8210857: Allow retiring TLABs and collecting statistics in parallel

Reviewed-by: sjohanss, eosterlund
This commit is contained in:
Per Lidén 2018-09-20 14:04:43 +02:00
parent 92575fc3f0
commit c0d0cbc4f1
5 changed files with 236 additions and 283 deletions

View file

@ -468,39 +468,25 @@ oop CollectedHeap::class_allocate(Klass* klass, int size, TRAPS) {
} }
void CollectedHeap::ensure_parsability(bool retire_tlabs) { void CollectedHeap::ensure_parsability(bool retire_tlabs) {
// The second disjunct in the assertion below makes a concession
// for the start-up verification done while the VM is being
// created. Callers be careful that you know that mutators
// aren't going to interfere -- for instance, this is permissible
// if we are still single-threaded and have either not yet
// started allocating (nothing much to verify) or we have
// started allocating but are now a full-fledged JavaThread
// (and have thus made our TLAB's) available for filling.
assert(SafepointSynchronize::is_at_safepoint() || !is_init_completed(), assert(SafepointSynchronize::is_at_safepoint() || !is_init_completed(),
"Should only be called at a safepoint or at start-up" "Should only be called at a safepoint or at start-up");
" otherwise concurrent mutator activity may make heap "
" unparsable again");
if (UseTLAB && retire_tlabs) { ThreadLocalAllocStats stats;
// Accumulate statistics before retiring
ThreadLocalAllocBuffer::accumulate_statistics_before_gc();
}
// The main thread starts allocating via a TLAB even before it for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next();) {
// has added itself to the threads list at vm boot-up. BarrierSet::barrier_set()->make_parsable(thread);
JavaThreadIteratorWithHandle jtiwh;
assert(jtiwh.length() > 0,
"Attempt to fill tlabs before main thread has been added"
" to threads list is doomed to failure!");
BarrierSet *bs = BarrierSet::barrier_set();
for (; JavaThread *thread = jtiwh.next(); ) {
if (UseTLAB) { if (UseTLAB) {
thread->tlab().make_parsable(retire_tlabs); if (retire_tlabs) {
thread->tlab().retire(&stats);
} else {
thread->tlab().make_parsable();
} }
bs->make_parsable(thread);
} }
} }
stats.publish();
}
void CollectedHeap::resize_all_tlabs() { void CollectedHeap::resize_all_tlabs() {
assert(SafepointSynchronize::is_at_safepoint() || !is_init_completed(), assert(SafepointSynchronize::is_at_safepoint() || !is_init_completed(),
"Should only resize tlabs at safepoint"); "Should only resize tlabs at safepoint");

View file

@ -320,7 +320,7 @@ HeapWord* MemAllocator::allocate_inside_tlab_slow(Allocation& allocation) const
// To minimize fragmentation, the last TLAB may be smaller than the rest. // To minimize fragmentation, the last TLAB may be smaller than the rest.
size_t new_tlab_size = tlab.compute_size(_word_size); size_t new_tlab_size = tlab.compute_size(_word_size);
tlab.clear_before_allocation(); tlab.retire_before_allocation();
if (new_tlab_size == 0) { if (new_tlab_size == 0) {
return NULL; return NULL;

View file

@ -32,18 +32,9 @@
#include "runtime/threadSMR.hpp" #include "runtime/threadSMR.hpp"
#include "utilities/copy.hpp" #include "utilities/copy.hpp"
// Thread-Local Edens support
// static member initialization
size_t ThreadLocalAllocBuffer::_max_size = 0; size_t ThreadLocalAllocBuffer::_max_size = 0;
int ThreadLocalAllocBuffer::_reserve_for_allocation_prefetch = 0; int ThreadLocalAllocBuffer::_reserve_for_allocation_prefetch = 0;
unsigned ThreadLocalAllocBuffer::_target_refills = 0; unsigned int ThreadLocalAllocBuffer::_target_refills = 0;
GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL;
void ThreadLocalAllocBuffer::clear_before_allocation() {
_slow_refill_waste += (unsigned)remaining();
make_parsable(true); // also retire the TLAB
}
size_t ThreadLocalAllocBuffer::remaining() { size_t ThreadLocalAllocBuffer::remaining() {
if (end() == NULL) { if (end() == NULL) {
@ -53,22 +44,7 @@ size_t ThreadLocalAllocBuffer::remaining() {
return pointer_delta(hard_end(), top()); return pointer_delta(hard_end(), top());
} }
void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() { void ThreadLocalAllocBuffer::accumulate_and_reset_statistics(ThreadLocalAllocStats* stats) {
global_stats()->initialize();
for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
thread->tlab().accumulate_statistics();
thread->tlab().initialize_statistics();
}
// Publish new stats if some allocation occurred.
if (global_stats()->allocation() != 0) {
global_stats()->publish();
global_stats()->print();
}
}
void ThreadLocalAllocBuffer::accumulate_statistics() {
Thread* thr = thread(); Thread* thr = thread();
size_t capacity = Universe::heap()->tlab_capacity(thr); size_t capacity = Universe::heap()->tlab_capacity(thr);
size_t used = Universe::heap()->tlab_used(thr); size_t used = Universe::heap()->tlab_used(thr);
@ -95,47 +71,55 @@ void ThreadLocalAllocBuffer::accumulate_statistics() {
double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used); double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
_allocation_fraction.sample(alloc_frac); _allocation_fraction.sample(alloc_frac);
} }
global_stats()->update_allocating_threads();
global_stats()->update_number_of_refills(_number_of_refills);
global_stats()->update_allocation(_allocated_size);
global_stats()->update_gc_waste(_gc_waste);
global_stats()->update_slow_refill_waste(_slow_refill_waste);
global_stats()->update_fast_refill_waste(_fast_refill_waste);
stats->update_fast_allocations(_number_of_refills,
_allocated_size,
_gc_waste,
_fast_refill_waste,
_slow_refill_waste);
} else { } else {
assert(_number_of_refills == 0 && _fast_refill_waste == 0 && assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
_slow_refill_waste == 0 && _gc_waste == 0, _slow_refill_waste == 0 && _gc_waste == 0,
"tlab stats == 0"); "tlab stats == 0");
} }
global_stats()->update_slow_allocations(_slow_allocations);
stats->update_slow_allocations(_slow_allocations);
reset_statistics();
} }
// Fills the current tlab with a dummy filler array to create void ThreadLocalAllocBuffer::insert_filler() {
// an illusion of a contiguous Eden and optionally retires the tlab. assert(end() != NULL, "Must not be retired");
// Waste accounting should be done in caller as appropriate; see, Universe::heap()->fill_with_dummy_object(top(), hard_end(), true);
// for example, clear_before_allocation(). }
void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
void ThreadLocalAllocBuffer::make_parsable() {
if (end() != NULL) { if (end() != NULL) {
invariants(); invariants();
if (ZeroTLAB) {
retire();
} else {
insert_filler();
}
}
}
if (retire) { void ThreadLocalAllocBuffer::retire(ThreadLocalAllocStats* stats) {
if (stats != NULL) {
accumulate_and_reset_statistics(stats);
}
if (end() != NULL) {
invariants();
thread()->incr_allocated_bytes(used_bytes()); thread()->incr_allocated_bytes(used_bytes());
insert_filler();
initialize(NULL, NULL, NULL);
}
} }
Universe::heap()->fill_with_dummy_object(top(), hard_end(), retire && zap); void ThreadLocalAllocBuffer::retire_before_allocation() {
_slow_refill_waste += (unsigned int)remaining();
if (retire || ZeroTLAB) { // "Reset" the TLAB retire();
set_start(NULL);
set_top(NULL);
set_pf_top(NULL);
set_end(NULL);
set_allocation_end(NULL);
}
}
assert(!(retire || ZeroTLAB) ||
(start() == NULL && end() == NULL && top() == NULL &&
_allocation_end == NULL),
"TLAB must be reset");
} }
void ThreadLocalAllocBuffer::resize() { void ThreadLocalAllocBuffer::resize() {
@ -158,7 +142,7 @@ void ThreadLocalAllocBuffer::resize() {
set_refill_waste_limit(initial_refill_waste_limit()); set_refill_waste_limit(initial_refill_waste_limit());
} }
void ThreadLocalAllocBuffer::initialize_statistics() { void ThreadLocalAllocBuffer::reset_statistics() {
_number_of_refills = 0; _number_of_refills = 0;
_fast_refill_waste = 0; _fast_refill_waste = 0;
_slow_refill_waste = 0; _slow_refill_waste = 0;
@ -199,21 +183,17 @@ void ThreadLocalAllocBuffer::initialize() {
set_desired_size(initial_desired_size()); set_desired_size(initial_desired_size());
// Following check is needed because at startup the main
// thread is initialized before the heap is. The initialization for
// this thread is redone in startup_initialization below.
if (Universe::heap() != NULL) {
size_t capacity = Universe::heap()->tlab_capacity(thread()) / HeapWordSize; size_t capacity = Universe::heap()->tlab_capacity(thread()) / HeapWordSize;
double alloc_frac = desired_size() * target_refills() / (double) capacity; double alloc_frac = desired_size() * target_refills() / (double) capacity;
_allocation_fraction.sample(alloc_frac); _allocation_fraction.sample(alloc_frac);
}
set_refill_waste_limit(initial_refill_waste_limit()); set_refill_waste_limit(initial_refill_waste_limit());
initialize_statistics(); reset_statistics();
} }
void ThreadLocalAllocBuffer::startup_initialization() { void ThreadLocalAllocBuffer::startup_initialization() {
ThreadLocalAllocStats::initialize();
// Assuming each thread's active tlab is, on average, // Assuming each thread's active tlab is, on average,
// 1/2 full at a GC // 1/2 full at a GC
@ -222,8 +202,6 @@ void ThreadLocalAllocBuffer::startup_initialization() {
// abort during VM initialization. // abort during VM initialization.
_target_refills = MAX2(_target_refills, 2U); _target_refills = MAX2(_target_refills, 2U);
_global_stats = new GlobalTLABStats();
#ifdef COMPILER2 #ifdef COMPILER2
// If the C2 compiler is present, extra space is needed at the end of // If the C2 compiler is present, extra space is needed at the end of
// TLABs, otherwise prefetching instructions generated by the C2 // TLABs, otherwise prefetching instructions generated by the C2
@ -262,9 +240,9 @@ size_t ThreadLocalAllocBuffer::initial_desired_size() {
if (TLABSize > 0) { if (TLABSize > 0) {
init_sz = TLABSize / HeapWordSize; init_sz = TLABSize / HeapWordSize;
} else if (global_stats() != NULL) { } else {
// Initial size is a function of the average number of allocating threads. // Initial size is a function of the average number of allocating threads.
unsigned nof_threads = global_stats()->allocating_threads_avg(); unsigned int nof_threads = ThreadLocalAllocStats::allocating_threads_avg();
init_sz = (Universe::heap()->tlab_capacity(thread()) / HeapWordSize) / init_sz = (Universe::heap()->tlab_capacity(thread()) / HeapWordSize) /
(nof_threads * target_refills()); (nof_threads * target_refills());
@ -338,123 +316,148 @@ HeapWord* ThreadLocalAllocBuffer::hard_end() {
return _allocation_end + alignment_reserve(); return _allocation_end + alignment_reserve();
} }
GlobalTLABStats::GlobalTLABStats() : PerfVariable* ThreadLocalAllocStats::_perf_allocating_threads;
_allocating_threads_avg(TLABAllocationWeight) { PerfVariable* ThreadLocalAllocStats::_perf_total_refills;
PerfVariable* ThreadLocalAllocStats::_perf_max_refills;
PerfVariable* ThreadLocalAllocStats::_perf_total_allocations;
PerfVariable* ThreadLocalAllocStats::_perf_total_gc_waste;
PerfVariable* ThreadLocalAllocStats::_perf_max_gc_waste;
PerfVariable* ThreadLocalAllocStats::_perf_total_slow_refill_waste;
PerfVariable* ThreadLocalAllocStats::_perf_max_slow_refill_waste;
PerfVariable* ThreadLocalAllocStats::_perf_total_fast_refill_waste;
PerfVariable* ThreadLocalAllocStats::_perf_max_fast_refill_waste;
PerfVariable* ThreadLocalAllocStats::_perf_total_slow_allocations;
PerfVariable* ThreadLocalAllocStats::_perf_max_slow_allocations;
AdaptiveWeightedAverage ThreadLocalAllocStats::_allocating_threads_avg(0);
initialize(); static PerfVariable* create_perf_variable(const char* name, PerfData::Units unit, TRAPS) {
ResourceMark rm;
return PerfDataManager::create_variable(SUN_GC, PerfDataManager::counter_name("tlab", name), unit, THREAD);
}
void ThreadLocalAllocStats::initialize() {
_allocating_threads_avg = AdaptiveWeightedAverage(TLABAllocationWeight);
_allocating_threads_avg.sample(1); // One allocating thread at startup _allocating_threads_avg.sample(1); // One allocating thread at startup
if (UsePerfData) { if (UsePerfData) {
EXCEPTION_MARK; EXCEPTION_MARK;
ResourceMark rm; _perf_allocating_threads = create_perf_variable("allocThreads", PerfData::U_None, CHECK);
_perf_total_refills = create_perf_variable("fills", PerfData::U_None, CHECK);
char* cname = PerfDataManager::counter_name("tlab", "allocThreads"); _perf_max_refills = create_perf_variable("maxFills", PerfData::U_None, CHECK);
_perf_allocating_threads = _perf_total_allocations = create_perf_variable("alloc", PerfData::U_Bytes, CHECK);
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); _perf_total_gc_waste = create_perf_variable("gcWaste", PerfData::U_Bytes, CHECK);
_perf_max_gc_waste = create_perf_variable("maxGcWaste", PerfData::U_Bytes, CHECK);
cname = PerfDataManager::counter_name("tlab", "fills"); _perf_total_slow_refill_waste = create_perf_variable("slowWaste", PerfData::U_Bytes, CHECK);
_perf_total_refills = _perf_max_slow_refill_waste = create_perf_variable("maxSlowWaste", PerfData::U_Bytes, CHECK);
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); _perf_total_fast_refill_waste = create_perf_variable("fastWaste", PerfData::U_Bytes, CHECK);
_perf_max_fast_refill_waste = create_perf_variable("maxFastWaste", PerfData::U_Bytes, CHECK);
cname = PerfDataManager::counter_name("tlab", "maxFills"); _perf_total_slow_allocations = create_perf_variable("slowAlloc", PerfData::U_None, CHECK);
_perf_max_refills = _perf_max_slow_allocations = create_perf_variable("maxSlowAlloc", PerfData::U_None, CHECK);
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
cname = PerfDataManager::counter_name("tlab", "alloc");
_perf_allocation =
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
cname = PerfDataManager::counter_name("tlab", "gcWaste");
_perf_gc_waste =
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
_perf_max_gc_waste =
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
cname = PerfDataManager::counter_name("tlab", "slowWaste");
_perf_slow_refill_waste =
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
_perf_max_slow_refill_waste =
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
cname = PerfDataManager::counter_name("tlab", "fastWaste");
_perf_fast_refill_waste =
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
_perf_max_fast_refill_waste =
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
cname = PerfDataManager::counter_name("tlab", "slowAlloc");
_perf_slow_allocations =
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
_perf_max_slow_allocations =
PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
} }
} }
void GlobalTLABStats::initialize() { ThreadLocalAllocStats::ThreadLocalAllocStats() :
// Clear counters summarizing info from all threads _allocating_threads(0),
_total_refills(0),
_max_refills(0),
_total_allocations(0),
_total_gc_waste(0),
_max_gc_waste(0),
_total_fast_refill_waste(0),
_max_fast_refill_waste(0),
_total_slow_refill_waste(0),
_max_slow_refill_waste(0),
_total_slow_allocations(0),
_max_slow_allocations(0) {}
unsigned int ThreadLocalAllocStats::allocating_threads_avg() {
return MAX2((unsigned int)(_allocating_threads_avg.average() + 0.5), 1U);
}
void ThreadLocalAllocStats::update_fast_allocations(unsigned int refills,
size_t allocations,
size_t gc_waste,
size_t fast_refill_waste,
size_t slow_refill_waste) {
_allocating_threads += 1;
_total_refills += refills;
_max_refills = MAX2(_max_refills, refills);
_total_allocations += allocations;
_total_gc_waste += gc_waste;
_max_gc_waste = MAX2(_max_gc_waste, gc_waste);
_total_fast_refill_waste += fast_refill_waste;
_max_fast_refill_waste = MAX2(_max_fast_refill_waste, fast_refill_waste);
_total_slow_refill_waste += slow_refill_waste;
_max_slow_refill_waste = MAX2(_max_slow_refill_waste, slow_refill_waste);
}
void ThreadLocalAllocStats::update_slow_allocations(unsigned int allocations) {
_total_slow_allocations += allocations;
_max_slow_allocations = MAX2(_max_slow_allocations, allocations);
}
void ThreadLocalAllocStats::update(const ThreadLocalAllocStats& other) {
_allocating_threads += other._allocating_threads;
_total_refills += other._total_refills;
_max_refills = MAX2(_max_refills, other._max_refills);
_total_allocations += other._total_allocations;
_total_gc_waste += other._total_gc_waste;
_max_gc_waste = MAX2(_max_gc_waste, other._max_gc_waste);
_total_fast_refill_waste += other._total_fast_refill_waste;
_max_fast_refill_waste = MAX2(_max_fast_refill_waste, other._max_fast_refill_waste);
_total_slow_refill_waste += other._total_slow_refill_waste;
_max_slow_refill_waste = MAX2(_max_slow_refill_waste, other._max_slow_refill_waste);
_total_slow_allocations += other._total_slow_allocations;
_max_slow_allocations = MAX2(_max_slow_allocations, other._max_slow_allocations);
}
void ThreadLocalAllocStats::reset() {
_allocating_threads = 0; _allocating_threads = 0;
_total_refills = 0; _total_refills = 0;
_max_refills = 0; _max_refills = 0;
_total_allocation = 0; _total_allocations = 0;
_total_gc_waste = 0; _total_gc_waste = 0;
_max_gc_waste = 0; _max_gc_waste = 0;
_total_slow_refill_waste = 0;
_max_slow_refill_waste = 0;
_total_fast_refill_waste = 0; _total_fast_refill_waste = 0;
_max_fast_refill_waste = 0; _max_fast_refill_waste = 0;
_total_slow_refill_waste = 0;
_max_slow_refill_waste = 0;
_total_slow_allocations = 0; _total_slow_allocations = 0;
_max_slow_allocations = 0; _max_slow_allocations = 0;
} }
void GlobalTLABStats::publish() { void ThreadLocalAllocStats::publish() {
_allocating_threads_avg.sample(_allocating_threads); if (_total_allocations == 0) {
if (UsePerfData) {
_perf_allocating_threads ->set_value(_allocating_threads);
_perf_total_refills ->set_value(_total_refills);
_perf_max_refills ->set_value(_max_refills);
_perf_allocation ->set_value(_total_allocation);
_perf_gc_waste ->set_value(_total_gc_waste);
_perf_max_gc_waste ->set_value(_max_gc_waste);
_perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
_perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
_perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
_perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
_perf_slow_allocations ->set_value(_total_slow_allocations);
_perf_max_slow_allocations ->set_value(_max_slow_allocations);
}
}
void GlobalTLABStats::print() {
Log(gc, tlab) log;
if (!log.is_debug()) {
return; return;
} }
size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste; _allocating_threads_avg.sample(_allocating_threads);
double waste_percent = percent_of(waste, _total_allocation);
log.debug("TLAB totals: thrds: %d refills: %d max: %d" const size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
const double waste_percent = percent_of(waste, _total_allocations);
log_debug(gc, tlab)("TLAB totals: thrds: %d refills: %d max: %d"
" slow allocs: %d max %d waste: %4.1f%%" " slow allocs: %d max %d waste: %4.1f%%"
" gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B" " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
" slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B" " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
" fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B", " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
_allocating_threads, _allocating_threads, _total_refills, _max_refills,
_total_refills, _max_refills, _total_slow_allocations, _max_slow_allocations, waste_percent,
_total_slow_allocations, _max_slow_allocations, _total_gc_waste * HeapWordSize, _max_gc_waste * HeapWordSize,
waste_percent, _total_slow_refill_waste * HeapWordSize, _max_slow_refill_waste * HeapWordSize,
_total_gc_waste * HeapWordSize, _total_fast_refill_waste * HeapWordSize, _max_fast_refill_waste * HeapWordSize);
_max_gc_waste * HeapWordSize,
_total_slow_refill_waste * HeapWordSize, if (UsePerfData) {
_max_slow_refill_waste * HeapWordSize, _perf_allocating_threads ->set_value(_allocating_threads);
_total_fast_refill_waste * HeapWordSize, _perf_total_refills ->set_value(_total_refills);
_max_fast_refill_waste * HeapWordSize); _perf_max_refills ->set_value(_max_refills);
_perf_total_allocations ->set_value(_total_allocations);
_perf_total_gc_waste ->set_value(_total_gc_waste);
_perf_max_gc_waste ->set_value(_max_gc_waste);
_perf_total_slow_refill_waste ->set_value(_total_slow_refill_waste);
_perf_max_slow_refill_waste ->set_value(_max_slow_refill_waste);
_perf_total_fast_refill_waste ->set_value(_total_fast_refill_waste);
_perf_max_fast_refill_waste ->set_value(_max_fast_refill_waste);
_perf_total_slow_allocations ->set_value(_total_slow_allocations);
_perf_max_slow_allocations ->set_value(_max_slow_allocations);
}
} }

View file

@ -30,7 +30,7 @@
#include "runtime/perfData.hpp" #include "runtime/perfData.hpp"
#include "runtime/vm_version.hpp" #include "runtime/vm_version.hpp"
class GlobalTLABStats; class ThreadLocalAllocStats;
// ThreadLocalAllocBuffer: a descriptor for thread-local storage used by // ThreadLocalAllocBuffer: a descriptor for thread-local storage used by
// the threads for allocation. // the threads for allocation.
@ -71,8 +71,7 @@ private:
AdaptiveWeightedAverage _allocation_fraction; // fraction of eden allocated in tlabs AdaptiveWeightedAverage _allocation_fraction; // fraction of eden allocated in tlabs
void accumulate_statistics(); void reset_statistics();
void initialize_statistics();
void set_start(HeapWord* start) { _start = start; } void set_start(HeapWord* start) { _start = start; }
void set_end(HeapWord* end) { _end = end; } void set_end(HeapWord* end) { _end = end; }
@ -96,6 +95,10 @@ private:
void initialize(HeapWord* start, HeapWord* top, HeapWord* end); void initialize(HeapWord* start, HeapWord* top, HeapWord* end);
void insert_filler();
void accumulate_and_reset_statistics(ThreadLocalAllocStats* stats);
void print_stats(const char* tag); void print_stats(const char* tag);
Thread* thread(); Thread* thread();
@ -108,9 +111,6 @@ private:
int gc_waste() const { return _gc_waste; } int gc_waste() const { return _gc_waste; }
int slow_allocations() const { return _slow_allocations; } int slow_allocations() const { return _slow_allocations; }
static GlobalTLABStats* _global_stats;
static GlobalTLABStats* global_stats() { return _global_stats; }
public: public:
ThreadLocalAllocBuffer() : _allocated_before_last_gc(0), _allocation_fraction(TLABAllocationWeight) { ThreadLocalAllocBuffer() : _allocated_before_last_gc(0), _allocation_fraction(TLABAllocationWeight) {
// do nothing. tlabs must be inited by initialize() calls // do nothing. tlabs must be inited by initialize() calls
@ -159,18 +159,18 @@ public:
// Initialization at startup // Initialization at startup
static void startup_initialization(); static void startup_initialization();
// Make an in-use tlab parsable, optionally retiring and/or zapping it. // Make an in-use tlab parsable.
void make_parsable(bool retire, bool zap = true); void make_parsable();
// Retire an in-use tlab and optionally collect statistics.
void retire(ThreadLocalAllocStats* stats = NULL);
// Retire in-use tlab before allocation of a new tlab // Retire in-use tlab before allocation of a new tlab
void clear_before_allocation(); void retire_before_allocation();
// Resize based on amount of allocation, etc. // Resize based on amount of allocation, etc.
void resize(); void resize();
// Accumulate statistics across all tlabs before gc
static void accumulate_statistics_before_gc();
void fill(HeapWord* start, HeapWord* top, size_t new_size); void fill(HeapWord* start, HeapWord* top, size_t new_size);
void initialize(); void initialize();
@ -196,88 +196,52 @@ public:
void verify(); void verify();
}; };
class GlobalTLABStats: public CHeapObj<mtThread> { class ThreadLocalAllocStats : public StackObj {
private: private:
static PerfVariable* _perf_allocating_threads;
static PerfVariable* _perf_total_refills;
static PerfVariable* _perf_max_refills;
static PerfVariable* _perf_total_allocations;
static PerfVariable* _perf_total_gc_waste;
static PerfVariable* _perf_max_gc_waste;
static PerfVariable* _perf_total_slow_refill_waste;
static PerfVariable* _perf_max_slow_refill_waste;
static PerfVariable* _perf_total_fast_refill_waste;
static PerfVariable* _perf_max_fast_refill_waste;
static PerfVariable* _perf_total_slow_allocations;
static PerfVariable* _perf_max_slow_allocations;
// Accumulate perfdata in private variables because static AdaptiveWeightedAverage _allocating_threads_avg;
// PerfData should be write-only for security reasons
// (see perfData.hpp) unsigned int _allocating_threads;
unsigned _allocating_threads; unsigned int _total_refills;
unsigned _total_refills; unsigned int _max_refills;
unsigned _max_refills; size_t _total_allocations;
size_t _total_allocation;
size_t _total_gc_waste; size_t _total_gc_waste;
size_t _max_gc_waste; size_t _max_gc_waste;
size_t _total_slow_refill_waste;
size_t _max_slow_refill_waste;
size_t _total_fast_refill_waste; size_t _total_fast_refill_waste;
size_t _max_fast_refill_waste; size_t _max_fast_refill_waste;
unsigned _total_slow_allocations; size_t _total_slow_refill_waste;
unsigned _max_slow_allocations; size_t _max_slow_refill_waste;
unsigned int _total_slow_allocations;
PerfVariable* _perf_allocating_threads; unsigned int _max_slow_allocations;
PerfVariable* _perf_total_refills;
PerfVariable* _perf_max_refills;
PerfVariable* _perf_allocation;
PerfVariable* _perf_gc_waste;
PerfVariable* _perf_max_gc_waste;
PerfVariable* _perf_slow_refill_waste;
PerfVariable* _perf_max_slow_refill_waste;
PerfVariable* _perf_fast_refill_waste;
PerfVariable* _perf_max_fast_refill_waste;
PerfVariable* _perf_slow_allocations;
PerfVariable* _perf_max_slow_allocations;
AdaptiveWeightedAverage _allocating_threads_avg;
public: public:
GlobalTLABStats(); static void initialize();
static unsigned int allocating_threads_avg();
// Initialize all counters ThreadLocalAllocStats();
void initialize();
// Write all perf counters to the perf_counters void update_fast_allocations(unsigned int refills,
size_t allocations,
size_t gc_waste,
size_t fast_refill_waste,
size_t slow_refill_waste);
void update_slow_allocations(unsigned int allocations);
void update(const ThreadLocalAllocStats& other);
void reset();
void publish(); void publish();
void print();
// Accessors
unsigned allocating_threads_avg() {
return MAX2((unsigned)(_allocating_threads_avg.average() + 0.5), 1U);
}
size_t allocation() {
return _total_allocation;
}
// Update methods
void update_allocating_threads() {
_allocating_threads++;
}
void update_number_of_refills(unsigned value) {
_total_refills += value;
_max_refills = MAX2(_max_refills, value);
}
void update_allocation(size_t value) {
_total_allocation += value;
}
void update_gc_waste(size_t value) {
_total_gc_waste += value;
_max_gc_waste = MAX2(_max_gc_waste, value);
}
void update_fast_refill_waste(size_t value) {
_total_fast_refill_waste += value;
_max_fast_refill_waste = MAX2(_max_fast_refill_waste, value);
}
void update_slow_refill_waste(size_t value) {
_total_slow_refill_waste += value;
_max_slow_refill_waste = MAX2(_max_slow_refill_waste, value);
}
void update_slow_allocations(unsigned value) {
_total_slow_allocations += value;
_max_slow_allocations = MAX2(_max_slow_allocations, value);
}
}; };
#endif // SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP #endif // SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP

View file

@ -1989,7 +1989,7 @@ void JavaThread::exit(bool destroy_vm, ExitType exit_type) {
remove_stack_guard_pages(); remove_stack_guard_pages();
if (UseTLAB) { if (UseTLAB) {
tlab().make_parsable(true); // retire TLAB tlab().retire();
} }
if (JvmtiEnv::environments_might_exist()) { if (JvmtiEnv::environments_might_exist()) {
@ -2045,7 +2045,7 @@ void JavaThread::cleanup_failed_attach_current_thread() {
remove_stack_guard_pages(); remove_stack_guard_pages();
if (UseTLAB) { if (UseTLAB) {
tlab().make_parsable(true); // retire TLAB, if any tlab().retire();
} }
BarrierSet::barrier_set()->on_thread_detach(this); BarrierSet::barrier_set()->on_thread_detach(this);