Merge

src/share/tools/MakeDeps/Database.java

@@ -36,6 +36,7 @@ public class Database {
   private FileList outerFiles;
   private FileList indivIncludes;
   private FileList grandInclude; // the results for the grand include file
+  private HashMap<String,String> platformDepFiles;
   private long threshold;
   private int nOuterFiles;
   private int nPrecompiledFiles;
@@ -57,6 +58,7 @@ public class Database {
     outerFiles      = new FileList("outerFiles", plat);
     indivIncludes   = new FileList("IndivIncludes", plat);
     grandInclude    = new FileList(plat.getGIFileTemplate().nameOfList(), plat);
+    platformDepFiles = new HashMap<String,String>();
 
     threshold = t;
     nOuterFiles = 0;
@@ -209,6 +211,10 @@ public class Database {
               FileList p = allFiles.listForFile(includer);
               p.setPlatformDependentInclude(pdName.dirPreStemSuff());
 
+              // Record the implicit include of this file so that the
+              // dependencies for precompiled headers can mention it.
+              platformDepFiles.put(newIncluder, includer);
+
               // Add an implicit dependency on platform
               // specific file for the generic file
 
@@ -408,6 +414,12 @@ public class Database {
       for (Iterator iter = grandInclude.iterator(); iter.hasNext(); ) {
         FileList list = (FileList) iter.next();
         gd.println(list.getName() + " \\");
+        String platformDep = platformDepFiles.get(list.getName());
+        if (platformDep != null) {
+            // make sure changes to the platform dependent file will
+            // cause regeneration of the pch file.
+            gd.println(platformDep + " \\");
+        }
       }
       gd.println();
       gd.println();

src/share/vm/adlc/formssel.cpp

@@ -729,6 +729,7 @@ bool InstructForm::captures_bottom_type() const {
         !strcmp(_matrule->_rChild->_opType,"DecodeN")    ||
         !strcmp(_matrule->_rChild->_opType,"EncodeP")    ||
         !strcmp(_matrule->_rChild->_opType,"LoadN")      ||
+        !strcmp(_matrule->_rChild->_opType,"LoadNKlass") ||
         !strcmp(_matrule->_rChild->_opType,"CreateEx")   ||  // type of exception
         !strcmp(_matrule->_rChild->_opType,"CheckCastPP")) ) return true;
   else if ( is_ideal_load() == Form::idealP )                return true;

src/share/vm/opto/connode.cpp

@@ -565,10 +565,12 @@ Node* DecodeNNode::Identity(PhaseTransform* phase) {
 }
 
 const Type *DecodeNNode::Value( PhaseTransform *phase ) const {
-  if (phase->type( in(1) ) == TypeNarrowOop::NULL_PTR) {
-    return TypePtr::NULL_PTR;
-  }
-  return bottom_type();
+  const Type *t = phase->type( in(1) );
+  if (t == Type::TOP) return Type::TOP;
+  if (t == TypeNarrowOop::NULL_PTR) return TypePtr::NULL_PTR;
+
+  assert(t->isa_narrowoop(), "only  narrowoop here");
+  return t->is_narrowoop()->make_oopptr();
 }
 
 Node* DecodeNNode::decode(PhaseTransform* phase, Node* value) {
@@ -599,10 +601,12 @@ Node* EncodePNode::Identity(PhaseTransform* phase) {
 }
 
 const Type *EncodePNode::Value( PhaseTransform *phase ) const {
-  if (phase->type( in(1) ) == TypePtr::NULL_PTR) {
-    return TypeNarrowOop::NULL_PTR;
-  }
-  return bottom_type();
+  const Type *t = phase->type( in(1) );
+  if (t == Type::TOP) return Type::TOP;
+  if (t == TypePtr::NULL_PTR) return TypeNarrowOop::NULL_PTR;
+
+  assert(t->isa_oopptr(), "only oopptr here");
+  return t->is_oopptr()->make_narrowoop();
 }
 
 Node* EncodePNode::encode(PhaseTransform* phase, Node* value) {

src/share/vm/opto/connode.hpp

@@ -549,10 +549,18 @@ class Opaque1Node : public Node {
   virtual uint hash() const ;                  // { return NO_HASH; }
   virtual uint cmp( const Node &n ) const;
 public:
-  Opaque1Node( Node *n ) : Node(0,n) {}
+  Opaque1Node( Compile* C, Node *n ) : Node(0,n) {
+    // Put it on the Macro nodes list to removed during macro nodes expansion.
+    init_flags(Flag_is_macro);
+    C->add_macro_node(this);
+  }
   // Special version for the pre-loop to hold the original loop limit
   // which is consumed by range check elimination.
-  Opaque1Node( Node *n, Node* orig_limit ) : Node(0,n,orig_limit) {}
+  Opaque1Node( Compile* C, Node *n, Node* orig_limit ) : Node(0,n,orig_limit) {
+    // Put it on the Macro nodes list to removed during macro nodes expansion.
+    init_flags(Flag_is_macro);
+    C->add_macro_node(this);
+  }
   Node* original_loop_limit() { return req()==3 ? in(2) : NULL; }
   virtual int Opcode() const;
   virtual const Type *bottom_type() const { return TypeInt::INT; }
@@ -572,7 +580,11 @@ class Opaque2Node : public Node {
   virtual uint hash() const ;                  // { return NO_HASH; }
   virtual uint cmp( const Node &n ) const;
 public:
-  Opaque2Node( Node *n ) : Node(0,n) {}
+  Opaque2Node( Compile* C, Node *n ) : Node(0,n) {
+    // Put it on the Macro nodes list to removed during macro nodes expansion.
+    init_flags(Flag_is_macro);
+    C->add_macro_node(this);
+  }
   virtual int Opcode() const;
   virtual const Type *bottom_type() const { return TypeInt::INT; }
 };

src/share/vm/opto/gcm.cpp

@@ -307,7 +307,6 @@ static Block* raise_LCA_above_marks(Block* LCA, node_idx_t mark,
 
     // Test and set the visited bit.
     if (mid->raise_LCA_visited() == mark)  continue;  // already visited
-    mid->set_raise_LCA_visited(mark);
 
     // Don't process the current LCA, otherwise the search may terminate early
     if (mid != LCA && mid->raise_LCA_mark() == mark) {
@@ -317,6 +316,8 @@ static Block* raise_LCA_above_marks(Block* LCA, node_idx_t mark,
       assert(early->dominates(LCA), "early is high enough");
       // Resume searching at that point, skipping intermediate levels.
       worklist.push(LCA);
+      if (LCA == mid)
+        continue; // Don't mark as visited to avoid early termination.
     } else {
       // Keep searching through this block's predecessors.
       for (uint j = 1, jmax = mid->num_preds(); j < jmax; j++) {
@@ -324,6 +325,7 @@ static Block* raise_LCA_above_marks(Block* LCA, node_idx_t mark,
         worklist.push(mid_parent);
       }
     }
+    mid->set_raise_LCA_visited(mark);
   }
   return LCA;
 }

src/share/vm/opto/loopTransform.cpp

@@ -690,7 +690,7 @@ void PhaseIdealLoop::insert_pre_post_loops( IdealLoopTree *loop, Node_List &old_
   // (the main-loop trip-counter exit value) because we will be changing
   // the exit value (via unrolling) so we cannot constant-fold away the zero
   // trip guard until all unrolling is done.
-  Node *zer_opaq = new (C, 2) Opaque1Node(incr);
+  Node *zer_opaq = new (C, 2) Opaque1Node(C, incr);
   Node *zer_cmp  = new (C, 3) CmpINode( zer_opaq, limit );
   Node *zer_bol  = new (C, 2) BoolNode( zer_cmp, b_test );
   register_new_node( zer_opaq, new_main_exit );
@@ -760,7 +760,7 @@ void PhaseIdealLoop::insert_pre_post_loops( IdealLoopTree *loop, Node_List &old_
   // pre-loop, the main-loop may not execute at all.  Later in life this
   // zero-trip guard will become the minimum-trip guard when we unroll
   // the main-loop.
-  Node *min_opaq = new (C, 2) Opaque1Node(limit);
+  Node *min_opaq = new (C, 2) Opaque1Node(C, limit);
   Node *min_cmp  = new (C, 3) CmpINode( pre_incr, min_opaq );
   Node *min_bol  = new (C, 2) BoolNode( min_cmp, b_test );
   register_new_node( min_opaq, new_pre_exit );
@@ -810,7 +810,7 @@ void PhaseIdealLoop::insert_pre_post_loops( IdealLoopTree *loop, Node_List &old_
 
   // Save the original loop limit in this Opaque1 node for
   // use by range check elimination.
-  Node *pre_opaq  = new (C, 3) Opaque1Node(pre_limit, limit);
+  Node *pre_opaq  = new (C, 3) Opaque1Node(C, pre_limit, limit);
 
   register_new_node( pre_limit, pre_head->in(0) );
   register_new_node( pre_opaq , pre_head->in(0) );

src/share/vm/opto/loopUnswitch.cpp

@@ -205,7 +205,7 @@ ProjNode* PhaseIdealLoop::create_slow_version_of_loop(IdealLoopTree *loop,
 
   Node *cont      = _igvn.intcon(1);
   set_ctrl(cont, C->root());
-  Node* opq       = new (C, 2) Opaque1Node(cont);
+  Node* opq       = new (C, 2) Opaque1Node(C, cont);
   register_node(opq, outer_loop, entry, dom_depth(entry));
   Node *bol       = new (C, 2) Conv2BNode(opq);
   register_node(bol, outer_loop, entry, dom_depth(entry));

src/share/vm/opto/loopopts.cpp

@@ -2685,7 +2685,7 @@ void PhaseIdealLoop::reorg_offsets( IdealLoopTree *loop ) {
       if( !cle->stride_is_con() ) continue;
       // Hit!  Refactor use to use the post-incremented tripcounter.
       // Compute a post-increment tripcounter.
-      Node *opaq = new (C, 2) Opaque2Node( cle->incr() );
+      Node *opaq = new (C, 2) Opaque2Node( C, cle->incr() );
       register_new_node( opaq, u_ctrl );
       Node *neg_stride = _igvn.intcon(-cle->stride_con());
       set_ctrl(neg_stride, C->root());

src/share/vm/opto/machnode.cpp

@@ -262,14 +262,16 @@ const Node* MachNode::get_base_and_disp(intptr_t &offset, const TypePtr* &adr_ty
     // Now we have collected every part of the ADLC MEMORY_INTER.
     // See if it adds up to a base + offset.
     if (index != NULL) {
-      if (!index->is_Con()) {
-        const TypeNarrowOop* narrowoop = index->bottom_type()->isa_narrowoop();
-        if (narrowoop != NULL) {
-          // Memory references through narrow oops have a
-          // funny base so grab the type from the index.
-          adr_type = narrowoop->make_oopptr();
-          return NULL;
-        }
+      const TypeNarrowOop* narrowoop = index->bottom_type()->isa_narrowoop();
+      if (narrowoop != NULL) { // EncodeN, LoadN, LoadConN, LoadNKlass.
+        // Memory references through narrow oops have a
+        // funny base so grab the type from the index:
+        // [R12 + narrow_oop_reg<<3 + offset]
+        assert(base == NULL, "Memory references through narrow oops have no base");
+        offset = disp;
+        adr_type = narrowoop->make_oopptr()->add_offset(offset);
+        return NULL;
+      } else if (!index->is_Con()) {
         disp = Type::OffsetBot;
       } else if (disp != Type::OffsetBot) {
         const TypeX* ti = index->bottom_type()->isa_intptr_t();

src/share/vm/opto/macro.cpp

@@ -1674,7 +1674,14 @@ bool PhaseMacroExpand::expand_macro_nodes() {
         success = eliminate_locking_node(n->as_AbstractLock());
         break;
       default:
-        assert(false, "unknown node type in macro list");
+        if (n->Opcode() == Op_Opaque1 || n->Opcode() == Op_Opaque2) {
+          _igvn.add_users_to_worklist(n);
+          _igvn.hash_delete(n);
+          _igvn.subsume_node(n, n->in(1));
+          success = true;
+        } else {
+          assert(false, "unknown node type in macro list");
+        }
       }
       assert(success == (C->macro_count() < old_macro_count), "elimination reduces macro count");
       progress = progress || success;

src/share/vm/opto/matcher.cpp

@@ -82,6 +82,7 @@ Matcher::Matcher( Node_List &proj_list ) :
   idealreg2debugmask[Op_RegF] = NULL;
   idealreg2debugmask[Op_RegD] = NULL;
   idealreg2debugmask[Op_RegP] = NULL;
+  debug_only(_mem_node = NULL;)   // Ideal memory node consumed by mach node
 }
 
 //------------------------------warp_incoming_stk_arg------------------------
@@ -1153,7 +1154,10 @@ MachNode *Matcher::match_tree( const Node *n ) {
 
   // StoreNodes require their Memory input to match any LoadNodes
   Node *mem = n->is_Store() ? n->in(MemNode::Memory) : (Node*)1 ;
-
+#ifdef ASSERT
+  Node* save_mem_node = _mem_node;
+  _mem_node = n->is_Store() ? (Node*)n : NULL;
+#endif
   // State object for root node of match tree
   // Allocate it on _states_arena - stack allocation can cause stack overflow.
   State *s = new (&_states_arena) State;
@@ -1205,6 +1209,7 @@ MachNode *Matcher::match_tree( const Node *n ) {
     }
   }
 
+  debug_only( _mem_node = save_mem_node; )
   return m;
 }
 
@@ -1445,8 +1450,30 @@ MachNode *Matcher::ReduceInst( State *s, int rule, Node *&mem ) {
   }
 
   // If a Memory was used, insert a Memory edge
-  if( mem != (Node*)1 )
+  if( mem != (Node*)1 ) {
     mach->ins_req(MemNode::Memory,mem);
+#ifdef ASSERT
+    // Verify adr type after matching memory operation
+    const MachOper* oper = mach->memory_operand();
+    if (oper != NULL && oper != (MachOper*)-1 &&
+        mach->adr_type() != TypeRawPtr::BOTTOM) { // non-direct addressing mode
+      // It has a unique memory operand.  Find corresponding ideal mem node.
+      Node* m = NULL;
+      if (leaf->is_Mem()) {
+        m = leaf;
+      } else {
+        m = _mem_node;
+        assert(m != NULL && m->is_Mem(), "expecting memory node");
+      }
+      if (m->adr_type() != mach->adr_type()) {
+        m->dump();
+        tty->print_cr("mach:");
+        mach->dump(1);
+      }
+      assert(m->adr_type() == mach->adr_type(), "matcher should not change adr type");
+    }
+#endif
+  }
 
   // If the _leaf is an AddP, insert the base edge
   if( leaf->is_AddP() )
@@ -1510,7 +1537,9 @@ void Matcher::ReduceInst_Chain_Rule( State *s, int rule, Node *&mem, MachNode *m
     assert( newrule >= _LAST_MACH_OPER, "Do NOT chain from internal operand");
     mach->_opnds[1] = s->MachOperGenerator( _reduceOp[catch_op], C );
     Node *mem1 = (Node*)1;
+    debug_only(Node *save_mem_node = _mem_node;)
     mach->add_req( ReduceInst(s, newrule, mem1) );
+    debug_only(_mem_node = save_mem_node;)
   }
   return;
 }
@@ -1520,6 +1549,7 @@ uint Matcher::ReduceInst_Interior( State *s, int rule, Node *&mem, MachNode *mac
   if( s->_leaf->is_Load() ) {
     Node *mem2 = s->_leaf->in(MemNode::Memory);
     assert( mem == (Node*)1 || mem == mem2, "multiple Memories being matched at once?" );
+    debug_only( if( mem == (Node*)1 ) _mem_node = s->_leaf;)
     mem = mem2;
   }
   if( s->_leaf->in(0) != NULL && s->_leaf->req() > 1) {
@@ -1563,7 +1593,9 @@ uint Matcher::ReduceInst_Interior( State *s, int rule, Node *&mem, MachNode *mac
         //             --> ReduceInst( newrule )
         mach->_opnds[num_opnds++] = s->MachOperGenerator( _reduceOp[catch_op], C );
         Node *mem1 = (Node*)1;
+        debug_only(Node *save_mem_node = _mem_node;)
         mach->add_req( ReduceInst( newstate, newrule, mem1 ) );
+        debug_only(_mem_node = save_mem_node;)
       }
     }
     assert( mach->_opnds[num_opnds-1], "" );
@@ -1594,6 +1626,7 @@ void Matcher::ReduceOper( State *s, int rule, Node *&mem, MachNode *mach ) {
   if( s->_leaf->is_Load() ) {
     assert( mem == (Node*)1, "multiple Memories being matched at once?" );
     mem = s->_leaf->in(MemNode::Memory);
+    debug_only(_mem_node = s->_leaf;)
   }
   if( s->_leaf->in(0) && s->_leaf->req() > 1) {
     if( !mach->in(0) )
@@ -1618,7 +1651,9 @@ void Matcher::ReduceOper( State *s, int rule, Node *&mem, MachNode *mach ) {
       // Reduce the instruction, and add a direct pointer from this
       // machine instruction to the newly reduced one.
       Node *mem1 = (Node*)1;
+      debug_only(Node *save_mem_node = _mem_node;)
       mach->add_req( ReduceInst( kid, newrule, mem1 ) );
+      debug_only(_mem_node = save_mem_node;)
     }
   }
 }

src/share/vm/opto/matcher.hpp

@@ -104,6 +104,8 @@ class Matcher : public PhaseTransform {
 #ifdef ASSERT
   // Make sure only new nodes are reachable from this node
   void verify_new_nodes_only(Node* root);
+
+  Node* _mem_node;   // Ideal memory node consumed by mach node
 #endif
 
 public:

src/share/vm/opto/subnode.cpp

@@ -45,10 +45,13 @@ Node *SubNode::Identity( PhaseTransform *phase ) {
     return in(2)->in(2);
   }
 
-  // Convert "(X+Y) - Y" into X
+  // Convert "(X+Y) - Y" into X and "(X+Y) - X" into Y
   if( in(1)->Opcode() == Op_AddI ) {
     if( phase->eqv(in(1)->in(2),in(2)) )
       return in(1)->in(1);
+    if (phase->eqv(in(1)->in(1),in(2)))
+      return in(1)->in(2);
+
     // Also catch: "(X + Opaque2(Y)) - Y".  In this case, 'Y' is a loop-varying
     // trip counter and X is likely to be loop-invariant (that's how O2 Nodes
     // are originally used, although the optimizer sometimes jiggers things).