src/share/vm/opto/lcm.cpp

changeset 5509
d1034bd8cefc
parent 5111
70120f47d403
child 5539
adb9a7d94cb5
     1.1 --- a/src/share/vm/opto/lcm.cpp	Mon Aug 05 15:03:40 2013 -0700
     1.2 +++ b/src/share/vm/opto/lcm.cpp	Wed Aug 07 17:56:19 2013 +0200
     1.3 @@ -237,7 +237,7 @@
     1.4      }
     1.5  
     1.6      // Check ctrl input to see if the null-check dominates the memory op
     1.7 -    Block *cb = cfg->_bbs[mach->_idx];
     1.8 +    Block *cb = cfg->get_block_for_node(mach);
     1.9      cb = cb->_idom;             // Always hoist at least 1 block
    1.10      if( !was_store ) {          // Stores can be hoisted only one block
    1.11        while( cb->_dom_depth > (_dom_depth + 1))
    1.12 @@ -262,7 +262,7 @@
    1.13          if( is_decoden ) continue;
    1.14        }
    1.15        // Block of memory-op input
    1.16 -      Block *inb = cfg->_bbs[mach->in(j)->_idx];
    1.17 +      Block *inb = cfg->get_block_for_node(mach->in(j));
    1.18        Block *b = this;          // Start from nul check
    1.19        while( b != inb && b->_dom_depth > inb->_dom_depth )
    1.20          b = b->_idom;           // search upwards for input
    1.21 @@ -272,7 +272,7 @@
    1.22      }
    1.23      if( j > 0 )
    1.24        continue;
    1.25 -    Block *mb = cfg->_bbs[mach->_idx];
    1.26 +    Block *mb = cfg->get_block_for_node(mach);
    1.27      // Hoisting stores requires more checks for the anti-dependence case.
    1.28      // Give up hoisting if we have to move the store past any load.
    1.29      if( was_store ) {
    1.30 @@ -291,7 +291,7 @@
    1.31            break;                // Found anti-dependent load
    1.32          // Make sure control does not do a merge (would have to check allpaths)
    1.33          if( b->num_preds() != 2 ) break;
    1.34 -        b = cfg->_bbs[b->pred(1)->_idx]; // Move up to predecessor block
    1.35 +        b = cfg->get_block_for_node(b->pred(1)); // Move up to predecessor block
    1.36        }
    1.37        if( b != this ) continue;
    1.38      }
    1.39 @@ -303,15 +303,15 @@
    1.40  
    1.41      // Found a candidate!  Pick one with least dom depth - the highest
    1.42      // in the dom tree should be closest to the null check.
    1.43 -    if( !best ||
    1.44 -        cfg->_bbs[mach->_idx]->_dom_depth < cfg->_bbs[best->_idx]->_dom_depth ) {
    1.45 +    if (best == NULL || cfg->get_block_for_node(mach)->_dom_depth < cfg->get_block_for_node(best)->_dom_depth) {
    1.46        best = mach;
    1.47        bidx = vidx;
    1.48 -
    1.49      }
    1.50    }
    1.51    // No candidate!
    1.52 -  if( !best ) return;
    1.53 +  if (best == NULL) {
    1.54 +    return;
    1.55 +  }
    1.56  
    1.57    // ---- Found an implicit null check
    1.58    extern int implicit_null_checks;
    1.59 @@ -319,29 +319,29 @@
    1.60  
    1.61    if( is_decoden ) {
    1.62      // Check if we need to hoist decodeHeapOop_not_null first.
    1.63 -    Block *valb = cfg->_bbs[val->_idx];
    1.64 +    Block *valb = cfg->get_block_for_node(val);
    1.65      if( this != valb && this->_dom_depth < valb->_dom_depth ) {
    1.66        // Hoist it up to the end of the test block.
    1.67        valb->find_remove(val);
    1.68        this->add_inst(val);
    1.69 -      cfg->_bbs.map(val->_idx,this);
    1.70 +      cfg->map_node_to_block(val, this);
    1.71        // DecodeN on x86 may kill flags. Check for flag-killing projections
    1.72        // that also need to be hoisted.
    1.73        for (DUIterator_Fast jmax, j = val->fast_outs(jmax); j < jmax; j++) {
    1.74          Node* n = val->fast_out(j);
    1.75          if( n->is_MachProj() ) {
    1.76 -          cfg->_bbs[n->_idx]->find_remove(n);
    1.77 +          cfg->get_block_for_node(n)->find_remove(n);
    1.78            this->add_inst(n);
    1.79 -          cfg->_bbs.map(n->_idx,this);
    1.80 +          cfg->map_node_to_block(n, this);
    1.81          }
    1.82        }
    1.83      }
    1.84    }
    1.85    // Hoist the memory candidate up to the end of the test block.
    1.86 -  Block *old_block = cfg->_bbs[best->_idx];
    1.87 +  Block *old_block = cfg->get_block_for_node(best);
    1.88    old_block->find_remove(best);
    1.89    add_inst(best);
    1.90 -  cfg->_bbs.map(best->_idx,this);
    1.91 +  cfg->map_node_to_block(best, this);
    1.92  
    1.93    // Move the control dependence
    1.94    if (best->in(0) && best->in(0) == old_block->_nodes[0])
    1.95 @@ -352,9 +352,9 @@
    1.96    for (DUIterator_Fast jmax, j = best->fast_outs(jmax); j < jmax; j++) {
    1.97      Node* n = best->fast_out(j);
    1.98      if( n->is_MachProj() ) {
    1.99 -      cfg->_bbs[n->_idx]->find_remove(n);
   1.100 +      cfg->get_block_for_node(n)->find_remove(n);
   1.101        add_inst(n);
   1.102 -      cfg->_bbs.map(n->_idx,this);
   1.103 +      cfg->map_node_to_block(n, this);
   1.104      }
   1.105    }
   1.106  
   1.107 @@ -385,7 +385,7 @@
   1.108    Node *old_tst = proj->in(0);
   1.109    MachNode *nul_chk = new (C) MachNullCheckNode(old_tst->in(0),best,bidx);
   1.110    _nodes.map(end_idx(),nul_chk);
   1.111 -  cfg->_bbs.map(nul_chk->_idx,this);
   1.112 +  cfg->map_node_to_block(nul_chk, this);
   1.113    // Redirect users of old_test to nul_chk
   1.114    for (DUIterator_Last i2min, i2 = old_tst->last_outs(i2min); i2 >= i2min; --i2)
   1.115      old_tst->last_out(i2)->set_req(0, nul_chk);
   1.116 @@ -468,7 +468,7 @@
   1.117          Node* use = n->fast_out(j);
   1.118  
   1.119          // The use is a conditional branch, make them adjacent
   1.120 -        if (use->is_MachIf() && cfg->_bbs[use->_idx]==this ) {
   1.121 +        if (use->is_MachIf() && cfg->get_block_for_node(use) == this) {
   1.122            found_machif = true;
   1.123            break;
   1.124          }
   1.125 @@ -529,13 +529,14 @@
   1.126  
   1.127  
   1.128  //------------------------------set_next_call----------------------------------
   1.129 -void Block::set_next_call( Node *n, VectorSet &next_call, Block_Array &bbs ) {
   1.130 +void Block::set_next_call( Node *n, VectorSet &next_call, PhaseCFG* cfg) {
   1.131    if( next_call.test_set(n->_idx) ) return;
   1.132    for( uint i=0; i<n->len(); i++ ) {
   1.133      Node *m = n->in(i);
   1.134      if( !m ) continue;  // must see all nodes in block that precede call
   1.135 -    if( bbs[m->_idx] == this )
   1.136 -      set_next_call( m, next_call, bbs );
   1.137 +    if (cfg->get_block_for_node(m) == this) {
   1.138 +      set_next_call(m, next_call, cfg);
   1.139 +    }
   1.140    }
   1.141  }
   1.142  
   1.143 @@ -545,12 +546,12 @@
   1.144  // next subroutine call get priority - basically it moves things NOT needed
   1.145  // for the next call till after the call.  This prevents me from trying to
   1.146  // carry lots of stuff live across a call.
   1.147 -void Block::needed_for_next_call(Node *this_call, VectorSet &next_call, Block_Array &bbs) {
   1.148 +void Block::needed_for_next_call(Node *this_call, VectorSet &next_call, PhaseCFG* cfg) {
   1.149    // Find the next control-defining Node in this block
   1.150    Node* call = NULL;
   1.151    for (DUIterator_Fast imax, i = this_call->fast_outs(imax); i < imax; i++) {
   1.152      Node* m = this_call->fast_out(i);
   1.153 -    if( bbs[m->_idx] == this && // Local-block user
   1.154 +    if(cfg->get_block_for_node(m) == this && // Local-block user
   1.155          m != this_call &&       // Not self-start node
   1.156          m->is_MachCall() )
   1.157        call = m;
   1.158 @@ -558,7 +559,7 @@
   1.159    }
   1.160    if (call == NULL)  return;    // No next call (e.g., block end is near)
   1.161    // Set next-call for all inputs to this call
   1.162 -  set_next_call(call, next_call, bbs);
   1.163 +  set_next_call(call, next_call, cfg);
   1.164  }
   1.165  
   1.166  //------------------------------add_call_kills-------------------------------------
   1.167 @@ -578,7 +579,7 @@
   1.168  
   1.169  
   1.170  //------------------------------sched_call-------------------------------------
   1.171 -uint Block::sched_call( Matcher &matcher, Block_Array &bbs, uint node_cnt, Node_List &worklist, GrowableArray<int> &ready_cnt, MachCallNode *mcall, VectorSet &next_call ) {
   1.172 +uint Block::sched_call( Matcher &matcher, PhaseCFG* cfg, uint node_cnt, Node_List &worklist, GrowableArray<int> &ready_cnt, MachCallNode *mcall, VectorSet &next_call ) {
   1.173    RegMask regs;
   1.174  
   1.175    // Schedule all the users of the call right now.  All the users are
   1.176 @@ -597,12 +598,14 @@
   1.177      // Check for scheduling the next control-definer
   1.178      if( n->bottom_type() == Type::CONTROL )
   1.179        // Warm up next pile of heuristic bits
   1.180 -      needed_for_next_call(n, next_call, bbs);
   1.181 +      needed_for_next_call(n, next_call, cfg);
   1.182  
   1.183      // Children of projections are now all ready
   1.184      for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
   1.185        Node* m = n->fast_out(j); // Get user
   1.186 -      if( bbs[m->_idx] != this ) continue;
   1.187 +      if(cfg->get_block_for_node(m) != this) {
   1.188 +        continue;
   1.189 +      }
   1.190        if( m->is_Phi() ) continue;
   1.191        int m_cnt = ready_cnt.at(m->_idx)-1;
   1.192        ready_cnt.at_put(m->_idx, m_cnt);
   1.193 @@ -620,7 +623,7 @@
   1.194    uint r_cnt = mcall->tf()->range()->cnt();
   1.195    int op = mcall->ideal_Opcode();
   1.196    MachProjNode *proj = new (matcher.C) MachProjNode( mcall, r_cnt+1, RegMask::Empty, MachProjNode::fat_proj );
   1.197 -  bbs.map(proj->_idx,this);
   1.198 +  cfg->map_node_to_block(proj, this);
   1.199    _nodes.insert(node_cnt++, proj);
   1.200  
   1.201    // Select the right register save policy.
   1.202 @@ -708,7 +711,7 @@
   1.203        uint local = 0;
   1.204        for( uint j=0; j<cnt; j++ ) {
   1.205          Node *m = n->in(j);
   1.206 -        if( m && cfg->_bbs[m->_idx] == this && !m->is_top() )
   1.207 +        if( m && cfg->get_block_for_node(m) == this && !m->is_top() )
   1.208            local++;              // One more block-local input
   1.209        }
   1.210        ready_cnt.at_put(n->_idx, local); // Count em up
   1.211 @@ -720,7 +723,7 @@
   1.212            for (uint prec = n->req(); prec < n->len(); prec++) {
   1.213              Node* oop_store = n->in(prec);
   1.214              if (oop_store != NULL) {
   1.215 -              assert(cfg->_bbs[oop_store->_idx]->_dom_depth <= this->_dom_depth, "oop_store must dominate card-mark");
   1.216 +              assert(cfg->get_block_for_node(oop_store)->_dom_depth <= this->_dom_depth, "oop_store must dominate card-mark");
   1.217              }
   1.218            }
   1.219          }
   1.220 @@ -753,7 +756,7 @@
   1.221      Node *n = _nodes[i3];       // Get pre-scheduled
   1.222      for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
   1.223        Node* m = n->fast_out(j);
   1.224 -      if( cfg->_bbs[m->_idx] ==this ) { // Local-block user
   1.225 +      if (cfg->get_block_for_node(m) == this) { // Local-block user
   1.226          int m_cnt = ready_cnt.at(m->_idx)-1;
   1.227          ready_cnt.at_put(m->_idx, m_cnt);   // Fix ready count
   1.228        }
   1.229 @@ -786,7 +789,7 @@
   1.230    }
   1.231  
   1.232    // Warm up the 'next_call' heuristic bits
   1.233 -  needed_for_next_call(_nodes[0], next_call, cfg->_bbs);
   1.234 +  needed_for_next_call(_nodes[0], next_call, cfg);
   1.235  
   1.236  #ifndef PRODUCT
   1.237      if (cfg->trace_opto_pipelining()) {
   1.238 @@ -837,7 +840,7 @@
   1.239  #endif
   1.240      if( n->is_MachCall() ) {
   1.241        MachCallNode *mcall = n->as_MachCall();
   1.242 -      phi_cnt = sched_call(matcher, cfg->_bbs, phi_cnt, worklist, ready_cnt, mcall, next_call);
   1.243 +      phi_cnt = sched_call(matcher, cfg, phi_cnt, worklist, ready_cnt, mcall, next_call);
   1.244        continue;
   1.245      }
   1.246  
   1.247 @@ -847,7 +850,7 @@
   1.248        regs.OR(n->out_RegMask());
   1.249  
   1.250        MachProjNode *proj = new (matcher.C) MachProjNode( n, 1, RegMask::Empty, MachProjNode::fat_proj );
   1.251 -      cfg->_bbs.map(proj->_idx,this);
   1.252 +      cfg->map_node_to_block(proj, this);
   1.253        _nodes.insert(phi_cnt++, proj);
   1.254  
   1.255        add_call_kills(proj, regs, matcher._c_reg_save_policy, false);
   1.256 @@ -856,7 +859,9 @@
   1.257      // Children are now all ready
   1.258      for (DUIterator_Fast i5max, i5 = n->fast_outs(i5max); i5 < i5max; i5++) {
   1.259        Node* m = n->fast_out(i5); // Get user
   1.260 -      if( cfg->_bbs[m->_idx] != this ) continue;
   1.261 +      if (cfg->get_block_for_node(m) != this) {
   1.262 +        continue;
   1.263 +      }
   1.264        if( m->is_Phi() ) continue;
   1.265        if (m->_idx >= max_idx) { // new node, skip it
   1.266          assert(m->is_MachProj() && n->is_Mach() && n->as_Mach()->has_call(), "unexpected node types");
   1.267 @@ -914,7 +919,7 @@
   1.268  }
   1.269  
   1.270  //------------------------------catch_cleanup_find_cloned_def------------------
   1.271 -static Node *catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def_blk, Block_Array &bbs, int n_clone_idx) {
   1.272 +static Node *catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def_blk, PhaseCFG* cfg, int n_clone_idx) {
   1.273    assert( use_blk != def_blk, "Inter-block cleanup only");
   1.274  
   1.275    // The use is some block below the Catch.  Find and return the clone of the def
   1.276 @@ -940,7 +945,8 @@
   1.277      // PhiNode, the PhiNode uses from the def and IT's uses need fixup.
   1.278      Node_Array inputs = new Node_List(Thread::current()->resource_area());
   1.279      for(uint k = 1; k < use_blk->num_preds(); k++) {
   1.280 -      inputs.map(k, catch_cleanup_find_cloned_def(bbs[use_blk->pred(k)->_idx], def, def_blk, bbs, n_clone_idx));
   1.281 +      Block* block = cfg->get_block_for_node(use_blk->pred(k));
   1.282 +      inputs.map(k, catch_cleanup_find_cloned_def(block, def, def_blk, cfg, n_clone_idx));
   1.283      }
   1.284  
   1.285      // Check to see if the use_blk already has an identical phi inserted.
   1.286 @@ -962,7 +968,7 @@
   1.287      if (fixup == NULL) {
   1.288        Node *new_phi = PhiNode::make(use_blk->head(), def);
   1.289        use_blk->_nodes.insert(1, new_phi);
   1.290 -      bbs.map(new_phi->_idx, use_blk);
   1.291 +      cfg->map_node_to_block(new_phi, use_blk);
   1.292        for (uint k = 1; k < use_blk->num_preds(); k++) {
   1.293          new_phi->set_req(k, inputs[k]);
   1.294        }
   1.295 @@ -1002,17 +1008,17 @@
   1.296  //------------------------------catch_cleanup_inter_block---------------------
   1.297  // Fix all input edges in use that reference "def".  The use is in a different
   1.298  // block than the def.
   1.299 -static void catch_cleanup_inter_block(Node *use, Block *use_blk, Node *def, Block *def_blk, Block_Array &bbs, int n_clone_idx) {
   1.300 +static void catch_cleanup_inter_block(Node *use, Block *use_blk, Node *def, Block *def_blk, PhaseCFG* cfg, int n_clone_idx) {
   1.301    if( !use_blk ) return;        // Can happen if the use is a precedence edge
   1.302  
   1.303 -  Node *new_def = catch_cleanup_find_cloned_def(use_blk, def, def_blk, bbs, n_clone_idx);
   1.304 +  Node *new_def = catch_cleanup_find_cloned_def(use_blk, def, def_blk, cfg, n_clone_idx);
   1.305    catch_cleanup_fix_all_inputs(use, def, new_def);
   1.306  }
   1.307  
   1.308  //------------------------------call_catch_cleanup-----------------------------
   1.309  // If we inserted any instructions between a Call and his CatchNode,
   1.310  // clone the instructions on all paths below the Catch.
   1.311 -void Block::call_catch_cleanup(Block_Array &bbs, Compile* C) {
   1.312 +void Block::call_catch_cleanup(PhaseCFG* cfg, Compile* C) {
   1.313  
   1.314    // End of region to clone
   1.315    uint end = end_idx();
   1.316 @@ -1037,7 +1043,7 @@
   1.317        // since clones dominate on each path.
   1.318        Node *clone = _nodes[j-1]->clone();
   1.319        sb->_nodes.insert( 1, clone );
   1.320 -      bbs.map(clone->_idx,sb);
   1.321 +      cfg->map_node_to_block(clone, sb);
   1.322      }
   1.323    }
   1.324  
   1.325 @@ -1054,18 +1060,19 @@
   1.326      uint max = out->size();
   1.327      for (uint j = 0; j < max; j++) {// For all users
   1.328        Node *use = out->pop();
   1.329 -      Block *buse = bbs[use->_idx];
   1.330 +      Block *buse = cfg->get_block_for_node(use);
   1.331        if( use->is_Phi() ) {
   1.332          for( uint k = 1; k < use->req(); k++ )
   1.333            if( use->in(k) == n ) {
   1.334 -            Node *fixup = catch_cleanup_find_cloned_def(bbs[buse->pred(k)->_idx], n, this, bbs, n_clone_idx);
   1.335 +            Block* block = cfg->get_block_for_node(buse->pred(k));
   1.336 +            Node *fixup = catch_cleanup_find_cloned_def(block, n, this, cfg, n_clone_idx);
   1.337              use->set_req(k, fixup);
   1.338            }
   1.339        } else {
   1.340          if (this == buse) {
   1.341            catch_cleanup_intra_block(use, n, this, beg, n_clone_idx);
   1.342          } else {
   1.343 -          catch_cleanup_inter_block(use, buse, n, this, bbs, n_clone_idx);
   1.344 +          catch_cleanup_inter_block(use, buse, n, this, cfg, n_clone_idx);
   1.345          }
   1.346        }
   1.347      } // End for all users

mercurial