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