1 /*
   2  * Copyright (c) 1999, 2010, 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 "incls/_precompiled.incl"
  26 # include "incls/_c1_IR.cpp.incl"
  27 
  28 
  29 // Implementation of XHandlers
  30 //
  31 // Note: This code could eventually go away if we are
  32 //       just using the ciExceptionHandlerStream.
  33 
  34 XHandlers::XHandlers(ciMethod* method) : _list(method->exception_table_length()) {
  35   ciExceptionHandlerStream s(method);
  36   while (!s.is_done()) {
  37     _list.append(new XHandler(s.handler()));
  38     s.next();
  39   }
  40   assert(s.count() == method->exception_table_length(), "exception table lengths inconsistent");
  41 }
  42 
  43 // deep copy of all XHandler contained in list
  44 XHandlers::XHandlers(XHandlers* other) :
  45   _list(other->length())
  46 {
  47   for (int i = 0; i < other->length(); i++) {
  48     _list.append(new XHandler(other->handler_at(i)));
  49   }
  50 }
  51 
  52 // Returns whether a particular exception type can be caught.  Also
  53 // returns true if klass is unloaded or any exception handler
  54 // classes are unloaded.  type_is_exact indicates whether the throw
  55 // is known to be exactly that class or it might throw a subtype.
  56 bool XHandlers::could_catch(ciInstanceKlass* klass, bool type_is_exact) const {
  57   // the type is unknown so be conservative
  58   if (!klass->is_loaded()) {
  59     return true;
  60   }
  61 
  62   for (int i = 0; i < length(); i++) {
  63     XHandler* handler = handler_at(i);
  64     if (handler->is_catch_all()) {
  65       // catch of ANY
  66       return true;
  67     }
  68     ciInstanceKlass* handler_klass = handler->catch_klass();
  69     // if it's unknown it might be catchable
  70     if (!handler_klass->is_loaded()) {
  71       return true;
  72     }
  73     // if the throw type is definitely a subtype of the catch type
  74     // then it can be caught.
  75     if (klass->is_subtype_of(handler_klass)) {
  76       return true;
  77     }
  78     if (!type_is_exact) {
  79       // If the type isn't exactly known then it can also be caught by
  80       // catch statements where the inexact type is a subtype of the
  81       // catch type.
  82       // given: foo extends bar extends Exception
  83       // throw bar can be caught by catch foo, catch bar, and catch
  84       // Exception, however it can't be caught by any handlers without
  85       // bar in its type hierarchy.
  86       if (handler_klass->is_subtype_of(klass)) {
  87         return true;
  88       }
  89     }
  90   }
  91 
  92   return false;
  93 }
  94 
  95 
  96 bool XHandlers::equals(XHandlers* others) const {
  97   if (others == NULL) return false;
  98   if (length() != others->length()) return false;
  99 
 100   for (int i = 0; i < length(); i++) {
 101     if (!handler_at(i)->equals(others->handler_at(i))) return false;
 102   }
 103   return true;
 104 }
 105 
 106 bool XHandler::equals(XHandler* other) const {
 107   assert(entry_pco() != -1 && other->entry_pco() != -1, "must have entry_pco");
 108 
 109   if (entry_pco() != other->entry_pco()) return false;
 110   if (scope_count() != other->scope_count()) return false;
 111   if (_desc != other->_desc) return false;
 112 
 113   assert(entry_block() == other->entry_block(), "entry_block must be equal when entry_pco is equal");
 114   return true;
 115 }
 116 
 117 
 118 // Implementation of IRScope
 119 
 120 BlockBegin* IRScope::header_block(BlockBegin* entry, BlockBegin::Flag f, ValueStack* state) {
 121   if (entry == NULL) return NULL;
 122   assert(entry->is_set(f), "entry/flag mismatch");
 123   // create header block
 124   BlockBegin* h = new BlockBegin(entry->bci());
 125   BlockEnd* g = new Goto(entry, false);
 126   h->set_next(g, entry->bci());
 127   h->set_end(g);
 128   h->set(f);
 129   // setup header block end state
 130   ValueStack* s = state->copy(); // can use copy since stack is empty (=> no phis)
 131   assert(s->stack_is_empty(), "must have empty stack at entry point");
 132   g->set_state(s);
 133   return h;
 134 }
 135 
 136 
 137 BlockBegin* IRScope::build_graph(Compilation* compilation, int osr_bci) {
 138   GraphBuilder gm(compilation, this);
 139   NOT_PRODUCT(if (PrintValueNumbering && Verbose) gm.print_stats());
 140   if (compilation->bailed_out()) return NULL;
 141   return gm.start();
 142 }
 143 
 144 
 145 IRScope::IRScope(Compilation* compilation, IRScope* caller, int caller_bci, ciMethod* method, int osr_bci, bool create_graph)
 146 : _callees(2)
 147 , _compilation(compilation)
 148 , _lock_stack_size(-1)
 149 , _requires_phi_function(method->max_locals())
 150 {
 151   _caller             = caller;
 152   _caller_bci         = caller == NULL ? -1 : caller_bci;
 153   _caller_state       = NULL; // Must be set later if needed
 154   _level              = caller == NULL ?  0 : caller->level() + 1;
 155   _method             = method;
 156   _xhandlers          = new XHandlers(method);
 157   _number_of_locks    = 0;
 158   _monitor_pairing_ok = method->has_balanced_monitors();
 159   _start              = NULL;
 160 
 161   if (osr_bci == -1) {
 162     _requires_phi_function.clear();
 163   } else {
 164         // selective creation of phi functions is not possibel in osr-methods
 165     _requires_phi_function.set_range(0, method->max_locals());
 166   }
 167 
 168   assert(method->holder()->is_loaded() , "method holder must be loaded");
 169 
 170   // build graph if monitor pairing is ok
 171   if (create_graph && monitor_pairing_ok()) _start = build_graph(compilation, osr_bci);
 172 }
 173 
 174 
 175 int IRScope::max_stack() const {
 176   int my_max = method()->max_stack();
 177   int callee_max = 0;
 178   for (int i = 0; i < number_of_callees(); i++) {
 179     callee_max = MAX2(callee_max, callee_no(i)->max_stack());
 180   }
 181   return my_max + callee_max;
 182 }
 183 
 184 
 185 void IRScope::compute_lock_stack_size() {
 186   if (!InlineMethodsWithExceptionHandlers) {
 187     _lock_stack_size = 0;
 188     return;
 189   }
 190 
 191   // Figure out whether we have to preserve expression stack elements
 192   // for parent scopes, and if so, how many
 193   IRScope* cur_scope = this;
 194   while (cur_scope != NULL && !cur_scope->xhandlers()->has_handlers()) {
 195     cur_scope = cur_scope->caller();
 196   }
 197   _lock_stack_size = (cur_scope == NULL ? 0 :
 198                       (cur_scope->caller_state() == NULL ? 0 :
 199                        cur_scope->caller_state()->stack_size()));
 200 }
 201 
 202 int IRScope::top_scope_bci() const {
 203   assert(!is_top_scope(), "no correct answer for top scope possible");
 204   const IRScope* scope = this;
 205   while (!scope->caller()->is_top_scope()) {
 206     scope = scope->caller();
 207   }
 208   return scope->caller_bci();
 209 }
 210 
 211 bool IRScopeDebugInfo::should_reexecute() {
 212   ciMethod* cur_method = scope()->method();
 213   int       cur_bci    = bci();
 214   if (cur_method != NULL && cur_bci != SynchronizationEntryBCI) {
 215     Bytecodes::Code code = cur_method->java_code_at_bci(cur_bci);
 216     return Interpreter::bytecode_should_reexecute(code);
 217   } else
 218     return false;
 219 }
 220 
 221 
 222 // Implementation of CodeEmitInfo
 223 
 224 // Stack must be NON-null
 225 CodeEmitInfo::CodeEmitInfo(int bci, ValueStack* stack, XHandlers* exception_handlers)
 226   : _scope(stack->scope())
 227   , _bci(bci)
 228   , _scope_debug_info(NULL)
 229   , _oop_map(NULL)
 230   , _stack(stack)
 231   , _exception_handlers(exception_handlers)
 232   , _next(NULL)
 233   , _id(-1)
 234   , _is_method_handle_invoke(false) {
 235   assert(_stack != NULL, "must be non null");
 236   assert(_bci == SynchronizationEntryBCI || Bytecodes::is_defined(scope()->method()->java_code_at_bci(_bci)), "make sure bci points at a real bytecode");
 237 }
 238 
 239 
 240 CodeEmitInfo::CodeEmitInfo(CodeEmitInfo* info, bool lock_stack_only)
 241   : _scope(info->_scope)
 242   , _exception_handlers(NULL)
 243   , _bci(info->_bci)
 244   , _scope_debug_info(NULL)
 245   , _oop_map(NULL)
 246   , _is_method_handle_invoke(info->_is_method_handle_invoke) {
 247   if (lock_stack_only) {
 248     if (info->_stack != NULL) {
 249       _stack = info->_stack->copy_locks();
 250     } else {
 251       _stack = NULL;
 252     }
 253   } else {
 254     _stack = info->_stack;
 255   }
 256 
 257   // deep copy of exception handlers
 258   if (info->_exception_handlers != NULL) {
 259     _exception_handlers = new XHandlers(info->_exception_handlers);
 260   }
 261 }
 262 
 263 
 264 void CodeEmitInfo::record_debug_info(DebugInformationRecorder* recorder, int pc_offset) {
 265   // record the safepoint before recording the debug info for enclosing scopes
 266   recorder->add_safepoint(pc_offset, _oop_map->deep_copy());
 267   _scope_debug_info->record_debug_info(recorder, pc_offset, true/*topmost*/, _is_method_handle_invoke);
 268   recorder->end_safepoint(pc_offset);
 269 }
 270 
 271 
 272 void CodeEmitInfo::add_register_oop(LIR_Opr opr) {
 273   assert(_oop_map != NULL, "oop map must already exist");
 274   assert(opr->is_single_cpu(), "should not call otherwise");
 275 
 276   int frame_size = frame_map()->framesize();
 277   int arg_count = frame_map()->oop_map_arg_count();
 278   VMReg name = frame_map()->regname(opr);
 279   _oop_map->set_oop(name);
 280 }
 281 
 282 
 283 
 284 
 285 // Implementation of IR
 286 
 287 IR::IR(Compilation* compilation, ciMethod* method, int osr_bci) :
 288     _locals_size(in_WordSize(-1))
 289   , _num_loops(0) {
 290   // setup IR fields
 291   _compilation = compilation;
 292   _top_scope   = new IRScope(compilation, NULL, -1, method, osr_bci, true);
 293   _code        = NULL;
 294 }
 295 
 296 
 297 void IR::optimize() {
 298   Optimizer opt(this);
 299   if (!compilation()->profile_branches()) {
 300     if (DoCEE) {
 301       opt.eliminate_conditional_expressions();
 302 #ifndef PRODUCT
 303       if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after CEE"); print(true); }
 304       if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after CEE"); print(false); }
 305 #endif
 306     }
 307     if (EliminateBlocks) {
 308       opt.eliminate_blocks();
 309 #ifndef PRODUCT
 310       if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after block elimination"); print(true); }
 311       if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after block elimination"); print(false); }
 312 #endif
 313     }
 314   }
 315   if (EliminateNullChecks) {
 316     opt.eliminate_null_checks();
 317 #ifndef PRODUCT
 318     if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after null check elimination"); print(true); }
 319     if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after null check elimination"); print(false); }
 320 #endif
 321   }
 322 }
 323 
 324 
 325 static int sort_pairs(BlockPair** a, BlockPair** b) {
 326   if ((*a)->from() == (*b)->from()) {
 327     return (*a)->to()->block_id() - (*b)->to()->block_id();
 328   } else {
 329     return (*a)->from()->block_id() - (*b)->from()->block_id();
 330   }
 331 }
 332 
 333 
 334 class CriticalEdgeFinder: public BlockClosure {
 335   BlockPairList blocks;
 336   IR*       _ir;
 337 
 338  public:
 339   CriticalEdgeFinder(IR* ir): _ir(ir) {}
 340   void block_do(BlockBegin* bb) {
 341     BlockEnd* be = bb->end();
 342     int nos = be->number_of_sux();
 343     if (nos >= 2) {
 344       for (int i = 0; i < nos; i++) {
 345         BlockBegin* sux = be->sux_at(i);
 346         if (sux->number_of_preds() >= 2) {
 347           blocks.append(new BlockPair(bb, sux));
 348         }
 349       }
 350     }
 351   }
 352 
 353   void split_edges() {
 354     BlockPair* last_pair = NULL;
 355     blocks.sort(sort_pairs);
 356     for (int i = 0; i < blocks.length(); i++) {
 357       BlockPair* pair = blocks.at(i);
 358       if (last_pair != NULL && pair->is_same(last_pair)) continue;
 359       BlockBegin* from = pair->from();
 360       BlockBegin* to = pair->to();
 361       BlockBegin* split = from->insert_block_between(to);
 362 #ifndef PRODUCT
 363       if ((PrintIR || PrintIR1) && Verbose) {
 364         tty->print_cr("Split critical edge B%d -> B%d (new block B%d)",
 365                       from->block_id(), to->block_id(), split->block_id());
 366       }
 367 #endif
 368       last_pair = pair;
 369     }
 370   }
 371 };
 372 
 373 void IR::split_critical_edges() {
 374   CriticalEdgeFinder cef(this);
 375 
 376   iterate_preorder(&cef);
 377   cef.split_edges();
 378 }
 379 
 380 
 381 class UseCountComputer: public ValueVisitor, BlockClosure {
 382  private:
 383   void visit(Value* n) {
 384     // Local instructions and Phis for expression stack values at the
 385     // start of basic blocks are not added to the instruction list
 386     if ((*n)->bci() == -99 && (*n)->as_Local() == NULL &&
 387         (*n)->as_Phi() == NULL) {
 388       assert(false, "a node was not appended to the graph");
 389       Compilation::current()->bailout("a node was not appended to the graph");
 390     }
 391     // use n's input if not visited before
 392     if (!(*n)->is_pinned() && !(*n)->has_uses()) {
 393       // note: a) if the instruction is pinned, it will be handled by compute_use_count
 394       //       b) if the instruction has uses, it was touched before
 395       //       => in both cases we don't need to update n's values
 396       uses_do(n);
 397     }
 398     // use n
 399     (*n)->_use_count++;
 400   }
 401 
 402   Values* worklist;
 403   int depth;
 404   enum {
 405     max_recurse_depth = 20
 406   };
 407 
 408   void uses_do(Value* n) {
 409     depth++;
 410     if (depth > max_recurse_depth) {
 411       // don't allow the traversal to recurse too deeply
 412       worklist->push(*n);
 413     } else {
 414       (*n)->input_values_do(this);
 415       // special handling for some instructions
 416       if ((*n)->as_BlockEnd() != NULL) {
 417         // note on BlockEnd:
 418         //   must 'use' the stack only if the method doesn't
 419         //   terminate, however, in those cases stack is empty
 420         (*n)->state_values_do(this);
 421       }
 422     }
 423     depth--;
 424   }
 425 
 426   void block_do(BlockBegin* b) {
 427     depth = 0;
 428     // process all pinned nodes as the roots of expression trees
 429     for (Instruction* n = b; n != NULL; n = n->next()) {
 430       if (n->is_pinned()) uses_do(&n);
 431     }
 432     assert(depth == 0, "should have counted back down");
 433 
 434     // now process any unpinned nodes which recursed too deeply
 435     while (worklist->length() > 0) {
 436       Value t = worklist->pop();
 437       if (!t->is_pinned()) {
 438         // compute the use count
 439         uses_do(&t);
 440 
 441         // pin the instruction so that LIRGenerator doesn't recurse
 442         // too deeply during it's evaluation.
 443         t->pin();
 444       }
 445     }
 446     assert(depth == 0, "should have counted back down");
 447   }
 448 
 449   UseCountComputer() {
 450     worklist = new Values();
 451     depth = 0;
 452   }
 453 
 454  public:
 455   static void compute(BlockList* blocks) {
 456     UseCountComputer ucc;
 457     blocks->iterate_backward(&ucc);
 458   }
 459 };
 460 
 461 
 462 // helper macro for short definition of trace-output inside code
 463 #ifndef PRODUCT
 464   #define TRACE_LINEAR_SCAN(level, code)       \
 465     if (TraceLinearScanLevel >= level) {       \
 466       code;                                    \
 467     }
 468 #else
 469   #define TRACE_LINEAR_SCAN(level, code)
 470 #endif
 471 
 472 class ComputeLinearScanOrder : public StackObj {
 473  private:
 474   int        _max_block_id;        // the highest block_id of a block
 475   int        _num_blocks;          // total number of blocks (smaller than _max_block_id)
 476   int        _num_loops;           // total number of loops
 477   bool       _iterative_dominators;// method requires iterative computation of dominatiors
 478 
 479   BlockList* _linear_scan_order;   // the resulting list of blocks in correct order
 480 
 481   BitMap     _visited_blocks;      // used for recursive processing of blocks
 482   BitMap     _active_blocks;       // used for recursive processing of blocks
 483   BitMap     _dominator_blocks;    // temproary BitMap used for computation of dominator
 484   intArray   _forward_branches;    // number of incoming forward branches for each block
 485   BlockList  _loop_end_blocks;     // list of all loop end blocks collected during count_edges
 486   BitMap2D   _loop_map;            // two-dimensional bit set: a bit is set if a block is contained in a loop
 487   BlockList  _work_list;           // temporary list (used in mark_loops and compute_order)
 488 
 489   Compilation* _compilation;
 490 
 491   // accessors for _visited_blocks and _active_blocks
 492   void init_visited()                     { _active_blocks.clear(); _visited_blocks.clear(); }
 493   bool is_visited(BlockBegin* b) const    { return _visited_blocks.at(b->block_id()); }
 494   bool is_active(BlockBegin* b) const     { return _active_blocks.at(b->block_id()); }
 495   void set_visited(BlockBegin* b)         { assert(!is_visited(b), "already set"); _visited_blocks.set_bit(b->block_id()); }
 496   void set_active(BlockBegin* b)          { assert(!is_active(b), "already set");  _active_blocks.set_bit(b->block_id()); }
 497   void clear_active(BlockBegin* b)        { assert(is_active(b), "not already");   _active_blocks.clear_bit(b->block_id()); }
 498 
 499   // accessors for _forward_branches
 500   void inc_forward_branches(BlockBegin* b) { _forward_branches.at_put(b->block_id(), _forward_branches.at(b->block_id()) + 1); }
 501   int  dec_forward_branches(BlockBegin* b) { _forward_branches.at_put(b->block_id(), _forward_branches.at(b->block_id()) - 1); return _forward_branches.at(b->block_id()); }
 502 
 503   // accessors for _loop_map
 504   bool is_block_in_loop   (int loop_idx, BlockBegin* b) const { return _loop_map.at(loop_idx, b->block_id()); }
 505   void set_block_in_loop  (int loop_idx, BlockBegin* b)       { _loop_map.set_bit(loop_idx, b->block_id()); }
 506   void clear_block_in_loop(int loop_idx, int block_id)        { _loop_map.clear_bit(loop_idx, block_id); }
 507 
 508   // count edges between blocks
 509   void count_edges(BlockBegin* cur, BlockBegin* parent);
 510 
 511   // loop detection
 512   void mark_loops();
 513   void clear_non_natural_loops(BlockBegin* start_block);
 514   void assign_loop_depth(BlockBegin* start_block);
 515 
 516   // computation of final block order
 517   BlockBegin* common_dominator(BlockBegin* a, BlockBegin* b);
 518   void compute_dominator(BlockBegin* cur, BlockBegin* parent);
 519   int  compute_weight(BlockBegin* cur);
 520   bool ready_for_processing(BlockBegin* cur);
 521   void sort_into_work_list(BlockBegin* b);
 522   void append_block(BlockBegin* cur);
 523   void compute_order(BlockBegin* start_block);
 524 
 525   // fixup of dominators for non-natural loops
 526   bool compute_dominators_iter();
 527   void compute_dominators();
 528 
 529   // debug functions
 530   NOT_PRODUCT(void print_blocks();)
 531   DEBUG_ONLY(void verify();)
 532 
 533   Compilation* compilation() const { return _compilation; }
 534  public:
 535   ComputeLinearScanOrder(Compilation* c, BlockBegin* start_block);
 536 
 537   // accessors for final result
 538   BlockList* linear_scan_order() const    { return _linear_scan_order; }
 539   int        num_loops() const            { return _num_loops; }
 540 };
 541 
 542 
 543 ComputeLinearScanOrder::ComputeLinearScanOrder(Compilation* c, BlockBegin* start_block) :
 544   _max_block_id(BlockBegin::number_of_blocks()),
 545   _num_blocks(0),
 546   _num_loops(0),
 547   _iterative_dominators(false),
 548   _visited_blocks(_max_block_id),
 549   _active_blocks(_max_block_id),
 550   _dominator_blocks(_max_block_id),
 551   _forward_branches(_max_block_id, 0),
 552   _loop_end_blocks(8),
 553   _work_list(8),
 554   _linear_scan_order(NULL), // initialized later with correct size
 555   _loop_map(0, 0),          // initialized later with correct size
 556   _compilation(c)
 557 {
 558   TRACE_LINEAR_SCAN(2, "***** computing linear-scan block order");
 559 
 560   init_visited();
 561   count_edges(start_block, NULL);
 562 
 563   if (compilation()->is_profiling()) {
 564     compilation()->method()->method_data()->set_compilation_stats(_num_loops, _num_blocks);
 565   }
 566 
 567   if (_num_loops > 0) {
 568     mark_loops();
 569     clear_non_natural_loops(start_block);
 570     assign_loop_depth(start_block);
 571   }
 572 
 573   compute_order(start_block);
 574   compute_dominators();
 575 
 576   NOT_PRODUCT(print_blocks());
 577   DEBUG_ONLY(verify());
 578 }
 579 
 580 
 581 // Traverse the CFG:
 582 // * count total number of blocks
 583 // * count all incoming edges and backward incoming edges
 584 // * number loop header blocks
 585 // * create a list with all loop end blocks
 586 void ComputeLinearScanOrder::count_edges(BlockBegin* cur, BlockBegin* parent) {
 587   TRACE_LINEAR_SCAN(3, tty->print_cr("Enter count_edges for block B%d coming from B%d", cur->block_id(), parent != NULL ? parent->block_id() : -1));
 588   assert(cur->dominator() == NULL, "dominator already initialized");
 589 
 590   if (is_active(cur)) {
 591     TRACE_LINEAR_SCAN(3, tty->print_cr("backward branch"));
 592     assert(is_visited(cur), "block must be visisted when block is active");
 593     assert(parent != NULL, "must have parent");
 594 
 595     cur->set(BlockBegin::linear_scan_loop_header_flag);
 596     cur->set(BlockBegin::backward_branch_target_flag);
 597 
 598     parent->set(BlockBegin::linear_scan_loop_end_flag);
 599 
 600     // When a loop header is also the start of an exception handler, then the backward branch is
 601     // an exception edge. Because such edges are usually critical edges which cannot be split, the
 602     // loop must be excluded here from processing.
 603     if (cur->is_set(BlockBegin::exception_entry_flag)) {
 604       // Make sure that dominators are correct in this weird situation
 605       _iterative_dominators = true;
 606       return;
 607     }
 608     assert(parent->number_of_sux() == 1 && parent->sux_at(0) == cur,
 609            "loop end blocks must have one successor (critical edges are split)");
 610 
 611     _loop_end_blocks.append(parent);
 612     return;
 613   }
 614 
 615   // increment number of incoming forward branches
 616   inc_forward_branches(cur);
 617 
 618   if (is_visited(cur)) {
 619     TRACE_LINEAR_SCAN(3, tty->print_cr("block already visited"));
 620     return;
 621   }
 622 
 623   _num_blocks++;
 624   set_visited(cur);
 625   set_active(cur);
 626 
 627   // recursive call for all successors
 628   int i;
 629   for (i = cur->number_of_sux() - 1; i >= 0; i--) {
 630     count_edges(cur->sux_at(i), cur);
 631   }
 632   for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
 633     count_edges(cur->exception_handler_at(i), cur);
 634   }
 635 
 636   clear_active(cur);
 637 
 638   // Each loop has a unique number.
 639   // When multiple loops are nested, assign_loop_depth assumes that the
 640   // innermost loop has the lowest number. This is guaranteed by setting
 641   // the loop number after the recursive calls for the successors above
 642   // have returned.
 643   if (cur->is_set(BlockBegin::linear_scan_loop_header_flag)) {
 644     assert(cur->loop_index() == -1, "cannot set loop-index twice");
 645     TRACE_LINEAR_SCAN(3, tty->print_cr("Block B%d is loop header of loop %d", cur->block_id(), _num_loops));
 646 
 647     cur->set_loop_index(_num_loops);
 648     _num_loops++;
 649   }
 650 
 651   TRACE_LINEAR_SCAN(3, tty->print_cr("Finished count_edges for block B%d", cur->block_id()));
 652 }
 653 
 654 
 655 void ComputeLinearScanOrder::mark_loops() {
 656   TRACE_LINEAR_SCAN(3, tty->print_cr("----- marking loops"));
 657 
 658   _loop_map = BitMap2D(_num_loops, _max_block_id);
 659   _loop_map.clear();
 660 
 661   for (int i = _loop_end_blocks.length() - 1; i >= 0; i--) {
 662     BlockBegin* loop_end   = _loop_end_blocks.at(i);
 663     BlockBegin* loop_start = loop_end->sux_at(0);
 664     int         loop_idx   = loop_start->loop_index();
 665 
 666     TRACE_LINEAR_SCAN(3, tty->print_cr("Processing loop from B%d to B%d (loop %d):", loop_start->block_id(), loop_end->block_id(), loop_idx));
 667     assert(loop_end->is_set(BlockBegin::linear_scan_loop_end_flag), "loop end flag must be set");
 668     assert(loop_end->number_of_sux() == 1, "incorrect number of successors");
 669     assert(loop_start->is_set(BlockBegin::linear_scan_loop_header_flag), "loop header flag must be set");
 670     assert(loop_idx >= 0 && loop_idx < _num_loops, "loop index not set");
 671     assert(_work_list.is_empty(), "work list must be empty before processing");
 672 
 673     // add the end-block of the loop to the working list
 674     _work_list.push(loop_end);
 675     set_block_in_loop(loop_idx, loop_end);
 676     do {
 677       BlockBegin* cur = _work_list.pop();
 678 
 679       TRACE_LINEAR_SCAN(3, tty->print_cr("    processing B%d", cur->block_id()));
 680       assert(is_block_in_loop(loop_idx, cur), "bit in loop map must be set when block is in work list");
 681 
 682       // recursive processing of all predecessors ends when start block of loop is reached
 683       if (cur != loop_start && !cur->is_set(BlockBegin::osr_entry_flag)) {
 684         for (int j = cur->number_of_preds() - 1; j >= 0; j--) {
 685           BlockBegin* pred = cur->pred_at(j);
 686 
 687           if (!is_block_in_loop(loop_idx, pred) /*&& !pred->is_set(BlockBeginosr_entry_flag)*/) {
 688             // this predecessor has not been processed yet, so add it to work list
 689             TRACE_LINEAR_SCAN(3, tty->print_cr("    pushing B%d", pred->block_id()));
 690             _work_list.push(pred);
 691             set_block_in_loop(loop_idx, pred);
 692           }
 693         }
 694       }
 695     } while (!_work_list.is_empty());
 696   }
 697 }
 698 
 699 
 700 // check for non-natural loops (loops where the loop header does not dominate
 701 // all other loop blocks = loops with mulitple entries).
 702 // such loops are ignored
 703 void ComputeLinearScanOrder::clear_non_natural_loops(BlockBegin* start_block) {
 704   for (int i = _num_loops - 1; i >= 0; i--) {
 705     if (is_block_in_loop(i, start_block)) {
 706       // loop i contains the entry block of the method
 707       // -> this is not a natural loop, so ignore it
 708       TRACE_LINEAR_SCAN(2, tty->print_cr("Loop %d is non-natural, so it is ignored", i));
 709 
 710       for (int block_id = _max_block_id - 1; block_id >= 0; block_id--) {
 711         clear_block_in_loop(i, block_id);
 712       }
 713       _iterative_dominators = true;
 714     }
 715   }
 716 }
 717 
 718 void ComputeLinearScanOrder::assign_loop_depth(BlockBegin* start_block) {
 719   TRACE_LINEAR_SCAN(3, "----- computing loop-depth and weight");
 720   init_visited();
 721 
 722   assert(_work_list.is_empty(), "work list must be empty before processing");
 723   _work_list.append(start_block);
 724 
 725   do {
 726     BlockBegin* cur = _work_list.pop();
 727 
 728     if (!is_visited(cur)) {
 729       set_visited(cur);
 730       TRACE_LINEAR_SCAN(4, tty->print_cr("Computing loop depth for block B%d", cur->block_id()));
 731 
 732       // compute loop-depth and loop-index for the block
 733       assert(cur->loop_depth() == 0, "cannot set loop-depth twice");
 734       int i;
 735       int loop_depth = 0;
 736       int min_loop_idx = -1;
 737       for (i = _num_loops - 1; i >= 0; i--) {
 738         if (is_block_in_loop(i, cur)) {
 739           loop_depth++;
 740           min_loop_idx = i;
 741         }
 742       }
 743       cur->set_loop_depth(loop_depth);
 744       cur->set_loop_index(min_loop_idx);
 745 
 746       // append all unvisited successors to work list
 747       for (i = cur->number_of_sux() - 1; i >= 0; i--) {
 748         _work_list.append(cur->sux_at(i));
 749       }
 750       for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
 751         _work_list.append(cur->exception_handler_at(i));
 752       }
 753     }
 754   } while (!_work_list.is_empty());
 755 }
 756 
 757 
 758 BlockBegin* ComputeLinearScanOrder::common_dominator(BlockBegin* a, BlockBegin* b) {
 759   assert(a != NULL && b != NULL, "must have input blocks");
 760 
 761   _dominator_blocks.clear();
 762   while (a != NULL) {
 763     _dominator_blocks.set_bit(a->block_id());
 764     assert(a->dominator() != NULL || a == _linear_scan_order->at(0), "dominator must be initialized");
 765     a = a->dominator();
 766   }
 767   while (b != NULL && !_dominator_blocks.at(b->block_id())) {
 768     assert(b->dominator() != NULL || b == _linear_scan_order->at(0), "dominator must be initialized");
 769     b = b->dominator();
 770   }
 771 
 772   assert(b != NULL, "could not find dominator");
 773   return b;
 774 }
 775 
 776 void ComputeLinearScanOrder::compute_dominator(BlockBegin* cur, BlockBegin* parent) {
 777   if (cur->dominator() == NULL) {
 778     TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: initializing dominator of B%d to B%d", cur->block_id(), parent->block_id()));
 779     cur->set_dominator(parent);
 780 
 781   } else if (!(cur->is_set(BlockBegin::linear_scan_loop_header_flag) && parent->is_set(BlockBegin::linear_scan_loop_end_flag))) {
 782     TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: computing dominator of B%d: common dominator of B%d and B%d is B%d", cur->block_id(), parent->block_id(), cur->dominator()->block_id(), common_dominator(cur->dominator(), parent)->block_id()));
 783     assert(cur->number_of_preds() > 1, "");
 784     cur->set_dominator(common_dominator(cur->dominator(), parent));
 785   }
 786 }
 787 
 788 
 789 int ComputeLinearScanOrder::compute_weight(BlockBegin* cur) {
 790   BlockBegin* single_sux = NULL;
 791   if (cur->number_of_sux() == 1) {
 792     single_sux = cur->sux_at(0);
 793   }
 794 
 795   // limit loop-depth to 15 bit (only for security reason, it will never be so big)
 796   int weight = (cur->loop_depth() & 0x7FFF) << 16;
 797 
 798   // general macro for short definition of weight flags
 799   // the first instance of INC_WEIGHT_IF has the highest priority
 800   int cur_bit = 15;
 801   #define INC_WEIGHT_IF(condition) if ((condition)) { weight |= (1 << cur_bit); } cur_bit--;
 802 
 803   // this is necessery for the (very rare) case that two successing blocks have
 804   // the same loop depth, but a different loop index (can happen for endless loops
 805   // with exception handlers)
 806   INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_header_flag));
 807 
 808   // loop end blocks (blocks that end with a backward branch) are added
 809   // after all other blocks of the loop.
 810   INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_end_flag));
 811 
 812   // critical edge split blocks are prefered because than they have a bigger
 813   // proability to be completely empty
 814   INC_WEIGHT_IF(cur->is_set(BlockBegin::critical_edge_split_flag));
 815 
 816   // exceptions should not be thrown in normal control flow, so these blocks
 817   // are added as late as possible
 818   INC_WEIGHT_IF(cur->end()->as_Throw() == NULL  && (single_sux == NULL || single_sux->end()->as_Throw()  == NULL));
 819   INC_WEIGHT_IF(cur->end()->as_Return() == NULL && (single_sux == NULL || single_sux->end()->as_Return() == NULL));
 820 
 821   // exceptions handlers are added as late as possible
 822   INC_WEIGHT_IF(!cur->is_set(BlockBegin::exception_entry_flag));
 823 
 824   // guarantee that weight is > 0
 825   weight |= 1;
 826 
 827   #undef INC_WEIGHT_IF
 828   assert(cur_bit >= 0, "too many flags");
 829   assert(weight > 0, "weight cannot become negative");
 830 
 831   return weight;
 832 }
 833 
 834 bool ComputeLinearScanOrder::ready_for_processing(BlockBegin* cur) {
 835   // Discount the edge just traveled.
 836   // When the number drops to zero, all forward branches were processed
 837   if (dec_forward_branches(cur) != 0) {
 838     return false;
 839   }
 840 
 841   assert(_linear_scan_order->index_of(cur) == -1, "block already processed (block can be ready only once)");
 842   assert(_work_list.index_of(cur) == -1, "block already in work-list (block can be ready only once)");
 843   return true;
 844 }
 845 
 846 void ComputeLinearScanOrder::sort_into_work_list(BlockBegin* cur) {
 847   assert(_work_list.index_of(cur) == -1, "block already in work list");
 848 
 849   int cur_weight = compute_weight(cur);
 850 
 851   // the linear_scan_number is used to cache the weight of a block
 852   cur->set_linear_scan_number(cur_weight);
 853 
 854 #ifndef PRODUCT
 855   if (StressLinearScan) {
 856     _work_list.insert_before(0, cur);
 857     return;
 858   }
 859 #endif
 860 
 861   _work_list.append(NULL); // provide space for new element
 862 
 863   int insert_idx = _work_list.length() - 1;
 864   while (insert_idx > 0 && _work_list.at(insert_idx - 1)->linear_scan_number() > cur_weight) {
 865     _work_list.at_put(insert_idx, _work_list.at(insert_idx - 1));
 866     insert_idx--;
 867   }
 868   _work_list.at_put(insert_idx, cur);
 869 
 870   TRACE_LINEAR_SCAN(3, tty->print_cr("Sorted B%d into worklist. new worklist:", cur->block_id()));
 871   TRACE_LINEAR_SCAN(3, for (int i = 0; i < _work_list.length(); i++) tty->print_cr("%8d B%2d  weight:%6x", i, _work_list.at(i)->block_id(), _work_list.at(i)->linear_scan_number()));
 872 
 873 #ifdef ASSERT
 874   for (int i = 0; i < _work_list.length(); i++) {
 875     assert(_work_list.at(i)->linear_scan_number() > 0, "weight not set");
 876     assert(i == 0 || _work_list.at(i - 1)->linear_scan_number() <= _work_list.at(i)->linear_scan_number(), "incorrect order in worklist");
 877   }
 878 #endif
 879 }
 880 
 881 void ComputeLinearScanOrder::append_block(BlockBegin* cur) {
 882   TRACE_LINEAR_SCAN(3, tty->print_cr("appending block B%d (weight 0x%6x) to linear-scan order", cur->block_id(), cur->linear_scan_number()));
 883   assert(_linear_scan_order->index_of(cur) == -1, "cannot add the same block twice");
 884 
 885   // currently, the linear scan order and code emit order are equal.
 886   // therefore the linear_scan_number and the weight of a block must also
 887   // be equal.
 888   cur->set_linear_scan_number(_linear_scan_order->length());
 889   _linear_scan_order->append(cur);
 890 }
 891 
 892 void ComputeLinearScanOrder::compute_order(BlockBegin* start_block) {
 893   TRACE_LINEAR_SCAN(3, "----- computing final block order");
 894 
 895   // the start block is always the first block in the linear scan order
 896   _linear_scan_order = new BlockList(_num_blocks);
 897   append_block(start_block);
 898 
 899   assert(start_block->end()->as_Base() != NULL, "start block must end with Base-instruction");
 900   BlockBegin* std_entry = ((Base*)start_block->end())->std_entry();
 901   BlockBegin* osr_entry = ((Base*)start_block->end())->osr_entry();
 902 
 903   BlockBegin* sux_of_osr_entry = NULL;
 904   if (osr_entry != NULL) {
 905     // special handling for osr entry:
 906     // ignore the edge between the osr entry and its successor for processing
 907     // the osr entry block is added manually below
 908     assert(osr_entry->number_of_sux() == 1, "osr entry must have exactly one successor");
 909     assert(osr_entry->sux_at(0)->number_of_preds() >= 2, "sucessor of osr entry must have two predecessors (otherwise it is not present in normal control flow");
 910 
 911     sux_of_osr_entry = osr_entry->sux_at(0);
 912     dec_forward_branches(sux_of_osr_entry);
 913 
 914     compute_dominator(osr_entry, start_block);
 915     _iterative_dominators = true;
 916   }
 917   compute_dominator(std_entry, start_block);
 918 
 919   // start processing with standard entry block
 920   assert(_work_list.is_empty(), "list must be empty before processing");
 921 
 922   if (ready_for_processing(std_entry)) {
 923     sort_into_work_list(std_entry);
 924   } else {
 925     assert(false, "the std_entry must be ready for processing (otherwise, the method has no start block)");
 926   }
 927 
 928   do {
 929     BlockBegin* cur = _work_list.pop();
 930 
 931     if (cur == sux_of_osr_entry) {
 932       // the osr entry block is ignored in normal processing, it is never added to the
 933       // work list. Instead, it is added as late as possible manually here.
 934       append_block(osr_entry);
 935       compute_dominator(cur, osr_entry);
 936     }
 937     append_block(cur);
 938 
 939     int i;
 940     int num_sux = cur->number_of_sux();
 941     // changed loop order to get "intuitive" order of if- and else-blocks
 942     for (i = 0; i < num_sux; i++) {
 943       BlockBegin* sux = cur->sux_at(i);
 944       compute_dominator(sux, cur);
 945       if (ready_for_processing(sux)) {
 946         sort_into_work_list(sux);
 947       }
 948     }
 949     num_sux = cur->number_of_exception_handlers();
 950     for (i = 0; i < num_sux; i++) {
 951       BlockBegin* sux = cur->exception_handler_at(i);
 952       compute_dominator(sux, cur);
 953       if (ready_for_processing(sux)) {
 954         sort_into_work_list(sux);
 955       }
 956     }
 957   } while (_work_list.length() > 0);
 958 }
 959 
 960 
 961 bool ComputeLinearScanOrder::compute_dominators_iter() {
 962   bool changed = false;
 963   int num_blocks = _linear_scan_order->length();
 964 
 965   assert(_linear_scan_order->at(0)->dominator() == NULL, "must not have dominator");
 966   assert(_linear_scan_order->at(0)->number_of_preds() == 0, "must not have predecessors");
 967   for (int i = 1; i < num_blocks; i++) {
 968     BlockBegin* block = _linear_scan_order->at(i);
 969 
 970     BlockBegin* dominator = block->pred_at(0);
 971     int num_preds = block->number_of_preds();
 972     for (int i = 1; i < num_preds; i++) {
 973       dominator = common_dominator(dominator, block->pred_at(i));
 974     }
 975 
 976     if (dominator != block->dominator()) {
 977       TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: updating dominator of B%d from B%d to B%d", block->block_id(), block->dominator()->block_id(), dominator->block_id()));
 978 
 979       block->set_dominator(dominator);
 980       changed = true;
 981     }
 982   }
 983   return changed;
 984 }
 985 
 986 void ComputeLinearScanOrder::compute_dominators() {
 987   TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing dominators (iterative computation reqired: %d)", _iterative_dominators));
 988 
 989   // iterative computation of dominators is only required for methods with non-natural loops
 990   // and OSR-methods. For all other methods, the dominators computed when generating the
 991   // linear scan block order are correct.
 992   if (_iterative_dominators) {
 993     do {
 994       TRACE_LINEAR_SCAN(1, tty->print_cr("DOM: next iteration of fix-point calculation"));
 995     } while (compute_dominators_iter());
 996   }
 997 
 998   // check that dominators are correct
 999   assert(!compute_dominators_iter(), "fix point not reached");
1000 }
1001 
1002 
1003 #ifndef PRODUCT
1004 void ComputeLinearScanOrder::print_blocks() {
1005   if (TraceLinearScanLevel >= 2) {
1006     tty->print_cr("----- loop information:");
1007     for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
1008       BlockBegin* cur = _linear_scan_order->at(block_idx);
1009 
1010       tty->print("%4d: B%2d: ", cur->linear_scan_number(), cur->block_id());
1011       for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
1012         tty->print ("%d ", is_block_in_loop(loop_idx, cur));
1013       }
1014       tty->print_cr(" -> loop_index: %2d, loop_depth: %2d", cur->loop_index(), cur->loop_depth());
1015     }
1016   }
1017 
1018   if (TraceLinearScanLevel >= 1) {
1019     tty->print_cr("----- linear-scan block order:");
1020     for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
1021       BlockBegin* cur = _linear_scan_order->at(block_idx);
1022       tty->print("%4d: B%2d    loop: %2d  depth: %2d", cur->linear_scan_number(), cur->block_id(), cur->loop_index(), cur->loop_depth());
1023 
1024       tty->print(cur->is_set(BlockBegin::exception_entry_flag)         ? " ex" : "   ");
1025       tty->print(cur->is_set(BlockBegin::critical_edge_split_flag)     ? " ce" : "   ");
1026       tty->print(cur->is_set(BlockBegin::linear_scan_loop_header_flag) ? " lh" : "   ");
1027       tty->print(cur->is_set(BlockBegin::linear_scan_loop_end_flag)    ? " le" : "   ");
1028 
1029       if (cur->dominator() != NULL) {
1030         tty->print("    dom: B%d ", cur->dominator()->block_id());
1031       } else {
1032         tty->print("    dom: NULL ");
1033       }
1034 
1035       if (cur->number_of_preds() > 0) {
1036         tty->print("    preds: ");
1037         for (int j = 0; j < cur->number_of_preds(); j++) {
1038           BlockBegin* pred = cur->pred_at(j);
1039           tty->print("B%d ", pred->block_id());
1040         }
1041       }
1042       if (cur->number_of_sux() > 0) {
1043         tty->print("    sux: ");
1044         for (int j = 0; j < cur->number_of_sux(); j++) {
1045           BlockBegin* sux = cur->sux_at(j);
1046           tty->print("B%d ", sux->block_id());
1047         }
1048       }
1049       if (cur->number_of_exception_handlers() > 0) {
1050         tty->print("    ex: ");
1051         for (int j = 0; j < cur->number_of_exception_handlers(); j++) {
1052           BlockBegin* ex = cur->exception_handler_at(j);
1053           tty->print("B%d ", ex->block_id());
1054         }
1055       }
1056       tty->cr();
1057     }
1058   }
1059 }
1060 #endif
1061 
1062 #ifdef ASSERT
1063 void ComputeLinearScanOrder::verify() {
1064   assert(_linear_scan_order->length() == _num_blocks, "wrong number of blocks in list");
1065 
1066   if (StressLinearScan) {
1067     // blocks are scrambled when StressLinearScan is used
1068     return;
1069   }
1070 
1071   // check that all successors of a block have a higher linear-scan-number
1072   // and that all predecessors of a block have a lower linear-scan-number
1073   // (only backward branches of loops are ignored)
1074   int i;
1075   for (i = 0; i < _linear_scan_order->length(); i++) {
1076     BlockBegin* cur = _linear_scan_order->at(i);
1077 
1078     assert(cur->linear_scan_number() == i, "incorrect linear_scan_number");
1079     assert(cur->linear_scan_number() >= 0 && cur->linear_scan_number() == _linear_scan_order->index_of(cur), "incorrect linear_scan_number");
1080 
1081     int j;
1082     for (j = cur->number_of_sux() - 1; j >= 0; j--) {
1083       BlockBegin* sux = cur->sux_at(j);
1084 
1085       assert(sux->linear_scan_number() >= 0 && sux->linear_scan_number() == _linear_scan_order->index_of(sux), "incorrect linear_scan_number");
1086       if (!cur->is_set(BlockBegin::linear_scan_loop_end_flag)) {
1087         assert(cur->linear_scan_number() < sux->linear_scan_number(), "invalid order");
1088       }
1089       if (cur->loop_depth() == sux->loop_depth()) {
1090         assert(cur->loop_index() == sux->loop_index() || sux->is_set(BlockBegin::linear_scan_loop_header_flag), "successing blocks with same loop depth must have same loop index");
1091       }
1092     }
1093 
1094     for (j = cur->number_of_preds() - 1; j >= 0; j--) {
1095       BlockBegin* pred = cur->pred_at(j);
1096 
1097       assert(pred->linear_scan_number() >= 0 && pred->linear_scan_number() == _linear_scan_order->index_of(pred), "incorrect linear_scan_number");
1098       if (!cur->is_set(BlockBegin::linear_scan_loop_header_flag)) {
1099         assert(cur->linear_scan_number() > pred->linear_scan_number(), "invalid order");
1100       }
1101       if (cur->loop_depth() == pred->loop_depth()) {
1102         assert(cur->loop_index() == pred->loop_index() || cur->is_set(BlockBegin::linear_scan_loop_header_flag), "successing blocks with same loop depth must have same loop index");
1103       }
1104 
1105       assert(cur->dominator()->linear_scan_number() <= cur->pred_at(j)->linear_scan_number(), "dominator must be before predecessors");
1106     }
1107 
1108     // check dominator
1109     if (i == 0) {
1110       assert(cur->dominator() == NULL, "first block has no dominator");
1111     } else {
1112       assert(cur->dominator() != NULL, "all but first block must have dominator");
1113     }
1114     assert(cur->number_of_preds() != 1 || cur->dominator() == cur->pred_at(0), "Single predecessor must also be dominator");
1115   }
1116 
1117   // check that all loops are continuous
1118   for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
1119     int block_idx = 0;
1120     assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "the first block must not be present in any loop");
1121 
1122     // skip blocks before the loop
1123     while (block_idx < _num_blocks && !is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
1124       block_idx++;
1125     }
1126     // skip blocks of loop
1127     while (block_idx < _num_blocks && is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
1128       block_idx++;
1129     }
1130     // after the first non-loop block, there must not be another loop-block
1131     while (block_idx < _num_blocks) {
1132       assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "loop not continuous in linear-scan order");
1133       block_idx++;
1134     }
1135   }
1136 }
1137 #endif
1138 
1139 
1140 void IR::compute_code() {
1141   assert(is_valid(), "IR must be valid");
1142 
1143   ComputeLinearScanOrder compute_order(compilation(), start());
1144   _num_loops = compute_order.num_loops();
1145   _code = compute_order.linear_scan_order();
1146 }
1147 
1148 
1149 void IR::compute_use_counts() {
1150   // make sure all values coming out of this block get evaluated.
1151   int num_blocks = _code->length();
1152   for (int i = 0; i < num_blocks; i++) {
1153     _code->at(i)->end()->state()->pin_stack_for_linear_scan();
1154   }
1155 
1156   // compute use counts
1157   UseCountComputer::compute(_code);
1158 }
1159 
1160 
1161 void IR::iterate_preorder(BlockClosure* closure) {
1162   assert(is_valid(), "IR must be valid");
1163   start()->iterate_preorder(closure);
1164 }
1165 
1166 
1167 void IR::iterate_postorder(BlockClosure* closure) {
1168   assert(is_valid(), "IR must be valid");
1169   start()->iterate_postorder(closure);
1170 }
1171 
1172 void IR::iterate_linear_scan_order(BlockClosure* closure) {
1173   linear_scan_order()->iterate_forward(closure);
1174 }
1175 
1176 
1177 #ifndef PRODUCT
1178 class BlockPrinter: public BlockClosure {
1179  private:
1180   InstructionPrinter* _ip;
1181   bool                _cfg_only;
1182   bool                _live_only;
1183 
1184  public:
1185   BlockPrinter(InstructionPrinter* ip, bool cfg_only, bool live_only = false) {
1186     _ip       = ip;
1187     _cfg_only = cfg_only;
1188     _live_only = live_only;
1189   }
1190 
1191   virtual void block_do(BlockBegin* block) {
1192     if (_cfg_only) {
1193       _ip->print_instr(block); tty->cr();
1194     } else {
1195       block->print_block(*_ip, _live_only);
1196     }
1197   }
1198 };
1199 
1200 
1201 void IR::print(BlockBegin* start, bool cfg_only, bool live_only) {
1202   ttyLocker ttyl;
1203   InstructionPrinter ip(!cfg_only);
1204   BlockPrinter bp(&ip, cfg_only, live_only);
1205   start->iterate_preorder(&bp);
1206   tty->cr();
1207 }
1208 
1209 void IR::print(bool cfg_only, bool live_only) {
1210   if (is_valid()) {
1211     print(start(), cfg_only, live_only);
1212   } else {
1213     tty->print_cr("invalid IR");
1214   }
1215 }
1216 
1217 
1218 define_array(BlockListArray, BlockList*)
1219 define_stack(BlockListList, BlockListArray)
1220 
1221 class PredecessorValidator : public BlockClosure {
1222  private:
1223   BlockListList* _predecessors;
1224   BlockList*     _blocks;
1225 
1226   static int cmp(BlockBegin** a, BlockBegin** b) {
1227     return (*a)->block_id() - (*b)->block_id();
1228   }
1229 
1230  public:
1231   PredecessorValidator(IR* hir) {
1232     ResourceMark rm;
1233     _predecessors = new BlockListList(BlockBegin::number_of_blocks(), NULL);
1234     _blocks = new BlockList();
1235 
1236     int i;
1237     hir->start()->iterate_preorder(this);
1238     if (hir->code() != NULL) {
1239       assert(hir->code()->length() == _blocks->length(), "must match");
1240       for (i = 0; i < _blocks->length(); i++) {
1241         assert(hir->code()->contains(_blocks->at(i)), "should be in both lists");
1242       }
1243     }
1244 
1245     for (i = 0; i < _blocks->length(); i++) {
1246       BlockBegin* block = _blocks->at(i);
1247       BlockList* preds = _predecessors->at(block->block_id());
1248       if (preds == NULL) {
1249         assert(block->number_of_preds() == 0, "should be the same");
1250         continue;
1251       }
1252 
1253       // clone the pred list so we can mutate it
1254       BlockList* pred_copy = new BlockList();
1255       int j;
1256       for (j = 0; j < block->number_of_preds(); j++) {
1257         pred_copy->append(block->pred_at(j));
1258       }
1259       // sort them in the same order
1260       preds->sort(cmp);
1261       pred_copy->sort(cmp);
1262       int length = MIN2(preds->length(), block->number_of_preds());
1263       for (j = 0; j < block->number_of_preds(); j++) {
1264         assert(preds->at(j) == pred_copy->at(j), "must match");
1265       }
1266 
1267       assert(preds->length() == block->number_of_preds(), "should be the same");
1268     }
1269   }
1270 
1271   virtual void block_do(BlockBegin* block) {
1272     _blocks->append(block);
1273     BlockEnd* be = block->end();
1274     int n = be->number_of_sux();
1275     int i;
1276     for (i = 0; i < n; i++) {
1277       BlockBegin* sux = be->sux_at(i);
1278       assert(!sux->is_set(BlockBegin::exception_entry_flag), "must not be xhandler");
1279 
1280       BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
1281       if (preds == NULL) {
1282         preds = new BlockList();
1283         _predecessors->at_put(sux->block_id(), preds);
1284       }
1285       preds->append(block);
1286     }
1287 
1288     n = block->number_of_exception_handlers();
1289     for (i = 0; i < n; i++) {
1290       BlockBegin* sux = block->exception_handler_at(i);
1291       assert(sux->is_set(BlockBegin::exception_entry_flag), "must be xhandler");
1292 
1293       BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
1294       if (preds == NULL) {
1295         preds = new BlockList();
1296         _predecessors->at_put(sux->block_id(), preds);
1297       }
1298       preds->append(block);
1299     }
1300   }
1301 };
1302 
1303 void IR::verify() {
1304 #ifdef ASSERT
1305   PredecessorValidator pv(this);
1306 #endif
1307 }
1308 
1309 #endif // PRODUCT
1310 
1311 void SubstitutionResolver::visit(Value* v) {
1312   Value v0 = *v;
1313   if (v0) {
1314     Value vs = v0->subst();
1315     if (vs != v0) {
1316       *v = v0->subst();
1317     }
1318   }
1319 }
1320 
1321 #ifdef ASSERT
1322 class SubstitutionChecker: public ValueVisitor {
1323   void visit(Value* v) {
1324     Value v0 = *v;
1325     if (v0) {
1326       Value vs = v0->subst();
1327       assert(vs == v0, "missed substitution");
1328     }
1329   }
1330 };
1331 #endif
1332 
1333 
1334 void SubstitutionResolver::block_do(BlockBegin* block) {
1335   Instruction* last = NULL;
1336   for (Instruction* n = block; n != NULL;) {
1337     n->values_do(this);
1338     // need to remove this instruction from the instruction stream
1339     if (n->subst() != n) {
1340       assert(last != NULL, "must have last");
1341       last->set_next(n->next(), n->next()->bci());
1342     } else {
1343       last = n;
1344     }
1345     n = last->next();
1346   }
1347 
1348 #ifdef ASSERT
1349   SubstitutionChecker check_substitute;
1350   if (block->state()) block->state()->values_do(&check_substitute);
1351   block->block_values_do(&check_substitute);
1352   if (block->end() && block->end()->state()) block->end()->state()->values_do(&check_substitute);
1353 #endif
1354 }