1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
|
/*
* Copyright 2023 The original authors
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package dev.morling.onebrc;
import jdk.incubator.vector.ShortVector;
import jdk.incubator.vector.VectorOperators;
import java.io.IOException;
import java.io.RandomAccessFile;
import java.util.*;
/**
* Approach:
* - Use memory-mapped file to speed up loading data into memory
* - Partition data, compute aggregates for partitions in parallel, and finally combine results from all partitions
* - Apply SIMD instructions for computing min/max/sum aggregates
* - Use Shorts for storing aggregates of partitions, so we maximize the SIMD parallelism
*/
public class CalculateAverage_flippingbits {
private static final String FILE = "./measurements.txt";
private static final long CHUNK_SIZE = 100 * 1024 * 1024; // 100 MB
private static final int SIMD_LANE_LENGTH = ShortVector.SPECIES_MAX.length();
public static void main(String[] args) throws IOException {
try (var file = new RandomAccessFile(FILE, "r")) {
// Calculate chunk boundaries
long[][] chunkBoundaries = getChunkBoundaries(file);
// Process chunks
var result = Arrays.asList(chunkBoundaries).stream()
.map(chunk -> {
try {
return processChunk(chunk[0], chunk[1]);
}
catch (IOException e) {
throw new RuntimeException(e);
}
})
.parallel()
.reduce((firstMap, secondMap) -> {
for (var entry : secondMap.entrySet()) {
PartitionAggregate firstAggregate = firstMap.get(entry.getKey());
if (firstAggregate == null) {
firstMap.put(entry.getKey(), entry.getValue());
}
else {
firstAggregate.mergeWith(entry.getValue());
}
}
return firstMap;
})
.map(hashMap -> new TreeMap(hashMap)).get();
System.out.println(result);
}
}
private static long[][] getChunkBoundaries(RandomAccessFile file) throws IOException {
var fileSize = file.length();
// Split file into chunks, so we can work around the size limitation of channels
var chunks = (int) (fileSize / CHUNK_SIZE);
long[][] chunkBoundaries = new long[chunks + 1][2];
var endPointer = 0L;
for (var i = 0; i <= chunks; i++) {
// Start of chunk
chunkBoundaries[i][0] = Math.min(Math.max(endPointer, i * CHUNK_SIZE), fileSize);
// Seek end of chunk, limited by the end of the file
file.seek(Math.min(chunkBoundaries[i][0] + CHUNK_SIZE - 1, fileSize));
// Extend chunk until end of line or file
while (true) {
var character = file.read();
if (character == '\n' || character == -1) {
break;
}
}
// End of chunk
endPointer = file.getFilePointer();
chunkBoundaries[i][1] = endPointer;
}
return chunkBoundaries;
}
private static Map<String, PartitionAggregate> processChunk(long startOfChunk, long endOfChunk)
throws IOException {
Map<String, PartitionAggregate> stationAggregates = new HashMap<>(10_000);
byte[] byteChunk = new byte[(int) (endOfChunk - startOfChunk)];
try (var file = new RandomAccessFile(FILE, "r")) {
file.seek(startOfChunk);
file.read(byteChunk);
var i = 0;
while (i < byteChunk.length) {
final var startPosStation = i;
// read station name
while (byteChunk[i] != ';') {
i++;
}
var station = new String(Arrays.copyOfRange(byteChunk, startPosStation, i));
i++;
// read measurement
final var startPosMeasurement = i;
while (byteChunk[i] != '\n') {
i++;
}
var measurement = Arrays.copyOfRange(byteChunk, startPosMeasurement, i);
var aggregate = stationAggregates.getOrDefault(station, new PartitionAggregate());
aggregate.addMeasurementAndComputeAggregate(measurement);
stationAggregates.put(station, aggregate);
i++;
}
stationAggregates.values().forEach(PartitionAggregate::aggregateRemainingMeasurements);
}
return stationAggregates;
}
private static class PartitionAggregate {
final short[] lane = new short[SIMD_LANE_LENGTH * 2];
// Assume that we do not have more than Integer.MAX_VALUE measurements for the same station per partition
int count = 0;
long sum = 0;
short min = Short.MAX_VALUE;
short max = Short.MIN_VALUE;
public void addMeasurementAndComputeAggregate(byte[] measurementBytes) {
// Parse measurement and exploit that we know the format of the floating-point values
var measurement = measurementBytes[measurementBytes.length - 1] - '0';
var digits = 1;
for (var i = measurementBytes.length - 3; i > 0; i--) {
var num = measurementBytes[i] - '0';
measurement = measurement + (num * (int) Math.pow(10, digits));
digits++;
}
// Check if measurement is negative
if (measurementBytes[0] == '-') {
measurement = measurement * -1;
}
else {
var num = measurementBytes[0] - '0';
measurement = measurement + (num * (int) Math.pow(10, digits));
}
// Add measurement to buffer, which is later processed by SIMD instructions
lane[count % lane.length] = (short) measurement;
count++;
// Once lane is full, use SIMD instructions to calculate aggregates
if (count % lane.length == 0) {
var firstVector = ShortVector.fromArray(ShortVector.SPECIES_MAX, lane, 0);
var secondVector = ShortVector.fromArray(ShortVector.SPECIES_MAX, lane, SIMD_LANE_LENGTH);
var simdMin = firstVector.min(secondVector).reduceLanes(VectorOperators.MIN);
min = (short) Math.min(min, simdMin);
var simdMax = firstVector.max(secondVector).reduceLanes(VectorOperators.MAX);
max = (short) Math.max(max, simdMax);
sum += firstVector.add(secondVector).reduceLanes(VectorOperators.ADD);
}
}
public void aggregateRemainingMeasurements() {
for (var i = 0; i < count % lane.length; i++) {
var measurement = lane[i];
min = (short) Math.min(min, measurement);
max = (short) Math.max(max, measurement);
sum += measurement;
}
}
public void mergeWith(PartitionAggregate otherAggregate) {
min = (short) Math.min(min, otherAggregate.min);
max = (short) Math.max(max, otherAggregate.max);
count = count + otherAggregate.count;
sum = sum + otherAggregate.sum;
}
public String toString() {
return String.format(
Locale.US,
"%.1f/%.1f/%.1f",
(min / 10.0),
(sum / 10.0) / count,
(max / 10.0));
}
}
}
|
