-
Notifications
You must be signed in to change notification settings - Fork 1.9k
/
CalculateAverage_flippingbits.java
321 lines (271 loc) · 11.9 KB
/
CalculateAverage_flippingbits.java
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
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
/*
* 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 sun.misc.Unsafe;
import java.lang.foreign.Arena;
import java.lang.reflect.Field;
import java.io.IOException;
import java.io.RandomAccessFile;
import java.nio.channels.FileChannel;
import java.nio.charset.StandardCharsets;
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 MINIMUM_FILE_SIZE_PARTITIONING = 10 * 1024 * 1024; // 10 MB
private static final int SIMD_LANE_LENGTH = ShortVector.SPECIES_MAX.length();
private static final int NUM_STATIONS = 10_000;
private static final int HASH_MAP_OFFSET_CAPACITY = 200_000;
private static final Unsafe UNSAFE = initUnsafe();
private static int HASH_PRIME_NUMBER = 31;
private static Unsafe initUnsafe() {
try {
Field theUnsafe = Unsafe.class.getDeclaredField("theUnsafe");
theUnsafe.setAccessible(true);
return (Unsafe) theUnsafe.get(Unsafe.class);
}
catch (NoSuchFieldException | IllegalAccessException e) {
throw new RuntimeException(e);
}
}
public static void main(String[] args) throws IOException {
var result = Arrays.asList(getSegments()).parallelStream()
.map(segment -> {
try {
return processSegment(segment[0], segment[1]);
}
catch (IOException e) {
throw new RuntimeException(e);
}
})
.reduce(FasterHashMap::mergeWith)
.get();
var sortedMap = new TreeMap<String, Station>();
for (Station station : result.getEntries()) {
sortedMap.put(station.getName(), station);
}
System.out.println(sortedMap);
}
private static long[][] getSegments() throws IOException {
try (var file = new RandomAccessFile(FILE, "r")) {
var channel = file.getChannel();
var fileSize = channel.size();
var startAddress = channel
.map(FileChannel.MapMode.READ_ONLY, 0, fileSize, Arena.global())
.address();
// Split file into segments, so we can work around the size limitation of channels
var numSegments = (fileSize > MINIMUM_FILE_SIZE_PARTITIONING)
? Runtime.getRuntime().availableProcessors()
: 1;
var segmentSize = fileSize / numSegments;
var boundaries = new long[numSegments][2];
var endPointer = startAddress;
for (var i = 0; i < numSegments - 1; i++) {
// Start of segment
boundaries[i][0] = endPointer;
// Extend segment until end of line or file
endPointer = endPointer + segmentSize;
while (UNSAFE.getByte(endPointer) != '\n') {
endPointer++;
}
// End of segment
boundaries[i][1] = endPointer++;
}
boundaries[numSegments - 1][0] = endPointer;
boundaries[numSegments - 1][1] = startAddress + fileSize;
return boundaries;
}
}
private static FasterHashMap processSegment(long startOfSegment, long endOfSegment) throws IOException {
var fasterHashMap = new FasterHashMap();
for (var i = startOfSegment; i < endOfSegment; i += 3) {
// Read station name
int nameHash = UNSAFE.getByte(i);
final var nameStartAddress = i++;
var character = UNSAFE.getByte(i);
while (character != ';') {
nameHash = nameHash * HASH_PRIME_NUMBER + character;
i++;
character = UNSAFE.getByte(i);
}
var nameLength = (int) (i - nameStartAddress);
i++;
// Read measurement
var isNegative = UNSAFE.getByte(i) == '-';
var measurement = 0;
if (isNegative) {
i++;
character = UNSAFE.getByte(i);
while (character != '.') {
measurement = measurement * 10 + character - '0';
i++;
character = UNSAFE.getByte(i);
}
measurement = (measurement * 10 + UNSAFE.getByte(i + 1) - '0') * -1;
}
else {
character = UNSAFE.getByte(i);
while (character != '.') {
measurement = measurement * 10 + character - '0';
i++;
character = UNSAFE.getByte(i);
}
measurement = measurement * 10 + UNSAFE.getByte(i + 1) - '0';
}
fasterHashMap.addEntry(nameHash, nameLength, nameStartAddress, (short) measurement);
}
for (Station station : fasterHashMap.getEntries()) {
station.aggregateRemainingMeasurements();
}
return fasterHashMap;
}
private static class Station {
final short[] measurements = 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 = 1;
long sum = 0;
short min = Short.MAX_VALUE;
short max = Short.MIN_VALUE;
final long nameAddress;
final int nameLength;
final int nameHash;
public Station(int nameHash, int nameLength, long nameAddress, short measurement) {
this.nameHash = nameHash;
this.nameLength = nameLength;
this.nameAddress = nameAddress;
measurements[0] = measurement;
}
public String getName() {
byte[] name = new byte[nameLength];
UNSAFE.copyMemory(null, nameAddress, name, Unsafe.ARRAY_BYTE_BASE_OFFSET, nameLength);
return new String(name, StandardCharsets.UTF_8);
}
public void addMeasurementAndComputeAggregate(short measurement) {
// Add measurement to buffer, which is later processed by SIMD instructions
measurements[count % measurements.length] = measurement;
count++;
// Once lane is full, use SIMD instructions to calculate aggregates
if (count % measurements.length == 0) {
var firstVector = ShortVector.fromArray(ShortVector.SPECIES_MAX, measurements, 0);
var secondVector = ShortVector.fromArray(ShortVector.SPECIES_MAX, measurements, 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 % measurements.length; i++) {
var measurement = measurements[i];
min = (short) Math.min(min, measurement);
max = (short) Math.max(max, measurement);
sum += measurement;
}
}
public void mergeWith(Station otherStation) {
min = (short) Math.min(min, otherStation.min);
max = (short) Math.max(max, otherStation.max);
count = count + otherStation.count;
sum = sum + otherStation.sum;
}
public boolean nameEquals(long otherNameAddress) {
var swarLimit = (nameLength / Long.BYTES) * Long.BYTES;
var i = 0;
for (; i < swarLimit; i += Long.BYTES) {
if (UNSAFE.getLong(nameAddress + i) != UNSAFE.getLong(otherNameAddress + i)) {
return false;
}
}
for (; i < nameLength; i++) {
if (UNSAFE.getByte(nameAddress + i) != UNSAFE.getByte(otherNameAddress + i)) {
return false;
}
}
return true;
}
public String toString() {
return String.format(
Locale.US,
"%.1f/%.1f/%.1f",
(min / 10.0),
((sum / 10.0) / count),
(max / 10.0));
}
}
/**
* Use two arrays for implementing the hash map:
* - The array `entries` holds the map values, in our case instances of the class Station.
* - The array `offsets` maps hashes of the keys to indexes in the `entries` array.
*
* We create `offsets` with a much larger capacity than `entries`, so we minimize collisions.
*/
private static class FasterHashMap {
// Using 16-bit integers (shorts) for offsets supports up to 2^15 (=32,767) entries
// If you need to store more entries, consider replacing short with int
short[] offsets = new short[HASH_MAP_OFFSET_CAPACITY];
Station[] entries = new Station[NUM_STATIONS + 1];
int slotsInUse = 0;
private int getOffsetIdx(int nameHash, int nameLength, long nameAddress) {
var offsetIdx = nameHash & (offsets.length - 1);
var offset = offsets[offsetIdx];
while (offset != 0 &&
(nameLength != entries[offset].nameLength || !entries[offset].nameEquals(nameAddress))) {
offsetIdx = (offsetIdx + 1) % offsets.length;
offset = offsets[offsetIdx];
}
return offsetIdx;
}
public void addEntry(int nameHash, int nameLength, long nameAddress, short measurement) {
var offsetIdx = getOffsetIdx(nameHash, nameLength, nameAddress);
var offset = offsets[offsetIdx];
if (offset == 0) {
slotsInUse++;
entries[slotsInUse] = new Station(nameHash, nameLength, nameAddress, measurement);
offsets[offsetIdx] = (short) slotsInUse;
}
else {
entries[offset].addMeasurementAndComputeAggregate(measurement);
}
}
public FasterHashMap mergeWith(FasterHashMap otherMap) {
for (Station station : otherMap.getEntries()) {
var offsetIdx = getOffsetIdx(station.nameHash, station.nameLength, station.nameAddress);
var offset = offsets[offsetIdx];
if (offset == 0) {
slotsInUse++;
entries[slotsInUse] = station;
offsets[offsetIdx] = (short) slotsInUse;
}
else {
entries[offset].mergeWith(station);
}
}
return this;
}
public List<Station> getEntries() {
return Arrays.asList(entries).subList(1, slotsInUse + 1);
}
}
}