1 /*
   2  * Copyright (c) 1999, 2013, 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   _start              = NULL;
 147 
 148   if (osr_bci == -1) {
 149     _requires_phi_function.clear();
 150   } else {
 151         // selective creation of phi functions is not possibel in osr-methods
 152     _requires_phi_function.set_range(0, method->max_locals());
 153   }
 154 
 155   assert(method->holder()->is_loaded() , "method holder must be loaded");
 156 
 157   // build graph if monitor pairing is ok
 158   if (create_graph && monitor_pairing_ok()) _start = build_graph(compilation, osr_bci);
 159 }
 160 
 161 
 162 int IRScope::max_stack() const {
 163   int my_max = method()->max_stack();
 164   int callee_max = 0;
 165   for (int i = 0; i < number_of_callees(); i++) {
 166     callee_max = MAX2(callee_max, callee_no(i)->max_stack());
 167   }
 168   return my_max + callee_max;
 169 }
 170 
 171 
 172 bool IRScopeDebugInfo::should_reexecute() {
 173   ciMethod* cur_method = scope()->method();
 174   int       cur_bci    = bci();
 175   if (cur_method != NULL && cur_bci != SynchronizationEntryBCI) {
 176     Bytecodes::Code code = cur_method->java_code_at_bci(cur_bci);
 177     return Interpreter::bytecode_should_reexecute(code);
 178   } else
 179     return false;
 180 }
 181 
 182 
 183 // Implementation of CodeEmitInfo
 184 
 185 // Stack must be NON-null
 186 CodeEmitInfo::CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers, bool deoptimize_on_exception)
 187   : _scope(stack->scope())
 188   , _scope_debug_info(NULL)
 189   , _oop_map(NULL)
 190   , _stack(stack)
 191   , _exception_handlers(exception_handlers)
 192   , _is_method_handle_invoke(false)
 193   , _deoptimize_on_exception(deoptimize_on_exception) {
 194   assert(_stack != NULL, "must be non null");
 195 }
 196 
 197 
 198 CodeEmitInfo::CodeEmitInfo(CodeEmitInfo* info, ValueStack* stack)
 199   : _scope(info->_scope)
 200   , _exception_handlers(NULL)
 201   , _scope_debug_info(NULL)
 202   , _oop_map(NULL)
 203   , _stack(stack == NULL ? info->_stack : stack)
 204   , _is_method_handle_invoke(info->_is_method_handle_invoke)
 205   , _deoptimize_on_exception(info->_deoptimize_on_exception) {
 206 
 207   // deep copy of exception handlers
 208   if (info->_exception_handlers != NULL) {
 209     _exception_handlers = new XHandlers(info->_exception_handlers);
 210   }
 211 }
 212 
 213 
 214 void CodeEmitInfo::record_debug_info(DebugInformationRecorder* recorder, int pc_offset) {
 215   // record the safepoint before recording the debug info for enclosing scopes
 216   recorder->add_safepoint(pc_offset, _oop_map->deep_copy());
 217   _scope_debug_info->record_debug_info(recorder, pc_offset, true/*topmost*/, _is_method_handle_invoke);
 218   recorder->end_safepoint(pc_offset);
 219 }
 220 
 221 
 222 void CodeEmitInfo::add_register_oop(LIR_Opr opr) {
 223   assert(_oop_map != NULL, "oop map must already exist");
 224   assert(opr->is_single_cpu(), "should not call otherwise");
 225 
 226   VMReg name = frame_map()->regname(opr);
 227   _oop_map->set_oop(name);
 228 }
 229 
 230 // Mirror the stack size calculation in the deopt code
 231 // How much stack space would we need at this point in the program in
 232 // case of deoptimization?
 233 int CodeEmitInfo::interpreter_frame_size() const {
 234   ValueStack* state = _stack;
 235   int size = 0;
 236   int callee_parameters = 0;
 237   int callee_locals = 0;
 238   int popframe_extra_args = 0;
 239 
 240   if (JvmtiExport::can_pop_frame() && state->kind() != ValueStack::StateBefore) {
 241     ciMethod* method = state->scope()->method();
 242     int bci = state->bci();
 243     popframe_extra_args = MAX2(-method->get_stack_effect_if_at_invoke(bci), 0);
 244   }
 245 
 246   while (state != NULL) {
 247     int locks = state->locks_size();
 248     int temps = state->stack_size();
 249     bool is_top_frame = (state->caller_state() == NULL);
 250     ciMethod* method = state->scope()->method();
 251 
 252     int frame_size = BytesPerWord * Interpreter::size_activation(method->max_stack(),
 253                                                                  temps + callee_parameters,
 254                                                                  popframe_extra_args,
 255                                                                  locks,
 256                                                                  callee_parameters,
 257                                                                  callee_locals,
 258                                                                  is_top_frame);
 259     size += frame_size;
 260 
 261     callee_parameters = method->size_of_parameters();
 262     callee_locals = method->max_locals();
 263     popframe_extra_args = 0;
 264     state = state->caller_state();
 265   }
 266   return size + Deoptimization::last_frame_adjust(0, callee_locals) * BytesPerWord;
 267 }
 268 
 269 // Implementation of IR
 270 
 271 IR::IR(Compilation* compilation, ciMethod* method, int osr_bci) :
 272     _locals_size(in_WordSize(-1))
 273   , _num_loops(0) {
 274   // setup IR fields
 275   _compilation = compilation;
 276   _top_scope   = new IRScope(compilation, NULL, -1, method, osr_bci, true);
 277   _code        = NULL;
 278 }
 279 
 280 
 281 void IR::optimize_blocks() {
 282   Optimizer opt(this);
 283   if (!compilation()->profile_branches()) {
 284     if (DoCEE) {
 285       opt.eliminate_conditional_expressions();
 286 #ifndef PRODUCT
 287       if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after CEE"); print(true); }
 288       if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after CEE"); print(false); }
 289 #endif
 290     }
 291     if (EliminateBlocks) {
 292       opt.eliminate_blocks();
 293 #ifndef PRODUCT
 294       if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after block elimination"); print(true); }
 295       if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after block elimination"); print(false); }
 296 #endif
 297     }
 298   }
 299 }
 300 
 301 void IR::eliminate_null_checks() {
 302   Optimizer opt(this);
 303   if (EliminateNullChecks) {
 304     opt.eliminate_null_checks();
 305 #ifndef PRODUCT
 306     if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after null check elimination"); print(true); }
 307     if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after null check elimination"); print(false); }
 308 #endif
 309   }
 310 }
 311 
 312 
 313 static int sort_pairs(BlockPair** a, BlockPair** b) {
 314   if ((*a)->from() == (*b)->from()) {
 315     return (*a)->to()->block_id() - (*b)->to()->block_id();
 316   } else {
 317     return (*a)->from()->block_id() - (*b)->from()->block_id();
 318   }
 319 }
 320 
 321 
 322 class CriticalEdgeFinder: public BlockClosure {
 323   BlockPairList blocks;
 324   IR*       _ir;
 325 
 326  public:
 327   CriticalEdgeFinder(IR* ir): _ir(ir) {}
 328   void block_do(BlockBegin* bb) {
 329     BlockEnd* be = bb->end();
 330     int nos = be->number_of_sux();
 331     if (nos >= 2) {
 332       for (int i = 0; i < nos; i++) {
 333         BlockBegin* sux = be->sux_at(i);
 334         if (sux->number_of_preds() >= 2) {
 335           blocks.append(new BlockPair(bb, sux));
 336         }
 337       }
 338     }
 339   }
 340 
 341   void split_edges() {
 342     BlockPair* last_pair = NULL;
 343     blocks.sort(sort_pairs);
 344     for (int i = 0; i < blocks.length(); i++) {
 345       BlockPair* pair = blocks.at(i);
 346       if (last_pair != NULL && pair->is_same(last_pair)) continue;
 347       BlockBegin* from = pair->from();
 348       BlockBegin* to = pair->to();
 349       BlockBegin* split = from->insert_block_between(to);
 350 #ifndef PRODUCT
 351       if ((PrintIR || PrintIR1) && Verbose) {
 352         tty->print_cr("Split critical edge B%d -> B%d (new block B%d)",
 353                       from->block_id(), to->block_id(), split->block_id());
 354       }
 355 #endif
 356       last_pair = pair;
 357     }
 358   }
 359 };
 360 
 361 void IR::split_critical_edges() {
 362   CriticalEdgeFinder cef(this);
 363 
 364   iterate_preorder(&cef);
 365   cef.split_edges();
 366 }
 367 
 368 
 369 class UseCountComputer: public ValueVisitor, BlockClosure {
 370  private:
 371   void visit(Value* n) {
 372     // Local instructions and Phis for expression stack values at the
 373     // start of basic blocks are not added to the instruction list
 374     if (!(*n)->is_linked() && (*n)->can_be_linked()) {
 375       assert(false, "a node was not appended to the graph");
 376       Compilation::current()->bailout("a node was not appended to the graph");
 377     }
 378     // use n's input if not visited before
 379     if (!(*n)->is_pinned() && !(*n)->has_uses()) {
 380       // note: a) if the instruction is pinned, it will be handled by compute_use_count
 381       //       b) if the instruction has uses, it was touched before
 382       //       => in both cases we don't need to update n's values
 383       uses_do(n);
 384     }
 385     // use n
 386     (*n)->_use_count++;
 387   }
 388 
 389   Values* worklist;
 390   int depth;
 391   enum {
 392     max_recurse_depth = 20
 393   };
 394 
 395   void uses_do(Value* n) {
 396     depth++;
 397     if (depth > max_recurse_depth) {
 398       // don't allow the traversal to recurse too deeply
 399       worklist->push(*n);
 400     } else {
 401       (*n)->input_values_do(this);
 402       // special handling for some instructions
 403       if ((*n)->as_BlockEnd() != NULL) {
 404         // note on BlockEnd:
 405         //   must 'use' the stack only if the method doesn't
 406         //   terminate, however, in those cases stack is empty
 407         (*n)->state_values_do(this);
 408       }
 409     }
 410     depth--;
 411   }
 412 
 413   void block_do(BlockBegin* b) {
 414     depth = 0;
 415     // process all pinned nodes as the roots of expression trees
 416     for (Instruction* n = b; n != NULL; n = n->next()) {
 417       if (n->is_pinned()) uses_do(&n);
 418     }
 419     assert(depth == 0, "should have counted back down");
 420 
 421     // now process any unpinned nodes which recursed too deeply
 422     while (worklist->length() > 0) {
 423       Value t = worklist->pop();
 424       if (!t->is_pinned()) {
 425         // compute the use count
 426         uses_do(&t);
 427 
 428         // pin the instruction so that LIRGenerator doesn't recurse
 429         // too deeply during it's evaluation.
 430         t->pin();
 431       }
 432     }
 433     assert(depth == 0, "should have counted back down");
 434   }
 435 
 436   UseCountComputer() {
 437     worklist = new Values();
 438     depth = 0;
 439   }
 440 
 441  public:
 442   static void compute(BlockList* blocks) {
 443     UseCountComputer ucc;
 444     blocks->iterate_backward(&ucc);
 445   }
 446 };
 447 
 448 
 449 // helper macro for short definition of trace-output inside code
 450 #ifndef PRODUCT
 451   #define TRACE_LINEAR_SCAN(level, code)       \
 452     if (TraceLinearScanLevel >= level) {       \
 453       code;                                    \
 454     }
 455 #else
 456   #define TRACE_LINEAR_SCAN(level, code)
 457 #endif
 458 
 459 class ComputeLinearScanOrder : public StackObj {
 460  private:
 461   int        _max_block_id;        // the highest block_id of a block
 462   int        _num_blocks;          // total number of blocks (smaller than _max_block_id)
 463   int        _num_loops;           // total number of loops
 464   bool       _iterative_dominators;// method requires iterative computation of dominatiors
 465 
 466   BlockList* _linear_scan_order;   // the resulting list of blocks in correct order
 467 
 468   BitMap     _visited_blocks;      // used for recursive processing of blocks
 469   BitMap     _active_blocks;       // used for recursive processing of blocks
 470   BitMap     _dominator_blocks;    // temproary BitMap used for computation of dominator
 471   intArray   _forward_branches;    // number of incoming forward branches for each block
 472   BlockList  _loop_end_blocks;     // list of all loop end blocks collected during count_edges
 473   BitMap2D   _loop_map;            // two-dimensional bit set: a bit is set if a block is contained in a loop
 474   BlockList  _work_list;           // temporary list (used in mark_loops and compute_order)
 475   BlockList  _loop_headers;
 476 
 477   Compilation* _compilation;
 478 
 479   // accessors for _visited_blocks and _active_blocks
 480   void init_visited()                     { _active_blocks.clear(); _visited_blocks.clear(); }
 481   bool is_visited(BlockBegin* b) const    { return _visited_blocks.at(b->block_id()); }
 482   bool is_active(BlockBegin* b) const     { return _active_blocks.at(b->block_id()); }
 483   void set_visited(BlockBegin* b)         { assert(!is_visited(b), "already set"); _visited_blocks.set_bit(b->block_id()); }
 484   void set_active(BlockBegin* b)          { assert(!is_active(b), "already set");  _active_blocks.set_bit(b->block_id()); }
 485   void clear_active(BlockBegin* b)        { assert(is_active(b), "not already");   _active_blocks.clear_bit(b->block_id()); }
 486 
 487   // accessors for _forward_branches
 488   void inc_forward_branches(BlockBegin* b) { _forward_branches.at_put(b->block_id(), _forward_branches.at(b->block_id()) + 1); }
 489   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()); }
 490 
 491   // accessors for _loop_map
 492   bool is_block_in_loop   (int loop_idx, BlockBegin* b) const { return _loop_map.at(loop_idx, b->block_id()); }
 493   void set_block_in_loop  (int loop_idx, BlockBegin* b)       { _loop_map.set_bit(loop_idx, b->block_id()); }
 494   void clear_block_in_loop(int loop_idx, int block_id)        { _loop_map.clear_bit(loop_idx, block_id); }
 495 
 496   // count edges between blocks
 497   void count_edges(BlockBegin* cur, BlockBegin* parent);
 498 
 499   // loop detection
 500   void mark_loops();
 501   void clear_non_natural_loops(BlockBegin* start_block);
 502   void assign_loop_depth(BlockBegin* start_block);
 503 
 504   // computation of final block order
 505   BlockBegin* common_dominator(BlockBegin* a, BlockBegin* b);
 506   void compute_dominator(BlockBegin* cur, BlockBegin* parent);
 507   int  compute_weight(BlockBegin* cur);
 508   bool ready_for_processing(BlockBegin* cur);
 509   void sort_into_work_list(BlockBegin* b);
 510   void append_block(BlockBegin* cur);
 511   void compute_order(BlockBegin* start_block);
 512 
 513   // fixup of dominators for non-natural loops
 514   bool compute_dominators_iter();
 515   void compute_dominators();
 516 
 517   // debug functions
 518   NOT_PRODUCT(void print_blocks();)
 519   DEBUG_ONLY(void verify();)
 520 
 521   Compilation* compilation() const { return _compilation; }
 522  public:
 523   ComputeLinearScanOrder(Compilation* c, BlockBegin* start_block);
 524 
 525   // accessors for final result
 526   BlockList* linear_scan_order() const    { return _linear_scan_order; }
 527   int        num_loops() const            { return _num_loops; }
 528 };
 529 
 530 
 531 ComputeLinearScanOrder::ComputeLinearScanOrder(Compilation* c, BlockBegin* start_block) :
 532   _max_block_id(BlockBegin::number_of_blocks()),
 533   _num_blocks(0),
 534   _num_loops(0),
 535   _iterative_dominators(false),
 536   _visited_blocks(_max_block_id),
 537   _active_blocks(_max_block_id),
 538   _dominator_blocks(_max_block_id),
 539   _forward_branches(_max_block_id, 0),
 540   _loop_end_blocks(8),
 541   _work_list(8),
 542   _linear_scan_order(NULL), // initialized later with correct size
 543   _loop_map(0, 0),          // initialized later with correct size
 544   _compilation(c)
 545 {
 546   TRACE_LINEAR_SCAN(2, tty->print_cr("***** computing linear-scan block order"));
 547 
 548   init_visited();
 549   count_edges(start_block, NULL);
 550 
 551   if (compilation()->is_profiling()) {
 552     ciMethod *method = compilation()->method();
 553     if (!method->is_accessor()) {
 554       ciMethodData* md = method->method_data_or_null();
 555       assert(md != NULL, "Sanity");
 556       md->set_compilation_stats(_num_loops, _num_blocks);
 557     }
 558   }
 559 
 560   if (_num_loops > 0) {
 561     mark_loops();
 562     clear_non_natural_loops(start_block);
 563     assign_loop_depth(start_block);
 564   }
 565 
 566   compute_order(start_block);
 567   compute_dominators();
 568 
 569   NOT_PRODUCT(print_blocks());
 570   DEBUG_ONLY(verify());
 571 }
 572 
 573 
 574 // Traverse the CFG:
 575 // * count total number of blocks
 576 // * count all incoming edges and backward incoming edges
 577 // * number loop header blocks
 578 // * create a list with all loop end blocks
 579 void ComputeLinearScanOrder::count_edges(BlockBegin* cur, BlockBegin* parent) {
 580   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));
 581   assert(cur->dominator() == NULL, "dominator already initialized");
 582 
 583   if (is_active(cur)) {
 584     TRACE_LINEAR_SCAN(3, tty->print_cr("backward branch"));
 585     assert(is_visited(cur), "block must be visisted when block is active");
 586     assert(parent != NULL, "must have parent");
 587 
 588     cur->set(BlockBegin::linear_scan_loop_header_flag);
 589     cur->set(BlockBegin::backward_branch_target_flag);
 590 
 591     parent->set(BlockBegin::linear_scan_loop_end_flag);
 592 
 593     // When a loop header is also the start of an exception handler, then the backward branch is
 594     // an exception edge. Because such edges are usually critical edges which cannot be split, the
 595     // loop must be excluded here from processing.
 596     if (cur->is_set(BlockBegin::exception_entry_flag)) {
 597       // Make sure that dominators are correct in this weird situation
 598       _iterative_dominators = true;
 599       return;
 600     }
 601     assert(parent->number_of_sux() == 1 && parent->sux_at(0) == cur,
 602            "loop end blocks must have one successor (critical edges are split)");
 603 
 604     _loop_end_blocks.append(parent);
 605     return;
 606   }
 607 
 608   // increment number of incoming forward branches
 609   inc_forward_branches(cur);
 610 
 611   if (is_visited(cur)) {
 612     TRACE_LINEAR_SCAN(3, tty->print_cr("block already visited"));
 613     return;
 614   }
 615 
 616   _num_blocks++;
 617   set_visited(cur);
 618   set_active(cur);
 619 
 620   // recursive call for all successors
 621   int i;
 622   for (i = cur->number_of_sux() - 1; i >= 0; i--) {
 623     count_edges(cur->sux_at(i), cur);
 624   }
 625   for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
 626     count_edges(cur->exception_handler_at(i), cur);
 627   }
 628 
 629   clear_active(cur);
 630 
 631   // Each loop has a unique number.
 632   // When multiple loops are nested, assign_loop_depth assumes that the
 633   // innermost loop has the lowest number. This is guaranteed by setting
 634   // the loop number after the recursive calls for the successors above
 635   // have returned.
 636   if (cur->is_set(BlockBegin::linear_scan_loop_header_flag)) {
 637     assert(cur->loop_index() == -1, "cannot set loop-index twice");
 638     TRACE_LINEAR_SCAN(3, tty->print_cr("Block B%d is loop header of loop %d", cur->block_id(), _num_loops));
 639 
 640     cur->set_loop_index(_num_loops);
 641     _loop_headers.append(cur);
 642     _num_loops++;
 643   }
 644 
 645   TRACE_LINEAR_SCAN(3, tty->print_cr("Finished count_edges for block B%d", cur->block_id()));
 646 }
 647 
 648 
 649 void ComputeLinearScanOrder::mark_loops() {
 650   TRACE_LINEAR_SCAN(3, tty->print_cr("----- marking loops"));
 651 
 652   _loop_map = BitMap2D(_num_loops, _max_block_id);
 653   _loop_map.clear();
 654 
 655   for (int i = _loop_end_blocks.length() - 1; i >= 0; i--) {
 656     BlockBegin* loop_end   = _loop_end_blocks.at(i);
 657     BlockBegin* loop_start = loop_end->sux_at(0);
 658     int         loop_idx   = loop_start->loop_index();
 659 
 660     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));
 661     assert(loop_end->is_set(BlockBegin::linear_scan_loop_end_flag), "loop end flag must be set");
 662     assert(loop_end->number_of_sux() == 1, "incorrect number of successors");
 663     assert(loop_start->is_set(BlockBegin::linear_scan_loop_header_flag), "loop header flag must be set");
 664     assert(loop_idx >= 0 && loop_idx < _num_loops, "loop index not set");
 665     assert(_work_list.is_empty(), "work list must be empty before processing");
 666 
 667     // add the end-block of the loop to the working list
 668     _work_list.push(loop_end);
 669     set_block_in_loop(loop_idx, loop_end);
 670     do {
 671       BlockBegin* cur = _work_list.pop();
 672 
 673       TRACE_LINEAR_SCAN(3, tty->print_cr("    processing B%d", cur->block_id()));
 674       assert(is_block_in_loop(loop_idx, cur), "bit in loop map must be set when block is in work list");
 675 
 676       // recursive processing of all predecessors ends when start block of loop is reached
 677       if (cur != loop_start && !cur->is_set(BlockBegin::osr_entry_flag)) {
 678         for (int j = cur->number_of_preds() - 1; j >= 0; j--) {
 679           BlockBegin* pred = cur->pred_at(j);
 680 
 681           if (!is_block_in_loop(loop_idx, pred) /*&& !pred->is_set(BlockBeginosr_entry_flag)*/) {
 682             // this predecessor has not been processed yet, so add it to work list
 683             TRACE_LINEAR_SCAN(3, tty->print_cr("    pushing B%d", pred->block_id()));
 684             _work_list.push(pred);
 685             set_block_in_loop(loop_idx, pred);
 686           }
 687         }
 688       }
 689     } while (!_work_list.is_empty());
 690   }
 691 }
 692 
 693 
 694 // check for non-natural loops (loops where the loop header does not dominate
 695 // all other loop blocks = loops with mulitple entries).
 696 // such loops are ignored
 697 void ComputeLinearScanOrder::clear_non_natural_loops(BlockBegin* start_block) {
 698   for (int i = _num_loops - 1; i >= 0; i--) {
 699     if (is_block_in_loop(i, start_block)) {
 700       // loop i contains the entry block of the method
 701       // -> this is not a natural loop, so ignore it
 702       TRACE_LINEAR_SCAN(2, tty->print_cr("Loop %d is non-natural, so it is ignored", i));
 703 
 704       BlockBegin *loop_header = _loop_headers.at(i);
 705       assert(loop_header->is_set(BlockBegin::linear_scan_loop_header_flag), "Must be loop header");
 706 
 707       for (int j = 0; j < loop_header->number_of_preds(); j++) {
 708         BlockBegin *pred = loop_header->pred_at(j);
 709         pred->clear(BlockBegin::linear_scan_loop_end_flag);
 710       }
 711 
 712       loop_header->clear(BlockBegin::linear_scan_loop_header_flag);
 713 
 714       for (int block_id = _max_block_id - 1; block_id >= 0; block_id--) {
 715         clear_block_in_loop(i, block_id);
 716       }
 717       _iterative_dominators = true;
 718     }
 719   }
 720 }
 721 
 722 void ComputeLinearScanOrder::assign_loop_depth(BlockBegin* start_block) {
 723   TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing loop-depth and weight"));
 724   init_visited();
 725 
 726   assert(_work_list.is_empty(), "work list must be empty before processing");
 727   _work_list.append(start_block);
 728 
 729   do {
 730     BlockBegin* cur = _work_list.pop();
 731 
 732     if (!is_visited(cur)) {
 733       set_visited(cur);
 734       TRACE_LINEAR_SCAN(4, tty->print_cr("Computing loop depth for block B%d", cur->block_id()));
 735 
 736       // compute loop-depth and loop-index for the block
 737       assert(cur->loop_depth() == 0, "cannot set loop-depth twice");
 738       int i;
 739       int loop_depth = 0;
 740       int min_loop_idx = -1;
 741       for (i = _num_loops - 1; i >= 0; i--) {
 742         if (is_block_in_loop(i, cur)) {
 743           loop_depth++;
 744           min_loop_idx = i;
 745         }
 746       }
 747       cur->set_loop_depth(loop_depth);
 748       cur->set_loop_index(min_loop_idx);
 749 
 750       // append all unvisited successors to work list
 751       for (i = cur->number_of_sux() - 1; i >= 0; i--) {
 752         _work_list.append(cur->sux_at(i));
 753       }
 754       for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
 755         _work_list.append(cur->exception_handler_at(i));
 756       }
 757     }
 758   } while (!_work_list.is_empty());
 759 }
 760 
 761 
 762 BlockBegin* ComputeLinearScanOrder::common_dominator(BlockBegin* a, BlockBegin* b) {
 763   assert(a != NULL && b != NULL, "must have input blocks");
 764 
 765   _dominator_blocks.clear();
 766   while (a != NULL) {
 767     _dominator_blocks.set_bit(a->block_id());
 768     assert(a->dominator() != NULL || a == _linear_scan_order->at(0), "dominator must be initialized");
 769     a = a->dominator();
 770   }
 771   while (b != NULL && !_dominator_blocks.at(b->block_id())) {
 772     assert(b->dominator() != NULL || b == _linear_scan_order->at(0), "dominator must be initialized");
 773     b = b->dominator();
 774   }
 775 
 776   assert(b != NULL, "could not find dominator");
 777   return b;
 778 }
 779 
 780 void ComputeLinearScanOrder::compute_dominator(BlockBegin* cur, BlockBegin* parent) {
 781   if (cur->dominator() == NULL) {
 782     TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: initializing dominator of B%d to B%d", cur->block_id(), parent->block_id()));
 783     cur->set_dominator(parent);
 784 
 785   } else if (!(cur->is_set(BlockBegin::linear_scan_loop_header_flag) && parent->is_set(BlockBegin::linear_scan_loop_end_flag))) {
 786     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()));
 787     // Does not hold for exception blocks
 788     assert(cur->number_of_preds() > 1 || cur->is_set(BlockBegin::exception_entry_flag), "");
 789     cur->set_dominator(common_dominator(cur->dominator(), parent));
 790   }
 791 
 792   // Additional edge to xhandler of all our successors
 793   // range check elimination needs that the state at the end of a
 794   // block be valid in every block it dominates so cur must dominate
 795   // the exception handlers of its successors.
 796   int num_cur_xhandler = cur->number_of_exception_handlers();
 797   for (int j = 0; j < num_cur_xhandler; j++) {
 798     BlockBegin* xhandler = cur->exception_handler_at(j);
 799     compute_dominator(xhandler, parent);
 800   }
 801 }
 802 
 803 
 804 int ComputeLinearScanOrder::compute_weight(BlockBegin* cur) {
 805   BlockBegin* single_sux = NULL;
 806   if (cur->number_of_sux() == 1) {
 807     single_sux = cur->sux_at(0);
 808   }
 809 
 810   // limit loop-depth to 15 bit (only for security reason, it will never be so big)
 811   int weight = (cur->loop_depth() & 0x7FFF) << 16;
 812 
 813   // general macro for short definition of weight flags
 814   // the first instance of INC_WEIGHT_IF has the highest priority
 815   int cur_bit = 15;
 816   #define INC_WEIGHT_IF(condition) if ((condition)) { weight |= (1 << cur_bit); } cur_bit--;
 817 
 818   // this is necessery for the (very rare) case that two successing blocks have
 819   // the same loop depth, but a different loop index (can happen for endless loops
 820   // with exception handlers)
 821   INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_header_flag));
 822 
 823   // loop end blocks (blocks that end with a backward branch) are added
 824   // after all other blocks of the loop.
 825   INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_end_flag));
 826 
 827   // critical edge split blocks are prefered because than they have a bigger
 828   // proability to be completely empty
 829   INC_WEIGHT_IF(cur->is_set(BlockBegin::critical_edge_split_flag));
 830 
 831   // exceptions should not be thrown in normal control flow, so these blocks
 832   // are added as late as possible
 833   INC_WEIGHT_IF(cur->end()->as_Throw() == NULL  && (single_sux == NULL || single_sux->end()->as_Throw()  == NULL));
 834   INC_WEIGHT_IF(cur->end()->as_Return() == NULL && (single_sux == NULL || single_sux->end()->as_Return() == NULL));
 835 
 836   // exceptions handlers are added as late as possible
 837   INC_WEIGHT_IF(!cur->is_set(BlockBegin::exception_entry_flag));
 838 
 839   // guarantee that weight is > 0
 840   weight |= 1;
 841 
 842   #undef INC_WEIGHT_IF
 843   assert(cur_bit >= 0, "too many flags");
 844   assert(weight > 0, "weight cannot become negative");
 845 
 846   return weight;
 847 }
 848 
 849 bool ComputeLinearScanOrder::ready_for_processing(BlockBegin* cur) {
 850   // Discount the edge just traveled.
 851   // When the number drops to zero, all forward branches were processed
 852   if (dec_forward_branches(cur) != 0) {
 853     return false;
 854   }
 855 
 856   assert(_linear_scan_order->index_of(cur) == -1, "block already processed (block can be ready only once)");
 857   assert(_work_list.index_of(cur) == -1, "block already in work-list (block can be ready only once)");
 858   return true;
 859 }
 860 
 861 void ComputeLinearScanOrder::sort_into_work_list(BlockBegin* cur) {
 862   assert(_work_list.index_of(cur) == -1, "block already in work list");
 863 
 864   int cur_weight = compute_weight(cur);
 865 
 866   // the linear_scan_number is used to cache the weight of a block
 867   cur->set_linear_scan_number(cur_weight);
 868 
 869 #ifndef PRODUCT
 870   if (StressLinearScan) {
 871     _work_list.insert_before(0, cur);
 872     return;
 873   }
 874 #endif
 875 
 876   _work_list.append(NULL); // provide space for new element
 877 
 878   int insert_idx = _work_list.length() - 1;
 879   while (insert_idx > 0 && _work_list.at(insert_idx - 1)->linear_scan_number() > cur_weight) {
 880     _work_list.at_put(insert_idx, _work_list.at(insert_idx - 1));
 881     insert_idx--;
 882   }
 883   _work_list.at_put(insert_idx, cur);
 884 
 885   TRACE_LINEAR_SCAN(3, tty->print_cr("Sorted B%d into worklist. new worklist:", cur->block_id()));
 886   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()));
 887 
 888 #ifdef ASSERT
 889   for (int i = 0; i < _work_list.length(); i++) {
 890     assert(_work_list.at(i)->linear_scan_number() > 0, "weight not set");
 891     assert(i == 0 || _work_list.at(i - 1)->linear_scan_number() <= _work_list.at(i)->linear_scan_number(), "incorrect order in worklist");
 892   }
 893 #endif
 894 }
 895 
 896 void ComputeLinearScanOrder::append_block(BlockBegin* cur) {
 897   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()));
 898   assert(_linear_scan_order->index_of(cur) == -1, "cannot add the same block twice");
 899 
 900   // currently, the linear scan order and code emit order are equal.
 901   // therefore the linear_scan_number and the weight of a block must also
 902   // be equal.
 903   cur->set_linear_scan_number(_linear_scan_order->length());
 904   _linear_scan_order->append(cur);
 905 }
 906 
 907 void ComputeLinearScanOrder::compute_order(BlockBegin* start_block) {
 908   TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing final block order"));
 909 
 910   // the start block is always the first block in the linear scan order
 911   _linear_scan_order = new BlockList(_num_blocks);
 912   append_block(start_block);
 913 
 914   assert(start_block->end()->as_Base() != NULL, "start block must end with Base-instruction");
 915   BlockBegin* std_entry = ((Base*)start_block->end())->std_entry();
 916   BlockBegin* osr_entry = ((Base*)start_block->end())->osr_entry();
 917 
 918   BlockBegin* sux_of_osr_entry = NULL;
 919   if (osr_entry != NULL) {
 920     // special handling for osr entry:
 921     // ignore the edge between the osr entry and its successor for processing
 922     // the osr entry block is added manually below
 923     assert(osr_entry->number_of_sux() == 1, "osr entry must have exactly one successor");
 924     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");
 925 
 926     sux_of_osr_entry = osr_entry->sux_at(0);
 927     dec_forward_branches(sux_of_osr_entry);
 928 
 929     compute_dominator(osr_entry, start_block);
 930     _iterative_dominators = true;
 931   }
 932   compute_dominator(std_entry, start_block);
 933 
 934   // start processing with standard entry block
 935   assert(_work_list.is_empty(), "list must be empty before processing");
 936 
 937   if (ready_for_processing(std_entry)) {
 938     sort_into_work_list(std_entry);
 939   } else {
 940     assert(false, "the std_entry must be ready for processing (otherwise, the method has no start block)");
 941   }
 942 
 943   do {
 944     BlockBegin* cur = _work_list.pop();
 945 
 946     if (cur == sux_of_osr_entry) {
 947       // the osr entry block is ignored in normal processing, it is never added to the
 948       // work list. Instead, it is added as late as possible manually here.
 949       append_block(osr_entry);
 950       compute_dominator(cur, osr_entry);
 951     }
 952     append_block(cur);
 953 
 954     int i;
 955     int num_sux = cur->number_of_sux();
 956     // changed loop order to get "intuitive" order of if- and else-blocks
 957     for (i = 0; i < num_sux; i++) {
 958       BlockBegin* sux = cur->sux_at(i);
 959       compute_dominator(sux, cur);
 960       if (ready_for_processing(sux)) {
 961         sort_into_work_list(sux);
 962       }
 963     }
 964     num_sux = cur->number_of_exception_handlers();
 965     for (i = 0; i < num_sux; i++) {
 966       BlockBegin* sux = cur->exception_handler_at(i);
 967       if (ready_for_processing(sux)) {
 968         sort_into_work_list(sux);
 969       }
 970     }
 971   } while (_work_list.length() > 0);
 972 }
 973 
 974 
 975 bool ComputeLinearScanOrder::compute_dominators_iter() {
 976   bool changed = false;
 977   int num_blocks = _linear_scan_order->length();
 978 
 979   assert(_linear_scan_order->at(0)->dominator() == NULL, "must not have dominator");
 980   assert(_linear_scan_order->at(0)->number_of_preds() == 0, "must not have predecessors");
 981   for (int i = 1; i < num_blocks; i++) {
 982     BlockBegin* block = _linear_scan_order->at(i);
 983 
 984     BlockBegin* dominator = block->pred_at(0);
 985     int num_preds = block->number_of_preds();
 986 
 987     TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: Processing B%d", block->block_id()));
 988 
 989     for (int j = 0; j < num_preds; j++) {
 990 
 991       BlockBegin *pred = block->pred_at(j);
 992       TRACE_LINEAR_SCAN(4, tty->print_cr("   DOM: Subrocessing B%d", pred->block_id()));
 993 
 994       if (block->is_set(BlockBegin::exception_entry_flag)) {
 995         dominator = common_dominator(dominator, pred);
 996         int num_pred_preds = pred->number_of_preds();
 997         for (int k = 0; k < num_pred_preds; k++) {
 998           dominator = common_dominator(dominator, pred->pred_at(k));
 999         }
1000       } else {
1001         dominator = common_dominator(dominator, pred);
1002       }
1003     }
1004 
1005     if (dominator != block->dominator()) {
1006       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()));
1007 
1008       block->set_dominator(dominator);
1009       changed = true;
1010     }
1011   }
1012   return changed;
1013 }
1014 
1015 void ComputeLinearScanOrder::compute_dominators() {
1016   TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing dominators (iterative computation reqired: %d)", _iterative_dominators));
1017 
1018   // iterative computation of dominators is only required for methods with non-natural loops
1019   // and OSR-methods. For all other methods, the dominators computed when generating the
1020   // linear scan block order are correct.
1021   if (_iterative_dominators) {
1022     do {
1023       TRACE_LINEAR_SCAN(1, tty->print_cr("DOM: next iteration of fix-point calculation"));
1024     } while (compute_dominators_iter());
1025   }
1026 
1027   // check that dominators are correct
1028   assert(!compute_dominators_iter(), "fix point not reached");
1029 
1030   // Add Blocks to dominates-Array
1031   int num_blocks = _linear_scan_order->length();
1032   for (int i = 0; i < num_blocks; i++) {
1033     BlockBegin* block = _linear_scan_order->at(i);
1034 
1035     BlockBegin *dom = block->dominator();
1036     if (dom) {
1037       assert(dom->dominator_depth() != -1, "Dominator must have been visited before");
1038       dom->dominates()->append(block);
1039       block->set_dominator_depth(dom->dominator_depth() + 1);
1040     } else {
1041       block->set_dominator_depth(0);
1042     }
1043   }
1044 }
1045 
1046 
1047 #ifndef PRODUCT
1048 void ComputeLinearScanOrder::print_blocks() {
1049   if (TraceLinearScanLevel >= 2) {
1050     tty->print_cr("----- loop information:");
1051     for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
1052       BlockBegin* cur = _linear_scan_order->at(block_idx);
1053 
1054       tty->print("%4d: B%2d: ", cur->linear_scan_number(), cur->block_id());
1055       for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
1056         tty->print ("%d ", is_block_in_loop(loop_idx, cur));
1057       }
1058       tty->print_cr(" -> loop_index: %2d, loop_depth: %2d", cur->loop_index(), cur->loop_depth());
1059     }
1060   }
1061 
1062   if (TraceLinearScanLevel >= 1) {
1063     tty->print_cr("----- linear-scan block order:");
1064     for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
1065       BlockBegin* cur = _linear_scan_order->at(block_idx);
1066       tty->print("%4d: B%2d    loop: %2d  depth: %2d", cur->linear_scan_number(), cur->block_id(), cur->loop_index(), cur->loop_depth());
1067 
1068       tty->print(cur->is_set(BlockBegin::exception_entry_flag)         ? " ex" : "   ");
1069       tty->print(cur->is_set(BlockBegin::critical_edge_split_flag)     ? " ce" : "   ");
1070       tty->print(cur->is_set(BlockBegin::linear_scan_loop_header_flag) ? " lh" : "   ");
1071       tty->print(cur->is_set(BlockBegin::linear_scan_loop_end_flag)    ? " le" : "   ");
1072 
1073       if (cur->dominator() != NULL) {
1074         tty->print("    dom: B%d ", cur->dominator()->block_id());
1075       } else {
1076         tty->print("    dom: NULL ");
1077       }
1078 
1079       if (cur->number_of_preds() > 0) {
1080         tty->print("    preds: ");
1081         for (int j = 0; j < cur->number_of_preds(); j++) {
1082           BlockBegin* pred = cur->pred_at(j);
1083           tty->print("B%d ", pred->block_id());
1084         }
1085       }
1086       if (cur->number_of_sux() > 0) {
1087         tty->print("    sux: ");
1088         for (int j = 0; j < cur->number_of_sux(); j++) {
1089           BlockBegin* sux = cur->sux_at(j);
1090           tty->print("B%d ", sux->block_id());
1091         }
1092       }
1093       if (cur->number_of_exception_handlers() > 0) {
1094         tty->print("    ex: ");
1095         for (int j = 0; j < cur->number_of_exception_handlers(); j++) {
1096           BlockBegin* ex = cur->exception_handler_at(j);
1097           tty->print("B%d ", ex->block_id());
1098         }
1099       }
1100       tty->cr();
1101     }
1102   }
1103 }
1104 #endif
1105 
1106 #ifdef ASSERT
1107 void ComputeLinearScanOrder::verify() {
1108   assert(_linear_scan_order->length() == _num_blocks, "wrong number of blocks in list");
1109 
1110   if (StressLinearScan) {
1111     // blocks are scrambled when StressLinearScan is used
1112     return;
1113   }
1114 
1115   // check that all successors of a block have a higher linear-scan-number
1116   // and that all predecessors of a block have a lower linear-scan-number
1117   // (only backward branches of loops are ignored)
1118   int i;
1119   for (i = 0; i < _linear_scan_order->length(); i++) {
1120     BlockBegin* cur = _linear_scan_order->at(i);
1121 
1122     assert(cur->linear_scan_number() == i, "incorrect linear_scan_number");
1123     assert(cur->linear_scan_number() >= 0 && cur->linear_scan_number() == _linear_scan_order->index_of(cur), "incorrect linear_scan_number");
1124 
1125     int j;
1126     for (j = cur->number_of_sux() - 1; j >= 0; j--) {
1127       BlockBegin* sux = cur->sux_at(j);
1128 
1129       assert(sux->linear_scan_number() >= 0 && sux->linear_scan_number() == _linear_scan_order->index_of(sux), "incorrect linear_scan_number");
1130       if (!sux->is_set(BlockBegin::backward_branch_target_flag)) {
1131         assert(cur->linear_scan_number() < sux->linear_scan_number(), "invalid order");
1132       }
1133       if (cur->loop_depth() == sux->loop_depth()) {
1134         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");
1135       }
1136     }
1137 
1138     for (j = cur->number_of_preds() - 1; j >= 0; j--) {
1139       BlockBegin* pred = cur->pred_at(j);
1140 
1141       assert(pred->linear_scan_number() >= 0 && pred->linear_scan_number() == _linear_scan_order->index_of(pred), "incorrect linear_scan_number");
1142       if (!cur->is_set(BlockBegin::backward_branch_target_flag)) {
1143         assert(cur->linear_scan_number() > pred->linear_scan_number(), "invalid order");
1144       }
1145       if (cur->loop_depth() == pred->loop_depth()) {
1146         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");
1147       }
1148 
1149       assert(cur->dominator()->linear_scan_number() <= cur->pred_at(j)->linear_scan_number(), "dominator must be before predecessors");
1150     }
1151 
1152     // check dominator
1153     if (i == 0) {
1154       assert(cur->dominator() == NULL, "first block has no dominator");
1155     } else {
1156       assert(cur->dominator() != NULL, "all but first block must have dominator");
1157     }
1158     // Assertion does not hold for exception handlers
1159     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");
1160   }
1161 
1162   // check that all loops are continuous
1163   for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
1164     int block_idx = 0;
1165     assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "the first block must not be present in any loop");
1166 
1167     // skip blocks before the loop
1168     while (block_idx < _num_blocks && !is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
1169       block_idx++;
1170     }
1171     // skip blocks of loop
1172     while (block_idx < _num_blocks && is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
1173       block_idx++;
1174     }
1175     // after the first non-loop block, there must not be another loop-block
1176     while (block_idx < _num_blocks) {
1177       assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "loop not continuous in linear-scan order");
1178       block_idx++;
1179     }
1180   }
1181 }
1182 #endif
1183 
1184 
1185 void IR::compute_code() {
1186   assert(is_valid(), "IR must be valid");
1187 
1188   ComputeLinearScanOrder compute_order(compilation(), start());
1189   _num_loops = compute_order.num_loops();
1190   _code = compute_order.linear_scan_order();
1191 }
1192 
1193 
1194 void IR::compute_use_counts() {
1195   // make sure all values coming out of this block get evaluated.
1196   int num_blocks = _code->length();
1197   for (int i = 0; i < num_blocks; i++) {
1198     _code->at(i)->end()->state()->pin_stack_for_linear_scan();
1199   }
1200 
1201   // compute use counts
1202   UseCountComputer::compute(_code);
1203 }
1204 
1205 
1206 void IR::iterate_preorder(BlockClosure* closure) {
1207   assert(is_valid(), "IR must be valid");
1208   start()->iterate_preorder(closure);
1209 }
1210 
1211 
1212 void IR::iterate_postorder(BlockClosure* closure) {
1213   assert(is_valid(), "IR must be valid");
1214   start()->iterate_postorder(closure);
1215 }
1216 
1217 void IR::iterate_linear_scan_order(BlockClosure* closure) {
1218   linear_scan_order()->iterate_forward(closure);
1219 }
1220 
1221 
1222 #ifndef PRODUCT
1223 class BlockPrinter: public BlockClosure {
1224  private:
1225   InstructionPrinter* _ip;
1226   bool                _cfg_only;
1227   bool                _live_only;
1228 
1229  public:
1230   BlockPrinter(InstructionPrinter* ip, bool cfg_only, bool live_only = false) {
1231     _ip       = ip;
1232     _cfg_only = cfg_only;
1233     _live_only = live_only;
1234   }
1235 
1236   virtual void block_do(BlockBegin* block) {
1237     if (_cfg_only) {
1238       _ip->print_instr(block); tty->cr();
1239     } else {
1240       block->print_block(*_ip, _live_only);
1241     }
1242   }
1243 };
1244 
1245 
1246 void IR::print(BlockBegin* start, bool cfg_only, bool live_only) {
1247   ttyLocker ttyl;
1248   InstructionPrinter ip(!cfg_only);
1249   BlockPrinter bp(&ip, cfg_only, live_only);
1250   start->iterate_preorder(&bp);
1251   tty->cr();
1252 }
1253 
1254 void IR::print(bool cfg_only, bool live_only) {
1255   if (is_valid()) {
1256     print(start(), cfg_only, live_only);
1257   } else {
1258     tty->print_cr("invalid IR");
1259   }
1260 }
1261 
1262 
1263 define_array(BlockListArray, BlockList*)
1264 define_stack(BlockListList, BlockListArray)
1265 
1266 class PredecessorValidator : public BlockClosure {
1267  private:
1268   BlockListList* _predecessors;
1269   BlockList*     _blocks;
1270 
1271   static int cmp(BlockBegin** a, BlockBegin** b) {
1272     return (*a)->block_id() - (*b)->block_id();
1273   }
1274 
1275  public:
1276   PredecessorValidator(IR* hir) {
1277     ResourceMark rm;
1278     _predecessors = new BlockListList(BlockBegin::number_of_blocks(), NULL);
1279     _blocks = new BlockList();
1280 
1281     int i;
1282     hir->start()->iterate_preorder(this);
1283     if (hir->code() != NULL) {
1284       assert(hir->code()->length() == _blocks->length(), "must match");
1285       for (i = 0; i < _blocks->length(); i++) {
1286         assert(hir->code()->contains(_blocks->at(i)), "should be in both lists");
1287       }
1288     }
1289 
1290     for (i = 0; i < _blocks->length(); i++) {
1291       BlockBegin* block = _blocks->at(i);
1292       BlockList* preds = _predecessors->at(block->block_id());
1293       if (preds == NULL) {
1294         assert(block->number_of_preds() == 0, "should be the same");
1295         continue;
1296       }
1297 
1298       // clone the pred list so we can mutate it
1299       BlockList* pred_copy = new BlockList();
1300       int j;
1301       for (j = 0; j < block->number_of_preds(); j++) {
1302         pred_copy->append(block->pred_at(j));
1303       }
1304       // sort them in the same order
1305       preds->sort(cmp);
1306       pred_copy->sort(cmp);
1307       int length = MIN2(preds->length(), block->number_of_preds());
1308       for (j = 0; j < block->number_of_preds(); j++) {
1309         assert(preds->at(j) == pred_copy->at(j), "must match");
1310       }
1311 
1312       assert(preds->length() == block->number_of_preds(), "should be the same");
1313     }
1314   }
1315 
1316   virtual void block_do(BlockBegin* block) {
1317     _blocks->append(block);
1318     BlockEnd* be = block->end();
1319     int n = be->number_of_sux();
1320     int i;
1321     for (i = 0; i < n; i++) {
1322       BlockBegin* sux = be->sux_at(i);
1323       assert(!sux->is_set(BlockBegin::exception_entry_flag), "must not be xhandler");
1324 
1325       BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
1326       if (preds == NULL) {
1327         preds = new BlockList();
1328         _predecessors->at_put(sux->block_id(), preds);
1329       }
1330       preds->append(block);
1331     }
1332 
1333     n = block->number_of_exception_handlers();
1334     for (i = 0; i < n; i++) {
1335       BlockBegin* sux = block->exception_handler_at(i);
1336       assert(sux->is_set(BlockBegin::exception_entry_flag), "must be xhandler");
1337 
1338       BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
1339       if (preds == NULL) {
1340         preds = new BlockList();
1341         _predecessors->at_put(sux->block_id(), preds);
1342       }
1343       preds->append(block);
1344     }
1345   }
1346 };
1347 
1348 class VerifyBlockBeginField : public BlockClosure {
1349 
1350 public:
1351 
1352   virtual void block_do(BlockBegin *block) {
1353     for ( Instruction *cur = block; cur != NULL; cur = cur->next()) {
1354       assert(cur->block() == block, "Block begin is not correct");
1355     }
1356   }
1357 };
1358 
1359 void IR::verify() {
1360 #ifdef ASSERT
1361   PredecessorValidator pv(this);
1362   VerifyBlockBeginField verifier;
1363   this->iterate_postorder(&verifier);
1364 #endif
1365 }
1366 
1367 #endif // PRODUCT
1368 
1369 void SubstitutionResolver::visit(Value* v) {
1370   Value v0 = *v;
1371   if (v0) {
1372     Value vs = v0->subst();
1373     if (vs != v0) {
1374       *v = v0->subst();
1375     }
1376   }
1377 }
1378 
1379 #ifdef ASSERT
1380 class SubstitutionChecker: public ValueVisitor {
1381   void visit(Value* v) {
1382     Value v0 = *v;
1383     if (v0) {
1384       Value vs = v0->subst();
1385       assert(vs == v0, "missed substitution");
1386     }
1387   }
1388 };
1389 #endif
1390 
1391 
1392 void SubstitutionResolver::block_do(BlockBegin* block) {
1393   Instruction* last = NULL;
1394   for (Instruction* n = block; n != NULL;) {
1395     n->values_do(this);
1396     // need to remove this instruction from the instruction stream
1397     if (n->subst() != n) {
1398       assert(last != NULL, "must have last");
1399       last->set_next(n->next());
1400     } else {
1401       last = n;
1402     }
1403     n = last->next();
1404   }
1405 
1406 #ifdef ASSERT
1407   SubstitutionChecker check_substitute;
1408   if (block->state()) block->state()->values_do(&check_substitute);
1409   block->block_values_do(&check_substitute);
1410   if (block->end() && block->end()->state()) block->end()->state()->values_do(&check_substitute);
1411 #endif
1412 }