1 /*
   2  * Copyright (c) 1999, 2016, 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 "c1/c1_Compilation.hpp"
  27 #include "c1/c1_FrameMap.hpp"
  28 #include "c1/c1_GraphBuilder.hpp"
  29 #include "c1/c1_IR.hpp"
  30 #include "c1/c1_InstructionPrinter.hpp"
  31 #include "c1/c1_Optimizer.hpp"
  32 #include "utilities/bitMap.inline.hpp"
  33 
  34 
  35 // Implementation of XHandlers
  36 //
  37 // Note: This code could eventually go away if we are
  38 //       just using the ciExceptionHandlerStream.
  39 
  40 XHandlers::XHandlers(ciMethod* method) : _list(method->exception_table_length()) {
  41   ciExceptionHandlerStream s(method);
  42   while (!s.is_done()) {
  43     _list.append(new XHandler(s.handler()));
  44     s.next();
  45   }
  46   assert(s.count() == method->exception_table_length(), "exception table lengths inconsistent");
  47 }
  48 
  49 // deep copy of all XHandler contained in list
  50 XHandlers::XHandlers(XHandlers* other) :
  51   _list(other->length())
  52 {
  53   for (int i = 0; i < other->length(); i++) {
  54     _list.append(new XHandler(other->handler_at(i)));
  55   }
  56 }
  57 
  58 // Returns whether a particular exception type can be caught.  Also
  59 // returns true if klass is unloaded or any exception handler
  60 // classes are unloaded.  type_is_exact indicates whether the throw
  61 // is known to be exactly that class or it might throw a subtype.
  62 bool XHandlers::could_catch(ciInstanceKlass* klass, bool type_is_exact) const {
  63   // the type is unknown so be conservative
  64   if (!klass->is_loaded()) {
  65     return true;
  66   }
  67 
  68   for (int i = 0; i < length(); i++) {
  69     XHandler* handler = handler_at(i);
  70     if (handler->is_catch_all()) {
  71       // catch of ANY
  72       return true;
  73     }
  74     ciInstanceKlass* handler_klass = handler->catch_klass();
  75     // if it's unknown it might be catchable
  76     if (!handler_klass->is_loaded()) {
  77       return true;
  78     }
  79     // if the throw type is definitely a subtype of the catch type
  80     // then it can be caught.
  81     if (klass->is_subtype_of(handler_klass)) {
  82       return true;
  83     }
  84     if (!type_is_exact) {
  85       // If the type isn't exactly known then it can also be caught by
  86       // catch statements where the inexact type is a subtype of the
  87       // catch type.
  88       // given: foo extends bar extends Exception
  89       // throw bar can be caught by catch foo, catch bar, and catch
  90       // Exception, however it can't be caught by any handlers without
  91       // bar in its type hierarchy.
  92       if (handler_klass->is_subtype_of(klass)) {
  93         return true;
  94       }
  95     }
  96   }
  97 
  98   return false;
  99 }
 100 
 101 
 102 bool XHandlers::equals(XHandlers* others) const {
 103   if (others == NULL) return false;
 104   if (length() != others->length()) return false;
 105 
 106   for (int i = 0; i < length(); i++) {
 107     if (!handler_at(i)->equals(others->handler_at(i))) return false;
 108   }
 109   return true;
 110 }
 111 
 112 bool XHandler::equals(XHandler* other) const {
 113   assert(entry_pco() != -1 && other->entry_pco() != -1, "must have entry_pco");
 114 
 115   if (entry_pco() != other->entry_pco()) return false;
 116   if (scope_count() != other->scope_count()) return false;
 117   if (_desc != other->_desc) return false;
 118 
 119   assert(entry_block() == other->entry_block(), "entry_block must be equal when entry_pco is equal");
 120   return true;
 121 }
 122 
 123 
 124 // Implementation of IRScope
 125 BlockBegin* IRScope::build_graph(Compilation* compilation, int osr_bci) {
 126   GraphBuilder gm(compilation, this);
 127   NOT_PRODUCT(if (PrintValueNumbering && Verbose) gm.print_stats());
 128   if (compilation->bailed_out()) return NULL;
 129   return gm.start();
 130 }
 131 
 132 
 133 IRScope::IRScope(Compilation* compilation, IRScope* caller, int caller_bci, ciMethod* method, int osr_bci, bool create_graph)
 134 : _callees(2)
 135 , _compilation(compilation)
 136 , _requires_phi_function(method->max_locals())
 137 {
 138   _caller             = caller;
 139   _level              = caller == NULL ?  0 : caller->level() + 1;
 140   _method             = method;
 141   _xhandlers          = new XHandlers(method);
 142   _number_of_locks    = 0;
 143   _monitor_pairing_ok = method->has_balanced_monitors();
 144   _wrote_final        = false;
 145   _wrote_fields       = false;
 146   _wrote_volatile     = false;
 147   _start              = NULL;
 148 
 149   if (osr_bci == -1) {
 150     _requires_phi_function.clear();
 151   } else {
 152         // selective creation of phi functions is not possibel in osr-methods
 153     _requires_phi_function.set_range(0, method->max_locals());
 154   }
 155 
 156   assert(method->holder()->is_loaded() , "method holder must be loaded");
 157 
 158   // build graph if monitor pairing is ok
 159   if (create_graph && monitor_pairing_ok()) _start = build_graph(compilation, osr_bci);
 160 }
 161 
 162 
 163 int IRScope::max_stack() const {
 164   int my_max = method()->max_stack();
 165   int callee_max = 0;
 166   for (int i = 0; i < number_of_callees(); i++) {
 167     callee_max = MAX2(callee_max, callee_no(i)->max_stack());
 168   }
 169   return my_max + callee_max;
 170 }
 171 
 172 
 173 bool IRScopeDebugInfo::should_reexecute() {
 174   ciMethod* cur_method = scope()->method();
 175   int       cur_bci    = bci();
 176   if (cur_method != NULL && cur_bci != SynchronizationEntryBCI) {
 177     Bytecodes::Code code = cur_method->java_code_at_bci(cur_bci);
 178     return Interpreter::bytecode_should_reexecute(code);
 179   } else
 180     return false;
 181 }
 182 
 183 
 184 // Implementation of CodeEmitInfo
 185 
 186 // Stack must be NON-null
 187 CodeEmitInfo::CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers, bool deoptimize_on_exception)
 188   : _scope(stack->scope())
 189   , _scope_debug_info(NULL)
 190   , _oop_map(NULL)
 191   , _stack(stack)
 192   , _exception_handlers(exception_handlers)
 193   , _is_method_handle_invoke(false)
 194   , _deoptimize_on_exception(deoptimize_on_exception) {
 195   assert(_stack != NULL, "must be non null");
 196 }
 197 
 198 
 199 CodeEmitInfo::CodeEmitInfo(CodeEmitInfo* info, ValueStack* stack)
 200   : _scope(info->_scope)
 201   , _exception_handlers(NULL)
 202   , _scope_debug_info(NULL)
 203   , _oop_map(NULL)
 204   , _stack(stack == NULL ? info->_stack : stack)
 205   , _is_method_handle_invoke(info->_is_method_handle_invoke)
 206   , _deoptimize_on_exception(info->_deoptimize_on_exception) {
 207 
 208   // deep copy of exception handlers
 209   if (info->_exception_handlers != NULL) {
 210     _exception_handlers = new XHandlers(info->_exception_handlers);
 211   }
 212 }
 213 
 214 
 215 void CodeEmitInfo::record_debug_info(DebugInformationRecorder* recorder, int pc_offset) {
 216   // record the safepoint before recording the debug info for enclosing scopes
 217   recorder->add_safepoint(pc_offset, _oop_map->deep_copy());
 218   _scope_debug_info->record_debug_info(recorder, pc_offset, true/*topmost*/, _is_method_handle_invoke);
 219   recorder->end_safepoint(pc_offset);
 220 }
 221 
 222 
 223 void CodeEmitInfo::add_register_oop(LIR_Opr opr) {
 224   assert(_oop_map != NULL, "oop map must already exist");
 225   assert(opr->is_single_cpu(), "should not call otherwise");
 226 
 227   VMReg name = frame_map()->regname(opr);
 228   _oop_map->set_oop(name);
 229 }
 230 
 231 // Mirror the stack size calculation in the deopt code
 232 // How much stack space would we need at this point in the program in
 233 // case of deoptimization?
 234 int CodeEmitInfo::interpreter_frame_size() const {
 235   ValueStack* state = _stack;
 236   int size = 0;
 237   int callee_parameters = 0;
 238   int callee_locals = 0;
 239   int extra_args = state->scope()->method()->max_stack() - state->stack_size();
 240 
 241   while (state != NULL) {
 242     int locks = state->locks_size();
 243     int temps = state->stack_size();
 244     bool is_top_frame = (state == _stack);
 245     ciMethod* method = state->scope()->method();
 246 
 247     int frame_size = BytesPerWord * Interpreter::size_activation(method->max_stack(),
 248                                                                  temps + callee_parameters,
 249                                                                  extra_args,
 250                                                                  locks,
 251                                                                  callee_parameters,
 252                                                                  callee_locals,
 253                                                                  is_top_frame);
 254     size += frame_size;
 255 
 256     callee_parameters = method->size_of_parameters();
 257     callee_locals = method->max_locals();
 258     extra_args = 0;
 259     state = state->caller_state();
 260   }
 261   return size + Deoptimization::last_frame_adjust(0, callee_locals) * BytesPerWord;
 262 }
 263 
 264 // Implementation of IR
 265 
 266 IR::IR(Compilation* compilation, ciMethod* method, int osr_bci) :
 267   _num_loops(0) {
 268   // setup IR fields
 269   _compilation = compilation;
 270   _top_scope   = new IRScope(compilation, NULL, -1, method, osr_bci, true);
 271   _code        = NULL;
 272 }
 273 
 274 
 275 void IR::optimize_blocks() {
 276   Optimizer opt(this);
 277   if (!compilation()->profile_branches()) {
 278     if (DoCEE) {
 279       opt.eliminate_conditional_expressions();
 280 #ifndef PRODUCT
 281       if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after CEE"); print(true); }
 282       if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after CEE"); print(false); }
 283 #endif
 284     }
 285     if (EliminateBlocks) {
 286       opt.eliminate_blocks();
 287 #ifndef PRODUCT
 288       if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after block elimination"); print(true); }
 289       if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after block elimination"); print(false); }
 290 #endif
 291     }
 292   }
 293 }
 294 
 295 void IR::eliminate_null_checks() {
 296   Optimizer opt(this);
 297   if (EliminateNullChecks) {
 298     opt.eliminate_null_checks();
 299 #ifndef PRODUCT
 300     if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after null check elimination"); print(true); }
 301     if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after null check elimination"); print(false); }
 302 #endif
 303   }
 304 }
 305 
 306 
 307 static int sort_pairs(BlockPair** a, BlockPair** b) {
 308   if ((*a)->from() == (*b)->from()) {
 309     return (*a)->to()->block_id() - (*b)->to()->block_id();
 310   } else {
 311     return (*a)->from()->block_id() - (*b)->from()->block_id();
 312   }
 313 }
 314 
 315 
 316 class CriticalEdgeFinder: public BlockClosure {
 317   BlockPairList blocks;
 318   IR*       _ir;
 319 
 320  public:
 321   CriticalEdgeFinder(IR* ir): _ir(ir) {}
 322   void block_do(BlockBegin* bb) {
 323     BlockEnd* be = bb->end();
 324     int nos = be->number_of_sux();
 325     if (nos >= 2) {
 326       for (int i = 0; i < nos; i++) {
 327         BlockBegin* sux = be->sux_at(i);
 328         if (sux->number_of_preds() >= 2) {
 329           blocks.append(new BlockPair(bb, sux));
 330         }
 331       }
 332     }
 333   }
 334 
 335   void split_edges() {
 336     BlockPair* last_pair = NULL;
 337     blocks.sort(sort_pairs);
 338     for (int i = 0; i < blocks.length(); i++) {
 339       BlockPair* pair = blocks.at(i);
 340       if (last_pair != NULL && pair->is_same(last_pair)) continue;
 341       BlockBegin* from = pair->from();
 342       BlockBegin* to = pair->to();
 343       BlockBegin* split = from->insert_block_between(to);
 344 #ifndef PRODUCT
 345       if ((PrintIR || PrintIR1) && Verbose) {
 346         tty->print_cr("Split critical edge B%d -> B%d (new block B%d)",
 347                       from->block_id(), to->block_id(), split->block_id());
 348       }
 349 #endif
 350       last_pair = pair;
 351     }
 352   }
 353 };
 354 
 355 void IR::split_critical_edges() {
 356   CriticalEdgeFinder cef(this);
 357 
 358   iterate_preorder(&cef);
 359   cef.split_edges();
 360 }
 361 
 362 
 363 class UseCountComputer: public ValueVisitor, BlockClosure {
 364  private:
 365   void visit(Value* n) {
 366     // Local instructions and Phis for expression stack values at the
 367     // start of basic blocks are not added to the instruction list
 368     if (!(*n)->is_linked() && (*n)->can_be_linked()) {
 369       assert(false, "a node was not appended to the graph");
 370       Compilation::current()->bailout("a node was not appended to the graph");
 371     }
 372     // use n's input if not visited before
 373     if (!(*n)->is_pinned() && !(*n)->has_uses()) {
 374       // note: a) if the instruction is pinned, it will be handled by compute_use_count
 375       //       b) if the instruction has uses, it was touched before
 376       //       => in both cases we don't need to update n's values
 377       uses_do(n);
 378     }
 379     // use n
 380     (*n)->_use_count++;
 381   }
 382 
 383   Values* worklist;
 384   int depth;
 385   enum {
 386     max_recurse_depth = 20
 387   };
 388 
 389   void uses_do(Value* n) {
 390     depth++;
 391     if (depth > max_recurse_depth) {
 392       // don't allow the traversal to recurse too deeply
 393       worklist->push(*n);
 394     } else {
 395       (*n)->input_values_do(this);
 396       // special handling for some instructions
 397       if ((*n)->as_BlockEnd() != NULL) {
 398         // note on BlockEnd:
 399         //   must 'use' the stack only if the method doesn't
 400         //   terminate, however, in those cases stack is empty
 401         (*n)->state_values_do(this);
 402       }
 403     }
 404     depth--;
 405   }
 406 
 407   void block_do(BlockBegin* b) {
 408     depth = 0;
 409     // process all pinned nodes as the roots of expression trees
 410     for (Instruction* n = b; n != NULL; n = n->next()) {
 411       if (n->is_pinned()) uses_do(&n);
 412     }
 413     assert(depth == 0, "should have counted back down");
 414 
 415     // now process any unpinned nodes which recursed too deeply
 416     while (worklist->length() > 0) {
 417       Value t = worklist->pop();
 418       if (!t->is_pinned()) {
 419         // compute the use count
 420         uses_do(&t);
 421 
 422         // pin the instruction so that LIRGenerator doesn't recurse
 423         // too deeply during it's evaluation.
 424         t->pin();
 425       }
 426     }
 427     assert(depth == 0, "should have counted back down");
 428   }
 429 
 430   UseCountComputer() {
 431     worklist = new Values();
 432     depth = 0;
 433   }
 434 
 435  public:
 436   static void compute(BlockList* blocks) {
 437     UseCountComputer ucc;
 438     blocks->iterate_backward(&ucc);
 439   }
 440 };
 441 
 442 
 443 // helper macro for short definition of trace-output inside code
 444 #ifndef PRODUCT
 445   #define TRACE_LINEAR_SCAN(level, code)       \
 446     if (TraceLinearScanLevel >= level) {       \
 447       code;                                    \
 448     }
 449 #else
 450   #define TRACE_LINEAR_SCAN(level, code)
 451 #endif
 452 
 453 class ComputeLinearScanOrder : public StackObj {
 454  private:
 455   int        _max_block_id;        // the highest block_id of a block
 456   int        _num_blocks;          // total number of blocks (smaller than _max_block_id)
 457   int        _num_loops;           // total number of loops
 458   bool       _iterative_dominators;// method requires iterative computation of dominatiors
 459 
 460   BlockList* _linear_scan_order;   // the resulting list of blocks in correct order
 461 
 462   BitMap     _visited_blocks;      // used for recursive processing of blocks
 463   BitMap     _active_blocks;       // used for recursive processing of blocks
 464   BitMap     _dominator_blocks;    // temproary BitMap used for computation of dominator
 465   intArray   _forward_branches;    // number of incoming forward branches for each block
 466   BlockList  _loop_end_blocks;     // list of all loop end blocks collected during count_edges
 467   BitMap2D   _loop_map;            // two-dimensional bit set: a bit is set if a block is contained in a loop
 468   BlockList  _work_list;           // temporary list (used in mark_loops and compute_order)
 469   BlockList  _loop_headers;
 470 
 471   Compilation* _compilation;
 472 
 473   // accessors for _visited_blocks and _active_blocks
 474   void init_visited()                     { _active_blocks.clear(); _visited_blocks.clear(); }
 475   bool is_visited(BlockBegin* b) const    { return _visited_blocks.at(b->block_id()); }
 476   bool is_active(BlockBegin* b) const     { return _active_blocks.at(b->block_id()); }
 477   void set_visited(BlockBegin* b)         { assert(!is_visited(b), "already set"); _visited_blocks.set_bit(b->block_id()); }
 478   void set_active(BlockBegin* b)          { assert(!is_active(b), "already set");  _active_blocks.set_bit(b->block_id()); }
 479   void clear_active(BlockBegin* b)        { assert(is_active(b), "not already");   _active_blocks.clear_bit(b->block_id()); }
 480 
 481   // accessors for _forward_branches
 482   void inc_forward_branches(BlockBegin* b) { _forward_branches.at_put(b->block_id(), _forward_branches.at(b->block_id()) + 1); }
 483   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()); }
 484 
 485   // accessors for _loop_map
 486   bool is_block_in_loop   (int loop_idx, BlockBegin* b) const { return _loop_map.at(loop_idx, b->block_id()); }
 487   void set_block_in_loop  (int loop_idx, BlockBegin* b)       { _loop_map.set_bit(loop_idx, b->block_id()); }
 488   void clear_block_in_loop(int loop_idx, int block_id)        { _loop_map.clear_bit(loop_idx, block_id); }
 489 
 490   // count edges between blocks
 491   void count_edges(BlockBegin* cur, BlockBegin* parent);
 492 
 493   // loop detection
 494   void mark_loops();
 495   void clear_non_natural_loops(BlockBegin* start_block);
 496   void assign_loop_depth(BlockBegin* start_block);
 497 
 498   // computation of final block order
 499   BlockBegin* common_dominator(BlockBegin* a, BlockBegin* b);
 500   void compute_dominator(BlockBegin* cur, BlockBegin* parent);
 501   int  compute_weight(BlockBegin* cur);
 502   bool ready_for_processing(BlockBegin* cur);
 503   void sort_into_work_list(BlockBegin* b);
 504   void append_block(BlockBegin* cur);
 505   void compute_order(BlockBegin* start_block);
 506 
 507   // fixup of dominators for non-natural loops
 508   bool compute_dominators_iter();
 509   void compute_dominators();
 510 
 511   // debug functions
 512   NOT_PRODUCT(void print_blocks();)
 513   DEBUG_ONLY(void verify();)
 514 
 515   Compilation* compilation() const { return _compilation; }
 516  public:
 517   ComputeLinearScanOrder(Compilation* c, BlockBegin* start_block);
 518 
 519   // accessors for final result
 520   BlockList* linear_scan_order() const    { return _linear_scan_order; }
 521   int        num_loops() const            { return _num_loops; }
 522 };
 523 
 524 
 525 ComputeLinearScanOrder::ComputeLinearScanOrder(Compilation* c, BlockBegin* start_block) :
 526   _max_block_id(BlockBegin::number_of_blocks()),
 527   _num_blocks(0),
 528   _num_loops(0),
 529   _iterative_dominators(false),
 530   _visited_blocks(_max_block_id),
 531   _active_blocks(_max_block_id),
 532   _dominator_blocks(_max_block_id),
 533   _forward_branches(_max_block_id, _max_block_id, 0),
 534   _loop_end_blocks(8),
 535   _work_list(8),
 536   _linear_scan_order(NULL), // initialized later with correct size
 537   _loop_map(0, 0),          // initialized later with correct size
 538   _compilation(c)
 539 {
 540   TRACE_LINEAR_SCAN(2, tty->print_cr("***** computing linear-scan block order"));
 541 
 542   init_visited();
 543   count_edges(start_block, NULL);
 544 
 545   if (compilation()->is_profiling()) {
 546     ciMethod *method = compilation()->method();
 547     if (!method->is_accessor()) {
 548       ciMethodData* md = method->method_data_or_null();
 549       assert(md != NULL, "Sanity");
 550       md->set_compilation_stats(_num_loops, _num_blocks);
 551     }
 552   }
 553 
 554   if (_num_loops > 0) {
 555     mark_loops();
 556     clear_non_natural_loops(start_block);
 557     assign_loop_depth(start_block);
 558   }
 559 
 560   compute_order(start_block);
 561   compute_dominators();
 562 
 563   NOT_PRODUCT(print_blocks());
 564   DEBUG_ONLY(verify());
 565 }
 566 
 567 
 568 // Traverse the CFG:
 569 // * count total number of blocks
 570 // * count all incoming edges and backward incoming edges
 571 // * number loop header blocks
 572 // * create a list with all loop end blocks
 573 void ComputeLinearScanOrder::count_edges(BlockBegin* cur, BlockBegin* parent) {
 574   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));
 575   assert(cur->dominator() == NULL, "dominator already initialized");
 576 
 577   if (is_active(cur)) {
 578     TRACE_LINEAR_SCAN(3, tty->print_cr("backward branch"));
 579     assert(is_visited(cur), "block must be visisted when block is active");
 580     assert(parent != NULL, "must have parent");
 581 
 582     cur->set(BlockBegin::backward_branch_target_flag);
 583 
 584     // When a loop header is also the start of an exception handler, then the backward branch is
 585     // an exception edge. Because such edges are usually critical edges which cannot be split, the
 586     // loop must be excluded here from processing.
 587     if (cur->is_set(BlockBegin::exception_entry_flag)) {
 588       // Make sure that dominators are correct in this weird situation
 589       _iterative_dominators = true;
 590       return;
 591     }
 592 
 593     cur->set(BlockBegin::linear_scan_loop_header_flag);
 594     parent->set(BlockBegin::linear_scan_loop_end_flag);
 595 
 596     assert(parent->number_of_sux() == 1 && parent->sux_at(0) == cur,
 597            "loop end blocks must have one successor (critical edges are split)");
 598 
 599     _loop_end_blocks.append(parent);
 600     return;
 601   }
 602 
 603   // increment number of incoming forward branches
 604   inc_forward_branches(cur);
 605 
 606   if (is_visited(cur)) {
 607     TRACE_LINEAR_SCAN(3, tty->print_cr("block already visited"));
 608     return;
 609   }
 610 
 611   _num_blocks++;
 612   set_visited(cur);
 613   set_active(cur);
 614 
 615   // recursive call for all successors
 616   int i;
 617   for (i = cur->number_of_sux() - 1; i >= 0; i--) {
 618     count_edges(cur->sux_at(i), cur);
 619   }
 620   for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
 621     count_edges(cur->exception_handler_at(i), cur);
 622   }
 623 
 624   clear_active(cur);
 625 
 626   // Each loop has a unique number.
 627   // When multiple loops are nested, assign_loop_depth assumes that the
 628   // innermost loop has the lowest number. This is guaranteed by setting
 629   // the loop number after the recursive calls for the successors above
 630   // have returned.
 631   if (cur->is_set(BlockBegin::linear_scan_loop_header_flag)) {
 632     assert(cur->loop_index() == -1, "cannot set loop-index twice");
 633     TRACE_LINEAR_SCAN(3, tty->print_cr("Block B%d is loop header of loop %d", cur->block_id(), _num_loops));
 634 
 635     cur->set_loop_index(_num_loops);
 636     _loop_headers.append(cur);
 637     _num_loops++;
 638   }
 639 
 640   TRACE_LINEAR_SCAN(3, tty->print_cr("Finished count_edges for block B%d", cur->block_id()));
 641 }
 642 
 643 
 644 void ComputeLinearScanOrder::mark_loops() {
 645   TRACE_LINEAR_SCAN(3, tty->print_cr("----- marking loops"));
 646 
 647   _loop_map = BitMap2D(_num_loops, _max_block_id);
 648   _loop_map.clear();
 649 
 650   for (int i = _loop_end_blocks.length() - 1; i >= 0; i--) {
 651     BlockBegin* loop_end   = _loop_end_blocks.at(i);
 652     BlockBegin* loop_start = loop_end->sux_at(0);
 653     int         loop_idx   = loop_start->loop_index();
 654 
 655     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));
 656     assert(loop_end->is_set(BlockBegin::linear_scan_loop_end_flag), "loop end flag must be set");
 657     assert(loop_end->number_of_sux() == 1, "incorrect number of successors");
 658     assert(loop_start->is_set(BlockBegin::linear_scan_loop_header_flag), "loop header flag must be set");
 659     assert(loop_idx >= 0 && loop_idx < _num_loops, "loop index not set");
 660     assert(_work_list.is_empty(), "work list must be empty before processing");
 661 
 662     // add the end-block of the loop to the working list
 663     _work_list.push(loop_end);
 664     set_block_in_loop(loop_idx, loop_end);
 665     do {
 666       BlockBegin* cur = _work_list.pop();
 667 
 668       TRACE_LINEAR_SCAN(3, tty->print_cr("    processing B%d", cur->block_id()));
 669       assert(is_block_in_loop(loop_idx, cur), "bit in loop map must be set when block is in work list");
 670 
 671       // recursive processing of all predecessors ends when start block of loop is reached
 672       if (cur != loop_start && !cur->is_set(BlockBegin::osr_entry_flag)) {
 673         for (int j = cur->number_of_preds() - 1; j >= 0; j--) {
 674           BlockBegin* pred = cur->pred_at(j);
 675 
 676           if (!is_block_in_loop(loop_idx, pred) /*&& !pred->is_set(BlockBeginosr_entry_flag)*/) {
 677             // this predecessor has not been processed yet, so add it to work list
 678             TRACE_LINEAR_SCAN(3, tty->print_cr("    pushing B%d", pred->block_id()));
 679             _work_list.push(pred);
 680             set_block_in_loop(loop_idx, pred);
 681           }
 682         }
 683       }
 684     } while (!_work_list.is_empty());
 685   }
 686 }
 687 
 688 
 689 // check for non-natural loops (loops where the loop header does not dominate
 690 // all other loop blocks = loops with mulitple entries).
 691 // such loops are ignored
 692 void ComputeLinearScanOrder::clear_non_natural_loops(BlockBegin* start_block) {
 693   for (int i = _num_loops - 1; i >= 0; i--) {
 694     if (is_block_in_loop(i, start_block)) {
 695       // loop i contains the entry block of the method
 696       // -> this is not a natural loop, so ignore it
 697       TRACE_LINEAR_SCAN(2, tty->print_cr("Loop %d is non-natural, so it is ignored", i));
 698 
 699       BlockBegin *loop_header = _loop_headers.at(i);
 700       assert(loop_header->is_set(BlockBegin::linear_scan_loop_header_flag), "Must be loop header");
 701 
 702       for (int j = 0; j < loop_header->number_of_preds(); j++) {
 703         BlockBegin *pred = loop_header->pred_at(j);
 704         pred->clear(BlockBegin::linear_scan_loop_end_flag);
 705       }
 706 
 707       loop_header->clear(BlockBegin::linear_scan_loop_header_flag);
 708 
 709       for (int block_id = _max_block_id - 1; block_id >= 0; block_id--) {
 710         clear_block_in_loop(i, block_id);
 711       }
 712       _iterative_dominators = true;
 713     }
 714   }
 715 }
 716 
 717 void ComputeLinearScanOrder::assign_loop_depth(BlockBegin* start_block) {
 718   TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing loop-depth and weight"));
 719   init_visited();
 720 
 721   assert(_work_list.is_empty(), "work list must be empty before processing");
 722   _work_list.append(start_block);
 723 
 724   do {
 725     BlockBegin* cur = _work_list.pop();
 726 
 727     if (!is_visited(cur)) {
 728       set_visited(cur);
 729       TRACE_LINEAR_SCAN(4, tty->print_cr("Computing loop depth for block B%d", cur->block_id()));
 730 
 731       // compute loop-depth and loop-index for the block
 732       assert(cur->loop_depth() == 0, "cannot set loop-depth twice");
 733       int i;
 734       int loop_depth = 0;
 735       int min_loop_idx = -1;
 736       for (i = _num_loops - 1; i >= 0; i--) {
 737         if (is_block_in_loop(i, cur)) {
 738           loop_depth++;
 739           min_loop_idx = i;
 740         }
 741       }
 742       cur->set_loop_depth(loop_depth);
 743       cur->set_loop_index(min_loop_idx);
 744 
 745       // append all unvisited successors to work list
 746       for (i = cur->number_of_sux() - 1; i >= 0; i--) {
 747         _work_list.append(cur->sux_at(i));
 748       }
 749       for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
 750         _work_list.append(cur->exception_handler_at(i));
 751       }
 752     }
 753   } while (!_work_list.is_empty());
 754 }
 755 
 756 
 757 BlockBegin* ComputeLinearScanOrder::common_dominator(BlockBegin* a, BlockBegin* b) {
 758   assert(a != NULL && b != NULL, "must have input blocks");
 759 
 760   _dominator_blocks.clear();
 761   while (a != NULL) {
 762     _dominator_blocks.set_bit(a->block_id());
 763     assert(a->dominator() != NULL || a == _linear_scan_order->at(0), "dominator must be initialized");
 764     a = a->dominator();
 765   }
 766   while (b != NULL && !_dominator_blocks.at(b->block_id())) {
 767     assert(b->dominator() != NULL || b == _linear_scan_order->at(0), "dominator must be initialized");
 768     b = b->dominator();
 769   }
 770 
 771   assert(b != NULL, "could not find dominator");
 772   return b;
 773 }
 774 
 775 void ComputeLinearScanOrder::compute_dominator(BlockBegin* cur, BlockBegin* parent) {
 776   if (cur->dominator() == NULL) {
 777     TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: initializing dominator of B%d to B%d", cur->block_id(), parent->block_id()));
 778     cur->set_dominator(parent);
 779 
 780   } else if (!(cur->is_set(BlockBegin::linear_scan_loop_header_flag) && parent->is_set(BlockBegin::linear_scan_loop_end_flag))) {
 781     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()));
 782     // Does not hold for exception blocks
 783     assert(cur->number_of_preds() > 1 || cur->is_set(BlockBegin::exception_entry_flag), "");
 784     cur->set_dominator(common_dominator(cur->dominator(), parent));
 785   }
 786 
 787   // Additional edge to xhandler of all our successors
 788   // range check elimination needs that the state at the end of a
 789   // block be valid in every block it dominates so cur must dominate
 790   // the exception handlers of its successors.
 791   int num_cur_xhandler = cur->number_of_exception_handlers();
 792   for (int j = 0; j < num_cur_xhandler; j++) {
 793     BlockBegin* xhandler = cur->exception_handler_at(j);
 794     compute_dominator(xhandler, parent);
 795   }
 796 }
 797 
 798 
 799 int ComputeLinearScanOrder::compute_weight(BlockBegin* cur) {
 800   BlockBegin* single_sux = NULL;
 801   if (cur->number_of_sux() == 1) {
 802     single_sux = cur->sux_at(0);
 803   }
 804 
 805   // limit loop-depth to 15 bit (only for security reason, it will never be so big)
 806   int weight = (cur->loop_depth() & 0x7FFF) << 16;
 807 
 808   // general macro for short definition of weight flags
 809   // the first instance of INC_WEIGHT_IF has the highest priority
 810   int cur_bit = 15;
 811   #define INC_WEIGHT_IF(condition) if ((condition)) { weight |= (1 << cur_bit); } cur_bit--;
 812 
 813   // this is necessery for the (very rare) case that two successing blocks have
 814   // the same loop depth, but a different loop index (can happen for endless loops
 815   // with exception handlers)
 816   INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_header_flag));
 817 
 818   // loop end blocks (blocks that end with a backward branch) are added
 819   // after all other blocks of the loop.
 820   INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_end_flag));
 821 
 822   // critical edge split blocks are prefered because than they have a bigger
 823   // proability to be completely empty
 824   INC_WEIGHT_IF(cur->is_set(BlockBegin::critical_edge_split_flag));
 825 
 826   // exceptions should not be thrown in normal control flow, so these blocks
 827   // are added as late as possible
 828   INC_WEIGHT_IF(cur->end()->as_Throw() == NULL  && (single_sux == NULL || single_sux->end()->as_Throw()  == NULL));
 829   INC_WEIGHT_IF(cur->end()->as_Return() == NULL && (single_sux == NULL || single_sux->end()->as_Return() == NULL));
 830 
 831   // exceptions handlers are added as late as possible
 832   INC_WEIGHT_IF(!cur->is_set(BlockBegin::exception_entry_flag));
 833 
 834   // guarantee that weight is > 0
 835   weight |= 1;
 836 
 837   #undef INC_WEIGHT_IF
 838   assert(cur_bit >= 0, "too many flags");
 839   assert(weight > 0, "weight cannot become negative");
 840 
 841   return weight;
 842 }
 843 
 844 bool ComputeLinearScanOrder::ready_for_processing(BlockBegin* cur) {
 845   // Discount the edge just traveled.
 846   // When the number drops to zero, all forward branches were processed
 847   if (dec_forward_branches(cur) != 0) {
 848     return false;
 849   }
 850 
 851   assert(_linear_scan_order->find(cur) == -1, "block already processed (block can be ready only once)");
 852   assert(_work_list.find(cur) == -1, "block already in work-list (block can be ready only once)");
 853   return true;
 854 }
 855 
 856 void ComputeLinearScanOrder::sort_into_work_list(BlockBegin* cur) {
 857   assert(_work_list.find(cur) == -1, "block already in work list");
 858 
 859   int cur_weight = compute_weight(cur);
 860 
 861   // the linear_scan_number is used to cache the weight of a block
 862   cur->set_linear_scan_number(cur_weight);
 863 
 864 #ifndef PRODUCT
 865   if (StressLinearScan) {
 866     _work_list.insert_before(0, cur);
 867     return;
 868   }
 869 #endif
 870 
 871   _work_list.append(NULL); // provide space for new element
 872 
 873   int insert_idx = _work_list.length() - 1;
 874   while (insert_idx > 0 && _work_list.at(insert_idx - 1)->linear_scan_number() > cur_weight) {
 875     _work_list.at_put(insert_idx, _work_list.at(insert_idx - 1));
 876     insert_idx--;
 877   }
 878   _work_list.at_put(insert_idx, cur);
 879 
 880   TRACE_LINEAR_SCAN(3, tty->print_cr("Sorted B%d into worklist. new worklist:", cur->block_id()));
 881   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()));
 882 
 883 #ifdef ASSERT
 884   for (int i = 0; i < _work_list.length(); i++) {
 885     assert(_work_list.at(i)->linear_scan_number() > 0, "weight not set");
 886     assert(i == 0 || _work_list.at(i - 1)->linear_scan_number() <= _work_list.at(i)->linear_scan_number(), "incorrect order in worklist");
 887   }
 888 #endif
 889 }
 890 
 891 void ComputeLinearScanOrder::append_block(BlockBegin* cur) {
 892   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()));
 893   assert(_linear_scan_order->find(cur) == -1, "cannot add the same block twice");
 894 
 895   // currently, the linear scan order and code emit order are equal.
 896   // therefore the linear_scan_number and the weight of a block must also
 897   // be equal.
 898   cur->set_linear_scan_number(_linear_scan_order->length());
 899   _linear_scan_order->append(cur);
 900 }
 901 
 902 void ComputeLinearScanOrder::compute_order(BlockBegin* start_block) {
 903   TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing final block order"));
 904 
 905   // the start block is always the first block in the linear scan order
 906   _linear_scan_order = new BlockList(_num_blocks);
 907   append_block(start_block);
 908 
 909   assert(start_block->end()->as_Base() != NULL, "start block must end with Base-instruction");
 910   BlockBegin* std_entry = ((Base*)start_block->end())->std_entry();
 911   BlockBegin* osr_entry = ((Base*)start_block->end())->osr_entry();
 912 
 913   BlockBegin* sux_of_osr_entry = NULL;
 914   if (osr_entry != NULL) {
 915     // special handling for osr entry:
 916     // ignore the edge between the osr entry and its successor for processing
 917     // the osr entry block is added manually below
 918     assert(osr_entry->number_of_sux() == 1, "osr entry must have exactly one successor");
 919     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");
 920 
 921     sux_of_osr_entry = osr_entry->sux_at(0);
 922     dec_forward_branches(sux_of_osr_entry);
 923 
 924     compute_dominator(osr_entry, start_block);
 925     _iterative_dominators = true;
 926   }
 927   compute_dominator(std_entry, start_block);
 928 
 929   // start processing with standard entry block
 930   assert(_work_list.is_empty(), "list must be empty before processing");
 931 
 932   if (ready_for_processing(std_entry)) {
 933     sort_into_work_list(std_entry);
 934   } else {
 935     assert(false, "the std_entry must be ready for processing (otherwise, the method has no start block)");
 936   }
 937 
 938   do {
 939     BlockBegin* cur = _work_list.pop();
 940 
 941     if (cur == sux_of_osr_entry) {
 942       // the osr entry block is ignored in normal processing, it is never added to the
 943       // work list. Instead, it is added as late as possible manually here.
 944       append_block(osr_entry);
 945       compute_dominator(cur, osr_entry);
 946     }
 947     append_block(cur);
 948 
 949     int i;
 950     int num_sux = cur->number_of_sux();
 951     // changed loop order to get "intuitive" order of if- and else-blocks
 952     for (i = 0; i < num_sux; i++) {
 953       BlockBegin* sux = cur->sux_at(i);
 954       compute_dominator(sux, cur);
 955       if (ready_for_processing(sux)) {
 956         sort_into_work_list(sux);
 957       }
 958     }
 959     num_sux = cur->number_of_exception_handlers();
 960     for (i = 0; i < num_sux; i++) {
 961       BlockBegin* sux = cur->exception_handler_at(i);
 962       if (ready_for_processing(sux)) {
 963         sort_into_work_list(sux);
 964       }
 965     }
 966   } while (_work_list.length() > 0);
 967 }
 968 
 969 
 970 bool ComputeLinearScanOrder::compute_dominators_iter() {
 971   bool changed = false;
 972   int num_blocks = _linear_scan_order->length();
 973 
 974   assert(_linear_scan_order->at(0)->dominator() == NULL, "must not have dominator");
 975   assert(_linear_scan_order->at(0)->number_of_preds() == 0, "must not have predecessors");
 976   for (int i = 1; i < num_blocks; i++) {
 977     BlockBegin* block = _linear_scan_order->at(i);
 978 
 979     BlockBegin* dominator = block->pred_at(0);
 980     int num_preds = block->number_of_preds();
 981 
 982     TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: Processing B%d", block->block_id()));
 983 
 984     for (int j = 0; j < num_preds; j++) {
 985 
 986       BlockBegin *pred = block->pred_at(j);
 987       TRACE_LINEAR_SCAN(4, tty->print_cr("   DOM: Subrocessing B%d", pred->block_id()));
 988 
 989       if (block->is_set(BlockBegin::exception_entry_flag)) {
 990         dominator = common_dominator(dominator, pred);
 991         int num_pred_preds = pred->number_of_preds();
 992         for (int k = 0; k < num_pred_preds; k++) {
 993           dominator = common_dominator(dominator, pred->pred_at(k));
 994         }
 995       } else {
 996         dominator = common_dominator(dominator, pred);
 997       }
 998     }
 999 
1000     if (dominator != block->dominator()) {
1001       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()));
1002 
1003       block->set_dominator(dominator);
1004       changed = true;
1005     }
1006   }
1007   return changed;
1008 }
1009 
1010 void ComputeLinearScanOrder::compute_dominators() {
1011   TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing dominators (iterative computation reqired: %d)", _iterative_dominators));
1012 
1013   // iterative computation of dominators is only required for methods with non-natural loops
1014   // and OSR-methods. For all other methods, the dominators computed when generating the
1015   // linear scan block order are correct.
1016   if (_iterative_dominators) {
1017     do {
1018       TRACE_LINEAR_SCAN(1, tty->print_cr("DOM: next iteration of fix-point calculation"));
1019     } while (compute_dominators_iter());
1020   }
1021 
1022   // check that dominators are correct
1023   assert(!compute_dominators_iter(), "fix point not reached");
1024 
1025   // Add Blocks to dominates-Array
1026   int num_blocks = _linear_scan_order->length();
1027   for (int i = 0; i < num_blocks; i++) {
1028     BlockBegin* block = _linear_scan_order->at(i);
1029 
1030     BlockBegin *dom = block->dominator();
1031     if (dom) {
1032       assert(dom->dominator_depth() != -1, "Dominator must have been visited before");
1033       dom->dominates()->append(block);
1034       block->set_dominator_depth(dom->dominator_depth() + 1);
1035     } else {
1036       block->set_dominator_depth(0);
1037     }
1038   }
1039 }
1040 
1041 
1042 #ifndef PRODUCT
1043 void ComputeLinearScanOrder::print_blocks() {
1044   if (TraceLinearScanLevel >= 2) {
1045     tty->print_cr("----- loop information:");
1046     for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
1047       BlockBegin* cur = _linear_scan_order->at(block_idx);
1048 
1049       tty->print("%4d: B%2d: ", cur->linear_scan_number(), cur->block_id());
1050       for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
1051         tty->print ("%d ", is_block_in_loop(loop_idx, cur));
1052       }
1053       tty->print_cr(" -> loop_index: %2d, loop_depth: %2d", cur->loop_index(), cur->loop_depth());
1054     }
1055   }
1056 
1057   if (TraceLinearScanLevel >= 1) {
1058     tty->print_cr("----- linear-scan block order:");
1059     for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
1060       BlockBegin* cur = _linear_scan_order->at(block_idx);
1061       tty->print("%4d: B%2d    loop: %2d  depth: %2d", cur->linear_scan_number(), cur->block_id(), cur->loop_index(), cur->loop_depth());
1062 
1063       tty->print(cur->is_set(BlockBegin::exception_entry_flag)         ? " ex" : "   ");
1064       tty->print(cur->is_set(BlockBegin::critical_edge_split_flag)     ? " ce" : "   ");
1065       tty->print(cur->is_set(BlockBegin::linear_scan_loop_header_flag) ? " lh" : "   ");
1066       tty->print(cur->is_set(BlockBegin::linear_scan_loop_end_flag)    ? " le" : "   ");
1067 
1068       if (cur->dominator() != NULL) {
1069         tty->print("    dom: B%d ", cur->dominator()->block_id());
1070       } else {
1071         tty->print("    dom: NULL ");
1072       }
1073 
1074       if (cur->number_of_preds() > 0) {
1075         tty->print("    preds: ");
1076         for (int j = 0; j < cur->number_of_preds(); j++) {
1077           BlockBegin* pred = cur->pred_at(j);
1078           tty->print("B%d ", pred->block_id());
1079         }
1080       }
1081       if (cur->number_of_sux() > 0) {
1082         tty->print("    sux: ");
1083         for (int j = 0; j < cur->number_of_sux(); j++) {
1084           BlockBegin* sux = cur->sux_at(j);
1085           tty->print("B%d ", sux->block_id());
1086         }
1087       }
1088       if (cur->number_of_exception_handlers() > 0) {
1089         tty->print("    ex: ");
1090         for (int j = 0; j < cur->number_of_exception_handlers(); j++) {
1091           BlockBegin* ex = cur->exception_handler_at(j);
1092           tty->print("B%d ", ex->block_id());
1093         }
1094       }
1095       tty->cr();
1096     }
1097   }
1098 }
1099 #endif
1100 
1101 #ifdef ASSERT
1102 void ComputeLinearScanOrder::verify() {
1103   assert(_linear_scan_order->length() == _num_blocks, "wrong number of blocks in list");
1104 
1105   if (StressLinearScan) {
1106     // blocks are scrambled when StressLinearScan is used
1107     return;
1108   }
1109 
1110   // check that all successors of a block have a higher linear-scan-number
1111   // and that all predecessors of a block have a lower linear-scan-number
1112   // (only backward branches of loops are ignored)
1113   int i;
1114   for (i = 0; i < _linear_scan_order->length(); i++) {
1115     BlockBegin* cur = _linear_scan_order->at(i);
1116 
1117     assert(cur->linear_scan_number() == i, "incorrect linear_scan_number");
1118     assert(cur->linear_scan_number() >= 0 && cur->linear_scan_number() == _linear_scan_order->find(cur), "incorrect linear_scan_number");
1119 
1120     int j;
1121     for (j = cur->number_of_sux() - 1; j >= 0; j--) {
1122       BlockBegin* sux = cur->sux_at(j);
1123 
1124       assert(sux->linear_scan_number() >= 0 && sux->linear_scan_number() == _linear_scan_order->find(sux), "incorrect linear_scan_number");
1125       if (!sux->is_set(BlockBegin::backward_branch_target_flag)) {
1126         assert(cur->linear_scan_number() < sux->linear_scan_number(), "invalid order");
1127       }
1128       if (cur->loop_depth() == sux->loop_depth()) {
1129         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");
1130       }
1131     }
1132 
1133     for (j = cur->number_of_preds() - 1; j >= 0; j--) {
1134       BlockBegin* pred = cur->pred_at(j);
1135 
1136       assert(pred->linear_scan_number() >= 0 && pred->linear_scan_number() == _linear_scan_order->find(pred), "incorrect linear_scan_number");
1137       if (!cur->is_set(BlockBegin::backward_branch_target_flag)) {
1138         assert(cur->linear_scan_number() > pred->linear_scan_number(), "invalid order");
1139       }
1140       if (cur->loop_depth() == pred->loop_depth()) {
1141         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");
1142       }
1143 
1144       assert(cur->dominator()->linear_scan_number() <= cur->pred_at(j)->linear_scan_number(), "dominator must be before predecessors");
1145     }
1146 
1147     // check dominator
1148     if (i == 0) {
1149       assert(cur->dominator() == NULL, "first block has no dominator");
1150     } else {
1151       assert(cur->dominator() != NULL, "all but first block must have dominator");
1152     }
1153     // Assertion does not hold for exception handlers
1154     assert(cur->number_of_preds() != 1 || cur->dominator() == cur->pred_at(0) || cur->is_set(BlockBegin::exception_entry_flag), "Single predecessor must also be dominator");
1155   }
1156 
1157   // check that all loops are continuous
1158   for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
1159     int block_idx = 0;
1160     assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "the first block must not be present in any loop");
1161 
1162     // skip blocks before the loop
1163     while (block_idx < _num_blocks && !is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
1164       block_idx++;
1165     }
1166     // skip blocks of loop
1167     while (block_idx < _num_blocks && is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
1168       block_idx++;
1169     }
1170     // after the first non-loop block, there must not be another loop-block
1171     while (block_idx < _num_blocks) {
1172       assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "loop not continuous in linear-scan order");
1173       block_idx++;
1174     }
1175   }
1176 }
1177 #endif
1178 
1179 
1180 void IR::compute_code() {
1181   assert(is_valid(), "IR must be valid");
1182 
1183   ComputeLinearScanOrder compute_order(compilation(), start());
1184   _num_loops = compute_order.num_loops();
1185   _code = compute_order.linear_scan_order();
1186 }
1187 
1188 
1189 void IR::compute_use_counts() {
1190   // make sure all values coming out of this block get evaluated.
1191   int num_blocks = _code->length();
1192   for (int i = 0; i < num_blocks; i++) {
1193     _code->at(i)->end()->state()->pin_stack_for_linear_scan();
1194   }
1195 
1196   // compute use counts
1197   UseCountComputer::compute(_code);
1198 }
1199 
1200 
1201 void IR::iterate_preorder(BlockClosure* closure) {
1202   assert(is_valid(), "IR must be valid");
1203   start()->iterate_preorder(closure);
1204 }
1205 
1206 
1207 void IR::iterate_postorder(BlockClosure* closure) {
1208   assert(is_valid(), "IR must be valid");
1209   start()->iterate_postorder(closure);
1210 }
1211 
1212 void IR::iterate_linear_scan_order(BlockClosure* closure) {
1213   linear_scan_order()->iterate_forward(closure);
1214 }
1215 
1216 
1217 #ifndef PRODUCT
1218 class BlockPrinter: public BlockClosure {
1219  private:
1220   InstructionPrinter* _ip;
1221   bool                _cfg_only;
1222   bool                _live_only;
1223 
1224  public:
1225   BlockPrinter(InstructionPrinter* ip, bool cfg_only, bool live_only = false) {
1226     _ip       = ip;
1227     _cfg_only = cfg_only;
1228     _live_only = live_only;
1229   }
1230 
1231   virtual void block_do(BlockBegin* block) {
1232     if (_cfg_only) {
1233       _ip->print_instr(block); tty->cr();
1234     } else {
1235       block->print_block(*_ip, _live_only);
1236     }
1237   }
1238 };
1239 
1240 
1241 void IR::print(BlockBegin* start, bool cfg_only, bool live_only) {
1242   ttyLocker ttyl;
1243   InstructionPrinter ip(!cfg_only);
1244   BlockPrinter bp(&ip, cfg_only, live_only);
1245   start->iterate_preorder(&bp);
1246   tty->cr();
1247 }
1248 
1249 void IR::print(bool cfg_only, bool live_only) {
1250   if (is_valid()) {
1251     print(start(), cfg_only, live_only);
1252   } else {
1253     tty->print_cr("invalid IR");
1254   }
1255 }
1256 
1257 
1258 typedef GrowableArray<BlockList*> BlockListList;
1259 
1260 class PredecessorValidator : public BlockClosure {
1261  private:
1262   BlockListList* _predecessors;
1263   BlockList*     _blocks;
1264 
1265   static int cmp(BlockBegin** a, BlockBegin** b) {
1266     return (*a)->block_id() - (*b)->block_id();
1267   }
1268 
1269  public:
1270   PredecessorValidator(IR* hir) {
1271     ResourceMark rm;
1272     _predecessors = new BlockListList(BlockBegin::number_of_blocks(), BlockBegin::number_of_blocks(), NULL);
1273     _blocks = new BlockList();
1274 
1275     int i;
1276     hir->start()->iterate_preorder(this);
1277     if (hir->code() != NULL) {
1278       assert(hir->code()->length() == _blocks->length(), "must match");
1279       for (i = 0; i < _blocks->length(); i++) {
1280         assert(hir->code()->contains(_blocks->at(i)), "should be in both lists");
1281       }
1282     }
1283 
1284     for (i = 0; i < _blocks->length(); i++) {
1285       BlockBegin* block = _blocks->at(i);
1286       BlockList* preds = _predecessors->at(block->block_id());
1287       if (preds == NULL) {
1288         assert(block->number_of_preds() == 0, "should be the same");
1289         continue;
1290       }
1291 
1292       // clone the pred list so we can mutate it
1293       BlockList* pred_copy = new BlockList();
1294       int j;
1295       for (j = 0; j < block->number_of_preds(); j++) {
1296         pred_copy->append(block->pred_at(j));
1297       }
1298       // sort them in the same order
1299       preds->sort(cmp);
1300       pred_copy->sort(cmp);
1301       int length = MIN2(preds->length(), block->number_of_preds());
1302       for (j = 0; j < block->number_of_preds(); j++) {
1303         assert(preds->at(j) == pred_copy->at(j), "must match");
1304       }
1305 
1306       assert(preds->length() == block->number_of_preds(), "should be the same");
1307     }
1308   }
1309 
1310   virtual void block_do(BlockBegin* block) {
1311     _blocks->append(block);
1312     BlockEnd* be = block->end();
1313     int n = be->number_of_sux();
1314     int i;
1315     for (i = 0; i < n; i++) {
1316       BlockBegin* sux = be->sux_at(i);
1317       assert(!sux->is_set(BlockBegin::exception_entry_flag), "must not be xhandler");
1318 
1319       BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
1320       if (preds == NULL) {
1321         preds = new BlockList();
1322         _predecessors->at_put(sux->block_id(), preds);
1323       }
1324       preds->append(block);
1325     }
1326 
1327     n = block->number_of_exception_handlers();
1328     for (i = 0; i < n; i++) {
1329       BlockBegin* sux = block->exception_handler_at(i);
1330       assert(sux->is_set(BlockBegin::exception_entry_flag), "must be xhandler");
1331 
1332       BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
1333       if (preds == NULL) {
1334         preds = new BlockList();
1335         _predecessors->at_put(sux->block_id(), preds);
1336       }
1337       preds->append(block);
1338     }
1339   }
1340 };
1341 
1342 class VerifyBlockBeginField : public BlockClosure {
1343 
1344 public:
1345 
1346   virtual void block_do(BlockBegin *block) {
1347     for ( Instruction *cur = block; cur != NULL; cur = cur->next()) {
1348       assert(cur->block() == block, "Block begin is not correct");
1349     }
1350   }
1351 };
1352 
1353 void IR::verify() {
1354 #ifdef ASSERT
1355   PredecessorValidator pv(this);
1356   VerifyBlockBeginField verifier;
1357   this->iterate_postorder(&verifier);
1358 #endif
1359 }
1360 
1361 #endif // PRODUCT
1362 
1363 void SubstitutionResolver::visit(Value* v) {
1364   Value v0 = *v;
1365   if (v0) {
1366     Value vs = v0->subst();
1367     if (vs != v0) {
1368       *v = v0->subst();
1369     }
1370   }
1371 }
1372 
1373 #ifdef ASSERT
1374 class SubstitutionChecker: public ValueVisitor {
1375   void visit(Value* v) {
1376     Value v0 = *v;
1377     if (v0) {
1378       Value vs = v0->subst();
1379       assert(vs == v0, "missed substitution");
1380     }
1381   }
1382 };
1383 #endif
1384 
1385 
1386 void SubstitutionResolver::block_do(BlockBegin* block) {
1387   Instruction* last = NULL;
1388   for (Instruction* n = block; n != NULL;) {
1389     n->values_do(this);
1390     // need to remove this instruction from the instruction stream
1391     if (n->subst() != n) {
1392       assert(last != NULL, "must have last");
1393       last->set_next(n->next());
1394     } else {
1395       last = n;
1396     }
1397     n = last->next();
1398   }
1399 
1400 #ifdef ASSERT
1401   SubstitutionChecker check_substitute;
1402   if (block->state()) block->state()->values_do(&check_substitute);
1403   block->block_values_do(&check_substitute);
1404   if (block->end() && block->end()->state()) block->end()->state()->values_do(&check_substitute);
1405 #endif
1406 }