-
Notifications
You must be signed in to change notification settings - Fork 0
/
HTMHashBuild.hpp
464 lines (415 loc) · 17.3 KB
/
HTMHashBuild.hpp
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
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
#pragma once
#include <cstdlib>
#include <iostream>
#ifdef __APPLE__
#include <rtm.h>
#else
#include <immintrin.h>
#endif
#include <tbb/tbb.h>
#include <memory>
#include <cmath>
#include <sys/time.h>
#include <cstring>
#include "config.h"
using namespace std;
using namespace tbb;
#define CACHE_LINE_SIZE 32
#ifndef NEXT_POW_2
/**
* compute the next number, greater than or equal to 32-bit unsigned v.
* taken from "bit twiddling hacks":
* http://graphics.stanford.edu/~seander/bithacks.html
*/
#define NEXT_POW_2(V) \
do { \
V--; \
V |= V >> 1; \
V |= V >> 2; \
V |= V >> 4; \
V |= V >> 8; \
V |= V >> 16; \
V++; \
} while(0)
#endif
struct Bucket {
uint64_t tuples[3];
uint32_t count;
uint32_t nextIndex;
};
// Number of rounds of 16k tuples to process.
#if HTM_SWITCH
#define K 5
#else
#define K 0
#endif
void
HTMHashBuild(uint64_t* relR, uint32_t rSize,
#if ENABLE_PROBE
uint64_t* relS, uint32_t sSize,
#endif
uint32_t transactionSize, uint32_t scaleOutput, uint32_t numPartitions,
uint32_t probeLength) {
uint32_t numBuckets = rSize / 3 + 1;
NEXT_POW_2(numBuckets);
uint32_t inputPartitionSize = rSize / numPartitions;
Bucket* buckets;
if (posix_memalign((void**)&buckets, CACHE_LINE_SIZE,
numBuckets * sizeof(Bucket))){
perror("Aligned allocation failed!\n");
exit(EXIT_FAILURE);
}
memset(buckets, 0, numBuckets * sizeof(Bucket));
int tCount = 0;
for (int i=0; i<numBuckets; i++) {
tCount += buckets[i].count;
}
uint64_t* conflicts = new uint64_t[rSize]{};
uint32_t* conflictCounts = new uint32_t[numPartitions]{};
uint64_t* conflictRanges = new uint64_t[rSize]{};
uint32_t* conflictRangeCounts = new uint32_t[numPartitions]{};
#if ENABLE_PROBE
uint32_t* matchCounter = new uint32_t[numPartitions];
#endif // ENABLE_PROBE
#if TM_TRACK
tbb::atomic<int> b1=0,b2=0,b3=0,b4=0,b5=0,b6=0,b7=0;
#endif // TM_TRACK
struct timeval before, after, afterFirstRound;
gettimeofday(&before, NULL);
uint32_t tableMask = numBuckets - 1;
// Run a first round to determine if there is sufficient locality in the data.
// 16k * K tuples run per partition with tSize = 16.
parallel_for(blocked_range<size_t>(0, rSize, inputPartitionSize),
[buckets, tableMask, transactionSize, inputPartitionSize,
#if TM_TRACK
&b1, &b2, &b3, &b4, &b5, &b6, &b7,
#endif
relR, conflicts, conflictCounts, conflictRanges, conflictRangeCounts](const auto range) {
uint32_t localConflictCount = 0;
uint32_t localConflictRangeCount = 0;
uint32_t localPartitionId = range.begin() / inputPartitionSize;
uint32_t conflictPartitionStart = inputPartitionSize * localPartitionId;
uint32_t localFailsCount = 0;
uint32_t tSize = 16;
for(size_t k = range.begin(); k < range.begin() + K*16384; k += 16384) {
for(size_t j = k; j < k + 16384; j += tSize) {
auto status = _xbegin();
if(status == _XBEGIN_STARTED) {
for(size_t i = j; i < j + tSize; i++) {
// Note that the hash function is (key / 3) & tableMask
// This ensures multiple nearby keys hit the same hash bucket.
uint32_t slot = (relR[i]/3) & tableMask;
if (buckets[slot].count != 3) {
buckets[slot].tuples[buckets[slot].count++] = relR[i];
} else {
conflicts[conflictPartitionStart + localConflictCount++] = relR[i];
}
}
_xend();
} else {
uint64_t entry = j;
entry = (entry << 32) + tSize;
conflictRanges[conflictPartitionStart + localConflictRangeCount++] = entry;
#if TM_TRACK
if (status & _XABORT_EXPLICIT) b1 += 1;
if (status & _XABORT_RETRY) b2 += 1;
if (status & _XABORT_CONFLICT) b3 += 1;
if (status & _XABORT_CAPACITY) b4 += 1;
if (status & _XABORT_DEBUG) b5 += 1;
if (status & _XABORT_NESTED) b6 += 1;
if (! (status & 0x3f)) b7 += 1;
#endif // TRACK_CONFLICT
}
}
}
conflictCounts[localPartitionId] = localConflictCount;
conflictRangeCounts[localPartitionId] = localConflictRangeCount;
});
uint32_t totalRun = 16384 * K * numPartitions;
uint32_t firstRoundConflicts = 0;
for (int i=0; i<numPartitions; i++) firstRoundConflicts += conflictRangeCounts[i];
firstRoundConflicts *= 16;
double firstRoundFailureFraction = totalRun > 0 ? (firstRoundConflicts / (1.0 * totalRun)) : 0;
gettimeofday(&afterFirstRound, NULL);
parallel_for(blocked_range<size_t>(0, rSize, inputPartitionSize),
[buckets, tableMask, transactionSize, inputPartitionSize,
#if TM_TRACK
&b1, &b2, &b3, &b4, &b5, &b6, &b7,
#endif
relR, conflicts, conflictCounts, conflictRanges, conflictRangeCounts](const auto range) {
uint32_t localConflictCount = 0;
uint32_t localConflictRangeCount = 0;
uint32_t localPartitionId = range.begin() / inputPartitionSize;
uint32_t conflictPartitionStart = inputPartitionSize * localPartitionId;
uint32_t tSize = transactionSize;
for(size_t k = range.begin() + K*16384; k < range.end(); k += 16384) {
uint32_t total = 16384 / tSize;
uint32_t prevConflictRangeCount = localConflictRangeCount;
for(size_t j = k; j < k + 16384; j += tSize) {
auto status = _xbegin();
if(status == _XBEGIN_STARTED) {
for(size_t i = j; i < j + tSize; i++) {
// Note that the hash function is (key / 3) & tableMask
// This ensures multiple nearby keys hit the same hash bucket.
uint32_t slot = (relR[i]/3) & tableMask;
if (buckets[slot].count != 3) {
buckets[slot].tuples[buckets[slot].count++] = relR[i];
} else {
conflicts[conflictPartitionStart + localConflictCount++] = relR[i];
}
}
_xend();
} else {
uint64_t entry = j;
entry = (entry << 32) + tSize;
conflictRanges[conflictPartitionStart + localConflictRangeCount++] = entry;
#if TM_TRACK
if (status & _XABORT_EXPLICIT) b1 += 1;
if (status & _XABORT_RETRY) b2 += 1;
if (status & _XABORT_CONFLICT) b3 += 1;
if (status & _XABORT_CAPACITY) b4 += 1;
if (status & _XABORT_DEBUG) b5 += 1;
if (status & _XABORT_NESTED) b6 += 1;
if (! (status & 0x3f)) b7 += 1;
#endif // TRACK_CONFLICT
}
}
double failureFraction = (localConflictRangeCount - prevConflictRangeCount);
failureFraction /= total;
// Adapt the transaction size.
#if HTM_ADAPT
if (failureFraction < 0.004) tSize = tSize > 16 ? 32: tSize * 2;
else if (failureFraction > 0.020) tSize = tSize > 2? tSize / 2: 1;
#endif
}
conflictCounts[localPartitionId] = localConflictCount;
conflictRangeCounts[localPartitionId] = localConflictRangeCount;
});
int conflictCount = 0;
#if TM_RETRY
for (int i=0; i<numPartitions; i++) {
uint32_t conflictPartitionStart = inputPartitionSize * i;
uint32_t localConflictCount = conflictCounts[i];
for (int j=inputPartitionSize*i; j<inputPartitionSize*i + conflictRangeCounts[i]; j++) {
uint64_t entry = conflictRanges[j];
uint64_t start = entry >> 32;
uint64_t tSize = ((entry << 32) >> 32);
for(size_t k = start; k < start + tSize; k++) {
uint32_t slot = (relR[k]/3) & tableMask;
if (buckets[slot].count != 3) {
buckets[slot].tuples[buckets[slot].count++] = relR[k];
} else {
// TODO: We can insert directly these entries with chaining.
conflicts[conflictPartitionStart + localConflictCount++] = relR[i];
}
}
}
conflictCounts[i] = localConflictCount;
}
for(int i = 0; i < numPartitions; i++) {
conflictCount += conflictCounts[i];
}
// Create a list of overflow buckets.
Bucket* overflows = new Bucket[conflictCount + 1]{};
int curCounter = 1;
for (int i=0; i<numPartitions; i++) {
uint32_t conflictPartitionStart = inputPartitionSize * i;
for (int j=conflictPartitionStart; j<conflictPartitionStart + conflictCounts[i]; j++) {
uint32_t slot = (relR[i]/3) & tableMask;
if (buckets[slot].count == 3) {
int nextIndex = buckets[slot].nextIndex;
if (nextIndex == 0) {
buckets[slot].nextIndex = curCounter;
overflows[curCounter].count = 1;
overflows[curCounter].tuples[0] = conflicts[j];
curCounter += 1;
} else {
Bucket& curBucket = overflows[nextIndex];
if (curBucket.count == 3) {
curBucket = overflows[curCounter];
curBucket.nextIndex = buckets[slot].nextIndex;
buckets[slot].nextIndex = curCounter;
curCounter += 1;
curBucket.count = 1;
curBucket.tuples[0] = conflicts[j];
} else {
curBucket.tuples[curBucket.count] = conflicts[j];
curBucket.count += 1;
}
}
} else {
buckets[slot].tuples[buckets[slot].count] = conflicts[j];
buckets[slot].count++;
}
}
}
/* for (int i=0; i<numBuckets; i++) {*/
//Bucket& b = buckets[i];
//cout<<"B " << b.count << " " << b.nextIndex << " ";
//for (int j=0; j<buckets[i].count; j++) cout<< b.tuples[j]<< " ";
//cout<<endl;
/*}*/
#endif // TM_RETRY
#if ENABLE_PROBE
uint32_t sPartitionSize = sSize/numPartitions;
parallel_for(blocked_range<size_t>(0, sSize, sPartitionSize),
[relS, matchCounter, sPartitionSize, tableMask, output, buckets, tMask, tbSize, overflows](auto range) {
uint32_t pId = range.begin() / sPartitionSize;
uint32_t matches = 0;
for(size_t i = range.begin(); i< range.end(); i++) {
uint32_t curSlot = (relS[i] / 3) & tableMask;
#if BUILD_OVERFLOW_TABLE
Bucket& curBucket = buckets[slot];
while (true) {
for (int j=0; j<curBucket.count; j++)
if (curBucket.tuples[j] == relS[i]) matches++;
if (curBucket.nextIndex == 0) break;
curBucket = overflows[curBucket.nextIndex];
}
#endif
}
matchCounter[pId] = matches;
});
#endif // ENABLE_PROBE
gettimeofday(&after, NULL);
auto inputSum =
parallel_deterministic_reduce(blocked_range<size_t>(0, rSize, 1024), 0ul,
[relR](auto range, auto init) {
for(size_t i = range.begin(); i < range.end(); i++) {
init += relR[i];
}
return init;
},
[](auto a, auto b) { return a + b; });
auto sum = parallel_deterministic_reduce(blocked_range<size_t>(0, numBuckets, 1024), 0ul,
[buckets
#if TM_RETRY
,overflows
#endif
](auto range, auto init) {
for(size_t i = range.begin(); i < range.end(); i++) {
Bucket& curBucket = buckets[i];
while (true) {
for (uint32_t j = 0; j < curBucket.count; j++) {
init += curBucket.tuples[j];
}
if (curBucket.nextIndex == 0) break;
#if TM_RETRY
curBucket = overflows[curBucket.nextIndex];
#endif
}
}
return init;
},
[](auto a, auto b) { return a + b; });
#if TM_RETRY == 0
auto failedTransactionSum = parallel_deterministic_reduce(
blocked_range<size_t>(0, numPartitions, 1), 0ul,
[relR, &conflictRanges, &conflictRangeCounts, inputPartitionSize](auto range, auto init) {
for(size_t i = range.begin(); i < range.end(); i++) {
for(int j = inputPartitionSize * i; j < inputPartitionSize * i + conflictRangeCounts[i]; j++) {
uint64_t entry = conflictRanges[j];
uint64_t start = entry >> 32;
uint64_t tSize = ((entry << 32) >> 32);
for(size_t k = start; k < start + tSize; k++) {
init += relR[k];
}
}
}
return init;
},
[](auto a, auto b) { return a + b; });
#else
uint64_t failedTransactionSum = 0;
#endif
int conflictRangeCount = 0;
for(int i = 0; i < numPartitions; i++) {
conflictRangeCount += conflictRangeCounts[i];
}
auto conflictSum = parallel_deterministic_reduce(
blocked_range<size_t>(0, numPartitions, 1), 0ul,
[&conflicts, &conflictCounts, inputPartitionSize](auto range, auto init) {
for(size_t i = range.begin(); i < range.end(); i++) {
for(int j = inputPartitionSize * i; j < inputPartitionSize * i + conflictCounts[i]; j++) {
init += conflicts[j];
}
}
return init;
},
[](auto a, auto b) { return a + b; });
#if TM_RETRY
#else
for(int i = 0; i < numPartitions; i++) {
conflictCount += conflictCounts[i];
}
#endif
uint32_t failedTransactions = parallel_deterministic_reduce(
blocked_range<size_t>(0, numPartitions, 1), 0ul,
[relR, &conflictRanges, &conflictRangeCounts, inputPartitionSize](auto range, auto init) {
for(size_t i = range.begin(); i < range.end(); i++) {
for(int j = inputPartitionSize * i; j < inputPartitionSize * i + conflictRangeCounts[i]; j++) {
uint64_t entry = conflictRanges[j];
uint64_t tSize = ((entry << 32) >> 32);
init += tSize;
}
}
return init;
},
[](auto a, auto b) { return a + b; });
#if ENABLE_PROBE
int totalMatches = 0;
for (int i=0; i<numPartitions; i++) {
totalMatches += matchCounter[i];
}
#endif //ENABLE_PROBE
double failedTransactionPercentage = (failedTransactions) / (1.0 * rSize);
#if TM_RETRY
double failedPercentage = (conflictCount) / (1.0 * rSize);
#else
double failedPercentage = (failedTransactions + conflictCount) / (1.0 * rSize);
#endif
cout << "{"
<< "\"algo\": \"htm\"",
cout << ","
<< "\"rSize\": " << rSize;
cout << ", "
<< "\"transactionSize\": " << transactionSize;
cout << ", "
<< "\"probeLength\": " << probeLength;
cout << ", \"hashBuildTimeInMicroseconds\": "
<< (after.tv_sec * 1000000 + after.tv_usec) -
(before.tv_sec * 1000000 + before.tv_usec);
cout << ", \"firstRoundTime\": "
<< (afterFirstRound.tv_sec * 1000000 + afterFirstRound.tv_usec) -
(before.tv_sec * 1000000 + before.tv_usec);
cout << ", \"firstRoundFailureFraction\": "
<< firstRoundFailureFraction;
cout << ", "
<< "\"conflictCount\": " << conflictCount;
cout << ", "
<< "\"failedTransactions\": " << failedTransactions;
cout << ", "
<< "\"failedTransactionPercentage\": " << failedTransactionPercentage;
cout << ", "
<< "\"totalFailedPercentage\": " << failedPercentage;
#if ENABLE_PROBE
cout << ", "
<< "\"totalMatches\": " << totalMatches;
#endif
cout << ", "
<< "\"inputSum\": " << inputSum;
cout << ", "
<< "\"outputSum\": " << sum + failedTransactionSum + conflictSum;
cout << "}" << endl;
#if TM_TRACK
printf("Conflict Reason: %d %d %d %d %d %d %d %d\n", b1, b2, b3, b4, b5, b6, b7);
#endif //TM_TRACK
free(buckets);
delete[] conflicts;
delete[] conflictCounts;
delete[] conflictRanges;
delete[] conflictRangeCounts;
#if TM_RETRY
delete[] overflows;
#endif
}