aboutsummaryrefslogtreecommitdiff
path: root/src/main/java/dev/morling/onebrc/CalculateAverage_SamuelYvon.java
blob: 45ffd0db6749aef009ad706f9874ab9c2687ec82 (plain)
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
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
/*
 *  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.ByteVector;

import java.io.IOException;
import java.io.RandomAccessFile;
import java.nio.MappedByteBuffer;
import java.nio.channels.FileChannel;
import java.nio.charset.StandardCharsets;
import java.util.*;
import java.util.stream.Collectors;

/**
 * Samuel Yvon's entry.
 * <p>
 * Explanation behind my reasoning:
 * - I want to make it as fast as possible without it being an unreadable mess; I want to avoid bit fiddling UTF-8
 * and use the provided facilities
 * - I use the fact that we know the number of stations to optimize HashMap creation (75% rule)
 * - I stole branch-less compare from royvanrijn
 * - I assume valid ASCII encoding for the number part, which allows me to parse it manually
 * (should hold for valid UTF-8)
 * - I have not done Java in forever. Especially what the heck it's become. I've looked at the other submissions and
 * the given sample to get inspiration. I did not even know about this Stream API thing.
 * </p>
 *
 * <p>
 * Future ideas:
 * - Probably can Vector-Apirize the number parsing (but it's three to four numbers, is it worth?)
 * </p>
 *
 * <p>
 * Observations:
 * - [2024-01-09] The branch-less code from royvarijn does not have a huge impact
 * </p>
 *
 * <p>
 * Changelogs:
 * 2024-01-09: Naive multi-threaded, no floats, manual line parsing
 * </p>
 */
public class CalculateAverage_SamuelYvon {

    private static final String FILE = "./measurements.txt";

    private static final int MAX_STATIONS = 10000;

    private static final byte SEMICOL = 0x3B;

    private static final byte DOT = '.';

    private static final byte MINUS = '-';

    private static final byte ZERO = '0';

    private static final String SLASH_S = "/";

    private static final byte NEWLINE = '\n';

    // The minimum line length in bytes (over-egg.)
    private static final int MIN_LINE_LENGTH_BYTES = 200;

    private static final int DJB2_INIT = 5381;

    /**
     * Branchless min (unprecise for large numbers, but good enough)
     *
     * @author royvanrijn
     */
    private static int branchlessMax(final int a, final int b) {
        final int diff = a - b;
        final int dsgn = diff >> 31;
        return a - (diff & dsgn);
    }

    /**
     * Branchless min (unprecise for large numbers, but good enough)
     *
     * @author royvanrijn
     */
    private static int branchlessMin(final int a, final int b) {
        final int diff = a - b;
        final int dsgn = diff >> 31;
        return b + (diff & dsgn);
    }

    /**
     * A magic key that contains references to the String and where it's located in memory
     */
    private static final class StationName {
        private final int hash;

        private final byte[] value;

        private StationName(int hash, MappedByteBuffer backing, int pos, int len) {
            this.hash = hash;
            this.value = new byte[len];
            backing.get(pos, this.value);
        }

        @Override
        public int hashCode() {
            return hash;
        }

        @SuppressWarnings("EqualsWhichDoesntCheckParameterClass")
        @Override
        public boolean equals(Object obj) {
            // Should NEVER be true
            // if (!(obj instanceof StationName)) {
            // return false;
            // }
            StationName other = (StationName) obj;

            if (this.value.length != other.value.length) {
                return false;
            }

            // Byte for byte compare. This actually is a bug! I'm assuming the input
            // is UTF-8 normalized, which in real life would probably not be the case.
            // TODO: SIMD?
            return Arrays.equals(this.value, other.value);
        }

        @Override
        public String toString() {
            return new String(this.value, StandardCharsets.UTF_8);
        }
    }

    private static class StationMeasureAgg {
        private int min;
        private int max;
        private long sum;
        private long count;

        private final StationName station;

        private String memoizedName;

        public StationMeasureAgg(StationName name) {
            // Actual numbers are between -99.9 and 99.9, but we *10 to avoid float
            this.station = name;
            min = 1000;
            max = -1000;
            sum = 0;
            count = 0;
        }

        @Override
        public int hashCode() {
            return this.station.hash;
        }

        /**
         * Get the city name, but also memoized it to avoid building it multiple times
         *
         * @return the city name
         */
        public String city() {
            if (null == this.memoizedName) {
                this.memoizedName = station.toString();
            }
            return this.memoizedName;
        }

        public StationMeasureAgg mergeWith(StationMeasureAgg other) {
            min = branchlessMin(min, other.min);
            max = branchlessMax(max, other.max);

            sum += other.sum;
            count += other.count;

            return this;
        }

        public void accumulate(int number) {
            min = branchlessMin(min, number);
            max = branchlessMax(max, number);

            sum += number;
            count++;
        }

        @Override
        public String toString() {
            double min = Math.round((double) this.min) / 10.0;
            double max = Math.round((double) this.max) / 10.0;
            double mean = Math.round((((double) this.sum / this.count))) / 10.0;
            return min + SLASH_S + mean + SLASH_S + max;
        }

        public StationName station() {
            return this.station;
        }
    }

    private static HashMap<StationName, StationMeasureAgg> parseChunk(MappedByteBuffer chunk) {
        HashMap<StationName, StationMeasureAgg> m = HashMap.newHashMap(MAX_STATIONS);

        int i = 0;
        while (i < chunk.limit()) {
            int j = i;

            int hash = DJB2_INIT;

            // Implement a version of djb2 while we read until the semi, we read anyways
            for (; j < chunk.limit(); ++j) {
                byte b = chunk.get(j);

                if (b == SEMICOL) {
                    break;
                }

                // How will this behave in java? Why do I get no control OF SIGNEDNESS FFS
                // Can I assume int is 32 bits? What is this language! In Java 1.7 it could be 16 bits?
                // Apparently not anymore??
                hash = (((hash << 5) + hash) + b);
            }

            StationName name = new StationName(hash, chunk, i, j - i);

            // Skip the `;`
            j++;

            // Parse the int ourselves, avoids a 'String::replace' to remove the
            // digit.
            int temp = 0;
            boolean neg = chunk.get(j) == MINUS;
            for (j = j + (neg ? 1 : 0); j < chunk.limit(); ++j) {
                temp *= 10;
                byte c = chunk.get(j);
                if (c != DOT) {
                    temp += (char) (c - ZERO);
                }
                else {
                    j++;
                    break;
                }
            }

            // The decimal point
            temp += (char) (chunk.get(j) - ZERO);

            i = j + 1;

            while (chunk.get(i++) != NEWLINE)
                ;

            if (neg)
                temp = -temp;

            m.computeIfAbsent(name, StationMeasureAgg::new).accumulate(temp);
        }

        return m;
    }

    private static int approximateChunks() {
        // https://stackoverflow.com/a/4759606
        // I don't remember Java :D
        return Runtime.getRuntime().availableProcessors();
    }

    private static List<MappedByteBuffer> getFileChunks() throws IOException {
        int approxChunkCount = approximateChunks();

        List<MappedByteBuffer> fileChunks = new ArrayList<>(approxChunkCount * 2);

        // Compute chunks offsets and lengths
        try (RandomAccessFile file = new RandomAccessFile(FILE, "r")) {
            long totalOffset = 0;
            long fLength = file.length();
            long approximateLength = Long.max(fLength / approxChunkCount, MIN_LINE_LENGTH_BYTES);

            while (totalOffset < fLength) {
                long offset = totalOffset;
                int length = (int) approximateLength;

                boolean eof = offset + length >= fLength;
                if (eof) {
                    length = (int) (fLength - offset);
                }

                MappedByteBuffer out = file.getChannel().map(FileChannel.MapMode.READ_ONLY, totalOffset, length);

                while (out.get(length - 1) != NEWLINE) {
                    length--;
                }

                out.position(0);
                out.limit(length);

                fileChunks.add(out);

                totalOffset += length;
            }
        }

        return fileChunks;
    }

    public static void main(String[] args) throws IOException {
        var fileChunks = getFileChunks();

        // Map per core, giving the non-overlapping memory slices
        final Map<StationName, StationMeasureAgg> allMeasures = fileChunks.parallelStream().map(CalculateAverage_SamuelYvon::parseChunk).flatMap(x -> x.values().stream())
                .collect(Collectors.toMap(StationMeasureAgg::station, x -> x, StationMeasureAgg::mergeWith, HashMap::new));

        // Give a capacity that should never be gone past
        StringBuilder sb = new StringBuilder(allMeasures.size() * (100 + 4 + 20));
        sb.append('{');

        allMeasures.values().stream().sorted(Comparator.comparing(StationMeasureAgg::city)).forEach(x -> {
            sb.append(x.city());
            sb.append('=');
            sb.append(x);
            sb.append(", ");
        });

        sb.delete(sb.length() - 2, sb.length()); // delete trailing comma and space

        sb.append('}');

        System.out.println(sb);
    }
}