1 /*
2 * Copyright (c) 2010, 2024, 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 #ifndef SHARE_COMPILER_COMPILATIONPOLICY_HPP
26 #define SHARE_COMPILER_COMPILATIONPOLICY_HPP
27
28 #include "code/nmethod.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "oops/methodData.hpp"
31 #include "utilities/globalDefinitions.hpp"
32
33 class CompileTask;
34 class CompileQueue;
35 /*
36 * The system supports 5 execution levels:
37 * * level 0 - interpreter (Profiling is tracked by a MethodData object, or MDO in short)
38 * * level 1 - C1 with full optimization (no profiling)
39 * * level 2 - C1 with invocation and backedge counters
40 * * level 3 - C1 with full profiling (level 2 + All other MDO profiling information)
41 * * level 4 - C2 with full profile guided optimization
42 *
43 * The MethodData object is created by both the interpreter or either compiler to store any
44 * profiling information collected on a method (ciMethod::ensure_method_data() for C1 and C2
45 * and CompilationPolicy::create_mdo() for the interpreter). Both the interpreter and code
46 * compiled by C1 at level 3 will constantly update profiling information in the MDO during
47 * execution. The information in the MDO is then used by C1 and C2 during compilation, via
48 * the compiler interface (ciMethodXXX).
49 * See ciMethod.cpp and ciMethodData.cpp for information transfer from an MDO to the compilers
50 * through the compiler interface.
51 *
52 * Levels 0, 2 and 3 periodically notify the runtime about the current value of the counters
53 * (invocation counters and backedge counters). The frequency of these notifications is
54 * different at each level. These notifications are used by the policy to decide what transition
55 * to make.
56 *
57 * Execution starts at level 0 (interpreter), then the policy can decide either to compile the
58 * method at level 3 or level 2. The decision is based on the following factors:
59 * 1. The length of the C2 queue determines the next level. The observation is that level 2
60 * is generally faster than level 3 by about 30%, therefore we would want to minimize the time
61 * a method spends at level 3. We should only spend the time at level 3 that is necessary to get
62 * adequate profiling. So, if the C2 queue is long enough it is more beneficial to go first to
63 * level 2, because if we transitioned to level 3 we would be stuck there until our C2 compile
64 * request makes its way through the long queue. When the load on C2 recedes we are going to
65 * recompile at level 3 and start gathering profiling information.
66 * 2. The length of C1 queue is used to dynamically adjust the thresholds, so as to introduce
67 * additional filtering if the compiler is overloaded. The rationale is that by the time a
68 * method gets compiled it can become unused, so it doesn't make sense to put too much onto the
69 * queue.
70 *
71 * After profiling is completed at level 3 the transition is made to level 4. Again, the length
72 * of the C2 queue is used as a feedback to adjust the thresholds.
73 *
74 * After the first C1 compile some basic information is determined about the code like the number
75 * of the blocks and the number of the loops. Based on that it can be decided that a method
76 * is trivial and compiling it with C1 will yield the same code. In this case the method is
77 * compiled at level 1 instead of 4.
78 *
79 * We also support profiling at level 0. If C1 is slow enough to produce the level 3 version of
80 * the code and the C2 queue is sufficiently small we can decide to start profiling in the
81 * interpreter (and continue profiling in the compiled code once the level 3 version arrives).
82 * If the profiling at level 0 is fully completed before level 3 version is produced, a level 2
83 * version is compiled instead in order to run faster waiting for a level 4 version.
84 *
85 * Compile queues are implemented as priority queues - for each method in the queue we compute
86 * the event rate (the number of invocation and backedge counter increments per unit of time).
87 * When getting an element off the queue we pick the one with the largest rate. Maintaining the
88 * rate also allows us to remove stale methods (the ones that got on the queue but stopped
89 * being used shortly after that).
90 */
91
92 /* Command line options:
93 * - Tier?InvokeNotifyFreqLog and Tier?BackedgeNotifyFreqLog control the frequency of method
94 * invocation and backedge notifications. Basically every n-th invocation or backedge a mutator thread
95 * makes a call into the runtime.
96 *
97 * - Tier?InvocationThreshold, Tier?CompileThreshold, Tier?BackEdgeThreshold, Tier?MinInvocationThreshold control
98 * compilation thresholds.
99 * Level 2 thresholds are not used and are provided for option-compatibility and potential future use.
100 * Other thresholds work as follows:
101 *
102 * Transition from interpreter (level 0) to C1 with full profiling (level 3) happens when
103 * the following predicate is true (X is the level):
104 *
105 * i > TierXInvocationThreshold * s || (i > TierXMinInvocationThreshold * s && i + b > TierXCompileThreshold * s),
106 *
107 * where $i$ is the number of method invocations, $b$ number of backedges and $s$ is the scaling
108 * coefficient that will be discussed further.
109 * The intuition is to equalize the time that is spend profiling each method.
110 * The same predicate is used to control the transition from level 3 to level 4 (C2). It should be
111 * noted though that the thresholds are relative. Moreover i and b for the 0->3 transition come
112 * from Method* and for 3->4 transition they come from MDO (since profiled invocations are
113 * counted separately). Finally, if a method does not contain anything worth profiling, a transition
114 * from level 3 to level 4 occurs without considering thresholds (e.g., with fewer invocations than
115 * what is specified by Tier4InvocationThreshold).
116 *
117 * OSR transitions are controlled simply with b > TierXBackEdgeThreshold * s predicates.
118 *
119 * - Tier?LoadFeedback options are used to automatically scale the predicates described above depending
120 * on the compiler load. The scaling coefficients are computed as follows:
121 *
122 * s = queue_size_X / (TierXLoadFeedback * compiler_count_X) + 1,
123 *
124 * where queue_size_X is the current size of the compiler queue of level X, and compiler_count_X
125 * is the number of level X compiler threads.
126 *
127 * Basically these parameters describe how many methods should be in the compile queue
128 * per compiler thread before the scaling coefficient increases by one.
129 *
130 * This feedback provides the mechanism to automatically control the flow of compilation requests
131 * depending on the machine speed, mutator load and other external factors.
132 *
133 * - Tier3DelayOn and Tier3DelayOff parameters control another important feedback loop.
134 * Consider the following observation: a method compiled with full profiling (level 3)
135 * is about 30% slower than a method at level 2 (just invocation and backedge counters, no MDO).
136 * Normally, the following transitions will occur: 0->3->4. The problem arises when the C2 queue
137 * gets congested and the 3->4 transition is delayed. While the method is the C2 queue it continues
138 * executing at level 3 for much longer time than is required by the predicate and at suboptimal speed.
139 * The idea is to dynamically change the behavior of the system in such a way that if a substantial
140 * load on C2 is detected we would first do the 0->2 transition allowing a method to run faster.
141 * And then when the load decreases to allow 2->3 transitions.
142 *
143 * Tier3Delay* parameters control this switching mechanism.
144 * Tier3DelayOn is the number of methods in the C2 queue per compiler thread after which the policy
145 * no longer does 0->3 transitions but does 0->2 transitions instead.
146 * Tier3DelayOff switches the original behavior back when the number of methods in the C2 queue
147 * per compiler thread falls below the specified amount.
148 * The hysteresis is necessary to avoid jitter.
149 *
150 * - TieredCompileTaskTimeout is the amount of time an idle method can spend in the compile queue.
151 * Basically, since we use the event rate d(i + b)/dt as a value of priority when selecting a method to
152 * compile from the compile queue, we also can detect stale methods for which the rate has been
153 * 0 for some time in the same iteration. Stale methods can appear in the queue when an application
154 * abruptly changes its behavior.
155 *
156 * - TieredStopAtLevel, is used mostly for testing. It allows to bypass the policy logic and stick
157 * to a given level. For example it's useful to set TieredStopAtLevel = 1 in order to compile everything
158 * with pure c1.
159 *
160 * - Tier0ProfilingStartPercentage allows the interpreter to start profiling when the inequalities in the
161 * 0->3 predicate are already exceeded by the given percentage but the level 3 version of the
162 * method is still not ready. We can even go directly from level 0 to 4 if c1 doesn't produce a compiled
163 * version in time. This reduces the overall transition to level 4 and decreases the startup time.
164 * Note that this behavior is also guarded by the Tier3Delay mechanism: when the c2 queue is too long
165 * these is not reason to start profiling prematurely.
166 *
167 * - TieredRateUpdateMinTime and TieredRateUpdateMaxTime are parameters of the rate computation.
168 * Basically, the rate is not computed more frequently than TieredRateUpdateMinTime and is considered
169 * to be zero if no events occurred in TieredRateUpdateMaxTime.
170 */
171
172 class CompilationPolicy : AllStatic {
173 friend class CallPredicate;
174 friend class LoopPredicate;
175
176 static jlong _start_time;
177 static int _c1_count, _c2_count;
178 static double _increase_threshold_at_ratio;
179
180 // Set carry flags in the counters (in Method* and MDO).
181 inline static void handle_counter_overflow(const methodHandle& method);
182 #ifdef ASSERT
183 // Verify that a level is consistent with the compilation mode
184 static bool verify_level(CompLevel level);
185 #endif
186 // Clamp the request level according to various constraints.
187 inline static CompLevel limit_level(CompLevel level);
188 // Common transition function. Given a predicate determines if a method should transition to another level.
189 template<typename Predicate>
190 static CompLevel common(const methodHandle& method, CompLevel cur_level, bool disable_feedback = false);
191 // Transition functions.
192 // call_event determines if a method should be compiled at a different
193 // level with a regular invocation entry.
194 static CompLevel call_event(const methodHandle& method, CompLevel cur_level, Thread* thread);
195 // loop_event checks if a method should be OSR compiled at a different
196 // level.
197 static CompLevel loop_event(const methodHandle& method, CompLevel cur_level, Thread* thread);
198 static void print_counters(const char* prefix, const Method* m);
199 // Has a method been long around?
200 // We don't remove old methods from the compile queue even if they have
201 // very low activity (see select_task()).
202 inline static bool is_old(const methodHandle& method);
203 // Was a given method inactive for a given number of milliseconds.
204 // If it is, we would remove it from the queue (see select_task()).
205 inline static bool is_stale(jlong t, jlong timeout, const methodHandle& method);
206 // Compute the weight of the method for the compilation scheduling
207 inline static double weight(Method* method);
208 // Apply heuristics and return true if x should be compiled before y
209 inline static bool compare_methods(Method* x, Method* y);
210 // Compute event rate for a given method. The rate is the number of event (invocations + backedges)
211 // per millisecond.
212 inline static void update_rate(jlong t, const methodHandle& method);
213 // Compute threshold scaling coefficient
214 inline static double threshold_scale(CompLevel level, int feedback_k);
215 // If a method is old enough and is still in the interpreter we would want to
216 // start profiling without waiting for the compiled method to arrive. This function
217 // determines whether we should do that.
218 inline static bool should_create_mdo(const methodHandle& method, CompLevel cur_level);
219 // Create MDO if necessary.
220 static void create_mdo(const methodHandle& mh, JavaThread* THREAD);
221 // Is method profiled enough?
222 static bool is_method_profiled(const methodHandle& method);
223
224 static void set_c1_count(int x) { _c1_count = x; }
225 static void set_c2_count(int x) { _c2_count = x; }
226
227 enum EventType { CALL, LOOP, COMPILE, REMOVE_FROM_QUEUE, UPDATE_IN_QUEUE, REPROFILE, MAKE_NOT_ENTRANT };
228 static void print_event(EventType type, const Method* m, const Method* im, int bci, CompLevel level);
229 // Check if the method can be compiled, change level if necessary
230 static void compile(const methodHandle& mh, int bci, CompLevel level, TRAPS);
231 // Simple methods are as good being compiled with C1 as C2.
232 // This function tells if it's such a function.
233 inline static bool is_trivial(const methodHandle& method);
234 // Force method to be compiled at CompLevel_simple?
235 inline static bool force_comp_at_level_simple(const methodHandle& method);
236
237 // Get a compilation level for a given method.
238 static CompLevel comp_level(Method* method);
239 static void method_invocation_event(const methodHandle& method, const methodHandle& inlinee,
240 CompLevel level, nmethod* nm, TRAPS);
241 static void method_back_branch_event(const methodHandle& method, const methodHandle& inlinee,
242 int bci, CompLevel level, nmethod* nm, TRAPS);
243
244 static void set_increase_threshold_at_ratio() { _increase_threshold_at_ratio = 100 / (100 - (double)IncreaseFirstTierCompileThresholdAt); }
245 static void set_start_time(jlong t) { _start_time = t; }
246 static jlong start_time() { return _start_time; }
247
248 // m must be compiled before executing it
249 static bool must_be_compiled(const methodHandle& m, int comp_level = CompLevel_any);
250 public:
251 static int min_invocations() { return Tier4MinInvocationThreshold; }
252 static int c1_count() { return _c1_count; }
253 static int c2_count() { return _c2_count; }
254 static int compiler_count(CompLevel comp_level);
255
256 // If m must_be_compiled then request a compilation from the CompileBroker.
257 // This supports the -Xcomp option.
258 static void compile_if_required(const methodHandle& m, TRAPS);
259
260 // m is allowed to be compiled
261 static bool can_be_compiled(const methodHandle& m, int comp_level = CompLevel_any);
262 // m is allowed to be osr compiled
263 static bool can_be_osr_compiled(const methodHandle& m, int comp_level = CompLevel_any);
264 static bool is_compilation_enabled();
265
266 static CompileTask* select_task_helper(CompileQueue* compile_queue);
267 // Return initial compile level to use with Xcomp (depends on compilation mode).
268 static void reprofile(ScopeDesc* trap_scope, bool is_osr);
269 static nmethod* event(const methodHandle& method, const methodHandle& inlinee,
270 int branch_bci, int bci, CompLevel comp_level, nmethod* nm, TRAPS);
271 // Select task is called by CompileBroker. We should return a task or nullptr.
272 static CompileTask* select_task(CompileQueue* compile_queue);
273 // Tell the runtime if we think a given method is adequately profiled.
274 static bool is_mature(Method* method);
275 // Initialize: set compiler thread count
276 static void initialize();
277 static bool should_not_inline(ciEnv* env, ciMethod* callee);
278
279 // Return desired initial compilation level for Xcomp
280 static CompLevel initial_compile_level(const methodHandle& method);
281 // Return highest level possible
282 static CompLevel highest_compile_level();
283 };
284
285 #endif // SHARE_COMPILER_COMPILATIONPOLICY_HPP