## OpenJDK / amber / amber

### view src/hotspot/share/gc/shared/gcUtil.cpp @ 48073:5caa1d5f74c1

Find changesets by keywords (author, files, the commit message), revision
number or hash, or revset expression.

8186571: Implementation: JEP 307: Parallel Full GC for G1
Summary: Improve G1 worst-case latencies by making the full GC parallel.
Reviewed-by: tschatzl, sangheki, ehelin

author | sjohanss |
---|---|

date | Tue, 14 Nov 2017 11:33:23 +0100 |

parents | fec48bf5a827 |

children |

line wrap: on

line source

/* * Copyright (c) 2002, 2015, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 only, as * published by the Free Software Foundation. * * This code is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * version 2 for more details (a copy is included in the LICENSE file that * accompanied this code). * * You should have received a copy of the GNU General Public License version * 2 along with this work; if not, write to the Free Software Foundation, * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. * * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA * or visit www.oracle.com if you need additional information or have any * questions. * */ #include "precompiled.hpp" #include "gc/shared/gcUtil.hpp" // Catch-all file for utility classes float AdaptiveWeightedAverage::compute_adaptive_average(float new_sample, float average) { // We smooth the samples by not using weight() directly until we've // had enough data to make it meaningful. We'd like the first weight // used to be 1, the second to be 1/2, etc until we have // OLD_THRESHOLD/weight samples. unsigned count_weight = 0; // Avoid division by zero if the counter wraps (7158457) if (!is_old()) { count_weight = OLD_THRESHOLD/count(); } unsigned adaptive_weight = (MAX2(weight(), count_weight)); float new_avg = exp_avg(average, new_sample, adaptive_weight); return new_avg; } void AdaptiveWeightedAverage::sample(float new_sample) { increment_count(); // Compute the new weighted average float new_avg = compute_adaptive_average(new_sample, average()); set_average(new_avg); _last_sample = new_sample; } void AdaptiveWeightedAverage::print() const { print_on(tty); } void AdaptiveWeightedAverage::print_on(outputStream* st) const { guarantee(false, "NYI"); } void AdaptivePaddedAverage::print() const { print_on(tty); } void AdaptivePaddedAverage::print_on(outputStream* st) const { guarantee(false, "NYI"); } void AdaptivePaddedNoZeroDevAverage::print() const { print_on(tty); } void AdaptivePaddedNoZeroDevAverage::print_on(outputStream* st) const { guarantee(false, "NYI"); } void AdaptivePaddedAverage::sample(float new_sample) { // Compute new adaptive weighted average based on new sample. AdaptiveWeightedAverage::sample(new_sample); // Now update the deviation and the padded average. float new_avg = average(); float new_dev = compute_adaptive_average(fabsd(new_sample - new_avg), deviation()); set_deviation(new_dev); set_padded_average(new_avg + padding() * new_dev); _last_sample = new_sample; } void AdaptivePaddedNoZeroDevAverage::sample(float new_sample) { // Compute our parent classes sample information AdaptiveWeightedAverage::sample(new_sample); float new_avg = average(); if (new_sample != 0) { // We only create a new deviation if the sample is non-zero float new_dev = compute_adaptive_average(fabsd(new_sample - new_avg), deviation()); set_deviation(new_dev); } set_padded_average(new_avg + padding() * deviation()); _last_sample = new_sample; } LinearLeastSquareFit::LinearLeastSquareFit(unsigned weight) : _sum_x(0), _sum_x_squared(0), _sum_y(0), _sum_xy(0), _intercept(0), _slope(0), _mean_x(weight), _mean_y(weight) {} void LinearLeastSquareFit::update(double x, double y) { _sum_x = _sum_x + x; _sum_x_squared = _sum_x_squared + x * x; _sum_y = _sum_y + y; _sum_xy = _sum_xy + x * y; _mean_x.sample(x); _mean_y.sample(y); assert(_mean_x.count() == _mean_y.count(), "Incorrect count"); if ( _mean_x.count() > 1 ) { double slope_denominator; slope_denominator = (_mean_x.count() * _sum_x_squared - _sum_x * _sum_x); // Some tolerance should be injected here. A denominator that is // nearly 0 should be avoided. if (slope_denominator != 0.0) { double slope_numerator; slope_numerator = (_mean_x.count() * _sum_xy - _sum_x * _sum_y); _slope = slope_numerator / slope_denominator; // The _mean_y and _mean_x are decaying averages and can // be used to discount earlier data. If they are used, // first consider whether all the quantities should be // kept as decaying averages. // _intercept = _mean_y.average() - _slope * _mean_x.average(); _intercept = (_sum_y - _slope * _sum_x) / ((double) _mean_x.count()); } } } double LinearLeastSquareFit::y(double x) { double new_y; if ( _mean_x.count() > 1 ) { new_y = (_intercept + _slope * x); return new_y; } else { return _mean_y.average(); } } // Both decrement_will_decrease() and increment_will_decrease() return // true for a slope of 0. That is because a change is necessary before // a slope can be calculated and a 0 slope will, in general, indicate // that no calculation of the slope has yet been done. Returning true // for a slope equal to 0 reflects the intuitive expectation of the // dependence on the slope. Don't use the complement of these functions // since that intuitive expectation is not built into the complement. bool LinearLeastSquareFit::decrement_will_decrease() { return (_slope >= 0.00); } bool LinearLeastSquareFit::increment_will_decrease() { return (_slope <= 0.00); }