1 /*
   2  * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "memory/allocation.inline.hpp"
  27 #include "opto/callnode.hpp"
  28 #include "opto/chaitin.hpp"
  29 #include "opto/live.hpp"
  30 #include "opto/machnode.hpp"
  31 
  32 
  33 // Compute live-in/live-out.  We use a totally incremental algorithm.  The LIVE
  34 // problem is monotonic.  The steady-state solution looks like this: pull a
  35 // block from the worklist.  It has a set of delta's - values which are newly
  36 // live-in from the block.  Push these to the live-out sets of all predecessor
  37 // blocks.  At each predecessor, the new live-out values are ANDed with what is
  38 // already live-out (extra stuff is added to the live-out sets).  Then the
  39 // remaining new live-out values are ANDed with what is locally defined.
  40 // Leftover bits become the new live-in for the predecessor block, and the pred
  41 // block is put on the worklist.
  42 //   The locally live-in stuff is computed once and added to predecessor
  43 // live-out sets.  This separate compilation is done in the outer loop below.
  44 PhaseLive::PhaseLive( const PhaseCFG &cfg, const LRG_List &names, Arena *arena ) : Phase(LIVE), _cfg(cfg), _names(names), _arena(arena), _live(0) {
  45 }
  46 
  47 void PhaseLive::compute(uint maxlrg) {
  48   _maxlrg   = maxlrg;
  49   _worklist = new (_arena) Block_List();
  50 
  51   // Init the sparse live arrays.  This data is live on exit from here!
  52   // The _live info is the live-out info.
  53   _live = (IndexSet*)_arena->Amalloc(sizeof(IndexSet) * _cfg.number_of_blocks());
  54   uint i;
  55   for (i = 0; i < _cfg.number_of_blocks(); i++) {
  56     _live[i].initialize(_maxlrg);
  57   }
  58 
  59   // Init the sparse arrays for delta-sets.
  60   ResourceMark rm;              // Nuke temp storage on exit
  61 
  62   // Does the memory used by _defs and _deltas get reclaimed?  Does it matter?  TT
  63 
  64   // Array of values defined locally in blocks
  65   _defs = NEW_RESOURCE_ARRAY(IndexSet,_cfg.number_of_blocks());
  66   for (i = 0; i < _cfg.number_of_blocks(); i++) {
  67     _defs[i].initialize(_maxlrg);
  68   }
  69 
  70   // Array of delta-set pointers, indexed by block pre_order-1.
  71   _deltas = NEW_RESOURCE_ARRAY(IndexSet*,_cfg.number_of_blocks());
  72   memset( _deltas, 0, sizeof(IndexSet*)* _cfg.number_of_blocks());
  73 
  74   _free_IndexSet = NULL;
  75 
  76   // Blocks having done pass-1
  77   VectorSet first_pass(Thread::current()->resource_area());
  78 
  79   // Outer loop: must compute local live-in sets and push into predecessors.
  80   for (uint j = _cfg.number_of_blocks(); j > 0; j--) {
  81     Block* block = _cfg.get_block(j - 1);
  82 
  83     // Compute the local live-in set.  Start with any new live-out bits.
  84     IndexSet* use = getset(block);
  85     IndexSet* def = &_defs[block->_pre_order-1];
  86     DEBUG_ONLY(IndexSet *def_outside = getfreeset();)
  87     uint i;
  88     for (i = block->number_of_nodes(); i > 1; i--) {
  89       Node* n = block->get_node(i-1);
  90       if (n->is_Phi()) {
  91         break;
  92       }
  93 
  94       uint r = _names.at(n->_idx);
  95       assert(!def_outside->member(r), "Use of external LRG overlaps the same LRG defined in this block");
  96       def->insert( r );
  97       use->remove( r );
  98       uint cnt = n->req();
  99       for (uint k = 1; k < cnt; k++) {
 100         Node *nk = n->in(k);
 101         uint nkidx = nk->_idx;
 102         if (_cfg.get_block_for_node(nk) != block) {
 103           uint u = _names.at(nkidx);
 104           use->insert(u);
 105           DEBUG_ONLY(def_outside->insert(u);)
 106         }
 107       }
 108     }
 109 #ifdef ASSERT
 110     def_outside->set_next(_free_IndexSet);
 111     _free_IndexSet = def_outside;     // Drop onto free list
 112 #endif
 113     // Remove anything defined by Phis and the block start instruction
 114     for (uint k = i; k > 0; k--) {
 115       uint r = _names.at(block->get_node(k - 1)->_idx);
 116       def->insert(r);
 117       use->remove(r);
 118     }
 119 
 120     // Push these live-in things to predecessors
 121     for (uint l = 1; l < block->num_preds(); l++) {
 122       Block* p = _cfg.get_block_for_node(block->pred(l));
 123       add_liveout(p, use, first_pass);
 124 
 125       // PhiNode uses go in the live-out set of prior blocks.
 126       for (uint k = i; k > 0; k--) {
 127         add_liveout(p, _names.at(block->get_node(k-1)->in(l)->_idx), first_pass);
 128       }
 129     }
 130     freeset(block);
 131     first_pass.set(block->_pre_order);
 132 
 133     // Inner loop: blocks that picked up new live-out values to be propagated
 134     while (_worklist->size()) {
 135       Block* block = _worklist->pop();
 136       IndexSet *delta = getset(block);
 137       assert( delta->count(), "missing delta set" );
 138 
 139       // Add new-live-in to predecessors live-out sets
 140       for (uint l = 1; l < block->num_preds(); l++) {
 141         Block* predecessor = _cfg.get_block_for_node(block->pred(l));
 142         add_liveout(predecessor, delta, first_pass);
 143       }
 144 
 145       freeset(block);
 146     } // End of while-worklist-not-empty
 147 
 148   } // End of for-all-blocks-outer-loop
 149 
 150   // We explicitly clear all of the IndexSets which we are about to release.
 151   // This allows us to recycle their internal memory into IndexSet's free list.
 152 
 153   for (i = 0; i < _cfg.number_of_blocks(); i++) {
 154     _defs[i].clear();
 155     if (_deltas[i]) {
 156       // Is this always true?
 157       _deltas[i]->clear();
 158     }
 159   }
 160   IndexSet *free = _free_IndexSet;
 161   while (free != NULL) {
 162     IndexSet *temp = free;
 163     free = free->next();
 164     temp->clear();
 165   }
 166 
 167 }
 168 
 169 #ifndef PRODUCT
 170 void PhaseLive::stats(uint iters) const {
 171 }
 172 #endif
 173 
 174 // Get an IndexSet for a block.  Return existing one, if any.  Make a new
 175 // empty one if a prior one does not exist.
 176 IndexSet *PhaseLive::getset( Block *p ) {
 177   IndexSet *delta = _deltas[p->_pre_order-1];
 178   if( !delta )                  // Not on worklist?
 179     // Get a free set; flag as being on worklist
 180     delta = _deltas[p->_pre_order-1] = getfreeset();
 181   return delta;                 // Return set of new live-out items
 182 }
 183 
 184 // Pull from free list, or allocate.  Internal allocation on the returned set
 185 // is always from thread local storage.
 186 IndexSet *PhaseLive::getfreeset( ) {
 187   IndexSet *f = _free_IndexSet;
 188   if( !f ) {
 189     f = new IndexSet;
 190 //    f->set_arena(Thread::current()->resource_area());
 191     f->initialize(_maxlrg, Thread::current()->resource_area());
 192   } else {
 193     // Pull from free list
 194     _free_IndexSet = f->next();
 195   //f->_cnt = 0;                        // Reset to empty
 196 //    f->set_arena(Thread::current()->resource_area());
 197     f->initialize(_maxlrg, Thread::current()->resource_area());
 198   }
 199   return f;
 200 }
 201 
 202 // Free an IndexSet from a block.
 203 void PhaseLive::freeset( const Block *p ) {
 204   IndexSet *f = _deltas[p->_pre_order-1];
 205   f->set_next(_free_IndexSet);
 206   _free_IndexSet = f;           // Drop onto free list
 207   _deltas[p->_pre_order-1] = NULL;
 208 }
 209 
 210 // Add a live-out value to a given blocks live-out set.  If it is new, then
 211 // also add it to the delta set and stick the block on the worklist.
 212 void PhaseLive::add_liveout( Block *p, uint r, VectorSet &first_pass ) {
 213   IndexSet *live = &_live[p->_pre_order-1];
 214   if( live->insert(r) ) {       // If actually inserted...
 215     // We extended the live-out set.  See if the value is generated locally.
 216     // If it is not, then we must extend the live-in set.
 217     if( !_defs[p->_pre_order-1].member( r ) ) {
 218       if( !_deltas[p->_pre_order-1] && // Not on worklist?
 219           first_pass.test(p->_pre_order) )
 220         _worklist->push(p);     // Actually go on worklist if already 1st pass
 221       getset(p)->insert(r);
 222     }
 223   }
 224 }
 225 
 226 // Add a vector of live-out values to a given blocks live-out set.
 227 void PhaseLive::add_liveout( Block *p, IndexSet *lo, VectorSet &first_pass ) {
 228   IndexSet *live = &_live[p->_pre_order-1];
 229   IndexSet *defs = &_defs[p->_pre_order-1];
 230   IndexSet *on_worklist = _deltas[p->_pre_order-1];
 231   IndexSet *delta = on_worklist ? on_worklist : getfreeset();
 232 
 233   IndexSetIterator elements(lo);
 234   uint r;
 235   while ((r = elements.next()) != 0) {
 236     if( live->insert(r) &&      // If actually inserted...
 237         !defs->member( r ) )    // and not defined locally
 238       delta->insert(r);         // Then add to live-in set
 239   }
 240 
 241   if( delta->count() ) {                // If actually added things
 242     _deltas[p->_pre_order-1] = delta; // Flag as on worklist now
 243     if( !on_worklist &&         // Not on worklist?
 244         first_pass.test(p->_pre_order) )
 245       _worklist->push(p);       // Actually go on worklist if already 1st pass
 246   } else {                      // Nothing there; just free it
 247     delta->set_next(_free_IndexSet);
 248     _free_IndexSet = delta;     // Drop onto free list
 249   }
 250 }
 251 
 252 #ifndef PRODUCT
 253 // Dump the live-out set for a block
 254 void PhaseLive::dump( const Block *b ) const {
 255   tty->print("Block %d: ",b->_pre_order);
 256   tty->print("LiveOut: ");  _live[b->_pre_order-1].dump();
 257   uint cnt = b->number_of_nodes();
 258   for( uint i=0; i<cnt; i++ ) {
 259     tty->print("L%d/", _names.at(b->get_node(i)->_idx));
 260     b->get_node(i)->dump();
 261   }
 262   tty->print("\n");
 263 }
 264 
 265 // Verify that base pointers and derived pointers are still sane.
 266 void PhaseChaitin::verify_base_ptrs( ResourceArea *a ) const {
 267 #ifdef ASSERT
 268   Unique_Node_List worklist(a);
 269   for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
 270     Block* block = _cfg.get_block(i);
 271     for (uint j = block->end_idx() + 1; j > 1; j--) {
 272       Node* n = block->get_node(j-1);
 273       if (n->is_Phi()) {
 274         break;
 275       }
 276       // Found a safepoint?
 277       if (n->is_MachSafePoint()) {
 278         MachSafePointNode *sfpt = n->as_MachSafePoint();
 279         JVMState* jvms = sfpt->jvms();
 280         if (jvms != NULL) {
 281           // Now scan for a live derived pointer
 282           if (jvms->oopoff() < sfpt->req()) {
 283             // Check each derived/base pair
 284             for (uint idx = jvms->oopoff(); idx < sfpt->req(); idx++) {
 285               Node *check = sfpt->in(idx);
 286               bool is_derived = ((idx - jvms->oopoff()) & 1) == 0;
 287               // search upwards through spills and spill phis for AddP
 288               worklist.clear();
 289               worklist.push(check);
 290               uint k = 0;
 291               while( k < worklist.size() ) {
 292                 check = worklist.at(k);
 293                 assert(check,"Bad base or derived pointer");
 294                 // See PhaseChaitin::find_base_for_derived() for all cases.
 295                 int isc = check->is_Copy();
 296                 if( isc ) {
 297                   worklist.push(check->in(isc));
 298                 } else if( check->is_Phi() ) {
 299                   for (uint m = 1; m < check->req(); m++)
 300                     worklist.push(check->in(m));
 301                 } else if( check->is_Con() ) {
 302                   if (is_derived) {
 303                     // Derived is NULL+offset
 304                     assert(!is_derived || check->bottom_type()->is_ptr()->ptr() == TypePtr::Null,"Bad derived pointer");
 305                   } else {
 306                     assert(check->bottom_type()->is_ptr()->_offset == 0,"Bad base pointer");
 307                     // Base either ConP(NULL) or loadConP
 308                     if (check->is_Mach()) {
 309                       assert(check->as_Mach()->ideal_Opcode() == Op_ConP,"Bad base pointer");
 310                     } else {
 311                       assert(check->Opcode() == Op_ConP &&
 312                              check->bottom_type()->is_ptr()->ptr() == TypePtr::Null,"Bad base pointer");
 313                     }
 314                   }
 315                 } else if( check->bottom_type()->is_ptr()->_offset == 0 ) {
 316                   if(check->is_Proj() || check->is_Mach() &&
 317                      (check->as_Mach()->ideal_Opcode() == Op_CreateEx ||
 318                       check->as_Mach()->ideal_Opcode() == Op_ThreadLocal ||
 319                       check->as_Mach()->ideal_Opcode() == Op_CMoveP ||
 320                       check->as_Mach()->ideal_Opcode() == Op_CheckCastPP ||
 321 #ifdef _LP64
 322                       UseCompressedOops && check->as_Mach()->ideal_Opcode() == Op_CastPP ||
 323                       UseCompressedOops && check->as_Mach()->ideal_Opcode() == Op_DecodeN ||
 324                       UseCompressedClassPointers && check->as_Mach()->ideal_Opcode() == Op_DecodeNKlass ||
 325 #endif
 326                       check->as_Mach()->ideal_Opcode() == Op_LoadP ||
 327                       check->as_Mach()->ideal_Opcode() == Op_LoadKlass)) {
 328                     // Valid nodes
 329                   } else {
 330                     check->dump();
 331                     assert(false,"Bad base or derived pointer");
 332                   }
 333                 } else {
 334                   assert(is_derived,"Bad base pointer");
 335                   assert(check->is_Mach() && check->as_Mach()->ideal_Opcode() == Op_AddP,"Bad derived pointer");
 336                 }
 337                 k++;
 338                 assert(k < 100000,"Derived pointer checking in infinite loop");
 339               } // End while
 340             }
 341           } // End of check for derived pointers
 342         } // End of Kcheck for debug info
 343       } // End of if found a safepoint
 344     } // End of forall instructions in block
 345   } // End of forall blocks
 346 #endif
 347 }
 348 
 349 // Verify that graphs and base pointers are still sane.
 350 void PhaseChaitin::verify( ResourceArea *a, bool verify_ifg ) const {
 351 #ifdef ASSERT
 352   if( VerifyOpto || VerifyRegisterAllocator ) {
 353     _cfg.verify();
 354     verify_base_ptrs(a);
 355     if(verify_ifg)
 356       _ifg->verify(this);
 357   }
 358 #endif
 359 }
 360 
 361 #endif