aoqi@0: /*
aoqi@0:  * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
aoqi@0:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
aoqi@0:  *
aoqi@0:  * This code is free software; you can redistribute it and/or modify it
aoqi@0:  * under the terms of the GNU General Public License version 2 only, as
aoqi@0:  * published by the Free Software Foundation.
aoqi@0:  *
aoqi@0:  * This code is distributed in the hope that it will be useful, but WITHOUT
aoqi@0:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
aoqi@0:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
aoqi@0:  * version 2 for more details (a copy is included in the LICENSE file that
aoqi@0:  * accompanied this code).
aoqi@0:  *
aoqi@0:  * You should have received a copy of the GNU General Public License version
aoqi@0:  * 2 along with this work; if not, write to the Free Software Foundation,
aoqi@0:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
aoqi@0:  *
aoqi@0:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
aoqi@0:  * or visit www.oracle.com if you need additional information or have any
aoqi@0:  * questions.
aoqi@0:  *
aoqi@0:  */
aoqi@0: 
aoqi@0: #include "precompiled.hpp"
aoqi@0: #include "c1/c1_Compilation.hpp"
aoqi@0: #include "c1/c1_FrameMap.hpp"
aoqi@0: #include "c1/c1_GraphBuilder.hpp"
aoqi@0: #include "c1/c1_IR.hpp"
aoqi@0: #include "c1/c1_InstructionPrinter.hpp"
aoqi@0: #include "c1/c1_Optimizer.hpp"
aoqi@0: #include "utilities/bitMap.inline.hpp"
aoqi@0: 
aoqi@0: 
aoqi@0: // Implementation of XHandlers
aoqi@0: //
aoqi@0: // Note: This code could eventually go away if we are
aoqi@0: //       just using the ciExceptionHandlerStream.
aoqi@0: 
aoqi@0: XHandlers::XHandlers(ciMethod* method) : _list(method->exception_table_length()) {
aoqi@0:   ciExceptionHandlerStream s(method);
aoqi@0:   while (!s.is_done()) {
aoqi@0:     _list.append(new XHandler(s.handler()));
aoqi@0:     s.next();
aoqi@0:   }
aoqi@0:   assert(s.count() == method->exception_table_length(), "exception table lengths inconsistent");
aoqi@0: }
aoqi@0: 
aoqi@0: // deep copy of all XHandler contained in list
aoqi@0: XHandlers::XHandlers(XHandlers* other) :
aoqi@0:   _list(other->length())
aoqi@0: {
aoqi@0:   for (int i = 0; i < other->length(); i++) {
aoqi@0:     _list.append(new XHandler(other->handler_at(i)));
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: // Returns whether a particular exception type can be caught.  Also
aoqi@0: // returns true if klass is unloaded or any exception handler
aoqi@0: // classes are unloaded.  type_is_exact indicates whether the throw
aoqi@0: // is known to be exactly that class or it might throw a subtype.
aoqi@0: bool XHandlers::could_catch(ciInstanceKlass* klass, bool type_is_exact) const {
aoqi@0:   // the type is unknown so be conservative
aoqi@0:   if (!klass->is_loaded()) {
aoqi@0:     return true;
aoqi@0:   }
aoqi@0: 
aoqi@0:   for (int i = 0; i < length(); i++) {
aoqi@0:     XHandler* handler = handler_at(i);
aoqi@0:     if (handler->is_catch_all()) {
aoqi@0:       // catch of ANY
aoqi@0:       return true;
aoqi@0:     }
aoqi@0:     ciInstanceKlass* handler_klass = handler->catch_klass();
aoqi@0:     // if it's unknown it might be catchable
aoqi@0:     if (!handler_klass->is_loaded()) {
aoqi@0:       return true;
aoqi@0:     }
aoqi@0:     // if the throw type is definitely a subtype of the catch type
aoqi@0:     // then it can be caught.
aoqi@0:     if (klass->is_subtype_of(handler_klass)) {
aoqi@0:       return true;
aoqi@0:     }
aoqi@0:     if (!type_is_exact) {
aoqi@0:       // If the type isn't exactly known then it can also be caught by
aoqi@0:       // catch statements where the inexact type is a subtype of the
aoqi@0:       // catch type.
aoqi@0:       // given: foo extends bar extends Exception
aoqi@0:       // throw bar can be caught by catch foo, catch bar, and catch
aoqi@0:       // Exception, however it can't be caught by any handlers without
aoqi@0:       // bar in its type hierarchy.
aoqi@0:       if (handler_klass->is_subtype_of(klass)) {
aoqi@0:         return true;
aoqi@0:       }
aoqi@0:     }
aoqi@0:   }
aoqi@0: 
aoqi@0:   return false;
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: bool XHandlers::equals(XHandlers* others) const {
aoqi@0:   if (others == NULL) return false;
aoqi@0:   if (length() != others->length()) return false;
aoqi@0: 
aoqi@0:   for (int i = 0; i < length(); i++) {
aoqi@0:     if (!handler_at(i)->equals(others->handler_at(i))) return false;
aoqi@0:   }
aoqi@0:   return true;
aoqi@0: }
aoqi@0: 
aoqi@0: bool XHandler::equals(XHandler* other) const {
aoqi@0:   assert(entry_pco() != -1 && other->entry_pco() != -1, "must have entry_pco");
aoqi@0: 
aoqi@0:   if (entry_pco() != other->entry_pco()) return false;
aoqi@0:   if (scope_count() != other->scope_count()) return false;
aoqi@0:   if (_desc != other->_desc) return false;
aoqi@0: 
aoqi@0:   assert(entry_block() == other->entry_block(), "entry_block must be equal when entry_pco is equal");
aoqi@0:   return true;
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: // Implementation of IRScope
aoqi@0: BlockBegin* IRScope::build_graph(Compilation* compilation, int osr_bci) {
aoqi@0:   GraphBuilder gm(compilation, this);
aoqi@0:   NOT_PRODUCT(if (PrintValueNumbering && Verbose) gm.print_stats());
aoqi@0:   if (compilation->bailed_out()) return NULL;
aoqi@0:   return gm.start();
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: IRScope::IRScope(Compilation* compilation, IRScope* caller, int caller_bci, ciMethod* method, int osr_bci, bool create_graph)
aoqi@0: : _callees(2)
aoqi@0: , _compilation(compilation)
aoqi@0: , _requires_phi_function(method->max_locals())
aoqi@0: {
aoqi@0:   _caller             = caller;
aoqi@0:   _level              = caller == NULL ?  0 : caller->level() + 1;
aoqi@0:   _method             = method;
aoqi@0:   _xhandlers          = new XHandlers(method);
aoqi@0:   _number_of_locks    = 0;
aoqi@0:   _monitor_pairing_ok = method->has_balanced_monitors();
aoqi@0:   _wrote_final        = false;
aoqi@0:   _start              = NULL;
aoqi@0: 
aoqi@0:   if (osr_bci == -1) {
aoqi@0:     _requires_phi_function.clear();
aoqi@0:   } else {
aoqi@0:         // selective creation of phi functions is not possibel in osr-methods
aoqi@0:     _requires_phi_function.set_range(0, method->max_locals());
aoqi@0:   }
aoqi@0: 
aoqi@0:   assert(method->holder()->is_loaded() , "method holder must be loaded");
aoqi@0: 
aoqi@0:   // build graph if monitor pairing is ok
aoqi@0:   if (create_graph && monitor_pairing_ok()) _start = build_graph(compilation, osr_bci);
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: int IRScope::max_stack() const {
aoqi@0:   int my_max = method()->max_stack();
aoqi@0:   int callee_max = 0;
aoqi@0:   for (int i = 0; i < number_of_callees(); i++) {
aoqi@0:     callee_max = MAX2(callee_max, callee_no(i)->max_stack());
aoqi@0:   }
aoqi@0:   return my_max + callee_max;
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: bool IRScopeDebugInfo::should_reexecute() {
aoqi@0:   ciMethod* cur_method = scope()->method();
aoqi@0:   int       cur_bci    = bci();
aoqi@0:   if (cur_method != NULL && cur_bci != SynchronizationEntryBCI) {
aoqi@0:     Bytecodes::Code code = cur_method->java_code_at_bci(cur_bci);
aoqi@0:     return Interpreter::bytecode_should_reexecute(code);
aoqi@0:   } else
aoqi@0:     return false;
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: // Implementation of CodeEmitInfo
aoqi@0: 
aoqi@0: // Stack must be NON-null
aoqi@0: CodeEmitInfo::CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers, bool deoptimize_on_exception)
aoqi@0:   : _scope(stack->scope())
aoqi@0:   , _scope_debug_info(NULL)
aoqi@0:   , _oop_map(NULL)
aoqi@0:   , _stack(stack)
aoqi@0:   , _exception_handlers(exception_handlers)
aoqi@0:   , _is_method_handle_invoke(false)
aoqi@0:   , _deoptimize_on_exception(deoptimize_on_exception) {
aoqi@0:   assert(_stack != NULL, "must be non null");
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: CodeEmitInfo::CodeEmitInfo(CodeEmitInfo* info, ValueStack* stack)
aoqi@0:   : _scope(info->_scope)
aoqi@0:   , _exception_handlers(NULL)
aoqi@0:   , _scope_debug_info(NULL)
aoqi@0:   , _oop_map(NULL)
aoqi@0:   , _stack(stack == NULL ? info->_stack : stack)
aoqi@0:   , _is_method_handle_invoke(info->_is_method_handle_invoke)
aoqi@0:   , _deoptimize_on_exception(info->_deoptimize_on_exception) {
aoqi@0: 
aoqi@0:   // deep copy of exception handlers
aoqi@0:   if (info->_exception_handlers != NULL) {
aoqi@0:     _exception_handlers = new XHandlers(info->_exception_handlers);
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: void CodeEmitInfo::record_debug_info(DebugInformationRecorder* recorder, int pc_offset) {
aoqi@0:   // record the safepoint before recording the debug info for enclosing scopes
aoqi@0:   recorder->add_safepoint(pc_offset, _oop_map->deep_copy());
aoqi@0:   _scope_debug_info->record_debug_info(recorder, pc_offset, true/*topmost*/, _is_method_handle_invoke);
aoqi@0:   recorder->end_safepoint(pc_offset);
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: void CodeEmitInfo::add_register_oop(LIR_Opr opr) {
aoqi@0:   assert(_oop_map != NULL, "oop map must already exist");
aoqi@0:   assert(opr->is_single_cpu(), "should not call otherwise");
aoqi@0: 
aoqi@0:   VMReg name = frame_map()->regname(opr);
aoqi@0:   _oop_map->set_oop(name);
aoqi@0: }
aoqi@0: 
aoqi@0: // Mirror the stack size calculation in the deopt code
aoqi@0: // How much stack space would we need at this point in the program in
aoqi@0: // case of deoptimization?
aoqi@0: int CodeEmitInfo::interpreter_frame_size() const {
aoqi@0:   ValueStack* state = _stack;
aoqi@0:   int size = 0;
aoqi@0:   int callee_parameters = 0;
aoqi@0:   int callee_locals = 0;
aoqi@0:   int extra_args = state->scope()->method()->max_stack() - state->stack_size();
aoqi@0: 
aoqi@0:   while (state != NULL) {
aoqi@0:     int locks = state->locks_size();
aoqi@0:     int temps = state->stack_size();
aoqi@0:     bool is_top_frame = (state == _stack);
aoqi@0:     ciMethod* method = state->scope()->method();
aoqi@0: 
aoqi@0:     int frame_size = BytesPerWord * Interpreter::size_activation(method->max_stack(),
aoqi@0:                                                                  temps + callee_parameters,
aoqi@0:                                                                  extra_args,
aoqi@0:                                                                  locks,
aoqi@0:                                                                  callee_parameters,
aoqi@0:                                                                  callee_locals,
aoqi@0:                                                                  is_top_frame);
aoqi@0:     size += frame_size;
aoqi@0: 
aoqi@0:     callee_parameters = method->size_of_parameters();
aoqi@0:     callee_locals = method->max_locals();
aoqi@0:     extra_args = 0;
aoqi@0:     state = state->caller_state();
aoqi@0:   }
aoqi@0:   return size + Deoptimization::last_frame_adjust(0, callee_locals) * BytesPerWord;
aoqi@0: }
aoqi@0: 
aoqi@0: // Implementation of IR
aoqi@0: 
aoqi@0: IR::IR(Compilation* compilation, ciMethod* method, int osr_bci) :
aoqi@0:     _locals_size(in_WordSize(-1))
aoqi@0:   , _num_loops(0) {
aoqi@0:   // setup IR fields
aoqi@0:   _compilation = compilation;
aoqi@0:   _top_scope   = new IRScope(compilation, NULL, -1, method, osr_bci, true);
aoqi@0:   _code        = NULL;
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: void IR::optimize_blocks() {
aoqi@0:   Optimizer opt(this);
aoqi@0:   if (!compilation()->profile_branches()) {
aoqi@0:     if (DoCEE) {
aoqi@0:       opt.eliminate_conditional_expressions();
aoqi@0: #ifndef PRODUCT
aoqi@0:       if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after CEE"); print(true); }
aoqi@0:       if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after CEE"); print(false); }
aoqi@0: #endif
aoqi@0:     }
aoqi@0:     if (EliminateBlocks) {
aoqi@0:       opt.eliminate_blocks();
aoqi@0: #ifndef PRODUCT
aoqi@0:       if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after block elimination"); print(true); }
aoqi@0:       if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after block elimination"); print(false); }
aoqi@0: #endif
aoqi@0:     }
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: void IR::eliminate_null_checks() {
aoqi@0:   Optimizer opt(this);
aoqi@0:   if (EliminateNullChecks) {
aoqi@0:     opt.eliminate_null_checks();
aoqi@0: #ifndef PRODUCT
aoqi@0:     if (PrintCFG || PrintCFG1) { tty->print_cr("CFG after null check elimination"); print(true); }
aoqi@0:     if (PrintIR  || PrintIR1 ) { tty->print_cr("IR after null check elimination"); print(false); }
aoqi@0: #endif
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: static int sort_pairs(BlockPair** a, BlockPair** b) {
aoqi@0:   if ((*a)->from() == (*b)->from()) {
aoqi@0:     return (*a)->to()->block_id() - (*b)->to()->block_id();
aoqi@0:   } else {
aoqi@0:     return (*a)->from()->block_id() - (*b)->from()->block_id();
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: class CriticalEdgeFinder: public BlockClosure {
aoqi@0:   BlockPairList blocks;
aoqi@0:   IR*       _ir;
aoqi@0: 
aoqi@0:  public:
aoqi@0:   CriticalEdgeFinder(IR* ir): _ir(ir) {}
aoqi@0:   void block_do(BlockBegin* bb) {
aoqi@0:     BlockEnd* be = bb->end();
aoqi@0:     int nos = be->number_of_sux();
aoqi@0:     if (nos >= 2) {
aoqi@0:       for (int i = 0; i < nos; i++) {
aoqi@0:         BlockBegin* sux = be->sux_at(i);
aoqi@0:         if (sux->number_of_preds() >= 2) {
aoqi@0:           blocks.append(new BlockPair(bb, sux));
aoqi@0:         }
aoqi@0:       }
aoqi@0:     }
aoqi@0:   }
aoqi@0: 
aoqi@0:   void split_edges() {
aoqi@0:     BlockPair* last_pair = NULL;
aoqi@0:     blocks.sort(sort_pairs);
aoqi@0:     for (int i = 0; i < blocks.length(); i++) {
aoqi@0:       BlockPair* pair = blocks.at(i);
aoqi@0:       if (last_pair != NULL && pair->is_same(last_pair)) continue;
aoqi@0:       BlockBegin* from = pair->from();
aoqi@0:       BlockBegin* to = pair->to();
aoqi@0:       BlockBegin* split = from->insert_block_between(to);
aoqi@0: #ifndef PRODUCT
aoqi@0:       if ((PrintIR || PrintIR1) && Verbose) {
aoqi@0:         tty->print_cr("Split critical edge B%d -> B%d (new block B%d)",
aoqi@0:                       from->block_id(), to->block_id(), split->block_id());
aoqi@0:       }
aoqi@0: #endif
aoqi@0:       last_pair = pair;
aoqi@0:     }
aoqi@0:   }
aoqi@0: };
aoqi@0: 
aoqi@0: void IR::split_critical_edges() {
aoqi@0:   CriticalEdgeFinder cef(this);
aoqi@0: 
aoqi@0:   iterate_preorder(&cef);
aoqi@0:   cef.split_edges();
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: class UseCountComputer: public ValueVisitor, BlockClosure {
aoqi@0:  private:
aoqi@0:   void visit(Value* n) {
aoqi@0:     // Local instructions and Phis for expression stack values at the
aoqi@0:     // start of basic blocks are not added to the instruction list
aoqi@0:     if (!(*n)->is_linked() && (*n)->can_be_linked()) {
aoqi@0:       assert(false, "a node was not appended to the graph");
aoqi@0:       Compilation::current()->bailout("a node was not appended to the graph");
aoqi@0:     }
aoqi@0:     // use n's input if not visited before
aoqi@0:     if (!(*n)->is_pinned() && !(*n)->has_uses()) {
aoqi@0:       // note: a) if the instruction is pinned, it will be handled by compute_use_count
aoqi@0:       //       b) if the instruction has uses, it was touched before
aoqi@0:       //       => in both cases we don't need to update n's values
aoqi@0:       uses_do(n);
aoqi@0:     }
aoqi@0:     // use n
aoqi@0:     (*n)->_use_count++;
aoqi@0:   }
aoqi@0: 
aoqi@0:   Values* worklist;
aoqi@0:   int depth;
aoqi@0:   enum {
aoqi@0:     max_recurse_depth = 20
aoqi@0:   };
aoqi@0: 
aoqi@0:   void uses_do(Value* n) {
aoqi@0:     depth++;
aoqi@0:     if (depth > max_recurse_depth) {
aoqi@0:       // don't allow the traversal to recurse too deeply
aoqi@0:       worklist->push(*n);
aoqi@0:     } else {
aoqi@0:       (*n)->input_values_do(this);
aoqi@0:       // special handling for some instructions
aoqi@0:       if ((*n)->as_BlockEnd() != NULL) {
aoqi@0:         // note on BlockEnd:
aoqi@0:         //   must 'use' the stack only if the method doesn't
aoqi@0:         //   terminate, however, in those cases stack is empty
aoqi@0:         (*n)->state_values_do(this);
aoqi@0:       }
aoqi@0:     }
aoqi@0:     depth--;
aoqi@0:   }
aoqi@0: 
aoqi@0:   void block_do(BlockBegin* b) {
aoqi@0:     depth = 0;
aoqi@0:     // process all pinned nodes as the roots of expression trees
aoqi@0:     for (Instruction* n = b; n != NULL; n = n->next()) {
aoqi@0:       if (n->is_pinned()) uses_do(&n);
aoqi@0:     }
aoqi@0:     assert(depth == 0, "should have counted back down");
aoqi@0: 
aoqi@0:     // now process any unpinned nodes which recursed too deeply
aoqi@0:     while (worklist->length() > 0) {
aoqi@0:       Value t = worklist->pop();
aoqi@0:       if (!t->is_pinned()) {
aoqi@0:         // compute the use count
aoqi@0:         uses_do(&t);
aoqi@0: 
aoqi@0:         // pin the instruction so that LIRGenerator doesn't recurse
aoqi@0:         // too deeply during it's evaluation.
aoqi@0:         t->pin();
aoqi@0:       }
aoqi@0:     }
aoqi@0:     assert(depth == 0, "should have counted back down");
aoqi@0:   }
aoqi@0: 
aoqi@0:   UseCountComputer() {
aoqi@0:     worklist = new Values();
aoqi@0:     depth = 0;
aoqi@0:   }
aoqi@0: 
aoqi@0:  public:
aoqi@0:   static void compute(BlockList* blocks) {
aoqi@0:     UseCountComputer ucc;
aoqi@0:     blocks->iterate_backward(&ucc);
aoqi@0:   }
aoqi@0: };
aoqi@0: 
aoqi@0: 
aoqi@0: // helper macro for short definition of trace-output inside code
aoqi@0: #ifndef PRODUCT
aoqi@0:   #define TRACE_LINEAR_SCAN(level, code)       \
aoqi@0:     if (TraceLinearScanLevel >= level) {       \
aoqi@0:       code;                                    \
aoqi@0:     }
aoqi@0: #else
aoqi@0:   #define TRACE_LINEAR_SCAN(level, code)
aoqi@0: #endif
aoqi@0: 
aoqi@0: class ComputeLinearScanOrder : public StackObj {
aoqi@0:  private:
aoqi@0:   int        _max_block_id;        // the highest block_id of a block
aoqi@0:   int        _num_blocks;          // total number of blocks (smaller than _max_block_id)
aoqi@0:   int        _num_loops;           // total number of loops
aoqi@0:   bool       _iterative_dominators;// method requires iterative computation of dominatiors
aoqi@0: 
aoqi@0:   BlockList* _linear_scan_order;   // the resulting list of blocks in correct order
aoqi@0: 
aoqi@0:   BitMap     _visited_blocks;      // used for recursive processing of blocks
aoqi@0:   BitMap     _active_blocks;       // used for recursive processing of blocks
aoqi@0:   BitMap     _dominator_blocks;    // temproary BitMap used for computation of dominator
aoqi@0:   intArray   _forward_branches;    // number of incoming forward branches for each block
aoqi@0:   BlockList  _loop_end_blocks;     // list of all loop end blocks collected during count_edges
aoqi@0:   BitMap2D   _loop_map;            // two-dimensional bit set: a bit is set if a block is contained in a loop
aoqi@0:   BlockList  _work_list;           // temporary list (used in mark_loops and compute_order)
aoqi@0:   BlockList  _loop_headers;
aoqi@0: 
aoqi@0:   Compilation* _compilation;
aoqi@0: 
aoqi@0:   // accessors for _visited_blocks and _active_blocks
aoqi@0:   void init_visited()                     { _active_blocks.clear(); _visited_blocks.clear(); }
aoqi@0:   bool is_visited(BlockBegin* b) const    { return _visited_blocks.at(b->block_id()); }
aoqi@0:   bool is_active(BlockBegin* b) const     { return _active_blocks.at(b->block_id()); }
aoqi@0:   void set_visited(BlockBegin* b)         { assert(!is_visited(b), "already set"); _visited_blocks.set_bit(b->block_id()); }
aoqi@0:   void set_active(BlockBegin* b)          { assert(!is_active(b), "already set");  _active_blocks.set_bit(b->block_id()); }
aoqi@0:   void clear_active(BlockBegin* b)        { assert(is_active(b), "not already");   _active_blocks.clear_bit(b->block_id()); }
aoqi@0: 
aoqi@0:   // accessors for _forward_branches
aoqi@0:   void inc_forward_branches(BlockBegin* b) { _forward_branches.at_put(b->block_id(), _forward_branches.at(b->block_id()) + 1); }
aoqi@0:   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()); }
aoqi@0: 
aoqi@0:   // accessors for _loop_map
aoqi@0:   bool is_block_in_loop   (int loop_idx, BlockBegin* b) const { return _loop_map.at(loop_idx, b->block_id()); }
aoqi@0:   void set_block_in_loop  (int loop_idx, BlockBegin* b)       { _loop_map.set_bit(loop_idx, b->block_id()); }
aoqi@0:   void clear_block_in_loop(int loop_idx, int block_id)        { _loop_map.clear_bit(loop_idx, block_id); }
aoqi@0: 
aoqi@0:   // count edges between blocks
aoqi@0:   void count_edges(BlockBegin* cur, BlockBegin* parent);
aoqi@0: 
aoqi@0:   // loop detection
aoqi@0:   void mark_loops();
aoqi@0:   void clear_non_natural_loops(BlockBegin* start_block);
aoqi@0:   void assign_loop_depth(BlockBegin* start_block);
aoqi@0: 
aoqi@0:   // computation of final block order
aoqi@0:   BlockBegin* common_dominator(BlockBegin* a, BlockBegin* b);
aoqi@0:   void compute_dominator(BlockBegin* cur, BlockBegin* parent);
aoqi@0:   int  compute_weight(BlockBegin* cur);
aoqi@0:   bool ready_for_processing(BlockBegin* cur);
aoqi@0:   void sort_into_work_list(BlockBegin* b);
aoqi@0:   void append_block(BlockBegin* cur);
aoqi@0:   void compute_order(BlockBegin* start_block);
aoqi@0: 
aoqi@0:   // fixup of dominators for non-natural loops
aoqi@0:   bool compute_dominators_iter();
aoqi@0:   void compute_dominators();
aoqi@0: 
aoqi@0:   // debug functions
aoqi@0:   NOT_PRODUCT(void print_blocks();)
aoqi@0:   DEBUG_ONLY(void verify();)
aoqi@0: 
aoqi@0:   Compilation* compilation() const { return _compilation; }
aoqi@0:  public:
aoqi@0:   ComputeLinearScanOrder(Compilation* c, BlockBegin* start_block);
aoqi@0: 
aoqi@0:   // accessors for final result
aoqi@0:   BlockList* linear_scan_order() const    { return _linear_scan_order; }
aoqi@0:   int        num_loops() const            { return _num_loops; }
aoqi@0: };
aoqi@0: 
aoqi@0: 
aoqi@0: ComputeLinearScanOrder::ComputeLinearScanOrder(Compilation* c, BlockBegin* start_block) :
aoqi@0:   _max_block_id(BlockBegin::number_of_blocks()),
aoqi@0:   _num_blocks(0),
aoqi@0:   _num_loops(0),
aoqi@0:   _iterative_dominators(false),
aoqi@0:   _visited_blocks(_max_block_id),
aoqi@0:   _active_blocks(_max_block_id),
aoqi@0:   _dominator_blocks(_max_block_id),
aoqi@0:   _forward_branches(_max_block_id, 0),
aoqi@0:   _loop_end_blocks(8),
aoqi@0:   _work_list(8),
aoqi@0:   _linear_scan_order(NULL), // initialized later with correct size
aoqi@0:   _loop_map(0, 0),          // initialized later with correct size
aoqi@0:   _compilation(c)
aoqi@0: {
aoqi@0:   TRACE_LINEAR_SCAN(2, tty->print_cr("***** computing linear-scan block order"));
aoqi@0: 
aoqi@0:   init_visited();
aoqi@0:   count_edges(start_block, NULL);
aoqi@0: 
aoqi@0:   if (compilation()->is_profiling()) {
aoqi@0:     ciMethod *method = compilation()->method();
aoqi@0:     if (!method->is_accessor()) {
aoqi@0:       ciMethodData* md = method->method_data_or_null();
aoqi@0:       assert(md != NULL, "Sanity");
aoqi@0:       md->set_compilation_stats(_num_loops, _num_blocks);
aoqi@0:     }
aoqi@0:   }
aoqi@0: 
aoqi@0:   if (_num_loops > 0) {
aoqi@0:     mark_loops();
aoqi@0:     clear_non_natural_loops(start_block);
aoqi@0:     assign_loop_depth(start_block);
aoqi@0:   }
aoqi@0: 
aoqi@0:   compute_order(start_block);
aoqi@0:   compute_dominators();
aoqi@0: 
aoqi@0:   NOT_PRODUCT(print_blocks());
aoqi@0:   DEBUG_ONLY(verify());
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: // Traverse the CFG:
aoqi@0: // * count total number of blocks
aoqi@0: // * count all incoming edges and backward incoming edges
aoqi@0: // * number loop header blocks
aoqi@0: // * create a list with all loop end blocks
aoqi@0: void ComputeLinearScanOrder::count_edges(BlockBegin* cur, BlockBegin* parent) {
aoqi@0:   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));
aoqi@0:   assert(cur->dominator() == NULL, "dominator already initialized");
aoqi@0: 
aoqi@0:   if (is_active(cur)) {
aoqi@0:     TRACE_LINEAR_SCAN(3, tty->print_cr("backward branch"));
aoqi@0:     assert(is_visited(cur), "block must be visisted when block is active");
aoqi@0:     assert(parent != NULL, "must have parent");
aoqi@0: 
aoqi@0:     cur->set(BlockBegin::linear_scan_loop_header_flag);
aoqi@0:     cur->set(BlockBegin::backward_branch_target_flag);
aoqi@0: 
aoqi@0:     parent->set(BlockBegin::linear_scan_loop_end_flag);
aoqi@0: 
aoqi@0:     // When a loop header is also the start of an exception handler, then the backward branch is
aoqi@0:     // an exception edge. Because such edges are usually critical edges which cannot be split, the
aoqi@0:     // loop must be excluded here from processing.
aoqi@0:     if (cur->is_set(BlockBegin::exception_entry_flag)) {
aoqi@0:       // Make sure that dominators are correct in this weird situation
aoqi@0:       _iterative_dominators = true;
aoqi@0:       return;
aoqi@0:     }
aoqi@0:     assert(parent->number_of_sux() == 1 && parent->sux_at(0) == cur,
aoqi@0:            "loop end blocks must have one successor (critical edges are split)");
aoqi@0: 
aoqi@0:     _loop_end_blocks.append(parent);
aoqi@0:     return;
aoqi@0:   }
aoqi@0: 
aoqi@0:   // increment number of incoming forward branches
aoqi@0:   inc_forward_branches(cur);
aoqi@0: 
aoqi@0:   if (is_visited(cur)) {
aoqi@0:     TRACE_LINEAR_SCAN(3, tty->print_cr("block already visited"));
aoqi@0:     return;
aoqi@0:   }
aoqi@0: 
aoqi@0:   _num_blocks++;
aoqi@0:   set_visited(cur);
aoqi@0:   set_active(cur);
aoqi@0: 
aoqi@0:   // recursive call for all successors
aoqi@0:   int i;
aoqi@0:   for (i = cur->number_of_sux() - 1; i >= 0; i--) {
aoqi@0:     count_edges(cur->sux_at(i), cur);
aoqi@0:   }
aoqi@0:   for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
aoqi@0:     count_edges(cur->exception_handler_at(i), cur);
aoqi@0:   }
aoqi@0: 
aoqi@0:   clear_active(cur);
aoqi@0: 
aoqi@0:   // Each loop has a unique number.
aoqi@0:   // When multiple loops are nested, assign_loop_depth assumes that the
aoqi@0:   // innermost loop has the lowest number. This is guaranteed by setting
aoqi@0:   // the loop number after the recursive calls for the successors above
aoqi@0:   // have returned.
aoqi@0:   if (cur->is_set(BlockBegin::linear_scan_loop_header_flag)) {
aoqi@0:     assert(cur->loop_index() == -1, "cannot set loop-index twice");
aoqi@0:     TRACE_LINEAR_SCAN(3, tty->print_cr("Block B%d is loop header of loop %d", cur->block_id(), _num_loops));
aoqi@0: 
aoqi@0:     cur->set_loop_index(_num_loops);
aoqi@0:     _loop_headers.append(cur);
aoqi@0:     _num_loops++;
aoqi@0:   }
aoqi@0: 
aoqi@0:   TRACE_LINEAR_SCAN(3, tty->print_cr("Finished count_edges for block B%d", cur->block_id()));
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: void ComputeLinearScanOrder::mark_loops() {
aoqi@0:   TRACE_LINEAR_SCAN(3, tty->print_cr("----- marking loops"));
aoqi@0: 
aoqi@0:   _loop_map = BitMap2D(_num_loops, _max_block_id);
aoqi@0:   _loop_map.clear();
aoqi@0: 
aoqi@0:   for (int i = _loop_end_blocks.length() - 1; i >= 0; i--) {
aoqi@0:     BlockBegin* loop_end   = _loop_end_blocks.at(i);
aoqi@0:     BlockBegin* loop_start = loop_end->sux_at(0);
aoqi@0:     int         loop_idx   = loop_start->loop_index();
aoqi@0: 
aoqi@0:     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));
aoqi@0:     assert(loop_end->is_set(BlockBegin::linear_scan_loop_end_flag), "loop end flag must be set");
aoqi@0:     assert(loop_end->number_of_sux() == 1, "incorrect number of successors");
aoqi@0:     assert(loop_start->is_set(BlockBegin::linear_scan_loop_header_flag), "loop header flag must be set");
aoqi@0:     assert(loop_idx >= 0 && loop_idx < _num_loops, "loop index not set");
aoqi@0:     assert(_work_list.is_empty(), "work list must be empty before processing");
aoqi@0: 
aoqi@0:     // add the end-block of the loop to the working list
aoqi@0:     _work_list.push(loop_end);
aoqi@0:     set_block_in_loop(loop_idx, loop_end);
aoqi@0:     do {
aoqi@0:       BlockBegin* cur = _work_list.pop();
aoqi@0: 
aoqi@0:       TRACE_LINEAR_SCAN(3, tty->print_cr("    processing B%d", cur->block_id()));
aoqi@0:       assert(is_block_in_loop(loop_idx, cur), "bit in loop map must be set when block is in work list");
aoqi@0: 
aoqi@0:       // recursive processing of all predecessors ends when start block of loop is reached
aoqi@0:       if (cur != loop_start && !cur->is_set(BlockBegin::osr_entry_flag)) {
aoqi@0:         for (int j = cur->number_of_preds() - 1; j >= 0; j--) {
aoqi@0:           BlockBegin* pred = cur->pred_at(j);
aoqi@0: 
aoqi@0:           if (!is_block_in_loop(loop_idx, pred) /*&& !pred->is_set(BlockBeginosr_entry_flag)*/) {
aoqi@0:             // this predecessor has not been processed yet, so add it to work list
aoqi@0:             TRACE_LINEAR_SCAN(3, tty->print_cr("    pushing B%d", pred->block_id()));
aoqi@0:             _work_list.push(pred);
aoqi@0:             set_block_in_loop(loop_idx, pred);
aoqi@0:           }
aoqi@0:         }
aoqi@0:       }
aoqi@0:     } while (!_work_list.is_empty());
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: // check for non-natural loops (loops where the loop header does not dominate
aoqi@0: // all other loop blocks = loops with mulitple entries).
aoqi@0: // such loops are ignored
aoqi@0: void ComputeLinearScanOrder::clear_non_natural_loops(BlockBegin* start_block) {
aoqi@0:   for (int i = _num_loops - 1; i >= 0; i--) {
aoqi@0:     if (is_block_in_loop(i, start_block)) {
aoqi@0:       // loop i contains the entry block of the method
aoqi@0:       // -> this is not a natural loop, so ignore it
aoqi@0:       TRACE_LINEAR_SCAN(2, tty->print_cr("Loop %d is non-natural, so it is ignored", i));
aoqi@0: 
aoqi@0:       BlockBegin *loop_header = _loop_headers.at(i);
aoqi@0:       assert(loop_header->is_set(BlockBegin::linear_scan_loop_header_flag), "Must be loop header");
aoqi@0: 
aoqi@0:       for (int j = 0; j < loop_header->number_of_preds(); j++) {
aoqi@0:         BlockBegin *pred = loop_header->pred_at(j);
aoqi@0:         pred->clear(BlockBegin::linear_scan_loop_end_flag);
aoqi@0:       }
aoqi@0: 
aoqi@0:       loop_header->clear(BlockBegin::linear_scan_loop_header_flag);
aoqi@0: 
aoqi@0:       for (int block_id = _max_block_id - 1; block_id >= 0; block_id--) {
aoqi@0:         clear_block_in_loop(i, block_id);
aoqi@0:       }
aoqi@0:       _iterative_dominators = true;
aoqi@0:     }
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: void ComputeLinearScanOrder::assign_loop_depth(BlockBegin* start_block) {
aoqi@0:   TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing loop-depth and weight"));
aoqi@0:   init_visited();
aoqi@0: 
aoqi@0:   assert(_work_list.is_empty(), "work list must be empty before processing");
aoqi@0:   _work_list.append(start_block);
aoqi@0: 
aoqi@0:   do {
aoqi@0:     BlockBegin* cur = _work_list.pop();
aoqi@0: 
aoqi@0:     if (!is_visited(cur)) {
aoqi@0:       set_visited(cur);
aoqi@0:       TRACE_LINEAR_SCAN(4, tty->print_cr("Computing loop depth for block B%d", cur->block_id()));
aoqi@0: 
aoqi@0:       // compute loop-depth and loop-index for the block
aoqi@0:       assert(cur->loop_depth() == 0, "cannot set loop-depth twice");
aoqi@0:       int i;
aoqi@0:       int loop_depth = 0;
aoqi@0:       int min_loop_idx = -1;
aoqi@0:       for (i = _num_loops - 1; i >= 0; i--) {
aoqi@0:         if (is_block_in_loop(i, cur)) {
aoqi@0:           loop_depth++;
aoqi@0:           min_loop_idx = i;
aoqi@0:         }
aoqi@0:       }
aoqi@0:       cur->set_loop_depth(loop_depth);
aoqi@0:       cur->set_loop_index(min_loop_idx);
aoqi@0: 
aoqi@0:       // append all unvisited successors to work list
aoqi@0:       for (i = cur->number_of_sux() - 1; i >= 0; i--) {
aoqi@0:         _work_list.append(cur->sux_at(i));
aoqi@0:       }
aoqi@0:       for (i = cur->number_of_exception_handlers() - 1; i >= 0; i--) {
aoqi@0:         _work_list.append(cur->exception_handler_at(i));
aoqi@0:       }
aoqi@0:     }
aoqi@0:   } while (!_work_list.is_empty());
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: BlockBegin* ComputeLinearScanOrder::common_dominator(BlockBegin* a, BlockBegin* b) {
aoqi@0:   assert(a != NULL && b != NULL, "must have input blocks");
aoqi@0: 
aoqi@0:   _dominator_blocks.clear();
aoqi@0:   while (a != NULL) {
aoqi@0:     _dominator_blocks.set_bit(a->block_id());
aoqi@0:     assert(a->dominator() != NULL || a == _linear_scan_order->at(0), "dominator must be initialized");
aoqi@0:     a = a->dominator();
aoqi@0:   }
aoqi@0:   while (b != NULL && !_dominator_blocks.at(b->block_id())) {
aoqi@0:     assert(b->dominator() != NULL || b == _linear_scan_order->at(0), "dominator must be initialized");
aoqi@0:     b = b->dominator();
aoqi@0:   }
aoqi@0: 
aoqi@0:   assert(b != NULL, "could not find dominator");
aoqi@0:   return b;
aoqi@0: }
aoqi@0: 
aoqi@0: void ComputeLinearScanOrder::compute_dominator(BlockBegin* cur, BlockBegin* parent) {
aoqi@0:   if (cur->dominator() == NULL) {
aoqi@0:     TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: initializing dominator of B%d to B%d", cur->block_id(), parent->block_id()));
aoqi@0:     cur->set_dominator(parent);
aoqi@0: 
aoqi@0:   } else if (!(cur->is_set(BlockBegin::linear_scan_loop_header_flag) && parent->is_set(BlockBegin::linear_scan_loop_end_flag))) {
aoqi@0:     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()));
aoqi@0:     // Does not hold for exception blocks
aoqi@0:     assert(cur->number_of_preds() > 1 || cur->is_set(BlockBegin::exception_entry_flag), "");
aoqi@0:     cur->set_dominator(common_dominator(cur->dominator(), parent));
aoqi@0:   }
aoqi@0: 
aoqi@0:   // Additional edge to xhandler of all our successors
aoqi@0:   // range check elimination needs that the state at the end of a
aoqi@0:   // block be valid in every block it dominates so cur must dominate
aoqi@0:   // the exception handlers of its successors.
aoqi@0:   int num_cur_xhandler = cur->number_of_exception_handlers();
aoqi@0:   for (int j = 0; j < num_cur_xhandler; j++) {
aoqi@0:     BlockBegin* xhandler = cur->exception_handler_at(j);
aoqi@0:     compute_dominator(xhandler, parent);
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: int ComputeLinearScanOrder::compute_weight(BlockBegin* cur) {
aoqi@0:   BlockBegin* single_sux = NULL;
aoqi@0:   if (cur->number_of_sux() == 1) {
aoqi@0:     single_sux = cur->sux_at(0);
aoqi@0:   }
aoqi@0: 
aoqi@0:   // limit loop-depth to 15 bit (only for security reason, it will never be so big)
aoqi@0:   int weight = (cur->loop_depth() & 0x7FFF) << 16;
aoqi@0: 
aoqi@0:   // general macro for short definition of weight flags
aoqi@0:   // the first instance of INC_WEIGHT_IF has the highest priority
aoqi@0:   int cur_bit = 15;
aoqi@0:   #define INC_WEIGHT_IF(condition) if ((condition)) { weight |= (1 << cur_bit); } cur_bit--;
aoqi@0: 
aoqi@0:   // this is necessery for the (very rare) case that two successing blocks have
aoqi@0:   // the same loop depth, but a different loop index (can happen for endless loops
aoqi@0:   // with exception handlers)
aoqi@0:   INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_header_flag));
aoqi@0: 
aoqi@0:   // loop end blocks (blocks that end with a backward branch) are added
aoqi@0:   // after all other blocks of the loop.
aoqi@0:   INC_WEIGHT_IF(!cur->is_set(BlockBegin::linear_scan_loop_end_flag));
aoqi@0: 
aoqi@0:   // critical edge split blocks are prefered because than they have a bigger
aoqi@0:   // proability to be completely empty
aoqi@0:   INC_WEIGHT_IF(cur->is_set(BlockBegin::critical_edge_split_flag));
aoqi@0: 
aoqi@0:   // exceptions should not be thrown in normal control flow, so these blocks
aoqi@0:   // are added as late as possible
aoqi@0:   INC_WEIGHT_IF(cur->end()->as_Throw() == NULL  && (single_sux == NULL || single_sux->end()->as_Throw()  == NULL));
aoqi@0:   INC_WEIGHT_IF(cur->end()->as_Return() == NULL && (single_sux == NULL || single_sux->end()->as_Return() == NULL));
aoqi@0: 
aoqi@0:   // exceptions handlers are added as late as possible
aoqi@0:   INC_WEIGHT_IF(!cur->is_set(BlockBegin::exception_entry_flag));
aoqi@0: 
aoqi@0:   // guarantee that weight is > 0
aoqi@0:   weight |= 1;
aoqi@0: 
aoqi@0:   #undef INC_WEIGHT_IF
aoqi@0:   assert(cur_bit >= 0, "too many flags");
aoqi@0:   assert(weight > 0, "weight cannot become negative");
aoqi@0: 
aoqi@0:   return weight;
aoqi@0: }
aoqi@0: 
aoqi@0: bool ComputeLinearScanOrder::ready_for_processing(BlockBegin* cur) {
aoqi@0:   // Discount the edge just traveled.
aoqi@0:   // When the number drops to zero, all forward branches were processed
aoqi@0:   if (dec_forward_branches(cur) != 0) {
aoqi@0:     return false;
aoqi@0:   }
aoqi@0: 
aoqi@0:   assert(_linear_scan_order->index_of(cur) == -1, "block already processed (block can be ready only once)");
aoqi@0:   assert(_work_list.index_of(cur) == -1, "block already in work-list (block can be ready only once)");
aoqi@0:   return true;
aoqi@0: }
aoqi@0: 
aoqi@0: void ComputeLinearScanOrder::sort_into_work_list(BlockBegin* cur) {
aoqi@0:   assert(_work_list.index_of(cur) == -1, "block already in work list");
aoqi@0: 
aoqi@0:   int cur_weight = compute_weight(cur);
aoqi@0: 
aoqi@0:   // the linear_scan_number is used to cache the weight of a block
aoqi@0:   cur->set_linear_scan_number(cur_weight);
aoqi@0: 
aoqi@0: #ifndef PRODUCT
aoqi@0:   if (StressLinearScan) {
aoqi@0:     _work_list.insert_before(0, cur);
aoqi@0:     return;
aoqi@0:   }
aoqi@0: #endif
aoqi@0: 
aoqi@0:   _work_list.append(NULL); // provide space for new element
aoqi@0: 
aoqi@0:   int insert_idx = _work_list.length() - 1;
aoqi@0:   while (insert_idx > 0 && _work_list.at(insert_idx - 1)->linear_scan_number() > cur_weight) {
aoqi@0:     _work_list.at_put(insert_idx, _work_list.at(insert_idx - 1));
aoqi@0:     insert_idx--;
aoqi@0:   }
aoqi@0:   _work_list.at_put(insert_idx, cur);
aoqi@0: 
aoqi@0:   TRACE_LINEAR_SCAN(3, tty->print_cr("Sorted B%d into worklist. new worklist:", cur->block_id()));
aoqi@0:   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()));
aoqi@0: 
aoqi@0: #ifdef ASSERT
aoqi@0:   for (int i = 0; i < _work_list.length(); i++) {
aoqi@0:     assert(_work_list.at(i)->linear_scan_number() > 0, "weight not set");
aoqi@0:     assert(i == 0 || _work_list.at(i - 1)->linear_scan_number() <= _work_list.at(i)->linear_scan_number(), "incorrect order in worklist");
aoqi@0:   }
aoqi@0: #endif
aoqi@0: }
aoqi@0: 
aoqi@0: void ComputeLinearScanOrder::append_block(BlockBegin* cur) {
aoqi@0:   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()));
aoqi@0:   assert(_linear_scan_order->index_of(cur) == -1, "cannot add the same block twice");
aoqi@0: 
aoqi@0:   // currently, the linear scan order and code emit order are equal.
aoqi@0:   // therefore the linear_scan_number and the weight of a block must also
aoqi@0:   // be equal.
aoqi@0:   cur->set_linear_scan_number(_linear_scan_order->length());
aoqi@0:   _linear_scan_order->append(cur);
aoqi@0: }
aoqi@0: 
aoqi@0: void ComputeLinearScanOrder::compute_order(BlockBegin* start_block) {
aoqi@0:   TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing final block order"));
aoqi@0: 
aoqi@0:   // the start block is always the first block in the linear scan order
aoqi@0:   _linear_scan_order = new BlockList(_num_blocks);
aoqi@0:   append_block(start_block);
aoqi@0: 
aoqi@0:   assert(start_block->end()->as_Base() != NULL, "start block must end with Base-instruction");
aoqi@0:   BlockBegin* std_entry = ((Base*)start_block->end())->std_entry();
aoqi@0:   BlockBegin* osr_entry = ((Base*)start_block->end())->osr_entry();
aoqi@0: 
aoqi@0:   BlockBegin* sux_of_osr_entry = NULL;
aoqi@0:   if (osr_entry != NULL) {
aoqi@0:     // special handling for osr entry:
aoqi@0:     // ignore the edge between the osr entry and its successor for processing
aoqi@0:     // the osr entry block is added manually below
aoqi@0:     assert(osr_entry->number_of_sux() == 1, "osr entry must have exactly one successor");
aoqi@0:     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");
aoqi@0: 
aoqi@0:     sux_of_osr_entry = osr_entry->sux_at(0);
aoqi@0:     dec_forward_branches(sux_of_osr_entry);
aoqi@0: 
aoqi@0:     compute_dominator(osr_entry, start_block);
aoqi@0:     _iterative_dominators = true;
aoqi@0:   }
aoqi@0:   compute_dominator(std_entry, start_block);
aoqi@0: 
aoqi@0:   // start processing with standard entry block
aoqi@0:   assert(_work_list.is_empty(), "list must be empty before processing");
aoqi@0: 
aoqi@0:   if (ready_for_processing(std_entry)) {
aoqi@0:     sort_into_work_list(std_entry);
aoqi@0:   } else {
aoqi@0:     assert(false, "the std_entry must be ready for processing (otherwise, the method has no start block)");
aoqi@0:   }
aoqi@0: 
aoqi@0:   do {
aoqi@0:     BlockBegin* cur = _work_list.pop();
aoqi@0: 
aoqi@0:     if (cur == sux_of_osr_entry) {
aoqi@0:       // the osr entry block is ignored in normal processing, it is never added to the
aoqi@0:       // work list. Instead, it is added as late as possible manually here.
aoqi@0:       append_block(osr_entry);
aoqi@0:       compute_dominator(cur, osr_entry);
aoqi@0:     }
aoqi@0:     append_block(cur);
aoqi@0: 
aoqi@0:     int i;
aoqi@0:     int num_sux = cur->number_of_sux();
aoqi@0:     // changed loop order to get "intuitive" order of if- and else-blocks
aoqi@0:     for (i = 0; i < num_sux; i++) {
aoqi@0:       BlockBegin* sux = cur->sux_at(i);
aoqi@0:       compute_dominator(sux, cur);
aoqi@0:       if (ready_for_processing(sux)) {
aoqi@0:         sort_into_work_list(sux);
aoqi@0:       }
aoqi@0:     }
aoqi@0:     num_sux = cur->number_of_exception_handlers();
aoqi@0:     for (i = 0; i < num_sux; i++) {
aoqi@0:       BlockBegin* sux = cur->exception_handler_at(i);
aoqi@0:       if (ready_for_processing(sux)) {
aoqi@0:         sort_into_work_list(sux);
aoqi@0:       }
aoqi@0:     }
aoqi@0:   } while (_work_list.length() > 0);
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: bool ComputeLinearScanOrder::compute_dominators_iter() {
aoqi@0:   bool changed = false;
aoqi@0:   int num_blocks = _linear_scan_order->length();
aoqi@0: 
aoqi@0:   assert(_linear_scan_order->at(0)->dominator() == NULL, "must not have dominator");
aoqi@0:   assert(_linear_scan_order->at(0)->number_of_preds() == 0, "must not have predecessors");
aoqi@0:   for (int i = 1; i < num_blocks; i++) {
aoqi@0:     BlockBegin* block = _linear_scan_order->at(i);
aoqi@0: 
aoqi@0:     BlockBegin* dominator = block->pred_at(0);
aoqi@0:     int num_preds = block->number_of_preds();
aoqi@0: 
aoqi@0:     TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: Processing B%d", block->block_id()));
aoqi@0: 
aoqi@0:     for (int j = 0; j < num_preds; j++) {
aoqi@0: 
aoqi@0:       BlockBegin *pred = block->pred_at(j);
aoqi@0:       TRACE_LINEAR_SCAN(4, tty->print_cr("   DOM: Subrocessing B%d", pred->block_id()));
aoqi@0: 
aoqi@0:       if (block->is_set(BlockBegin::exception_entry_flag)) {
aoqi@0:         dominator = common_dominator(dominator, pred);
aoqi@0:         int num_pred_preds = pred->number_of_preds();
aoqi@0:         for (int k = 0; k < num_pred_preds; k++) {
aoqi@0:           dominator = common_dominator(dominator, pred->pred_at(k));
aoqi@0:         }
aoqi@0:       } else {
aoqi@0:         dominator = common_dominator(dominator, pred);
aoqi@0:       }
aoqi@0:     }
aoqi@0: 
aoqi@0:     if (dominator != block->dominator()) {
aoqi@0:       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()));
aoqi@0: 
aoqi@0:       block->set_dominator(dominator);
aoqi@0:       changed = true;
aoqi@0:     }
aoqi@0:   }
aoqi@0:   return changed;
aoqi@0: }
aoqi@0: 
aoqi@0: void ComputeLinearScanOrder::compute_dominators() {
aoqi@0:   TRACE_LINEAR_SCAN(3, tty->print_cr("----- computing dominators (iterative computation reqired: %d)", _iterative_dominators));
aoqi@0: 
aoqi@0:   // iterative computation of dominators is only required for methods with non-natural loops
aoqi@0:   // and OSR-methods. For all other methods, the dominators computed when generating the
aoqi@0:   // linear scan block order are correct.
aoqi@0:   if (_iterative_dominators) {
aoqi@0:     do {
aoqi@0:       TRACE_LINEAR_SCAN(1, tty->print_cr("DOM: next iteration of fix-point calculation"));
aoqi@0:     } while (compute_dominators_iter());
aoqi@0:   }
aoqi@0: 
aoqi@0:   // check that dominators are correct
aoqi@0:   assert(!compute_dominators_iter(), "fix point not reached");
aoqi@0: 
aoqi@0:   // Add Blocks to dominates-Array
aoqi@0:   int num_blocks = _linear_scan_order->length();
aoqi@0:   for (int i = 0; i < num_blocks; i++) {
aoqi@0:     BlockBegin* block = _linear_scan_order->at(i);
aoqi@0: 
aoqi@0:     BlockBegin *dom = block->dominator();
aoqi@0:     if (dom) {
aoqi@0:       assert(dom->dominator_depth() != -1, "Dominator must have been visited before");
aoqi@0:       dom->dominates()->append(block);
aoqi@0:       block->set_dominator_depth(dom->dominator_depth() + 1);
aoqi@0:     } else {
aoqi@0:       block->set_dominator_depth(0);
aoqi@0:     }
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: #ifndef PRODUCT
aoqi@0: void ComputeLinearScanOrder::print_blocks() {
aoqi@0:   if (TraceLinearScanLevel >= 2) {
aoqi@0:     tty->print_cr("----- loop information:");
aoqi@0:     for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
aoqi@0:       BlockBegin* cur = _linear_scan_order->at(block_idx);
aoqi@0: 
aoqi@0:       tty->print("%4d: B%2d: ", cur->linear_scan_number(), cur->block_id());
aoqi@0:       for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
aoqi@0:         tty->print ("%d ", is_block_in_loop(loop_idx, cur));
aoqi@0:       }
aoqi@0:       tty->print_cr(" -> loop_index: %2d, loop_depth: %2d", cur->loop_index(), cur->loop_depth());
aoqi@0:     }
aoqi@0:   }
aoqi@0: 
aoqi@0:   if (TraceLinearScanLevel >= 1) {
aoqi@0:     tty->print_cr("----- linear-scan block order:");
aoqi@0:     for (int block_idx = 0; block_idx < _linear_scan_order->length(); block_idx++) {
aoqi@0:       BlockBegin* cur = _linear_scan_order->at(block_idx);
aoqi@0:       tty->print("%4d: B%2d    loop: %2d  depth: %2d", cur->linear_scan_number(), cur->block_id(), cur->loop_index(), cur->loop_depth());
aoqi@0: 
aoqi@0:       tty->print(cur->is_set(BlockBegin::exception_entry_flag)         ? " ex" : "   ");
aoqi@0:       tty->print(cur->is_set(BlockBegin::critical_edge_split_flag)     ? " ce" : "   ");
aoqi@0:       tty->print(cur->is_set(BlockBegin::linear_scan_loop_header_flag) ? " lh" : "   ");
aoqi@0:       tty->print(cur->is_set(BlockBegin::linear_scan_loop_end_flag)    ? " le" : "   ");
aoqi@0: 
aoqi@0:       if (cur->dominator() != NULL) {
aoqi@0:         tty->print("    dom: B%d ", cur->dominator()->block_id());
aoqi@0:       } else {
aoqi@0:         tty->print("    dom: NULL ");
aoqi@0:       }
aoqi@0: 
aoqi@0:       if (cur->number_of_preds() > 0) {
aoqi@0:         tty->print("    preds: ");
aoqi@0:         for (int j = 0; j < cur->number_of_preds(); j++) {
aoqi@0:           BlockBegin* pred = cur->pred_at(j);
aoqi@0:           tty->print("B%d ", pred->block_id());
aoqi@0:         }
aoqi@0:       }
aoqi@0:       if (cur->number_of_sux() > 0) {
aoqi@0:         tty->print("    sux: ");
aoqi@0:         for (int j = 0; j < cur->number_of_sux(); j++) {
aoqi@0:           BlockBegin* sux = cur->sux_at(j);
aoqi@0:           tty->print("B%d ", sux->block_id());
aoqi@0:         }
aoqi@0:       }
aoqi@0:       if (cur->number_of_exception_handlers() > 0) {
aoqi@0:         tty->print("    ex: ");
aoqi@0:         for (int j = 0; j < cur->number_of_exception_handlers(); j++) {
aoqi@0:           BlockBegin* ex = cur->exception_handler_at(j);
aoqi@0:           tty->print("B%d ", ex->block_id());
aoqi@0:         }
aoqi@0:       }
aoqi@0:       tty->cr();
aoqi@0:     }
aoqi@0:   }
aoqi@0: }
aoqi@0: #endif
aoqi@0: 
aoqi@0: #ifdef ASSERT
aoqi@0: void ComputeLinearScanOrder::verify() {
aoqi@0:   assert(_linear_scan_order->length() == _num_blocks, "wrong number of blocks in list");
aoqi@0: 
aoqi@0:   if (StressLinearScan) {
aoqi@0:     // blocks are scrambled when StressLinearScan is used
aoqi@0:     return;
aoqi@0:   }
aoqi@0: 
aoqi@0:   // check that all successors of a block have a higher linear-scan-number
aoqi@0:   // and that all predecessors of a block have a lower linear-scan-number
aoqi@0:   // (only backward branches of loops are ignored)
aoqi@0:   int i;
aoqi@0:   for (i = 0; i < _linear_scan_order->length(); i++) {
aoqi@0:     BlockBegin* cur = _linear_scan_order->at(i);
aoqi@0: 
aoqi@0:     assert(cur->linear_scan_number() == i, "incorrect linear_scan_number");
aoqi@0:     assert(cur->linear_scan_number() >= 0 && cur->linear_scan_number() == _linear_scan_order->index_of(cur), "incorrect linear_scan_number");
aoqi@0: 
aoqi@0:     int j;
aoqi@0:     for (j = cur->number_of_sux() - 1; j >= 0; j--) {
aoqi@0:       BlockBegin* sux = cur->sux_at(j);
aoqi@0: 
aoqi@0:       assert(sux->linear_scan_number() >= 0 && sux->linear_scan_number() == _linear_scan_order->index_of(sux), "incorrect linear_scan_number");
aoqi@0:       if (!sux->is_set(BlockBegin::backward_branch_target_flag)) {
aoqi@0:         assert(cur->linear_scan_number() < sux->linear_scan_number(), "invalid order");
aoqi@0:       }
aoqi@0:       if (cur->loop_depth() == sux->loop_depth()) {
aoqi@0:         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");
aoqi@0:       }
aoqi@0:     }
aoqi@0: 
aoqi@0:     for (j = cur->number_of_preds() - 1; j >= 0; j--) {
aoqi@0:       BlockBegin* pred = cur->pred_at(j);
aoqi@0: 
aoqi@0:       assert(pred->linear_scan_number() >= 0 && pred->linear_scan_number() == _linear_scan_order->index_of(pred), "incorrect linear_scan_number");
aoqi@0:       if (!cur->is_set(BlockBegin::backward_branch_target_flag)) {
aoqi@0:         assert(cur->linear_scan_number() > pred->linear_scan_number(), "invalid order");
aoqi@0:       }
aoqi@0:       if (cur->loop_depth() == pred->loop_depth()) {
aoqi@0:         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");
aoqi@0:       }
aoqi@0: 
aoqi@0:       assert(cur->dominator()->linear_scan_number() <= cur->pred_at(j)->linear_scan_number(), "dominator must be before predecessors");
aoqi@0:     }
aoqi@0: 
aoqi@0:     // check dominator
aoqi@0:     if (i == 0) {
aoqi@0:       assert(cur->dominator() == NULL, "first block has no dominator");
aoqi@0:     } else {
aoqi@0:       assert(cur->dominator() != NULL, "all but first block must have dominator");
aoqi@0:     }
aoqi@0:     // Assertion does not hold for exception handlers
aoqi@0:     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");
aoqi@0:   }
aoqi@0: 
aoqi@0:   // check that all loops are continuous
aoqi@0:   for (int loop_idx = 0; loop_idx < _num_loops; loop_idx++) {
aoqi@0:     int block_idx = 0;
aoqi@0:     assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "the first block must not be present in any loop");
aoqi@0: 
aoqi@0:     // skip blocks before the loop
aoqi@0:     while (block_idx < _num_blocks && !is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
aoqi@0:       block_idx++;
aoqi@0:     }
aoqi@0:     // skip blocks of loop
aoqi@0:     while (block_idx < _num_blocks && is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx))) {
aoqi@0:       block_idx++;
aoqi@0:     }
aoqi@0:     // after the first non-loop block, there must not be another loop-block
aoqi@0:     while (block_idx < _num_blocks) {
aoqi@0:       assert(!is_block_in_loop(loop_idx, _linear_scan_order->at(block_idx)), "loop not continuous in linear-scan order");
aoqi@0:       block_idx++;
aoqi@0:     }
aoqi@0:   }
aoqi@0: }
aoqi@0: #endif
aoqi@0: 
aoqi@0: 
aoqi@0: void IR::compute_code() {
aoqi@0:   assert(is_valid(), "IR must be valid");
aoqi@0: 
aoqi@0:   ComputeLinearScanOrder compute_order(compilation(), start());
aoqi@0:   _num_loops = compute_order.num_loops();
aoqi@0:   _code = compute_order.linear_scan_order();
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: void IR::compute_use_counts() {
aoqi@0:   // make sure all values coming out of this block get evaluated.
aoqi@0:   int num_blocks = _code->length();
aoqi@0:   for (int i = 0; i < num_blocks; i++) {
aoqi@0:     _code->at(i)->end()->state()->pin_stack_for_linear_scan();
aoqi@0:   }
aoqi@0: 
aoqi@0:   // compute use counts
aoqi@0:   UseCountComputer::compute(_code);
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: void IR::iterate_preorder(BlockClosure* closure) {
aoqi@0:   assert(is_valid(), "IR must be valid");
aoqi@0:   start()->iterate_preorder(closure);
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: void IR::iterate_postorder(BlockClosure* closure) {
aoqi@0:   assert(is_valid(), "IR must be valid");
aoqi@0:   start()->iterate_postorder(closure);
aoqi@0: }
aoqi@0: 
aoqi@0: void IR::iterate_linear_scan_order(BlockClosure* closure) {
aoqi@0:   linear_scan_order()->iterate_forward(closure);
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: #ifndef PRODUCT
aoqi@0: class BlockPrinter: public BlockClosure {
aoqi@0:  private:
aoqi@0:   InstructionPrinter* _ip;
aoqi@0:   bool                _cfg_only;
aoqi@0:   bool                _live_only;
aoqi@0: 
aoqi@0:  public:
aoqi@0:   BlockPrinter(InstructionPrinter* ip, bool cfg_only, bool live_only = false) {
aoqi@0:     _ip       = ip;
aoqi@0:     _cfg_only = cfg_only;
aoqi@0:     _live_only = live_only;
aoqi@0:   }
aoqi@0: 
aoqi@0:   virtual void block_do(BlockBegin* block) {
aoqi@0:     if (_cfg_only) {
aoqi@0:       _ip->print_instr(block); tty->cr();
aoqi@0:     } else {
aoqi@0:       block->print_block(*_ip, _live_only);
aoqi@0:     }
aoqi@0:   }
aoqi@0: };
aoqi@0: 
aoqi@0: 
aoqi@0: void IR::print(BlockBegin* start, bool cfg_only, bool live_only) {
aoqi@0:   ttyLocker ttyl;
aoqi@0:   InstructionPrinter ip(!cfg_only);
aoqi@0:   BlockPrinter bp(&ip, cfg_only, live_only);
aoqi@0:   start->iterate_preorder(&bp);
aoqi@0:   tty->cr();
aoqi@0: }
aoqi@0: 
aoqi@0: void IR::print(bool cfg_only, bool live_only) {
aoqi@0:   if (is_valid()) {
aoqi@0:     print(start(), cfg_only, live_only);
aoqi@0:   } else {
aoqi@0:     tty->print_cr("invalid IR");
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: 
aoqi@0: define_array(BlockListArray, BlockList*)
aoqi@0: define_stack(BlockListList, BlockListArray)
aoqi@0: 
aoqi@0: class PredecessorValidator : public BlockClosure {
aoqi@0:  private:
aoqi@0:   BlockListList* _predecessors;
aoqi@0:   BlockList*     _blocks;
aoqi@0: 
aoqi@0:   static int cmp(BlockBegin** a, BlockBegin** b) {
aoqi@0:     return (*a)->block_id() - (*b)->block_id();
aoqi@0:   }
aoqi@0: 
aoqi@0:  public:
aoqi@0:   PredecessorValidator(IR* hir) {
aoqi@0:     ResourceMark rm;
aoqi@0:     _predecessors = new BlockListList(BlockBegin::number_of_blocks(), NULL);
aoqi@0:     _blocks = new BlockList();
aoqi@0: 
aoqi@0:     int i;
aoqi@0:     hir->start()->iterate_preorder(this);
aoqi@0:     if (hir->code() != NULL) {
aoqi@0:       assert(hir->code()->length() == _blocks->length(), "must match");
aoqi@0:       for (i = 0; i < _blocks->length(); i++) {
aoqi@0:         assert(hir->code()->contains(_blocks->at(i)), "should be in both lists");
aoqi@0:       }
aoqi@0:     }
aoqi@0: 
aoqi@0:     for (i = 0; i < _blocks->length(); i++) {
aoqi@0:       BlockBegin* block = _blocks->at(i);
aoqi@0:       BlockList* preds = _predecessors->at(block->block_id());
aoqi@0:       if (preds == NULL) {
aoqi@0:         assert(block->number_of_preds() == 0, "should be the same");
aoqi@0:         continue;
aoqi@0:       }
aoqi@0: 
aoqi@0:       // clone the pred list so we can mutate it
aoqi@0:       BlockList* pred_copy = new BlockList();
aoqi@0:       int j;
aoqi@0:       for (j = 0; j < block->number_of_preds(); j++) {
aoqi@0:         pred_copy->append(block->pred_at(j));
aoqi@0:       }
aoqi@0:       // sort them in the same order
aoqi@0:       preds->sort(cmp);
aoqi@0:       pred_copy->sort(cmp);
aoqi@0:       int length = MIN2(preds->length(), block->number_of_preds());
aoqi@0:       for (j = 0; j < block->number_of_preds(); j++) {
aoqi@0:         assert(preds->at(j) == pred_copy->at(j), "must match");
aoqi@0:       }
aoqi@0: 
aoqi@0:       assert(preds->length() == block->number_of_preds(), "should be the same");
aoqi@0:     }
aoqi@0:   }
aoqi@0: 
aoqi@0:   virtual void block_do(BlockBegin* block) {
aoqi@0:     _blocks->append(block);
aoqi@0:     BlockEnd* be = block->end();
aoqi@0:     int n = be->number_of_sux();
aoqi@0:     int i;
aoqi@0:     for (i = 0; i < n; i++) {
aoqi@0:       BlockBegin* sux = be->sux_at(i);
aoqi@0:       assert(!sux->is_set(BlockBegin::exception_entry_flag), "must not be xhandler");
aoqi@0: 
aoqi@0:       BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
aoqi@0:       if (preds == NULL) {
aoqi@0:         preds = new BlockList();
aoqi@0:         _predecessors->at_put(sux->block_id(), preds);
aoqi@0:       }
aoqi@0:       preds->append(block);
aoqi@0:     }
aoqi@0: 
aoqi@0:     n = block->number_of_exception_handlers();
aoqi@0:     for (i = 0; i < n; i++) {
aoqi@0:       BlockBegin* sux = block->exception_handler_at(i);
aoqi@0:       assert(sux->is_set(BlockBegin::exception_entry_flag), "must be xhandler");
aoqi@0: 
aoqi@0:       BlockList* preds = _predecessors->at_grow(sux->block_id(), NULL);
aoqi@0:       if (preds == NULL) {
aoqi@0:         preds = new BlockList();
aoqi@0:         _predecessors->at_put(sux->block_id(), preds);
aoqi@0:       }
aoqi@0:       preds->append(block);
aoqi@0:     }
aoqi@0:   }
aoqi@0: };
aoqi@0: 
aoqi@0: class VerifyBlockBeginField : public BlockClosure {
aoqi@0: 
aoqi@0: public:
aoqi@0: 
aoqi@0:   virtual void block_do(BlockBegin *block) {
aoqi@0:     for ( Instruction *cur = block; cur != NULL; cur = cur->next()) {
aoqi@0:       assert(cur->block() == block, "Block begin is not correct");
aoqi@0:     }
aoqi@0:   }
aoqi@0: };
aoqi@0: 
aoqi@0: void IR::verify() {
aoqi@0: #ifdef ASSERT
aoqi@0:   PredecessorValidator pv(this);
aoqi@0:   VerifyBlockBeginField verifier;
aoqi@0:   this->iterate_postorder(&verifier);
aoqi@0: #endif
aoqi@0: }
aoqi@0: 
aoqi@0: #endif // PRODUCT
aoqi@0: 
aoqi@0: void SubstitutionResolver::visit(Value* v) {
aoqi@0:   Value v0 = *v;
aoqi@0:   if (v0) {
aoqi@0:     Value vs = v0->subst();
aoqi@0:     if (vs != v0) {
aoqi@0:       *v = v0->subst();
aoqi@0:     }
aoqi@0:   }
aoqi@0: }
aoqi@0: 
aoqi@0: #ifdef ASSERT
aoqi@0: class SubstitutionChecker: public ValueVisitor {
aoqi@0:   void visit(Value* v) {
aoqi@0:     Value v0 = *v;
aoqi@0:     if (v0) {
aoqi@0:       Value vs = v0->subst();
aoqi@0:       assert(vs == v0, "missed substitution");
aoqi@0:     }
aoqi@0:   }
aoqi@0: };
aoqi@0: #endif
aoqi@0: 
aoqi@0: 
aoqi@0: void SubstitutionResolver::block_do(BlockBegin* block) {
aoqi@0:   Instruction* last = NULL;
aoqi@0:   for (Instruction* n = block; n != NULL;) {
aoqi@0:     n->values_do(this);
aoqi@0:     // need to remove this instruction from the instruction stream
aoqi@0:     if (n->subst() != n) {
aoqi@0:       assert(last != NULL, "must have last");
aoqi@0:       last->set_next(n->next());
aoqi@0:     } else {
aoqi@0:       last = n;
aoqi@0:     }
aoqi@0:     n = last->next();
aoqi@0:   }
aoqi@0: 
aoqi@0: #ifdef ASSERT
aoqi@0:   SubstitutionChecker check_substitute;
aoqi@0:   if (block->state()) block->state()->values_do(&check_substitute);
aoqi@0:   block->block_values_do(&check_substitute);
aoqi@0:   if (block->end() && block->end()->state()) block->end()->state()->values_do(&check_substitute);
aoqi@0: #endif
aoqi@0: }