• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2013 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #ifndef ART_RUNTIME_BASE_HISTOGRAM_INL_H_
18 #define ART_RUNTIME_BASE_HISTOGRAM_INL_H_
19 
20 #include <algorithm>
21 #include <cmath>
22 #include <limits>
23 #include <ostream>
24 
25 #include "histogram.h"
26 
27 #include "base/bit_utils.h"
28 #include "base/time_utils.h"
29 #include "utils.h"
30 
31 namespace art {
32 
AddValue(Value value)33 template <class Value> inline void Histogram<Value>::AddValue(Value value) {
34   CHECK_GE(value, static_cast<Value>(0));
35   if (value >= max_) {
36     Value new_max = ((value + 1) / bucket_width_ + 1) * bucket_width_;
37     DCHECK_GT(new_max, max_);
38     GrowBuckets(new_max);
39   }
40   BucketiseValue(value);
41 }
42 
AdjustAndAddValue(Value value)43 template <class Value> inline void Histogram<Value>::AdjustAndAddValue(Value value) {
44   AddValue(value / kAdjust);
45 }
46 
Histogram(const char * name)47 template <class Value> inline Histogram<Value>::Histogram(const char* name)
48     : kAdjust(0),
49       kInitialBucketCount(0),
50       name_(name),
51       max_buckets_(0) {
52 }
53 
54 template <class Value>
Histogram(const char * name,Value initial_bucket_width,size_t max_buckets)55 inline Histogram<Value>::Histogram(const char* name, Value initial_bucket_width,
56                                    size_t max_buckets)
57     : kAdjust(1000),
58       kInitialBucketCount(8),
59       name_(name),
60       max_buckets_(max_buckets),
61       bucket_width_(initial_bucket_width) {
62   Reset();
63 }
64 
65 template <class Value>
GrowBuckets(Value new_max)66 inline void Histogram<Value>::GrowBuckets(Value new_max) {
67   while (max_ < new_max) {
68     // If we have reached the maximum number of buckets, merge buckets together.
69     if (frequency_.size() >= max_buckets_) {
70       CHECK_ALIGNED(frequency_.size(), 2);
71       // We double the width of each bucket to reduce the number of buckets by a factor of 2.
72       bucket_width_ *= 2;
73       const size_t limit = frequency_.size() / 2;
74       // Merge the frequencies by adding each adjacent two together.
75       for (size_t i = 0; i < limit; ++i) {
76         frequency_[i] = frequency_[i * 2] + frequency_[i * 2 + 1];
77       }
78       // Remove frequencies in the second half of the array which were added to the first half.
79       while (frequency_.size() > limit) {
80         frequency_.pop_back();
81       }
82     }
83     max_ += bucket_width_;
84     frequency_.push_back(0);
85   }
86 }
87 
FindBucket(Value val)88 template <class Value> inline size_t Histogram<Value>::FindBucket(Value val) const {
89   // Since this is only a linear histogram, bucket index can be found simply with
90   // dividing the value by the bucket width.
91   DCHECK_GE(val, min_);
92   DCHECK_LE(val, max_);
93   const size_t bucket_idx = static_cast<size_t>((val - min_) / bucket_width_);
94   DCHECK_GE(bucket_idx, 0ul);
95   DCHECK_LE(bucket_idx, GetBucketCount());
96   return bucket_idx;
97 }
98 
99 template <class Value>
BucketiseValue(Value val)100 inline void Histogram<Value>::BucketiseValue(Value val) {
101   CHECK_LT(val, max_);
102   sum_ += val;
103   sum_of_squares_ += val * val;
104   ++sample_size_;
105   ++frequency_[FindBucket(val)];
106   max_value_added_ = std::max(val, max_value_added_);
107   min_value_added_ = std::min(val, min_value_added_);
108 }
109 
Initialize()110 template <class Value> inline void Histogram<Value>::Initialize() {
111   for (size_t idx = 0; idx < kInitialBucketCount; idx++) {
112     frequency_.push_back(0);
113   }
114   // Cumulative frequency and ranges has a length of 1 over frequency.
115   max_ = bucket_width_ * GetBucketCount();
116 }
117 
GetBucketCount()118 template <class Value> inline size_t Histogram<Value>::GetBucketCount() const {
119   return frequency_.size();
120 }
121 
Reset()122 template <class Value> inline void Histogram<Value>::Reset() {
123   sum_of_squares_ = 0;
124   sample_size_ = 0;
125   min_ = 0;
126   sum_ = 0;
127   min_value_added_ = std::numeric_limits<Value>::max();
128   max_value_added_ = std::numeric_limits<Value>::min();
129   frequency_.clear();
130   Initialize();
131 }
132 
GetRange(size_t bucket_idx)133 template <class Value> inline Value Histogram<Value>::GetRange(size_t bucket_idx) const {
134   DCHECK_LE(bucket_idx, GetBucketCount());
135   return min_ + bucket_idx * bucket_width_;
136 }
137 
Mean()138 template <class Value> inline double Histogram<Value>::Mean() const {
139   DCHECK_GT(sample_size_, 0ull);
140   return static_cast<double>(sum_) / static_cast<double>(sample_size_);
141 }
142 
Variance()143 template <class Value> inline double Histogram<Value>::Variance() const {
144   DCHECK_GT(sample_size_, 0ull);
145   // Using algorithms for calculating variance over a population:
146   // http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance
147   Value sum_squared = sum_ * sum_;
148   double sum_squared_by_n_squared =
149       static_cast<double>(sum_squared) /
150       static_cast<double>(sample_size_ * sample_size_);
151   double sum_of_squares_by_n =
152       static_cast<double>(sum_of_squares_) / static_cast<double>(sample_size_);
153   return sum_of_squares_by_n - sum_squared_by_n_squared;
154 }
155 
156 template <class Value>
PrintBins(std::ostream & os,const CumulativeData & data)157 inline void Histogram<Value>::PrintBins(std::ostream& os, const CumulativeData& data) const {
158   DCHECK_GT(sample_size_, 0ull);
159   for (size_t bin_idx = 0; bin_idx < data.freq_.size(); ++bin_idx) {
160     if (bin_idx > 0 && data.perc_[bin_idx] == data.perc_[bin_idx - 1]) {
161       bin_idx++;
162       continue;
163     }
164     os << GetRange(bin_idx) << ": " << data.freq_[bin_idx] << "\t"
165        << data.perc_[bin_idx] * 100.0 << "%\n";
166   }
167 }
168 
169 template <class Value>
DumpBins(std::ostream & os)170 inline void Histogram<Value>::DumpBins(std::ostream& os) const {
171   DCHECK_GT(sample_size_, 0ull);
172   bool dumped_one = false;
173   for (size_t bin_idx = 0; bin_idx < frequency_.size(); ++bin_idx) {
174     if (frequency_[bin_idx] != 0U) {
175       if (dumped_one) {
176         // Prepend a comma if not the first bin.
177         os << ",";
178       } else {
179         dumped_one = true;
180       }
181       os << GetRange(bin_idx) << ":" << frequency_[bin_idx];
182     }
183   }
184 }
185 
186 template <class Value>
PrintConfidenceIntervals(std::ostream & os,double interval,const CumulativeData & data)187 inline void Histogram<Value>::PrintConfidenceIntervals(std::ostream &os, double interval,
188                                                        const CumulativeData& data) const {
189   static constexpr size_t kFractionalDigits = 3;
190   DCHECK_GT(interval, 0);
191   DCHECK_LT(interval, 1.0);
192   const double per_0 = (1.0 - interval) / 2.0;
193   const double per_1 = per_0 + interval;
194   const TimeUnit unit = GetAppropriateTimeUnit(Mean() * kAdjust);
195   os << Name() << ":\tSum: " << PrettyDuration(Sum() * kAdjust) << " "
196      << (interval * 100) << "% C.I. " << FormatDuration(Percentile(per_0, data) * kAdjust, unit,
197                                                         kFractionalDigits)
198      << "-" << FormatDuration(Percentile(per_1, data) * kAdjust, unit, kFractionalDigits) << " "
199      << "Avg: " << FormatDuration(Mean() * kAdjust, unit, kFractionalDigits) << " Max: "
200      << FormatDuration(Max() * kAdjust, unit, kFractionalDigits) << "\n";
201 }
202 
203 template <class Value>
PrintMemoryUse(std::ostream & os)204 inline void Histogram<Value>::PrintMemoryUse(std::ostream &os) const {
205   os << Name()
206      << ": Avg: " << PrettySize(Mean()) << " Max: "
207      << PrettySize(Max()) << " Min: " << PrettySize(Min()) << "\n";
208 }
209 
210 template <class Value>
CreateHistogram(CumulativeData * out_data)211 inline void Histogram<Value>::CreateHistogram(CumulativeData* out_data) const {
212   DCHECK_GT(sample_size_, 0ull);
213   out_data->freq_.clear();
214   out_data->perc_.clear();
215   uint64_t accumulated = 0;
216   out_data->freq_.push_back(accumulated);
217   out_data->perc_.push_back(0.0);
218   for (size_t idx = 0; idx < frequency_.size(); idx++) {
219     accumulated += frequency_[idx];
220     out_data->freq_.push_back(accumulated);
221     out_data->perc_.push_back(static_cast<double>(accumulated) / static_cast<double>(sample_size_));
222   }
223   DCHECK_EQ(out_data->freq_.back(), sample_size_);
224   DCHECK_LE(std::abs(out_data->perc_.back() - 1.0), 0.001);
225 }
226 
227 #if defined(__clang__)
228 #pragma clang diagnostic push
229 #pragma clang diagnostic ignored "-Wfloat-equal"
230 #endif
231 
232 template <class Value>
Percentile(double per,const CumulativeData & data)233 inline double Histogram<Value>::Percentile(double per, const CumulativeData& data) const {
234   DCHECK_GT(data.perc_.size(), 0ull);
235   size_t upper_idx = 0, lower_idx = 0;
236   for (size_t idx = 0; idx < data.perc_.size(); idx++) {
237     if (per <= data.perc_[idx]) {
238       upper_idx = idx;
239       break;
240     }
241 
242     if (per >= data.perc_[idx] && idx != 0 && data.perc_[idx] != data.perc_[idx - 1]) {
243       lower_idx = idx;
244     }
245   }
246 
247   const double lower_perc = data.perc_[lower_idx];
248   const double lower_value = static_cast<double>(GetRange(lower_idx));
249   if (per == lower_perc) {
250     return lower_value;
251   }
252 
253   const double upper_perc = data.perc_[upper_idx];
254   const double upper_value = static_cast<double>(GetRange(upper_idx));
255   if (per == upper_perc) {
256     return upper_value;
257   }
258   DCHECK_GT(upper_perc, lower_perc);
259 
260   double value = lower_value + (upper_value - lower_value) *
261                                (per - lower_perc) / (upper_perc - lower_perc);
262 
263   if (value < min_value_added_) {
264     value = min_value_added_;
265   } else if (value > max_value_added_) {
266     value = max_value_added_;
267   }
268 
269   return value;
270 }
271 
272 #if defined(__clang__)
273 #pragma clang diagnostic pop
274 #endif
275 
276 }  // namespace art
277 #endif  // ART_RUNTIME_BASE_HISTOGRAM_INL_H_
278