iveresov@2630: /*
mikael@6198:  * Copyright (c) 2010, 2013, Oracle and/or its affiliates. All rights reserved.
iveresov@2890:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
iveresov@2890:  *
iveresov@2890:  * This code is free software; you can redistribute it and/or modify it
iveresov@2890:  * under the terms of the GNU General Public License version 2 only, as
iveresov@2890:  * published by the Free Software Foundation.
iveresov@2890:  *
iveresov@2890:  * This code is distributed in the hope that it will be useful, but WITHOUT
iveresov@2890:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
iveresov@2890:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
iveresov@2890:  * version 2 for more details (a copy is included in the LICENSE file that
iveresov@2890:  * accompanied this code).
iveresov@2890:  *
iveresov@2890:  * You should have received a copy of the GNU General Public License version
iveresov@2890:  * 2 along with this work; if not, write to the Free Software Foundation,
iveresov@2890:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
iveresov@2890:  *
iveresov@2890:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
iveresov@2890:  * or visit www.oracle.com if you need additional information or have any
iveresov@2890:  * questions.
iveresov@2890:  *
iveresov@2890:  */
iveresov@2630: 
iveresov@2630: #ifndef SHARE_VM_RUNTIME_ADVANCEDTHRESHOLDPOLICY_HPP
iveresov@2630: #define SHARE_VM_RUNTIME_ADVANCEDTHRESHOLDPOLICY_HPP
iveresov@2630: 
iveresov@2630: #include "runtime/simpleThresholdPolicy.hpp"
iveresov@2630: 
iveresov@2630: #ifdef TIERED
iveresov@2630: class CompileTask;
iveresov@2630: class CompileQueue;
iveresov@2630: 
iveresov@2630: /*
iveresov@2630:  *  The system supports 5 execution levels:
iveresov@2630:  *  * level 0 - interpreter
iveresov@2630:  *  * level 1 - C1 with full optimization (no profiling)
iveresov@2630:  *  * level 2 - C1 with invocation and backedge counters
iveresov@2630:  *  * level 3 - C1 with full profiling (level 2 + MDO)
iveresov@2630:  *  * level 4 - C2
iveresov@2630:  *
iveresov@2630:  * Levels 0, 2 and 3 periodically notify the runtime about the current value of the counters
iveresov@2630:  * (invocation counters and backedge counters). The frequency of these notifications is
iveresov@2630:  * different at each level. These notifications are used by the policy to decide what transition
iveresov@2630:  * to make.
iveresov@2630:  *
iveresov@2630:  * Execution starts at level 0 (interpreter), then the policy can decide either to compile the
iveresov@2630:  * method at level 3 or level 2. The decision is based on the following factors:
iveresov@2630:  *    1. The length of the C2 queue determines the next level. The observation is that level 2
iveresov@2630:  * is generally faster than level 3 by about 30%, therefore we would want to minimize the time
iveresov@2630:  * a method spends at level 3. We should only spend the time at level 3 that is necessary to get
iveresov@2630:  * adequate profiling. So, if the C2 queue is long enough it is more beneficial to go first to
iveresov@2630:  * level 2, because if we transitioned to level 3 we would be stuck there until our C2 compile
iveresov@2630:  * request makes its way through the long queue. When the load on C2 recedes we are going to
iveresov@2630:  * recompile at level 3 and start gathering profiling information.
iveresov@2630:  *    2. The length of C1 queue is used to dynamically adjust the thresholds, so as to introduce
iveresov@2630:  * additional filtering if the compiler is overloaded. The rationale is that by the time a
iveresov@2630:  * method gets compiled it can become unused, so it doesn't make sense to put too much onto the
iveresov@2630:  * queue.
iveresov@2630:  *
iveresov@2630:  * After profiling is completed at level 3 the transition is made to level 4. Again, the length
iveresov@2630:  * of the C2 queue is used as a feedback to adjust the thresholds.
iveresov@2630:  *
iveresov@2630:  * After the first C1 compile some basic information is determined about the code like the number
iveresov@2630:  * of the blocks and the number of the loops. Based on that it can be decided that a method
iveresov@2630:  * is trivial and compiling it with C1 will yield the same code. In this case the method is
iveresov@2630:  * compiled at level 1 instead of 4.
iveresov@2630:  *
iveresov@2630:  * We also support profiling at level 0. If C1 is slow enough to produce the level 3 version of
iveresov@2630:  * the code and the C2 queue is sufficiently small we can decide to start profiling in the
iveresov@2630:  * interpreter (and continue profiling in the compiled code once the level 3 version arrives).
iveresov@2630:  * If the profiling at level 0 is fully completed before level 3 version is produced, a level 2
iveresov@2630:  * version is compiled instead in order to run faster waiting for a level 4 version.
iveresov@2630:  *
iveresov@2630:  * Compile queues are implemented as priority queues - for each method in the queue we compute
iveresov@2630:  * the event rate (the number of invocation and backedge counter increments per unit of time).
iveresov@2630:  * When getting an element off the queue we pick the one with the largest rate. Maintaining the
iveresov@2630:  * rate also allows us to remove stale methods (the ones that got on the queue but stopped
iveresov@2630:  * being used shortly after that).
iveresov@2630: */
iveresov@2630: 
iveresov@2630: /* Command line options:
iveresov@2630:  * - Tier?InvokeNotifyFreqLog and Tier?BackedgeNotifyFreqLog control the frequency of method
iveresov@2630:  *   invocation and backedge notifications. Basically every n-th invocation or backedge a mutator thread
iveresov@2630:  *   makes a call into the runtime.
iveresov@2630:  *
iveresov@2630:  * - Tier?CompileThreshold, Tier?BackEdgeThreshold, Tier?MinInvocationThreshold control
iveresov@2630:  *   compilation thresholds.
iveresov@2630:  *   Level 2 thresholds are not used and are provided for option-compatibility and potential future use.
iveresov@2630:  *   Other thresholds work as follows:
iveresov@2630:  *
iveresov@2630:  *   Transition from interpreter (level 0) to C1 with full profiling (level 3) happens when
iveresov@2630:  *   the following predicate is true (X is the level):
iveresov@2630:  *
iveresov@2630:  *   i > TierXInvocationThreshold * s || (i > TierXMinInvocationThreshold * s  && i + b > TierXCompileThreshold * s),
iveresov@2630:  *
iveresov@2630:  *   where $i$ is the number of method invocations, $b$ number of backedges and $s$ is the scaling
iveresov@2630:  *   coefficient that will be discussed further.
iveresov@2630:  *   The intuition is to equalize the time that is spend profiling each method.
iveresov@2630:  *   The same predicate is used to control the transition from level 3 to level 4 (C2). It should be
iveresov@2630:  *   noted though that the thresholds are relative. Moreover i and b for the 0->3 transition come
coleenp@4037:  *   from Method* and for 3->4 transition they come from MDO (since profiled invocations are
iveresov@2630:  *   counted separately).
iveresov@2630:  *
iveresov@2630:  *   OSR transitions are controlled simply with b > TierXBackEdgeThreshold * s predicates.
iveresov@2630:  *
iveresov@2630:  * - Tier?LoadFeedback options are used to automatically scale the predicates described above depending
iveresov@2630:  *   on the compiler load. The scaling coefficients are computed as follows:
iveresov@2630:  *
iveresov@2630:  *   s = queue_size_X / (TierXLoadFeedback * compiler_count_X) + 1,
iveresov@2630:  *
iveresov@2630:  *   where queue_size_X is the current size of the compiler queue of level X, and compiler_count_X
iveresov@2630:  *   is the number of level X compiler threads.
iveresov@2630:  *
iveresov@2630:  *   Basically these parameters describe how many methods should be in the compile queue
iveresov@2630:  *   per compiler thread before the scaling coefficient increases by one.
iveresov@2630:  *
iveresov@2630:  *   This feedback provides the mechanism to automatically control the flow of compilation requests
iveresov@2630:  *   depending on the machine speed, mutator load and other external factors.
iveresov@2630:  *
iveresov@2630:  * - Tier3DelayOn and Tier3DelayOff parameters control another important feedback loop.
iveresov@2630:  *   Consider the following observation: a method compiled with full profiling (level 3)
iveresov@2630:  *   is about 30% slower than a method at level 2 (just invocation and backedge counters, no MDO).
iveresov@2630:  *   Normally, the following transitions will occur: 0->3->4. The problem arises when the C2 queue
iveresov@2630:  *   gets congested and the 3->4 transition is delayed. While the method is the C2 queue it continues
iveresov@2630:  *   executing at level 3 for much longer time than is required by the predicate and at suboptimal speed.
iveresov@2630:  *   The idea is to dynamically change the behavior of the system in such a way that if a substantial
iveresov@2630:  *   load on C2 is detected we would first do the 0->2 transition allowing a method to run faster.
iveresov@2630:  *   And then when the load decreases to allow 2->3 transitions.
iveresov@2630:  *
iveresov@2630:  *   Tier3Delay* parameters control this switching mechanism.
iveresov@2630:  *   Tier3DelayOn is the number of methods in the C2 queue per compiler thread after which the policy
iveresov@2630:  *   no longer does 0->3 transitions but does 0->2 transitions instead.
iveresov@2630:  *   Tier3DelayOff switches the original behavior back when the number of methods in the C2 queue
iveresov@2630:  *   per compiler thread falls below the specified amount.
iveresov@2630:  *   The hysteresis is necessary to avoid jitter.
iveresov@2630:  *
iveresov@2630:  * - TieredCompileTaskTimeout is the amount of time an idle method can spend in the compile queue.
iveresov@2630:  *   Basically, since we use the event rate d(i + b)/dt as a value of priority when selecting a method to
iveresov@2630:  *   compile from the compile queue, we also can detect stale methods for which the rate has been
iveresov@2630:  *   0 for some time in the same iteration. Stale methods can appear in the queue when an application
iveresov@2630:  *   abruptly changes its behavior.
iveresov@2630:  *
iveresov@2630:  * - TieredStopAtLevel, is used mostly for testing. It allows to bypass the policy logic and stick
iveresov@2630:  *   to a given level. For example it's useful to set TieredStopAtLevel = 1 in order to compile everything
iveresov@2630:  *   with pure c1.
iveresov@2630:  *
iveresov@2630:  * - Tier0ProfilingStartPercentage allows the interpreter to start profiling when the inequalities in the
iveresov@2630:  *   0->3 predicate are already exceeded by the given percentage but the level 3 version of the
iveresov@2630:  *   method is still not ready. We can even go directly from level 0 to 4 if c1 doesn't produce a compiled
iveresov@2630:  *   version in time. This reduces the overall transition to level 4 and decreases the startup time.
iveresov@2630:  *   Note that this behavior is also guarded by the Tier3Delay mechanism: when the c2 queue is too long
iveresov@2630:  *   these is not reason to start profiling prematurely.
iveresov@2630:  *
iveresov@2630:  * - TieredRateUpdateMinTime and TieredRateUpdateMaxTime are parameters of the rate computation.
iveresov@2630:  *   Basically, the rate is not computed more frequently than TieredRateUpdateMinTime and is considered
iveresov@2630:  *   to be zero if no events occurred in TieredRateUpdateMaxTime.
iveresov@2630:  */
iveresov@2630: 
iveresov@2630: 
iveresov@2630: class AdvancedThresholdPolicy : public SimpleThresholdPolicy {
iveresov@2630:   jlong _start_time;
iveresov@2630: 
iveresov@2630:   // Call and loop predicates determine whether a transition to a higher compilation
iveresov@2630:   // level should be performed (pointers to predicate functions are passed to common().
iveresov@2630:   // Predicates also take compiler load into account.
iveresov@2630:   typedef bool (AdvancedThresholdPolicy::*Predicate)(int i, int b, CompLevel cur_level);
iveresov@2630:   bool call_predicate(int i, int b, CompLevel cur_level);
iveresov@2630:   bool loop_predicate(int i, int b, CompLevel cur_level);
iveresov@2630:   // Common transition function. Given a predicate determines if a method should transition to another level.
coleenp@4037:   CompLevel common(Predicate p, Method* method, CompLevel cur_level, bool disable_feedback = false);
iveresov@2630:   // Transition functions.
iveresov@2630:   // call_event determines if a method should be compiled at a different
iveresov@2630:   // level with a regular invocation entry.
coleenp@4037:   CompLevel call_event(Method* method, CompLevel cur_level);
iveresov@2630:   // loop_event checks if a method should be OSR compiled at a different
iveresov@2630:   // level.
coleenp@4037:   CompLevel loop_event(Method* method, CompLevel cur_level);
iveresov@2630:   // Has a method been long around?
iveresov@2630:   // We don't remove old methods from the compile queue even if they have
iveresov@2630:   // very low activity (see select_task()).
coleenp@4037:   inline bool is_old(Method* method);
iveresov@2630:   // Was a given method inactive for a given number of milliseconds.
iveresov@2630:   // If it is, we would remove it from the queue (see select_task()).
coleenp@4037:   inline bool is_stale(jlong t, jlong timeout, Method* m);
iveresov@2630:   // Compute the weight of the method for the compilation scheduling
coleenp@4037:   inline double weight(Method* method);
iveresov@2630:   // Apply heuristics and return true if x should be compiled before y
coleenp@4037:   inline bool compare_methods(Method* x, Method* y);
iveresov@2630:   // Compute event rate for a given method. The rate is the number of event (invocations + backedges)
iveresov@2630:   // per millisecond.
coleenp@4037:   inline void update_rate(jlong t, Method* m);
iveresov@2630:   // Compute threshold scaling coefficient
iveresov@2630:   inline double threshold_scale(CompLevel level, int feedback_k);
iveresov@2630:   // If a method is old enough and is still in the interpreter we would want to
iveresov@2630:   // start profiling without waiting for the compiled method to arrive. This function
iveresov@2630:   // determines whether we should do that.
coleenp@4037:   inline bool should_create_mdo(Method* method, CompLevel cur_level);
iveresov@2630:   // Create MDO if necessary.
iveresov@3452:   void create_mdo(methodHandle mh, JavaThread* thread);
iveresov@2630:   // Is method profiled enough?
coleenp@4037:   bool is_method_profiled(Method* method);
iveresov@2630: 
anoll@5151:   double _increase_threshold_at_ratio;
anoll@5151: 
iveresov@2630: protected:
iveresov@2630:   void print_specific(EventType type, methodHandle mh, methodHandle imh, int bci, CompLevel level);
iveresov@2630: 
anoll@5151:   void set_increase_threshold_at_ratio() { _increase_threshold_at_ratio = 100 / (100 - (double)IncreaseFirstTierCompileThresholdAt); }
iveresov@2630:   void set_start_time(jlong t) { _start_time = t;    }
iveresov@2630:   jlong start_time() const     { return _start_time; }
iveresov@2630: 
iveresov@2630:   // Submit a given method for compilation (and update the rate).
iveresov@3452:   virtual void submit_compile(methodHandle mh, int bci, CompLevel level, JavaThread* thread);
iveresov@2630:   // event() from SimpleThresholdPolicy would call these.
iveresov@2630:   virtual void method_invocation_event(methodHandle method, methodHandle inlinee,
iveresov@3452:                                        CompLevel level, nmethod* nm, JavaThread* thread);
iveresov@2630:   virtual void method_back_branch_event(methodHandle method, methodHandle inlinee,
iveresov@3452:                                         int bci, CompLevel level, nmethod* nm, JavaThread* thread);
iveresov@2630: public:
iveresov@2630:   AdvancedThresholdPolicy() : _start_time(0) { }
iveresov@2630:   // Select task is called by CompileBroker. We should return a task or NULL.
iveresov@2630:   virtual CompileTask* select_task(CompileQueue* compile_queue);
iveresov@2630:   virtual void initialize();
iveresov@2988:   virtual bool should_not_inline(ciEnv* env, ciMethod* callee);
iveresov@2988: 
iveresov@2630: };
iveresov@2630: 
iveresov@2630: #endif // TIERED
iveresov@2630: 
iveresov@2630: #endif // SHARE_VM_RUNTIME_ADVANCEDTHRESHOLDPOLICY_HPP