. 2622 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { 2623 Node* m = n->fast_out(i); // Child 2624 if( n == m ) continue; // Ignore control self-cycles 2625 if( !m->is_CFG() ) continue;// Ignore non-CFG edges 2626 2627 IdealLoopTree *l; // Child's loop 2628 if( !is_postvisited(m) ) { // Child visited but not post-visited? 2629 // Found a backedge 2630 assert( get_preorder(m) < pre_order, "should be backedge" ); 2631 // Check for the RootNode, which is already a LoopNode and is allowed 2632 // to have multiple "backedges". 2633 if( m == C->root()) { // Found the root? 2634 l = _ltree_root; // Root is the outermost LoopNode 2635 } else { // Else found a nested loop 2636 // Insert a LoopNode to mark this loop. 2637 l = new IdealLoopTree(this, m, n); 2638 } // End of Else found a nested loop 2639 if( !has_loop(m) ) // If 'm' does not already have a loop set 2640 set_loop(m, l); // Set loop header to loop now 2641 2642 } else { // Else not a nested loop 2643 if( !_nodes[m->_idx] ) continue; // Dead code has no loop 2644 l = get_loop(m); // Get previously determined loop 2645 // If successor is header of a loop (nest), move up-loop till it 2646 // is a member of some outer enclosing loop. Since there are no 2647 // shared headers (I've split them already) I only need to go up 2648 // at most 1 level. 2649 while( l && l->_head == m ) // Successor heads loop? 2650 l = l->_parent; // Move up 1 for me 2651 // If this loop is not properly parented, then this loop 2652 // has no exit path out, i.e. its an infinite loop. 2653 if( !l ) { 2654 // Make loop "reachable" from root so the CFG is reachable. Basically 2655 // insert a bogus loop exit that is never taken. 'm', the loop head, 2656 // points to 'n', one (of possibly many) fall-in paths. There may be 2657 // many backedges as well. 2658 2659 // Here I set the loop to be the root loop. I could have, after 2660 // inserting a bogus loop exit, restarted the recursion and found my 2661 // new loop exit. This would make the infinite loop a first-class 2662 // loop and it would then get properly optimized. What's the use of 2663 // optimizing an infinite loop? 2664 l = _ltree_root; // Oops, found infinite loop 2665 2666 if (!_verify_only) { 2667 // Insert the NeverBranch between 'm' and it's control user. 2668 NeverBranchNode *iff = new (C, 1) NeverBranchNode( m ); 2669 _igvn.register_new_node_with_optimizer(iff); 2670 set_loop(iff, l); 2671 Node *if_t = new (C, 1) CProjNode( iff, 0 ); 2672 _igvn.register_new_node_with_optimizer(if_t); 2673 set_loop(if_t, l); 2674 2675 Node* cfg = NULL; // Find the One True Control User of m 2676 for (DUIterator_Fast jmax, j = m->fast_outs(jmax); j < jmax; j++) { 2677 Node* x = m->fast_out(j); 2678 if (x->is_CFG() && x != m && x != iff) 2679 { cfg = x; break; } 2680 } 2681 assert(cfg != NULL, "must find the control user of m"); 2682 uint k = 0; // Probably cfg->in(0) 2683 while( cfg->in(k) != m ) k++; // But check incase cfg is a Region 2684 cfg->set_req( k, if_t ); // Now point to NeverBranch 2685 2686 // Now create the never-taken loop exit 2687 Node *if_f = new (C, 1) CProjNode( iff, 1 ); 2688 _igvn.register_new_node_with_optimizer(if_f); 2689 set_loop(if_f, l); 2690 // Find frame ptr for Halt. Relies on the optimizer 2691 // V-N'ing. Easier and quicker than searching through 2692 // the program structure. 2693 Node *frame = new (C, 1) ParmNode( C->start(), TypeFunc::FramePtr ); 2694 _igvn.register_new_node_with_optimizer(frame); 2695 // Halt & Catch Fire 2696 Node *halt = new (C, TypeFunc::Parms) HaltNode( if_f, frame ); 2697 _igvn.register_new_node_with_optimizer(halt); 2698 set_loop(halt, l); 2699 C->root()->add_req(halt); 2700 } 2701 set_loop(C->root(), _ltree_root); 2702 } 2703 } 2704 // Weeny check for irreducible. This child was already visited (this 2705 // IS the post-work phase). Is this child's loop header post-visited 2706 // as well? If so, then I found another entry into the loop. 2707 if (!_verify_only) { 2708 while( is_postvisited(l->_head) ) { 2709 // found irreducible 2710 l->_irreducible = 1; // = true 2711 l = l->_parent; 2712 _has_irreducible_loops = true; 2713 // Check for bad CFG here to prevent crash, and bailout of compile 2714 if (l == NULL) { 2715 C->record_method_not_compilable("unhandled CFG detected during loop optimization"); 2716 return pre_order; 2717 } 2718 } 2719 } 2720 2721 // This Node might be a decision point for loops. It is only if 2722 // it's children belong to several different loops. The sort call 2723 // does a trivial amount of work if there is only 1 child or all 2724 // children belong to the same loop. If however, the children 2725 // belong to different loops, the sort call will properly set the 2726 // _parent pointers to show how the loops nest. 2727 // 2728 // In any case, it returns the tightest enclosing loop. 2729 innermost = sort( l, innermost ); 2730 } 2731 2732 // Def-use info will have some dead stuff; dead stuff will have no 2733 // loop decided on. 2734 2735 // Am I a loop header? If so fix up my parent's child and next ptrs. 2736 if( innermost && innermost->_head == n ) { 2737 assert( get_loop(n) == innermost, "" ); 2738 IdealLoopTree *p = innermost->_parent; 2739 IdealLoopTree *l = innermost; 2740 while( p && l->_head == n ) { 2741 l->_next = p->_child; // Put self on parents 'next child' 2742 p->_child = l; // Make self as first child of parent 2743 l = p; // Now walk up the parent chain 2744 p = l->_parent; 2745 } 2746 } else { 2747 // Note that it is possible for a LoopNode to reach here, if the 2748 // backedge has been made unreachable (hence the LoopNode no longer 2749 // denotes a Loop, and will eventually be removed). 2750 2751 // Record tightest enclosing loop for self. Mark as post-visited. 2752 set_loop(n, innermost); 2753 // Also record has_call flag early on 2754 if( innermost ) { 2755 if( n->is_Call() && !n->is_CallLeaf() && !n->is_macro() ) { 2756 // Do not count uncommon calls 2757 if( !n->is_CallStaticJava() || !n->as_CallStaticJava()->_name ) { 2758 Node *iff = n->in(0)->in(0); 2759 if( !iff->is_If() || 2760 (n->in(0)->Opcode() == Op_IfFalse && 2761 (1.0 - iff->as_If()->_prob) >= 0.01) || 2762 (iff->as_If()->_prob >= 0.01) ) 2763 innermost->_has_call = 1; 2764 } 2765 } else if( n->is_Allocate() && n->as_Allocate()->_is_scalar_replaceable ) { 2766 // Disable loop optimizations if the loop has a scalar replaceable 2767 // allocation. This disabling may cause a potential performance lost 2768 // if the allocation is not eliminated for some reason. 2769 innermost->_allow_optimizations = false; 2770 innermost->_has_call = 1; // = true 2771 } 2772 } 2773 } 2774 2775 // Flag as post-visited now 2776 set_postvisited(n); 2777 return pre_order; 2778 } 2779 2780 2781 //------------------------------build_loop_early------------------------------- 2782 // Put Data nodes into some loop nest, by setting the _nodes[]->loop mapping. 2783 // First pass computes the earliest controlling node possible. This is the 2784 // controlling input with the deepest dominating depth. 2785 void PhaseIdealLoop::build_loop_early( VectorSet &visited, Node_List &worklist, Node_Stack &nstack ) { 2786 while (worklist.size() != 0) { 2787 // Use local variables nstack_top_n & nstack_top_i to cache values 2788 // on nstack's top. 2789 Node *nstack_top_n = worklist.pop(); 2790 uint nstack_top_i = 0; 2791 //while_nstack_nonempty: 2792 while (true) { 2793 // Get parent node and next input's index from stack's top. 2794 Node *n = nstack_top_n; 2795 uint i = nstack_top_i; 2796 uint cnt = n->req(); // Count of inputs 2797 if (i == 0) { // Pre-process the node. 2798 if( has_node(n) && // Have either loop or control already? 2799 !has_ctrl(n) ) { // Have loop picked out already? 2800 // During "merge_many_backedges" we fold up several nested loops 2801 // into a single loop. This makes the members of the original 2802 // loop bodies pointing to dead loops; they need to move up 2803 // to the new UNION'd larger loop. I set the _head field of these 2804 // dead loops to NULL and the _parent field points to the owning 2805 // loop. Shades of UNION-FIND algorithm. 2806 IdealLoopTree *ilt; 2807 while( !(ilt = get_loop(n))->_head ) { 2808 // Normally I would use a set_loop here. But in this one special 2809 // case, it is legal (and expected) to change what loop a Node 2810 // belongs to. 2811 _nodes.map(n->_idx, (Node*)(ilt->_parent) ); 2812 } 2813 // Remove safepoints ONLY if I've already seen I don't need one. 2814 // (the old code here would yank a 2nd safepoint after seeing a 2815 // first one, even though the 1st did not dominate in the loop body 2816 // and thus could be avoided indefinitely) 2817 if( !_verify_only && !_verify_me && ilt->_has_sfpt && n->Opcode() == Op_SafePoint && 2818 is_deleteable_safept(n)) { 2819 Node *in = n->in(TypeFunc::Control); 2820 lazy_replace(n,in); // Pull safepoint now 2821 // Carry on with the recursion "as if" we are walking 2822 // only the control input 2823 if( !visited.test_set( in->_idx ) ) { 2824 worklist.push(in); // Visit this guy later, using worklist 2825 } 2826 // Get next node from nstack: 2827 // - skip n's inputs processing by setting i > cnt; 2828 // - we also will not call set_early_ctrl(n) since 2829 // has_node(n) == true (see the condition above). 2830 i = cnt + 1; 2831 } 2832 } 2833 } // if (i == 0) 2834 2835 // Visit all inputs 2836 bool done = true; // Assume all n's inputs will be processed 2837 while (i < cnt) { 2838 Node *in = n->in(i); 2839 ++i; 2840 if (in == NULL) continue; 2841 if (in->pinned() && !in->is_CFG()) 2842 set_ctrl(in, in->in(0)); 2843 int is_visited = visited.test_set( in->_idx ); 2844 if (!has_node(in)) { // No controlling input yet? 2845 assert( !in->is_CFG(), "CFG Node with no controlling input?" ); 2846 assert( !is_visited, "visit only once" ); 2847 nstack.push(n, i); // Save parent node and next input's index. 2848 nstack_top_n = in; // Process current input now. 2849 nstack_top_i = 0; 2850 done = false; // Not all n's inputs processed. 2851 break; // continue while_nstack_nonempty; 2852 } else if (!is_visited) { 2853 // This guy has a location picked out for him, but has not yet 2854 // been visited. Happens to all CFG nodes, for instance. 2855 // Visit him using the worklist instead of recursion, to break 2856 // cycles. Since he has a location already we do not need to 2857 // find his location before proceeding with the current Node. 2858 worklist.push(in); // Visit this guy later, using worklist 2859 } 2860 } 2861 if (done) { 2862 // All of n's inputs have been processed, complete post-processing. 2863 2864 // Compute earliest point this Node can go. 2865 // CFG, Phi, pinned nodes already know their controlling input. 2866 if (!has_node(n)) { 2867 // Record earliest legal location 2868 set_early_ctrl( n ); 2869 } 2870 if (nstack.is_empty()) { 2871 // Finished all nodes on stack. 2872 // Process next node on the worklist. 2873 break; 2874 } 2875 // Get saved parent node and next input's index. 2876 nstack_top_n = nstack.node(); 2877 nstack_top_i = nstack.index(); 2878 nstack.pop(); 2879 } 2880 } // while (true) 2881 } 2882 } 2883 2884 //------------------------------dom_lca_internal-------------------------------- 2885 // Pair-wise LCA 2886 Node *PhaseIdealLoop::dom_lca_internal( Node *n1, Node *n2 ) const { 2887 if( !n1 ) return n2; // Handle NULL original LCA 2888 assert( n1->is_CFG(), "" ); 2889 assert( n2->is_CFG(), "" ); 2890 // find LCA of all uses 2891 uint d1 = dom_depth(n1); 2892 uint d2 = dom_depth(n2); 2893 while (n1 != n2) { 2894 if (d1 > d2) { 2895 n1 = idom(n1); 2896 d1 = dom_depth(n1); 2897 } else if (d1 < d2) { 2898 n2 = idom(n2); 2899 d2 = dom_depth(n2); 2900 } else { 2901 // Here d1 == d2. Due to edits of the dominator-tree, sections 2902 // of the tree might have the same depth. These sections have 2903 // to be searched more carefully. 2904 2905 // Scan up all the n1's with equal depth, looking for n2. 2906 Node *t1 = idom(n1); 2907 while (dom_depth(t1) == d1) { 2908 if (t1 == n2) return n2; 2909 t1 = idom(t1); 2910 } 2911 // Scan up all the n2's with equal depth, looking for n1. 2912 Node *t2 = idom(n2); 2913 while (dom_depth(t2) == d2) { 2914 if (t2 == n1) return n1; 2915 t2 = idom(t2); 2916 } 2917 // Move up to a new dominator-depth value as well as up the dom-tree. 2918 n1 = t1; 2919 n2 = t2; 2920 d1 = dom_depth(n1); 2921 d2 = dom_depth(n2); 2922 } 2923 } 2924 return n1; 2925 } 2926 2927 //------------------------------compute_idom----------------------------------- 2928 // Locally compute IDOM using dom_lca call. Correct only if the incoming 2929 // IDOMs are correct. 2930 Node *PhaseIdealLoop::compute_idom( Node *region ) const { 2931 assert( region->is_Region(), "" ); 2932 Node *LCA = NULL; 2933 for( uint i = 1; i < region->req(); i++ ) { 2934 if( region->in(i) != C->top() ) 2935 LCA = dom_lca( LCA, region->in(i) ); 2936 } 2937 return LCA; 2938 } 2939 2940 bool PhaseIdealLoop::verify_dominance(Node* n, Node* use, Node* LCA, Node* early) { 2941 bool had_error = false; 2942 #ifdef ASSERT 2943 if (early != C->root()) { 2944 // Make sure that there's a dominance path from use to LCA 2945 Node* d = use; 2946 while (d != LCA) { 2947 d = idom(d); 2948 if (d == C->root()) { 2949 tty->print_cr("*** Use %d isn't dominated by def %s", use->_idx, n->_idx); 2950 n->dump(); 2951 use->dump(); 2952 had_error = true; 2953 break; 2954 } 2955 } 2956 } 2957 #endif 2958 return had_error; 2959 } 2960 2961 2962 Node* PhaseIdealLoop::compute_lca_of_uses(Node* n, Node* early, bool verify) { 2963 // Compute LCA over list of uses 2964 bool had_error = false; 2965 Node *LCA = NULL; 2966 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax && LCA != early; i++) { 2967 Node* c = n->fast_out(i); 2968 if (_nodes[c->_idx] == NULL) 2969 continue; // Skip the occasional dead node 2970 if( c->is_Phi() ) { // For Phis, we must land above on the path 2971 for( uint j=1; j<c->req(); j++ ) {// For all inputs 2972 if( c->in(j) == n ) { // Found matching input? 2973 Node *use = c->in(0)->in(j); 2974 if (_verify_only && use->is_top()) continue; 2975 LCA = dom_lca_for_get_late_ctrl( LCA, use, n ); 2976 if (verify) had_error = verify_dominance(n, use, LCA, early) || had_error; 2977 } 2978 } 2979 } else { 2980 // For CFG data-users, use is in the block just prior 2981 Node *use = has_ctrl(c) ? get_ctrl(c) : c->in(0); 2982 LCA = dom_lca_for_get_late_ctrl( LCA, use, n ); 2983 if (verify) had_error = verify_dominance(n, use, LCA, early) || had_error; 2984 } 2985 } 2986 assert(!had_error, "bad dominance"); 2987 return LCA; 2988 } 2989 2990 //------------------------------get_late_ctrl---------------------------------- 2991 // Compute latest legal control. 2992 Node *PhaseIdealLoop::get_late_ctrl( Node *n, Node *early ) { 2993 assert(early != NULL, "early control should not be NULL"); 2994 2995 Node* LCA = compute_lca_of_uses(n, early); 2996 #ifdef ASSERT 2997 if (LCA == C->root() && LCA != early) { 2998 // def doesn't dominate uses so print some useful debugging output 2999 compute_lca_of_uses(n, early, true); 3000 } 3001 #endif 3002 3003 // if this is a load, check for anti-dependent stores 3004 // We use a conservative algorithm to identify potential interfering 3005 // instructions and for rescheduling the load. The users of the memory 3006 // input of this load are examined. Any use which is not a load and is 3007 // dominated by early is considered a potentially interfering store. 3008 // This can produce false positives. 3009 if (n->is_Load() && LCA != early) { 3010 Node_List worklist; 3011 3012 Node *mem = n->in(MemNode::Memory); 3013 for (DUIterator_Fast imax, i = mem->fast_outs(imax); i < imax; i++) { 3014 Node* s = mem->fast_out(i); 3015 worklist.push(s); 3016 } 3017 while(worklist.size() != 0 && LCA != early) { 3018 Node* s = worklist.pop(); 3019 if (s->is_Load()) { 3020 continue; 3021 } else if (s->is_MergeMem()) { 3022 for (DUIterator_Fast imax, i = s->fast_outs(imax); i < imax; i++) { 3023 Node* s1 = s->fast_out(i); 3024 worklist.push(s1); 3025 } 3026 } else { 3027 Node *sctrl = has_ctrl(s) ? get_ctrl(s) : s->in(0); 3028 assert(sctrl != NULL || s->outcnt() == 0, "must have control"); 3029 if (sctrl != NULL && !sctrl->is_top() && is_dominator(early, sctrl)) { 3030 LCA = dom_lca_for_get_late_ctrl(LCA, sctrl, n); 3031 } 3032 } 3033 } 3034 } 3035 3036 assert(LCA == find_non_split_ctrl(LCA), "unexpected late control"); 3037 return LCA; 3038 } 3039 3040 // true if CFG node d dominates CFG node n 3041 bool PhaseIdealLoop::is_dominator(Node *d, Node *n) { 3042 if (d == n) 3043 return true; 3044 assert(d->is_CFG() && n->is_CFG(), "must have CFG nodes"); 3045 uint dd = dom_depth(d); 3046 while (dom_depth(n) >= dd) { 3047 if (n == d) 3048 return true; 3049 n = idom(n); 3050 } 3051 return false; 3052 } 3053 3054 //------------------------------dom_lca_for_get_late_ctrl_internal------------- 3055 // Pair-wise LCA with tags. 3056 // Tag each index with the node 'tag' currently being processed 3057 // before advancing up the dominator chain using idom(). 3058 // Later calls that find a match to 'tag' know that this path has already 3059 // been considered in the current LCA (which is input 'n1' by convention). 3060 // Since get_late_ctrl() is only called once for each node, the tag array 3061 // does not need to be cleared between calls to get_late_ctrl(). 3062 // Algorithm trades a larger constant factor for better asymptotic behavior 3063 // 3064 Node *PhaseIdealLoop::dom_lca_for_get_late_ctrl_internal( Node *n1, Node *n2, Node *tag ) { 3065 uint d1 = dom_depth(n1); 3066 uint d2 = dom_depth(n2); 3067 3068 do { 3069 if (d1 > d2) { 3070 // current lca is deeper than n2 3071 _dom_lca_tags.map(n1->_idx, tag); 3072 n1 = idom(n1); 3073 d1 = dom_depth(n1); 3074 } else if (d1 < d2) { 3075 // n2 is deeper than current lca 3076 Node *memo = _dom_lca_tags[n2->_idx]; 3077 if( memo == tag ) { 3078 return n1; // Return the current LCA 3079 } 3080 _dom_lca_tags.map(n2->_idx, tag); 3081 n2 = idom(n2); 3082 d2 = dom_depth(n2); 3083 } else { 3084 // Here d1 == d2. Due to edits of the dominator-tree, sections 3085 // of the tree might have the same depth. These sections have 3086 // to be searched more carefully. 3087 3088 // Scan up all the n1's with equal depth, looking for n2. 3089 _dom_lca_tags.map(n1->_idx, tag); 3090 Node *t1 = idom(n1); 3091 while (dom_depth(t1) == d1) { 3092 if (t1 == n2) return n2; 3093 _dom_lca_tags.map(t1->_idx, tag); 3094 t1 = idom(t1); 3095 } 3096 // Scan up all the n2's with equal depth, looking for n1. 3097 _dom_lca_tags.map(n2->_idx, tag); 3098 Node *t2 = idom(n2); 3099 while (dom_depth(t2) == d2) { 3100 if (t2 == n1) return n1; 3101 _dom_lca_tags.map(t2->_idx, tag); 3102 t2 = idom(t2); 3103 } 3104 // Move up to a new dominator-depth value as well as up the dom-tree. 3105 n1 = t1; 3106 n2 = t2; 3107 d1 = dom_depth(n1); 3108 d2 = dom_depth(n2); 3109 } 3110 } while (n1 != n2); 3111 return n1; 3112 } 3113 3114 //------------------------------init_dom_lca_tags------------------------------ 3115 // Tag could be a node's integer index, 32bits instead of 64bits in some cases 3116 // Intended use does not involve any growth for the array, so it could 3117 // be of fixed size. 3118 void PhaseIdealLoop::init_dom_lca_tags() { 3119 uint limit = C->unique() + 1; 3120 _dom_lca_tags.map( limit, NULL ); 3121 #ifdef ASSERT 3122 for( uint i = 0; i < limit; ++i ) { 3123 assert(_dom_lca_tags[i] == NULL, "Must be distinct from each node pointer"); 3124 } 3125 #endif // ASSERT 3126 } 3127 3128 //------------------------------clear_dom_lca_tags------------------------------ 3129 // Tag could be a node's integer index, 32bits instead of 64bits in some cases 3130 // Intended use does not involve any growth for the array, so it could 3131 // be of fixed size. 3132 void PhaseIdealLoop::clear_dom_lca_tags() { 3133 uint limit = C->unique() + 1; 3134 _dom_lca_tags.map( limit, NULL ); 3135 _dom_lca_tags.clear(); 3136 #ifdef ASSERT 3137 for( uint i = 0; i < limit; ++i ) { 3138 assert(_dom_lca_tags[i] == NULL, "Must be distinct from each node pointer"); 3139 } 3140 #endif // ASSERT 3141 } 3142 3143 //------------------------------build_loop_late-------------------------------- 3144 // Put Data nodes into some loop nest, by setting the _nodes[]->loop mapping. 3145 // Second pass finds latest legal placement, and ideal loop placement. 3146 void PhaseIdealLoop::build_loop_late( VectorSet &visited, Node_List &worklist, Node_Stack &nstack ) { 3147 while (worklist.size() != 0) { 3148 Node *n = worklist.pop(); 3149 // Only visit once 3150 if (visited.test_set(n->_idx)) continue; 3151 uint cnt = n->outcnt(); 3152 uint i = 0; 3153 while (true) { 3154 assert( _nodes[n->_idx], "no dead nodes" ); 3155 // Visit all children 3156 if (i < cnt) { 3157 Node* use = n->raw_out(i); 3158 ++i; 3159 // Check for dead uses. Aggressively prune such junk. It might be 3160 // dead in the global sense, but still have local uses so I cannot 3161 // easily call 'remove_dead_node'. 3162 if( _nodes[use->_idx] != NULL || use->is_top() ) { // Not dead? 3163 // Due to cycles, we might not hit the same fixed point in the verify 3164 // pass as we do in the regular pass. Instead, visit such phis as 3165 // simple uses of the loop head. 3166 if( use->in(0) && (use->is_CFG() || use->is_Phi()) ) { 3167 if( !visited.test(use->_idx) ) 3168 worklist.push(use); 3169 } else if( !visited.test_set(use->_idx) ) { 3170 nstack.push(n, i); // Save parent and next use's index. 3171 n = use; // Process all children of current use. 3172 cnt = use->outcnt(); 3173 i = 0; 3174 } 3175 } else { 3176 // Do not visit around the backedge of loops via data edges. 3177 // push dead code onto a worklist 3178 _deadlist.push(use); 3179 } 3180 } else { 3181 // All of n's children have been processed, complete post-processing. 3182 build_loop_late_post(n); 3183 if (nstack.is_empty()) { 3184 // Finished all nodes on stack. 3185 // Process next node on the worklist. 3186 break; 3187 } 3188 // Get saved parent node and next use's index. Visit the rest of uses. 3189 n = nstack.node(); 3190 cnt = n->outcnt(); 3191 i = nstack.index(); 3192 nstack.pop(); 3193 } 3194 } 3195 } 3196 } 3197 3198 //------------------------------build_loop_late_post--------------------------- 3199 // Put Data nodes into some loop nest, by setting the _nodes[]->loop mapping. 3200 // Second pass finds latest legal placement, and ideal loop placement. 3201 void PhaseIdealLoop::build_loop_late_post( Node *n ) { 3202 3203 if (n->req() == 2 && n->Opcode() == Op_ConvI2L && !C->major_progress() && !_verify_only) { 3204 _igvn._worklist.push(n); // Maybe we'll normalize it, if no more loops. 3205 } 3206 3207 // CFG and pinned nodes already handled 3208 if( n->in(0) ) { 3209 if( n->in(0)->is_top() ) return; // Dead? 3210 3211 // We'd like +VerifyLoopOptimizations to not believe that Mod's/Loads 3212 // _must_ be pinned (they have to observe their control edge of course). 3213 // Unlike Stores (which modify an unallocable resource, the memory 3214 // state), Mods/Loads can float around. So free them up. 3215 bool pinned = true; 3216 switch( n->Opcode() ) { 3217 case Op_DivI: 3218 case Op_DivF: 3219 case Op_DivD: 3220 case Op_ModI: 3221 case Op_ModF: 3222 case Op_ModD: 3223 case Op_LoadB: // Same with Loads; they can sink 3224 case Op_LoadUS: // during loop optimizations. 3225 case Op_LoadD: 3226 case Op_LoadF: 3227 case Op_LoadI: 3228 case Op_LoadKlass: 3229 case Op_LoadNKlass: 3230 case Op_LoadL: 3231 case Op_LoadS: 3232 case Op_LoadP: 3233 case Op_LoadN: 3234 case Op_LoadRange: 3235 case Op_LoadD_unaligned: 3236 case Op_LoadL_unaligned: 3237 case Op_StrComp: // Does a bunch of load-like effects 3238 case Op_StrEquals: 3239 case Op_StrIndexOf: 3240 case Op_AryEq: 3241 pinned = false; 3242 } 3243 if( pinned ) { 3244 IdealLoopTree *chosen_loop = get_loop(n->is_CFG() ? n : get_ctrl(n)); 3245 if( !chosen_loop->_child ) // Inner loop? 3246 chosen_loop->_body.push(n); // Collect inner loops 3247 return; 3248 } 3249 } else { // No slot zero 3250 if( n->is_CFG() ) { // CFG with no slot 0 is dead 3251 _nodes.map(n->_idx,0); // No block setting, it's globally dead 3252 return; 3253 } 3254 assert(!n->is_CFG() || n->outcnt() == 0, ""); 3255 } 3256 3257 // Do I have a "safe range" I can select over? 3258 Node *early = get_ctrl(n);// Early location already computed 3259 3260 // Compute latest point this Node can go 3261 Node *LCA = get_late_ctrl( n, early ); 3262 // LCA is NULL due to uses being dead 3263 if( LCA == NULL ) { 3264 #ifdef ASSERT 3265 for (DUIterator i1 = n->outs(); n->has_out(i1); i1++) { 3266 assert( _nodes[n->out(i1)->_idx] == NULL, "all uses must also be dead"); 3267 } 3268 #endif 3269 _nodes.map(n->_idx, 0); // This node is useless 3270 _deadlist.push(n); 3271 return; 3272 } 3273 assert(LCA != NULL && !LCA->is_top(), "no dead nodes"); 3274 3275 Node *legal = LCA; // Walk 'legal' up the IDOM chain 3276 Node *least = legal; // Best legal position so far 3277 while( early != legal ) { // While not at earliest legal 3278 #ifdef ASSERT 3279 if (legal->is_Start() && !early->is_Root()) { 3280 // Bad graph. Print idom path and fail. 3281 dump_bad_graph(n, early, LCA); 3282 assert(false, "Bad graph detected in build_loop_late"); 3283 } 3284 #endif 3285 // Find least loop nesting depth 3286 legal = idom(legal); // Bump up the IDOM tree 3287 // Check for lower nesting depth 3288 if( get_loop(legal)->_nest < get_loop(least)->_nest ) 3289 least = legal; 3290 } 3291 assert(early == legal || legal != C->root(), "bad dominance of inputs"); 3292 3293 // Try not to place code on a loop entry projection 3294 // which can inhibit range check elimination. 3295 if (least != early) { 3296 Node* ctrl_out = least->unique_ctrl_out(); 3297 if (ctrl_out && ctrl_out->is_CountedLoop() && 3298 least == ctrl_out->in(LoopNode::EntryControl)) { 3299 Node* least_dom = idom(least); 3300 if (get_loop(least_dom)->is_member(get_loop(least))) { 3301 least = least_dom; 3302 } 3303 } 3304 } 3305 3306 #ifdef ASSERT 3307 // If verifying, verify that 'verify_me' has a legal location 3308 // and choose it as our location. 3309 if( _verify_me ) { 3310 Node *v_ctrl = _verify_me->get_ctrl_no_update(n); 3311 Node *legal = LCA; 3312 while( early != legal ) { // While not at earliest legal 3313 if( legal == v_ctrl ) break; // Check for prior good location 3314 legal = idom(legal) ;// Bump up the IDOM tree 3315 } 3316 // Check for prior good location 3317 if( legal == v_ctrl ) least = legal; // Keep prior if found 3318 } 3319 #endif 3320 3321 // Assign discovered "here or above" point 3322 least = find_non_split_ctrl(least); 3323 set_ctrl(n, least); 3324 3325 // Collect inner loop bodies 3326 IdealLoopTree *chosen_loop = get_loop(least); 3327 if( !chosen_loop->_child ) // Inner loop? 3328 chosen_loop->_body.push(n);// Collect inner loops 3329 } 3330 3331 #ifdef ASSERT 3332 void PhaseIdealLoop::dump_bad_graph(Node* n, Node* early, Node* LCA) { 3333 tty->print_cr( "Bad graph detected in build_loop_late"); 3334 tty->print("n: "); n->dump(); 3335 tty->print("early(n): "); early->dump(); 3336 if (n->in(0) != NULL && !n->in(0)->is_top() && 3337 n->in(0) != early && !n->in(0)->is_Root()) { 3338 tty->print("n->in(0): "); n->in(0)->dump(); 3339 } 3340 for (uint i = 1; i < n->req(); i++) { 3341 Node* in1 = n->in(i); 3342 if (in1 != NULL && in1 != n && !in1->is_top()) { 3343 tty->print("n->in(%d): ", i); in1->dump(); 3344 Node* in1_early = get_ctrl(in1); 3345 tty->print("early(n->in(%d)): ", i); in1_early->dump(); 3346 if (in1->in(0) != NULL && !in1->in(0)->is_top() && 3347 in1->in(0) != in1_early && !in1->in(0)->is_Root()) { 3348 tty->print("n->in(%d)->in(0): ", i); in1->in(0)->dump(); 3349 } 3350 for (uint j = 1; j < in1->req(); j++) { 3351 Node* in2 = in1->in(j); 3352 if (in2 != NULL && in2 != n && in2 != in1 && !in2->is_top()) { 3353 tty->print("n->in(%d)->in(%d): ", i, j); in2->dump(); 3354 Node* in2_early = get_ctrl(in2); 3355 tty->print("early(n->in(%d)->in(%d)): ", i, j); in2_early->dump(); 3356 if (in2->in(0) != NULL && !in2->in(0)->is_top() && 3357 in2->in(0) != in2_early && !in2->in(0)->is_Root()) { 3358 tty->print("n->in(%d)->in(%d)->in(0): ", i, j); in2->in(0)->dump(); 3359 } 3360 } 3361 } 3362 } 3363 } 3364 tty->cr(); 3365 tty->print("LCA(n): "); LCA->dump(); 3366 for (uint i = 0; i < n->outcnt(); i++) { 3367 Node* u1 = n->raw_out(i); 3368 if (u1 == n) 3369 continue; 3370 tty->print("n->out(%d): ", i); u1->dump(); 3371 if (u1->is_CFG()) { 3372 for (uint j = 0; j < u1->outcnt(); j++) { 3373 Node* u2 = u1->raw_out(j); 3374 if (u2 != u1 && u2 != n && u2->is_CFG()) { 3375 tty->print("n->out(%d)->out(%d): ", i, j); u2->dump(); 3376 } 3377 } 3378 } else { 3379 Node* u1_later = get_ctrl(u1); 3380 tty->print("later(n->out(%d)): ", i); u1_later->dump(); 3381 if (u1->in(0) != NULL && !u1->in(0)->is_top() && 3382 u1->in(0) != u1_later && !u1->in(0)->is_Root()) { 3383 tty->print("n->out(%d)->in(0): ", i); u1->in(0)->dump(); 3384 } 3385 for (uint j = 0; j < u1->outcnt(); j++) { 3386 Node* u2 = u1->raw_out(j); 3387 if (u2 == n || u2 == u1) 3388 continue; 3389 tty->print("n->out(%d)->out(%d): ", i, j); u2->dump(); 3390 if (!u2->is_CFG()) { 3391 Node* u2_later = get_ctrl(u2); 3392 tty->print("later(n->out(%d)->out(%d)): ", i, j); u2_later->dump(); 3393 if (u2->in(0) != NULL && !u2->in(0)->is_top() && 3394 u2->in(0) != u2_later && !u2->in(0)->is_Root()) { 3395 tty->print("n->out(%d)->in(0): ", i); u2->in(0)->dump(); 3396 } 3397 } 3398 } 3399 } 3400 } 3401 tty->cr(); 3402 int ct = 0; 3403 Node *dbg_legal = LCA; 3404 while(!dbg_legal->is_Start() && ct < 100) { 3405 tty->print("idom[%d] ",ct); dbg_legal->dump(); 3406 ct++; 3407 dbg_legal = idom(dbg_legal); 3408 } 3409 tty->cr(); 3410 } 3411 #endif 3412 3413 #ifndef PRODUCT 3414 //------------------------------dump------------------------------------------- 3415 void PhaseIdealLoop::dump( ) const { 3416 ResourceMark rm; 3417 Arena* arena = Thread::current()->resource_area(); 3418 Node_Stack stack(arena, C->unique() >> 2); 3419 Node_List rpo_list; 3420 VectorSet visited(arena); 3421 visited.set(C->top()->_idx); 3422 rpo( C->root(), stack, visited, rpo_list ); 3423 // Dump root loop indexed by last element in PO order 3424 dump( _ltree_root, rpo_list.size(), rpo_list ); 3425 } 3426 3427 void PhaseIdealLoop::dump( IdealLoopTree *loop, uint idx, Node_List &rpo_list ) const { 3428 loop->dump_head(); 3429 3430 // Now scan for CFG nodes in the same loop 3431 for( uint j=idx; j > 0; j-- ) { 3432 Node *n = rpo_list[j-1]; 3433 if( !_nodes[n->_idx] ) // Skip dead nodes 3434 continue; 3435 if( get_loop(n) != loop ) { // Wrong loop nest 3436 if( get_loop(n)->_head == n && // Found nested loop? 3437 get_loop(n)->_parent == loop ) 3438 dump(get_loop(n),rpo_list.size(),rpo_list); // Print it nested-ly 3439 continue; 3440 } 3441 3442 // Dump controlling node 3443 for( uint x = 0; x < loop->_nest; x++ ) 3444 tty->print(" "); 3445 tty->print("C"); 3446 if( n == C->root() ) { 3447 n->dump(); 3448 } else { 3449 Node* cached_idom = idom_no_update(n); 3450 Node *computed_idom = n->in(0); 3451 if( n->is_Region() ) { 3452 computed_idom = compute_idom(n); 3453 // computed_idom() will return n->in(0) when idom(n) is an IfNode (or 3454 // any MultiBranch ctrl node), so apply a similar transform to 3455 // the cached idom returned from idom_no_update. 3456 cached_idom = find_non_split_ctrl(cached_idom); 3457 } 3458 tty->print(" ID:%d",computed_idom->_idx); 3459 n->dump(); 3460 if( cached_idom != computed_idom ) { 3461 tty->print_cr("*** BROKEN IDOM! Computed as: %d, cached as: %d", 3462 computed_idom->_idx, cached_idom->_idx); 3463 } 3464 } 3465 // Dump nodes it controls 3466 for( uint k = 0; k < _nodes.Size(); k++ ) { 3467 // (k < C->unique() && get_ctrl(find(k)) == n) 3468 if (k < C->unique() && _nodes[k] == (Node*)((intptr_t)n + 1)) { 3469 Node *m = C->root()->find(k); 3470 if( m && m->outcnt() > 0 ) { 3471 if (!(has_ctrl(m) && get_ctrl_no_update(m) == n)) { 3472 tty->print_cr("*** BROKEN CTRL ACCESSOR! _nodes[k] is %p, ctrl is %p", 3473 _nodes[k], has_ctrl(m) ? get_ctrl_no_update(m) : NULL); 3474 } 3475 for( uint j = 0; j < loop->_nest; j++ ) 3476 tty->print(" "); 3477 tty->print(" "); 3478 m->dump(); 3479 } 3480 } 3481 } 3482 } 3483 } 3484 3485 // Collect a R-P-O for the whole CFG. 3486 // Result list is in post-order (scan backwards for RPO) 3487 void PhaseIdealLoop::rpo( Node *start, Node_Stack &stk, VectorSet &visited, Node_List &rpo_list ) const { 3488 stk.push(start, 0); 3489 visited.set(start->_idx); 3490 3491 while (stk.is_nonempty()) { 3492 Node* m = stk.node(); 3493 uint idx = stk.index(); 3494 if (idx < m->outcnt()) { 3495 stk.set_index(idx + 1); 3496 Node* n = m->raw_out(idx); 3497 if (n->is_CFG() && !visited.test_set(n->_idx)) { 3498 stk.push(n, 0); 3499 } 3500 } else { 3501 rpo_list.push(m); 3502 stk.pop(); 3503 } 3504 } 3505 } 3506 #endif 3507 3508 3509 //============================================================================= 3510 //------------------------------LoopTreeIterator----------------------------------- 3511 3512 // Advance to next loop tree using a preorder, left-to-right traversal. 3513 void LoopTreeIterator::next() { 3514 assert(!done(), "must not be done."); 3515 if (_curnt->_child != NULL) { 3516 _curnt = _curnt->_child; 3517 } else if (_curnt->_next != NULL) { 3518 _curnt = _curnt->_next; 3519 } else { 3520 while (_curnt != _root && _curnt->_next == NULL) { 3521 _curnt = _curnt->_parent; 3522 } 3523 if (_curnt == _root) { 3524 _curnt = NULL; 3525 assert(done(), "must be done."); 3526 } else { 3527 assert(_curnt->_next != NULL, "must be more to do"); 3528 _curnt = _curnt->_next; 3529 } 3530 } 3531 }