src/share/vm/opto/macro.cpp

Tue, 15 Apr 2008 10:49:32 -0700

author
kvn
date
Tue, 15 Apr 2008 10:49:32 -0700
changeset 520
f3b3fe64f59f
parent 508
a8880a78d355
child 548
ba764ed4b6f2
permissions
-rw-r--r--

6692301: Side effect in NumberFormat tests with -server -Xcomp
Summary: Optimization in CmpPNode::sub() removed the valid compare instruction because of false positive answer from detect_dominating_control().
Reviewed-by: jrose, sgoldman

     1 /*
     2  * Copyright 2005-2007 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
    20  * CA 95054 USA or visit www.sun.com if you need additional information or
    21  * have any questions.
    22  *
    23  */
    25 #include "incls/_precompiled.incl"
    26 #include "incls/_macro.cpp.incl"
    29 //
    30 // Replace any references to "oldref" in inputs to "use" with "newref".
    31 // Returns the number of replacements made.
    32 //
    33 int PhaseMacroExpand::replace_input(Node *use, Node *oldref, Node *newref) {
    34   int nreplacements = 0;
    35   uint req = use->req();
    36   for (uint j = 0; j < use->len(); j++) {
    37     Node *uin = use->in(j);
    38     if (uin == oldref) {
    39       if (j < req)
    40         use->set_req(j, newref);
    41       else
    42         use->set_prec(j, newref);
    43       nreplacements++;
    44     } else if (j >= req && uin == NULL) {
    45       break;
    46     }
    47   }
    48   return nreplacements;
    49 }
    51 void PhaseMacroExpand::copy_call_debug_info(CallNode *oldcall, CallNode * newcall) {
    52   // Copy debug information and adjust JVMState information
    53   uint old_dbg_start = oldcall->tf()->domain()->cnt();
    54   uint new_dbg_start = newcall->tf()->domain()->cnt();
    55   int jvms_adj  = new_dbg_start - old_dbg_start;
    56   assert (new_dbg_start == newcall->req(), "argument count mismatch");
    58   Dict* sosn_map = new Dict(cmpkey,hashkey);
    59   for (uint i = old_dbg_start; i < oldcall->req(); i++) {
    60     Node* old_in = oldcall->in(i);
    61     // Clone old SafePointScalarObjectNodes, adjusting their field contents.
    62     if (old_in->is_SafePointScalarObject()) {
    63       SafePointScalarObjectNode* old_sosn = old_in->as_SafePointScalarObject();
    64       uint old_unique = C->unique();
    65       Node* new_in = old_sosn->clone(jvms_adj, sosn_map);
    66       if (old_unique != C->unique()) {
    67         new_in = transform_later(new_in); // Register new node.
    68       }
    69       old_in = new_in;
    70     }
    71     newcall->add_req(old_in);
    72   }
    74   newcall->set_jvms(oldcall->jvms());
    75   for (JVMState *jvms = newcall->jvms(); jvms != NULL; jvms = jvms->caller()) {
    76     jvms->set_map(newcall);
    77     jvms->set_locoff(jvms->locoff()+jvms_adj);
    78     jvms->set_stkoff(jvms->stkoff()+jvms_adj);
    79     jvms->set_monoff(jvms->monoff()+jvms_adj);
    80     jvms->set_scloff(jvms->scloff()+jvms_adj);
    81     jvms->set_endoff(jvms->endoff()+jvms_adj);
    82   }
    83 }
    85 Node* PhaseMacroExpand::opt_iff(Node* region, Node* iff) {
    86   IfNode *opt_iff = transform_later(iff)->as_If();
    88   // Fast path taken; set region slot 2
    89   Node *fast_taken = transform_later( new (C, 1) IfFalseNode(opt_iff) );
    90   region->init_req(2,fast_taken); // Capture fast-control
    92   // Fast path not-taken, i.e. slow path
    93   Node *slow_taken = transform_later( new (C, 1) IfTrueNode(opt_iff) );
    94   return slow_taken;
    95 }
    97 //--------------------copy_predefined_input_for_runtime_call--------------------
    98 void PhaseMacroExpand::copy_predefined_input_for_runtime_call(Node * ctrl, CallNode* oldcall, CallNode* call) {
    99   // Set fixed predefined input arguments
   100   call->init_req( TypeFunc::Control, ctrl );
   101   call->init_req( TypeFunc::I_O    , oldcall->in( TypeFunc::I_O) );
   102   call->init_req( TypeFunc::Memory , oldcall->in( TypeFunc::Memory ) ); // ?????
   103   call->init_req( TypeFunc::ReturnAdr, oldcall->in( TypeFunc::ReturnAdr ) );
   104   call->init_req( TypeFunc::FramePtr, oldcall->in( TypeFunc::FramePtr ) );
   105 }
   107 //------------------------------make_slow_call---------------------------------
   108 CallNode* PhaseMacroExpand::make_slow_call(CallNode *oldcall, const TypeFunc* slow_call_type, address slow_call, const char* leaf_name, Node* slow_path, Node* parm0, Node* parm1) {
   110   // Slow-path call
   111   int size = slow_call_type->domain()->cnt();
   112  CallNode *call = leaf_name
   113    ? (CallNode*)new (C, size) CallLeafNode      ( slow_call_type, slow_call, leaf_name, TypeRawPtr::BOTTOM )
   114    : (CallNode*)new (C, size) CallStaticJavaNode( slow_call_type, slow_call, OptoRuntime::stub_name(slow_call), oldcall->jvms()->bci(), TypeRawPtr::BOTTOM );
   116   // Slow path call has no side-effects, uses few values
   117   copy_predefined_input_for_runtime_call(slow_path, oldcall, call );
   118   if (parm0 != NULL)  call->init_req(TypeFunc::Parms+0, parm0);
   119   if (parm1 != NULL)  call->init_req(TypeFunc::Parms+1, parm1);
   120   copy_call_debug_info(oldcall, call);
   121   call->set_cnt(PROB_UNLIKELY_MAG(4));  // Same effect as RC_UNCOMMON.
   122   _igvn.hash_delete(oldcall);
   123   _igvn.subsume_node(oldcall, call);
   124   transform_later(call);
   126   return call;
   127 }
   129 void PhaseMacroExpand::extract_call_projections(CallNode *call) {
   130   _fallthroughproj = NULL;
   131   _fallthroughcatchproj = NULL;
   132   _ioproj_fallthrough = NULL;
   133   _ioproj_catchall = NULL;
   134   _catchallcatchproj = NULL;
   135   _memproj_fallthrough = NULL;
   136   _memproj_catchall = NULL;
   137   _resproj = NULL;
   138   for (DUIterator_Fast imax, i = call->fast_outs(imax); i < imax; i++) {
   139     ProjNode *pn = call->fast_out(i)->as_Proj();
   140     switch (pn->_con) {
   141       case TypeFunc::Control:
   142       {
   143         // For Control (fallthrough) and I_O (catch_all_index) we have CatchProj -> Catch -> Proj
   144         _fallthroughproj = pn;
   145         DUIterator_Fast jmax, j = pn->fast_outs(jmax);
   146         const Node *cn = pn->fast_out(j);
   147         if (cn->is_Catch()) {
   148           ProjNode *cpn = NULL;
   149           for (DUIterator_Fast kmax, k = cn->fast_outs(kmax); k < kmax; k++) {
   150             cpn = cn->fast_out(k)->as_Proj();
   151             assert(cpn->is_CatchProj(), "must be a CatchProjNode");
   152             if (cpn->_con == CatchProjNode::fall_through_index)
   153               _fallthroughcatchproj = cpn;
   154             else {
   155               assert(cpn->_con == CatchProjNode::catch_all_index, "must be correct index.");
   156               _catchallcatchproj = cpn;
   157             }
   158           }
   159         }
   160         break;
   161       }
   162       case TypeFunc::I_O:
   163         if (pn->_is_io_use)
   164           _ioproj_catchall = pn;
   165         else
   166           _ioproj_fallthrough = pn;
   167         break;
   168       case TypeFunc::Memory:
   169         if (pn->_is_io_use)
   170           _memproj_catchall = pn;
   171         else
   172           _memproj_fallthrough = pn;
   173         break;
   174       case TypeFunc::Parms:
   175         _resproj = pn;
   176         break;
   177       default:
   178         assert(false, "unexpected projection from allocation node.");
   179     }
   180   }
   182 }
   184 // Eliminate a card mark sequence.  p2x is a ConvP2XNode
   185 void PhaseMacroExpand::eliminate_card_mark(Node *p2x) {
   186   assert(p2x->Opcode() == Op_CastP2X, "ConvP2XNode required");
   187   Node *shift = p2x->unique_out();
   188   Node *addp = shift->unique_out();
   189   for (DUIterator_Last jmin, j = addp->last_outs(jmin); j >= jmin; --j) {
   190     Node *st = addp->last_out(j);
   191     assert(st->is_Store(), "store required");
   192     _igvn.replace_node(st, st->in(MemNode::Memory));
   193   }
   194 }
   196 // Search for a memory operation for the specified memory slice.
   197 static Node *scan_mem_chain(Node *mem, int alias_idx, int offset, Node *start_mem, Node *alloc) {
   198   Node *orig_mem = mem;
   199   Node *alloc_mem = alloc->in(TypeFunc::Memory);
   200   while (true) {
   201     if (mem == alloc_mem || mem == start_mem ) {
   202       return mem;  // hit one of our sentinals
   203     } else if (mem->is_MergeMem()) {
   204       mem = mem->as_MergeMem()->memory_at(alias_idx);
   205     } else if (mem->is_Proj() && mem->as_Proj()->_con == TypeFunc::Memory) {
   206       Node *in = mem->in(0);
   207       // we can safely skip over safepoints, calls, locks and membars because we
   208       // already know that the object is safe to eliminate.
   209       if (in->is_Initialize() && in->as_Initialize()->allocation() == alloc) {
   210         return in;
   211       } else if (in->is_Call() || in->is_MemBar()) {
   212         mem = in->in(TypeFunc::Memory);
   213       } else {
   214         assert(false, "unexpected projection");
   215       }
   216     } else if (mem->is_Store()) {
   217       const TypePtr* atype = mem->as_Store()->adr_type();
   218       int adr_idx = Compile::current()->get_alias_index(atype);
   219       if (adr_idx == alias_idx) {
   220         assert(atype->isa_oopptr(), "address type must be oopptr");
   221         int adr_offset = atype->offset();
   222         uint adr_iid = atype->is_oopptr()->instance_id();
   223         // Array elements references have the same alias_idx
   224         // but different offset and different instance_id.
   225         if (adr_offset == offset && adr_iid == alloc->_idx)
   226           return mem;
   227       } else {
   228         assert(adr_idx == Compile::AliasIdxRaw, "address must match or be raw");
   229       }
   230       mem = mem->in(MemNode::Memory);
   231     } else {
   232       return mem;
   233     }
   234     if (mem == orig_mem)
   235       return mem;
   236   }
   237 }
   239 //
   240 // Given a Memory Phi, compute a value Phi containing the values from stores
   241 // on the input paths.
   242 // Note: this function is recursive, its depth is limied by the "level" argument
   243 // Returns the computed Phi, or NULL if it cannot compute it.
   244 Node *PhaseMacroExpand::value_from_mem_phi(Node *mem, BasicType ft, const Type *phi_type, const TypeOopPtr *adr_t, Node *alloc, int level) {
   246   if (level <= 0) {
   247     return NULL;
   248   }
   249   int alias_idx = C->get_alias_index(adr_t);
   250   int offset = adr_t->offset();
   251   int instance_id = adr_t->instance_id();
   253   Node *start_mem = C->start()->proj_out(TypeFunc::Memory);
   254   Node *alloc_mem = alloc->in(TypeFunc::Memory);
   256   uint length = mem->req();
   257   GrowableArray <Node *> values(length, length, NULL);
   259   for (uint j = 1; j < length; j++) {
   260     Node *in = mem->in(j);
   261     if (in == NULL || in->is_top()) {
   262       values.at_put(j, in);
   263     } else  {
   264       Node *val = scan_mem_chain(in, alias_idx, offset, start_mem, alloc);
   265       if (val == start_mem || val == alloc_mem) {
   266         // hit a sentinel, return appropriate 0 value
   267         values.at_put(j, _igvn.zerocon(ft));
   268         continue;
   269       }
   270       if (val->is_Initialize()) {
   271         val = val->as_Initialize()->find_captured_store(offset, type2aelembytes(ft), &_igvn);
   272       }
   273       if (val == NULL) {
   274         return NULL;  // can't find a value on this path
   275       }
   276       if (val == mem) {
   277         values.at_put(j, mem);
   278       } else if (val->is_Store()) {
   279         values.at_put(j, val->in(MemNode::ValueIn));
   280       } else if(val->is_Proj() && val->in(0) == alloc) {
   281         values.at_put(j, _igvn.zerocon(ft));
   282       } else if (val->is_Phi()) {
   283         // Check if an appropriate node already exists.
   284         Node* region = val->in(0);
   285         Node* old_phi = NULL;
   286         for (DUIterator_Fast kmax, k = region->fast_outs(kmax); k < kmax; k++) {
   287           Node* phi = region->fast_out(k);
   288           if (phi->is_Phi() && phi != val &&
   289               phi->as_Phi()->is_same_inst_field(phi_type, instance_id, alias_idx, offset)) {
   290             old_phi = phi;
   291             break;
   292           }
   293         }
   294         if (old_phi == NULL) {
   295           val = value_from_mem_phi(val, ft, phi_type, adr_t, alloc, level-1);
   296           if (val == NULL) {
   297             return NULL;
   298           }
   299           values.at_put(j, val);
   300         } else {
   301           values.at_put(j, old_phi);
   302         }
   303       } else {
   304         return NULL;  // unknown node  on this path
   305       }
   306     }
   307   }
   308   // create a new Phi for the value
   309   PhiNode *phi = new (C, length) PhiNode(mem->in(0), phi_type, NULL, instance_id, alias_idx, offset);
   310   for (uint j = 1; j < length; j++) {
   311     if (values.at(j) == mem) {
   312       phi->init_req(j, phi);
   313     } else {
   314       phi->init_req(j, values.at(j));
   315     }
   316   }
   317   transform_later(phi);
   318   return phi;
   319 }
   321 // Search the last value stored into the object's field.
   322 Node *PhaseMacroExpand::value_from_mem(Node *sfpt_mem, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, Node *alloc) {
   323   assert(adr_t->is_instance_field(), "instance required");
   324   uint instance_id = adr_t->instance_id();
   325   assert(instance_id == alloc->_idx, "wrong allocation");
   327   int alias_idx = C->get_alias_index(adr_t);
   328   int offset = adr_t->offset();
   329   Node *start_mem = C->start()->proj_out(TypeFunc::Memory);
   330   Node *alloc_ctrl = alloc->in(TypeFunc::Control);
   331   Node *alloc_mem = alloc->in(TypeFunc::Memory);
   332   VectorSet visited(Thread::current()->resource_area());
   335   bool done = sfpt_mem == alloc_mem;
   336   Node *mem = sfpt_mem;
   337   while (!done) {
   338     if (visited.test_set(mem->_idx)) {
   339       return NULL;  // found a loop, give up
   340     }
   341     mem = scan_mem_chain(mem, alias_idx, offset, start_mem, alloc);
   342     if (mem == start_mem || mem == alloc_mem) {
   343       done = true;  // hit a sentinel, return appropriate 0 value
   344     } else if (mem->is_Initialize()) {
   345       mem = mem->as_Initialize()->find_captured_store(offset, type2aelembytes(ft), &_igvn);
   346       if (mem == NULL) {
   347         done = true; // Something go wrong.
   348       } else if (mem->is_Store()) {
   349         const TypePtr* atype = mem->as_Store()->adr_type();
   350         assert(C->get_alias_index(atype) == Compile::AliasIdxRaw, "store is correct memory slice");
   351         done = true;
   352       }
   353     } else if (mem->is_Store()) {
   354       const TypeOopPtr* atype = mem->as_Store()->adr_type()->isa_oopptr();
   355       assert(atype != NULL, "address type must be oopptr");
   356       assert(C->get_alias_index(atype) == alias_idx &&
   357              atype->is_instance_field() && atype->offset() == offset &&
   358              atype->instance_id() == instance_id, "store is correct memory slice");
   359       done = true;
   360     } else if (mem->is_Phi()) {
   361       // try to find a phi's unique input
   362       Node *unique_input = NULL;
   363       Node *top = C->top();
   364       for (uint i = 1; i < mem->req(); i++) {
   365         Node *n = scan_mem_chain(mem->in(i), alias_idx, offset, start_mem, alloc);
   366         if (n == NULL || n == top || n == mem) {
   367           continue;
   368         } else if (unique_input == NULL) {
   369           unique_input = n;
   370         } else if (unique_input != n) {
   371           unique_input = top;
   372           break;
   373         }
   374       }
   375       if (unique_input != NULL && unique_input != top) {
   376         mem = unique_input;
   377       } else {
   378         done = true;
   379       }
   380     } else {
   381       assert(false, "unexpected node");
   382     }
   383   }
   384   if (mem != NULL) {
   385     if (mem == start_mem || mem == alloc_mem) {
   386       // hit a sentinel, return appropriate 0 value
   387       return _igvn.zerocon(ft);
   388     } else if (mem->is_Store()) {
   389       return mem->in(MemNode::ValueIn);
   390     } else if (mem->is_Phi()) {
   391       // attempt to produce a Phi reflecting the values on the input paths of the Phi
   392       Node * phi = value_from_mem_phi(mem, ft, ftype, adr_t, alloc, 8);
   393       if (phi != NULL) {
   394         return phi;
   395       }
   396     }
   397   }
   398   // Something go wrong.
   399   return NULL;
   400 }
   402 // Check the possibility of scalar replacement.
   403 bool PhaseMacroExpand::can_eliminate_allocation(AllocateNode *alloc, GrowableArray <SafePointNode *>& safepoints) {
   404   //  Scan the uses of the allocation to check for anything that would
   405   //  prevent us from eliminating it.
   406   NOT_PRODUCT( const char* fail_eliminate = NULL; )
   407   DEBUG_ONLY( Node* disq_node = NULL; )
   408   bool  can_eliminate = true;
   410   Node* res = alloc->result_cast();
   411   const TypeOopPtr* res_type = NULL;
   412   if (res == NULL) {
   413     // All users were eliminated.
   414   } else if (!res->is_CheckCastPP()) {
   415     alloc->_is_scalar_replaceable = false;  // don't try again
   416     NOT_PRODUCT(fail_eliminate = "Allocation does not have unique CheckCastPP";)
   417     can_eliminate = false;
   418   } else {
   419     res_type = _igvn.type(res)->isa_oopptr();
   420     if (res_type == NULL) {
   421       NOT_PRODUCT(fail_eliminate = "Neither instance or array allocation";)
   422       can_eliminate = false;
   423     } else if (res_type->isa_aryptr()) {
   424       int length = alloc->in(AllocateNode::ALength)->find_int_con(-1);
   425       if (length < 0) {
   426         NOT_PRODUCT(fail_eliminate = "Array's size is not constant";)
   427         can_eliminate = false;
   428       }
   429     }
   430   }
   432   if (can_eliminate && res != NULL) {
   433     for (DUIterator_Fast jmax, j = res->fast_outs(jmax);
   434                                j < jmax && can_eliminate; j++) {
   435       Node* use = res->fast_out(j);
   437       if (use->is_AddP()) {
   438         const TypePtr* addp_type = _igvn.type(use)->is_ptr();
   439         int offset = addp_type->offset();
   441         if (offset == Type::OffsetTop || offset == Type::OffsetBot) {
   442           NOT_PRODUCT(fail_eliminate = "Undefined field referrence";)
   443           can_eliminate = false;
   444           break;
   445         }
   446         for (DUIterator_Fast kmax, k = use->fast_outs(kmax);
   447                                    k < kmax && can_eliminate; k++) {
   448           Node* n = use->fast_out(k);
   449           if (!n->is_Store() && n->Opcode() != Op_CastP2X) {
   450             DEBUG_ONLY(disq_node = n;)
   451             if (n->is_Load()) {
   452               NOT_PRODUCT(fail_eliminate = "Field load";)
   453             } else {
   454               NOT_PRODUCT(fail_eliminate = "Not store field referrence";)
   455             }
   456             can_eliminate = false;
   457           }
   458         }
   459       } else if (use->is_SafePoint()) {
   460         SafePointNode* sfpt = use->as_SafePoint();
   461         if (sfpt->has_non_debug_use(res)) {
   462           // Object is passed as argument.
   463           DEBUG_ONLY(disq_node = use;)
   464           NOT_PRODUCT(fail_eliminate = "Object is passed as argument";)
   465           can_eliminate = false;
   466         }
   467         Node* sfptMem = sfpt->memory();
   468         if (sfptMem == NULL || sfptMem->is_top()) {
   469           DEBUG_ONLY(disq_node = use;)
   470           NOT_PRODUCT(fail_eliminate = "NULL or TOP memory";)
   471           can_eliminate = false;
   472         } else {
   473           safepoints.append_if_missing(sfpt);
   474         }
   475       } else if (use->Opcode() != Op_CastP2X) { // CastP2X is used by card mark
   476         if (use->is_Phi()) {
   477           if (use->outcnt() == 1 && use->unique_out()->Opcode() == Op_Return) {
   478             NOT_PRODUCT(fail_eliminate = "Object is return value";)
   479           } else {
   480             NOT_PRODUCT(fail_eliminate = "Object is referenced by Phi";)
   481           }
   482           DEBUG_ONLY(disq_node = use;)
   483         } else {
   484           if (use->Opcode() == Op_Return) {
   485             NOT_PRODUCT(fail_eliminate = "Object is return value";)
   486           }else {
   487             NOT_PRODUCT(fail_eliminate = "Object is referenced by node";)
   488           }
   489           DEBUG_ONLY(disq_node = use;)
   490         }
   491         can_eliminate = false;
   492       }
   493     }
   494   }
   496 #ifndef PRODUCT
   497   if (PrintEliminateAllocations) {
   498     if (can_eliminate) {
   499       tty->print("Scalar ");
   500       if (res == NULL)
   501         alloc->dump();
   502       else
   503         res->dump();
   504     } else {
   505       tty->print("NotScalar (%s)", fail_eliminate);
   506       if (res == NULL)
   507         alloc->dump();
   508       else
   509         res->dump();
   510 #ifdef ASSERT
   511       if (disq_node != NULL) {
   512           tty->print("  >>>> ");
   513           disq_node->dump();
   514       }
   515 #endif /*ASSERT*/
   516     }
   517   }
   518 #endif
   519   return can_eliminate;
   520 }
   522 // Do scalar replacement.
   523 bool PhaseMacroExpand::scalar_replacement(AllocateNode *alloc, GrowableArray <SafePointNode *>& safepoints) {
   524   GrowableArray <SafePointNode *> safepoints_done;
   526   ciKlass* klass = NULL;
   527   ciInstanceKlass* iklass = NULL;
   528   int nfields = 0;
   529   int array_base;
   530   int element_size;
   531   BasicType basic_elem_type;
   532   ciType* elem_type;
   534   Node* res = alloc->result_cast();
   535   const TypeOopPtr* res_type = NULL;
   536   if (res != NULL) { // Could be NULL when there are no users
   537     res_type = _igvn.type(res)->isa_oopptr();
   538   }
   540   if (res != NULL) {
   541     klass = res_type->klass();
   542     if (res_type->isa_instptr()) {
   543       // find the fields of the class which will be needed for safepoint debug information
   544       assert(klass->is_instance_klass(), "must be an instance klass.");
   545       iklass = klass->as_instance_klass();
   546       nfields = iklass->nof_nonstatic_fields();
   547     } else {
   548       // find the array's elements which will be needed for safepoint debug information
   549       nfields = alloc->in(AllocateNode::ALength)->find_int_con(-1);
   550       assert(klass->is_array_klass() && nfields >= 0, "must be an array klass.");
   551       elem_type = klass->as_array_klass()->element_type();
   552       basic_elem_type = elem_type->basic_type();
   553       array_base = arrayOopDesc::base_offset_in_bytes(basic_elem_type);
   554       element_size = type2aelembytes(basic_elem_type);
   555     }
   556   }
   557   //
   558   // Process the safepoint uses
   559   //
   560   while (safepoints.length() > 0) {
   561     SafePointNode* sfpt = safepoints.pop();
   562     Node* mem = sfpt->memory();
   563     uint first_ind = sfpt->req();
   564     SafePointScalarObjectNode* sobj = new (C, 1) SafePointScalarObjectNode(res_type,
   565 #ifdef ASSERT
   566                                                  alloc,
   567 #endif
   568                                                  first_ind, nfields);
   569     sobj->init_req(0, sfpt->in(TypeFunc::Control));
   570     transform_later(sobj);
   572     // Scan object's fields adding an input to the safepoint for each field.
   573     for (int j = 0; j < nfields; j++) {
   574       int offset;
   575       ciField* field = NULL;
   576       if (iklass != NULL) {
   577         field = iklass->nonstatic_field_at(j);
   578         offset = field->offset();
   579         elem_type = field->type();
   580         basic_elem_type = field->layout_type();
   581       } else {
   582         offset = array_base + j * element_size;
   583       }
   585       const Type *field_type;
   586       // The next code is taken from Parse::do_get_xxx().
   587       if (basic_elem_type == T_OBJECT) {
   588         if (!elem_type->is_loaded()) {
   589           field_type = TypeInstPtr::BOTTOM;
   590         } else if (field != NULL && field->is_constant()) {
   591           // This can happen if the constant oop is non-perm.
   592           ciObject* con = field->constant_value().as_object();
   593           // Do not "join" in the previous type; it doesn't add value,
   594           // and may yield a vacuous result if the field is of interface type.
   595           field_type = TypeOopPtr::make_from_constant(con)->isa_oopptr();
   596           assert(field_type != NULL, "field singleton type must be consistent");
   597         } else {
   598           field_type = TypeOopPtr::make_from_klass(elem_type->as_klass());
   599         }
   600       } else {
   601         field_type = Type::get_const_basic_type(basic_elem_type);
   602       }
   604       const TypeOopPtr *field_addr_type = res_type->add_offset(offset)->isa_oopptr();
   606       Node *field_val = value_from_mem(mem, basic_elem_type, field_type, field_addr_type, alloc);
   607       if (field_val == NULL) {
   608         // we weren't able to find a value for this field,
   609         // give up on eliminating this allocation
   610         alloc->_is_scalar_replaceable = false;  // don't try again
   611         // remove any extra entries we added to the safepoint
   612         uint last = sfpt->req() - 1;
   613         for (int k = 0;  k < j; k++) {
   614           sfpt->del_req(last--);
   615         }
   616         // rollback processed safepoints
   617         while (safepoints_done.length() > 0) {
   618           SafePointNode* sfpt_done = safepoints_done.pop();
   619           // remove any extra entries we added to the safepoint
   620           last = sfpt_done->req() - 1;
   621           for (int k = 0;  k < nfields; k++) {
   622             sfpt_done->del_req(last--);
   623           }
   624           JVMState *jvms = sfpt_done->jvms();
   625           jvms->set_endoff(sfpt_done->req());
   626           // Now make a pass over the debug information replacing any references
   627           // to SafePointScalarObjectNode with the allocated object.
   628           int start = jvms->debug_start();
   629           int end   = jvms->debug_end();
   630           for (int i = start; i < end; i++) {
   631             if (sfpt_done->in(i)->is_SafePointScalarObject()) {
   632               SafePointScalarObjectNode* scobj = sfpt_done->in(i)->as_SafePointScalarObject();
   633               if (scobj->first_index() == sfpt_done->req() &&
   634                   scobj->n_fields() == (uint)nfields) {
   635                 assert(scobj->alloc() == alloc, "sanity");
   636                 sfpt_done->set_req(i, res);
   637               }
   638             }
   639           }
   640         }
   641 #ifndef PRODUCT
   642         if (PrintEliminateAllocations) {
   643           if (field != NULL) {
   644             tty->print("=== At SafePoint node %d can't find value of Field: ",
   645                        sfpt->_idx);
   646             field->print();
   647             int field_idx = C->get_alias_index(field_addr_type);
   648             tty->print(" (alias_idx=%d)", field_idx);
   649           } else { // Array's element
   650             tty->print("=== At SafePoint node %d can't find value of array element [%d]",
   651                        sfpt->_idx, j);
   652           }
   653           tty->print(", which prevents elimination of: ");
   654           if (res == NULL)
   655             alloc->dump();
   656           else
   657             res->dump();
   658         }
   659 #endif
   660         return false;
   661       }
   662       sfpt->add_req(field_val);
   663     }
   664     JVMState *jvms = sfpt->jvms();
   665     jvms->set_endoff(sfpt->req());
   666     // Now make a pass over the debug information replacing any references
   667     // to the allocated object with "sobj"
   668     int start = jvms->debug_start();
   669     int end   = jvms->debug_end();
   670     for (int i = start; i < end; i++) {
   671       if (sfpt->in(i) == res) {
   672         sfpt->set_req(i, sobj);
   673       }
   674     }
   675     safepoints_done.append_if_missing(sfpt); // keep it for rollback
   676   }
   677   return true;
   678 }
   680 // Process users of eliminated allocation.
   681 void PhaseMacroExpand::process_users_of_allocation(AllocateNode *alloc) {
   682   Node* res = alloc->result_cast();
   683   if (res != NULL) {
   684     for (DUIterator_Last jmin, j = res->last_outs(jmin); j >= jmin; ) {
   685       Node *use = res->last_out(j);
   686       uint oc1 = res->outcnt();
   688       if (use->is_AddP()) {
   689         for (DUIterator_Last kmin, k = use->last_outs(kmin); k >= kmin; ) {
   690           Node *n = use->last_out(k);
   691           uint oc2 = use->outcnt();
   692           if (n->is_Store()) {
   693             _igvn.replace_node(n, n->in(MemNode::Memory));
   694           } else {
   695             assert( n->Opcode() == Op_CastP2X, "CastP2X required");
   696             eliminate_card_mark(n);
   697           }
   698           k -= (oc2 - use->outcnt());
   699         }
   700       } else {
   701         assert( !use->is_SafePoint(), "safepoint uses must have been already elimiated");
   702         assert( use->Opcode() == Op_CastP2X, "CastP2X required");
   703         eliminate_card_mark(use);
   704       }
   705       j -= (oc1 - res->outcnt());
   706     }
   707     assert(res->outcnt() == 0, "all uses of allocated objects must be deleted");
   708     _igvn.remove_dead_node(res);
   709   }
   711   //
   712   // Process other users of allocation's projections
   713   //
   714   if (_resproj != NULL && _resproj->outcnt() != 0) {
   715     for (DUIterator_Last jmin, j = _resproj->last_outs(jmin); j >= jmin; ) {
   716       Node *use = _resproj->last_out(j);
   717       uint oc1 = _resproj->outcnt();
   718       if (use->is_Initialize()) {
   719         // Eliminate Initialize node.
   720         InitializeNode *init = use->as_Initialize();
   721         assert(init->outcnt() <= 2, "only a control and memory projection expected");
   722         Node *ctrl_proj = init->proj_out(TypeFunc::Control);
   723         if (ctrl_proj != NULL) {
   724            assert(init->in(TypeFunc::Control) == _fallthroughcatchproj, "allocation control projection");
   725           _igvn.replace_node(ctrl_proj, _fallthroughcatchproj);
   726         }
   727         Node *mem_proj = init->proj_out(TypeFunc::Memory);
   728         if (mem_proj != NULL) {
   729           Node *mem = init->in(TypeFunc::Memory);
   730 #ifdef ASSERT
   731           if (mem->is_MergeMem()) {
   732             assert(mem->in(TypeFunc::Memory) == _memproj_fallthrough, "allocation memory projection");
   733           } else {
   734             assert(mem == _memproj_fallthrough, "allocation memory projection");
   735           }
   736 #endif
   737           _igvn.replace_node(mem_proj, mem);
   738         }
   739       } else if (use->is_AddP()) {
   740         // raw memory addresses used only by the initialization
   741         _igvn.hash_delete(use);
   742         _igvn.subsume_node(use, C->top());
   743       } else  {
   744         assert(false, "only Initialize or AddP expected");
   745       }
   746       j -= (oc1 - _resproj->outcnt());
   747     }
   748   }
   749   if (_fallthroughcatchproj != NULL) {
   750     _igvn.replace_node(_fallthroughcatchproj, alloc->in(TypeFunc::Control));
   751   }
   752   if (_memproj_fallthrough != NULL) {
   753     _igvn.replace_node(_memproj_fallthrough, alloc->in(TypeFunc::Memory));
   754   }
   755   if (_memproj_catchall != NULL) {
   756     _igvn.replace_node(_memproj_catchall, C->top());
   757   }
   758   if (_ioproj_fallthrough != NULL) {
   759     _igvn.replace_node(_ioproj_fallthrough, alloc->in(TypeFunc::I_O));
   760   }
   761   if (_ioproj_catchall != NULL) {
   762     _igvn.replace_node(_ioproj_catchall, C->top());
   763   }
   764   if (_catchallcatchproj != NULL) {
   765     _igvn.replace_node(_catchallcatchproj, C->top());
   766   }
   767 }
   769 bool PhaseMacroExpand::eliminate_allocate_node(AllocateNode *alloc) {
   771   if (!EliminateAllocations || !alloc->_is_scalar_replaceable) {
   772     return false;
   773   }
   775   extract_call_projections(alloc);
   777   GrowableArray <SafePointNode *> safepoints;
   778   if (!can_eliminate_allocation(alloc, safepoints)) {
   779     return false;
   780   }
   782   if (!scalar_replacement(alloc, safepoints)) {
   783     return false;
   784   }
   786   process_users_of_allocation(alloc);
   788 #ifndef PRODUCT
   789 if (PrintEliminateAllocations) {
   790   if (alloc->is_AllocateArray())
   791     tty->print_cr("++++ Eliminated: %d AllocateArray", alloc->_idx);
   792   else
   793     tty->print_cr("++++ Eliminated: %d Allocate", alloc->_idx);
   794 }
   795 #endif
   797   return true;
   798 }
   801 //---------------------------set_eden_pointers-------------------------
   802 void PhaseMacroExpand::set_eden_pointers(Node* &eden_top_adr, Node* &eden_end_adr) {
   803   if (UseTLAB) {                // Private allocation: load from TLS
   804     Node* thread = transform_later(new (C, 1) ThreadLocalNode());
   805     int tlab_top_offset = in_bytes(JavaThread::tlab_top_offset());
   806     int tlab_end_offset = in_bytes(JavaThread::tlab_end_offset());
   807     eden_top_adr = basic_plus_adr(top()/*not oop*/, thread, tlab_top_offset);
   808     eden_end_adr = basic_plus_adr(top()/*not oop*/, thread, tlab_end_offset);
   809   } else {                      // Shared allocation: load from globals
   810     CollectedHeap* ch = Universe::heap();
   811     address top_adr = (address)ch->top_addr();
   812     address end_adr = (address)ch->end_addr();
   813     eden_top_adr = makecon(TypeRawPtr::make(top_adr));
   814     eden_end_adr = basic_plus_adr(eden_top_adr, end_adr - top_adr);
   815   }
   816 }
   819 Node* PhaseMacroExpand::make_load(Node* ctl, Node* mem, Node* base, int offset, const Type* value_type, BasicType bt) {
   820   Node* adr = basic_plus_adr(base, offset);
   821   const TypePtr* adr_type = TypeRawPtr::BOTTOM;
   822   Node* value = LoadNode::make(C, ctl, mem, adr, adr_type, value_type, bt);
   823   transform_later(value);
   824   return value;
   825 }
   828 Node* PhaseMacroExpand::make_store(Node* ctl, Node* mem, Node* base, int offset, Node* value, BasicType bt) {
   829   Node* adr = basic_plus_adr(base, offset);
   830   mem = StoreNode::make(C, ctl, mem, adr, NULL, value, bt);
   831   transform_later(mem);
   832   return mem;
   833 }
   835 //=============================================================================
   836 //
   837 //                              A L L O C A T I O N
   838 //
   839 // Allocation attempts to be fast in the case of frequent small objects.
   840 // It breaks down like this:
   841 //
   842 // 1) Size in doublewords is computed.  This is a constant for objects and
   843 // variable for most arrays.  Doubleword units are used to avoid size
   844 // overflow of huge doubleword arrays.  We need doublewords in the end for
   845 // rounding.
   846 //
   847 // 2) Size is checked for being 'too large'.  Too-large allocations will go
   848 // the slow path into the VM.  The slow path can throw any required
   849 // exceptions, and does all the special checks for very large arrays.  The
   850 // size test can constant-fold away for objects.  For objects with
   851 // finalizers it constant-folds the otherway: you always go slow with
   852 // finalizers.
   853 //
   854 // 3) If NOT using TLABs, this is the contended loop-back point.
   855 // Load-Locked the heap top.  If using TLABs normal-load the heap top.
   856 //
   857 // 4) Check that heap top + size*8 < max.  If we fail go the slow ` route.
   858 // NOTE: "top+size*8" cannot wrap the 4Gig line!  Here's why: for largish
   859 // "size*8" we always enter the VM, where "largish" is a constant picked small
   860 // enough that there's always space between the eden max and 4Gig (old space is
   861 // there so it's quite large) and large enough that the cost of entering the VM
   862 // is dwarfed by the cost to initialize the space.
   863 //
   864 // 5) If NOT using TLABs, Store-Conditional the adjusted heap top back
   865 // down.  If contended, repeat at step 3.  If using TLABs normal-store
   866 // adjusted heap top back down; there is no contention.
   867 //
   868 // 6) If !ZeroTLAB then Bulk-clear the object/array.  Fill in klass & mark
   869 // fields.
   870 //
   871 // 7) Merge with the slow-path; cast the raw memory pointer to the correct
   872 // oop flavor.
   873 //
   874 //=============================================================================
   875 // FastAllocateSizeLimit value is in DOUBLEWORDS.
   876 // Allocations bigger than this always go the slow route.
   877 // This value must be small enough that allocation attempts that need to
   878 // trigger exceptions go the slow route.  Also, it must be small enough so
   879 // that heap_top + size_in_bytes does not wrap around the 4Gig limit.
   880 //=============================================================================j//
   881 // %%% Here is an old comment from parseHelper.cpp; is it outdated?
   882 // The allocator will coalesce int->oop copies away.  See comment in
   883 // coalesce.cpp about how this works.  It depends critically on the exact
   884 // code shape produced here, so if you are changing this code shape
   885 // make sure the GC info for the heap-top is correct in and around the
   886 // slow-path call.
   887 //
   889 void PhaseMacroExpand::expand_allocate_common(
   890             AllocateNode* alloc, // allocation node to be expanded
   891             Node* length,  // array length for an array allocation
   892             const TypeFunc* slow_call_type, // Type of slow call
   893             address slow_call_address  // Address of slow call
   894     )
   895 {
   897   Node* ctrl = alloc->in(TypeFunc::Control);
   898   Node* mem  = alloc->in(TypeFunc::Memory);
   899   Node* i_o  = alloc->in(TypeFunc::I_O);
   900   Node* size_in_bytes     = alloc->in(AllocateNode::AllocSize);
   901   Node* klass_node        = alloc->in(AllocateNode::KlassNode);
   902   Node* initial_slow_test = alloc->in(AllocateNode::InitialTest);
   904   // With escape analysis, the entire memory state was needed to be able to
   905   // eliminate the allocation.  Since the allocations cannot be eliminated,
   906   // optimize it to the raw slice.
   907   if (mem->is_MergeMem()) {
   908     mem = mem->as_MergeMem()->memory_at(Compile::AliasIdxRaw);
   909   }
   911   Node* eden_top_adr;
   912   Node* eden_end_adr;
   913   set_eden_pointers(eden_top_adr, eden_end_adr);
   915   uint raw_idx = C->get_alias_index(TypeRawPtr::BOTTOM);
   916   assert(ctrl != NULL, "must have control");
   918   // Load Eden::end.  Loop invariant and hoisted.
   919   //
   920   // Note: We set the control input on "eden_end" and "old_eden_top" when using
   921   //       a TLAB to work around a bug where these values were being moved across
   922   //       a safepoint.  These are not oops, so they cannot be include in the oop
   923   //       map, but the can be changed by a GC.   The proper way to fix this would
   924   //       be to set the raw memory state when generating a  SafepointNode.  However
   925   //       this will require extensive changes to the loop optimization in order to
   926   //       prevent a degradation of the optimization.
   927   //       See comment in memnode.hpp, around line 227 in class LoadPNode.
   928   Node* eden_end = make_load(ctrl, mem, eden_end_adr, 0, TypeRawPtr::BOTTOM, T_ADDRESS);
   930   // We need a Region and corresponding Phi's to merge the slow-path and fast-path results.
   931   // they will not be used if "always_slow" is set
   932   enum { slow_result_path = 1, fast_result_path = 2 };
   933   Node *result_region;
   934   Node *result_phi_rawmem;
   935   Node *result_phi_rawoop;
   936   Node *result_phi_i_o;
   938   // The initial slow comparison is a size check, the comparison
   939   // we want to do is a BoolTest::gt
   940   bool always_slow = false;
   941   int tv = _igvn.find_int_con(initial_slow_test, -1);
   942   if (tv >= 0) {
   943     always_slow = (tv == 1);
   944     initial_slow_test = NULL;
   945   } else {
   946     initial_slow_test = BoolNode::make_predicate(initial_slow_test, &_igvn);
   947   }
   949   if (DTraceAllocProbes) {
   950     // Force slow-path allocation
   951     always_slow = true;
   952     initial_slow_test = NULL;
   953   }
   955   enum { too_big_or_final_path = 1, need_gc_path = 2 };
   956   Node *slow_region = NULL;
   957   Node *toobig_false = ctrl;
   959   assert (initial_slow_test == NULL || !always_slow, "arguments must be consistent");
   960   // generate the initial test if necessary
   961   if (initial_slow_test != NULL ) {
   962     slow_region = new (C, 3) RegionNode(3);
   964     // Now make the initial failure test.  Usually a too-big test but
   965     // might be a TRUE for finalizers or a fancy class check for
   966     // newInstance0.
   967     IfNode *toobig_iff = new (C, 2) IfNode(ctrl, initial_slow_test, PROB_MIN, COUNT_UNKNOWN);
   968     transform_later(toobig_iff);
   969     // Plug the failing-too-big test into the slow-path region
   970     Node *toobig_true = new (C, 1) IfTrueNode( toobig_iff );
   971     transform_later(toobig_true);
   972     slow_region    ->init_req( too_big_or_final_path, toobig_true );
   973     toobig_false = new (C, 1) IfFalseNode( toobig_iff );
   974     transform_later(toobig_false);
   975   } else {         // No initial test, just fall into next case
   976     toobig_false = ctrl;
   977     debug_only(slow_region = NodeSentinel);
   978   }
   980   Node *slow_mem = mem;  // save the current memory state for slow path
   981   // generate the fast allocation code unless we know that the initial test will always go slow
   982   if (!always_slow) {
   983     // allocate the Region and Phi nodes for the result
   984     result_region = new (C, 3) RegionNode(3);
   985     result_phi_rawmem = new (C, 3) PhiNode( result_region, Type::MEMORY, TypeRawPtr::BOTTOM );
   986     result_phi_rawoop = new (C, 3) PhiNode( result_region, TypeRawPtr::BOTTOM );
   987     result_phi_i_o    = new (C, 3) PhiNode( result_region, Type::ABIO ); // I/O is used for Prefetch
   989     // We need a Region for the loop-back contended case.
   990     enum { fall_in_path = 1, contended_loopback_path = 2 };
   991     Node *contended_region;
   992     Node *contended_phi_rawmem;
   993     if( UseTLAB ) {
   994       contended_region = toobig_false;
   995       contended_phi_rawmem = mem;
   996     } else {
   997       contended_region = new (C, 3) RegionNode(3);
   998       contended_phi_rawmem = new (C, 3) PhiNode( contended_region, Type::MEMORY, TypeRawPtr::BOTTOM);
   999       // Now handle the passing-too-big test.  We fall into the contended
  1000       // loop-back merge point.
  1001       contended_region    ->init_req( fall_in_path, toobig_false );
  1002       contended_phi_rawmem->init_req( fall_in_path, mem );
  1003       transform_later(contended_region);
  1004       transform_later(contended_phi_rawmem);
  1007     // Load(-locked) the heap top.
  1008     // See note above concerning the control input when using a TLAB
  1009     Node *old_eden_top = UseTLAB
  1010       ? new (C, 3) LoadPNode     ( ctrl, contended_phi_rawmem, eden_top_adr, TypeRawPtr::BOTTOM, TypeRawPtr::BOTTOM )
  1011       : new (C, 3) LoadPLockedNode( contended_region, contended_phi_rawmem, eden_top_adr );
  1013     transform_later(old_eden_top);
  1014     // Add to heap top to get a new heap top
  1015     Node *new_eden_top = new (C, 4) AddPNode( top(), old_eden_top, size_in_bytes );
  1016     transform_later(new_eden_top);
  1017     // Check for needing a GC; compare against heap end
  1018     Node *needgc_cmp = new (C, 3) CmpPNode( new_eden_top, eden_end );
  1019     transform_later(needgc_cmp);
  1020     Node *needgc_bol = new (C, 2) BoolNode( needgc_cmp, BoolTest::ge );
  1021     transform_later(needgc_bol);
  1022     IfNode *needgc_iff = new (C, 2) IfNode(contended_region, needgc_bol, PROB_UNLIKELY_MAG(4), COUNT_UNKNOWN );
  1023     transform_later(needgc_iff);
  1025     // Plug the failing-heap-space-need-gc test into the slow-path region
  1026     Node *needgc_true = new (C, 1) IfTrueNode( needgc_iff );
  1027     transform_later(needgc_true);
  1028     if( initial_slow_test ) {
  1029       slow_region    ->init_req( need_gc_path, needgc_true );
  1030       // This completes all paths into the slow merge point
  1031       transform_later(slow_region);
  1032     } else {                      // No initial slow path needed!
  1033       // Just fall from the need-GC path straight into the VM call.
  1034       slow_region    = needgc_true;
  1036     // No need for a GC.  Setup for the Store-Conditional
  1037     Node *needgc_false = new (C, 1) IfFalseNode( needgc_iff );
  1038     transform_later(needgc_false);
  1040     // Grab regular I/O before optional prefetch may change it.
  1041     // Slow-path does no I/O so just set it to the original I/O.
  1042     result_phi_i_o->init_req( slow_result_path, i_o );
  1044     i_o = prefetch_allocation(i_o, needgc_false, contended_phi_rawmem,
  1045                               old_eden_top, new_eden_top, length);
  1047     // Store (-conditional) the modified eden top back down.
  1048     // StorePConditional produces flags for a test PLUS a modified raw
  1049     // memory state.
  1050     Node *store_eden_top;
  1051     Node *fast_oop_ctrl;
  1052     if( UseTLAB ) {
  1053       store_eden_top = new (C, 4) StorePNode( needgc_false, contended_phi_rawmem, eden_top_adr, TypeRawPtr::BOTTOM, new_eden_top );
  1054       transform_later(store_eden_top);
  1055       fast_oop_ctrl = needgc_false; // No contention, so this is the fast path
  1056     } else {
  1057       store_eden_top = new (C, 5) StorePConditionalNode( needgc_false, contended_phi_rawmem, eden_top_adr, new_eden_top, old_eden_top );
  1058       transform_later(store_eden_top);
  1059       Node *contention_check = new (C, 2) BoolNode( store_eden_top, BoolTest::ne );
  1060       transform_later(contention_check);
  1061       store_eden_top = new (C, 1) SCMemProjNode(store_eden_top);
  1062       transform_later(store_eden_top);
  1064       // If not using TLABs, check to see if there was contention.
  1065       IfNode *contention_iff = new (C, 2) IfNode ( needgc_false, contention_check, PROB_MIN, COUNT_UNKNOWN );
  1066       transform_later(contention_iff);
  1067       Node *contention_true = new (C, 1) IfTrueNode( contention_iff );
  1068       transform_later(contention_true);
  1069       // If contention, loopback and try again.
  1070       contended_region->init_req( contended_loopback_path, contention_true );
  1071       contended_phi_rawmem->init_req( contended_loopback_path, store_eden_top );
  1073       // Fast-path succeeded with no contention!
  1074       Node *contention_false = new (C, 1) IfFalseNode( contention_iff );
  1075       transform_later(contention_false);
  1076       fast_oop_ctrl = contention_false;
  1079     // Rename successful fast-path variables to make meaning more obvious
  1080     Node* fast_oop        = old_eden_top;
  1081     Node* fast_oop_rawmem = store_eden_top;
  1082     fast_oop_rawmem = initialize_object(alloc,
  1083                                         fast_oop_ctrl, fast_oop_rawmem, fast_oop,
  1084                                         klass_node, length, size_in_bytes);
  1086     if (ExtendedDTraceProbes) {
  1087       // Slow-path call
  1088       int size = TypeFunc::Parms + 2;
  1089       CallLeafNode *call = new (C, size) CallLeafNode(OptoRuntime::dtrace_object_alloc_Type(),
  1090                                                       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc_base),
  1091                                                       "dtrace_object_alloc",
  1092                                                       TypeRawPtr::BOTTOM);
  1094       // Get base of thread-local storage area
  1095       Node* thread = new (C, 1) ThreadLocalNode();
  1096       transform_later(thread);
  1098       call->init_req(TypeFunc::Parms+0, thread);
  1099       call->init_req(TypeFunc::Parms+1, fast_oop);
  1100       call->init_req( TypeFunc::Control, fast_oop_ctrl );
  1101       call->init_req( TypeFunc::I_O    , top() )        ;   // does no i/o
  1102       call->init_req( TypeFunc::Memory , fast_oop_rawmem );
  1103       call->init_req( TypeFunc::ReturnAdr, alloc->in(TypeFunc::ReturnAdr) );
  1104       call->init_req( TypeFunc::FramePtr, alloc->in(TypeFunc::FramePtr) );
  1105       transform_later(call);
  1106       fast_oop_ctrl = new (C, 1) ProjNode(call,TypeFunc::Control);
  1107       transform_later(fast_oop_ctrl);
  1108       fast_oop_rawmem = new (C, 1) ProjNode(call,TypeFunc::Memory);
  1109       transform_later(fast_oop_rawmem);
  1112     // Plug in the successful fast-path into the result merge point
  1113     result_region    ->init_req( fast_result_path, fast_oop_ctrl );
  1114     result_phi_rawoop->init_req( fast_result_path, fast_oop );
  1115     result_phi_i_o   ->init_req( fast_result_path, i_o );
  1116     result_phi_rawmem->init_req( fast_result_path, fast_oop_rawmem );
  1117   } else {
  1118     slow_region = ctrl;
  1121   // Generate slow-path call
  1122   CallNode *call = new (C, slow_call_type->domain()->cnt())
  1123     CallStaticJavaNode(slow_call_type, slow_call_address,
  1124                        OptoRuntime::stub_name(slow_call_address),
  1125                        alloc->jvms()->bci(),
  1126                        TypePtr::BOTTOM);
  1127   call->init_req( TypeFunc::Control, slow_region );
  1128   call->init_req( TypeFunc::I_O    , top() )     ;   // does no i/o
  1129   call->init_req( TypeFunc::Memory , slow_mem ); // may gc ptrs
  1130   call->init_req( TypeFunc::ReturnAdr, alloc->in(TypeFunc::ReturnAdr) );
  1131   call->init_req( TypeFunc::FramePtr, alloc->in(TypeFunc::FramePtr) );
  1133   call->init_req(TypeFunc::Parms+0, klass_node);
  1134   if (length != NULL) {
  1135     call->init_req(TypeFunc::Parms+1, length);
  1138   // Copy debug information and adjust JVMState information, then replace
  1139   // allocate node with the call
  1140   copy_call_debug_info((CallNode *) alloc,  call);
  1141   if (!always_slow) {
  1142     call->set_cnt(PROB_UNLIKELY_MAG(4));  // Same effect as RC_UNCOMMON.
  1144   _igvn.hash_delete(alloc);
  1145   _igvn.subsume_node(alloc, call);
  1146   transform_later(call);
  1148   // Identify the output projections from the allocate node and
  1149   // adjust any references to them.
  1150   // The control and io projections look like:
  1151   //
  1152   //        v---Proj(ctrl) <-----+   v---CatchProj(ctrl)
  1153   //  Allocate                   Catch
  1154   //        ^---Proj(io) <-------+   ^---CatchProj(io)
  1155   //
  1156   //  We are interested in the CatchProj nodes.
  1157   //
  1158   extract_call_projections(call);
  1160   // An allocate node has separate memory projections for the uses on the control and i_o paths
  1161   // Replace uses of the control memory projection with result_phi_rawmem (unless we are only generating a slow call)
  1162   if (!always_slow && _memproj_fallthrough != NULL) {
  1163     for (DUIterator_Fast imax, i = _memproj_fallthrough->fast_outs(imax); i < imax; i++) {
  1164       Node *use = _memproj_fallthrough->fast_out(i);
  1165       _igvn.hash_delete(use);
  1166       imax -= replace_input(use, _memproj_fallthrough, result_phi_rawmem);
  1167       _igvn._worklist.push(use);
  1168       // back up iterator
  1169       --i;
  1172   // Now change uses of _memproj_catchall to use _memproj_fallthrough and delete _memproj_catchall so
  1173   // we end up with a call that has only 1 memory projection
  1174   if (_memproj_catchall != NULL ) {
  1175     if (_memproj_fallthrough == NULL) {
  1176       _memproj_fallthrough = new (C, 1) ProjNode(call, TypeFunc::Memory);
  1177       transform_later(_memproj_fallthrough);
  1179     for (DUIterator_Fast imax, i = _memproj_catchall->fast_outs(imax); i < imax; i++) {
  1180       Node *use = _memproj_catchall->fast_out(i);
  1181       _igvn.hash_delete(use);
  1182       imax -= replace_input(use, _memproj_catchall, _memproj_fallthrough);
  1183       _igvn._worklist.push(use);
  1184       // back up iterator
  1185       --i;
  1189   mem = result_phi_rawmem;
  1191   // An allocate node has separate i_o projections for the uses on the control and i_o paths
  1192   // Replace uses of the control i_o projection with result_phi_i_o (unless we are only generating a slow call)
  1193   if (_ioproj_fallthrough == NULL) {
  1194     _ioproj_fallthrough = new (C, 1) ProjNode(call, TypeFunc::I_O);
  1195     transform_later(_ioproj_fallthrough);
  1196   } else if (!always_slow) {
  1197     for (DUIterator_Fast imax, i = _ioproj_fallthrough->fast_outs(imax); i < imax; i++) {
  1198       Node *use = _ioproj_fallthrough->fast_out(i);
  1200       _igvn.hash_delete(use);
  1201       imax -= replace_input(use, _ioproj_fallthrough, result_phi_i_o);
  1202       _igvn._worklist.push(use);
  1203       // back up iterator
  1204       --i;
  1207   // Now change uses of _ioproj_catchall to use _ioproj_fallthrough and delete _ioproj_catchall so
  1208   // we end up with a call that has only 1 control projection
  1209   if (_ioproj_catchall != NULL ) {
  1210     for (DUIterator_Fast imax, i = _ioproj_catchall->fast_outs(imax); i < imax; i++) {
  1211       Node *use = _ioproj_catchall->fast_out(i);
  1212       _igvn.hash_delete(use);
  1213       imax -= replace_input(use, _ioproj_catchall, _ioproj_fallthrough);
  1214       _igvn._worklist.push(use);
  1215       // back up iterator
  1216       --i;
  1220   // if we generated only a slow call, we are done
  1221   if (always_slow)
  1222     return;
  1225   if (_fallthroughcatchproj != NULL) {
  1226     ctrl = _fallthroughcatchproj->clone();
  1227     transform_later(ctrl);
  1228     _igvn.hash_delete(_fallthroughcatchproj);
  1229     _igvn.subsume_node(_fallthroughcatchproj, result_region);
  1230   } else {
  1231     ctrl = top();
  1233   Node *slow_result;
  1234   if (_resproj == NULL) {
  1235     // no uses of the allocation result
  1236     slow_result = top();
  1237   } else {
  1238     slow_result = _resproj->clone();
  1239     transform_later(slow_result);
  1240     _igvn.hash_delete(_resproj);
  1241     _igvn.subsume_node(_resproj, result_phi_rawoop);
  1244   // Plug slow-path into result merge point
  1245   result_region    ->init_req( slow_result_path, ctrl );
  1246   result_phi_rawoop->init_req( slow_result_path, slow_result);
  1247   result_phi_rawmem->init_req( slow_result_path, _memproj_fallthrough );
  1248   transform_later(result_region);
  1249   transform_later(result_phi_rawoop);
  1250   transform_later(result_phi_rawmem);
  1251   transform_later(result_phi_i_o);
  1252   // This completes all paths into the result merge point
  1256 // Helper for PhaseMacroExpand::expand_allocate_common.
  1257 // Initializes the newly-allocated storage.
  1258 Node*
  1259 PhaseMacroExpand::initialize_object(AllocateNode* alloc,
  1260                                     Node* control, Node* rawmem, Node* object,
  1261                                     Node* klass_node, Node* length,
  1262                                     Node* size_in_bytes) {
  1263   InitializeNode* init = alloc->initialization();
  1264   // Store the klass & mark bits
  1265   Node* mark_node = NULL;
  1266   // For now only enable fast locking for non-array types
  1267   if (UseBiasedLocking && (length == NULL)) {
  1268     mark_node = make_load(NULL, rawmem, klass_node, Klass::prototype_header_offset_in_bytes() + sizeof(oopDesc), TypeRawPtr::BOTTOM, T_ADDRESS);
  1269   } else {
  1270     mark_node = makecon(TypeRawPtr::make((address)markOopDesc::prototype()));
  1272   rawmem = make_store(control, rawmem, object, oopDesc::mark_offset_in_bytes(), mark_node, T_ADDRESS);
  1273   rawmem = make_store(control, rawmem, object, oopDesc::klass_offset_in_bytes(), klass_node, T_OBJECT);
  1274   int header_size = alloc->minimum_header_size();  // conservatively small
  1276   // Array length
  1277   if (length != NULL) {         // Arrays need length field
  1278     rawmem = make_store(control, rawmem, object, arrayOopDesc::length_offset_in_bytes(), length, T_INT);
  1279     // conservatively small header size:
  1280     header_size = sizeof(arrayOopDesc);
  1281     ciKlass* k = _igvn.type(klass_node)->is_klassptr()->klass();
  1282     if (k->is_array_klass())    // we know the exact header size in most cases:
  1283       header_size = Klass::layout_helper_header_size(k->layout_helper());
  1286   // Clear the object body, if necessary.
  1287   if (init == NULL) {
  1288     // The init has somehow disappeared; be cautious and clear everything.
  1289     //
  1290     // This can happen if a node is allocated but an uncommon trap occurs
  1291     // immediately.  In this case, the Initialize gets associated with the
  1292     // trap, and may be placed in a different (outer) loop, if the Allocate
  1293     // is in a loop.  If (this is rare) the inner loop gets unrolled, then
  1294     // there can be two Allocates to one Initialize.  The answer in all these
  1295     // edge cases is safety first.  It is always safe to clear immediately
  1296     // within an Allocate, and then (maybe or maybe not) clear some more later.
  1297     if (!ZeroTLAB)
  1298       rawmem = ClearArrayNode::clear_memory(control, rawmem, object,
  1299                                             header_size, size_in_bytes,
  1300                                             &_igvn);
  1301   } else {
  1302     if (!init->is_complete()) {
  1303       // Try to win by zeroing only what the init does not store.
  1304       // We can also try to do some peephole optimizations,
  1305       // such as combining some adjacent subword stores.
  1306       rawmem = init->complete_stores(control, rawmem, object,
  1307                                      header_size, size_in_bytes, &_igvn);
  1310     // We have no more use for this link, since the AllocateNode goes away:
  1311     init->set_req(InitializeNode::RawAddress, top());
  1312     // (If we keep the link, it just confuses the register allocator,
  1313     // who thinks he sees a real use of the address by the membar.)
  1316   return rawmem;
  1319 // Generate prefetch instructions for next allocations.
  1320 Node* PhaseMacroExpand::prefetch_allocation(Node* i_o, Node*& needgc_false,
  1321                                         Node*& contended_phi_rawmem,
  1322                                         Node* old_eden_top, Node* new_eden_top,
  1323                                         Node* length) {
  1324    if( UseTLAB && AllocatePrefetchStyle == 2 ) {
  1325       // Generate prefetch allocation with watermark check.
  1326       // As an allocation hits the watermark, we will prefetch starting
  1327       // at a "distance" away from watermark.
  1328       enum { fall_in_path = 1, pf_path = 2 };
  1330       Node *pf_region = new (C, 3) RegionNode(3);
  1331       Node *pf_phi_rawmem = new (C, 3) PhiNode( pf_region, Type::MEMORY,
  1332                                                 TypeRawPtr::BOTTOM );
  1333       // I/O is used for Prefetch
  1334       Node *pf_phi_abio = new (C, 3) PhiNode( pf_region, Type::ABIO );
  1336       Node *thread = new (C, 1) ThreadLocalNode();
  1337       transform_later(thread);
  1339       Node *eden_pf_adr = new (C, 4) AddPNode( top()/*not oop*/, thread,
  1340                    _igvn.MakeConX(in_bytes(JavaThread::tlab_pf_top_offset())) );
  1341       transform_later(eden_pf_adr);
  1343       Node *old_pf_wm = new (C, 3) LoadPNode( needgc_false,
  1344                                    contended_phi_rawmem, eden_pf_adr,
  1345                                    TypeRawPtr::BOTTOM, TypeRawPtr::BOTTOM );
  1346       transform_later(old_pf_wm);
  1348       // check against new_eden_top
  1349       Node *need_pf_cmp = new (C, 3) CmpPNode( new_eden_top, old_pf_wm );
  1350       transform_later(need_pf_cmp);
  1351       Node *need_pf_bol = new (C, 2) BoolNode( need_pf_cmp, BoolTest::ge );
  1352       transform_later(need_pf_bol);
  1353       IfNode *need_pf_iff = new (C, 2) IfNode( needgc_false, need_pf_bol,
  1354                                        PROB_UNLIKELY_MAG(4), COUNT_UNKNOWN );
  1355       transform_later(need_pf_iff);
  1357       // true node, add prefetchdistance
  1358       Node *need_pf_true = new (C, 1) IfTrueNode( need_pf_iff );
  1359       transform_later(need_pf_true);
  1361       Node *need_pf_false = new (C, 1) IfFalseNode( need_pf_iff );
  1362       transform_later(need_pf_false);
  1364       Node *new_pf_wmt = new (C, 4) AddPNode( top(), old_pf_wm,
  1365                                     _igvn.MakeConX(AllocatePrefetchDistance) );
  1366       transform_later(new_pf_wmt );
  1367       new_pf_wmt->set_req(0, need_pf_true);
  1369       Node *store_new_wmt = new (C, 4) StorePNode( need_pf_true,
  1370                                        contended_phi_rawmem, eden_pf_adr,
  1371                                        TypeRawPtr::BOTTOM, new_pf_wmt );
  1372       transform_later(store_new_wmt);
  1374       // adding prefetches
  1375       pf_phi_abio->init_req( fall_in_path, i_o );
  1377       Node *prefetch_adr;
  1378       Node *prefetch;
  1379       uint lines = AllocatePrefetchDistance / AllocatePrefetchStepSize;
  1380       uint step_size = AllocatePrefetchStepSize;
  1381       uint distance = 0;
  1383       for ( uint i = 0; i < lines; i++ ) {
  1384         prefetch_adr = new (C, 4) AddPNode( old_pf_wm, new_pf_wmt,
  1385                                             _igvn.MakeConX(distance) );
  1386         transform_later(prefetch_adr);
  1387         prefetch = new (C, 3) PrefetchWriteNode( i_o, prefetch_adr );
  1388         transform_later(prefetch);
  1389         distance += step_size;
  1390         i_o = prefetch;
  1392       pf_phi_abio->set_req( pf_path, i_o );
  1394       pf_region->init_req( fall_in_path, need_pf_false );
  1395       pf_region->init_req( pf_path, need_pf_true );
  1397       pf_phi_rawmem->init_req( fall_in_path, contended_phi_rawmem );
  1398       pf_phi_rawmem->init_req( pf_path, store_new_wmt );
  1400       transform_later(pf_region);
  1401       transform_later(pf_phi_rawmem);
  1402       transform_later(pf_phi_abio);
  1404       needgc_false = pf_region;
  1405       contended_phi_rawmem = pf_phi_rawmem;
  1406       i_o = pf_phi_abio;
  1407    } else if( AllocatePrefetchStyle > 0 ) {
  1408       // Insert a prefetch for each allocation only on the fast-path
  1409       Node *prefetch_adr;
  1410       Node *prefetch;
  1411       // Generate several prefetch instructions only for arrays.
  1412       uint lines = (length != NULL) ? AllocatePrefetchLines : 1;
  1413       uint step_size = AllocatePrefetchStepSize;
  1414       uint distance = AllocatePrefetchDistance;
  1415       for ( uint i = 0; i < lines; i++ ) {
  1416         prefetch_adr = new (C, 4) AddPNode( old_eden_top, new_eden_top,
  1417                                             _igvn.MakeConX(distance) );
  1418         transform_later(prefetch_adr);
  1419         prefetch = new (C, 3) PrefetchWriteNode( i_o, prefetch_adr );
  1420         // Do not let it float too high, since if eden_top == eden_end,
  1421         // both might be null.
  1422         if( i == 0 ) { // Set control for first prefetch, next follows it
  1423           prefetch->init_req(0, needgc_false);
  1425         transform_later(prefetch);
  1426         distance += step_size;
  1427         i_o = prefetch;
  1430    return i_o;
  1434 void PhaseMacroExpand::expand_allocate(AllocateNode *alloc) {
  1435   expand_allocate_common(alloc, NULL,
  1436                          OptoRuntime::new_instance_Type(),
  1437                          OptoRuntime::new_instance_Java());
  1440 void PhaseMacroExpand::expand_allocate_array(AllocateArrayNode *alloc) {
  1441   Node* length = alloc->in(AllocateNode::ALength);
  1442   expand_allocate_common(alloc, length,
  1443                          OptoRuntime::new_array_Type(),
  1444                          OptoRuntime::new_array_Java());
  1448 // we have determined that this lock/unlock can be eliminated, we simply
  1449 // eliminate the node without expanding it.
  1450 //
  1451 // Note:  The membar's associated with the lock/unlock are currently not
  1452 //        eliminated.  This should be investigated as a future enhancement.
  1453 //
  1454 bool PhaseMacroExpand::eliminate_locking_node(AbstractLockNode *alock) {
  1456   if (!alock->is_eliminated()) {
  1457     return false;
  1459   // Mark the box lock as eliminated if all correspondent locks are eliminated
  1460   // to construct correct debug info.
  1461   BoxLockNode* box = alock->box_node()->as_BoxLock();
  1462   if (!box->is_eliminated()) {
  1463     bool eliminate = true;
  1464     for (DUIterator_Fast imax, i = box->fast_outs(imax); i < imax; i++) {
  1465       Node *lck = box->fast_out(i);
  1466       if (lck->is_Lock() && !lck->as_AbstractLock()->is_eliminated()) {
  1467         eliminate = false;
  1468         break;
  1471     if (eliminate)
  1472       box->set_eliminated();
  1475   #ifndef PRODUCT
  1476   if (PrintEliminateLocks) {
  1477     if (alock->is_Lock()) {
  1478       tty->print_cr("++++ Eliminating: %d Lock", alock->_idx);
  1479     } else {
  1480       tty->print_cr("++++ Eliminating: %d Unlock", alock->_idx);
  1483   #endif
  1485   Node* mem  = alock->in(TypeFunc::Memory);
  1486   Node* ctrl = alock->in(TypeFunc::Control);
  1488   extract_call_projections(alock);
  1489   // There are 2 projections from the lock.  The lock node will
  1490   // be deleted when its last use is subsumed below.
  1491   assert(alock->outcnt() == 2 &&
  1492          _fallthroughproj != NULL &&
  1493          _memproj_fallthrough != NULL,
  1494          "Unexpected projections from Lock/Unlock");
  1496   Node* fallthroughproj = _fallthroughproj;
  1497   Node* memproj_fallthrough = _memproj_fallthrough;
  1499   // The memory projection from a lock/unlock is RawMem
  1500   // The input to a Lock is merged memory, so extract its RawMem input
  1501   // (unless the MergeMem has been optimized away.)
  1502   if (alock->is_Lock()) {
  1503     // Seach for MemBarAcquire node and delete it also.
  1504     MemBarNode* membar = fallthroughproj->unique_ctrl_out()->as_MemBar();
  1505     assert(membar != NULL && membar->Opcode() == Op_MemBarAcquire, "");
  1506     Node* ctrlproj = membar->proj_out(TypeFunc::Control);
  1507     Node* memproj = membar->proj_out(TypeFunc::Memory);
  1508     _igvn.hash_delete(ctrlproj);
  1509     _igvn.subsume_node(ctrlproj, fallthroughproj);
  1510     _igvn.hash_delete(memproj);
  1511     _igvn.subsume_node(memproj, memproj_fallthrough);
  1514   // Seach for MemBarRelease node and delete it also.
  1515   if (alock->is_Unlock() && ctrl != NULL && ctrl->is_Proj() &&
  1516       ctrl->in(0)->is_MemBar()) {
  1517     MemBarNode* membar = ctrl->in(0)->as_MemBar();
  1518     assert(membar->Opcode() == Op_MemBarRelease &&
  1519            mem->is_Proj() && membar == mem->in(0), "");
  1520     _igvn.hash_delete(fallthroughproj);
  1521     _igvn.subsume_node(fallthroughproj, ctrl);
  1522     _igvn.hash_delete(memproj_fallthrough);
  1523     _igvn.subsume_node(memproj_fallthrough, mem);
  1524     fallthroughproj = ctrl;
  1525     memproj_fallthrough = mem;
  1526     ctrl = membar->in(TypeFunc::Control);
  1527     mem  = membar->in(TypeFunc::Memory);
  1530   _igvn.hash_delete(fallthroughproj);
  1531   _igvn.subsume_node(fallthroughproj, ctrl);
  1532   _igvn.hash_delete(memproj_fallthrough);
  1533   _igvn.subsume_node(memproj_fallthrough, mem);
  1534   return true;
  1538 //------------------------------expand_lock_node----------------------
  1539 void PhaseMacroExpand::expand_lock_node(LockNode *lock) {
  1541   Node* ctrl = lock->in(TypeFunc::Control);
  1542   Node* mem = lock->in(TypeFunc::Memory);
  1543   Node* obj = lock->obj_node();
  1544   Node* box = lock->box_node();
  1545   Node* flock = lock->fastlock_node();
  1547   // Make the merge point
  1548   Node *region = new (C, 3) RegionNode(3);
  1550   Node *bol = transform_later(new (C, 2) BoolNode(flock,BoolTest::ne));
  1551   Node *iff = new (C, 2) IfNode( ctrl, bol, PROB_MIN, COUNT_UNKNOWN );
  1552   // Optimize test; set region slot 2
  1553   Node *slow_path = opt_iff(region,iff);
  1555   // Make slow path call
  1556   CallNode *call = make_slow_call( (CallNode *) lock, OptoRuntime::complete_monitor_enter_Type(), OptoRuntime::complete_monitor_locking_Java(), NULL, slow_path, obj, box );
  1558   extract_call_projections(call);
  1560   // Slow path can only throw asynchronous exceptions, which are always
  1561   // de-opted.  So the compiler thinks the slow-call can never throw an
  1562   // exception.  If it DOES throw an exception we would need the debug
  1563   // info removed first (since if it throws there is no monitor).
  1564   assert ( _ioproj_fallthrough == NULL && _ioproj_catchall == NULL &&
  1565            _memproj_catchall == NULL && _catchallcatchproj == NULL, "Unexpected projection from Lock");
  1567   // Capture slow path
  1568   // disconnect fall-through projection from call and create a new one
  1569   // hook up users of fall-through projection to region
  1570   Node *slow_ctrl = _fallthroughproj->clone();
  1571   transform_later(slow_ctrl);
  1572   _igvn.hash_delete(_fallthroughproj);
  1573   _fallthroughproj->disconnect_inputs(NULL);
  1574   region->init_req(1, slow_ctrl);
  1575   // region inputs are now complete
  1576   transform_later(region);
  1577   _igvn.subsume_node(_fallthroughproj, region);
  1579   // create a Phi for the memory state
  1580   Node *mem_phi = new (C, 3) PhiNode( region, Type::MEMORY, TypeRawPtr::BOTTOM);
  1581   Node *memproj = transform_later( new (C, 1) ProjNode(call, TypeFunc::Memory) );
  1582   mem_phi->init_req(1, memproj );
  1583   mem_phi->init_req(2, mem);
  1584   transform_later(mem_phi);
  1585     _igvn.hash_delete(_memproj_fallthrough);
  1586   _igvn.subsume_node(_memproj_fallthrough, mem_phi);
  1591 //------------------------------expand_unlock_node----------------------
  1592 void PhaseMacroExpand::expand_unlock_node(UnlockNode *unlock) {
  1594   Node* ctrl = unlock->in(TypeFunc::Control);
  1595   Node* mem = unlock->in(TypeFunc::Memory);
  1596   Node* obj = unlock->obj_node();
  1597   Node* box = unlock->box_node();
  1599   // No need for a null check on unlock
  1601   // Make the merge point
  1602   RegionNode *region = new (C, 3) RegionNode(3);
  1604   FastUnlockNode *funlock = new (C, 3) FastUnlockNode( ctrl, obj, box );
  1605   funlock = transform_later( funlock )->as_FastUnlock();
  1606   Node *bol = transform_later(new (C, 2) BoolNode(funlock,BoolTest::ne));
  1607   Node *iff = new (C, 2) IfNode( ctrl, bol, PROB_MIN, COUNT_UNKNOWN );
  1608   // Optimize test; set region slot 2
  1609   Node *slow_path = opt_iff(region,iff);
  1611   CallNode *call = make_slow_call( (CallNode *) unlock, OptoRuntime::complete_monitor_exit_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C), "complete_monitor_unlocking_C", slow_path, obj, box );
  1613   extract_call_projections(call);
  1615   assert ( _ioproj_fallthrough == NULL && _ioproj_catchall == NULL &&
  1616            _memproj_catchall == NULL && _catchallcatchproj == NULL, "Unexpected projection from Lock");
  1618   // No exceptions for unlocking
  1619   // Capture slow path
  1620   // disconnect fall-through projection from call and create a new one
  1621   // hook up users of fall-through projection to region
  1622   Node *slow_ctrl = _fallthroughproj->clone();
  1623   transform_later(slow_ctrl);
  1624   _igvn.hash_delete(_fallthroughproj);
  1625   _fallthroughproj->disconnect_inputs(NULL);
  1626   region->init_req(1, slow_ctrl);
  1627   // region inputs are now complete
  1628   transform_later(region);
  1629   _igvn.subsume_node(_fallthroughproj, region);
  1631   // create a Phi for the memory state
  1632   Node *mem_phi = new (C, 3) PhiNode( region, Type::MEMORY, TypeRawPtr::BOTTOM);
  1633   Node *memproj = transform_later( new(C, 1) ProjNode(call, TypeFunc::Memory) );
  1634   mem_phi->init_req(1, memproj );
  1635   mem_phi->init_req(2, mem);
  1636   transform_later(mem_phi);
  1637     _igvn.hash_delete(_memproj_fallthrough);
  1638   _igvn.subsume_node(_memproj_fallthrough, mem_phi);
  1643 //------------------------------expand_macro_nodes----------------------
  1644 //  Returns true if a failure occurred.
  1645 bool PhaseMacroExpand::expand_macro_nodes() {
  1646   if (C->macro_count() == 0)
  1647     return false;
  1648   // attempt to eliminate allocations
  1649   bool progress = true;
  1650   while (progress) {
  1651     progress = false;
  1652     for (int i = C->macro_count(); i > 0; i--) {
  1653       Node * n = C->macro_node(i-1);
  1654       bool success = false;
  1655       debug_only(int old_macro_count = C->macro_count(););
  1656       switch (n->class_id()) {
  1657       case Node::Class_Allocate:
  1658       case Node::Class_AllocateArray:
  1659         success = eliminate_allocate_node(n->as_Allocate());
  1660         break;
  1661       case Node::Class_Lock:
  1662       case Node::Class_Unlock:
  1663         success = eliminate_locking_node(n->as_AbstractLock());
  1664         break;
  1665       default:
  1666         assert(false, "unknown node type in macro list");
  1668       assert(success == (C->macro_count() < old_macro_count), "elimination reduces macro count");
  1669       progress = progress || success;
  1672   // Make sure expansion will not cause node limit to be exceeded.
  1673   // Worst case is a macro node gets expanded into about 50 nodes.
  1674   // Allow 50% more for optimization.
  1675   if (C->check_node_count(C->macro_count() * 75, "out of nodes before macro expansion" ) )
  1676     return true;
  1678   // expand "macro" nodes
  1679   // nodes are removed from the macro list as they are processed
  1680   while (C->macro_count() > 0) {
  1681     int macro_count = C->macro_count();
  1682     Node * n = C->macro_node(macro_count-1);
  1683     assert(n->is_macro(), "only macro nodes expected here");
  1684     if (_igvn.type(n) == Type::TOP || n->in(0)->is_top() ) {
  1685       // node is unreachable, so don't try to expand it
  1686       C->remove_macro_node(n);
  1687       continue;
  1689     switch (n->class_id()) {
  1690     case Node::Class_Allocate:
  1691       expand_allocate(n->as_Allocate());
  1692       break;
  1693     case Node::Class_AllocateArray:
  1694       expand_allocate_array(n->as_AllocateArray());
  1695       break;
  1696     case Node::Class_Lock:
  1697       expand_lock_node(n->as_Lock());
  1698       break;
  1699     case Node::Class_Unlock:
  1700       expand_unlock_node(n->as_Unlock());
  1701       break;
  1702     default:
  1703       assert(false, "unknown node type in macro list");
  1705     assert(C->macro_count() < macro_count, "must have deleted a node from macro list");
  1706     if (C->failing())  return true;
  1708   _igvn.optimize();
  1709   return false;

mercurial