/* * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 only, as * published by the Free Software Foundation. * * This code is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * version 2 for more details (a copy is included in the LICENSE file that * accompanied this code). * * You should have received a copy of the GNU General Public License version * 2 along with this work; if not, write to the Free Software Foundation, * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. * * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA * or visit www.oracle.com if you need additional information or have any * questions. * */ #include "gc/shared/barrierSet.hpp" #include "gc/shared/c2/barrierSetC2.hpp" #include "memory/allocation.inline.hpp" #include "memory/resourceArea.hpp" #include "opto/addnode.hpp" #include "opto/block.hpp" #include "opto/callnode.hpp" #include "opto/castnode.hpp" #include "opto/cfgnode.hpp" #include "opto/idealGraphPrinter.hpp" #include "opto/loopnode.hpp" #include "opto/machnode.hpp" #include "opto/opcodes.hpp" #include "opto/phaseX.hpp" #include "opto/regalloc.hpp" #include "opto/rootnode.hpp" #include "utilities/macros.hpp" #include "utilities/powerOfTwo.hpp" //============================================================================= #define NODE_HASH_MINIMUM_SIZE 255 //------------------------------NodeHash--------------------------------------- NodeHash::NodeHash(Arena *arena, uint est_max_size) : _a(arena), _max( round_up(est_max_size < NODE_HASH_MINIMUM_SIZE ? NODE_HASH_MINIMUM_SIZE : est_max_size) ), _inserts(0), _insert_limit( insert_limit() ), _table( NEW_ARENA_ARRAY( _a , Node* , _max ) ) #ifndef PRODUCT , _grows(0),_look_probes(0), _lookup_hits(0), _lookup_misses(0), _insert_probes(0), _delete_probes(0), _delete_hits(0), _delete_misses(0), _total_inserts(0), _total_insert_probes(0) #endif { // _sentinel must be in the current node space _sentinel = new ProjNode(nullptr, TypeFunc::Control); memset(_table,0,sizeof(Node*)*_max); } //------------------------------hash_find-------------------------------------- // Find in hash table Node *NodeHash::hash_find( const Node *n ) { // ((Node*)n)->set_hash( n->hash() ); uint hash = n->hash(); if (hash == Node::NO_HASH) { NOT_PRODUCT( _lookup_misses++ ); return nullptr; } uint key = hash & (_max-1); uint stride = key | 0x01; NOT_PRODUCT( _look_probes++ ); Node *k = _table[key]; // Get hashed value if( !k ) { // ?Miss? NOT_PRODUCT( _lookup_misses++ ); return nullptr; // Miss! } int op = n->Opcode(); uint req = n->req(); while( 1 ) { // While probing hash table if( k->req() == req && // Same count of inputs k->Opcode() == op ) { // Same Opcode for( uint i=0; iin(i)!=k->in(i)) // Different inputs? goto collision; // "goto" is a speed hack... if( n->cmp(*k) ) { // Check for any special bits NOT_PRODUCT( _lookup_hits++ ); return k; // Hit! } } collision: NOT_PRODUCT( _look_probes++ ); key = (key + stride/*7*/) & (_max-1); // Stride through table with relative prime k = _table[key]; // Get hashed value if( !k ) { // ?Miss? NOT_PRODUCT( _lookup_misses++ ); return nullptr; // Miss! } } ShouldNotReachHere(); return nullptr; } //------------------------------hash_find_insert------------------------------- // Find in hash table, insert if not already present // Used to preserve unique entries in hash table Node *NodeHash::hash_find_insert( Node *n ) { // n->set_hash( ); uint hash = n->hash(); if (hash == Node::NO_HASH) { NOT_PRODUCT( _lookup_misses++ ); return nullptr; } uint key = hash & (_max-1); uint stride = key | 0x01; // stride must be relatively prime to table siz uint first_sentinel = 0; // replace a sentinel if seen. NOT_PRODUCT( _look_probes++ ); Node *k = _table[key]; // Get hashed value if( !k ) { // ?Miss? NOT_PRODUCT( _lookup_misses++ ); _table[key] = n; // Insert into table! DEBUG_ONLY(n->enter_hash_lock()); // Lock down the node while in the table. check_grow(); // Grow table if insert hit limit return nullptr; // Miss! } else if( k == _sentinel ) { first_sentinel = key; // Can insert here } int op = n->Opcode(); uint req = n->req(); while( 1 ) { // While probing hash table if( k->req() == req && // Same count of inputs k->Opcode() == op ) { // Same Opcode for( uint i=0; iin(i)!=k->in(i)) // Different inputs? goto collision; // "goto" is a speed hack... if( n->cmp(*k) ) { // Check for any special bits NOT_PRODUCT( _lookup_hits++ ); return k; // Hit! } } collision: NOT_PRODUCT( _look_probes++ ); key = (key + stride) & (_max-1); // Stride through table w/ relative prime k = _table[key]; // Get hashed value if( !k ) { // ?Miss? NOT_PRODUCT( _lookup_misses++ ); key = (first_sentinel == 0) ? key : first_sentinel; // ?saw sentinel? _table[key] = n; // Insert into table! DEBUG_ONLY(n->enter_hash_lock()); // Lock down the node while in the table. check_grow(); // Grow table if insert hit limit return nullptr; // Miss! } else if( first_sentinel == 0 && k == _sentinel ) { first_sentinel = key; // Can insert here } } ShouldNotReachHere(); return nullptr; } //------------------------------hash_insert------------------------------------ // Insert into hash table void NodeHash::hash_insert( Node *n ) { // // "conflict" comments -- print nodes that conflict // bool conflict = false; // n->set_hash(); uint hash = n->hash(); if (hash == Node::NO_HASH) { return; } check_grow(); uint key = hash & (_max-1); uint stride = key | 0x01; while( 1 ) { // While probing hash table NOT_PRODUCT( _insert_probes++ ); Node *k = _table[key]; // Get hashed value if( !k || (k == _sentinel) ) break; // Found a slot assert( k != n, "already inserted" ); // if( PrintCompilation && PrintOptoStatistics && Verbose ) { tty->print(" conflict: "); k->dump(); conflict = true; } key = (key + stride) & (_max-1); // Stride through table w/ relative prime } _table[key] = n; // Insert into table! DEBUG_ONLY(n->enter_hash_lock()); // Lock down the node while in the table. // if( conflict ) { n->dump(); } } //------------------------------hash_delete------------------------------------ // Replace in hash table with sentinel bool NodeHash::hash_delete( const Node *n ) { Node *k; uint hash = n->hash(); if (hash == Node::NO_HASH) { NOT_PRODUCT( _delete_misses++ ); return false; } uint key = hash & (_max-1); uint stride = key | 0x01; DEBUG_ONLY( uint counter = 0; ); for( ; /* (k != nullptr) && (k != _sentinel) */; ) { DEBUG_ONLY( counter++ ); NOT_PRODUCT( _delete_probes++ ); k = _table[key]; // Get hashed value if( !k ) { // Miss? NOT_PRODUCT( _delete_misses++ ); return false; // Miss! Not in chain } else if( n == k ) { NOT_PRODUCT( _delete_hits++ ); _table[key] = _sentinel; // Hit! Label as deleted entry DEBUG_ONLY(((Node*)n)->exit_hash_lock()); // Unlock the node upon removal from table. return true; } else { // collision: move through table with prime offset key = (key + stride/*7*/) & (_max-1); assert( counter <= _insert_limit, "Cycle in hash-table"); } } ShouldNotReachHere(); return false; } //------------------------------round_up--------------------------------------- // Round up to nearest power of 2 uint NodeHash::round_up(uint x) { x += (x >> 2); // Add 25% slop return MAX2(16U, round_up_power_of_2(x)); } //------------------------------grow------------------------------------------- // Grow _table to next power of 2 and insert old entries void NodeHash::grow() { // Record old state uint old_max = _max; Node **old_table = _table; // Construct new table with twice the space #ifndef PRODUCT _grows++; _total_inserts += _inserts; _total_insert_probes += _insert_probes; _insert_probes = 0; #endif _inserts = 0; _max = _max << 1; _table = NEW_ARENA_ARRAY( _a , Node* , _max ); // (Node**)_a->Amalloc( _max * sizeof(Node*) ); memset(_table,0,sizeof(Node*)*_max); _insert_limit = insert_limit(); // Insert old entries into the new table for( uint i = 0; i < old_max; i++ ) { Node *m = *old_table++; if( !m || m == _sentinel ) continue; DEBUG_ONLY(m->exit_hash_lock()); // Unlock the node upon removal from old table. hash_insert(m); } } //------------------------------clear------------------------------------------ // Clear all entries in _table to null but keep storage void NodeHash::clear() { #ifdef ASSERT // Unlock all nodes upon removal from table. for (uint i = 0; i < _max; i++) { Node* n = _table[i]; if (!n || n == _sentinel) continue; n->exit_hash_lock(); } #endif memset( _table, 0, _max * sizeof(Node*) ); } //-----------------------remove_useless_nodes---------------------------------- // Remove useless nodes from value table, // implementation does not depend on hash function void NodeHash::remove_useless_nodes(VectorSet &useful) { // Dead nodes in the hash table inherited from GVN should not replace // existing nodes, remove dead nodes. uint max = size(); Node *sentinel_node = sentinel(); for( uint i = 0; i < max; ++i ) { Node *n = at(i); if(n != nullptr && n != sentinel_node && !useful.test(n->_idx)) { DEBUG_ONLY(n->exit_hash_lock()); // Unlock the node when removed _table[i] = sentinel_node; // Replace with placeholder } } } void NodeHash::check_no_speculative_types() { #ifdef ASSERT uint max = size(); Unique_Node_List live_nodes; Compile::current()->identify_useful_nodes(live_nodes); Node *sentinel_node = sentinel(); for (uint i = 0; i < max; ++i) { Node *n = at(i); if (n != nullptr && n != sentinel_node && n->is_Type() && live_nodes.member(n)) { TypeNode* tn = n->as_Type(); const Type* t = tn->type(); const Type* t_no_spec = t->remove_speculative(); assert(t == t_no_spec, "dead node in hash table or missed node during speculative cleanup"); } } #endif } #ifndef PRODUCT //------------------------------dump------------------------------------------- // Dump statistics for the hash table void NodeHash::dump() { _total_inserts += _inserts; _total_insert_probes += _insert_probes; if (PrintCompilation && PrintOptoStatistics && Verbose && (_inserts > 0)) { if (WizardMode) { for (uint i=0; i<_max; i++) { if (_table[i]) tty->print("%d/%d/%d ",i,_table[i]->hash()&(_max-1),_table[i]->_idx); } } tty->print("\nGVN Hash stats: %d grows to %d max_size\n", _grows, _max); tty->print(" %d/%d (%8.1f%% full)\n", _inserts, _max, (double)_inserts/_max*100.0); tty->print(" %dp/(%dh+%dm) (%8.2f probes/lookup)\n", _look_probes, _lookup_hits, _lookup_misses, (double)_look_probes/(_lookup_hits+_lookup_misses)); tty->print(" %dp/%di (%8.2f probes/insert)\n", _total_insert_probes, _total_inserts, (double)_total_insert_probes/_total_inserts); // sentinels increase lookup cost, but not insert cost assert((_lookup_misses+_lookup_hits)*4+100 >= _look_probes, "bad hash function"); assert( _inserts+(_inserts>>3) < _max, "table too full" ); assert( _inserts*3+100 >= _insert_probes, "bad hash function" ); } } Node *NodeHash::find_index(uint idx) { // For debugging // Find an entry by its index value for( uint i = 0; i < _max; i++ ) { Node *m = _table[i]; if( !m || m == _sentinel ) continue; if( m->_idx == (uint)idx ) return m; } return nullptr; } #endif #ifdef ASSERT NodeHash::~NodeHash() { // Unlock all nodes upon destruction of table. if (_table != (Node**)badAddress) clear(); } #endif //============================================================================= //------------------------------PhaseRemoveUseless----------------------------- // 1) Use a breadthfirst walk to collect useful nodes reachable from root. PhaseRemoveUseless::PhaseRemoveUseless(PhaseGVN* gvn, Unique_Node_List& worklist, PhaseNumber phase_num) : Phase(phase_num) { C->print_method(PHASE_BEFORE_REMOVEUSELESS, 3); // Implementation requires an edge from root to each SafePointNode // at a backward branch. Inserted in add_safepoint(). // Identify nodes that are reachable from below, useful. C->identify_useful_nodes(_useful); // Update dead node list C->update_dead_node_list(_useful); // Remove all useless nodes from PhaseValues' recorded types // Must be done before disconnecting nodes to preserve hash-table-invariant gvn->remove_useless_nodes(_useful.member_set()); // Remove all useless nodes from future worklist worklist.remove_useless_nodes(_useful.member_set()); // Disconnect 'useless' nodes that are adjacent to useful nodes C->disconnect_useless_nodes(_useful, worklist); } //============================================================================= //------------------------------PhaseRenumberLive------------------------------ // First, remove useless nodes (equivalent to identifying live nodes). // Then, renumber live nodes. // // The set of live nodes is returned by PhaseRemoveUseless in the _useful structure. // If the number of live nodes is 'x' (where 'x' == _useful.size()), then the // PhaseRenumberLive updates the node ID of each node (the _idx field) with a unique // value in the range [0, x). // // At the end of the PhaseRenumberLive phase, the compiler's count of unique nodes is // updated to 'x' and the list of dead nodes is reset (as there are no dead nodes). // // The PhaseRenumberLive phase updates two data structures with the new node IDs. // (1) The "worklist" is "C->igvn_worklist()", which is to collect which nodes need to // be processed by IGVN after removal of the useless nodes. // (2) Type information "gvn->types()" (same as "C->types()") maps every node ID to // the node's type. The mapping is updated to use the new node IDs as well. We // create a new map, and swap it with the old one. // // Other data structures used by the compiler are not updated. The hash table for value // numbering ("C->node_hash()", referenced by PhaseValue::_table) is not updated because // computing the hash values is not based on node IDs. PhaseRenumberLive::PhaseRenumberLive(PhaseGVN* gvn, Unique_Node_List& worklist, PhaseNumber phase_num) : PhaseRemoveUseless(gvn, worklist, Remove_Useless_And_Renumber_Live), _new_type_array(C->comp_arena()), _old2new_map(C->unique(), C->unique(), -1), _is_pass_finished(false), _live_node_count(C->live_nodes()) { assert(RenumberLiveNodes, "RenumberLiveNodes must be set to true for node renumbering to take place"); assert(C->live_nodes() == _useful.size(), "the number of live nodes must match the number of useful nodes"); assert(_delayed.size() == 0, "should be empty"); assert(&worklist == C->igvn_worklist(), "reference still same as the one from Compile"); assert(&gvn->types() == C->types(), "reference still same as that from Compile"); GrowableArray* old_node_note_array = C->node_note_array(); if (old_node_note_array != nullptr) { int new_size = (_useful.size() >> 8) + 1; // The node note array uses blocks, see C->_log2_node_notes_block_size new_size = MAX2(8, new_size); C->set_node_note_array(new (C->comp_arena()) GrowableArray (C->comp_arena(), new_size, 0, nullptr)); C->grow_node_notes(C->node_note_array(), new_size); } assert(worklist.is_subset_of(_useful), "only useful nodes should still be in the worklist"); // Iterate over the set of live nodes. for (uint current_idx = 0; current_idx < _useful.size(); current_idx++) { Node* n = _useful.at(current_idx); const Type* type = gvn->type_or_null(n); _new_type_array.map(current_idx, type); assert(_old2new_map.at(n->_idx) == -1, "already seen"); _old2new_map.at_put(n->_idx, current_idx); if (old_node_note_array != nullptr) { Node_Notes* nn = C->locate_node_notes(old_node_note_array, n->_idx); C->set_node_notes_at(current_idx, nn); } n->set_idx(current_idx); // Update node ID. if (update_embedded_ids(n) < 0) { _delayed.push(n); // has embedded IDs; handle later } } // VectorSet in Unique_Node_Set must be recomputed, since IDs have changed. worklist.recompute_idx_set(); assert(_live_node_count == _useful.size(), "all live nodes must be processed"); _is_pass_finished = true; // pass finished; safe to process delayed updates while (_delayed.size() > 0) { Node* n = _delayed.pop(); int no_of_updates = update_embedded_ids(n); assert(no_of_updates > 0, "should be updated"); } // Replace the compiler's type information with the updated type information. gvn->types().swap(_new_type_array); // Update the unique node count of the compilation to the number of currently live nodes. C->set_unique(_live_node_count); // Set the dead node count to 0 and reset dead node list. C->reset_dead_node_list(); } int PhaseRenumberLive::new_index(int old_idx) { assert(_is_pass_finished, "not finished"); if (_old2new_map.at(old_idx) == -1) { // absent // Allocate a placeholder to preserve uniqueness _old2new_map.at_put(old_idx, _live_node_count); _live_node_count++; } return _old2new_map.at(old_idx); } int PhaseRenumberLive::update_embedded_ids(Node* n) { int no_of_updates = 0; if (n->is_Phi()) { PhiNode* phi = n->as_Phi(); if (phi->_inst_id != -1) { if (!_is_pass_finished) { return -1; // delay } int new_idx = new_index(phi->_inst_id); assert(new_idx != -1, ""); phi->_inst_id = new_idx; no_of_updates++; } if (phi->_inst_mem_id != -1) { if (!_is_pass_finished) { return -1; // delay } int new_idx = new_index(phi->_inst_mem_id); assert(new_idx != -1, ""); phi->_inst_mem_id = new_idx; no_of_updates++; } } const Type* type = _new_type_array.fast_lookup(n->_idx); if (type != nullptr && type->isa_oopptr() && type->is_oopptr()->is_known_instance()) { if (!_is_pass_finished) { return -1; // delay } int old_idx = type->is_oopptr()->instance_id(); int new_idx = new_index(old_idx); const Type* new_type = type->is_oopptr()->with_instance_id(new_idx); _new_type_array.map(n->_idx, new_type); no_of_updates++; } return no_of_updates; } void PhaseValues::init_con_caches() { memset(_icons,0,sizeof(_icons)); memset(_lcons,0,sizeof(_lcons)); memset(_zcons,0,sizeof(_zcons)); } //--------------------------------find_int_type-------------------------------- const TypeInt* PhaseValues::find_int_type(Node* n) { if (n == nullptr) return nullptr; // Call type_or_null(n) to determine node's type since we might be in // parse phase and call n->Value() may return wrong type. // (For example, a phi node at the beginning of loop parsing is not ready.) const Type* t = type_or_null(n); if (t == nullptr) return nullptr; return t->isa_int(); } //-------------------------------find_long_type-------------------------------- const TypeLong* PhaseValues::find_long_type(Node* n) { if (n == nullptr) return nullptr; // (See comment above on type_or_null.) const Type* t = type_or_null(n); if (t == nullptr) return nullptr; return t->isa_long(); } //------------------------------~PhaseValues----------------------------------- #ifndef PRODUCT PhaseValues::~PhaseValues() { // Statistics for NodeHash _table.dump(); // Statistics for value progress and efficiency if( PrintCompilation && Verbose && WizardMode ) { tty->print("\n%sValues: %d nodes ---> %d/%d (%d)", is_IterGVN() ? "Iter" : " ", C->unique(), made_progress(), made_transforms(), made_new_values()); if( made_transforms() != 0 ) { tty->print_cr(" ratio %f", made_progress()/(float)made_transforms() ); } else { tty->cr(); } } } #endif //------------------------------makecon---------------------------------------- ConNode* PhaseValues::makecon(const Type* t) { assert(t->singleton(), "must be a constant"); assert(!t->empty() || t == Type::TOP, "must not be vacuous range"); switch (t->base()) { // fast paths case Type::Half: case Type::Top: return (ConNode*) C->top(); case Type::Int: return intcon( t->is_int()->get_con() ); case Type::Long: return longcon( t->is_long()->get_con() ); default: break; } if (t->is_zero_type()) return zerocon(t->basic_type()); return uncached_makecon(t); } //--------------------------uncached_makecon----------------------------------- // Make an idealized constant - one of ConINode, ConPNode, etc. ConNode* PhaseValues::uncached_makecon(const Type *t) { assert(t->singleton(), "must be a constant"); ConNode* x = ConNode::make(t); ConNode* k = (ConNode*)hash_find_insert(x); // Value numbering if (k == nullptr) { set_type(x, t); // Missed, provide type mapping GrowableArray* nna = C->node_note_array(); if (nna != nullptr) { Node_Notes* loc = C->locate_node_notes(nna, x->_idx, true); loc->clear(); // do not put debug info on constants } } else { x->destruct(this); // Hit, destroy duplicate constant x = k; // use existing constant } return x; } //------------------------------intcon----------------------------------------- // Fast integer constant. Same as "transform(new ConINode(TypeInt::make(i)))" ConINode* PhaseValues::intcon(jint i) { // Small integer? Check cache! Check that cached node is not dead if (i >= _icon_min && i <= _icon_max) { ConINode* icon = _icons[i-_icon_min]; if (icon != nullptr && icon->in(TypeFunc::Control) != nullptr) return icon; } ConINode* icon = (ConINode*) uncached_makecon(TypeInt::make(i)); assert(icon->is_Con(), ""); if (i >= _icon_min && i <= _icon_max) _icons[i-_icon_min] = icon; // Cache small integers return icon; } //------------------------------longcon---------------------------------------- // Fast long constant. ConLNode* PhaseValues::longcon(jlong l) { // Small integer? Check cache! Check that cached node is not dead if (l >= _lcon_min && l <= _lcon_max) { ConLNode* lcon = _lcons[l-_lcon_min]; if (lcon != nullptr && lcon->in(TypeFunc::Control) != nullptr) return lcon; } ConLNode* lcon = (ConLNode*) uncached_makecon(TypeLong::make(l)); assert(lcon->is_Con(), ""); if (l >= _lcon_min && l <= _lcon_max) _lcons[l-_lcon_min] = lcon; // Cache small integers return lcon; } ConNode* PhaseValues::integercon(jlong l, BasicType bt) { if (bt == T_INT) { return intcon(checked_cast(l)); } assert(bt == T_LONG, "not an integer"); return longcon(l); } //------------------------------zerocon----------------------------------------- // Fast zero or null constant. Same as "transform(ConNode::make(Type::get_zero_type(bt)))" ConNode* PhaseValues::zerocon(BasicType bt) { assert((uint)bt <= _zcon_max, "domain check"); ConNode* zcon = _zcons[bt]; if (zcon != nullptr && zcon->in(TypeFunc::Control) != nullptr) return zcon; zcon = (ConNode*) uncached_makecon(Type::get_zero_type(bt)); _zcons[bt] = zcon; return zcon; } //============================================================================= Node* PhaseGVN::apply_ideal(Node* k, bool can_reshape) { Node* i = BarrierSet::barrier_set()->barrier_set_c2()->ideal_node(this, k, can_reshape); if (i == nullptr) { i = k->Ideal(this, can_reshape); } return i; } //------------------------------transform-------------------------------------- // Return a node which computes the same function as this node, but // in a faster or cheaper fashion. Node* PhaseGVN::transform(Node* n) { NOT_PRODUCT( set_transforms(); ) // Apply the Ideal call in a loop until it no longer applies Node* k = n; Node* i = apply_ideal(k, /*can_reshape=*/false); NOT_PRODUCT(uint loop_count = 1;) while (i != nullptr) { assert(i->_idx >= k->_idx, "Idealize should return new nodes, use Identity to return old nodes" ); k = i; #ifdef ASSERT if (loop_count >= K + C->live_nodes()) { dump_infinite_loop_info(i, "PhaseGVN::transform"); } #endif i = apply_ideal(k, /*can_reshape=*/false); NOT_PRODUCT(loop_count++;) } NOT_PRODUCT(if (loop_count != 0) { set_progress(); }) // If brand new node, make space in type array. ensure_type_or_null(k); // Since I just called 'Value' to compute the set of run-time values // for this Node, and 'Value' is non-local (and therefore expensive) I'll // cache Value. Later requests for the local phase->type of this Node can // use the cached Value instead of suffering with 'bottom_type'. const Type* t = k->Value(this); // Get runtime Value set assert(t != nullptr, "value sanity"); if (type_or_null(k) != t) { #ifndef PRODUCT // Do not count initial visit to node as a transformation if (type_or_null(k) == nullptr) { inc_new_values(); set_progress(); } #endif set_type(k, t); // If k is a TypeNode, capture any more-precise type permanently into Node k->raise_bottom_type(t); } if (t->singleton() && !k->is_Con()) { NOT_PRODUCT(set_progress();) return makecon(t); // Turn into a constant } // Now check for Identities i = k->Identity(this); // Look for a nearby replacement if (i != k) { // Found? Return replacement! NOT_PRODUCT(set_progress();) return i; } // Global Value Numbering i = hash_find_insert(k); // Insert if new if (i && (i != k)) { // Return the pre-existing node NOT_PRODUCT(set_progress();) return i; } // Return Idealized original return k; } bool PhaseGVN::is_dominator_helper(Node *d, Node *n, bool linear_only) { if (d->is_top() || (d->is_Proj() && d->in(0)->is_top())) { return false; } if (n->is_top() || (n->is_Proj() && n->in(0)->is_top())) { return false; } assert(d->is_CFG() && n->is_CFG(), "must have CFG nodes"); int i = 0; while (d != n) { n = IfNode::up_one_dom(n, linear_only); i++; if (n == nullptr || i >= 100) { return false; } } return true; } #ifdef ASSERT //------------------------------dead_loop_check-------------------------------- // Check for a simple dead loop when a data node references itself directly // or through an other data node excluding cons and phis. void PhaseGVN::dead_loop_check( Node *n ) { // Phi may reference itself in a loop if (n != nullptr && !n->is_dead_loop_safe() && !n->is_CFG()) { // Do 2 levels check and only data inputs. bool no_dead_loop = true; uint cnt = n->req(); for (uint i = 1; i < cnt && no_dead_loop; i++) { Node *in = n->in(i); if (in == n) { no_dead_loop = false; } else if (in != nullptr && !in->is_dead_loop_safe()) { uint icnt = in->req(); for (uint j = 1; j < icnt && no_dead_loop; j++) { if (in->in(j) == n || in->in(j) == in) no_dead_loop = false; } } } if (!no_dead_loop) n->dump_bfs(100,nullptr,"#"); assert(no_dead_loop, "dead loop detected"); } } /** * Dumps information that can help to debug the problem. A debug * build fails with an assert. */ void PhaseGVN::dump_infinite_loop_info(Node* n, const char* where) { n->dump(4); assert(false, "infinite loop in %s", where); } #endif //============================================================================= //------------------------------PhaseIterGVN----------------------------------- // Initialize with previous PhaseIterGVN info; used by PhaseCCP PhaseIterGVN::PhaseIterGVN(PhaseIterGVN* igvn) : _delay_transform(igvn->_delay_transform), _worklist(*C->igvn_worklist()) { _iterGVN = true; assert(&_worklist == &igvn->_worklist, "sanity"); } //------------------------------PhaseIterGVN----------------------------------- // Initialize with previous PhaseGVN info from Parser PhaseIterGVN::PhaseIterGVN(PhaseGVN* gvn) : _delay_transform(false), _worklist(*C->igvn_worklist()) { _iterGVN = true; uint max; // Dead nodes in the hash table inherited from GVN were not treated as // roots during def-use info creation; hence they represent an invisible // use. Clear them out. max = _table.size(); for( uint i = 0; i < max; ++i ) { Node *n = _table.at(i); if(n != nullptr && n != _table.sentinel() && n->outcnt() == 0) { if( n->is_top() ) continue; // If remove_useless_nodes() has run, we expect no such nodes left. assert(false, "remove_useless_nodes missed this node"); hash_delete(n); } } // Any Phis or Regions on the worklist probably had uses that could not // make more progress because the uses were made while the Phis and Regions // were in half-built states. Put all uses of Phis and Regions on worklist. max = _worklist.size(); for( uint j = 0; j < max; j++ ) { Node *n = _worklist.at(j); uint uop = n->Opcode(); if( uop == Op_Phi || uop == Op_Region || n->is_Type() || n->is_Mem() ) add_users_to_worklist(n); } } void PhaseIterGVN::shuffle_worklist() { if (_worklist.size() < 2) return; for (uint i = _worklist.size() - 1; i >= 1; i--) { uint j = C->random() % (i + 1); swap(_worklist.adr()[i], _worklist.adr()[j]); } } #ifndef PRODUCT void PhaseIterGVN::verify_step(Node* n) { if (is_verify_def_use()) { ResourceMark rm; VectorSet visited; Node_List worklist; _verify_window[_verify_counter % _verify_window_size] = n; ++_verify_counter; if (C->unique() < 1000 || 0 == _verify_counter % (C->unique() < 10000 ? 10 : 100)) { ++_verify_full_passes; worklist.push(C->root()); Node::verify(-1, visited, worklist); return; } for (int i = 0; i < _verify_window_size; i++) { Node* n = _verify_window[i]; if (n == nullptr) { continue; } if (n->in(0) == NodeSentinel) { // xform_idom _verify_window[i] = n->in(1); --i; continue; } // Typical fanout is 1-2, so this call visits about 6 nodes. if (!visited.test_set(n->_idx)) { worklist.push(n); } } Node::verify(4, visited, worklist); } } void PhaseIterGVN::trace_PhaseIterGVN(Node* n, Node* nn, const Type* oldtype) { const Type* newtype = type_or_null(n); if (nn != n || oldtype != newtype) { C->print_method(PHASE_AFTER_ITER_GVN_STEP, 5, n); } if (TraceIterativeGVN) { uint wlsize = _worklist.size(); if (nn != n) { // print old node tty->print("< "); if (oldtype != newtype && oldtype != nullptr) { oldtype->dump(); } do { tty->print("\t"); } while (tty->position() < 16); tty->print("<"); n->dump(); } if (oldtype != newtype || nn != n) { // print new node and/or new type if (oldtype == nullptr) { tty->print("* "); } else if (nn != n) { tty->print("> "); } else { tty->print("= "); } if (newtype == nullptr) { tty->print("null"); } else { newtype->dump(); } do { tty->print("\t"); } while (tty->position() < 16); nn->dump(); } if (Verbose && wlsize < _worklist.size()) { tty->print(" Push {"); while (wlsize != _worklist.size()) { Node* pushed = _worklist.at(wlsize++); tty->print(" %d", pushed->_idx); } tty->print_cr(" }"); } if (nn != n) { // ignore n, it might be subsumed verify_step((Node*) nullptr); } } } void PhaseIterGVN::init_verifyPhaseIterGVN() { _verify_counter = 0; _verify_full_passes = 0; for (int i = 0; i < _verify_window_size; i++) { _verify_window[i] = nullptr; } #ifdef ASSERT // Verify that all modified nodes are on _worklist Unique_Node_List* modified_list = C->modified_nodes(); while (modified_list != nullptr && modified_list->size()) { Node* n = modified_list->pop(); if (!n->is_Con() && !_worklist.member(n)) { n->dump(); fatal("modified node is not on IGVN._worklist"); } } #endif } void PhaseIterGVN::verify_PhaseIterGVN() { #ifdef ASSERT // Verify nodes with changed inputs. Unique_Node_List* modified_list = C->modified_nodes(); while (modified_list != nullptr && modified_list->size()) { Node* n = modified_list->pop(); if (!n->is_Con()) { // skip Con nodes n->dump(); fatal("modified node was not processed by IGVN.transform_old()"); } } #endif C->verify_graph_edges(); if (is_verify_def_use() && PrintOpto) { if (_verify_counter == _verify_full_passes) { tty->print_cr("VerifyIterativeGVN: %d transforms and verify passes", (int) _verify_full_passes); } else { tty->print_cr("VerifyIterativeGVN: %d transforms, %d full verify passes", (int) _verify_counter, (int) _verify_full_passes); } } #ifdef ASSERT if (modified_list != nullptr) { while (modified_list->size() > 0) { Node* n = modified_list->pop(); n->dump(); assert(false, "VerifyIterativeGVN: new modified node was added"); } } verify_optimize(); #endif } #endif /* PRODUCT */ #ifdef ASSERT /** * Dumps information that can help to debug the problem. A debug * build fails with an assert. */ void PhaseIterGVN::dump_infinite_loop_info(Node* n, const char* where) { n->dump(4); _worklist.dump(); assert(false, "infinite loop in %s", where); } /** * Prints out information about IGVN if the 'verbose' option is used. */ void PhaseIterGVN::trace_PhaseIterGVN_verbose(Node* n, int num_processed) { if (TraceIterativeGVN && Verbose) { tty->print(" Pop "); n->dump(); if ((num_processed % 100) == 0) { _worklist.print_set(); } } } #endif /* ASSERT */ void PhaseIterGVN::optimize() { DEBUG_ONLY(uint num_processed = 0;) NOT_PRODUCT(init_verifyPhaseIterGVN();) NOT_PRODUCT(C->reset_igv_phase_iter(PHASE_AFTER_ITER_GVN_STEP);) C->print_method(PHASE_BEFORE_ITER_GVN, 3); if (StressIGVN) { shuffle_worklist(); } // The node count check in the loop below (check_node_count) assumes that we // increase the live node count with at most // max_live_nodes_increase_per_iteration in between checks. If this // assumption does not hold, there is a risk that we exceed the max node // limit in between checks and trigger an assert during node creation. const int max_live_nodes_increase_per_iteration = NodeLimitFudgeFactor * 3; uint loop_count = 0; // Pull from worklist and transform the node. If the node has changed, // update edge info and put uses on worklist. while (_worklist.size() > 0) { if (C->check_node_count(max_live_nodes_increase_per_iteration, "Out of nodes")) { C->print_method(PHASE_AFTER_ITER_GVN, 3); return; } Node* n = _worklist.pop(); if (loop_count >= K * C->live_nodes()) { DEBUG_ONLY(dump_infinite_loop_info(n, "PhaseIterGVN::optimize");) C->record_method_not_compilable("infinite loop in PhaseIterGVN::optimize"); C->print_method(PHASE_AFTER_ITER_GVN, 3); return; } DEBUG_ONLY(trace_PhaseIterGVN_verbose(n, num_processed++);) if (n->outcnt() != 0) { NOT_PRODUCT(const Type* oldtype = type_or_null(n)); // Do the transformation DEBUG_ONLY(int live_nodes_before = C->live_nodes();) Node* nn = transform_old(n); DEBUG_ONLY(int live_nodes_after = C->live_nodes();) // Ensure we did not increase the live node count with more than // max_live_nodes_increase_per_iteration during the call to transform_old DEBUG_ONLY(int increase = live_nodes_after - live_nodes_before;) assert(increase < max_live_nodes_increase_per_iteration, "excessive live node increase in single iteration of IGVN: %d " "(should be at most %d)", increase, max_live_nodes_increase_per_iteration); NOT_PRODUCT(trace_PhaseIterGVN(n, nn, oldtype);) } else if (!n->is_top()) { remove_dead_node(n); } loop_count++; } NOT_PRODUCT(verify_PhaseIterGVN();) C->print_method(PHASE_AFTER_ITER_GVN, 3); } #ifdef ASSERT void PhaseIterGVN::verify_optimize() { assert(_worklist.size() == 0, "igvn worklist must be empty before verify"); if (is_verify_Value() || is_verify_Ideal() || is_verify_Identity()) { ResourceMark rm; Unique_Node_List worklist; bool failure = false; // BFS all nodes, starting at root worklist.push(C->root()); for (uint j = 0; j < worklist.size(); ++j) { Node* n = worklist.at(j); if (is_verify_Value()) { failure |= verify_Value_for(n); } if (is_verify_Ideal()) { failure |= verify_Ideal_for(n, false); } if (is_verify_Ideal()) { failure |= verify_Ideal_for(n, true); } if (is_verify_Identity()) { failure |= verify_Identity_for(n); } // traverse all inputs and outputs for (uint i = 0; i < n->req(); i++) { if (n->in(i) != nullptr) { worklist.push(n->in(i)); } } for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { worklist.push(n->fast_out(i)); } } // If we get this assert, check why the reported nodes were not processed again in IGVN. // We should either make sure that these nodes are properly added back to the IGVN worklist // in PhaseIterGVN::add_users_to_worklist to update them again or add an exception // in the verification code above if that is not possible for some reason (like Load nodes). assert(!failure, "Missed optimization opportunity in PhaseIterGVN"); } verify_empty_worklist(nullptr); } void PhaseIterGVN::verify_empty_worklist(Node* node) { // Verify that the igvn worklist is empty. If no optimization happened, then // nothing needs to be on the worklist. if (_worklist.size() == 0) { return; } stringStream ss; // Print as a block without tty lock. for (uint j = 0; j < _worklist.size(); j++) { Node* n = _worklist.at(j); ss.print("igvn.worklist[%d] ", j); n->dump("\n", false, &ss); } if (_worklist.size() != 0 && node != nullptr) { ss.print_cr("Previously optimized:"); node->dump("\n", false, &ss); } tty->print_cr("%s", ss.as_string()); assert(false, "igvn worklist must still be empty after verify"); } // Check that type(n) == n->Value(), return true if we have a failure. // We have a list of exceptions, see detailed comments in code. // (1) Integer "widen" changes, but the range is the same. // (2) LoadNode performs deep traversals. Load is not notified for changes far away. // (3) CmpPNode performs deep traversals if it compares oopptr. CmpP is not notified for changes far away. bool PhaseIterGVN::verify_Value_for(Node* n) { // If we assert inside type(n), because the type is still a null, then maybe // the node never went through gvn.transform, which would be a bug. const Type* told = type(n); const Type* tnew = n->Value(this); if (told == tnew) { return false; } // Exception (1) // Integer "widen" changes, but range is the same. if (told->isa_integer(tnew->basic_type()) != nullptr) { // both either int or long const TypeInteger* t0 = told->is_integer(tnew->basic_type()); const TypeInteger* t1 = tnew->is_integer(tnew->basic_type()); if (t0->lo_as_long() == t1->lo_as_long() && t0->hi_as_long() == t1->hi_as_long()) { return false; // ignore integer widen } } // Exception (2) // LoadNode performs deep traversals. Load is not notified for changes far away. if (n->is_Load() && !told->singleton()) { // MemNode::can_see_stored_value looks up through many memory nodes, // which means we would need to notify modifications from far up in // the inputs all the way down to the LoadNode. We don't do that. return false; } // Exception (3) // CmpPNode performs deep traversals if it compares oopptr. CmpP is not notified for changes far away. if (n->Opcode() == Op_CmpP && type(n->in(1))->isa_oopptr() && type(n->in(2))->isa_oopptr()) { // SubNode::Value // CmpPNode::sub // MemNode::detect_ptr_independence // MemNode::all_controls_dominate // We find all controls of a pointer load, and see if they dominate the control of // an allocation. If they all dominate, we know the allocation is after (independent) // of the pointer load, and we can say the pointers are different. For this we call // n->dominates(sub, nlist) to check if controls n of the pointer load dominate the // control sub of the allocation. The problems is that sometimes dominates answers // false conservatively, and later it can determine that it is indeed true. Loops with // Region heads can lead to giving up, whereas LoopNodes can be skipped easier, and // so the traversal becomes more powerful. This is difficult to remidy, we would have // to notify the CmpP of CFG updates. Luckily, we recompute CmpP::Value during CCP // after loop-opts, so that should take care of many of these cases. return false; } stringStream ss; // Print as a block without tty lock. ss.cr(); ss.print_cr("Missed Value optimization:"); n->dump_bfs(1, nullptr, "", &ss); ss.print_cr("Current type:"); told->dump_on(&ss); ss.cr(); ss.print_cr("Optimized type:"); tnew->dump_on(&ss); ss.cr(); tty->print_cr("%s", ss.as_string()); return true; } // Check that all Ideal optimizations that could be done were done. // Returns true if it found missed optimization opportunities and // false otherwise (no missed optimization, or skipped verification). bool PhaseIterGVN::verify_Ideal_for(Node* n, bool can_reshape) { // First, we check a list of exceptions, where we skip verification, // because there are known cases where Ideal can optimize after IGVN. // Some may be expected and cannot be fixed, and others should be fixed. switch (n->Opcode()) { // RangeCheckNode::Ideal looks up the chain for about 999 nodes // (see "Range-Check scan limit"). So, it is possible that something // is optimized in that input subgraph, and the RangeCheck was not // added to the worklist because it would be too expensive to walk // down the graph for 1000 nodes and put all on the worklist. // // Found with: // java -XX:VerifyIterativeGVN=0100 -Xbatch --version case Op_RangeCheck: return false; // IfNode::Ideal does: // Node* prev_dom = search_identical(dist, igvn); // which means we seach up the CFG, traversing at most up to a distance. // If anything happens rather far away from the If, we may not put the If // back on the worklist. // // Found with: // java -XX:VerifyIterativeGVN=0100 -Xcomp --version case Op_If: return false; // IfNode::simple_subsuming // Looks for dominating test that subsumes the current test. // Notification could be difficult because of larger distance. // // Found with: // runtime/exceptionMsgs/ArrayIndexOutOfBoundsException/ArrayIndexOutOfBoundsExceptionTest.java#id1 // -XX:VerifyIterativeGVN=1110 case Op_CountedLoopEnd: return false; // LongCountedLoopEndNode::Ideal // Probably same issue as above. // // Found with: // compiler/predicates/assertion/TestAssertionPredicates.java#NoLoopPredicationXbatch // -XX:StressLongCountedLoop=2000000 -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 case Op_LongCountedLoopEnd: return false; // RegionNode::Ideal does "Skip around the useless IF diamond". // 245 IfTrue === 244 // 258 If === 245 257 // 259 IfTrue === 258 [[ 263 ]] // 260 IfFalse === 258 [[ 263 ]] // 263 Region === 263 260 259 [[ 263 268 ]] // to // 245 IfTrue === 244 // 263 Region === 263 245 _ [[ 263 268 ]] // // "Useless" means that there is no code in either branch of the If. // I found a case where this was not done yet during IGVN. // Why does the Region not get added to IGVN worklist when the If diamond becomes useless? // // Found with: // java -XX:VerifyIterativeGVN=0100 -Xcomp --version case Op_Region: return false; // In AddNode::Ideal, we call "commute", which swaps the inputs so // that smaller idx are first. Tracking it back, it led me to // PhaseIdealLoop::remix_address_expressions which swapped the edges. // // Example: // Before PhaseIdealLoop::remix_address_expressions // 154 AddI === _ 12 144 // After PhaseIdealLoop::remix_address_expressions // 154 AddI === _ 144 12 // After AddNode::Ideal // 154 AddI === _ 12 144 // // I suspect that the node should be added to the IGVN worklist after // PhaseIdealLoop::remix_address_expressions // // This is the only case I looked at, there may be others. Found like this: // java -XX:VerifyIterativeGVN=0100 -Xbatch --version // // The following hit the same logic in PhaseIdealLoop::remix_address_expressions. // // Note: currently all of these fail also for other reasons, for example // because of "commute" doing the reordering with the phi below. Once // that is resolved, we can come back to this issue here. // // case Op_AddD: // case Op_AddI: // case Op_AddL: // case Op_AddF: // case Op_MulI: // case Op_MulL: // case Op_MulF: // case Op_MulD: // if (n->in(1)->_idx > n->in(2)->_idx) { // // Expect "commute" to revert this case. // return false; // } // break; // keep verifying // AddFNode::Ideal calls "commute", which can reorder the inputs for this: // Check for tight loop increments: Loop-phi of Add of loop-phi // It wants to take the phi into in(1): // 471 Phi === 435 38 390 // 390 AddF === _ 471 391 // // Other Associative operators are also affected equally. // // Investigate why this does not happen earlier during IGVN. // // Found with: // test/hotspot/jtreg/compiler/loopopts/superword/ReductionPerf.java // -XX:VerifyIterativeGVN=1110 case Op_AddD: //case Op_AddI: // Also affected for other reasons, see case further down. //case Op_AddL: // Also affected for other reasons, see case further down. case Op_AddF: case Op_MulI: case Op_MulL: case Op_MulF: case Op_MulD: case Op_MinF: case Op_MinD: case Op_MaxF: case Op_MaxD: // XorINode::Ideal // Found with: // compiler/intrinsics/chacha/TestChaCha20.java // -XX:VerifyIterativeGVN=1110 case Op_XorI: case Op_XorL: // It seems we may have similar issues with the HF cases. // Found with aarch64: // compiler/vectorization/TestFloat16VectorOperations.java // -XX:VerifyIterativeGVN=1110 case Op_AddHF: case Op_MulHF: case Op_MaxHF: case Op_MinHF: return false; // In MulNode::Ideal the edges can be swapped to help value numbering: // // // We are OK if right is a constant, or right is a load and // // left is a non-constant. // if( !(t2->singleton() || // (in(2)->is_Load() && !(t1->singleton() || in(1)->is_Load())) ) ) { // if( t1->singleton() || // Left input is a constant? // // Otherwise, sort inputs (commutativity) to help value numbering. // (in(1)->_idx > in(2)->_idx) ) { // swap_edges(1, 2); // // Why was this not done earlier during IGVN? // // Found with: // test/hotspot/jtreg/gc/stress/gcbasher/TestGCBasherWithG1.java // -XX:VerifyIterativeGVN=1110 case Op_AndI: // Same for AndL. // Found with: // compiler/intrinsics/bigInteger/MontgomeryMultiplyTest.java // -XX:VerifyIterativeGVN=1110 case Op_AndL: return false; // SubLNode::Ideal does transform like: // Convert "c1 - (y+c0)" into "(c1-c0) - y" // // In IGVN before verification: // 8423 ConvI2L === _ 3519 [[ 8424 ]] #long:-2 // 8422 ConvI2L === _ 8399 [[ 8424 ]] #long:3..256:www // 8424 AddL === _ 8422 8423 [[ 8383 ]] !orig=[8382] // 8016 ConL === 0 [[ 8383 ]] #long:0 // 8383 SubL === _ 8016 8424 [[ 8156 ]] !orig=[8154] // // And then in verification: // 8338 ConL === 0 [[ 8339 8424 ]] #long:-2 <----- Was constant folded. // 8422 ConvI2L === _ 8399 [[ 8424 ]] #long:3..256:www // 8424 AddL === _ 8422 8338 [[ 8383 ]] !orig=[8382] // 8016 ConL === 0 [[ 8383 ]] #long:0 // 8383 SubL === _ 8016 8424 [[ 8156 ]] !orig=[8154] // // So the form changed from: // c1 - (y + [8423 ConvI2L]) // to // c1 - (y + -2) // but the SubL was not added to the IGVN worklist. Investigate why. // There could be other issues too. // // There seems to be a related AddL IGVN optimization that triggers // the same SubL optimization, so investigate that too. // // Found with: // java -XX:VerifyIterativeGVN=0100 -Xcomp --version case Op_SubL: return false; // SubINode::Ideal does // Convert "x - (y+c0)" into "(x-y) - c0" AND // Convert "c1 - (y+c0)" into "(c1-c0) - y" // // Investigate why this does not yet happen during IGVN. // // Found with: // test/hotspot/jtreg/compiler/c2/IVTest.java // -XX:VerifyIterativeGVN=1110 case Op_SubI: return false; // AddNode::IdealIL does transform like: // Convert x + (con - y) into "(x - y) + con" // // In IGVN before verification: // 8382 ConvI2L // 8381 ConvI2L === _ 791 [[ 8383 ]] #long:0 // 8383 SubL === _ 8381 8382 // 8168 ConvI2L // 8156 AddL === _ 8168 8383 [[ 8158 ]] // // And then in verification: // 8424 AddL // 8016 ConL === 0 [[ 8383 ]] #long:0 <--- Was constant folded. // 8383 SubL === _ 8016 8424 // 8168 ConvI2L // 8156 AddL === _ 8168 8383 [[ 8158 ]] // // So the form changed from: // x + (ConvI2L(0) - [8382 ConvI2L]) // to // x + (0 - [8424 AddL]) // but the AddL was not added to the IGVN worklist. Investigate why. // There could be other issues, too. For example with "commute", see above. // // Found with: // java -XX:VerifyIterativeGVN=0100 -Xcomp --version case Op_AddL: return false; // SubTypeCheckNode::Ideal calls SubTypeCheckNode::verify_helper, which does // Node* cmp = phase->transform(new CmpPNode(subklass, in(SuperKlass))); // record_for_cleanup(cmp, phase); // This verification code in the Ideal code creates new nodes, and checks // if they fold in unexpected ways. This means some nodes are created and // added to the worklist, even if the SubTypeCheck is not optimized. This // goes agains the assumption of the verification here, which assumes that // if the node is not optimized, then no new nodes should be created, and // also no nodes should be added to the worklist. // I see two options: // 1) forbid what verify_helper does, because for each Ideal call it // uses memory and that is suboptimal. But it is not clear how that // verification can be done otherwise. // 2) Special case the verification here. Probably the new nodes that // were just created are dead, i.e. they are not connected down to // root. We could verify that, and remove those nodes from the graph // by setting all their inputs to nullptr. And of course we would // have to remove those nodes from the worklist. // Maybe there are other options too, I did not dig much deeper yet. // // Found with: // java -XX:VerifyIterativeGVN=0100 -Xbatch --version case Op_SubTypeCheck: return false; // LoopLimitNode::Ideal when stride is constant power-of-2, we can do a lowering // to other nodes: Conv, Add, Sub, Mul, And ... // // 107 ConI === 0 [[ ... ]] #int:2 // 84 LoadRange === _ 7 83 // 50 ConI === 0 [[ ... ]] #int:0 // 549 LoopLimit === _ 50 84 107 // // I stepped backward, to see how the node was generated, and I found that it was // created in PhaseIdealLoop::exact_limit and not changed since. It is added to the // IGVN worklist. I quickly checked when it goes into LoopLimitNode::Ideal after // that, and it seems we want to skip lowering it until after loop-opts, but never // add call record_for_post_loop_opts_igvn. This would be an easy fix, but there // could be other issues too. // // Fond with: // java -XX:VerifyIterativeGVN=0100 -Xcomp --version case Op_LoopLimit: return false; // PhiNode::Ideal calls split_flow_path, which tries to do this: // "This optimization tries to find two or more inputs of phi with the same constant // value. It then splits them into a separate Phi, and according Region." // // Example: // 130 DecodeN === _ 129 // 50 ConP === 0 [[ 18 91 99 18 ]] #null // 18 Phi === 14 50 130 50 [[ 133 ]] #java/lang/Object * Oop:java/lang/Object * // // turns into: // // 50 ConP === 0 [[ 99 91 18 ]] #null // 130 DecodeN === _ 129 [[ 18 ]] // 18 Phi === 14 130 50 [[ 133 ]] #java/lang/Object * Oop:java/lang/Object * // // We would have to investigate why this optimization does not happen during IGVN. // There could also be other issues - I did not investigate further yet. // // Found with: // java -XX:VerifyIterativeGVN=0100 -Xcomp --version case Op_Phi: return false; // MemBarNode::Ideal does "Eliminate volatile MemBars for scalar replaced objects". // For examle "The allocated object does not escape". // // It seems the difference to earlier calls to MemBarNode::Ideal, is that there // alloc->as_Allocate()->does_not_escape_thread() returned false, but in verification // it returned true. Why does the MemBarStoreStore not get added to the IGVN // worklist when this change happens? // // Found with: // java -XX:VerifyIterativeGVN=0100 -Xcomp --version case Op_MemBarStoreStore: return false; // ConvI2LNode::Ideal converts // 648 AddI === _ 583 645 [[ 661 ]] // 661 ConvI2L === _ 648 [[ 664 ]] #long:0..maxint-1:www // into // 772 ConvI2L === _ 645 [[ 773 ]] #long:-120..maxint-61:www // 771 ConvI2L === _ 583 [[ 773 ]] #long:60..120:www // 773 AddL === _ 771 772 [[ ]] // // We have to investigate why this does not happen during IGVN in this case. // There could also be other issues - I did not investigate further yet. // // Found with: // java -XX:VerifyIterativeGVN=0100 -Xcomp --version case Op_ConvI2L: return false; // AddNode::IdealIL can do this transform (and similar other ones): // Convert "a*b+a*c into a*(b+c) // The example had AddI(MulI(a, b), MulI(a, c)). Why did this not happen // during IGVN? There was a mutation for one of the MulI, and only // after that the pattern was as needed for the optimization. The MulI // was added to the IGVN worklist, but not the AddI. This probably // can be fixed by adding the correct pattern in add_users_of_use_to_worklist. // // Found with: // test/hotspot/jtreg/compiler/loopopts/superword/ReductionPerf.java // -XX:VerifyIterativeGVN=1110 case Op_AddI: return false; // ArrayCopyNode::Ideal // calls ArrayCopyNode::prepare_array_copy // calls Compile::conv_I2X_index -> is called with sizetype = intcon(0), I think that // is not expected, and we create a range int:0..-1 // calls Compile::constrained_convI2L -> creates ConvI2L(intcon(1), int:0..-1) // note: the type is already empty! // calls PhaseIterGVN::transform // calls PhaseIterGVN::transform_old // calls PhaseIterGVN::subsume_node -> subsume ConvI2L with TOP // calls Unique_Node_List::push -> pushes TOP to worklist // // Once we get back to ArrayCopyNode::prepare_array_copy, we get back TOP, and // return false. This means we eventually return nullptr from ArrayCopyNode::Ideal. // // Question: is it ok to push anything to the worklist during ::Ideal, if we will // return nullptr, indicating nothing happened? // Is it smart to do transform in Compile::constrained_convI2L, and then // check for TOP in calls ArrayCopyNode::prepare_array_copy? // Should we just allow TOP to land on the worklist, as an exception? // // Found with: // compiler/arraycopy/TestArrayCopyAsLoadsStores.java // -XX:VerifyIterativeGVN=1110 case Op_ArrayCopy: return false; // CastLLNode::Ideal // calls ConstraintCastNode::optimize_integer_cast -> pushes CastLL through SubL // // Could be a notification issue, where updates inputs of CastLL do not notify // down through SubL to CastLL. // // Found With: // compiler/c2/TestMergeStoresMemorySegment.java#byte-array // -XX:VerifyIterativeGVN=1110 case Op_CastLL: return false; // Similar case happens to CastII // // Found With: // compiler/c2/TestScalarReplacementMaxLiveNodes.java // -XX:VerifyIterativeGVN=1110 case Op_CastII: return false; // MaxLNode::Ideal // calls AddNode::Ideal // calls commute -> decides to swap edges // // Another notification issue, because we check inputs of inputs? // MaxL -> Phi -> Loop // MaxL -> Phi -> MaxL // // Found with: // compiler/c2/irTests/TestIfMinMax.java // -XX:VerifyIterativeGVN=1110 case Op_MaxL: case Op_MinL: return false; // OrINode::Ideal // calls AddNode::Ideal // calls commute -> left is Load, right not -> commute. // // Not sure why notification does not work here, seems like // the depth is only 1, so it should work. Needs investigation. // // Found with: // compiler/codegen/TestCharVect2.java#id0 // -XX:VerifyIterativeGVN=1110 case Op_OrI: case Op_OrL: return false; // Bool -> constant folded to 1. // Issue with notification? // // Found with: // compiler/c2/irTests/TestVectorizationMismatchedAccess.java // -XX:VerifyIterativeGVN=1110 case Op_Bool: return false; // LShiftLNode::Ideal // Looks at pattern: "(x + x) << c0", converts it to "x << (c0 + 1)" // Probably a notification issue. // // Found with: // compiler/conversions/TestMoveConvI2LOrCastIIThruAddIs.java // -ea -esa -XX:CompileThreshold=100 -XX:+UnlockExperimentalVMOptions -server -XX:-TieredCompilation -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 case Op_LShiftL: return false; // LShiftINode::Ideal // pattern: ((x + con1) << con2) -> x << con2 + con1 << con2 // Could be issue with notification of inputs of inputs // // Side-note: should cases like these not be shared between // LShiftI and LShiftL? // // Found with: // compiler/escapeAnalysis/Test6689060.java // -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 -ea -esa -XX:CompileThreshold=100 -XX:+UnlockExperimentalVMOptions -server -XX:-TieredCompilation -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 case Op_LShiftI: return false; // AddPNode::Ideal seems to do set_req without removing lock first. // Found with various vector tests tier1-tier3. case Op_AddP: return false; // StrIndexOfNode::Ideal // Found in tier1-3. case Op_StrIndexOf: case Op_StrIndexOfChar: return false; // StrEqualsNode::Identity // // Found (linux x64 only?) with: // serviceability/sa/ClhsdbThreadContext.java // -XX:+UnlockExperimentalVMOptions -XX:LockingMode=1 -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 case Op_StrEquals: return false; // AryEqNode::Ideal // Not investigated. Reshapes itself and adds lots of nodes to the worklist. // // Found with: // vmTestbase/vm/mlvm/meth/stress/compiler/i2c_c2i/Test.java // -XX:+UnlockDiagnosticVMOptions -XX:-TieredCompilation -XX:+StressUnstableIfTraps -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 case Op_AryEq: return false; // MergeMemNode::Ideal // Found in tier1-3. Did not investigate further yet. case Op_MergeMem: return false; // URShiftINode::Ideal // Found in tier1-3. Did not investigate further yet. case Op_URShiftI: return false; // CMoveINode::Ideal // Found in tier1-3. Did not investigate further yet. case Op_CMoveI: return false; // CmpPNode::Ideal calls isa_const_java_mirror // and generates new constant nodes, even if no progress is made. // We can probably rewrite this so that only types are generated. // It seems that object types are not hashed, we could investigate // if that is an option as well. // // Found with: // java -XX:VerifyIterativeGVN=1110 -Xcomp --version case Op_CmpP: return false; // MinINode::Ideal // Did not investigate, but there are some patterns that might // need more notification. case Op_MinI: case Op_MaxI: // preemptively removed it as well. return false; } if (n->is_Load()) { // LoadNode::Ideal uses tries to find an earlier memory state, and // checks can_see_stored_value for it. // // Investigate why this was not already done during IGVN. // A similar issue happens with Identity. // // There seem to be other cases where loads go up some steps, like // LoadNode::Ideal going up 10x steps to find dominating load. // // Found with: // test/hotspot/jtreg/compiler/arraycopy/TestCloneAccess.java // -XX:VerifyIterativeGVN=1110 return false; } if (n->is_Store()) { // StoreNode::Ideal can do this: // // Capture an unaliased, unconditional, simple store into an initializer. // // Or, if it is independent of the allocation, hoist it above the allocation. // That replaces the Store with a MergeMem. // // We have to investigate why this does not happen during IGVN in this case. // There could also be other issues - I did not investigate further yet. // // Found with: // java -XX:VerifyIterativeGVN=0100 -Xcomp --version return false; } if (n->is_Vector()) { // VectorNode::Ideal swaps edges, but only for ops // that are deemed commutable. But swap_edges // requires the hash to be invariant when the edges // are swapped, which is not implemented for these // vector nodes. This seems not to create any trouble // usually, but we can also get graphs where in the // end the nodes are not all commuted, so there is // definitively an issue here. // // Probably we have two options: kill the hash, or // properly make the hash commutation friendly. // // Found with: // compiler/vectorapi/TestMaskedMacroLogicVector.java // -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 -XX:+UseParallelGC -XX:+UseNUMA return false; } if (n->is_Region()) { // LoopNode::Ideal calls RegionNode::Ideal. // CountedLoopNode::Ideal calls RegionNode::Ideal too. // But I got an issue because RegionNode::optimize_trichotomy // then modifies another node, and pushes nodes to the worklist // Not sure if this is ok, modifying another node like that. // Maybe it is, then we need to look into what to do with // the nodes that are now on the worklist, maybe just clear // them out again. But maybe modifying other nodes like that // is also bad design. In the end, we return nullptr for // the current CountedLoop. But the extra nodes on the worklist // trip the asserts later on. // // Found with: // compiler/eliminateAutobox/TestShortBoxing.java // -ea -esa -XX:CompileThreshold=100 -XX:+UnlockExperimentalVMOptions -server -XX:-TieredCompilation -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 return false; } if (n->is_CallJava()) { // CallStaticJavaNode::Ideal // Led to a crash: // assert((is_CallStaticJava() && cg->is_mh_late_inline()) || (is_CallDynamicJava() && cg->is_virtual_late_inline())) failed: mismatch // // Did not investigate yet, could be a bug. // Or maybe it does not expect to be called during verification. // // Found with: // test/jdk/jdk/incubator/vector/VectorRuns.java // -XX:VerifyIterativeGVN=1110 // CallDynamicJavaNode::Ideal, and I think also for CallStaticJavaNode::Ideal // and possibly their subclasses. // During late inlining it can call CallJavaNode::register_for_late_inline // That means we do more rounds of late inlining, but might fail. // Then we do IGVN again, and register the node again for late inlining. // This creates an endless cycle. Everytime we try late inlining, we // are also creating more nodes, especially SafePoint and MergeMem. // These nodes are immediately rejected when the inlining fails in the // do_late_inline_check, but they still grow the memory, until we hit // the MemLimit and crash. // The assumption here seems that CallDynamicJavaNode::Ideal does not get // called repeatedly, and eventually we terminate. I fear this is not // a great assumption to make. We should investigate more. // // Found with: // compiler/loopopts/superword/TestDependencyOffsets.java#vanilla-U // -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 return false; } // The number of nodes shoud not increase. uint old_unique = C->unique(); Node* i = n->Ideal(this, can_reshape); // If there was no new Idealization, we are probably happy. if (i == nullptr) { if (old_unique < C->unique()) { stringStream ss; // Print as a block without tty lock. ss.cr(); ss.print_cr("Ideal optimization did not make progress but created new unused nodes."); ss.print_cr(" old_unique = %d, unique = %d", old_unique, C->unique()); n->dump_bfs(1, nullptr, "", &ss); tty->print_cr("%s", ss.as_string()); return true; } verify_empty_worklist(n); // Everything is good. return false; } // We just saw a new Idealization which was not done during IGVN. stringStream ss; // Print as a block without tty lock. ss.cr(); ss.print_cr("Missed Ideal optimization (can_reshape=%s):", can_reshape ? "true": "false"); if (i == n) { ss.print_cr("The node was reshaped by Ideal."); } else { ss.print_cr("The node was replaced by Ideal."); ss.print_cr("Old node:"); n->dump_bfs(1, nullptr, "", &ss); } ss.print_cr("The result after Ideal:"); i->dump_bfs(1, nullptr, "", &ss); tty->print_cr("%s", ss.as_string()); return true; } // Check that all Identity optimizations that could be done were done. // Returns true if it found missed optimization opportunities and // false otherwise (no missed optimization, or skipped verification). bool PhaseIterGVN::verify_Identity_for(Node* n) { // First, we check a list of exceptions, where we skip verification, // because there are known cases where Ideal can optimize after IGVN. // Some may be expected and cannot be fixed, and others should be fixed. switch (n->Opcode()) { // SafePointNode::Identity can remove SafePoints, but wants to wait until // after loopopts: // // Transforming long counted loops requires a safepoint node. Do not // // eliminate a safepoint until loop opts are over. // if (in(0)->is_Proj() && !phase->C->major_progress()) { // // I think the check for major_progress does delay it until after loopopts // but it does not ensure that the node is on the IGVN worklist after // loopopts. I think we should try to instead check for // phase->C->post_loop_opts_phase() and call record_for_post_loop_opts_igvn. // // Found with: // java -XX:VerifyIterativeGVN=1000 -Xcomp --version case Op_SafePoint: return false; // MergeMemNode::Identity replaces the MergeMem with its base_memory if it // does not record any other memory splits. // // I did not deeply investigate, but it looks like MergeMemNode::Identity // never got called during IGVN for this node, investigate why. // // Found with: // java -XX:VerifyIterativeGVN=1000 -Xcomp --version case Op_MergeMem: return false; // ConstraintCastNode::Identity finds casts that are the same, except that // the control is "higher up", i.e. dominates. The call goes via // ConstraintCastNode::dominating_cast to PhaseGVN::is_dominator_helper, // which traverses up to 100 idom steps. If anything gets optimized somewhere // away from the cast, but within 100 idom steps, the cast may not be // put on the IGVN worklist any more. // // Found with: // java -XX:VerifyIterativeGVN=1000 -Xcomp --version case Op_CastPP: case Op_CastII: case Op_CastLL: return false; // Same issue for CheckCastPP, uses ConstraintCastNode::Identity and // checks dominator, which may be changed, but too far up for notification // to work. // // Found with: // compiler/c2/irTests/TestSkeletonPredicates.java // -XX:VerifyIterativeGVN=1110 case Op_CheckCastPP: return false; // In SubNode::Identity, we do: // Convert "(X+Y) - Y" into X and "(X+Y) - X" into Y // In the example, the AddI had an input replaced, the AddI is // added to the IGVN worklist, but the SubI is one link further // down and is not added. I checked add_users_of_use_to_worklist // where I would expect the SubI would be added, and I cannot // find the pattern, only this one: // If changed AddI/SubI inputs, check CmpU for range check optimization. // // Fix this "notification" issue and check if there are any other // issues. // // Found with: // java -XX:VerifyIterativeGVN=1000 -Xcomp --version case Op_SubI: case Op_SubL: return false; // PhiNode::Identity checks for patterns like: // r = (x != con) ? x : con; // that can be constant folded to "x". // // Call goes through PhiNode::is_cmove_id and CMoveNode::is_cmove_id. // I suspect there was some earlier change to one of the inputs, but // not all relevant outputs were put on the IGVN worklist. // // Found with: // test/hotspot/jtreg/gc/stress/gcbasher/TestGCBasherWithG1.java // -XX:VerifyIterativeGVN=1110 case Op_Phi: return false; // ConvI2LNode::Identity does // convert I2L(L2I(x)) => x // // Investigate why this did not already happen during IGVN. // // Found with: // compiler/loopopts/superword/TestDependencyOffsets.java#vanilla-A // -XX:VerifyIterativeGVN=1110 case Op_ConvI2L: return false; // MaxNode::find_identity_operation // Finds patterns like Max(A, Max(A, B)) -> Max(A, B) // This can be a 2-hop search, so maybe notification is not // good enough. // // Found with: // compiler/codegen/TestBooleanVect.java // -XX:VerifyIterativeGVN=1110 case Op_MaxL: case Op_MinL: case Op_MaxI: case Op_MinI: case Op_MaxF: case Op_MinF: case Op_MaxHF: case Op_MinHF: case Op_MaxD: case Op_MinD: return false; // AddINode::Identity // Converts (x-y)+y to x // Could be issue with notification // // Turns out AddL does the same. // // Found with: // compiler/c2/Test6792161.java // -ea -esa -XX:CompileThreshold=100 -XX:+UnlockExperimentalVMOptions -server -XX:-TieredCompilation -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 case Op_AddI: case Op_AddL: return false; // AbsINode::Identity // Not investigated yet. case Op_AbsI: return false; } if (n->is_Load()) { // LoadNode::Identity tries to look for an earlier store value via // can_see_stored_value. I found an example where this led to // an Allocation, where we could assume the value was still zero. // So the LoadN can be replaced with a zerocon. // // Investigate why this was not already done during IGVN. // A similar issue happens with Ideal. // // Found with: // java -XX:VerifyIterativeGVN=1000 -Xcomp --version return false; } if (n->is_Store()) { // StoreNode::Identity // Not investigated, but found missing optimization for StoreI. // Looks like a StoreI is replaced with an InitializeNode. // // Found with: // applications/ctw/modules/java_base_2.java // -ea -esa -XX:CompileThreshold=100 -XX:+UnlockExperimentalVMOptions -server -XX:-TieredCompilation -Djava.awt.headless=true -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=1110 return false; } if (n->is_Vector()) { // Found with tier1-3. Not investigated yet. // The observed issue was with AndVNode::Identity return false; } Node* i = n->Identity(this); // If we cannot find any other Identity, we are happy. if (i == n) { verify_empty_worklist(n); return false; } // The verification just found a new Identity that was not found during IGVN. stringStream ss; // Print as a block without tty lock. ss.cr(); ss.print_cr("Missed Identity optimization:"); ss.print_cr("Old node:"); n->dump_bfs(1, nullptr, "", &ss); ss.print_cr("New node:"); i->dump_bfs(1, nullptr, "", &ss); tty->print_cr("%s", ss.as_string()); return true; } #endif /** * Register a new node with the optimizer. Update the types array, the def-use * info. Put on worklist. */ Node* PhaseIterGVN::register_new_node_with_optimizer(Node* n, Node* orig) { set_type_bottom(n); _worklist.push(n); if (orig != nullptr) C->copy_node_notes_to(n, orig); return n; } //------------------------------transform-------------------------------------- // Non-recursive: idealize Node 'n' with respect to its inputs and its value Node *PhaseIterGVN::transform( Node *n ) { if (_delay_transform) { // Register the node but don't optimize for now register_new_node_with_optimizer(n); return n; } // If brand new node, make space in type array, and give it a type. ensure_type_or_null(n); if (type_or_null(n) == nullptr) { set_type_bottom(n); } return transform_old(n); } Node *PhaseIterGVN::transform_old(Node* n) { NOT_PRODUCT(set_transforms()); // Remove 'n' from hash table in case it gets modified _table.hash_delete(n); #ifdef ASSERT if (is_verify_def_use()) { assert(!_table.find_index(n->_idx), "found duplicate entry in table"); } #endif // Allow Bool -> Cmp idealisation in late inlining intrinsics that return a bool if (n->is_Cmp()) { add_users_to_worklist(n); } // Apply the Ideal call in a loop until it no longer applies Node* k = n; DEBUG_ONLY(dead_loop_check(k);) DEBUG_ONLY(bool is_new = (k->outcnt() == 0);) C->remove_modified_node(k); Node* i = apply_ideal(k, /*can_reshape=*/true); assert(i != k || is_new || i->outcnt() > 0, "don't return dead nodes"); #ifndef PRODUCT verify_step(k); #endif DEBUG_ONLY(uint loop_count = 1;) while (i != nullptr) { #ifdef ASSERT if (loop_count >= K + C->live_nodes()) { dump_infinite_loop_info(i, "PhaseIterGVN::transform_old"); } #endif assert((i->_idx >= k->_idx) || i->is_top(), "Idealize should return new nodes, use Identity to return old nodes"); // Made a change; put users of original Node on worklist add_users_to_worklist(k); // Replacing root of transform tree? if (k != i) { // Make users of old Node now use new. subsume_node(k, i); k = i; } DEBUG_ONLY(dead_loop_check(k);) // Try idealizing again DEBUG_ONLY(is_new = (k->outcnt() == 0);) C->remove_modified_node(k); i = apply_ideal(k, /*can_reshape=*/true); assert(i != k || is_new || (i->outcnt() > 0), "don't return dead nodes"); #ifndef PRODUCT verify_step(k); #endif DEBUG_ONLY(loop_count++;) } // If brand new node, make space in type array. ensure_type_or_null(k); // See what kind of values 'k' takes on at runtime const Type* t = k->Value(this); assert(t != nullptr, "value sanity"); // Since I just called 'Value' to compute the set of run-time values // for this Node, and 'Value' is non-local (and therefore expensive) I'll // cache Value. Later requests for the local phase->type of this Node can // use the cached Value instead of suffering with 'bottom_type'. if (type_or_null(k) != t) { #ifndef PRODUCT inc_new_values(); set_progress(); #endif set_type(k, t); // If k is a TypeNode, capture any more-precise type permanently into Node k->raise_bottom_type(t); // Move users of node to worklist add_users_to_worklist(k); } // If 'k' computes a constant, replace it with a constant if (t->singleton() && !k->is_Con()) { NOT_PRODUCT(set_progress();) Node* con = makecon(t); // Make a constant add_users_to_worklist(k); subsume_node(k, con); // Everybody using k now uses con return con; } // Now check for Identities i = k->Identity(this); // Look for a nearby replacement if (i != k) { // Found? Return replacement! NOT_PRODUCT(set_progress();) add_users_to_worklist(k); subsume_node(k, i); // Everybody using k now uses i return i; } // Global Value Numbering i = hash_find_insert(k); // Check for pre-existing node if (i && (i != k)) { // Return the pre-existing node if it isn't dead NOT_PRODUCT(set_progress();) add_users_to_worklist(k); subsume_node(k, i); // Everybody using k now uses i return i; } // Return Idealized original return k; } //---------------------------------saturate------------------------------------ const Type* PhaseIterGVN::saturate(const Type* new_type, const Type* old_type, const Type* limit_type) const { return new_type->narrow(old_type); } //------------------------------remove_globally_dead_node---------------------- // Kill a globally dead Node. All uses are also globally dead and are // aggressively trimmed. void PhaseIterGVN::remove_globally_dead_node( Node *dead ) { enum DeleteProgress { PROCESS_INPUTS, PROCESS_OUTPUTS }; ResourceMark rm; Node_Stack stack(32); stack.push(dead, PROCESS_INPUTS); while (stack.is_nonempty()) { dead = stack.node(); if (dead->Opcode() == Op_SafePoint) { dead->as_SafePoint()->disconnect_from_root(this); } uint progress_state = stack.index(); assert(dead != C->root(), "killing root, eh?"); assert(!dead->is_top(), "add check for top when pushing"); NOT_PRODUCT( set_progress(); ) if (progress_state == PROCESS_INPUTS) { // After following inputs, continue to outputs stack.set_index(PROCESS_OUTPUTS); if (!dead->is_Con()) { // Don't kill cons but uses bool recurse = false; // Remove from hash table _table.hash_delete( dead ); // Smash all inputs to 'dead', isolating him completely for (uint i = 0; i < dead->req(); i++) { Node *in = dead->in(i); if (in != nullptr && in != C->top()) { // Points to something? int nrep = dead->replace_edge(in, nullptr, this); // Kill edges assert((nrep > 0), "sanity"); if (in->outcnt() == 0) { // Made input go dead? stack.push(in, PROCESS_INPUTS); // Recursively remove recurse = true; } else if (in->outcnt() == 1 && in->has_special_unique_user()) { _worklist.push(in->unique_out()); } else if (in->outcnt() <= 2 && dead->is_Phi()) { if (in->Opcode() == Op_Region) { _worklist.push(in); } else if (in->is_Store()) { DUIterator_Fast imax, i = in->fast_outs(imax); _worklist.push(in->fast_out(i)); i++; if (in->outcnt() == 2) { _worklist.push(in->fast_out(i)); i++; } assert(!(i < imax), "sanity"); } } else if (dead->is_data_proj_of_pure_function(in)) { _worklist.push(in); } else { BarrierSet::barrier_set()->barrier_set_c2()->enqueue_useful_gc_barrier(this, in); } if (ReduceFieldZeroing && dead->is_Load() && i == MemNode::Memory && in->is_Proj() && in->in(0) != nullptr && in->in(0)->is_Initialize()) { // A Load that directly follows an InitializeNode is // going away. The Stores that follow are candidates // again to be captured by the InitializeNode. for (DUIterator_Fast jmax, j = in->fast_outs(jmax); j < jmax; j++) { Node *n = in->fast_out(j); if (n->is_Store()) { _worklist.push(n); } } } } // if (in != nullptr && in != C->top()) } // for (uint i = 0; i < dead->req(); i++) if (recurse) { continue; } } // if (!dead->is_Con()) } // if (progress_state == PROCESS_INPUTS) // Aggressively kill globally dead uses // (Rather than pushing all the outs at once, we push one at a time, // plus the parent to resume later, because of the indefinite number // of edge deletions per loop trip.) if (dead->outcnt() > 0) { // Recursively remove output edges stack.push(dead->raw_out(0), PROCESS_INPUTS); } else { // Finished disconnecting all input and output edges. stack.pop(); // Remove dead node from iterative worklist _worklist.remove(dead); C->remove_useless_node(dead); } } // while (stack.is_nonempty()) } //------------------------------subsume_node----------------------------------- // Remove users from node 'old' and add them to node 'nn'. void PhaseIterGVN::subsume_node( Node *old, Node *nn ) { if (old->Opcode() == Op_SafePoint) { old->as_SafePoint()->disconnect_from_root(this); } assert( old != hash_find(old), "should already been removed" ); assert( old != C->top(), "cannot subsume top node"); // Copy debug or profile information to the new version: C->copy_node_notes_to(nn, old); // Move users of node 'old' to node 'nn' for (DUIterator_Last imin, i = old->last_outs(imin); i >= imin; ) { Node* use = old->last_out(i); // for each use... // use might need re-hashing (but it won't if it's a new node) rehash_node_delayed(use); // Update use-def info as well // We remove all occurrences of old within use->in, // so as to avoid rehashing any node more than once. // The hash table probe swamps any outer loop overhead. uint num_edges = 0; for (uint jmax = use->len(), j = 0; j < jmax; j++) { if (use->in(j) == old) { use->set_req(j, nn); ++num_edges; } } i -= num_edges; // we deleted 1 or more copies of this edge } // Search for instance field data PhiNodes in the same region pointing to the old // memory PhiNode and update their instance memory ids to point to the new node. if (old->is_Phi() && old->as_Phi()->type()->has_memory() && old->in(0) != nullptr) { Node* region = old->in(0); for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) { PhiNode* phi = region->fast_out(i)->isa_Phi(); if (phi != nullptr && phi->inst_mem_id() == (int)old->_idx) { phi->set_inst_mem_id((int)nn->_idx); } } } // Smash all inputs to 'old', isolating him completely Node *temp = new Node(1); temp->init_req(0,nn); // Add a use to nn to prevent him from dying remove_dead_node( old ); temp->del_req(0); // Yank bogus edge if (nn != nullptr && nn->outcnt() == 0) { _worklist.push(nn); } #ifndef PRODUCT if (is_verify_def_use()) { for ( int i = 0; i < _verify_window_size; i++ ) { if ( _verify_window[i] == old ) _verify_window[i] = nn; } } #endif temp->destruct(this); // reuse the _idx of this little guy } //------------------------------add_users_to_worklist-------------------------- void PhaseIterGVN::add_users_to_worklist0(Node* n, Unique_Node_List& worklist) { for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { worklist.push(n->fast_out(i)); // Push on worklist } } // Return counted loop Phi if as a counted loop exit condition, cmp // compares the induction variable with n static PhiNode* countedloop_phi_from_cmp(CmpNode* cmp, Node* n) { for (DUIterator_Fast imax, i = cmp->fast_outs(imax); i < imax; i++) { Node* bol = cmp->fast_out(i); for (DUIterator_Fast i2max, i2 = bol->fast_outs(i2max); i2 < i2max; i2++) { Node* iff = bol->fast_out(i2); if (iff->is_BaseCountedLoopEnd()) { BaseCountedLoopEndNode* cle = iff->as_BaseCountedLoopEnd(); if (cle->limit() == n) { PhiNode* phi = cle->phi(); if (phi != nullptr) { return phi; } } } } } return nullptr; } void PhaseIterGVN::add_users_to_worklist(Node *n) { add_users_to_worklist0(n, _worklist); Unique_Node_List& worklist = _worklist; // Move users of node to worklist for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { Node* use = n->fast_out(i); // Get use add_users_of_use_to_worklist(n, use, worklist); } } void PhaseIterGVN::add_users_of_use_to_worklist(Node* n, Node* use, Unique_Node_List& worklist) { if(use->is_Multi() || // Multi-definer? Push projs on worklist use->is_Store() ) // Enable store/load same address add_users_to_worklist0(use, worklist); // If we changed the receiver type to a call, we need to revisit // the Catch following the call. It's looking for a non-null // receiver to know when to enable the regular fall-through path // in addition to the NullPtrException path. if (use->is_CallDynamicJava() && n == use->in(TypeFunc::Parms)) { Node* p = use->as_CallDynamicJava()->proj_out_or_null(TypeFunc::Control); if (p != nullptr) { add_users_to_worklist0(p, worklist); } } uint use_op = use->Opcode(); if(use->is_Cmp()) { // Enable CMP/BOOL optimization add_users_to_worklist0(use, worklist); // Put Bool on worklist if (use->outcnt() > 0) { Node* bol = use->raw_out(0); if (bol->outcnt() > 0) { Node* iff = bol->raw_out(0); if (iff->outcnt() == 2) { // Look for the 'is_x2logic' pattern: "x ? : 0 : 1" and put the // phi merging either 0 or 1 onto the worklist Node* ifproj0 = iff->raw_out(0); Node* ifproj1 = iff->raw_out(1); if (ifproj0->outcnt() > 0 && ifproj1->outcnt() > 0) { Node* region0 = ifproj0->raw_out(0); Node* region1 = ifproj1->raw_out(0); if( region0 == region1 ) add_users_to_worklist0(region0, worklist); } } } } if (use_op == Op_CmpI || use_op == Op_CmpL) { Node* phi = countedloop_phi_from_cmp(use->as_Cmp(), n); if (phi != nullptr) { // Input to the cmp of a loop exit check has changed, thus // the loop limit may have changed, which can then change the // range values of the trip-count Phi. worklist.push(phi); } } if (use_op == Op_CmpI) { Node* cmp = use; Node* in1 = cmp->in(1); Node* in2 = cmp->in(2); // Notify CmpI / If pattern from CastIINode::Value (left pattern). // Must also notify if in1 is modified and possibly turns into X (right pattern). // // in1 in2 in1 in2 // | | | | // +--- | --+ | | // | | | | | // CmpINode | CmpINode // | | | // BoolNode | BoolNode // | | OR | // IfNode | IfNode // | | | // IfProj | IfProj X // | | | | // CastIINode CastIINode // if (in1 != in2) { // if they are equal, the CmpI can fold them away if (in1 == n) { // in1 modified -> could turn into X -> do traversal based on right pattern. for (DUIterator_Fast i2max, i2 = cmp->fast_outs(i2max); i2 < i2max; i2++) { Node* bol = cmp->fast_out(i2); // For each Bool if (bol->is_Bool()) { for (DUIterator_Fast i3max, i3 = bol->fast_outs(i3max); i3 < i3max; i3++) { Node* iff = bol->fast_out(i3); // For each If if (iff->is_If()) { for (DUIterator_Fast i4max, i4 = iff->fast_outs(i4max); i4 < i4max; i4++) { Node* if_proj = iff->fast_out(i4); // For each IfProj assert(if_proj->is_IfProj(), "If only has IfTrue and IfFalse as outputs"); for (DUIterator_Fast i5max, i5 = if_proj->fast_outs(i5max); i5 < i5max; i5++) { Node* castii = if_proj->fast_out(i5); // For each CastII if (castii->is_CastII() && castii->as_CastII()->carry_dependency()) { worklist.push(castii); } } } } } } } } else { // Only in2 modified -> can assume X == in2 (left pattern). assert(n == in2, "only in2 modified"); // Find all CastII with input in1. for (DUIterator_Fast jmax, j = in1->fast_outs(jmax); j < jmax; j++) { Node* castii = in1->fast_out(j); if (castii->is_CastII() && castii->as_CastII()->carry_dependency()) { // Find If. if (castii->in(0) != nullptr && castii->in(0)->in(0) != nullptr && castii->in(0)->in(0)->is_If()) { Node* ifnode = castii->in(0)->in(0); // Check that if connects to the cmp if (ifnode->in(1) != nullptr && ifnode->in(1)->is_Bool() && ifnode->in(1)->in(1) == cmp) { worklist.push(castii); } } } } } } } } // If changed Cast input, notify down for Phi, Sub, and Xor - all do "uncast" // Patterns: // ConstraintCast+ -> Sub // ConstraintCast+ -> Phi // ConstraintCast+ -> Xor if (use->is_ConstraintCast()) { auto push_the_uses_to_worklist = [&](Node* n){ if (n->is_Phi() || n->is_Sub() || n->Opcode() == Op_XorI || n->Opcode() == Op_XorL) { worklist.push(n); } }; auto is_boundary = [](Node* n){ return !n->is_ConstraintCast(); }; use->visit_uses(push_the_uses_to_worklist, is_boundary); } // If changed LShift inputs, check RShift users for useless sign-ext if (use_op == Op_LShiftI || use_op == Op_LShiftL) { for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) { Node* u = use->fast_out(i2); if (u->Opcode() == Op_RShiftI || u->Opcode() == Op_RShiftL) worklist.push(u); } } // If changed LShift inputs, check And users for shift and mask (And) operation if (use_op == Op_LShiftI || use_op == Op_LShiftL) { for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) { Node* u = use->fast_out(i2); if (u->Opcode() == Op_AndI || u->Opcode() == Op_AndL) { worklist.push(u); } } } // If changed AddI/SubI inputs, check CmpU for range check optimization. if (use_op == Op_AddI || use_op == Op_SubI) { for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) { Node* u = use->fast_out(i2); if (u->is_Cmp() && (u->Opcode() == Op_CmpU)) { worklist.push(u); } } } // If changed AddP inputs: // - check Stores for loop invariant, and // - if the changed input is the offset, check constant-offset AddP users for // address expression flattening. if (use_op == Op_AddP) { bool offset_changed = n == use->in(AddPNode::Offset); for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) { Node* u = use->fast_out(i2); if (u->is_Mem()) { worklist.push(u); } else if (offset_changed && u->is_AddP() && u->in(AddPNode::Offset)->is_Con()) { worklist.push(u); } } } // If changed initialization activity, check dependent Stores if (use_op == Op_Allocate || use_op == Op_AllocateArray) { InitializeNode* init = use->as_Allocate()->initialization(); if (init != nullptr) { Node* imem = init->proj_out_or_null(TypeFunc::Memory); if (imem != nullptr) add_users_to_worklist0(imem, worklist); } } // If the ValidLengthTest input changes then the fallthrough path out of the AllocateArray may have become dead. // CatchNode::Value() is responsible for killing that path. The CatchNode has to be explicitly enqueued for igvn // to guarantee the change is not missed. if (use_op == Op_AllocateArray && n == use->in(AllocateNode::ValidLengthTest)) { Node* p = use->as_AllocateArray()->proj_out_or_null(TypeFunc::Control); if (p != nullptr) { add_users_to_worklist0(p, worklist); } } if (use_op == Op_Initialize) { Node* imem = use->as_Initialize()->proj_out_or_null(TypeFunc::Memory); if (imem != nullptr) add_users_to_worklist0(imem, worklist); } // Loading the java mirror from a Klass requires two loads and the type // of the mirror load depends on the type of 'n'. See LoadNode::Value(). // LoadBarrier?(LoadP(LoadP(AddP(foo:Klass, #java_mirror)))) BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); bool has_load_barrier_nodes = bs->has_load_barrier_nodes(); if (use_op == Op_LoadP && use->bottom_type()->isa_rawptr()) { for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) { Node* u = use->fast_out(i2); const Type* ut = u->bottom_type(); if (u->Opcode() == Op_LoadP && ut->isa_instptr()) { if (has_load_barrier_nodes) { // Search for load barriers behind the load for (DUIterator_Fast i3max, i3 = u->fast_outs(i3max); i3 < i3max; i3++) { Node* b = u->fast_out(i3); if (bs->is_gc_barrier_node(b)) { worklist.push(b); } } } worklist.push(u); } } } if (use->Opcode() == Op_OpaqueZeroTripGuard) { assert(use->outcnt() <= 1, "OpaqueZeroTripGuard can't be shared"); if (use->outcnt() == 1) { Node* cmp = use->unique_out(); worklist.push(cmp); } } if (use->Opcode() == Op_AddX) { for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) { Node* u = use->fast_out(i2); if (u->Opcode() == Op_CastX2P) { worklist.push(u); } } } /* AndNode has a special handling when one of the operands is a LShiftNode: * (LHS << s) & RHS * if RHS fits in less than s bits, the value of this expression is 0. * The difficulty is that there might be a conversion node (ConvI2L) between * the LShiftINode and the AndLNode, like so: * AndLNode(ConvI2L(LShiftI(LHS, s)), RHS) * This case is handled by And[IL]Node::Value(PhaseGVN*) * (see `AndIL_min_trailing_zeros`). * * But, when the shift is updated during IGVN, pushing the user (ConvI2L) * is not enough: there might be no update happening there. We need to * directly push the And[IL]Node on the worklist, jumping over ConvI2L. * * Moreover we can have ConstraintCasts in between. It may look like * ConstraintCast+ -> ConvI2L -> ConstraintCast+ -> And * and And[IL]Node::Value(PhaseGVN*) still handles that by looking through casts. * So we must deal with that as well. */ if (use->is_ConstraintCast() || use_op == Op_ConvI2L) { auto is_boundary = [](Node* n){ return !n->is_ConstraintCast() && n->Opcode() != Op_ConvI2L; }; auto push_and_to_worklist = [&worklist](Node* n){ if (n->Opcode() == Op_AndL || n->Opcode() == Op_AndI) { worklist.push(n); } }; use->visit_uses(push_and_to_worklist, is_boundary); } } /** * Remove the speculative part of all types that we know of */ void PhaseIterGVN::remove_speculative_types() { assert(UseTypeSpeculation, "speculation is off"); for (uint i = 0; i < _types.Size(); i++) { const Type* t = _types.fast_lookup(i); if (t != nullptr) { _types.map(i, t->remove_speculative()); } } _table.check_no_speculative_types(); } // Check if the type of a divisor of a Div or Mod node includes zero. bool PhaseIterGVN::no_dependent_zero_check(Node* n) const { switch (n->Opcode()) { case Op_DivI: case Op_ModI: case Op_UDivI: case Op_UModI: { // Type of divisor includes 0? if (type(n->in(2)) == Type::TOP) { // 'n' is dead. Treat as if zero check is still there to avoid any further optimizations. return false; } const TypeInt* type_divisor = type(n->in(2))->is_int(); return (type_divisor->_hi < 0 || type_divisor->_lo > 0); } case Op_DivL: case Op_ModL: case Op_UDivL: case Op_UModL: { // Type of divisor includes 0? if (type(n->in(2)) == Type::TOP) { // 'n' is dead. Treat as if zero check is still there to avoid any further optimizations. return false; } const TypeLong* type_divisor = type(n->in(2))->is_long(); return (type_divisor->_hi < 0 || type_divisor->_lo > 0); } } return true; } //============================================================================= #ifndef PRODUCT uint PhaseCCP::_total_invokes = 0; uint PhaseCCP::_total_constants = 0; #endif //------------------------------PhaseCCP--------------------------------------- // Conditional Constant Propagation, ala Wegman & Zadeck PhaseCCP::PhaseCCP( PhaseIterGVN *igvn ) : PhaseIterGVN(igvn) { NOT_PRODUCT( clear_constants(); ) assert( _worklist.size() == 0, "" ); analyze(); } #ifndef PRODUCT //------------------------------~PhaseCCP-------------------------------------- PhaseCCP::~PhaseCCP() { inc_invokes(); _total_constants += count_constants(); } #endif #ifdef ASSERT void PhaseCCP::verify_type(Node* n, const Type* tnew, const Type* told) { if (tnew->meet(told) != tnew->remove_speculative()) { n->dump(1); tty->print("told = "); told->dump(); tty->cr(); tty->print("tnew = "); tnew->dump(); tty->cr(); fatal("Not monotonic"); } assert(!told->isa_int() || !tnew->isa_int() || told->is_int()->_widen <= tnew->is_int()->_widen, "widen increases"); assert(!told->isa_long() || !tnew->isa_long() || told->is_long()->_widen <= tnew->is_long()->_widen, "widen increases"); } #endif //ASSERT // In this analysis, all types are initially set to TOP. We iteratively call Value() on all nodes of the graph until // we reach a fixed-point (i.e. no types change anymore). We start with a list that only contains the root node. Each time // a new type is set, we push all uses of that node back to the worklist (in some cases, we also push grandchildren // or nodes even further down back to the worklist because their type could change as a result of the current type // change). void PhaseCCP::analyze() { // Initialize all types to TOP, optimistic analysis for (uint i = 0; i < C->unique(); i++) { _types.map(i, Type::TOP); } // CCP worklist is placed on a local arena, so that we can allow ResourceMarks on "Compile::current()->resource_arena()". // We also do not want to put the worklist on "Compile::current()->comp_arena()", as that one only gets de-allocated after // Compile is over. The local arena gets de-allocated at the end of its scope. ResourceArea local_arena(mtCompiler); Unique_Node_List worklist(&local_arena); DEBUG_ONLY(Unique_Node_List worklist_verify(&local_arena);) // Push root onto worklist worklist.push(C->root()); assert(_root_and_safepoints.size() == 0, "must be empty (unused)"); _root_and_safepoints.push(C->root()); // Pull from worklist; compute new value; push changes out. // This loop is the meat of CCP. while (worklist.size() != 0) { Node* n = fetch_next_node(worklist); DEBUG_ONLY(worklist_verify.push(n);) if (n->is_SafePoint()) { // Make sure safepoints are processed by PhaseCCP::transform even if they are // not reachable from the bottom. Otherwise, infinite loops would be removed. _root_and_safepoints.push(n); } const Type* new_type = n->Value(this); if (new_type != type(n)) { DEBUG_ONLY(verify_type(n, new_type, type(n));) dump_type_and_node(n, new_type); set_type(n, new_type); push_child_nodes_to_worklist(worklist, n); } if (KillPathsReachableByDeadTypeNode && n->is_Type() && new_type == Type::TOP) { // Keep track of Type nodes to kill CFG paths that use Type // nodes that become dead. _maybe_top_type_nodes.push(n); } } DEBUG_ONLY(verify_analyze(worklist_verify);) } #ifdef ASSERT // For every node n on verify list, check if type(n) == n->Value() // We have a list of exceptions, see comments in verify_Value_for. void PhaseCCP::verify_analyze(Unique_Node_List& worklist_verify) { bool failure = false; while (worklist_verify.size()) { Node* n = worklist_verify.pop(); failure |= verify_Value_for(n); } // If we get this assert, check why the reported nodes were not processed again in CCP. // We should either make sure that these nodes are properly added back to the CCP worklist // in PhaseCCP::push_child_nodes_to_worklist() to update their type or add an exception // in the verification code above if that is not possible for some reason (like Load nodes). assert(!failure, "PhaseCCP not at fixpoint: analysis result may be unsound."); } #endif // Fetch next node from worklist to be examined in this iteration. Node* PhaseCCP::fetch_next_node(Unique_Node_List& worklist) { if (StressCCP) { return worklist.remove(C->random() % worklist.size()); } else { return worklist.pop(); } } #ifndef PRODUCT void PhaseCCP::dump_type_and_node(const Node* n, const Type* t) { if (TracePhaseCCP) { t->dump(); do { tty->print("\t"); } while (tty->position() < 16); n->dump(); } } #endif // We need to propagate the type change of 'n' to all its uses. Depending on the kind of node, additional nodes // (grandchildren or even further down) need to be revisited as their types could also be improved as a result // of the new type of 'n'. Push these nodes to the worklist. void PhaseCCP::push_child_nodes_to_worklist(Unique_Node_List& worklist, Node* n) const { for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { Node* use = n->fast_out(i); push_if_not_bottom_type(worklist, use); push_more_uses(worklist, n, use); } } void PhaseCCP::push_if_not_bottom_type(Unique_Node_List& worklist, Node* n) const { if (n->bottom_type() != type(n)) { worklist.push(n); } } // For some nodes, we need to propagate the type change to grandchildren or even further down. // Add them back to the worklist. void PhaseCCP::push_more_uses(Unique_Node_List& worklist, Node* parent, const Node* use) const { push_phis(worklist, use); push_catch(worklist, use); push_cmpu(worklist, use); push_counted_loop_phi(worklist, parent, use); push_loadp(worklist, use); push_and(worklist, parent, use); push_cast_ii(worklist, parent, use); push_opaque_zero_trip_guard(worklist, use); } // We must recheck Phis too if use is a Region. void PhaseCCP::push_phis(Unique_Node_List& worklist, const Node* use) const { if (use->is_Region()) { for (DUIterator_Fast imax, i = use->fast_outs(imax); i < imax; i++) { push_if_not_bottom_type(worklist, use->fast_out(i)); } } } // If we changed the receiver type to a call, we need to revisit the Catch node following the call. It's looking for a // non-null receiver to know when to enable the regular fall-through path in addition to the NullPtrException path. // Same is true if the type of a ValidLengthTest input to an AllocateArrayNode changes. void PhaseCCP::push_catch(Unique_Node_List& worklist, const Node* use) { if (use->is_Call()) { for (DUIterator_Fast imax, i = use->fast_outs(imax); i < imax; i++) { Node* proj = use->fast_out(i); if (proj->is_Proj() && proj->as_Proj()->_con == TypeFunc::Control) { Node* catch_node = proj->find_out_with(Op_Catch); if (catch_node != nullptr) { worklist.push(catch_node); } } } } } // CmpU nodes can get their type information from two nodes up in the graph (instead of from the nodes immediately // above). Make sure they are added to the worklist if nodes they depend on are updated since they could be missed // and get wrong types otherwise. void PhaseCCP::push_cmpu(Unique_Node_List& worklist, const Node* use) const { uint use_op = use->Opcode(); if (use_op == Op_AddI || use_op == Op_SubI) { for (DUIterator_Fast imax, i = use->fast_outs(imax); i < imax; i++) { Node* cmpu = use->fast_out(i); const uint cmpu_opcode = cmpu->Opcode(); if (cmpu_opcode == Op_CmpU || cmpu_opcode == Op_CmpU3) { // Got a CmpU or CmpU3 which might need the new type information from node n. push_if_not_bottom_type(worklist, cmpu); } } } } // If n is used in a counted loop exit condition, then the type of the counted loop's Phi depends on the type of 'n'. // Seem PhiNode::Value(). void PhaseCCP::push_counted_loop_phi(Unique_Node_List& worklist, Node* parent, const Node* use) { uint use_op = use->Opcode(); if (use_op == Op_CmpI || use_op == Op_CmpL) { PhiNode* phi = countedloop_phi_from_cmp(use->as_Cmp(), parent); if (phi != nullptr) { worklist.push(phi); } } } // Loading the java mirror from a Klass requires two loads and the type of the mirror load depends on the type of 'n'. // See LoadNode::Value(). void PhaseCCP::push_loadp(Unique_Node_List& worklist, const Node* use) const { BarrierSetC2* barrier_set = BarrierSet::barrier_set()->barrier_set_c2(); bool has_load_barrier_nodes = barrier_set->has_load_barrier_nodes(); if (use->Opcode() == Op_LoadP && use->bottom_type()->isa_rawptr()) { for (DUIterator_Fast imax, i = use->fast_outs(imax); i < imax; i++) { Node* loadp = use->fast_out(i); const Type* ut = loadp->bottom_type(); if (loadp->Opcode() == Op_LoadP && ut->isa_instptr() && ut != type(loadp)) { if (has_load_barrier_nodes) { // Search for load barriers behind the load push_load_barrier(worklist, barrier_set, loadp); } worklist.push(loadp); } } } } void PhaseCCP::push_load_barrier(Unique_Node_List& worklist, const BarrierSetC2* barrier_set, const Node* use) { for (DUIterator_Fast imax, i = use->fast_outs(imax); i < imax; i++) { Node* barrier_node = use->fast_out(i); if (barrier_set->is_gc_barrier_node(barrier_node)) { worklist.push(barrier_node); } } } // AndI/L::Value() optimizes patterns similar to (v << 2) & 3, or CON & 3 to zero if they are bitwise disjoint. // Add the AndI/L nodes back to the worklist to re-apply Value() in case the value is now a constant or shift // value changed. void PhaseCCP::push_and(Unique_Node_List& worklist, const Node* parent, const Node* use) const { const TypeInteger* parent_type = type(parent)->isa_integer(type(parent)->basic_type()); uint use_op = use->Opcode(); if ( // Pattern: parent (now constant) -> (ConstraintCast | ConvI2L)* -> And (parent_type != nullptr && parent_type->is_con()) || // Pattern: parent -> LShift (use) -> (ConstraintCast | ConvI2L)* -> And ((use_op == Op_LShiftI || use_op == Op_LShiftL) && use->in(2) == parent)) { auto push_and_uses_to_worklist = [&](Node* n) { uint opc = n->Opcode(); if (opc == Op_AndI || opc == Op_AndL) { push_if_not_bottom_type(worklist, n); } }; auto is_boundary = [](Node* n) { return !(n->is_ConstraintCast() || n->Opcode() == Op_ConvI2L); }; use->visit_uses(push_and_uses_to_worklist, is_boundary); } } // CastII::Value() optimizes CmpI/If patterns if the right input of the CmpI has a constant type. If the CastII input is // the same node as the left input into the CmpI node, the type of the CastII node can be improved accordingly. Add the // CastII node back to the worklist to re-apply Value() to either not miss this optimization or to undo it because it // cannot be applied anymore. We could have optimized the type of the CastII before but now the type of the right input // of the CmpI (i.e. 'parent') is no longer constant. The type of the CastII must be widened in this case. void PhaseCCP::push_cast_ii(Unique_Node_List& worklist, const Node* parent, const Node* use) const { if (use->Opcode() == Op_CmpI && use->in(2) == parent) { Node* other_cmp_input = use->in(1); for (DUIterator_Fast imax, i = other_cmp_input->fast_outs(imax); i < imax; i++) { Node* cast_ii = other_cmp_input->fast_out(i); if (cast_ii->is_CastII()) { push_if_not_bottom_type(worklist, cast_ii); } } } } void PhaseCCP::push_opaque_zero_trip_guard(Unique_Node_List& worklist, const Node* use) const { if (use->Opcode() == Op_OpaqueZeroTripGuard) { push_if_not_bottom_type(worklist, use->unique_out()); } } //------------------------------do_transform----------------------------------- // Top level driver for the recursive transformer void PhaseCCP::do_transform() { // Correct leaves of new-space Nodes; they point to old-space. C->set_root( transform(C->root())->as_Root() ); assert( C->top(), "missing TOP node" ); assert( C->root(), "missing root" ); } //------------------------------transform-------------------------------------- // Given a Node in old-space, clone him into new-space. // Convert any of his old-space children into new-space children. Node *PhaseCCP::transform( Node *n ) { assert(n->is_Root(), "traversal must start at root"); assert(_root_and_safepoints.member(n), "root (n) must be in list"); ResourceMark rm; // Map: old node idx -> node after CCP (or nullptr if not yet transformed or useless). Node_List node_map; // Pre-allocate to avoid frequent realloc GrowableArray transform_stack(C->live_nodes() >> 1); // track all visited nodes, so that we can remove the complement Unique_Node_List useful; if (KillPathsReachableByDeadTypeNode) { for (uint i = 0; i < _maybe_top_type_nodes.size(); ++i) { Node* type_node = _maybe_top_type_nodes.at(i); if (type(type_node) == Type::TOP) { ResourceMark rm; type_node->as_Type()->make_paths_from_here_dead(this, nullptr, "ccp"); } } } else { assert(_maybe_top_type_nodes.size() == 0, "we don't need type nodes"); } // Initialize the traversal. // This CCP pass may prove that no exit test for a loop ever succeeds (i.e. the loop is infinite). In that case, // the logic below doesn't follow any path from Root to the loop body: there's at least one such path but it's proven // never taken (its type is TOP). As a consequence the node on the exit path that's input to Root (let's call it n) is // replaced by the top node and the inputs of that node n are not enqueued for further processing. If CCP only works // through the graph from Root, this causes the loop body to never be processed here even when it's not dead (that // is reachable from Root following its uses). To prevent that issue, transform() starts walking the graph from Root // and all safepoints. for (uint i = 0; i < _root_and_safepoints.size(); ++i) { Node* nn = _root_and_safepoints.at(i); Node* new_node = node_map[nn->_idx]; assert(new_node == nullptr, ""); new_node = transform_once(nn); // Check for constant node_map.map(nn->_idx, new_node); // Flag as having been cloned transform_stack.push(new_node); // Process children of cloned node useful.push(new_node); } while (transform_stack.is_nonempty()) { Node* clone = transform_stack.pop(); uint cnt = clone->req(); for( uint i = 0; i < cnt; i++ ) { // For all inputs do Node *input = clone->in(i); if( input != nullptr ) { // Ignore nulls Node *new_input = node_map[input->_idx]; // Check for cloned input node if( new_input == nullptr ) { new_input = transform_once(input); // Check for constant node_map.map( input->_idx, new_input );// Flag as having been cloned transform_stack.push(new_input); // Process children of cloned node useful.push(new_input); } assert( new_input == clone->in(i), "insanity check"); } } } // The above transformation might lead to subgraphs becoming unreachable from the // bottom while still being reachable from the top. As a result, nodes in that // subgraph are not transformed and their bottom types are not updated, leading to // an inconsistency between bottom_type() and type(). In rare cases, LoadNodes in // such a subgraph, might be re-enqueued for IGVN indefinitely by MemNode::Ideal_common // because their address type is inconsistent. Therefore, we aggressively remove // all useless nodes here even before PhaseIdealLoop::build_loop_late gets a chance // to remove them anyway. if (C->cached_top_node()) { useful.push(C->cached_top_node()); } C->update_dead_node_list(useful); remove_useless_nodes(useful.member_set()); _worklist.remove_useless_nodes(useful.member_set()); C->disconnect_useless_nodes(useful, _worklist, &_root_and_safepoints); Node* new_root = node_map[n->_idx]; assert(new_root->is_Root(), "transformed root node must be a root node"); return new_root; } //------------------------------transform_once--------------------------------- // For PhaseCCP, transformation is IDENTITY unless Node computed a constant. Node *PhaseCCP::transform_once( Node *n ) { const Type *t = type(n); // Constant? Use constant Node instead if( t->singleton() ) { Node *nn = n; // Default is to return the original constant if( t == Type::TOP ) { // cache my top node on the Compile instance if( C->cached_top_node() == nullptr || C->cached_top_node()->in(0) == nullptr ) { C->set_cached_top_node(ConNode::make(Type::TOP)); set_type(C->top(), Type::TOP); } nn = C->top(); } if( !n->is_Con() ) { if( t != Type::TOP ) { nn = makecon(t); // ConNode::make(t); NOT_PRODUCT( inc_constants(); ) } else if( n->is_Region() ) { // Unreachable region // Note: nn == C->top() n->set_req(0, nullptr); // Cut selfreference bool progress = true; uint max = n->outcnt(); DUIterator i; while (progress) { progress = false; // Eagerly remove dead phis to avoid phis copies creation. for (i = n->outs(); n->has_out(i); i++) { Node* m = n->out(i); if (m->is_Phi()) { assert(type(m) == Type::TOP, "Unreachable region should not have live phis."); replace_node(m, nn); if (max != n->outcnt()) { progress = true; i = n->refresh_out_pos(i); max = n->outcnt(); } } } } } replace_node(n,nn); // Update DefUse edges for new constant } return nn; } // If x is a TypeNode, capture any more-precise type permanently into Node if (t != n->bottom_type()) { hash_delete(n); // changing bottom type may force a rehash n->raise_bottom_type(t); _worklist.push(n); // n re-enters the hash table via the worklist } // TEMPORARY fix to ensure that 2nd GVN pass eliminates null checks switch( n->Opcode() ) { case Op_CallStaticJava: // Give post-parse call devirtualization a chance case Op_CallDynamicJava: case Op_FastLock: // Revisit FastLocks for lock coarsening case Op_If: case Op_CountedLoopEnd: case Op_Region: case Op_Loop: case Op_CountedLoop: case Op_Conv2B: case Op_Opaque1: _worklist.push(n); break; default: break; } return n; } //---------------------------------saturate------------------------------------ const Type* PhaseCCP::saturate(const Type* new_type, const Type* old_type, const Type* limit_type) const { const Type* wide_type = new_type->widen(old_type, limit_type); if (wide_type != new_type) { // did we widen? // If so, we may have widened beyond the limit type. Clip it back down. new_type = wide_type->filter(limit_type); } return new_type; } //------------------------------print_statistics------------------------------- #ifndef PRODUCT void PhaseCCP::print_statistics() { tty->print_cr("CCP: %d constants found: %d", _total_invokes, _total_constants); } #endif //============================================================================= #ifndef PRODUCT uint PhasePeephole::_total_peepholes = 0; #endif //------------------------------PhasePeephole---------------------------------- // Conditional Constant Propagation, ala Wegman & Zadeck PhasePeephole::PhasePeephole( PhaseRegAlloc *regalloc, PhaseCFG &cfg ) : PhaseTransform(Peephole), _regalloc(regalloc), _cfg(cfg) { NOT_PRODUCT( clear_peepholes(); ) } #ifndef PRODUCT //------------------------------~PhasePeephole--------------------------------- PhasePeephole::~PhasePeephole() { _total_peepholes += count_peepholes(); } #endif //------------------------------transform-------------------------------------- Node *PhasePeephole::transform( Node *n ) { ShouldNotCallThis(); return nullptr; } //------------------------------do_transform----------------------------------- void PhasePeephole::do_transform() { bool method_name_not_printed = true; // Examine each basic block for (uint block_number = 1; block_number < _cfg.number_of_blocks(); ++block_number) { Block* block = _cfg.get_block(block_number); bool block_not_printed = true; for (bool progress = true; progress;) { progress = false; // block->end_idx() not valid after PhaseRegAlloc uint end_index = block->number_of_nodes(); for( uint instruction_index = end_index - 1; instruction_index > 0; --instruction_index ) { Node *n = block->get_node(instruction_index); if( n->is_Mach() ) { MachNode *m = n->as_Mach(); // check for peephole opportunities int result = m->peephole(block, instruction_index, &_cfg, _regalloc); if( result != -1 ) { #ifndef PRODUCT if( PrintOptoPeephole ) { // Print method, first time only if( C->method() && method_name_not_printed ) { C->method()->print_short_name(); tty->cr(); method_name_not_printed = false; } // Print this block if( Verbose && block_not_printed) { tty->print_cr("in block"); block->dump(); block_not_printed = false; } // Print the peephole number tty->print_cr("peephole number: %d", result); } inc_peepholes(); #endif // Set progress, start again progress = true; break; } } } } } } //------------------------------print_statistics------------------------------- #ifndef PRODUCT void PhasePeephole::print_statistics() { tty->print_cr("Peephole: peephole rules applied: %d", _total_peepholes); } #endif //============================================================================= //------------------------------set_req_X-------------------------------------- void Node::set_req_X( uint i, Node *n, PhaseIterGVN *igvn ) { assert( is_not_dead(n), "can not use dead node"); #ifdef ASSERT if (igvn->hash_find(this) == this) { tty->print_cr("Need to remove from hash before changing edges"); this->dump(1); tty->print_cr("Set at i = %d", i); n->dump(); assert(false, "Need to remove from hash before changing edges"); } #endif Node *old = in(i); set_req(i, n); // old goes dead? if( old ) { switch (old->outcnt()) { case 0: // Put into the worklist to kill later. We do not kill it now because the // recursive kill will delete the current node (this) if dead-loop exists if (!old->is_top()) igvn->_worklist.push( old ); break; case 1: if( old->is_Store() || old->has_special_unique_user() ) igvn->add_users_to_worklist( old ); break; case 2: if( old->is_Store() ) igvn->add_users_to_worklist( old ); if( old->Opcode() == Op_Region ) igvn->_worklist.push(old); break; case 3: if( old->Opcode() == Op_Region ) { igvn->_worklist.push(old); igvn->add_users_to_worklist( old ); } break; default: break; } BarrierSet::barrier_set()->barrier_set_c2()->enqueue_useful_gc_barrier(igvn, old); } } void Node::set_req_X(uint i, Node *n, PhaseGVN *gvn) { PhaseIterGVN* igvn = gvn->is_IterGVN(); if (igvn == nullptr) { set_req(i, n); return; } set_req_X(i, n, igvn); } //-------------------------------replace_by----------------------------------- // Using def-use info, replace one node for another. Follow the def-use info // to all users of the OLD node. Then make all uses point to the NEW node. void Node::replace_by(Node *new_node) { assert(!is_top(), "top node has no DU info"); for (DUIterator_Last imin, i = last_outs(imin); i >= imin; ) { Node* use = last_out(i); uint uses_found = 0; for (uint j = 0; j < use->len(); j++) { if (use->in(j) == this) { if (j < use->req()) use->set_req(j, new_node); else use->set_prec(j, new_node); uses_found++; } } i -= uses_found; // we deleted 1 or more copies of this edge } } //============================================================================= //----------------------------------------------------------------------------- void Type_Array::grow( uint i ) { assert(_a == Compile::current()->comp_arena(), "Should be allocated in comp_arena"); if( !_max ) { _max = 1; _types = (const Type**)_a->Amalloc( _max * sizeof(Type*) ); _types[0] = nullptr; } uint old = _max; _max = next_power_of_2(i); _types = (const Type**)_a->Arealloc( _types, old*sizeof(Type*),_max*sizeof(Type*)); memset( &_types[old], 0, (_max-old)*sizeof(Type*) ); } //------------------------------dump------------------------------------------- #ifndef PRODUCT void Type_Array::dump() const { uint max = Size(); for( uint i = 0; i < max; i++ ) { if( _types[i] != nullptr ) { tty->print(" %d\t== ", i); _types[i]->dump(); tty->cr(); } } } #endif