1 /*
2 * Copyright (c) 2011, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 */
23
24
25 package org.graalvm.compiler.hotspot.phases;
26
27 import static jdk.vm.ci.meta.SpeculationLog.SpeculationReason;
28 import static org.graalvm.compiler.phases.common.DeadCodeEliminationPhase.Optionality.Required;
29
30 import org.graalvm.compiler.core.common.PermanentBailoutException;
31 import org.graalvm.compiler.core.common.cfg.Loop;
32 import org.graalvm.compiler.core.common.type.ObjectStamp;
33 import org.graalvm.compiler.core.common.type.Stamp;
34 import org.graalvm.compiler.debug.CounterKey;
35 import org.graalvm.compiler.debug.DebugCloseable;
36 import org.graalvm.compiler.debug.DebugContext;
37 import org.graalvm.compiler.debug.GraalError;
38 import org.graalvm.compiler.graph.Node;
39 import org.graalvm.compiler.graph.iterators.NodeIterable;
40 import org.graalvm.compiler.loop.LoopsData;
41 import org.graalvm.compiler.loop.phases.LoopTransformations;
42 import org.graalvm.compiler.nodeinfo.InputType;
43 import org.graalvm.compiler.nodeinfo.Verbosity;
44 import org.graalvm.compiler.nodes.AbstractBeginNode;
45 import org.graalvm.compiler.nodes.EntryMarkerNode;
46 import org.graalvm.compiler.nodes.EntryProxyNode;
47 import org.graalvm.compiler.nodes.FixedGuardNode;
48 import org.graalvm.compiler.nodes.FixedNode;
49 import org.graalvm.compiler.nodes.FrameState;
50 import org.graalvm.compiler.nodes.LogicNode;
51 import org.graalvm.compiler.nodes.LoopBeginNode;
52 import org.graalvm.compiler.nodes.NodeView;
53 import org.graalvm.compiler.nodes.ParameterNode;
54 import org.graalvm.compiler.nodes.PiNode;
55 import org.graalvm.compiler.nodes.StartNode;
56 import org.graalvm.compiler.nodes.StructuredGraph;
57 import org.graalvm.compiler.nodes.ValueNode;
58 import org.graalvm.compiler.nodes.cfg.Block;
59 import org.graalvm.compiler.nodes.extended.OSRLocalNode;
60 import org.graalvm.compiler.nodes.extended.OSRLockNode;
61 import org.graalvm.compiler.nodes.extended.OSRMonitorEnterNode;
62 import org.graalvm.compiler.nodes.extended.OSRStartNode;
63 import org.graalvm.compiler.nodes.java.AccessMonitorNode;
64 import org.graalvm.compiler.nodes.java.InstanceOfNode;
65 import org.graalvm.compiler.nodes.java.MonitorEnterNode;
66 import org.graalvm.compiler.nodes.java.MonitorExitNode;
67 import org.graalvm.compiler.nodes.java.MonitorIdNode;
68 import org.graalvm.compiler.nodes.util.GraphUtil;
69 import org.graalvm.compiler.options.Option;
70 import org.graalvm.compiler.options.OptionKey;
71 import org.graalvm.compiler.options.OptionType;
72 import org.graalvm.compiler.options.OptionValues;
73 import org.graalvm.compiler.phases.Phase;
74 import org.graalvm.compiler.phases.common.DeadCodeEliminationPhase;
75
76 import jdk.vm.ci.meta.DeoptimizationAction;
77 import jdk.vm.ci.meta.DeoptimizationReason;
78 import jdk.vm.ci.meta.JavaKind;
79 import jdk.vm.ci.meta.SpeculationLog;
80 import jdk.vm.ci.runtime.JVMCICompiler;
81
82 public class OnStackReplacementPhase extends Phase {
83
84 public static class Options {
85 // @formatter:off
86 @Option(help = "Deoptimize OSR compiled code when the OSR entry loop is finished " +
87 "if there is no mature profile available for the rest of the method.", type = OptionType.Debug)
88 public static final OptionKey<Boolean> DeoptAfterOSR = new OptionKey<>(true);
89 @Option(help = "Support OSR compilations with locks. If DeoptAfterOSR is true we can per definition not have " +
90 "unbalaced enter/extis mappings. If DeoptAfterOSR is false insert artificial monitor enters after " +
91 "the OSRStart to have balanced enter/exits in the graph.", type = OptionType.Debug)
92 public static final OptionKey<Boolean> SupportOSRWithLocks = new OptionKey<>(true);
93 // @formatter:on
94 }
95
96 private static final CounterKey OsrWithLocksCount = DebugContext.counter("OSRWithLocks");
97
98 private static boolean supportOSRWithLocks(OptionValues options) {
99 return Options.SupportOSRWithLocks.getValue(options);
100 }
101
102 @Override
103 @SuppressWarnings("try")
104 protected void run(StructuredGraph graph) {
105 DebugContext debug = graph.getDebug();
106 if (graph.getEntryBCI() == JVMCICompiler.INVOCATION_ENTRY_BCI) {
107 // This happens during inlining in a OSR method, because the same phase plan will be
108 // used.
109 assert graph.getNodes(EntryMarkerNode.TYPE).isEmpty();
110 return;
111 }
112 debug.dump(DebugContext.DETAILED_LEVEL, graph, "OnStackReplacement initial at bci %d", graph.getEntryBCI());
113
114 EntryMarkerNode osr;
115 int maxIterations = -1;
116 int iterations = 0;
117
118 final EntryMarkerNode originalOSRNode = getEntryMarker(graph);
119 final LoopBeginNode originalOSRLoop = osrLoop(originalOSRNode);
120 final boolean currentOSRWithLocks = osrWithLocks(originalOSRNode);
121
122 if (originalOSRLoop == null) {
123 /*
124 * OSR with Locks: We do not have an OSR loop for the original OSR bci. Therefore we
125 * cannot decide where to deopt and which framestate will be used. In the worst case the
126 * framestate of the OSR entry would be used.
127 */
128 throw new PermanentBailoutException("OSR compilation without OSR entry loop.");
129 }
130
131 if (!supportOSRWithLocks(graph.getOptions()) && currentOSRWithLocks) {
132 throw new PermanentBailoutException("OSR with locks disabled.");
133 }
134
135 do {
136 osr = getEntryMarker(graph);
137 LoopsData loops = new LoopsData(graph);
138 // Find the loop that contains the EntryMarker
139 Loop<Block> l = loops.getCFG().getNodeToBlock().get(osr).getLoop();
140 if (l == null) {
141 break;
142 }
143
144 iterations++;
145 if (maxIterations == -1) {
146 maxIterations = l.getDepth();
147 } else if (iterations > maxIterations) {
148 throw GraalError.shouldNotReachHere();
149 }
150 // Peel the outermost loop first
151 while (l.getParent() != null) {
152 l = l.getParent();
153 }
154
155 LoopTransformations.peel(loops.loop(l));
156 osr.replaceAtUsages(InputType.Guard, AbstractBeginNode.prevBegin((FixedNode) osr.predecessor()));
157 for (Node usage : osr.usages().snapshot()) {
158 EntryProxyNode proxy = (EntryProxyNode) usage;
159 proxy.replaceAndDelete(proxy.value());
160 }
161 GraphUtil.removeFixedWithUnusedInputs(osr);
162 debug.dump(DebugContext.DETAILED_LEVEL, graph, "OnStackReplacement loop peeling result");
163 } while (true);
164
165 StartNode start = graph.start();
166 FrameState osrState = osr.stateAfter();
167 OSRStartNode osrStart;
168 try (DebugCloseable context = osr.withNodeSourcePosition()) {
169 osr.setStateAfter(null);
170 osrStart = graph.add(new OSRStartNode());
171 FixedNode next = osr.next();
172 osr.setNext(null);
173 osrStart.setNext(next);
174 graph.setStart(osrStart);
175 osrStart.setStateAfter(osrState);
176
177 debug.dump(DebugContext.DETAILED_LEVEL, graph, "OnStackReplacement after setting OSR start");
178 final int localsSize = osrState.localsSize();
179 final int locksSize = osrState.locksSize();
180
181 for (int i = 0; i < localsSize + locksSize; i++) {
182 ValueNode value = null;
183 if (i >= localsSize) {
184 value = osrState.lockAt(i - localsSize);
185 } else {
186 value = osrState.localAt(i);
187 }
188 if (value instanceof EntryProxyNode) {
189 EntryProxyNode proxy = (EntryProxyNode) value;
190 /*
191 * We need to drop the stamp since the types we see during OSR may be too
192 * precise (if a branch was not parsed for example). In cases when this is
193 * possible, we insert a guard and narrow the OSRLocal stamp at its usages.
194 */
195 Stamp narrowedStamp = proxy.value().stamp(NodeView.DEFAULT);
196 Stamp unrestrictedStamp = proxy.stamp(NodeView.DEFAULT).unrestricted();
197 ValueNode osrLocal;
198 if (i >= localsSize) {
199 osrLocal = graph.addOrUnique(new OSRLockNode(i - localsSize, unrestrictedStamp));
200 } else {
201 osrLocal = graph.addOrUnique(new OSRLocalNode(i, unrestrictedStamp));
202 }
203 // Speculate on the OSRLocal stamps that could be more precise.
204 OSRLocalSpeculationReason reason = new OSRLocalSpeculationReason(osrState.bci, narrowedStamp, i);
205 if (graph.getSpeculationLog().maySpeculate(reason) && osrLocal instanceof OSRLocalNode && value.getStackKind().equals(JavaKind.Object) && !narrowedStamp.isUnrestricted()) {
206 // Add guard.
207 LogicNode check = graph.addOrUniqueWithInputs(InstanceOfNode.createHelper((ObjectStamp) narrowedStamp, osrLocal, null, null));
208 SpeculationLog.Speculation constant = graph.getSpeculationLog().speculate(reason);
209 FixedGuardNode guard = graph.add(new FixedGuardNode(check, DeoptimizationReason.OptimizedTypeCheckViolated, DeoptimizationAction.InvalidateRecompile, constant, false));
210 graph.addAfterFixed(osrStart, guard);
211
212 // Replace with a more specific type at usages.
213 // We know that we are at the root,
214 // so we need to replace the proxy in the state.
215 proxy.replaceAtMatchingUsages(osrLocal, n -> n == osrState);
216 osrLocal = graph.addOrUnique(new PiNode(osrLocal, narrowedStamp, guard));
217 }
218 proxy.replaceAndDelete(osrLocal);
219 } else {
220 assert value == null || value instanceof OSRLocalNode;
221 }
222 }
223
224 osr.replaceAtUsages(InputType.Guard, osrStart);
225 }
226 debug.dump(DebugContext.DETAILED_LEVEL, graph, "OnStackReplacement after replacing entry proxies");
227 GraphUtil.killCFG(start);
228 debug.dump(DebugContext.DETAILED_LEVEL, graph, "OnStackReplacement result");
229 new DeadCodeEliminationPhase(Required).apply(graph);
230
231 if (currentOSRWithLocks) {
232 OsrWithLocksCount.increment(debug);
233 try (DebugCloseable context = osrStart.withNodeSourcePosition()) {
234 for (int i = osrState.monitorIdCount() - 1; i >= 0; --i) {
235 MonitorIdNode id = osrState.monitorIdAt(i);
236 ValueNode lockedObject = osrState.lockAt(i);
237 OSRMonitorEnterNode osrMonitorEnter = graph.add(new OSRMonitorEnterNode(lockedObject, id));
238 for (Node usage : id.usages()) {
239 if (usage instanceof AccessMonitorNode) {
240 AccessMonitorNode access = (AccessMonitorNode) usage;
241 access.setObject(lockedObject);
242 }
243 }
244 FixedNode oldNext = osrStart.next();
245 oldNext.replaceAtPredecessor(null);
246 osrMonitorEnter.setNext(oldNext);
247 osrStart.setNext(osrMonitorEnter);
248 }
249 }
250
251 debug.dump(DebugContext.DETAILED_LEVEL, graph, "After inserting OSR monitor enters");
252 /*
253 * Ensure balanced monitorenter - monitorexit
254 *
255 * Ensure that there is no monitor exit without a monitor enter in the graph. If there
256 * is one this can only be done by bytecode as we have the monitor enter before the OSR
257 * loop but the exit in a path of the loop that must be under a condition, else it will
258 * throw an IllegalStateException anyway in the 2.iteration
259 */
260 for (MonitorExitNode exit : graph.getNodes(MonitorExitNode.TYPE)) {
261 MonitorIdNode id = exit.getMonitorId();
262 if (id.usages().filter(MonitorEnterNode.class).count() != 1) {
263 throw new PermanentBailoutException("Unbalanced monitor enter-exit in OSR compilation with locks. Object is locked before the loop but released inside the loop.");
264 }
265 }
266 }
267 debug.dump(DebugContext.DETAILED_LEVEL, graph, "OnStackReplacement result");
268 new DeadCodeEliminationPhase(Required).apply(graph);
269 /*
270 * There must not be any parameter nodes left after OSR compilation.
271 */
272 assert graph.getNodes(ParameterNode.TYPE).count() == 0 : "OSR Compilation contains references to parameters.";
273 }
274
275 private static EntryMarkerNode getEntryMarker(StructuredGraph graph) {
276 NodeIterable<EntryMarkerNode> osrNodes = graph.getNodes(EntryMarkerNode.TYPE);
277 EntryMarkerNode osr = osrNodes.first();
278 if (osr == null) {
279 throw new PermanentBailoutException("No OnStackReplacementNode generated");
280 }
281 if (osrNodes.count() > 1) {
282 throw new GraalError("Multiple OnStackReplacementNodes generated");
283 }
284 if (osr.stateAfter().stackSize() != 0) {
285 throw new PermanentBailoutException("OSR with stack entries not supported: %s", osr.stateAfter().toString(Verbosity.Debugger));
286 }
287 return osr;
288 }
289
290 private static LoopBeginNode osrLoop(EntryMarkerNode osr) {
291 // Check that there is an OSR loop for the OSR begin
292 LoopsData loops = new LoopsData(osr.graph());
293 Loop<Block> l = loops.getCFG().getNodeToBlock().get(osr).getLoop();
294 if (l == null) {
295 return null;
296 }
297 return (LoopBeginNode) l.getHeader().getBeginNode();
298 }
299
300 private static boolean osrWithLocks(EntryMarkerNode osr) {
301 return osr.stateAfter().locksSize() != 0;
302 }
303
304 @Override
305 public float codeSizeIncrease() {
306 return 5.0f;
307 }
308
309 private static class OSRLocalSpeculationReason implements SpeculationReason {
310 private int bci;
311 private Stamp speculatedStamp;
312 private int localIndex;
313
314 OSRLocalSpeculationReason(int bci, Stamp speculatedStamp, int localIndex) {
315 this.bci = bci;
316 this.speculatedStamp = speculatedStamp;
317 this.localIndex = localIndex;
318 }
319
320 @Override
321 public boolean equals(Object obj) {
322 if (obj instanceof OSRLocalSpeculationReason) {
323 OSRLocalSpeculationReason that = (OSRLocalSpeculationReason) obj;
324 return this.bci == that.bci && this.speculatedStamp.equals(that.speculatedStamp) && this.localIndex == that.localIndex;
325 }
326 return false;
327 }
328
329 @Override
330 public int hashCode() {
331 return (bci << 16) ^ speculatedStamp.hashCode() ^ localIndex;
332 }
333 }
334 }