1.1 --- a/src/share/vm/opto/lcm.cpp Wed Aug 28 11:22:43 2013 +0200
1.2 +++ b/src/share/vm/opto/lcm.cpp Sun Sep 01 19:21:05 2013 +0200
1.3 @@ -58,14 +58,14 @@
1.4 // The proj is the control projection for the not-null case.
1.5 // The val is the pointer being checked for nullness or
1.6 // decodeHeapOop_not_null node if it did not fold into address.
1.7 -void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowed_reasons) {
1.8 +void PhaseCFG::implicit_null_check(Block* block, Node *proj, Node *val, int allowed_reasons) {
1.9 // Assume if null check need for 0 offset then always needed
1.10 // Intel solaris doesn't support any null checks yet and no
1.11 // mechanism exists (yet) to set the switches at an os_cpu level
1.12 if( !ImplicitNullChecks || MacroAssembler::needs_explicit_null_check(0)) return;
1.13
1.14 // Make sure the ptr-is-null path appears to be uncommon!
1.15 - float f = end()->as_MachIf()->_prob;
1.16 + float f = block->end()->as_MachIf()->_prob;
1.17 if( proj->Opcode() == Op_IfTrue ) f = 1.0f - f;
1.18 if( f > PROB_UNLIKELY_MAG(4) ) return;
1.19
1.20 @@ -75,13 +75,13 @@
1.21 // Get the successor block for if the test ptr is non-null
1.22 Block* not_null_block; // this one goes with the proj
1.23 Block* null_block;
1.24 - if (get_node(number_of_nodes()-1) == proj) {
1.25 - null_block = _succs[0];
1.26 - not_null_block = _succs[1];
1.27 + if (block->get_node(block->number_of_nodes()-1) == proj) {
1.28 + null_block = block->_succs[0];
1.29 + not_null_block = block->_succs[1];
1.30 } else {
1.31 - assert(get_node(number_of_nodes()-2) == proj, "proj is one or the other");
1.32 - not_null_block = _succs[0];
1.33 - null_block = _succs[1];
1.34 + assert(block->get_node(block->number_of_nodes()-2) == proj, "proj is one or the other");
1.35 + not_null_block = block->_succs[0];
1.36 + null_block = block->_succs[1];
1.37 }
1.38 while (null_block->is_Empty() == Block::empty_with_goto) {
1.39 null_block = null_block->_succs[0];
1.40 @@ -93,7 +93,7 @@
1.41 // detect failure of this optimization, as in 6366351.)
1.42 {
1.43 bool found_trap = false;
1.44 - for (uint i1 = 0; i1 < null_block->_nodes.size(); i1++) {
1.45 + for (uint i1 = 0; i1 < null_block->number_of_nodes(); i1++) {
1.46 Node* nn = null_block->get_node(i1);
1.47 if (nn->is_MachCall() &&
1.48 nn->as_MachCall()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point()) {
1.49 @@ -237,20 +237,20 @@
1.50 }
1.51
1.52 // Check ctrl input to see if the null-check dominates the memory op
1.53 - Block *cb = cfg->get_block_for_node(mach);
1.54 + Block *cb = get_block_for_node(mach);
1.55 cb = cb->_idom; // Always hoist at least 1 block
1.56 if( !was_store ) { // Stores can be hoisted only one block
1.57 - while( cb->_dom_depth > (_dom_depth + 1))
1.58 + while( cb->_dom_depth > (block->_dom_depth + 1))
1.59 cb = cb->_idom; // Hoist loads as far as we want
1.60 // The non-null-block should dominate the memory op, too. Live
1.61 // range spilling will insert a spill in the non-null-block if it is
1.62 // needs to spill the memory op for an implicit null check.
1.63 - if (cb->_dom_depth == (_dom_depth + 1)) {
1.64 + if (cb->_dom_depth == (block->_dom_depth + 1)) {
1.65 if (cb != not_null_block) continue;
1.66 cb = cb->_idom;
1.67 }
1.68 }
1.69 - if( cb != this ) continue;
1.70 + if( cb != block ) continue;
1.71
1.72 // Found a memory user; see if it can be hoisted to check-block
1.73 uint vidx = 0; // Capture index of value into memop
1.74 @@ -262,8 +262,8 @@
1.75 if( is_decoden ) continue;
1.76 }
1.77 // Block of memory-op input
1.78 - Block *inb = cfg->get_block_for_node(mach->in(j));
1.79 - Block *b = this; // Start from nul check
1.80 + Block *inb = get_block_for_node(mach->in(j));
1.81 + Block *b = block; // Start from nul check
1.82 while( b != inb && b->_dom_depth > inb->_dom_depth )
1.83 b = b->_idom; // search upwards for input
1.84 // See if input dominates null check
1.85 @@ -272,28 +272,28 @@
1.86 }
1.87 if( j > 0 )
1.88 continue;
1.89 - Block *mb = cfg->get_block_for_node(mach);
1.90 + Block *mb = get_block_for_node(mach);
1.91 // Hoisting stores requires more checks for the anti-dependence case.
1.92 // Give up hoisting if we have to move the store past any load.
1.93 if( was_store ) {
1.94 Block *b = mb; // Start searching here for a local load
1.95 // mach use (faulting) trying to hoist
1.96 // n might be blocker to hoisting
1.97 - while( b != this ) {
1.98 + while( b != block ) {
1.99 uint k;
1.100 - for( k = 1; k < b->_nodes.size(); k++ ) {
1.101 + for( k = 1; k < b->number_of_nodes(); k++ ) {
1.102 Node *n = b->get_node(k);
1.103 if( n->needs_anti_dependence_check() &&
1.104 n->in(LoadNode::Memory) == mach->in(StoreNode::Memory) )
1.105 break; // Found anti-dependent load
1.106 }
1.107 - if( k < b->_nodes.size() )
1.108 + if( k < b->number_of_nodes() )
1.109 break; // Found anti-dependent load
1.110 // Make sure control does not do a merge (would have to check allpaths)
1.111 if( b->num_preds() != 2 ) break;
1.112 - b = cfg->get_block_for_node(b->pred(1)); // Move up to predecessor block
1.113 + b = get_block_for_node(b->pred(1)); // Move up to predecessor block
1.114 }
1.115 - if( b != this ) continue;
1.116 + if( b != block ) continue;
1.117 }
1.118
1.119 // Make sure this memory op is not already being used for a NullCheck
1.120 @@ -303,7 +303,7 @@
1.121
1.122 // Found a candidate! Pick one with least dom depth - the highest
1.123 // in the dom tree should be closest to the null check.
1.124 - if (best == NULL || cfg->get_block_for_node(mach)->_dom_depth < cfg->get_block_for_node(best)->_dom_depth) {
1.125 + if (best == NULL || get_block_for_node(mach)->_dom_depth < get_block_for_node(best)->_dom_depth) {
1.126 best = mach;
1.127 bidx = vidx;
1.128 }
1.129 @@ -319,46 +319,45 @@
1.130
1.131 if( is_decoden ) {
1.132 // Check if we need to hoist decodeHeapOop_not_null first.
1.133 - Block *valb = cfg->get_block_for_node(val);
1.134 - if( this != valb && this->_dom_depth < valb->_dom_depth ) {
1.135 + Block *valb = get_block_for_node(val);
1.136 + if( block != valb && block->_dom_depth < valb->_dom_depth ) {
1.137 // Hoist it up to the end of the test block.
1.138 valb->find_remove(val);
1.139 - this->add_inst(val);
1.140 - cfg->map_node_to_block(val, this);
1.141 + block->add_inst(val);
1.142 + map_node_to_block(val, block);
1.143 // DecodeN on x86 may kill flags. Check for flag-killing projections
1.144 // that also need to be hoisted.
1.145 for (DUIterator_Fast jmax, j = val->fast_outs(jmax); j < jmax; j++) {
1.146 Node* n = val->fast_out(j);
1.147 if( n->is_MachProj() ) {
1.148 - cfg->get_block_for_node(n)->find_remove(n);
1.149 - this->add_inst(n);
1.150 - cfg->map_node_to_block(n, this);
1.151 + get_block_for_node(n)->find_remove(n);
1.152 + block->add_inst(n);
1.153 + map_node_to_block(n, block);
1.154 }
1.155 }
1.156 }
1.157 }
1.158 // Hoist the memory candidate up to the end of the test block.
1.159 - Block *old_block = cfg->get_block_for_node(best);
1.160 + Block *old_block = get_block_for_node(best);
1.161 old_block->find_remove(best);
1.162 - add_inst(best);
1.163 - cfg->map_node_to_block(best, this);
1.164 + block->add_inst(best);
1.165 + map_node_to_block(best, block);
1.166
1.167 // Move the control dependence
1.168 if (best->in(0) && best->in(0) == old_block->head())
1.169 - best->set_req(0, head());
1.170 + best->set_req(0, block->head());
1.171
1.172 // Check for flag-killing projections that also need to be hoisted
1.173 // Should be DU safe because no edge updates.
1.174 for (DUIterator_Fast jmax, j = best->fast_outs(jmax); j < jmax; j++) {
1.175 Node* n = best->fast_out(j);
1.176 if( n->is_MachProj() ) {
1.177 - cfg->get_block_for_node(n)->find_remove(n);
1.178 - add_inst(n);
1.179 - cfg->map_node_to_block(n, this);
1.180 + get_block_for_node(n)->find_remove(n);
1.181 + block->add_inst(n);
1.182 + map_node_to_block(n, block);
1.183 }
1.184 }
1.185
1.186 - Compile *C = cfg->C;
1.187 // proj==Op_True --> ne test; proj==Op_False --> eq test.
1.188 // One of two graph shapes got matched:
1.189 // (IfTrue (If (Bool NE (CmpP ptr NULL))))
1.190 @@ -368,10 +367,10 @@
1.191 // We need to flip the projections to keep the same semantics.
1.192 if( proj->Opcode() == Op_IfTrue ) {
1.193 // Swap order of projections in basic block to swap branch targets
1.194 - Node *tmp1 = get_node(end_idx()+1);
1.195 - Node *tmp2 = get_node(end_idx()+2);
1.196 - _nodes.map(end_idx()+1, tmp2);
1.197 - _nodes.map(end_idx()+2, tmp1);
1.198 + Node *tmp1 = block->get_node(block->end_idx()+1);
1.199 + Node *tmp2 = block->get_node(block->end_idx()+2);
1.200 + block->map_node(tmp2, block->end_idx()+1);
1.201 + block->map_node(tmp1, block->end_idx()+2);
1.202 Node *tmp = new (C) Node(C->top()); // Use not NULL input
1.203 tmp1->replace_by(tmp);
1.204 tmp2->replace_by(tmp1);
1.205 @@ -384,8 +383,8 @@
1.206 // it as well.
1.207 Node *old_tst = proj->in(0);
1.208 MachNode *nul_chk = new (C) MachNullCheckNode(old_tst->in(0),best,bidx);
1.209 - _nodes.map(end_idx(),nul_chk);
1.210 - cfg->map_node_to_block(nul_chk, this);
1.211 + block->map_node(nul_chk, block->end_idx());
1.212 + map_node_to_block(nul_chk, block);
1.213 // Redirect users of old_test to nul_chk
1.214 for (DUIterator_Last i2min, i2 = old_tst->last_outs(i2min); i2 >= i2min; --i2)
1.215 old_tst->last_out(i2)->set_req(0, nul_chk);
1.216 @@ -393,8 +392,8 @@
1.217 for (uint i3 = 0; i3 < old_tst->req(); i3++)
1.218 old_tst->set_req(i3, NULL);
1.219
1.220 - cfg->latency_from_uses(nul_chk);
1.221 - cfg->latency_from_uses(best);
1.222 + latency_from_uses(nul_chk);
1.223 + latency_from_uses(best);
1.224 }
1.225
1.226
1.227 @@ -408,7 +407,7 @@
1.228 // remaining cases (most), choose the instruction with the greatest latency
1.229 // (that is, the most number of pseudo-cycles required to the end of the
1.230 // routine). If there is a tie, choose the instruction with the most inputs.
1.231 -Node *Block::select(PhaseCFG *cfg, Node_List &worklist, GrowableArray<int> &ready_cnt, VectorSet &next_call, uint sched_slot) {
1.232 +Node* PhaseCFG::select(Block* block, Node_List &worklist, GrowableArray<int> &ready_cnt, VectorSet &next_call, uint sched_slot) {
1.233
1.234 // If only a single entry on the stack, use it
1.235 uint cnt = worklist.size();
1.236 @@ -442,7 +441,7 @@
1.237 }
1.238
1.239 // Final call in a block must be adjacent to 'catch'
1.240 - Node *e = end();
1.241 + Node *e = block->end();
1.242 if( e->is_Catch() && e->in(0)->in(0) == n )
1.243 continue;
1.244
1.245 @@ -468,7 +467,7 @@
1.246 Node* use = n->fast_out(j);
1.247
1.248 // The use is a conditional branch, make them adjacent
1.249 - if (use->is_MachIf() && cfg->get_block_for_node(use) == this) {
1.250 + if (use->is_MachIf() && get_block_for_node(use) == block) {
1.251 found_machif = true;
1.252 break;
1.253 }
1.254 @@ -501,7 +500,7 @@
1.255 n_choice = 1;
1.256 }
1.257
1.258 - uint n_latency = cfg->get_latency_for_node(n);
1.259 + uint n_latency = get_latency_for_node(n);
1.260 uint n_score = n->req(); // Many inputs get high score to break ties
1.261
1.262 // Keep best latency found
1.263 @@ -529,13 +528,13 @@
1.264
1.265
1.266 //------------------------------set_next_call----------------------------------
1.267 -void Block::set_next_call( Node *n, VectorSet &next_call, PhaseCFG* cfg) {
1.268 +void PhaseCFG::set_next_call(Block* block, Node* n, VectorSet& next_call) {
1.269 if( next_call.test_set(n->_idx) ) return;
1.270 for( uint i=0; i<n->len(); i++ ) {
1.271 Node *m = n->in(i);
1.272 if( !m ) continue; // must see all nodes in block that precede call
1.273 - if (cfg->get_block_for_node(m) == this) {
1.274 - set_next_call(m, next_call, cfg);
1.275 + if (get_block_for_node(m) == block) {
1.276 + set_next_call(block, m, next_call);
1.277 }
1.278 }
1.279 }
1.280 @@ -546,12 +545,12 @@
1.281 // next subroutine call get priority - basically it moves things NOT needed
1.282 // for the next call till after the call. This prevents me from trying to
1.283 // carry lots of stuff live across a call.
1.284 -void Block::needed_for_next_call(Node *this_call, VectorSet &next_call, PhaseCFG* cfg) {
1.285 +void PhaseCFG::needed_for_next_call(Block* block, Node* this_call, VectorSet& next_call) {
1.286 // Find the next control-defining Node in this block
1.287 Node* call = NULL;
1.288 for (DUIterator_Fast imax, i = this_call->fast_outs(imax); i < imax; i++) {
1.289 Node* m = this_call->fast_out(i);
1.290 - if(cfg->get_block_for_node(m) == this && // Local-block user
1.291 + if(get_block_for_node(m) == block && // Local-block user
1.292 m != this_call && // Not self-start node
1.293 m->is_MachCall() )
1.294 call = m;
1.295 @@ -559,11 +558,12 @@
1.296 }
1.297 if (call == NULL) return; // No next call (e.g., block end is near)
1.298 // Set next-call for all inputs to this call
1.299 - set_next_call(call, next_call, cfg);
1.300 + set_next_call(block, call, next_call);
1.301 }
1.302
1.303 //------------------------------add_call_kills-------------------------------------
1.304 -void Block::add_call_kills(MachProjNode *proj, RegMask& regs, const char* save_policy, bool exclude_soe) {
1.305 +// helper function that adds caller save registers to MachProjNode
1.306 +static void add_call_kills(MachProjNode *proj, RegMask& regs, const char* save_policy, bool exclude_soe) {
1.307 // Fill in the kill mask for the call
1.308 for( OptoReg::Name r = OptoReg::Name(0); r < _last_Mach_Reg; r=OptoReg::add(r,1) ) {
1.309 if( !regs.Member(r) ) { // Not already defined by the call
1.310 @@ -579,7 +579,7 @@
1.311
1.312
1.313 //------------------------------sched_call-------------------------------------
1.314 -uint Block::sched_call( Matcher &matcher, PhaseCFG* cfg, uint node_cnt, Node_List &worklist, GrowableArray<int> &ready_cnt, MachCallNode *mcall, VectorSet &next_call ) {
1.315 +uint PhaseCFG::sched_call(Block* block, uint node_cnt, Node_List& worklist, GrowableArray<int>& ready_cnt, MachCallNode* mcall, VectorSet& next_call) {
1.316 RegMask regs;
1.317
1.318 // Schedule all the users of the call right now. All the users are
1.319 @@ -592,18 +592,18 @@
1.320 ready_cnt.at_put(n->_idx, n_cnt);
1.321 assert( n_cnt == 0, "" );
1.322 // Schedule next to call
1.323 - _nodes.map(node_cnt++, n);
1.324 + block->map_node(n, node_cnt++);
1.325 // Collect defined registers
1.326 regs.OR(n->out_RegMask());
1.327 // Check for scheduling the next control-definer
1.328 if( n->bottom_type() == Type::CONTROL )
1.329 // Warm up next pile of heuristic bits
1.330 - needed_for_next_call(n, next_call, cfg);
1.331 + needed_for_next_call(block, n, next_call);
1.332
1.333 // Children of projections are now all ready
1.334 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
1.335 Node* m = n->fast_out(j); // Get user
1.336 - if(cfg->get_block_for_node(m) != this) {
1.337 + if(get_block_for_node(m) != block) {
1.338 continue;
1.339 }
1.340 if( m->is_Phi() ) continue;
1.341 @@ -617,14 +617,14 @@
1.342
1.343 // Act as if the call defines the Frame Pointer.
1.344 // Certainly the FP is alive and well after the call.
1.345 - regs.Insert(matcher.c_frame_pointer());
1.346 + regs.Insert(_matcher.c_frame_pointer());
1.347
1.348 // Set all registers killed and not already defined by the call.
1.349 uint r_cnt = mcall->tf()->range()->cnt();
1.350 int op = mcall->ideal_Opcode();
1.351 - MachProjNode *proj = new (matcher.C) MachProjNode( mcall, r_cnt+1, RegMask::Empty, MachProjNode::fat_proj );
1.352 - cfg->map_node_to_block(proj, this);
1.353 - insert_node(proj, node_cnt++);
1.354 + MachProjNode *proj = new (C) MachProjNode( mcall, r_cnt+1, RegMask::Empty, MachProjNode::fat_proj );
1.355 + map_node_to_block(proj, block);
1.356 + block->insert_node(proj, node_cnt++);
1.357
1.358 // Select the right register save policy.
1.359 const char * save_policy;
1.360 @@ -633,13 +633,13 @@
1.361 case Op_CallLeaf:
1.362 case Op_CallLeafNoFP:
1.363 // Calling C code so use C calling convention
1.364 - save_policy = matcher._c_reg_save_policy;
1.365 + save_policy = _matcher._c_reg_save_policy;
1.366 break;
1.367
1.368 case Op_CallStaticJava:
1.369 case Op_CallDynamicJava:
1.370 // Calling Java code so use Java calling convention
1.371 - save_policy = matcher._register_save_policy;
1.372 + save_policy = _matcher._register_save_policy;
1.373 break;
1.374
1.375 default:
1.376 @@ -674,44 +674,46 @@
1.377
1.378 //------------------------------schedule_local---------------------------------
1.379 // Topological sort within a block. Someday become a real scheduler.
1.380 -bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, GrowableArray<int> &ready_cnt, VectorSet &next_call) {
1.381 +bool PhaseCFG::schedule_local(Block* block, GrowableArray<int>& ready_cnt, VectorSet& next_call) {
1.382 // Already "sorted" are the block start Node (as the first entry), and
1.383 // the block-ending Node and any trailing control projections. We leave
1.384 // these alone. PhiNodes and ParmNodes are made to follow the block start
1.385 // Node. Everything else gets topo-sorted.
1.386
1.387 #ifndef PRODUCT
1.388 - if (cfg->trace_opto_pipelining()) {
1.389 - tty->print_cr("# --- schedule_local B%d, before: ---", _pre_order);
1.390 - for (uint i = 0;i < _nodes.size();i++) {
1.391 + if (trace_opto_pipelining()) {
1.392 + tty->print_cr("# --- schedule_local B%d, before: ---", block->_pre_order);
1.393 + for (uint i = 0;i < block->number_of_nodes(); i++) {
1.394 tty->print("# ");
1.395 - get_node(i)->fast_dump();
1.396 + block->get_node(i)->fast_dump();
1.397 }
1.398 tty->print_cr("#");
1.399 }
1.400 #endif
1.401
1.402 // RootNode is already sorted
1.403 - if( _nodes.size() == 1 ) return true;
1.404 + if (block->number_of_nodes() == 1) {
1.405 + return true;
1.406 + }
1.407
1.408 // Move PhiNodes and ParmNodes from 1 to cnt up to the start
1.409 - uint node_cnt = end_idx();
1.410 + uint node_cnt = block->end_idx();
1.411 uint phi_cnt = 1;
1.412 uint i;
1.413 for( i = 1; i<node_cnt; i++ ) { // Scan for Phi
1.414 - Node *n = get_node(i);
1.415 + Node *n = block->get_node(i);
1.416 if( n->is_Phi() || // Found a PhiNode or ParmNode
1.417 - (n->is_Proj() && n->in(0) == head()) ) {
1.418 + (n->is_Proj() && n->in(0) == block->head()) ) {
1.419 // Move guy at 'phi_cnt' to the end; makes a hole at phi_cnt
1.420 - _nodes.map(i,get_node(phi_cnt));
1.421 - _nodes.map(phi_cnt++,n); // swap Phi/Parm up front
1.422 + block->map_node(block->get_node(phi_cnt), i);
1.423 + block->map_node(n, phi_cnt++); // swap Phi/Parm up front
1.424 } else { // All others
1.425 // Count block-local inputs to 'n'
1.426 uint cnt = n->len(); // Input count
1.427 uint local = 0;
1.428 for( uint j=0; j<cnt; j++ ) {
1.429 Node *m = n->in(j);
1.430 - if( m && cfg->get_block_for_node(m) == this && !m->is_top() )
1.431 + if( m && get_block_for_node(m) == block && !m->is_top() )
1.432 local++; // One more block-local input
1.433 }
1.434 ready_cnt.at_put(n->_idx, local); // Count em up
1.435 @@ -723,7 +725,7 @@
1.436 for (uint prec = n->req(); prec < n->len(); prec++) {
1.437 Node* oop_store = n->in(prec);
1.438 if (oop_store != NULL) {
1.439 - assert(cfg->get_block_for_node(oop_store)->_dom_depth <= this->_dom_depth, "oop_store must dominate card-mark");
1.440 + assert(get_block_for_node(oop_store)->_dom_depth <= block->_dom_depth, "oop_store must dominate card-mark");
1.441 }
1.442 }
1.443 }
1.444 @@ -747,16 +749,16 @@
1.445 }
1.446 }
1.447 }
1.448 - for(uint i2=i; i2<_nodes.size(); i2++ ) // Trailing guys get zapped count
1.449 - ready_cnt.at_put(get_node(i2)->_idx, 0);
1.450 + for(uint i2=i; i2< block->number_of_nodes(); i2++ ) // Trailing guys get zapped count
1.451 + ready_cnt.at_put(block->get_node(i2)->_idx, 0);
1.452
1.453 // All the prescheduled guys do not hold back internal nodes
1.454 uint i3;
1.455 for(i3 = 0; i3<phi_cnt; i3++ ) { // For all pre-scheduled
1.456 - Node *n = get_node(i3); // Get pre-scheduled
1.457 + Node *n = block->get_node(i3); // Get pre-scheduled
1.458 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
1.459 Node* m = n->fast_out(j);
1.460 - if (cfg->get_block_for_node(m) == this) { // Local-block user
1.461 + if (get_block_for_node(m) == block) { // Local-block user
1.462 int m_cnt = ready_cnt.at(m->_idx)-1;
1.463 ready_cnt.at_put(m->_idx, m_cnt); // Fix ready count
1.464 }
1.465 @@ -767,7 +769,7 @@
1.466 // Make a worklist
1.467 Node_List worklist;
1.468 for(uint i4=i3; i4<node_cnt; i4++ ) { // Put ready guys on worklist
1.469 - Node *m = get_node(i4);
1.470 + Node *m = block->get_node(i4);
1.471 if( !ready_cnt.at(m->_idx) ) { // Zero ready count?
1.472 if (m->is_iteratively_computed()) {
1.473 // Push induction variable increments last to allow other uses
1.474 @@ -789,15 +791,15 @@
1.475 }
1.476
1.477 // Warm up the 'next_call' heuristic bits
1.478 - needed_for_next_call(head(), next_call, cfg);
1.479 + needed_for_next_call(block, block->head(), next_call);
1.480
1.481 #ifndef PRODUCT
1.482 - if (cfg->trace_opto_pipelining()) {
1.483 - for (uint j=0; j<_nodes.size(); j++) {
1.484 - Node *n = get_node(j);
1.485 + if (trace_opto_pipelining()) {
1.486 + for (uint j=0; j< block->number_of_nodes(); j++) {
1.487 + Node *n = block->get_node(j);
1.488 int idx = n->_idx;
1.489 tty->print("# ready cnt:%3d ", ready_cnt.at(idx));
1.490 - tty->print("latency:%3d ", cfg->get_latency_for_node(n));
1.491 + tty->print("latency:%3d ", get_latency_for_node(n));
1.492 tty->print("%4d: %s\n", idx, n->Name());
1.493 }
1.494 }
1.495 @@ -808,7 +810,7 @@
1.496 while( worklist.size() ) { // Worklist is not ready
1.497
1.498 #ifndef PRODUCT
1.499 - if (cfg->trace_opto_pipelining()) {
1.500 + if (trace_opto_pipelining()) {
1.501 tty->print("# ready list:");
1.502 for( uint i=0; i<worklist.size(); i++ ) { // Inspect entire worklist
1.503 Node *n = worklist[i]; // Get Node on worklist
1.504 @@ -819,13 +821,13 @@
1.505 #endif
1.506
1.507 // Select and pop a ready guy from worklist
1.508 - Node* n = select(cfg, worklist, ready_cnt, next_call, phi_cnt);
1.509 - _nodes.map(phi_cnt++,n); // Schedule him next
1.510 + Node* n = select(block, worklist, ready_cnt, next_call, phi_cnt);
1.511 + block->map_node(n, phi_cnt++); // Schedule him next
1.512
1.513 #ifndef PRODUCT
1.514 - if (cfg->trace_opto_pipelining()) {
1.515 + if (trace_opto_pipelining()) {
1.516 tty->print("# select %d: %s", n->_idx, n->Name());
1.517 - tty->print(", latency:%d", cfg->get_latency_for_node(n));
1.518 + tty->print(", latency:%d", get_latency_for_node(n));
1.519 n->dump();
1.520 if (Verbose) {
1.521 tty->print("# ready list:");
1.522 @@ -840,26 +842,26 @@
1.523 #endif
1.524 if( n->is_MachCall() ) {
1.525 MachCallNode *mcall = n->as_MachCall();
1.526 - phi_cnt = sched_call(matcher, cfg, phi_cnt, worklist, ready_cnt, mcall, next_call);
1.527 + phi_cnt = sched_call(block, phi_cnt, worklist, ready_cnt, mcall, next_call);
1.528 continue;
1.529 }
1.530
1.531 if (n->is_Mach() && n->as_Mach()->has_call()) {
1.532 RegMask regs;
1.533 - regs.Insert(matcher.c_frame_pointer());
1.534 + regs.Insert(_matcher.c_frame_pointer());
1.535 regs.OR(n->out_RegMask());
1.536
1.537 - MachProjNode *proj = new (matcher.C) MachProjNode( n, 1, RegMask::Empty, MachProjNode::fat_proj );
1.538 - cfg->map_node_to_block(proj, this);
1.539 - insert_node(proj, phi_cnt++);
1.540 + MachProjNode *proj = new (C) MachProjNode( n, 1, RegMask::Empty, MachProjNode::fat_proj );
1.541 + map_node_to_block(proj, block);
1.542 + block->insert_node(proj, phi_cnt++);
1.543
1.544 - add_call_kills(proj, regs, matcher._c_reg_save_policy, false);
1.545 + add_call_kills(proj, regs, _matcher._c_reg_save_policy, false);
1.546 }
1.547
1.548 // Children are now all ready
1.549 for (DUIterator_Fast i5max, i5 = n->fast_outs(i5max); i5 < i5max; i5++) {
1.550 Node* m = n->fast_out(i5); // Get user
1.551 - if (cfg->get_block_for_node(m) != this) {
1.552 + if (get_block_for_node(m) != block) {
1.553 continue;
1.554 }
1.555 if( m->is_Phi() ) continue;
1.556 @@ -874,9 +876,8 @@
1.557 }
1.558 }
1.559
1.560 - if( phi_cnt != end_idx() ) {
1.561 + if( phi_cnt != block->end_idx() ) {
1.562 // did not schedule all. Retry, Bailout, or Die
1.563 - Compile* C = matcher.C;
1.564 if (C->subsume_loads() == true && !C->failing()) {
1.565 // Retry with subsume_loads == false
1.566 // If this is the first failure, the sentinel string will "stick"
1.567 @@ -888,12 +889,12 @@
1.568 }
1.569
1.570 #ifndef PRODUCT
1.571 - if (cfg->trace_opto_pipelining()) {
1.572 + if (trace_opto_pipelining()) {
1.573 tty->print_cr("#");
1.574 tty->print_cr("# after schedule_local");
1.575 - for (uint i = 0;i < _nodes.size();i++) {
1.576 + for (uint i = 0;i < block->number_of_nodes();i++) {
1.577 tty->print("# ");
1.578 - get_node(i)->fast_dump();
1.579 + block->get_node(i)->fast_dump();
1.580 }
1.581 tty->cr();
1.582 }
1.583 @@ -919,7 +920,7 @@
1.584 }
1.585
1.586 //------------------------------catch_cleanup_find_cloned_def------------------
1.587 -static Node *catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def_blk, PhaseCFG* cfg, int n_clone_idx) {
1.588 +Node* PhaseCFG::catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def_blk, int n_clone_idx) {
1.589 assert( use_blk != def_blk, "Inter-block cleanup only");
1.590
1.591 // The use is some block below the Catch. Find and return the clone of the def
1.592 @@ -945,8 +946,8 @@
1.593 // PhiNode, the PhiNode uses from the def and IT's uses need fixup.
1.594 Node_Array inputs = new Node_List(Thread::current()->resource_area());
1.595 for(uint k = 1; k < use_blk->num_preds(); k++) {
1.596 - Block* block = cfg->get_block_for_node(use_blk->pred(k));
1.597 - inputs.map(k, catch_cleanup_find_cloned_def(block, def, def_blk, cfg, n_clone_idx));
1.598 + Block* block = get_block_for_node(use_blk->pred(k));
1.599 + inputs.map(k, catch_cleanup_find_cloned_def(block, def, def_blk, n_clone_idx));
1.600 }
1.601
1.602 // Check to see if the use_blk already has an identical phi inserted.
1.603 @@ -968,7 +969,7 @@
1.604 if (fixup == NULL) {
1.605 Node *new_phi = PhiNode::make(use_blk->head(), def);
1.606 use_blk->insert_node(new_phi, 1);
1.607 - cfg->map_node_to_block(new_phi, use_blk);
1.608 + map_node_to_block(new_phi, use_blk);
1.609 for (uint k = 1; k < use_blk->num_preds(); k++) {
1.610 new_phi->set_req(k, inputs[k]);
1.611 }
1.612 @@ -1008,25 +1009,25 @@
1.613 //------------------------------catch_cleanup_inter_block---------------------
1.614 // Fix all input edges in use that reference "def". The use is in a different
1.615 // block than the def.
1.616 -static void catch_cleanup_inter_block(Node *use, Block *use_blk, Node *def, Block *def_blk, PhaseCFG* cfg, int n_clone_idx) {
1.617 +void PhaseCFG::catch_cleanup_inter_block(Node *use, Block *use_blk, Node *def, Block *def_blk, int n_clone_idx) {
1.618 if( !use_blk ) return; // Can happen if the use is a precedence edge
1.619
1.620 - Node *new_def = catch_cleanup_find_cloned_def(use_blk, def, def_blk, cfg, n_clone_idx);
1.621 + Node *new_def = catch_cleanup_find_cloned_def(use_blk, def, def_blk, n_clone_idx);
1.622 catch_cleanup_fix_all_inputs(use, def, new_def);
1.623 }
1.624
1.625 //------------------------------call_catch_cleanup-----------------------------
1.626 // If we inserted any instructions between a Call and his CatchNode,
1.627 // clone the instructions on all paths below the Catch.
1.628 -void Block::call_catch_cleanup(PhaseCFG* cfg, Compile* C) {
1.629 +void PhaseCFG::call_catch_cleanup(Block* block) {
1.630
1.631 // End of region to clone
1.632 - uint end = end_idx();
1.633 - if( !get_node(end)->is_Catch() ) return;
1.634 + uint end = block->end_idx();
1.635 + if( !block->get_node(end)->is_Catch() ) return;
1.636 // Start of region to clone
1.637 uint beg = end;
1.638 - while(!get_node(beg-1)->is_MachProj() ||
1.639 - !get_node(beg-1)->in(0)->is_MachCall() ) {
1.640 + while(!block->get_node(beg-1)->is_MachProj() ||
1.641 + !block->get_node(beg-1)->in(0)->is_MachCall() ) {
1.642 beg--;
1.643 assert(beg > 0,"Catch cleanup walking beyond block boundary");
1.644 }
1.645 @@ -1035,15 +1036,15 @@
1.646
1.647 // Clone along all Catch output paths. Clone area between the 'beg' and
1.648 // 'end' indices.
1.649 - for( uint i = 0; i < _num_succs; i++ ) {
1.650 - Block *sb = _succs[i];
1.651 + for( uint i = 0; i < block->_num_succs; i++ ) {
1.652 + Block *sb = block->_succs[i];
1.653 // Clone the entire area; ignoring the edge fixup for now.
1.654 for( uint j = end; j > beg; j-- ) {
1.655 // It is safe here to clone a node with anti_dependence
1.656 // since clones dominate on each path.
1.657 - Node *clone = get_node(j-1)->clone();
1.658 + Node *clone = block->get_node(j-1)->clone();
1.659 sb->insert_node(clone, 1);
1.660 - cfg->map_node_to_block(clone, sb);
1.661 + map_node_to_block(clone, sb);
1.662 }
1.663 }
1.664
1.665 @@ -1051,7 +1052,7 @@
1.666 // Fixup edges. Check the def-use info per cloned Node
1.667 for(uint i2 = beg; i2 < end; i2++ ) {
1.668 uint n_clone_idx = i2-beg+1; // Index of clone of n in each successor block
1.669 - Node *n = get_node(i2); // Node that got cloned
1.670 + Node *n = block->get_node(i2); // Node that got cloned
1.671 // Need DU safe iterator because of edge manipulation in calls.
1.672 Unique_Node_List *out = new Unique_Node_List(Thread::current()->resource_area());
1.673 for (DUIterator_Fast j1max, j1 = n->fast_outs(j1max); j1 < j1max; j1++) {
1.674 @@ -1060,19 +1061,19 @@
1.675 uint max = out->size();
1.676 for (uint j = 0; j < max; j++) {// For all users
1.677 Node *use = out->pop();
1.678 - Block *buse = cfg->get_block_for_node(use);
1.679 + Block *buse = get_block_for_node(use);
1.680 if( use->is_Phi() ) {
1.681 for( uint k = 1; k < use->req(); k++ )
1.682 if( use->in(k) == n ) {
1.683 - Block* block = cfg->get_block_for_node(buse->pred(k));
1.684 - Node *fixup = catch_cleanup_find_cloned_def(block, n, this, cfg, n_clone_idx);
1.685 + Block* b = get_block_for_node(buse->pred(k));
1.686 + Node *fixup = catch_cleanup_find_cloned_def(b, n, block, n_clone_idx);
1.687 use->set_req(k, fixup);
1.688 }
1.689 } else {
1.690 - if (this == buse) {
1.691 - catch_cleanup_intra_block(use, n, this, beg, n_clone_idx);
1.692 + if (block == buse) {
1.693 + catch_cleanup_intra_block(use, n, block, beg, n_clone_idx);
1.694 } else {
1.695 - catch_cleanup_inter_block(use, buse, n, this, cfg, n_clone_idx);
1.696 + catch_cleanup_inter_block(use, buse, n, block, n_clone_idx);
1.697 }
1.698 }
1.699 } // End for all users
1.700 @@ -1081,13 +1082,13 @@
1.701
1.702 // Remove the now-dead cloned ops
1.703 for(uint i3 = beg; i3 < end; i3++ ) {
1.704 - get_node(beg)->disconnect_inputs(NULL, C);
1.705 - remove_node(beg);
1.706 + block->get_node(beg)->disconnect_inputs(NULL, C);
1.707 + block->remove_node(beg);
1.708 }
1.709
1.710 // If the successor blocks have a CreateEx node, move it back to the top
1.711 - for(uint i4 = 0; i4 < _num_succs; i4++ ) {
1.712 - Block *sb = _succs[i4];
1.713 + for(uint i4 = 0; i4 < block->_num_succs; i4++ ) {
1.714 + Block *sb = block->_succs[i4];
1.715 uint new_cnt = end - beg;
1.716 // Remove any newly created, but dead, nodes.
1.717 for( uint j = new_cnt; j > 0; j-- ) {
