/*
 * Copyright (c) 2011, 2020, 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.
 */


package org.graalvm.compiler.nodes.util;

import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collection;
import java.util.Collections;
import java.util.Iterator;
import java.util.List;
import java.util.Objects;
import java.util.function.BiFunction;

import jdk.internal.vm.compiler.collections.EconomicMap;
import jdk.internal.vm.compiler.collections.EconomicSet;
import jdk.internal.vm.compiler.collections.Equivalence;
import jdk.internal.vm.compiler.collections.MapCursor;
import org.graalvm.compiler.bytecode.Bytecode;
import org.graalvm.compiler.code.SourceStackTraceBailoutException;
import org.graalvm.compiler.core.common.type.ObjectStamp;
import org.graalvm.compiler.debug.DebugContext;
import org.graalvm.compiler.debug.GraalError;
import org.graalvm.compiler.graph.Graph;
import org.graalvm.compiler.graph.Node;
import org.graalvm.compiler.graph.NodeBitMap;
import org.graalvm.compiler.graph.NodeSourcePosition;
import org.graalvm.compiler.graph.NodeStack;
import org.graalvm.compiler.graph.Position;
import org.graalvm.compiler.graph.iterators.NodeIterable;
import org.graalvm.compiler.graph.spi.SimplifierTool;
import org.graalvm.compiler.nodes.AbstractBeginNode;
import org.graalvm.compiler.nodes.AbstractEndNode;
import org.graalvm.compiler.nodes.AbstractMergeNode;
import org.graalvm.compiler.nodes.ConstantNode;
import org.graalvm.compiler.nodes.ControlSinkNode;
import org.graalvm.compiler.nodes.ControlSplitNode;
import org.graalvm.compiler.nodes.EndNode;
import org.graalvm.compiler.nodes.FixedNode;
import org.graalvm.compiler.nodes.FixedWithNextNode;
import org.graalvm.compiler.nodes.FrameState;
import org.graalvm.compiler.nodes.GuardNode;
import org.graalvm.compiler.nodes.IfNode;
import org.graalvm.compiler.nodes.LoopBeginNode;
import org.graalvm.compiler.nodes.LoopEndNode;
import org.graalvm.compiler.nodes.LoopExitNode;
import org.graalvm.compiler.nodes.MergeNode;
import org.graalvm.compiler.nodes.NodeView;
import org.graalvm.compiler.nodes.PhiNode;
import org.graalvm.compiler.nodes.PiNode;
import org.graalvm.compiler.nodes.ProxyNode;
import org.graalvm.compiler.nodes.StateSplit;
import org.graalvm.compiler.nodes.StructuredGraph;
import org.graalvm.compiler.nodes.ValueNode;
import org.graalvm.compiler.nodes.ValuePhiNode;
import org.graalvm.compiler.nodes.ValueProxyNode;
import org.graalvm.compiler.nodes.WithExceptionNode;
import org.graalvm.compiler.nodes.extended.MultiGuardNode;
import org.graalvm.compiler.nodes.java.LoadIndexedNode;
import org.graalvm.compiler.nodes.java.MethodCallTargetNode;
import org.graalvm.compiler.nodes.java.MonitorIdNode;
import org.graalvm.compiler.nodes.memory.MemoryPhiNode;
import org.graalvm.compiler.nodes.spi.ArrayLengthProvider;
import org.graalvm.compiler.nodes.spi.ArrayLengthProvider.FindLengthMode;
import org.graalvm.compiler.nodes.spi.CoreProviders;
import org.graalvm.compiler.nodes.spi.CoreProvidersDelegate;
import org.graalvm.compiler.nodes.spi.LimitedValueProxy;
import org.graalvm.compiler.nodes.spi.ValueProxy;
import org.graalvm.compiler.nodes.spi.VirtualizerTool;
import org.graalvm.compiler.nodes.type.StampTool;
import org.graalvm.compiler.nodes.virtual.VirtualArrayNode;
import org.graalvm.compiler.nodes.virtual.VirtualObjectNode;
import org.graalvm.compiler.options.Option;
import org.graalvm.compiler.options.OptionKey;
import org.graalvm.compiler.options.OptionType;
import org.graalvm.compiler.options.OptionValues;

import jdk.vm.ci.code.BailoutException;
import jdk.vm.ci.code.BytecodePosition;
import jdk.vm.ci.meta.Assumptions;
import jdk.vm.ci.meta.Constant;
import jdk.vm.ci.meta.ConstantReflectionProvider;
import jdk.vm.ci.meta.JavaKind;
import jdk.vm.ci.meta.ResolvedJavaMethod;
import jdk.vm.ci.meta.ResolvedJavaType;

public class GraphUtil {

    public static class Options {
        @Option(help = "Verify that there are no new unused nodes when performing killCFG", type = OptionType.Debug)//
        public static final OptionKey<Boolean> VerifyKillCFGUnusedNodes = new OptionKey<>(false);
    }

    public static final int MAX_FRAMESTATE_SEARCH_DEPTH = 4;

    private static void killCFGInner(FixedNode node) {
        EconomicSet<Node> markedNodes = EconomicSet.create();
        EconomicMap<AbstractMergeNode, List<AbstractEndNode>> unmarkedMerges = EconomicMap.create();

        // Detach this node from CFG
        node.replaceAtPredecessor(null);

        markFixedNodes(node, markedNodes, unmarkedMerges);

        fixSurvivingAffectedMerges(markedNodes, unmarkedMerges);

        DebugContext debug = node.getDebug();
        debug.dump(DebugContext.DETAILED_LEVEL, node.graph(), "After fixing merges (killCFG %s)", node);

        // Mark non-fixed nodes
        markUsagesForKill(markedNodes);

        // Detach marked nodes from non-marked nodes
        for (Node marked : markedNodes) {
            for (Node input : marked.inputs()) {
                if (!markedNodes.contains(input)) {
                    marked.replaceFirstInput(input, null);
                    tryKillUnused(input);
                }
            }
        }
        debug.dump(DebugContext.VERY_DETAILED_LEVEL, node.graph(), "After disconnecting non-marked inputs (killCFG %s)", node);
        // Kill marked nodes
        for (Node marked : markedNodes) {
            if (marked.isAlive()) {
                marked.markDeleted();
            }
        }
    }

    private static void markFixedNodes(FixedNode node, EconomicSet<Node> markedNodes, EconomicMap<AbstractMergeNode, List<AbstractEndNode>> unmarkedMerges) {
        NodeStack workStack = new NodeStack();
        workStack.push(node);
        while (!workStack.isEmpty()) {
            Node fixedNode = workStack.pop();
            markedNodes.add(fixedNode);
            if (fixedNode instanceof AbstractMergeNode) {
                unmarkedMerges.removeKey((AbstractMergeNode) fixedNode);
            }
            while (fixedNode instanceof FixedWithNextNode) {
                fixedNode = ((FixedWithNextNode) fixedNode).next();
                if (fixedNode != null) {
                    markedNodes.add(fixedNode);
                }
            }
            if (fixedNode instanceof ControlSplitNode) {
                for (Node successor : fixedNode.successors()) {
                    workStack.push(successor);
                }
            } else if (fixedNode instanceof AbstractEndNode) {
                AbstractEndNode end = (AbstractEndNode) fixedNode;
                AbstractMergeNode merge = end.merge();
                if (merge != null) {
                    assert !markedNodes.contains(merge) || (merge instanceof LoopBeginNode && end instanceof LoopEndNode) : merge;
                    if (merge instanceof LoopBeginNode) {
                        if (end == ((LoopBeginNode) merge).forwardEnd()) {
                            workStack.push(merge);
                            continue;
                        }
                        if (markedNodes.contains(merge)) {
                            continue;
                        }
                    }
                    List<AbstractEndNode> endsSeen = unmarkedMerges.get(merge);
                    if (endsSeen == null) {
                        endsSeen = new ArrayList<>(merge.forwardEndCount());
                        unmarkedMerges.put(merge, endsSeen);
                    }
                    endsSeen.add(end);
                    if (!(end instanceof LoopEndNode) && endsSeen.size() == merge.forwardEndCount()) {
                        assert merge.forwardEnds().filter(n -> !markedNodes.contains(n)).isEmpty();
                        // all this merge's forward ends are marked: it needs to be killed
                        workStack.push(merge);
                    }
                }
            }
        }
    }

    private static void fixSurvivingAffectedMerges(EconomicSet<Node> markedNodes, EconomicMap<AbstractMergeNode, List<AbstractEndNode>> unmarkedMerges) {
        MapCursor<AbstractMergeNode, List<AbstractEndNode>> cursor = unmarkedMerges.getEntries();
        while (cursor.advance()) {
            AbstractMergeNode merge = cursor.getKey();
            for (AbstractEndNode end : cursor.getValue()) {
                merge.removeEnd(end);
            }
            if (merge.phiPredecessorCount() == 1) {
                if (merge instanceof LoopBeginNode) {
                    LoopBeginNode loopBegin = (LoopBeginNode) merge;
                    assert merge.forwardEndCount() == 1;
                    for (LoopExitNode loopExit : loopBegin.loopExits().snapshot()) {
                        if (markedNodes.contains(loopExit)) {
                            /*
                             * disconnect from loop begin so that reduceDegenerateLoopBegin doesn't
                             * transform it into a new beginNode
                             */
                            loopExit.replaceFirstInput(loopBegin, null);
                        }
                    }
                    merge.graph().reduceDegenerateLoopBegin(loopBegin);
                } else {
                    merge.graph().reduceTrivialMerge(merge);
                }
            } else {
                assert merge.phiPredecessorCount() > 1 : merge;
            }
        }
    }

    private static void markUsagesForKill(EconomicSet<Node> markedNodes) {
        NodeStack workStack = new NodeStack(markedNodes.size() + 4);
        for (Node marked : markedNodes) {
            workStack.push(marked);
        }
        ArrayList<MultiGuardNode> unmarkedMultiGuards = new ArrayList<>();
        while (!workStack.isEmpty()) {
            Node marked = workStack.pop();
            for (Node usage : marked.usages()) {
                boolean doMark = true;
                if (usage instanceof MultiGuardNode) {
                    // Only mark a MultiGuardNode for deletion if all of its guards are marked for
                    // deletion. Otherwise, we would kill nodes outside the path to be killed.
                    MultiGuardNode multiGuard = (MultiGuardNode) usage;
                    for (Node guard : multiGuard.inputs()) {
                        if (!markedNodes.contains(guard)) {
                            doMark = false;
                            unmarkedMultiGuards.add(multiGuard);
                        }
                    }
                }
                if (doMark && !markedNodes.contains(usage)) {
                    workStack.push(usage);
                    markedNodes.add(usage);
                }
            }
            // Detach unmarked multi guards from the marked node
            for (MultiGuardNode multiGuard : unmarkedMultiGuards) {
                multiGuard.replaceFirstInput(marked, null);
            }
            unmarkedMultiGuards.clear();
        }
    }

    @SuppressWarnings("try")
    public static void killCFG(FixedNode node) {
        DebugContext debug = node.getDebug();
        try (DebugContext.Scope scope = debug.scope("KillCFG", node)) {
            EconomicSet<Node> unusedNodes = null;
            EconomicSet<Node> unsafeNodes = null;
            Graph.NodeEventScope nodeEventScope = null;
            OptionValues options = node.getOptions();
            boolean verifyGraalGraphEdges = Graph.Options.VerifyGraalGraphEdges.getValue(options);
            boolean verifyKillCFGUnusedNodes = GraphUtil.Options.VerifyKillCFGUnusedNodes.getValue(options);
            if (verifyGraalGraphEdges) {
                unsafeNodes = collectUnsafeNodes(node.graph());
            }
            if (verifyKillCFGUnusedNodes) {
                EconomicSet<Node> collectedUnusedNodes = unusedNodes = EconomicSet.create(Equivalence.IDENTITY);
                nodeEventScope = node.graph().trackNodeEvents(new Graph.NodeEventListener() {
                    @Override
                    public void changed(Graph.NodeEvent e, Node n) {
                        if (e == Graph.NodeEvent.ZERO_USAGES && isFloatingNode(n) && !(n instanceof GuardNode)) {
                            collectedUnusedNodes.add(n);
                        }
                    }
                });
            }
            debug.dump(DebugContext.VERY_DETAILED_LEVEL, node.graph(), "Before killCFG %s", node);
            killCFGInner(node);
            debug.dump(DebugContext.VERY_DETAILED_LEVEL, node.graph(), "After killCFG %s", node);
            if (verifyGraalGraphEdges) {
                EconomicSet<Node> newUnsafeNodes = collectUnsafeNodes(node.graph());
                newUnsafeNodes.removeAll(unsafeNodes);
                assert newUnsafeNodes.isEmpty() : "New unsafe nodes: " + newUnsafeNodes;
            }
            if (verifyKillCFGUnusedNodes) {
                nodeEventScope.close();
                Iterator<Node> iterator = unusedNodes.iterator();
                while (iterator.hasNext()) {
                    Node curNode = iterator.next();
                    if (curNode.isDeleted()) {
                        iterator.remove();
                    }
                }
                assert unusedNodes.isEmpty() : "New unused nodes: " + unusedNodes;
            }
        } catch (Throwable t) {
            throw debug.handle(t);
        }
    }

    
Collects all node in the graph which have non-optional inputs that are null.
/** * Collects all node in the graph which have non-optional inputs that are null. */
private static EconomicSet<Node> collectUnsafeNodes(Graph graph) { EconomicSet<Node> unsafeNodes = EconomicSet.create(Equivalence.IDENTITY); for (Node n : graph.getNodes()) { for (Position pos : n.inputPositions()) { Node input = pos.get(n); if (input == null) { if (!pos.isInputOptional()) { unsafeNodes.add(n); } } } } return unsafeNodes; } public static boolean isFloatingNode(Node n) { return !(n instanceof FixedNode); } private static boolean checkKill(Node node, boolean mayKillGuard) { node.assertTrue(mayKillGuard || !(node instanceof GuardNode), "must not be a guard node %s", node); node.assertTrue(node.isAlive(), "must be alive"); node.assertTrue(node.hasNoUsages(), "cannot kill node %s because of usages: %s", node, node.usages()); node.assertTrue(node.predecessor() == null, "cannot kill node %s because of predecessor: %s", node, node.predecessor()); return true; } public static void killWithUnusedFloatingInputs(Node node) { killWithUnusedFloatingInputs(node, false); } public static void killWithUnusedFloatingInputs(Node node, boolean mayKillGuard) { assert checkKill(node, mayKillGuard); node.markDeleted(); outer: for (Node in : node.inputs()) { if (in.isAlive()) { in.removeUsage(node); if (in.hasNoUsages()) { node.maybeNotifyZeroUsages(in); } if (isFloatingNode(in)) { if (in.hasNoUsages()) { if (in instanceof GuardNode) { // Guard nodes are only killed if their anchor dies. } else { killWithUnusedFloatingInputs(in); } } else if (in instanceof PhiNode) { for (Node use : in.usages()) { if (use != in) { continue outer; } } in.replaceAtUsages(null); killWithUnusedFloatingInputs(in); } } } } }
Removes all nodes created after the mark, assuming no "old" nodes point to "new" nodes.
/** * Removes all nodes created after the {@code mark}, assuming no "old" nodes point to "new" * nodes. */
public static void removeNewNodes(Graph graph, Graph.Mark mark) { assert checkNoOldToNewEdges(graph, mark); for (Node n : graph.getNewNodes(mark)) { n.markDeleted(); for (Node in : n.inputs()) { in.removeUsage(n); } } } private static boolean checkNoOldToNewEdges(Graph graph, Graph.Mark mark) { for (Node old : graph.getNodes()) { if (graph.isNew(mark, old)) { break; } for (Node n : old.successors()) { assert !graph.isNew(mark, n) : old + " -> " + n; } for (Node n : old.inputs()) { assert !graph.isNew(mark, n) : old + " -> " + n; } } return true; } public static void removeFixedWithUnusedInputs(FixedWithNextNode fixed) { if (fixed instanceof StateSplit) { FrameState stateAfter = ((StateSplit) fixed).stateAfter(); if (stateAfter != null) { ((StateSplit) fixed).setStateAfter(null); if (stateAfter.hasNoUsages()) { killWithUnusedFloatingInputs(stateAfter); } } } unlinkFixedNode(fixed); killWithUnusedFloatingInputs(fixed); } public static void unlinkFixedNode(FixedWithNextNode fixed) { assert fixed.next() != null && fixed.predecessor() != null && fixed.isAlive() : fixed; FixedNode next = fixed.next(); fixed.setNext(null); fixed.replaceAtPredecessor(next); } public static void unlinkAndKillExceptionEdge(WithExceptionNode withException) { assert withException.next() != null && withException.predecessor() != null && withException.isAlive() : withException; FixedNode next = withException.next(); withException.setNext(null); withException.replaceAtPredecessor(next); withException.killExceptionEdge(); } public static void checkRedundantPhi(PhiNode phiNode) { if (phiNode.isDeleted() || phiNode.valueCount() == 1) { return; } ValueNode singleValue = phiNode.singleValueOrThis(); if (singleValue != phiNode) { Collection<PhiNode> phiUsages = phiNode.usages().filter(PhiNode.class).snapshot(); Collection<ProxyNode> proxyUsages = phiNode.usages().filter(ProxyNode.class).snapshot(); phiNode.replaceAtUsagesAndDelete(singleValue); for (PhiNode phi : phiUsages) { checkRedundantPhi(phi); } for (ProxyNode proxy : proxyUsages) { checkRedundantProxy(proxy); } } } public static void checkRedundantProxy(ProxyNode vpn) { if (vpn.isDeleted()) { return; } AbstractBeginNode proxyPoint = vpn.proxyPoint(); if (proxyPoint instanceof LoopExitNode) { LoopExitNode exit = (LoopExitNode) proxyPoint; LoopBeginNode loopBegin = exit.loopBegin(); Node vpnValue = vpn.value(); for (ValueNode v : loopBegin.stateAfter().values()) { ValueNode v2 = v; if (loopBegin.isPhiAtMerge(v2)) { v2 = ((PhiNode) v2).valueAt(loopBegin.forwardEnd()); } if (vpnValue == v2) { Collection<PhiNode> phiUsages = vpn.usages().filter(PhiNode.class).snapshot(); Collection<ProxyNode> proxyUsages = vpn.usages().filter(ProxyNode.class).snapshot(); vpn.replaceAtUsagesAndDelete(vpnValue); for (PhiNode phi : phiUsages) { checkRedundantPhi(phi); } for (ProxyNode proxy : proxyUsages) { checkRedundantProxy(proxy); } return; } } } }
Remove loop header without loop ends. This can happen with degenerated loops like this one:
for (;;) {
    try {
        break;
    } catch (UnresolvedException iioe) {
    }
}
/** * Remove loop header without loop ends. This can happen with degenerated loops like this one: * * <pre> * for (;;) { * try { * break; * } catch (UnresolvedException iioe) { * } * } * </pre> */
public static void normalizeLoops(StructuredGraph graph) { boolean loopRemoved = false; for (LoopBeginNode begin : graph.getNodes(LoopBeginNode.TYPE)) { if (begin.loopEnds().isEmpty()) { assert begin.forwardEndCount() == 1; graph.reduceDegenerateLoopBegin(begin); loopRemoved = true; } else { normalizeLoopBegin(begin); } } if (loopRemoved) { /* * Removing a degenerated loop can make non-loop phi functions unnecessary. Therefore, * we re-check all phi functions and remove redundant ones. */ for (Node node : graph.getNodes()) { if (node instanceof PhiNode) { checkRedundantPhi((PhiNode) node); } } } } private static void normalizeLoopBegin(LoopBeginNode begin) { // Delete unnecessary loop phi functions, i.e., phi functions where all inputs are either // the same or the phi itself. for (PhiNode phi : begin.phis().snapshot()) { GraphUtil.checkRedundantPhi(phi); } for (LoopExitNode exit : begin.loopExits()) { for (ProxyNode vpn : exit.proxies().snapshot()) { GraphUtil.checkRedundantProxy(vpn); } } }
Gets an approximate source code location for a node if possible.
Returns:the StackTraceElements if an approximate source location is found, null otherwise
/** * Gets an approximate source code location for a node if possible. * * @return the StackTraceElements if an approximate source location is found, null otherwise */
public static StackTraceElement[] approxSourceStackTraceElement(Node node) { NodeSourcePosition position = node.getNodeSourcePosition(); if (position != null) { // use GraphBuilderConfiguration and enable trackNodeSourcePosition to get better source // positions. return approxSourceStackTraceElement(position); } ArrayList<StackTraceElement> elements = new ArrayList<>(); Node n = node; while (n != null) { if (n instanceof MethodCallTargetNode) { elements.add(((MethodCallTargetNode) n).targetMethod().asStackTraceElement(-1)); n = ((MethodCallTargetNode) n).invoke().asNode(); } if (n instanceof StateSplit) { FrameState state = ((StateSplit) n).stateAfter(); elements.addAll(Arrays.asList(approxSourceStackTraceElement(state))); break; } n = n.predecessor(); } return elements.toArray(new StackTraceElement[elements.size()]); }
Gets an approximate source code location for frame state.
Returns:the StackTraceElements if an approximate source location is found, null otherwise
/** * Gets an approximate source code location for frame state. * * @return the StackTraceElements if an approximate source location is found, null otherwise */
public static StackTraceElement[] approxSourceStackTraceElement(FrameState frameState) { ArrayList<StackTraceElement> elements = new ArrayList<>(); FrameState state = frameState; while (state != null) { Bytecode code = state.getCode(); if (code != null) { elements.add(code.asStackTraceElement(state.bci - 1)); } state = state.outerFrameState(); } return elements.toArray(new StackTraceElement[0]); }
Gets approximate stack trace elements for a bytecode position.
/** * Gets approximate stack trace elements for a bytecode position. */
public static StackTraceElement[] approxSourceStackTraceElement(BytecodePosition bytecodePosition) { ArrayList<StackTraceElement> elements = new ArrayList<>(); BytecodePosition position = bytecodePosition; while (position != null) { ResolvedJavaMethod method = position.getMethod(); if (method != null) { elements.add(method.asStackTraceElement(position.getBCI())); } position = position.getCaller(); } return elements.toArray(new StackTraceElement[0]); }
Gets an approximate source code location for a node, encoded as an exception, if possible.
Returns:the exception with the location
/** * Gets an approximate source code location for a node, encoded as an exception, if possible. * * @return the exception with the location */
public static RuntimeException approxSourceException(Node node, Throwable cause) { final StackTraceElement[] elements = approxSourceStackTraceElement(node); return createBailoutException(cause == null ? "" : cause.getMessage(), cause, elements); }
Creates a bailout exception with the given stack trace elements and message.
Params:
  • message – the message of the exception
  • elements – the stack trace elements
Returns:the exception
/** * Creates a bailout exception with the given stack trace elements and message. * * @param message the message of the exception * @param elements the stack trace elements * @return the exception */
public static BailoutException createBailoutException(String message, Throwable cause, StackTraceElement[] elements) { return SourceStackTraceBailoutException.create(cause, message, elements); }
Gets an approximate source code location for a node if possible.
Returns:a file name and source line number in stack trace format (e.g. "String.java:32") if an approximate source location is found, null otherwise
/** * Gets an approximate source code location for a node if possible. * * @return a file name and source line number in stack trace format (e.g. "String.java:32") if * an approximate source location is found, null otherwise */
public static String approxSourceLocation(Node node) { StackTraceElement[] stackTraceElements = approxSourceStackTraceElement(node); if (stackTraceElements != null && stackTraceElements.length > 0) { StackTraceElement top = stackTraceElements[0]; if (top.getFileName() != null && top.getLineNumber() >= 0) { return top.getFileName() + ":" + top.getLineNumber(); } } return null; }
Returns a string representation of the given collection of objects.
Params:
  • objects – The Iterable that will be used to iterate over the objects.
Returns:A string of the format "[a, b, ...]".
/** * Returns a string representation of the given collection of objects. * * @param objects The {@link Iterable} that will be used to iterate over the objects. * @return A string of the format "[a, b, ...]". */
public static String toString(Iterable<?> objects) { StringBuilder str = new StringBuilder(); str.append("["); for (Object o : objects) { str.append(o).append(", "); } if (str.length() > 1) { str.setLength(str.length() - 2); } str.append("]"); return str.toString(); }
Gets the original value by iterating through all ValueProxies.
Params:
  • value – the start value.
Returns:the first non-proxy value encountered
/** * Gets the original value by iterating through all {@link ValueProxy ValueProxies}. * * @param value the start value. * @return the first non-proxy value encountered */
public static ValueNode unproxify(ValueNode value) { if (value instanceof ValueProxy) { return unproxify((ValueProxy) value); } else { return value; } }
Gets the original value by iterating through all ValueProxies.
Params:
  • value – the start value proxy.
Returns:the first non-proxy value encountered
/** * Gets the original value by iterating through all {@link ValueProxy ValueProxies}. * * @param value the start value proxy. * @return the first non-proxy value encountered */
public static ValueNode unproxify(ValueProxy value) { if (value != null) { ValueNode result = value.getOriginalNode(); while (result instanceof ValueProxy) { result = ((ValueProxy) result).getOriginalNode(); } return result; } else { return null; } } public static ValueNode skipPi(ValueNode node) { ValueNode n = node; while (n instanceof PiNode) { PiNode piNode = (PiNode) n; n = piNode.getOriginalNode(); } return n; } public static ValueNode skipPiWhileNonNull(ValueNode node) { ValueNode n = node; while (n instanceof PiNode) { PiNode piNode = (PiNode) n; ObjectStamp originalStamp = (ObjectStamp) piNode.getOriginalNode().stamp(NodeView.DEFAULT); if (originalStamp.nonNull()) { n = piNode.getOriginalNode(); } else { break; } } return n; }
Returns the length of the array described by the value parameter, or null if it is not available. Details of the different modes are documented in FindLengthMode.
Params:
  • value – The start value.
  • mode – The mode as documented in FindLengthMode.
Returns:The array length if one was found, or null otherwise.
/** * Returns the length of the array described by the value parameter, or null if it is not * available. Details of the different modes are documented in {@link FindLengthMode}. * * @param value The start value. * @param mode The mode as documented in {@link FindLengthMode}. * @return The array length if one was found, or null otherwise. */
public static ValueNode arrayLength(ValueNode value, FindLengthMode mode, ConstantReflectionProvider constantReflection) { Objects.requireNonNull(mode); ValueNode current = value; do { /* * PiArrayNode implements ArrayLengthProvider and ValueProxy. We want to treat it as an * ArrayLengthProvider, therefore we check this case first. */ if (current instanceof ArrayLengthProvider) { return ((ArrayLengthProvider) current).findLength(mode, constantReflection); } else if (current instanceof ValuePhiNode) { return phiArrayLength((ValuePhiNode) current, mode, constantReflection); } else if (current instanceof ValueProxyNode) { ValueProxyNode proxy = (ValueProxyNode) current; ValueNode length = arrayLength(proxy.getOriginalNode(), mode, constantReflection); if (mode == ArrayLengthProvider.FindLengthMode.CANONICALIZE_READ && length != null && !length.isConstant()) { length = new ValueProxyNode(length, proxy.proxyPoint()); } return length; } else if (current instanceof ValueProxy) { /* Written as a loop instead of a recursive call to reduce recursion depth. */ current = ((ValueProxy) current).getOriginalNode(); } else { return null; } } while (true); } private static ValueNode phiArrayLength(ValuePhiNode phi, ArrayLengthProvider.FindLengthMode mode, ConstantReflectionProvider constantReflection) { if (phi.merge() instanceof LoopBeginNode) { /* Avoid cycle detection by not processing phi functions that could introduce cycles. */ return null; } ValueNode singleLength = null; for (int i = 0; i < phi.values().count(); i++) { ValueNode input = phi.values().get(i); ValueNode length = arrayLength(input, mode, constantReflection); if (length == null) { return null; } assert length.stamp(NodeView.DEFAULT).getStackKind() == JavaKind.Int; if (i == 0) { assert singleLength == null; singleLength = length; } else if (singleLength == length) { /* Nothing to do, still having a single length. */ } else { return null; } } return singleLength; }
Tries to find an original value of the given node by traversing through proxies and unambiguous phis. Note that this method will perform an exhaustive search through phis.
Params:
  • value – the node whose original value should be determined
  • abortOnLoopPhi – specifies if the traversal through phis should stop and return value if it hits a loop phi. This argument must be true if used during graph building as loop phi nodes may not yet have all their inputs computed.
Returns:the original value (which might be value itself)
/** * Tries to find an original value of the given node by traversing through proxies and * unambiguous phis. Note that this method will perform an exhaustive search through phis. * * @param value the node whose original value should be determined * @param abortOnLoopPhi specifies if the traversal through phis should stop and return * {@code value} if it hits a {@linkplain PhiNode#isLoopPhi loop phi}. This argument * must be {@code true} if used during graph building as loop phi nodes may not yet * have all their inputs computed. * @return the original value (which might be {@code value} itself) */
public static ValueNode originalValue(ValueNode value, boolean abortOnLoopPhi) { ValueNode result = originalValueSimple(value, abortOnLoopPhi); assert result != null; return result; } private static ValueNode originalValueSimple(ValueNode value, boolean abortOnLoopPhi) { /* The very simple case: look through proxies. */ ValueNode cur = originalValueForProxy(value); while (cur instanceof PhiNode) { /* * We found a phi function. Check if we can analyze it without allocating temporary data * structures. */ PhiNode phi = (PhiNode) cur; if (abortOnLoopPhi && phi.isLoopPhi()) { return value; } ValueNode phiSingleValue = null; int count = phi.valueCount(); for (int i = 0; i < count; ++i) { ValueNode phiCurValue = originalValueForProxy(phi.valueAt(i)); if (phiCurValue == phi) { /* Simple cycle, we can ignore the input value. */ } else if (phiSingleValue == null) { /* The first input. */ phiSingleValue = phiCurValue; } else if (phiSingleValue != phiCurValue) { /* Another input that is different from the first input. */ if (phiSingleValue instanceof PhiNode || phiCurValue instanceof PhiNode) { /* * We have two different input values for the phi function, and at least one * of the inputs is another phi function. We need to do a complicated * exhaustive check. */ return originalValueForComplicatedPhi(value, phi, new NodeBitMap(value.graph()), abortOnLoopPhi); } else { /* * We have two different input values for the phi function, but none of them * is another phi function. This phi function cannot be reduce any further, * so the phi function is the original value. */ return phi; } } } /* * Successfully reduced the phi function to a single input value. The single input value * can itself be a phi function again, so we might take another loop iteration. */ assert phiSingleValue != null; cur = phiSingleValue; } /* We reached a "normal" node, which is the original value. */ assert !(cur instanceof LimitedValueProxy) && !(cur instanceof PhiNode); return cur; } private static ValueNode originalValueForProxy(ValueNode value) { ValueNode cur = value; while (cur instanceof LimitedValueProxy) { cur = ((LimitedValueProxy) cur).getOriginalNode(); } return cur; }
Handling for complicated nestings of phi functions. We need to reduce phi functions recursively, and need a temporary map of visited nodes to avoid endless recursion of cycles.
Params:
  • value – the node whose original value is being determined
  • abortOnLoopPhi – specifies if the traversal through phis should stop and return value if it hits a loop phi
/** * Handling for complicated nestings of phi functions. We need to reduce phi functions * recursively, and need a temporary map of visited nodes to avoid endless recursion of cycles. * * @param value the node whose original value is being determined * @param abortOnLoopPhi specifies if the traversal through phis should stop and return * {@code value} if it hits a {@linkplain PhiNode#isLoopPhi loop phi} */
private static ValueNode originalValueForComplicatedPhi(ValueNode value, PhiNode phi, NodeBitMap visited, boolean abortOnLoopPhi) { if (visited.isMarked(phi)) { /* * Found a phi function that was already seen. Either a cycle, or just a second phi * input to a path we have already processed. */ return null; } visited.mark(phi); ValueNode phiSingleValue = null; int count = phi.valueCount(); for (int i = 0; i < count; ++i) { ValueNode phiCurValue = originalValueForProxy(phi.valueAt(i)); if (phiCurValue instanceof PhiNode) { /* Recursively process a phi function input. */ PhiNode curPhi = (PhiNode) phiCurValue; if (abortOnLoopPhi && curPhi.isLoopPhi()) { return value; } phiCurValue = originalValueForComplicatedPhi(value, curPhi, visited, abortOnLoopPhi); if (phiCurValue == value) { // Hit a loop phi assert abortOnLoopPhi; return value; } } if (phiCurValue == null) { /* Cycle to a phi function that was already seen. We can ignore this input. */ } else if (phiSingleValue == null) { /* The first input. */ phiSingleValue = phiCurValue; } else if (phiCurValue != phiSingleValue) { /* * Another input that is different from the first input. Since we already * recursively looked through other phi functions, we now know that this phi * function cannot be reduce any further, so the phi function is the original value. */ return phi; } } return phiSingleValue; } public static boolean tryKillUnused(Node node) { if (node.isAlive() && isFloatingNode(node) && node.hasNoUsages() && !(node instanceof GuardNode)) { killWithUnusedFloatingInputs(node); return true; } return false; }
Returns an iterator that will return the given node followed by all its predecessors, up until the point where Node.predecessor() returns null.
Params:
  • start – the node at which to start iterating
/** * Returns an iterator that will return the given node followed by all its predecessors, up * until the point where {@link Node#predecessor()} returns null. * * @param start the node at which to start iterating */
public static NodeIterable<FixedNode> predecessorIterable(final FixedNode start) { return new NodeIterable<FixedNode>() { @Override public Iterator<FixedNode> iterator() { return new Iterator<FixedNode>() { public FixedNode current = start; @Override public boolean hasNext() { return current != null; } @Override public FixedNode next() { try { return current; } finally { current = (FixedNode) current.predecessor(); } } }; } }; } private static final class DefaultSimplifierTool extends CoreProvidersDelegate implements SimplifierTool { private final boolean canonicalizeReads; private final Assumptions assumptions; private final OptionValues options; DefaultSimplifierTool(CoreProviders providers, boolean canonicalizeReads, Assumptions assumptions, OptionValues options) { super(providers); this.canonicalizeReads = canonicalizeReads; this.assumptions = assumptions; this.options = options; } @Override public boolean canonicalizeReads() { return canonicalizeReads; } @Override public boolean allUsagesAvailable() { return true; } @Override public void deleteBranch(Node branch) { FixedNode fixedBranch = (FixedNode) branch; fixedBranch.predecessor().replaceFirstSuccessor(fixedBranch, null); GraphUtil.killCFG(fixedBranch); } @Override public void removeIfUnused(Node node) { GraphUtil.tryKillUnused(node); } @Override public void addToWorkList(Node node) { } @Override public void addToWorkList(Iterable<? extends Node> nodes) { } @Override public Assumptions getAssumptions() { return assumptions; } @Override public OptionValues getOptions() { return options; } @Override public Integer smallestCompareWidth() { if (getLowerer() != null) { return getLowerer().smallestCompareWidth(); } else { return null; } } } public static SimplifierTool getDefaultSimplifier(CoreProviders providers, boolean canonicalizeReads, Assumptions assumptions, OptionValues options) { return new DefaultSimplifierTool(providers, canonicalizeReads, assumptions, options); } public static Constant foldIfConstantAndRemove(ValueNode node, ValueNode constant) { assert node.inputs().contains(constant); if (constant.isConstant()) { node.replaceFirstInput(constant, null); Constant result = constant.asConstant(); tryKillUnused(constant); return result; } return null; }
Virtualize an array copy.
Params:
  • tool – the virtualization tool
  • source – the source array
  • sourceLength – the length of the source array
  • newLength – the length of the new array
  • from – the start index in the source array
  • newComponentType – the component type of the new array
  • elementKind – the kind of the new array elements
  • graph – the node graph
  • virtualArrayProvider – a functional provider that returns a new virtual array given the component type and length
/** * Virtualize an array copy. * * @param tool the virtualization tool * @param source the source array * @param sourceLength the length of the source array * @param newLength the length of the new array * @param from the start index in the source array * @param newComponentType the component type of the new array * @param elementKind the kind of the new array elements * @param graph the node graph * @param virtualArrayProvider a functional provider that returns a new virtual array given the * component type and length */
public static void virtualizeArrayCopy(VirtualizerTool tool, ValueNode source, ValueNode sourceLength, ValueNode newLength, ValueNode from, ResolvedJavaType newComponentType, JavaKind elementKind, StructuredGraph graph, BiFunction<ResolvedJavaType, Integer, VirtualArrayNode> virtualArrayProvider) { ValueNode sourceAlias = tool.getAlias(source); ValueNode replacedSourceLength = tool.getAlias(sourceLength); ValueNode replacedNewLength = tool.getAlias(newLength); ValueNode replacedFrom = tool.getAlias(from); if (!replacedNewLength.isConstant() || !replacedFrom.isConstant() || !replacedSourceLength.isConstant()) { return; } assert newComponentType != null : "An array copy can be virtualized only if the real type of the resulting array is known statically."; int fromInt = replacedFrom.asJavaConstant().asInt(); int newLengthInt = replacedNewLength.asJavaConstant().asInt(); int sourceLengthInt = replacedSourceLength.asJavaConstant().asInt(); if (sourceAlias instanceof VirtualObjectNode) { VirtualObjectNode sourceVirtual = (VirtualObjectNode) sourceAlias; assert sourceLengthInt == sourceVirtual.entryCount(); } if (fromInt < 0 || newLengthInt < 0 || fromInt > sourceLengthInt) { /* Illegal values for either from index, the new length or the source length. */ return; } if (newLengthInt > tool.getMaximumEntryCount()) { /* The new array size is higher than maximum allowed size of virtualized objects. */ return; } ValueNode[] newEntryState = new ValueNode[newLengthInt]; int readLength = Math.min(newLengthInt, sourceLengthInt - fromInt); if (sourceAlias instanceof VirtualObjectNode) { /* The source array is virtualized, just copy over the values. */ VirtualObjectNode sourceVirtual = (VirtualObjectNode) sourceAlias; boolean alwaysAssignable = newComponentType.getJavaKind() == JavaKind.Object && newComponentType.isJavaLangObject(); for (int i = 0; i < readLength; i++) { ValueNode entry = tool.getEntry(sourceVirtual, fromInt + i); if (!alwaysAssignable) { ResolvedJavaType entryType = StampTool.typeOrNull(entry, tool.getMetaAccess()); if (entryType == null) { return; } if (!newComponentType.isAssignableFrom(entryType)) { return; } } newEntryState[i] = entry; } } else { /* The source array is not virtualized, emit index loads. */ ResolvedJavaType sourceType = StampTool.typeOrNull(sourceAlias, tool.getMetaAccess()); if (sourceType == null || !sourceType.isArray() || !newComponentType.isAssignableFrom(sourceType.getElementalType())) { return; } for (int i = 0; i < readLength; i++) { LoadIndexedNode load = new LoadIndexedNode(null, sourceAlias, ConstantNode.forInt(i + fromInt, graph), null, elementKind); tool.addNode(load); newEntryState[i] = load; } } if (readLength < newLengthInt) { /* Pad the copy with the default value of its elment kind. */ ValueNode defaultValue = ConstantNode.defaultForKind(elementKind, graph); for (int i = readLength; i < newLengthInt; i++) { newEntryState[i] = defaultValue; } } /* Perform the replacement. */ VirtualArrayNode newVirtualArray = virtualArrayProvider.apply(newComponentType, newLengthInt); tool.createVirtualObject(newVirtualArray, newEntryState, Collections.<MonitorIdNode> emptyList(), false); tool.replaceWithVirtual(newVirtualArray); }
Snippet lowerings may produce patterns without a frame state on the merge. We need to take extra care when optimizing these patterns.
/** * Snippet lowerings may produce patterns without a frame state on the merge. We need to take * extra care when optimizing these patterns. */
public static boolean checkFrameState(FixedNode start, int maxDepth) { if (maxDepth == 0) { return false; } FixedNode node = start; while (true) { if (node instanceof AbstractMergeNode) { AbstractMergeNode mergeNode = (AbstractMergeNode) node; if (mergeNode.stateAfter() == null) { return false; } else { return true; } } else if (node instanceof StateSplit) { StateSplit stateSplitNode = (StateSplit) node; if (stateSplitNode.stateAfter() != null) { return true; } } if (node instanceof ControlSplitNode) { ControlSplitNode controlSplitNode = (ControlSplitNode) node; for (Node succ : controlSplitNode.cfgSuccessors()) { if (checkFrameState((FixedNode) succ, maxDepth - 1)) { return true; } } return false; } else if (node instanceof FixedWithNextNode) { FixedWithNextNode fixedWithNextNode = (FixedWithNextNode) node; node = fixedWithNextNode.next(); } else if (node instanceof AbstractEndNode) { AbstractEndNode endNode = (AbstractEndNode) node; node = endNode.merge(); } else if (node instanceof ControlSinkNode) { return true; } else { assert false : "unexpected node"; return false; } } } public static boolean mayRemoveSplit(IfNode ifNode) { return GraphUtil.checkFrameState(ifNode.trueSuccessor(), MAX_FRAMESTATE_SEARCH_DEPTH) && GraphUtil.checkFrameState(ifNode.falseSuccessor(), MAX_FRAMESTATE_SEARCH_DEPTH); }
An if node with an empty body at the end of a loop is represented with a LoopEndNode at the end of each path. For some optimizations it is more useful to have a representation of the if statement as a proper diamond with a merge after the two bodies, followed by a LoopEndNode. This method tries to transform the given ifNode into such a form, introducing new phi nodes for the diamond and patching up the loop's phis accordingly. On success, the newly introduced loop end node is returned. If the given ifNode is not an if statement with empty bodies at the end of the loop, the graph is not modified, and null is returned. Note that the diamond representation is not canonical and will be undone by the next application of Node.simplify(SimplifierTool).
/** * An if node with an empty body at the end of a loop is represented with a {@link LoopEndNode} * at the end of each path. For some optimizations it is more useful to have a representation of * the if statement as a proper diamond with a merge after the two bodies, followed by a * {@link LoopEndNode}. This method tries to transform the given {@code ifNode} into such a * form, introducing new phi nodes for the diamond and patching up the loop's phis accordingly. * On success, the newly introduced loop end node is returned. If the given {@code ifNode} is * not an if statement with empty bodies at the end of the loop, the graph is not modified, and * {@code null} is returned. * * Note that the diamond representation is not canonical and will be undone by the next * application of {@link LoopEndNode#simplify(SimplifierTool)}. */
public static LoopEndNode tryToTransformToEmptyLoopDiamond(IfNode ifNode, LoopBeginNode loopBegin) { if (ifNode.trueSuccessor().next() instanceof AbstractEndNode && ifNode.falseSuccessor().next() instanceof AbstractEndNode) { AbstractEndNode trueEnd = (AbstractEndNode) ifNode.trueSuccessor().next(); AbstractEndNode falseEnd = (AbstractEndNode) ifNode.falseSuccessor().next(); if (trueEnd.merge() == loopBegin && falseEnd.merge() == loopBegin) { StructuredGraph graph = loopBegin.graph(); for (PhiNode phi : loopBegin.phis()) { if (!(phi instanceof ValuePhiNode || phi instanceof MemoryPhiNode)) { return null; } } EndNode newTrueEnd = graph.add(new EndNode()); EndNode newFalseEnd = graph.add(new EndNode()); MergeNode merge = graph.add(new MergeNode()); merge.addForwardEnd(newTrueEnd); merge.addForwardEnd(newFalseEnd); ArrayList<PhiNode> replacementPhis = new ArrayList<>(loopBegin.phis().count()); for (PhiNode phi : loopBegin.phis()) { if (phi instanceof ValuePhiNode) { ValuePhiNode valuePhi = (ValuePhiNode) phi; ValuePhiNode newPhi = phi.graph().unique(new ValuePhiNode(valuePhi.stamp(NodeView.DEFAULT), merge, new ValueNode[]{valuePhi.valueAt(trueEnd), valuePhi.valueAt(falseEnd)})); replacementPhis.add(newPhi); } else if (phi instanceof MemoryPhiNode) { MemoryPhiNode memoryPhi = (MemoryPhiNode) phi; MemoryPhiNode newPhi = phi.graph().unique(new MemoryPhiNode(merge, memoryPhi.getLocationIdentity(), new ValueNode[]{memoryPhi.valueAt(trueEnd), memoryPhi.valueAt(falseEnd)})); replacementPhis.add(newPhi); } else { GraalError.shouldNotReachHere(); } } loopBegin.removeEnd(trueEnd); loopBegin.removeEnd(falseEnd); ifNode.trueSuccessor().setNext(newTrueEnd); ifNode.falseSuccessor().setNext(newFalseEnd); trueEnd.safeDelete(); falseEnd.safeDelete(); LoopEndNode newEnd = graph.add(new LoopEndNode(loopBegin)); merge.setNext(newEnd); int i = 0; for (PhiNode phi : loopBegin.phis()) { ValueNode replacementPhi = replacementPhis.get(i); assert (phi instanceof ValuePhiNode && replacementPhi instanceof ValuePhiNode) || (phi instanceof MemoryPhiNode && replacementPhi instanceof MemoryPhiNode); phi.addInput(replacementPhi); i++; } assert i == replacementPhis.size() : "did not consume all values"; return newEnd; } } return null; } }