/* * Copyright (c) 2011, 2023, 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 "precompiled.hpp" #include "memory/allocation.hpp" #include "opto/loopnode.hpp" #include "opto/addnode.hpp" #include "opto/callnode.hpp" #include "opto/castnode.hpp" #include "opto/connode.hpp" #include "opto/convertnode.hpp" #include "opto/loopnode.hpp" #include "opto/matcher.hpp" #include "opto/mulnode.hpp" #include "opto/opaquenode.hpp" #include "opto/predicates.hpp" #include "opto/rootnode.hpp" #include "opto/subnode.hpp" #include #include /* * The general idea of Loop Predication is to hoist a check inside a loop body by inserting a Hoisted Check Predicate with * an uncommon trap on the entry path to the loop. The old check inside the loop can be eliminated. If the condition of * the Hoisted Check Predicate fails at runtime, we'll execute the uncommon trap to avoid entering the loop which misses * the check. Loop Predication can currently remove array range checks and loop invariant checks (such as null checks). * * On top of these predicates added by Loop Predication, there are other kinds of predicates. A detailed description * about all predicates can be found in predicates.hpp. */ //-------------------------------register_control------------------------- void PhaseIdealLoop::register_control(Node* n, IdealLoopTree *loop, Node* pred, bool update_body) { assert(n->is_CFG(), "msust be control node"); _igvn.register_new_node_with_optimizer(n); if (update_body) { loop->_body.push(n); } set_loop(n, loop); // When called from beautify_loops() idom is not constructed yet. if (_idom != nullptr) { set_idom(n, pred, dom_depth(pred)); } } //------------------------------create_new_if_for_predicate------------------------ // create a new if above the uct_if_pattern for the predicate to be promoted. // // before after // ---------- ---------- // ctrl ctrl // | | // | | // v v // iff new_iff // / \ / \ // / \ / \ // v v v v // uncommon_proj cont_proj if_uct if_cont // \ | | | | // \ | | | | // v v v | v // rgn loop | iff // | | / \ // | | / \ // v | v v // uncommon_trap | uncommon_proj cont_proj // \ \ | | // \ \ | | // v v v v // rgn loop // | // | // v // uncommon_trap // // // We will create a region to guard the uct call if there is no one there. // The continuation projection (if_cont) of the new_iff is returned which // is an IfTrue projection. This code is also used to clone predicates to cloned loops. IfProjNode* PhaseIdealLoop::create_new_if_for_predicate(ParsePredicateSuccessProj* parse_predicate_proj, Node* new_entry, Deoptimization::DeoptReason reason, const int opcode, const bool rewire_uncommon_proj_phi_inputs) { assert(parse_predicate_proj->is_uncommon_trap_if_pattern(reason), "must be a uct if pattern!"); ParsePredicateNode* parse_predicate = parse_predicate_proj->in(0)->as_ParsePredicate(); ProjNode* uncommon_proj = parse_predicate->proj_out(false); Node* uct_region = uncommon_proj->unique_ctrl_out(); assert(uct_region->is_Region() || uct_region->is_Call(), "must be a region or call uct"); uint proj_index = 1; // region's edge corresponding to uncommon_proj if (!uct_region->is_Region()) { // create a region to guard the call assert(uct_region->is_Call(), "must be call uct"); CallNode* call = uct_region->as_Call(); IdealLoopTree* loop = get_loop(call); uct_region = new RegionNode(1); Node* uncommon_proj_orig = uncommon_proj; uncommon_proj = uncommon_proj->clone()->as_Proj(); register_control(uncommon_proj, loop, parse_predicate); uct_region->add_req(uncommon_proj); register_control(uct_region, loop, uncommon_proj); _igvn.replace_input_of(call, 0, uct_region); // When called from beautify_loops() idom is not constructed yet. if (_idom != nullptr) { set_idom(call, uct_region, dom_depth(uct_region)); } // Move nodes pinned on the projection or whose control is set to // the projection to the region. lazy_replace(uncommon_proj_orig, uct_region); } else { // Find region's edge corresponding to uncommon_proj for (; proj_index < uct_region->req(); proj_index++) if (uct_region->in(proj_index) == uncommon_proj) break; assert(proj_index < uct_region->req(), "sanity"); } Node* entry = parse_predicate->in(0); if (new_entry != nullptr) { // Cloning the predicate to new location. entry = new_entry; } // Create new_iff IdealLoopTree* lp = get_loop(entry); IfNode* new_iff = nullptr; switch (opcode) { case Op_If: new_iff = new IfNode(entry, parse_predicate->in(1), parse_predicate->_prob, parse_predicate->_fcnt); break; case Op_RangeCheck: new_iff = new RangeCheckNode(entry, parse_predicate->in(1), parse_predicate->_prob, parse_predicate->_fcnt); break; case Op_ParsePredicate: new_iff = new ParsePredicateNode(entry, reason, &_igvn); break; default: fatal("no other If variant here"); } register_control(new_iff, lp, entry); IfProjNode* if_cont = new IfTrueNode(new_iff); IfProjNode* if_uct = new IfFalseNode(new_iff); register_control(if_cont, lp, new_iff); register_control(if_uct, get_loop(uct_region), new_iff); _igvn.add_input_to(uct_region, if_uct); // If rgn has phis add new edges which has the same // value as on original uncommon_proj pass. assert(uct_region->in(uct_region->req() - 1) == if_uct, "new edge should be last"); bool has_phi = false; for (DUIterator_Fast imax, i = uct_region->fast_outs(imax); i < imax; i++) { Node* use = uct_region->fast_out(i); if (use->is_Phi() && use->outcnt() > 0) { assert(use->in(0) == uct_region, ""); _igvn.rehash_node_delayed(use); Node* phi_input = use->in(proj_index); if (uncommon_proj->outcnt() > 1 && !phi_input->is_CFG() && !phi_input->is_Phi() && get_ctrl(phi_input) == uncommon_proj) { // There are some control dependent nodes on the uncommon projection. We cannot simply reuse these data nodes. // We either need to rewire them from the old uncommon projection to the newly created uncommon proj (if the old // If is dying) or clone them and update their control (if the old If is not dying). if (rewire_uncommon_proj_phi_inputs) { // Replace phi input for the old uncommon projection with TOP as the If is dying anyways. Reuse the old data // nodes by simply updating control inputs and ctrl. _igvn.replace_input_of(use, proj_index, C->top()); set_ctrl_of_nodes_with_same_ctrl(phi_input, uncommon_proj, if_uct); } else { phi_input = clone_nodes_with_same_ctrl(phi_input, uncommon_proj, if_uct); } } use->add_req(phi_input); has_phi = true; } } assert(!has_phi || uct_region->req() > 3, "no phis when region is created"); if (new_entry == nullptr) { // Attach if_cont to iff _igvn.replace_input_of(parse_predicate, 0, if_cont); if (_idom != nullptr) { set_idom(parse_predicate, if_cont, dom_depth(parse_predicate)); } } // When called from beautify_loops() idom is not constructed yet. if (_idom != nullptr) { Node* ridom = idom(uct_region); Node* nrdom = dom_lca_internal(ridom, new_iff); set_idom(uct_region, nrdom, dom_depth(uct_region)); } return if_cont->as_IfProj(); } // Update ctrl and control inputs of all data nodes starting from 'node' to 'new_ctrl' which have 'old_ctrl' as // current ctrl. void PhaseIdealLoop::set_ctrl_of_nodes_with_same_ctrl(Node* node, ProjNode* old_ctrl, Node* new_ctrl) { Unique_Node_List nodes_with_same_ctrl = find_nodes_with_same_ctrl(node, old_ctrl); for (uint j = 0; j < nodes_with_same_ctrl.size(); j++) { Node* next = nodes_with_same_ctrl[j]; if (next->in(0) == old_ctrl) { _igvn.replace_input_of(next, 0, new_ctrl); } set_ctrl(next, new_ctrl); } } // Recursively find all input nodes with the same ctrl. Unique_Node_List PhaseIdealLoop::find_nodes_with_same_ctrl(Node* node, const ProjNode* ctrl) { Unique_Node_List nodes_with_same_ctrl; nodes_with_same_ctrl.push(node); for (uint j = 0; j < nodes_with_same_ctrl.size(); j++) { Node* next = nodes_with_same_ctrl[j]; for (uint k = 1; k < next->req(); k++) { Node* in = next->in(k); if (!in->is_Phi() && get_ctrl(in) == ctrl) { nodes_with_same_ctrl.push(in); } } } return nodes_with_same_ctrl; } // Clone all nodes with the same ctrl as 'old_ctrl' starting from 'node' by following its inputs. Rewire the cloned nodes // to 'new_ctrl'. Returns the clone of 'node'. Node* PhaseIdealLoop::clone_nodes_with_same_ctrl(Node* node, ProjNode* old_ctrl, Node* new_ctrl) { DEBUG_ONLY(uint last_idx = C->unique();) Unique_Node_List nodes_with_same_ctrl = find_nodes_with_same_ctrl(node, old_ctrl); Dict old_new_mapping = clone_nodes(nodes_with_same_ctrl); // Cloned but not rewired, yet rewire_cloned_nodes_to_ctrl(old_ctrl, new_ctrl, nodes_with_same_ctrl, old_new_mapping); Node* clone_phi_input = static_cast(old_new_mapping[node]); assert(clone_phi_input != nullptr && clone_phi_input->_idx >= last_idx, "must exist and be a proper clone"); return clone_phi_input; } // Clone all the nodes on 'list_to_clone' and return an old->new mapping. Dict PhaseIdealLoop::clone_nodes(const Node_List& list_to_clone) { Dict old_new_mapping(cmpkey, hashkey); for (uint i = 0; i < list_to_clone.size(); i++) { Node* next = list_to_clone[i]; Node* clone = next->clone(); _igvn.register_new_node_with_optimizer(clone); old_new_mapping.Insert(next, clone); } return old_new_mapping; } // Rewire inputs of the unprocessed cloned nodes (inputs are not updated, yet, and still point to the old nodes) by // using the old_new_mapping. void PhaseIdealLoop::rewire_cloned_nodes_to_ctrl(const ProjNode* old_ctrl, Node* new_ctrl, const Node_List& nodes_with_same_ctrl, const Dict& old_new_mapping) { for (uint i = 0; i < nodes_with_same_ctrl.size(); i++) { Node* next = nodes_with_same_ctrl[i]; Node* clone = static_cast(old_new_mapping[next]); if (next->in(0) == old_ctrl) { // All data nodes with a control input to the uncommon projection in the chain need to be rewired to the new uncommon // projection (could not only be the last data node in the chain but also, for example, a DivNode within the chain). _igvn.replace_input_of(clone, 0, new_ctrl); set_ctrl(clone, new_ctrl); } rewire_inputs_of_clones_to_clones(new_ctrl, clone, old_new_mapping, next); } } // Rewire the inputs of the cloned nodes to the old nodes to the new clones. void PhaseIdealLoop::rewire_inputs_of_clones_to_clones(Node* new_ctrl, Node* clone, const Dict& old_new_mapping, const Node* next) { for (uint i = 1; i < next->req(); i++) { Node* in = next->in(i); if (!in->is_Phi()) { assert(!in->is_CFG(), "must be data node"); Node* in_clone = static_cast(old_new_mapping[in]); if (in_clone != nullptr) { _igvn.replace_input_of(clone, i, in_clone); set_ctrl(clone, new_ctrl); } } } } IfProjNode* PhaseIdealLoop::clone_parse_predicate_to_unswitched_loop(ParsePredicateSuccessProj* parse_predicate_proj, Node* new_entry, Deoptimization::DeoptReason reason, const bool slow_loop) { IfProjNode* new_predicate_proj = create_new_if_for_predicate(parse_predicate_proj, new_entry, reason, Op_ParsePredicate, slow_loop); assert(new_predicate_proj->is_IfTrue(), "the success projection of a Parse Predicate is a true projection"); ParsePredicateNode* parse_predicate = new_predicate_proj->in(0)->as_ParsePredicate(); return new_predicate_proj; } // Clones Assertion Predicates to both unswitched loops starting at 'old_predicate_proj' by following its control inputs. // It also rewires the control edges of data nodes with dependencies in the loop from the old predicates to the new // cloned predicates. void PhaseIdealLoop::clone_assertion_predicates_to_unswitched_loop(IdealLoopTree* loop, const Node_List& old_new, Deoptimization::DeoptReason reason, IfProjNode* old_predicate_proj, ParsePredicateSuccessProj* fast_loop_parse_predicate_proj, ParsePredicateSuccessProj* slow_loop_parse_predicate_proj) { assert(fast_loop_parse_predicate_proj->in(0)->is_ParsePredicate() && slow_loop_parse_predicate_proj->in(0)->is_ParsePredicate(), "sanity check"); // Only need to clone range check predicates as those can be changed and duplicated by inserting pre/main/post loops // and doing loop unrolling. Push the original predicates on a list to later process them in reverse order to keep the // original predicate order. Unique_Node_List list; get_assertion_predicates(old_predicate_proj, list); Node_List to_process; IfNode* iff = old_predicate_proj->in(0)->as_If(); IfProjNode* uncommon_proj = iff->proj_out(1 - old_predicate_proj->as_Proj()->_con)->as_IfProj(); // Process in reverse order such that 'create_new_if_for_predicate' can be used in // 'clone_assertion_predicate_for_unswitched_loops' and the original order is maintained. for (int i = list.size() - 1; i >= 0; i--) { Node* predicate = list.at(i); assert(predicate->in(0)->is_If(), "must be If node"); iff = predicate->in(0)->as_If(); assert(predicate->is_Proj() && predicate->as_Proj()->is_IfProj(), "predicate must be a projection of an if node"); IfProjNode* predicate_proj = predicate->as_IfProj(); IfProjNode* fast_proj = clone_assertion_predicate_for_unswitched_loops(iff, predicate_proj, reason, fast_loop_parse_predicate_proj); assert(assertion_predicate_has_loop_opaque_node(fast_proj->in(0)->as_If()), "must find Assertion Predicate for fast loop"); IfProjNode* slow_proj = clone_assertion_predicate_for_unswitched_loops(iff, predicate_proj, reason, slow_loop_parse_predicate_proj); assert(assertion_predicate_has_loop_opaque_node(slow_proj->in(0)->as_If()), "must find Assertion Predicate for slow loop"); // Update control dependent data nodes. for (DUIterator j = predicate->outs(); predicate->has_out(j); j++) { Node* fast_node = predicate->out(j); if (loop->is_member(get_loop(ctrl_or_self(fast_node)))) { assert(fast_node->in(0) == predicate, "only control edge"); Node* slow_node = old_new[fast_node->_idx]; assert(slow_node->in(0) == predicate, "only control edge"); _igvn.replace_input_of(fast_node, 0, fast_proj); to_process.push(slow_node); --j; } } // Have to delay updates to the slow loop so uses of predicate are not modified while we iterate on them. while (to_process.size() > 0) { Node* slow_node = to_process.pop(); _igvn.replace_input_of(slow_node, 0, slow_proj); } } } // Put all Assertion Predicate projections on a list, starting at 'predicate' and going up in the tree. If 'get_opaque' // is set, then the Opaque4 nodes of the Assertion Predicates are put on the list instead of the projections. void PhaseIdealLoop::get_assertion_predicates(Node* predicate, Unique_Node_List& list, bool get_opaque) { ParsePredicateNode* parse_predicate = predicate->in(0)->as_ParsePredicate(); ProjNode* uncommon_proj = parse_predicate->proj_out(1 - predicate->as_Proj()->_con); Node* rgn = uncommon_proj->unique_ctrl_out(); assert(rgn->is_Region() || rgn->is_Call(), "must be a region or call uct"); predicate = parse_predicate->in(0); while (predicate != nullptr && predicate->is_Proj() && predicate->in(0)->is_If()) { IfNode* iff = predicate->in(0)->as_If(); uncommon_proj = iff->proj_out(1 - predicate->as_Proj()->_con); if (uncommon_proj->unique_ctrl_out() != rgn) { break; } if (iff->in(1)->Opcode() == Op_Opaque4 && assertion_predicate_has_loop_opaque_node(iff)) { if (get_opaque) { // Collect the predicate Opaque4 node. list.push(iff->in(1)); } else { // Collect the predicate projection. list.push(predicate); } } predicate = predicate->in(0)->in(0); } } // Clone an Assertion Predicate for an unswitched loop. OpaqueLoopInit and OpaqueLoopStride nodes are cloned and uncommon // traps are kept for the predicate (a Halt node is used later when creating pre/main/post loops and copying this cloned // predicate again). IfProjNode* PhaseIdealLoop::clone_assertion_predicate_for_unswitched_loops(Node* iff, IfProjNode* predicate, Deoptimization::DeoptReason reason, ParsePredicateSuccessProj* parse_predicate_proj) { Node* bol = create_bool_from_template_assertion_predicate(iff, nullptr, nullptr, parse_predicate_proj); IfProjNode* if_proj = create_new_if_for_predicate(parse_predicate_proj, nullptr, reason, iff->Opcode(), false); _igvn.replace_input_of(if_proj->in(0), 1, bol); _igvn.replace_input_of(parse_predicate_proj->in(0), 0, if_proj); set_idom(parse_predicate_proj->in(0), if_proj, dom_depth(if_proj)); return if_proj; } // Clone the old Parse Predicates and Assertion Predicates before the unswitch If to the unswitched loops after the // unswitch If. void PhaseIdealLoop::clone_parse_and_assertion_predicates_to_unswitched_loop(IdealLoopTree* loop, Node_List& old_new, IfProjNode*& iffast_pred, IfProjNode*& ifslow_pred) { LoopNode* head = loop->_head->as_Loop(); Node* entry = head->skip_strip_mined()->in(LoopNode::EntryControl); const Predicates predicates(entry); clone_loop_predication_predicates_to_unswitched_loop(loop, old_new, predicates.loop_predicate_block(), Deoptimization::Reason_predicate, iffast_pred, ifslow_pred); clone_loop_predication_predicates_to_unswitched_loop(loop, old_new, predicates.profiled_loop_predicate_block(), Deoptimization::Reason_profile_predicate, iffast_pred, ifslow_pred); const PredicateBlock* loop_limit_check_predicate_block = predicates.loop_limit_check_predicate_block(); if (loop_limit_check_predicate_block->has_parse_predicate() && !head->is_CountedLoop()) { // Don't clone the Loop Limit Check Parse Predicate if we already have a counted loop (a Loop Limit Check Predicate // is only created when converting a LoopNode to a CountedLoopNode). clone_parse_predicate_to_unswitched_loops(loop_limit_check_predicate_block, Deoptimization::Reason_loop_limit_check, iffast_pred, ifslow_pred); } } // Clone the Parse Predicate and Template Assertion Predicates of a Loop Predication related Predicate Block. void PhaseIdealLoop::clone_loop_predication_predicates_to_unswitched_loop(IdealLoopTree* loop, const Node_List& old_new, const PredicateBlock* predicate_block, Deoptimization::DeoptReason reason, IfProjNode*& iffast_pred, IfProjNode*& ifslow_pred) { if (predicate_block->has_parse_predicate()) { // We currently only clone Assertion Predicates if there are Parse Predicates. This is not entirely correct and will // be changed with the complete fix for Assertion Predicates. clone_parse_predicate_to_unswitched_loops(predicate_block, reason, iffast_pred, ifslow_pred); assert(iffast_pred->in(0)->is_ParsePredicate() && ifslow_pred->in(0)->is_ParsePredicate(), "must be success projections of the cloned Parse Predicates"); clone_assertion_predicates_to_unswitched_loop(loop, old_new, reason, predicate_block->parse_predicate_success_proj(), iffast_pred->as_IfTrue(), ifslow_pred->as_IfTrue()); } } void PhaseIdealLoop::clone_parse_predicate_to_unswitched_loops(const PredicateBlock* predicate_block, Deoptimization::DeoptReason reason, IfProjNode*& iffast_pred, IfProjNode*& ifslow_pred) { assert(predicate_block->has_parse_predicate(), "must have parse predicate"); ParsePredicateSuccessProj* parse_predicate_proj = predicate_block->parse_predicate_success_proj(); iffast_pred = clone_parse_predicate_to_unswitched_loop(parse_predicate_proj, iffast_pred, reason, false); check_cloned_parse_predicate_for_unswitching(iffast_pred, true); ifslow_pred = clone_parse_predicate_to_unswitched_loop(parse_predicate_proj, ifslow_pred, reason, true); check_cloned_parse_predicate_for_unswitching(ifslow_pred, false); } #ifndef PRODUCT void PhaseIdealLoop::check_cloned_parse_predicate_for_unswitching(const Node* new_entry, const bool is_fast_loop) { assert(new_entry != nullptr, "IfTrue or IfFalse after clone predicate"); if (TraceLoopPredicate) { tty->print("Parse Predicate cloned to %s loop: ", is_fast_loop ? "fast" : "slow"); new_entry->in(0)->dump(); } } #endif //------------------------------Invariance----------------------------------- // Helper class for loop_predication_impl to compute invariance on the fly and // clone invariants. class Invariance : public StackObj { VectorSet _visited, _invariant; Node_Stack _stack; VectorSet _clone_visited; Node_List _old_new; // map of old to new (clone) IdealLoopTree* _lpt; PhaseIdealLoop* _phase; Node* _data_dependency_on; // The projection into the loop on which data nodes are dependent or null otherwise // Helper function to set up the invariance for invariance computation // If n is a known invariant, set up directly. Otherwise, look up the // the possibility to push n onto the stack for further processing. void visit(Node* use, Node* n) { if (_lpt->is_invariant(n)) { // known invariant _invariant.set(n->_idx); } else if (!n->is_CFG()) { Node *n_ctrl = _phase->ctrl_or_self(n); Node *u_ctrl = _phase->ctrl_or_self(use); // self if use is a CFG if (_phase->is_dominator(n_ctrl, u_ctrl)) { _stack.push(n, n->in(0) == nullptr ? 1 : 0); } } } // Compute invariance for "the_node" and (possibly) all its inputs recursively // on the fly void compute_invariance(Node* n) { assert(_visited.test(n->_idx), "must be"); visit(n, n); while (_stack.is_nonempty()) { Node* n = _stack.node(); uint idx = _stack.index(); if (idx == n->req()) { // all inputs are processed _stack.pop(); // n is invariant if it's inputs are all invariant bool all_inputs_invariant = true; for (uint i = 0; i < n->req(); i++) { Node* in = n->in(i); if (in == nullptr) continue; assert(_visited.test(in->_idx), "must have visited input"); if (!_invariant.test(in->_idx)) { // bad guy all_inputs_invariant = false; break; } } if (all_inputs_invariant) { // If n's control is a predicate that was moved out of the // loop, it was marked invariant but n is only invariant if // it depends only on that test. Otherwise, unless that test // is out of the loop, it's not invariant. if (n->is_CFG() || n->depends_only_on_test() || n->in(0) == nullptr || !_phase->is_member(_lpt, n->in(0))) { _invariant.set(n->_idx); // I am a invariant too } } } else { // process next input _stack.set_index(idx + 1); Node* m = n->in(idx); if (m != nullptr && !_visited.test_set(m->_idx)) { visit(n, m); } } } } // Helper function to set up _old_new map for clone_nodes. // If n is a known invariant, set up directly ("clone" of n == n). // Otherwise, push n onto the stack for real cloning. void clone_visit(Node* n) { assert(_invariant.test(n->_idx), "must be invariant"); if (_lpt->is_invariant(n)) { // known invariant _old_new.map(n->_idx, n); } else { // to be cloned assert(!n->is_CFG(), "should not see CFG here"); _stack.push(n, n->in(0) == nullptr ? 1 : 0); } } // Clone "n" and (possibly) all its inputs recursively void clone_nodes(Node* n, Node* ctrl) { clone_visit(n); while (_stack.is_nonempty()) { Node* n = _stack.node(); uint idx = _stack.index(); if (idx == n->req()) { // all inputs processed, clone n! _stack.pop(); // clone invariant node Node* n_cl = n->clone(); _old_new.map(n->_idx, n_cl); _phase->register_new_node(n_cl, ctrl); for (uint i = 0; i < n->req(); i++) { Node* in = n_cl->in(i); if (in == nullptr) continue; n_cl->set_req(i, _old_new[in->_idx]); } } else { // process next input _stack.set_index(idx + 1); Node* m = n->in(idx); if (m != nullptr && !_clone_visited.test_set(m->_idx)) { clone_visit(m); // visit the input } } } } public: Invariance(Arena* area, IdealLoopTree* lpt) : _visited(area), _invariant(area), _stack(area, 10 /* guess */), _clone_visited(area), _old_new(area), _lpt(lpt), _phase(lpt->_phase), _data_dependency_on(nullptr) { LoopNode* head = _lpt->_head->as_Loop(); Node* entry = head->skip_strip_mined()->in(LoopNode::EntryControl); if (entry->outcnt() != 1) { // If a node is pinned between the predicates and the loop // entry, we won't be able to move any node in the loop that // depends on it above it in a predicate. Mark all those nodes // as non-loop-invariant. // Loop predication could create new nodes for which the below // invariant information is missing. Mark the 'entry' node to // later check again if a node needs to be treated as non-loop- // invariant as well. _data_dependency_on = entry; Unique_Node_List wq; wq.push(entry); for (uint next = 0; next < wq.size(); ++next) { Node *n = wq.at(next); for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { Node* u = n->fast_out(i); if (!u->is_CFG()) { Node* c = _phase->get_ctrl(u); if (_lpt->is_member(_phase->get_loop(c)) || _phase->is_dominator(c, head)) { _visited.set(u->_idx); wq.push(u); } } } } } } // Did we explicitly mark some nodes non-loop-invariant? If so, return the entry node on which some data nodes // are dependent that prevent loop predication. Otherwise, return null. Node* data_dependency_on() { return _data_dependency_on; } // Map old to n for invariance computation and clone void map_ctrl(Node* old, Node* n) { assert(old->is_CFG() && n->is_CFG(), "must be"); _old_new.map(old->_idx, n); // "clone" of old is n _invariant.set(old->_idx); // old is invariant _clone_visited.set(old->_idx); } // Driver function to compute invariance bool is_invariant(Node* n) { if (!_visited.test_set(n->_idx)) compute_invariance(n); return (_invariant.test(n->_idx) != 0); } // Driver function to clone invariant Node* clone(Node* n, Node* ctrl) { assert(ctrl->is_CFG(), "must be"); assert(_invariant.test(n->_idx), "must be an invariant"); if (!_clone_visited.test(n->_idx)) clone_nodes(n, ctrl); return _old_new[n->_idx]; } }; //------------------------------is_range_check_if ----------------------------------- // Returns true if the predicate of iff is in "scale*iv + offset u< load_range(ptr)" format // Note: this function is particularly designed for loop predication. We require load_range // and offset to be loop invariant computed on the fly by "invar" bool IdealLoopTree::is_range_check_if(IfProjNode* if_success_proj, PhaseIdealLoop *phase, BasicType bt, Node *iv, Node *&range, Node *&offset, jlong &scale) const { IfNode* iff = if_success_proj->in(0)->as_If(); if (!is_loop_exit(iff)) { return false; } if (!iff->in(1)->is_Bool()) { return false; } const BoolNode *bol = iff->in(1)->as_Bool(); if (bol->_test._test != BoolTest::lt || if_success_proj->is_IfFalse()) { // We don't have the required range check pattern: // if (scale*iv + offset =u limit) { // // } else { // trap(); // } // // If we create a Range Check Predicate for this wrong pattern, it could succeed at runtime (i.e. true for the // value of "scale*iv + offset" in the first loop iteration and true for the value of "scale*iv + offset" in the // last loop iteration) while the check to be hoisted could fail in other loop iterations. // // Example: // Loop: "for (int i = -1; i < 1000; i++)" // init = "scale*iv + offset" in the first loop iteration = 1*-1 + 0 = -1 // last = "scale*iv + offset" in the last loop iteration = 1*999 + 0 = 999 // limit = 100 // // Range Check Predicate is always true: // init >=u limit && last >=u limit <=> // -1 >=u 100 && 999 >= u 100 // // But for 0 <= x < 100: x >=u 100 is false. // We would wrongly skip the branch with the trap() and possibly miss to execute some other statements inside that // trap() branch. return false; } if (!bol->in(1)->is_Cmp()) { return false; } const CmpNode *cmp = bol->in(1)->as_Cmp(); if (cmp->Opcode() != Op_Cmp_unsigned(bt)) { return false; } range = cmp->in(2); if (range->Opcode() != Op_LoadRange) { const TypeInteger* tinteger = phase->_igvn.type(range)->isa_integer(bt); if (tinteger == nullptr || tinteger->empty() || tinteger->lo_as_long() < 0) { // Allow predication on positive values that aren't LoadRanges. // This allows optimization of loops where the length of the // array is a known value and doesn't need to be loaded back // from the array. return false; } } else { assert(bt == T_INT, "no LoadRange for longs"); } scale = 0; offset = nullptr; if (!phase->is_scaled_iv_plus_offset(cmp->in(1), iv, bt, &scale, &offset)) { return false; } return true; } bool IdealLoopTree::is_range_check_if(IfProjNode* if_success_proj, PhaseIdealLoop *phase, Invariance& invar DEBUG_ONLY(COMMA ProjNode *predicate_proj)) const { Node* range = nullptr; Node* offset = nullptr; jlong scale = 0; Node* iv = _head->as_BaseCountedLoop()->phi(); Compile* C = Compile::current(); const uint old_unique_idx = C->unique(); if (!is_range_check_if(if_success_proj, phase, T_INT, iv, range, offset, scale)) { return false; } if (!invar.is_invariant(range)) { return false; } if (offset != nullptr) { if (!invar.is_invariant(offset)) { // offset must be invariant return false; } Node* data_dependency_on = invar.data_dependency_on(); if (data_dependency_on != nullptr && old_unique_idx < C->unique()) { // 'offset' node was newly created in is_range_check_if(). Check that it does not depend on the entry projection // into the loop. If it does, we cannot perform loop predication (see Invariant::Invariant()). assert(!offset->is_CFG(), "offset must be a data node"); if (_phase->get_ctrl(offset) == data_dependency_on) { return false; } } } #ifdef ASSERT if (offset && phase->has_ctrl(offset)) { Node* offset_ctrl = phase->get_ctrl(offset); if (phase->get_loop(predicate_proj) == phase->get_loop(offset_ctrl) && phase->is_dominator(predicate_proj, offset_ctrl)) { // If the control of offset is loop predication promoted by previous pass, // then it will lead to cyclic dependency. // Previously promoted loop predication is in the same loop of predication // point. // This situation can occur when pinning nodes too conservatively - can we do better? assert(false, "cyclic dependency prevents range check elimination, idx: offset %d, offset_ctrl %d, predicate_proj %d", offset->_idx, offset_ctrl->_idx, predicate_proj->_idx); } } #endif return true; } //------------------------------rc_predicate----------------------------------- // Create a range check predicate // // for (i = init; i < limit; i += stride) { // a[scale*i+offset] // } // // Compute max(scale*i + offset) for init <= i < limit and build the predicate // as "max(scale*i + offset) u< a.length". // // There are two cases for max(scale*i + offset): // (1) stride*scale > 0 // max(scale*i + offset) = scale*(limit-stride) + offset // (2) stride*scale < 0 // max(scale*i + offset) = scale*init + offset BoolNode* PhaseIdealLoop::rc_predicate(IdealLoopTree* loop, Node* ctrl, int scale, Node* offset, Node* init, Node* limit, jint stride, Node* range, bool upper, bool& overflow) { jint con_limit = (limit != nullptr && limit->is_Con()) ? limit->get_int() : 0; jint con_init = init->is_Con() ? init->get_int() : 0; jint con_offset = offset->is_Con() ? offset->get_int() : 0; stringStream* predString = nullptr; if (TraceLoopPredicate) { predString = new (mtCompiler) stringStream(); predString->print("rc_predicate "); } overflow = false; Node* max_idx_expr = nullptr; const TypeInt* idx_type = TypeInt::INT; if ((stride > 0) == (scale > 0) == upper) { guarantee(limit != nullptr, "sanity"); if (TraceLoopPredicate) { if (limit->is_Con()) { predString->print("(%d ", con_limit); } else { predString->print("(limit "); } predString->print("- %d) ", stride); } // Check if (limit - stride) may overflow const TypeInt* limit_type = _igvn.type(limit)->isa_int(); jint limit_lo = limit_type->_lo; jint limit_hi = limit_type->_hi; if ((stride > 0 && (java_subtract(limit_lo, stride) < limit_lo)) || (stride < 0 && (java_subtract(limit_hi, stride) > limit_hi))) { // No overflow possible ConINode* con_stride = _igvn.intcon(stride); set_ctrl(con_stride, C->root()); max_idx_expr = new SubINode(limit, con_stride); idx_type = TypeInt::make(limit_lo - stride, limit_hi - stride, limit_type->_widen); } else { // May overflow overflow = true; limit = new ConvI2LNode(limit); register_new_node(limit, ctrl); ConLNode* con_stride = _igvn.longcon(stride); set_ctrl(con_stride, C->root()); max_idx_expr = new SubLNode(limit, con_stride); } register_new_node(max_idx_expr, ctrl); } else { if (TraceLoopPredicate) { if (init->is_Con()) { predString->print("%d ", con_init); } else { predString->print("init "); } } idx_type = _igvn.type(init)->isa_int(); max_idx_expr = init; } if (scale != 1) { ConNode* con_scale = _igvn.intcon(scale); set_ctrl(con_scale, C->root()); if (TraceLoopPredicate) { predString->print("* %d ", scale); } // Check if (scale * max_idx_expr) may overflow const TypeInt* scale_type = TypeInt::make(scale); MulINode* mul = new MulINode(max_idx_expr, con_scale); idx_type = (TypeInt*)mul->mul_ring(idx_type, scale_type); if (overflow || TypeInt::INT->higher_equal(idx_type)) { // May overflow mul->destruct(&_igvn); if (!overflow) { max_idx_expr = new ConvI2LNode(max_idx_expr); register_new_node(max_idx_expr, ctrl); } overflow = true; con_scale = _igvn.longcon(scale); set_ctrl(con_scale, C->root()); max_idx_expr = new MulLNode(max_idx_expr, con_scale); } else { // No overflow possible max_idx_expr = mul; } register_new_node(max_idx_expr, ctrl); } if (offset && (!offset->is_Con() || con_offset != 0)){ if (TraceLoopPredicate) { if (offset->is_Con()) { predString->print("+ %d ", con_offset); } else { predString->print("+ offset"); } } // Check if (max_idx_expr + offset) may overflow const TypeInt* offset_type = _igvn.type(offset)->isa_int(); jint lo = java_add(idx_type->_lo, offset_type->_lo); jint hi = java_add(idx_type->_hi, offset_type->_hi); if (overflow || (lo > hi) || ((idx_type->_lo & offset_type->_lo) < 0 && lo >= 0) || ((~(idx_type->_hi | offset_type->_hi)) < 0 && hi < 0)) { // May overflow if (!overflow) { max_idx_expr = new ConvI2LNode(max_idx_expr); register_new_node(max_idx_expr, ctrl); } overflow = true; offset = new ConvI2LNode(offset); register_new_node(offset, ctrl); max_idx_expr = new AddLNode(max_idx_expr, offset); } else { // No overflow possible max_idx_expr = new AddINode(max_idx_expr, offset); } register_new_node(max_idx_expr, ctrl); } CmpNode* cmp = nullptr; if (overflow) { // Integer expressions may overflow, do long comparison range = new ConvI2LNode(range); register_new_node(range, ctrl); cmp = new CmpULNode(max_idx_expr, range); } else { cmp = new CmpUNode(max_idx_expr, range); } register_new_node(cmp, ctrl); BoolNode* bol = new BoolNode(cmp, BoolTest::lt); register_new_node(bol, ctrl); if (TraceLoopPredicate) { predString->print_cr("print("%s", predString->base()); delete predString; } return bol; } // Should loop predication look not only in the path from tail to head // but also in branches of the loop body? bool PhaseIdealLoop::loop_predication_should_follow_branches(IdealLoopTree* loop, float& loop_trip_cnt) { if (!UseProfiledLoopPredicate) { return false; } LoopNode* head = loop->_head->as_Loop(); bool follow_branches = true; IdealLoopTree* l = loop->_child; // For leaf loops and loops with a single inner loop while (l != nullptr && follow_branches) { IdealLoopTree* child = l; if (child->_child != nullptr && child->_head->is_OuterStripMinedLoop()) { assert(child->_child->_next == nullptr, "only one inner loop for strip mined loop"); assert(child->_child->_head->is_CountedLoop() && child->_child->_head->as_CountedLoop()->is_strip_mined(), "inner loop should be strip mined"); child = child->_child; } if (child->_child != nullptr || child->_irreducible) { follow_branches = false; } l = l->_next; } if (follow_branches) { loop->compute_profile_trip_cnt(this); if (head->is_profile_trip_failed()) { follow_branches = false; } else { loop_trip_cnt = head->profile_trip_cnt(); if (head->is_CountedLoop()) { CountedLoopNode* cl = head->as_CountedLoop(); if (cl->phi() != nullptr) { const TypeInt* t = _igvn.type(cl->phi())->is_int(); float worst_case_trip_cnt = ((float)t->_hi - t->_lo) / ABS(cl->stride_con()); if (worst_case_trip_cnt < loop_trip_cnt) { loop_trip_cnt = worst_case_trip_cnt; } } } } } return follow_branches; } float PathFrequency::to(Node* n) { // post order walk on the CFG graph from n to _dom IdealLoopTree* loop = _phase->get_loop(_dom); Node* c = n; for (;;) { assert(_phase->get_loop(c) == loop, "have to be in the same loop"); if (c == _dom || _freqs.at_grow(c->_idx, -1) >= 0) { float f = c == _dom ? 1 : _freqs.at(c->_idx); Node* prev = c; while (_stack.size() > 0 && prev == c) { Node* n = _stack.node(); if (!n->is_Region()) { if (_phase->get_loop(n) != _phase->get_loop(n->in(0))) { // Found an inner loop: compute frequency of reaching this // exit from the loop head by looking at the number of // times each loop exit was taken IdealLoopTree* inner_loop = _phase->get_loop(n->in(0)); LoopNode* inner_head = inner_loop->_head->as_Loop(); assert(_phase->get_loop(n) == loop, "only 1 inner loop"); if (inner_head->is_OuterStripMinedLoop()) { inner_head->verify_strip_mined(1); if (n->in(0) == inner_head->in(LoopNode::LoopBackControl)->in(0)) { n = n->in(0)->in(0)->in(0); } inner_loop = inner_loop->_child; inner_head = inner_loop->_head->as_Loop(); inner_head->verify_strip_mined(1); } float loop_exit_cnt = 0.0f; for (uint i = 0; i < inner_loop->_body.size(); i++) { Node *n = inner_loop->_body[i]; float c = inner_loop->compute_profile_trip_cnt_helper(n); loop_exit_cnt += c; } float cnt = -1; if (n->in(0)->is_If()) { IfNode* iff = n->in(0)->as_If(); float p = n->in(0)->as_If()->_prob; if (n->Opcode() == Op_IfFalse) { p = 1 - p; } if (p > PROB_MIN) { cnt = p * iff->_fcnt; } else { cnt = 0; } } else { assert(n->in(0)->is_Jump(), "unsupported node kind"); JumpNode* jmp = n->in(0)->as_Jump(); float p = n->in(0)->as_Jump()->_probs[n->as_JumpProj()->_con]; cnt = p * jmp->_fcnt; } float this_exit_f = cnt > 0 ? cnt / loop_exit_cnt : 0; this_exit_f = check_and_truncate_frequency(this_exit_f); f = f * this_exit_f; f = check_and_truncate_frequency(f); } else { float p = -1; if (n->in(0)->is_If()) { p = n->in(0)->as_If()->_prob; if (n->Opcode() == Op_IfFalse) { p = 1 - p; } } else { assert(n->in(0)->is_Jump(), "unsupported node kind"); p = n->in(0)->as_Jump()->_probs[n->as_JumpProj()->_con]; } f = f * p; f = check_and_truncate_frequency(f); } _freqs.at_put_grow(n->_idx, (float)f, -1); _stack.pop(); } else { float prev_f = _freqs_stack.pop(); float new_f = f; f = new_f + prev_f; f = check_and_truncate_frequency(f); uint i = _stack.index(); if (i < n->req()) { c = n->in(i); _stack.set_index(i+1); _freqs_stack.push(f); } else { _freqs.at_put_grow(n->_idx, f, -1); _stack.pop(); } } } if (_stack.size() == 0) { return check_and_truncate_frequency(f); } } else if (c->is_Loop()) { ShouldNotReachHere(); c = c->in(LoopNode::EntryControl); } else if (c->is_Region()) { _freqs_stack.push(0); _stack.push(c, 2); c = c->in(1); } else { if (c->is_IfProj()) { IfNode* iff = c->in(0)->as_If(); if (iff->_prob == PROB_UNKNOWN) { // assume never taken _freqs.at_put_grow(c->_idx, 0, -1); } else if (_phase->get_loop(c) != _phase->get_loop(iff)) { if (iff->_fcnt == COUNT_UNKNOWN) { // assume never taken _freqs.at_put_grow(c->_idx, 0, -1); } else { // skip over loop _stack.push(c, 1); c = _phase->get_loop(c->in(0))->_head->as_Loop()->skip_strip_mined()->in(LoopNode::EntryControl); } } else { _stack.push(c, 1); c = iff; } } else if (c->is_JumpProj()) { JumpNode* jmp = c->in(0)->as_Jump(); if (_phase->get_loop(c) != _phase->get_loop(jmp)) { if (jmp->_fcnt == COUNT_UNKNOWN) { // assume never taken _freqs.at_put_grow(c->_idx, 0, -1); } else { // skip over loop _stack.push(c, 1); c = _phase->get_loop(c->in(0))->_head->as_Loop()->skip_strip_mined()->in(LoopNode::EntryControl); } } else { _stack.push(c, 1); c = jmp; } } else if (c->Opcode() == Op_CatchProj && c->in(0)->Opcode() == Op_Catch && c->in(0)->in(0)->is_Proj() && c->in(0)->in(0)->in(0)->is_Call()) { // assume exceptions are never thrown uint con = c->as_Proj()->_con; if (con == CatchProjNode::fall_through_index) { Node* call = c->in(0)->in(0)->in(0)->in(0); if (_phase->get_loop(call) != _phase->get_loop(c)) { _freqs.at_put_grow(c->_idx, 0, -1); } else { c = call; } } else { assert(con >= CatchProjNode::catch_all_index, "what else?"); _freqs.at_put_grow(c->_idx, 0, -1); } } else if (c->unique_ctrl_out_or_null() == nullptr && !c->is_If() && !c->is_Jump()) { ShouldNotReachHere(); } else { c = c->in(0); } } } ShouldNotReachHere(); return -1; } void PhaseIdealLoop::loop_predication_follow_branches(Node *n, IdealLoopTree *loop, float loop_trip_cnt, PathFrequency& pf, Node_Stack& stack, VectorSet& seen, Node_List& if_proj_list) { assert(n->is_Region(), "start from a region"); Node* tail = loop->tail(); stack.push(n, 1); do { Node* c = stack.node(); assert(c->is_Region() || c->is_IfProj(), "only region here"); uint i = stack.index(); if (i < c->req()) { stack.set_index(i+1); Node* in = c->in(i); while (!is_dominator(in, tail) && !seen.test_set(in->_idx)) { IdealLoopTree* in_loop = get_loop(in); if (in_loop != loop) { in = in_loop->_head->in(LoopNode::EntryControl); } else if (in->is_Region()) { stack.push(in, 1); break; } else if (in->is_IfProj() && in->as_Proj()->is_uncommon_trap_if_pattern() && (in->in(0)->Opcode() == Op_If || in->in(0)->Opcode() == Op_RangeCheck)) { if (pf.to(in) * loop_trip_cnt >= 1) { stack.push(in, 1); } in = in->in(0); } else { in = in->in(0); } } } else { if (c->is_IfProj()) { if_proj_list.push(c); } stack.pop(); } } while (stack.size() > 0); } bool PhaseIdealLoop::loop_predication_impl_helper(IdealLoopTree* loop, IfProjNode* if_success_proj, ParsePredicateSuccessProj* parse_predicate_proj, CountedLoopNode* cl, ConNode* zero, Invariance& invar, Deoptimization::DeoptReason reason) { // Following are changed to nonnull when a predicate can be hoisted IfProjNode* new_predicate_proj = nullptr; IfNode* iff = if_success_proj->in(0)->as_If(); Node* test = iff->in(1); if (!test->is_Bool()) { //Conv2B, ... return false; } BoolNode* bol = test->as_Bool(); if (invar.is_invariant(bol)) { C->print_method(PHASE_BEFORE_LOOP_PREDICATION_IC, 4, iff); // Invariant test new_predicate_proj = create_new_if_for_predicate(parse_predicate_proj, nullptr, reason, iff->Opcode()); Node* ctrl = new_predicate_proj->in(0)->as_If()->in(0); BoolNode* new_predicate_bol = invar.clone(bol, ctrl)->as_Bool(); // Negate test if necessary (Parse Predicates always have IfTrue as success projection and IfFalse as uncommon trap) bool negated = false; if (if_success_proj->is_IfFalse()) { new_predicate_bol = new BoolNode(new_predicate_bol->in(1), new_predicate_bol->_test.negate()); register_new_node(new_predicate_bol, ctrl); negated = true; } IfNode* new_predicate_iff = new_predicate_proj->in(0)->as_If(); _igvn.hash_delete(new_predicate_iff); new_predicate_iff->set_req(1, new_predicate_bol); C->print_method(PHASE_AFTER_LOOP_PREDICATION_IC, 4, new_predicate_proj->in(0)); #ifndef PRODUCT if (TraceLoopPredicate) { tty->print("Predicate invariant if%s: %d ", negated ? " negated" : "", new_predicate_iff->_idx); loop->dump_head(); } else if (TraceLoopOpts) { tty->print("Predicate IC "); loop->dump_head(); } #endif } else if (cl != nullptr && loop->is_range_check_if(if_success_proj, this, invar DEBUG_ONLY(COMMA parse_predicate_proj))) { C->print_method(PHASE_BEFORE_LOOP_PREDICATION_RC, 4, iff); // Range check for counted loops assert(if_success_proj->is_IfTrue(), "trap must be on false projection for a range check"); const Node* cmp = bol->in(1)->as_Cmp(); Node* idx = cmp->in(1); assert(!invar.is_invariant(idx), "index is variant"); Node* rng = cmp->in(2); assert(rng->Opcode() == Op_LoadRange || iff->is_RangeCheck() || _igvn.type(rng)->is_int()->_lo >= 0, "must be"); assert(invar.is_invariant(rng), "range must be invariant"); int scale = 1; Node* offset = zero; bool ok = is_scaled_iv_plus_offset(idx, cl->phi(), &scale, &offset); assert(ok, "must be index expression"); Node* init = cl->init_trip(); // Limit is not exact. // Calculate exact limit here. // Note, counted loop's test is '<' or '>'. loop->compute_trip_count(this); Node* limit = exact_limit(loop); int stride = cl->stride()->get_int(); // Build if's for the upper and lower bound tests. The // lower_bound test will dominate the upper bound test and all // cloned or created nodes will use the lower bound test as // their declared control. // Perform cloning to keep Invariance state correct since the // late schedule will place invariant things in the loop. Node* ctrl = parse_predicate_proj->in(0)->as_If()->in(0); rng = invar.clone(rng, ctrl); if (offset && offset != zero) { assert(invar.is_invariant(offset), "offset must be loop invariant"); offset = invar.clone(offset, ctrl); } // If predicate expressions may overflow in the integer range, longs are used. bool overflow = false; // Test the lower bound BoolNode* lower_bound_bol = rc_predicate(loop, ctrl, scale, offset, init, limit, stride, rng, false, overflow); const int if_opcode = iff->Opcode(); IfProjNode* lower_bound_proj = create_new_if_for_predicate(parse_predicate_proj, nullptr, reason, overflow ? Op_If : if_opcode); IfNode* lower_bound_iff = lower_bound_proj->in(0)->as_If(); _igvn.hash_delete(lower_bound_iff); lower_bound_iff->set_req(1, lower_bound_bol); if (TraceLoopPredicate) tty->print_cr("lower bound check if: %d", lower_bound_iff->_idx); // Test the upper bound BoolNode* upper_bound_bol = rc_predicate(loop, lower_bound_proj, scale, offset, init, limit, stride, rng, true, overflow); IfProjNode* upper_bound_proj = create_new_if_for_predicate(parse_predicate_proj, nullptr, reason, overflow ? Op_If : if_opcode); assert(upper_bound_proj->in(0)->as_If()->in(0) == lower_bound_proj, "should dominate"); IfNode* upper_bound_iff = upper_bound_proj->in(0)->as_If(); _igvn.hash_delete(upper_bound_iff); upper_bound_iff->set_req(1, upper_bound_bol); if (TraceLoopPredicate) tty->print_cr("upper bound check if: %d", lower_bound_iff->_idx); // Fall through into rest of the cleanup code which will move any dependent nodes to the skeleton predicates of the // upper bound test. We always need to create skeleton predicates in order to properly remove dead loops when later // splitting the predicated loop into (unreachable) sub-loops (i.e. done by unrolling, peeling, pre/main/post etc.). new_predicate_proj = add_template_assertion_predicate(iff, loop, if_success_proj, parse_predicate_proj, upper_bound_proj, scale, offset, init, limit, stride, rng, overflow, reason); C->print_method(PHASE_AFTER_LOOP_PREDICATION_RC, 4, new_predicate_proj->in(0)); #ifndef PRODUCT if (TraceLoopOpts && !TraceLoopPredicate) { tty->print("Predicate RC "); loop->dump_head(); } #endif } else { // Loop variant check (for example, range check in non-counted loop) // with uncommon trap. return false; } assert(new_predicate_proj != nullptr, "sanity"); // Success - attach condition (new_predicate_bol) to predicate if invar.map_ctrl(if_success_proj, new_predicate_proj); // so that invariance test can be appropriate // Eliminate the old If in the loop body dominated_by(new_predicate_proj, iff, if_success_proj->_con != new_predicate_proj->_con); C->set_major_progress(); return true; } // Each newly created Hoisted Check Predicate is accompanied by two Template Assertion Predicates. Later, we initialize // them by making a copy of them when splitting a loop into sub loops. The Assertion Predicates ensure that dead sub // loops are removed properly. IfProjNode* PhaseIdealLoop::add_template_assertion_predicate(IfNode* iff, IdealLoopTree* loop, IfProjNode* if_proj, ParsePredicateSuccessProj* parse_predicate_proj, IfProjNode* upper_bound_proj, const int scale, Node* offset, Node* init, Node* limit, const jint stride, Node* rng, bool& overflow, Deoptimization::DeoptReason reason) { // First predicate for the initial value on first loop iteration Node* opaque_init = new OpaqueLoopInitNode(C, init); register_new_node(opaque_init, upper_bound_proj); bool negate = (if_proj->_con != parse_predicate_proj->_con); BoolNode* bol = rc_predicate(loop, upper_bound_proj, scale, offset, opaque_init, limit, stride, rng, (stride > 0) != (scale > 0), overflow); Node* opaque_bol = new Opaque4Node(C, bol, _igvn.intcon(1)); // This will go away once loop opts are over C->add_template_assertion_predicate_opaq(opaque_bol); register_new_node(opaque_bol, upper_bound_proj); IfProjNode* new_proj = create_new_if_for_predicate(parse_predicate_proj, nullptr, reason, overflow ? Op_If : iff->Opcode()); _igvn.replace_input_of(new_proj->in(0), 1, opaque_bol); assert(opaque_init->outcnt() > 0, "should be used"); // Second predicate for init + (current stride - initial stride) // This is identical to the previous predicate initially but as // unrolling proceeds current stride is updated. Node* init_stride = loop->_head->as_CountedLoop()->stride(); Node* opaque_stride = new OpaqueLoopStrideNode(C, init_stride); register_new_node(opaque_stride, new_proj); Node* max_value = new SubINode(opaque_stride, init_stride); register_new_node(max_value, new_proj); max_value = new AddINode(opaque_init, max_value); register_new_node(max_value, new_proj); // init + (current stride - initial stride) is within the loop so narrow its type by leveraging the type of the iv Phi max_value = new CastIINode(max_value, loop->_head->as_CountedLoop()->phi()->bottom_type()); register_new_node(max_value, parse_predicate_proj); bol = rc_predicate(loop, new_proj, scale, offset, max_value, limit, stride, rng, (stride > 0) != (scale > 0), overflow); opaque_bol = new Opaque4Node(C, bol, _igvn.intcon(1)); C->add_template_assertion_predicate_opaq(opaque_bol); register_new_node(opaque_bol, new_proj); new_proj = create_new_if_for_predicate(parse_predicate_proj, nullptr, reason, overflow ? Op_If : iff->Opcode()); _igvn.replace_input_of(new_proj->in(0), 1, opaque_bol); assert(max_value->outcnt() > 0, "should be used"); assert(assertion_predicate_has_loop_opaque_node(new_proj->in(0)->as_If()), "unexpected"); return new_proj; } // Insert Hoisted Check Predicates for null checks and range checks and additional Template Assertion Predicates for // range checks. bool PhaseIdealLoop::loop_predication_impl(IdealLoopTree* loop) { LoopNode* head = loop->_head->as_Loop(); if (head->unique_ctrl_out()->is_NeverBranch()) { // do nothing for infinite loops return false; } if (head->is_OuterStripMinedLoop()) { return false; } CountedLoopNode *cl = nullptr; if (head->is_valid_counted_loop(T_INT)) { cl = head->as_CountedLoop(); // do nothing for iteration-splitted loops if (!cl->is_normal_loop()) return false; // Avoid RCE if Counted loop's test is '!='. BoolTest::mask bt = cl->loopexit()->test_trip(); if (bt != BoolTest::lt && bt != BoolTest::gt) cl = nullptr; } Node* entry = head->skip_strip_mined()->in(LoopNode::EntryControl); const Predicates predicates(entry); const PredicateBlock* loop_predicate_block = predicates.loop_predicate_block(); const PredicateBlock* profiled_loop_predicate_block = predicates.profiled_loop_predicate_block(); float loop_trip_cnt = -1; bool follow_branches = profiled_loop_predicate_block->has_parse_predicate() && loop_predication_should_follow_branches(loop, loop_trip_cnt); assert(!follow_branches || loop_trip_cnt >= 0, "negative trip count?"); if (!loop_predicate_block->has_parse_predicate() && !follow_branches) { #ifndef PRODUCT if (TraceLoopPredicate) { tty->print("Missing Parse Predicates:"); loop->dump_head(); head->dump(1); } #endif return false; } ConNode* zero = _igvn.intcon(0); set_ctrl(zero, C->root()); ResourceArea* area = Thread::current()->resource_area(); Invariance invar(area, loop); // Create list of if-projs such that a newer proj dominates all older // projs in the list, and they all dominate loop->tail() Node_List if_proj_list; Node_List regions; Node* current_proj = loop->tail(); // start from tail Node_List controls; while (current_proj != head) { if (loop == get_loop(current_proj) && // still in the loop ? current_proj->is_Proj() && // is a projection ? (current_proj->in(0)->Opcode() == Op_If || current_proj->in(0)->Opcode() == Op_RangeCheck)) { // is a if projection ? if_proj_list.push(current_proj); } if (follow_branches && current_proj->Opcode() == Op_Region && loop == get_loop(current_proj)) { regions.push(current_proj); } current_proj = idom(current_proj); } bool hoisted = false; // true if at least one proj is promoted if (can_create_loop_predicates(profiled_loop_predicate_block)) { while (if_proj_list.size() > 0) { Node* n = if_proj_list.pop(); IfProjNode* if_proj = n->as_IfProj(); IfNode* iff = if_proj->in(0)->as_If(); CallStaticJavaNode* call = if_proj->is_uncommon_trap_if_pattern(); if (call == nullptr) { if (loop->is_loop_exit(iff)) { // stop processing the remaining projs in the list because the execution of them // depends on the condition of "iff" (iff->in(1)). break; } else { // Both arms are inside the loop. There are two cases: // (1) there is one backward branch. In this case, any remaining proj // in the if_proj list post-dominates "iff". So, the condition of "iff" // does not determine the execution the remaining projs directly, and we // can safely continue. // (2) both arms are forwarded, i.e. a diamond shape. In this case, "proj" // does not dominate loop->tail(), so it can not be in the if_proj list. continue; } } Deoptimization::DeoptReason reason = Deoptimization::trap_request_reason(call->uncommon_trap_request()); if (reason == Deoptimization::Reason_predicate) { break; } if (loop_predicate_block->has_parse_predicate()) { ParsePredicateSuccessProj* loop_parse_predicate_proj = loop_predicate_block->parse_predicate_success_proj(); hoisted = loop_predication_impl_helper(loop, if_proj, loop_parse_predicate_proj, cl, zero, invar, Deoptimization::Reason_predicate) | hoisted; } } // end while } if (follow_branches) { assert(profiled_loop_predicate_block->has_parse_predicate(), "sanity check"); PathFrequency pf(loop->_head, this); // Some projections were skipped due to an early loop exit. Try them with profile data. while (if_proj_list.size() > 0) { Node* if_proj = if_proj_list.pop(); float f = pf.to(if_proj); if (if_proj->as_Proj()->is_uncommon_trap_if_pattern() && f * loop_trip_cnt >= 1) { ParsePredicateSuccessProj* profiled_loop_parse_predicate_proj = profiled_loop_predicate_block->parse_predicate_success_proj(); hoisted = loop_predication_impl_helper(loop, if_proj->as_IfProj(), profiled_loop_parse_predicate_proj, cl, zero, invar, Deoptimization::Reason_profile_predicate) | hoisted; } } // And look into all branches Node_Stack stack(0); VectorSet seen; Node_List if_proj_list_freq(area); while (regions.size() > 0) { Node* c = regions.pop(); loop_predication_follow_branches(c, loop, loop_trip_cnt, pf, stack, seen, if_proj_list_freq); } for (uint i = 0; i < if_proj_list_freq.size(); i++) { IfProjNode* if_proj = if_proj_list_freq.at(i)->as_IfProj(); ParsePredicateSuccessProj* profiled_loop_parse_predicate_proj = profiled_loop_predicate_block->parse_predicate_success_proj(); hoisted = loop_predication_impl_helper(loop, if_proj, profiled_loop_parse_predicate_proj, cl, zero, invar, Deoptimization::Reason_profile_predicate) | hoisted; } } #ifndef PRODUCT // report that the loop predication has been actually performed // for this loop if (TraceLoopPredicate && hoisted) { tty->print("Loop Predication Performed:"); loop->dump_head(); } #endif head->verify_strip_mined(1); return hoisted; } // We cannot add Loop Predicates if: // (1) Already added Profiled Loop Predicates (Loop Predicates and Profiled Loop Predicates can be dependent // through a data node, and thus we should only add new Profiled Loop Predicates which are below Loop Predicates // in the graph). // (2) There are currently no Profiled Loop Predicates, but we have a data node with a control dependency on the Loop // Parse Predicate (could happen, for example, if we've removed an earlier created Profiled Loop Predicate with // dominated_by()). We should not create a Loop Predicate for a check that is dependent on this data node because // the Loop Predicate would end up above the data node with its dependency on the Loop Parse Predicate below. This // would become unschedulable. However, we can still hoist the check as Profiled Loop Predicate which would end up // below the Loop Parse Predicate. bool PhaseIdealLoop::can_create_loop_predicates(const PredicateBlock* profiled_loop_predicate_block) const { bool has_profiled_loop_predicate_block = profiled_loop_predicate_block != nullptr; bool can_create_loop_predicates = true; if (has_profiled_loop_predicate_block && (profiled_loop_predicate_block->has_runtime_predicates() // (1) || profiled_loop_predicate_block->entry()->outcnt() != 1)) { // (2) can_create_loop_predicates = false; } return can_create_loop_predicates; } //------------------------------loop_predication-------------------------------- // driver routine for loop predication optimization bool IdealLoopTree::loop_predication( PhaseIdealLoop *phase) { bool hoisted = false; // Recursively promote predicates if (_child) { hoisted = _child->loop_predication( phase); } // self if (can_apply_loop_predication()) { hoisted |= phase->loop_predication_impl(this); } if (_next) { //sibling hoisted |= _next->loop_predication( phase); } return hoisted; } bool IdealLoopTree::can_apply_loop_predication() { return _head->is_Loop() && !_irreducible && !tail()->is_top(); }