-
Notifications
You must be signed in to change notification settings - Fork 24
/
sort_manager.h
930 lines (761 loc) · 36.1 KB
/
sort_manager.h
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
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
//
// File: sort_manager.h
//
// $Id: sort_manager.h,v 1.7 2006-05-20 22:19:53 adanner Exp $
//
// This file contains the class sort_manager that actually performs sorting
// given an internal sort implementation and merge heap implementation
// The merge heap classes can be found in the file mergeheap.h,
// and the internal sort classes can be found in the file internal_sort.h
#ifndef _TPIE_AMI_SORT_MANAGER_H
#define _TPIE_AMI_SORT_MANAGER_H
// Get definitions for working with Unix and Windows
#include <tpie/portability.h>
#include <tpie/stream.h>
#include <tpie/tempname.h>
#include <tpie/merge_sorted_runs.h>
#include <tpie/mergeheap.h> //For templated heaps
#include <tpie/internal_sort.h> // Contains classes for sorting internal runs
// using different comparison types
#include <cmath> //for log, ceil, etc.
#include <string>
#include <tpie/progress_indicator_base.h>
namespace tpie {
namespace ami {
#ifndef AMI_STREAM_IMP_SINGLE
#warning Including __FILE__ when AMI_STREAM_IMP_SINGLE undefined.
#endif
// A class of manager objects for merge sorting objects of type T. We
// will actually use one of two subclasses of this class which use
// either a comparison object, or the binary comparison operator <.
template <class T, class I, class M>
class sort_manager {
public:
sort_manager(I* isort, M* mheap);
~sort_manager() {
// No code in this destructor.
};
// Sort in stream to out stream an save in stream (uses 3x space)
err sort(stream<T>* in, stream<T>* out,
progress_indicator_base* indicator = NULL);
//Sort in stream and overwrite unsorted input with sorted output
//(uses 2x space)
err sort(stream<T>* in, progress_indicator_base* indicator = NULL);
private:
// *************
// * Functions *
// *************
err start_sort(); // high level wrapper to full sort
err compute_sort_params(); // compute nInputItems, mrgArity, nRuns
err partition_and_sort_runs(); // make initial sorted runs
err merge_to_output(); // loop over merge tree, create output stream
// Merge a single group mrgArity streams to an output stream
err single_merge(stream<T>**, arity_t, stream<T>*, TPIE_OS_OFFSET = -1);
// helper function for creating filename
inline void make_name(
const std::string& prepre,
const std::string& pre, int id, std::string& dest);
// **************
// * Attributes *
// **************
I* m_internalSorter; // Method for sorting runs in memory
M* m_mergeHeap; // Merge heap implementation
stream<T>* inStream;
stream<T>* outStream;
err ae; // For catching error codes
TPIE_OS_OFFSET nInputItems; // Number of items in inStream;
TPIE_OS_SIZE_T mmBytesAvail; // Amount of spare memory we can use
TPIE_OS_SIZE_T mmBytesPerStream; // Memory consumed by each Stream obj
progress_indicator_base* m_indicator; // pointer to progress indicator
TPIE_OS_OFFSET progCount; //counter for showing progress
bool use2xSpace; //flag to indicate if we are doing a 2x sort
// The maximum number of stream items of type T that we can
// sort in internal memory
TPIE_OS_SIZE_T nItemsPerRun;
TPIE_OS_OFFSET nRuns; //The number of sorted runs left to merge
arity_t mrgArity; //Max runs we can merge at one time
// The output stream to which we are currently writing runs
stream<T>* curOutputRunStream;
// The mininum number of runs in each output stream
// some streams can have one additional run
TPIE_OS_OFFSET minRunsPerStream;
// The number of extra runs or the number of streams that
// get one additional run.
arity_t nXtraRuns;
// The last run can have fewer than nItemsPerRun;
TPIE_OS_SIZE_T nItemsInLastRun;
// How many items we will sort in a given run
TPIE_OS_SIZE_T nItemsInThisRun;
// For each output stream, how many runs it should get
TPIE_OS_OFFSET runsInStream;
// A suffix to use in forming output file names. During the merge phase
// we keep two sets of files, the input files and the output files to
// which we are merging. The input file suffix is the opposite of the
// output file suffix. After merging one level, the output streams
// become the input for the next level.
std::string suffixName[2];
// A buffer for building the output file names
std::string newName;
//prefix of temp files created during sort
std::string working_disk;
private:
sort_manager(const sort_manager<T,I,M>& other);
sort_manager<T,I,M>& operator=(const sort_manager<T,I,M>& other);
};
template <class T, class I, class M>
sort_manager<T, I, M>::sort_manager(I* isort, M* mheap):
m_internalSorter(isort),
m_mergeHeap(mheap),
inStream(NULL),
outStream(NULL),
ae(NO_ERROR),
nInputItems(0),
mmBytesAvail(0),
mmBytesPerStream(0),
m_indicator(NULL),
progCount(0),
use2xSpace(false),
nItemsPerRun(0),
nRuns(0),
mrgArity(0),
curOutputRunStream(NULL),
minRunsPerStream(0),
nXtraRuns(0),
nItemsInLastRun(0),
nItemsInThisRun(0),
runsInStream(0) {
suffixName[0]="_0_";
suffixName[1]="_1_";
// Prefix of temp files created during sort
working_disk = std::string(tpie_tempnam("sort"));
};
template<class T, class I, class M>
err sort_manager<T,I,M>::sort(stream<T>* in, stream<T>* out,
progress_indicator_base* indicator){
// This version saves the original input and uses 3x space
// (input, current temp runs, output runs)
m_indicator = indicator;
inStream=in;
outStream=out;
use2xSpace=false;
// Basic checks that input is ok
if (inStream==NULL || outStream==NULL) {
return NULL_POINTER;
}
if (!inStream || !outStream) {
return OBJECT_INVALID;
}
if (inStream->stream_len() < 2) {
return SORT_ALREADY_SORTED;
}
// Else, there is something to sort, do it
return start_sort();
}
template<class T, class I, class M>
err sort_manager<T,I,M>::sort(stream<T>* in, progress_indicator_base* indicator){
//This version overwrites the original input and uses 2x space
//The input stream is truncated to length 0 after forming initial runs
//and only two levels of the merge tree are on disk at any one time.
m_indicator = indicator;
inStream=in;
outStream=in; //output destination is same as input
use2xSpace=true;
// Basic checks that input is ok
if (inStream==NULL) {
return NULL_POINTER;
}
if (!inStream) {
return OBJECT_INVALID;
}
if (inStream->stream_len() < 2) {
return SORT_ALREADY_SORTED;
}
// Else, there is something to sort, do it
return start_sort();
}
template<class T, class I, class M>
err sort_manager<T,I,M>::start_sort(){
TP_LOG_DEBUG_ID ("sort_manager::sort START");
// ********************************************************************
// * PHASE 1: See if we can sort the entire stream in internal memory *
// * without the need to use general merge sort *
// ********************************************************************
// Figure out how much memory we've got to work with.
mmBytesAvail = MM_manager.memory_available();
// Space for internal buffers for the input and output stream may not
// have been allocated yet. Query the space usage and subtract.
if ((ae = inStream->main_memory_usage
(&mmBytesPerStream,mem::STREAM_USAGE_MAXIMUM))
!= NO_ERROR) {
TP_LOG_DEBUG_ID ("Error returned from main_memory_usage");
return ae;
}
TP_LOG_DEBUG ("BTE says we use at most "<< mmBytesPerStream
<< "bytes per stream\n");
// This is how much we can use for internal sort if
// we are not doing general merge sort
mmBytesAvail -= 2 * mmBytesPerStream;
// Check if all input items can be sorted internally using less than
// mmBytesAvail
nInputItems = inStream->stream_len();
if (m_indicator) {
m_indicator->set_title("\nStarting TPIE Sort");
m_indicator->set_percentage_range(0, nInputItems, 1000);
}
inStream->seek (0);
if (nInputItems < m_internalSorter->MaxItemCount(mmBytesAvail)){
if (m_indicator) {
m_indicator->init("Sorting items internally");
}
// allocate the internal array items
m_internalSorter->allocate(static_cast<TPIE_OS_SIZE_T>(nInputItems));
// load the items into main memory, sort, and write to output.
// m_internalSorter also checks if inStream/outStream are the same and
// truncates/rewrites inStream if they are. This probably should not
// be the job of m_internalSorter-> TODO: build a cleaner interface
if ((ae = m_internalSorter->sort(inStream, outStream, static_cast<TPIE_OS_SIZE_T>(nInputItems)))
!= NO_ERROR) {
TP_LOG_FATAL_ID ("main_mem_operate failed");
return ae;
}
// de-allocate the internal array of items
m_internalSorter->deallocate();
if (m_indicator) {
m_indicator->done();
}
return NO_ERROR;
}
// ******************************************************************
// * Input stream too large for main memory, use general merge sort *
// ******************************************************************
// PHASE 2: compute nItemsPerRun, nItemsPerRun, nRuns
ae=compute_sort_params();
if (ae != NO_ERROR){
return ae;
}
// ********************************************************************
// * By this point we have checked that we have valid input, checked *
// * that we indeed need an external memory sort, verified that we *
// * have enough memory to partition and at least do a binary merge. *
// * Also checked that we have enough file descriptors to merge, *
// * and calculated the mrgArity and nItemsPerRun given memory *
// * constraints. We have also calculated nRuns for the initial *
// * number of runs we will partition into. Let's sort! *
// ********************************************************************
// ********************************************************************
// * WARNING: Since we accounted for all known memory usage in PHASE 2*
// * be very wary of memory allocation via "new" or constructors from *
// * this point on and make sure it was accounted for in PHASE 2 *
// ********************************************************************
// PHASE 3: partition and form sorted runs
TP_LOG_DEBUG_ID ("Beginning general merge sort.");
ae=partition_and_sort_runs();
if (ae != NO_ERROR){
return ae;
}
// PHASE 4: merge sorted runs to a single output stream
ae=merge_to_output();
if (ae != NO_ERROR){
return ae;
}
// That's it
if (m_indicator) {
m_indicator->done();
}
return NO_ERROR;
}
template<class T, class I, class M>
err sort_manager<T,I,M>::compute_sort_params(void){
// ********************************************************************
// * PHASE 2: Compute/check limits *
// * Compute the maximum number of items we can sort in main memory *
// * and the maximium number of sorted runs we can merge at one time *
// * Before doing any sorting, check that we can fit at least one item*
// * in internal memory for sorting and that we can merge at least two*
// * runs at at time *
// * *
// * Memory needed for the run formation phase: *
// * 2*mmBytesPerStream + {for input/output streams} *
// * nItemsPerRun*space_per_sort_item() + {for each item sorted } *
// * space_overhead_sort() {constant overhead in *
// * sort management object *
// * during sorting } *
// * *
// * Memory needed for a D-way merge: *
// * Cost per merge stream: *
// * mmBytesPerStream+ {a open stream to read from} *
// * space_per_merge_item()+ {used in internal merge heap} *
// * sizeof(T*)+sizeof(off_t) {arrays in single_merge()} *
// * sizeof(stream<T>*) {array element that points to *
// * merge stream} *
// * Fixed costs: *
// * 2*mmBytesPerStream+ {original input stream + output *
// * of current merge} *
// * space_overhead_merge()+ {fixed dynamic memory costs of *
// * merge heap} *
// * 3*space_overhead() {overhead per "new" memory request *
// * for allocating array of streams *
// * in merge_to_output and two arrays *
// * in single_merge} *
// * *
// * Total cost for D-way Merge: *
// * D*(Cost per merge stream)+(Fixed costs) *
// * *
// * Any additional memory requests that call "new" directly or *
// * indirectly should be documented and accounted for in this phase *
// ********************************************************************
TP_LOG_DEBUG_ID ("Computing merge sort parameters.");
TPIE_OS_SIZE_T mmBytesAvailSort; // Bytes available for sorting
TP_LOG_DEBUG ("Each object of size " << sizeof(T) << " uses "
<< m_internalSorter->space_per_item () << " bytes "
<< "for sorting in memory\n");
// Subtract off size of temp output stream
// The size of the input stream was already subtracted from
// mmBytesAvail
mmBytesAvailSort=mmBytesAvail - mmBytesPerStream;
nItemsPerRun=m_internalSorter->MaxItemCount(mmBytesAvailSort);
if(nItemsPerRun<1){
TP_LOG_FATAL_ID ("Insufficient Memory for forming sorted runs");
return INSUFFICIENT_MAIN_MEMORY;
}
// Now we know the max number of Items we can sort in a single
// internal memory run. Next, compute the number of runs we can
// merge together at one time
TPIE_OS_SIZE_T mmBytesPerMergeItem = mmBytesPerStream +
m_mergeHeap->space_per_item() + sizeof(T*) +
sizeof(TPIE_OS_OFFSET)+sizeof(stream<T>*);
// Fixed cost of mergheap impl. + MM_manager overhead of allocating
// an array of stream<T> ptrs (pending)
// cost of Input stream already accounted for in mmBytesAvail..
TPIE_OS_SIZE_T mmBytesFixedForMerge = m_mergeHeap->space_overhead() +
mmBytesPerStream + 3*MM_manager.space_overhead();
TPIE_OS_SIZE_T mmBytesAvailMerge = mmBytesAvail - mmBytesFixedForMerge;
// Need to support at least binary merge
if(mmBytesAvailMerge<2*mmBytesPerMergeItem){
TP_LOG_FATAL_ID ("Merge arity < 2 -- Insufficient memory for a merge.");
return INSUFFICIENT_MAIN_MEMORY;
}
// Cast down from TPIE_OS_OFFSET (type of mmBytesAvail).
// mmBytesPerMergeItem is at least 1KB, so we are OK unless we
// have more than 2 TerraBytes of memory, assuming 64 bit
// (or smaller) TPIE_OS_OFFSETS. I look forward to the day
// this comment seems silly and wrong
mrgArity = static_cast<arity_t>(mmBytesAvail-mmBytesFixedForMerge) /
mmBytesPerMergeItem;
TP_LOG_DEBUG("mem avail=" << mmBytesAvail-mmBytesFixedForMerge
<< " bytes per merge item=" << mmBytesPerMergeItem
<< " initial mrgArity=" << mrgArity << "\n");
// Make sure that the AMI is willing to provide us with the
// number of substreams we want. It may not be able to due to
// operating system restrictions, such as on the number of regions
// that can be mmap()ed in, max number of file descriptors, etc.
int availableStreams = inStream->available_streams ();
// Merging requires an available stream/file decriptor for
// each of the mrgArity input strems. We need one additional file descriptor
// for the output of the current merge, so binary merge requires
// three available streams.
if (availableStreams < 3) {
TP_LOG_FATAL_ID ("Not enough stream descriptors available " <<
"to perform merge.");
return INSUFFICIENT_AVAILABLE_STREAMS;
}
// Can at least do binary merge. See if availableStreams limits
// maximum mrgArity
// Due to the previous test, we know that available_streams >= 3.
if (mrgArity > static_cast<arity_t>(availableStreams - 1)) {
mrgArity = static_cast<arity_t>(availableStreams - 1);
TP_LOG_WARNING_ID ("Reduced merge arity due to AMI restrictions.");
}
// The number of memory-sized runs that the original input stream
// will be partitioned into.
nRuns = ((nInputItems + nItemsPerRun - 1) / nItemsPerRun);
#ifdef TPIE_SORT_SMALL_MRGARITY
// KEEP OUT!!!
// This should not be done by the typical user and is only for
// testing/debugging purposes. ONLY define this flag and set a value
// if you know what you are doing.
TP_LOG_WARNING_ID("TPIE_SORT_SMALL_MRGARITY flag is set."
" Did you mean to do this?");
if(mrgArity > TPIE_SORT_SMALL_MRGARITY) {
TP_LOG_WARNING_ID("Reducing merge arity due to compiler specified flag");
mrgArity=TPIE_SORT_SMALL_MRGARITY;
}
#endif // TPIE_SORT_SMALL_MRGARITY
#ifdef TPIE_SORT_SMALL_RUNSIZE
// KEEP OUT!!!
// This should not be done by the typical user and is only for
// testing/debugging purposes ONLY define this flag and set a value
// if you know what you are doing.
TP_LOG_WARNING_ID("TPIE_SORT_SMALL_RUNSIZE flag is set."
" Did you mean to do this?");
if(nItemsPerRun > TPIE_SORT_SMALL_RUNSIZE) {
TP_LOG_WARNING_ID("Reducing run size due to compiler specified flag");
nItemsPerRun=TPIE_SORT_SMALL_RUNSIZE;
}
// need to adjust nRuns
nRuns = ((nInputItems + nItemsPerRun - 1) / nItemsPerRun);
#endif // TPIE_SORT_SMALL_RUNSIZE
//#define MINIMIZE_INITIAL_RUN_LENGTH
#ifdef MINIMIZE_INITIAL_RUN_LENGTH
// If compiled with the above flag, try to reduce the length of
// the initial sorted runs without increasing the merge tree height
// This could be a speed-up if it is faster to quicksort many small
// runs and merge them than it is to quicksort fewer long
// runs and merge them.
TP_LOG_DEBUG_ID ("Minimizing initial run lengths without increasing" <<
" the height of the merge tree.");
// The tree height is the ceiling of the log base mrgArity of the
// number of original runs.
double tree_height = log((double)nRuns) / log((double)mrgArity);
tp_assert (tree_height > 0, "Negative or zero tree height!");
tree_height = ceil (tree_height);
// See how many runs we could possibly fit in the tree without
// increasing the height.
double maxOrigRuns = pow ((double) mrgArity, tree_height);
tp_assert (maxOrigRuns >= nRuns "Number of permitted runs was reduced!");
// How big will such runs be?
double new_nItemsPerRun = ceil (nInputItems/ maxOrigRuns);
tp_assert (new_nItemsPerRun <= nItemsPerRun,
"Size of original runs increased!");
// Update the number of items per run and the number of original runs
nItemsPerRun = (TPIE_OS_SIZE_T) new_nItemsPerRun;
TP_LOG_DEBUG_ID ("With long internal memory runs, nRuns = "
<< nRuns);
nRuns = (nInputItems + nItemsPerRun - 1) / nItemsPerRun;
TP_LOG_DEBUG_ID ("With shorter internal memory runs "
<< "and the same merge tree height, nRuns = "
<< nRuns );
tp_assert (maxOrigRuns >= nRuns,
"We increased the merge height when we weren't supposed to do so!");
#endif // MINIMIZE_INITIAL_SUBSTREAM_LENGTH
// If we have just a few runs, we don't need the
// full mrgArity. This is the last change to mrgArity
// N.B. We need to "up"-cast mrgArity here!
if(static_cast<TPIE_OS_OFFSET>(mrgArity)>nRuns){
// N.B. We know that nRuns is small, so
// it is safr to downcast.
mrgArity=static_cast<TPIE_OS_SIZE_T>(nRuns);
}
// We should always end up with at least two runs
// otherwise why are we doing it externally?
tp_assert (nRuns > 1, "Less than two runs to merge!");
// Check that numbers are consistent with input size
tp_assert (nRuns * nItemsPerRun - nInputItems < nItemsPerRun,
"Total expected output size is too large.");
tp_assert (nInputItems - (nRuns - 1) * nItemsPerRun <= nItemsPerRun,
"Total expected output size is too small.");
TP_LOG_DEBUG_ID ("Input stream has " << nInputItems << " items");
TP_LOG_DEBUG ("Max number of items per runs " << nItemsPerRun );
TP_LOG_DEBUG ("\nInitial number of runs " << nRuns );
TP_LOG_DEBUG ("\nMerge arity is " << mrgArity << "\n" );
return NO_ERROR;
}
template<class T, class I, class M>
err sort_manager<T,I,M>::partition_and_sort_runs(void){
// ********************************************************************
// * PHASE 3: Partition *
// * Partition the input stream into nRuns of at most nItemsPerRun *
// * and sort them, and write them to temporay output files. *
// * The last run may have fewer than nItemsPerRun. To keep the number*
// * of files down and to support sequential I/O, we distribute the *
// * nRuns evenly across mrgArity files, thus each file on disk holds *
// * multiple sorted runs. *
// ********************************************************************
// The mininum number of runs in each output stream
// some streams can have one additional run
minRunsPerStream = nRuns/mrgArity;
// The number of extra runs or the number of streams that
// get one additional run. This is less than mrgArity and
// it is OK to downcast to an arity_t.
nXtraRuns = static_cast<arity_t>(nRuns - minRunsPerStream*mrgArity);
tp_assert(nXtraRuns<mrgArity, "Too many extra runs");
// The last run can have fewer than nItemsPerRun;
// general case
nItemsInLastRun = static_cast<TPIE_OS_SIZE_T>(nInputItems % nItemsPerRun);
if(nItemsInLastRun==0){
// Input size is an exact multiple of nItemsPerStream
nItemsInLastRun=nItemsPerRun;
}
// Initialize memory for the internal memory runs
// accounted for in phase 2: (nItemsPerRun*size_of_sort_item) +
// space_overhead_sort
m_internalSorter->allocate(nItemsPerRun);
TP_LOG_DEBUG_ID ("Partitioning and forming sorted runs.");
// nItemsPerRun except for last run.
nItemsInThisRun=nItemsPerRun;
arity_t ii,jj; //Some index vars
// Rewind the input stream, we are about to begin
inStream->seek(0);
// ********************************************************************
// * Partition and make initial sorted runs *
// ********************************************************************
TPIE_OS_OFFSET check_size = 0; //for debugging
if (m_indicator) {
m_indicator->set_description("Forming runs ");
m_indicator->set_range(0,mrgArity,1);
m_indicator->refresh();
}
for( ii=0; ii<mrgArity; ii++){ //For each output stream
// Make the output file name
make_name(working_disk, suffixName[0], ii, newName);
// Dynamically allocate the stream
// We account for these mmBytesPerStream in phase 2 (output stream)
curOutputRunStream = new stream<T>(newName);
// How many runs should this stream get?
// extra runs go in the LAST nXtraRuns streams so that
// the one short run is always in the LAST output stream
runsInStream = minRunsPerStream + ((ii >= mrgArity-nXtraRuns)?1:0);
for( jj=0; jj < runsInStream; jj++ ) { // For each run in this stream
// See if this is the last run
if( (ii==mrgArity-1) && (jj==runsInStream-1)) {
nItemsInThisRun=nItemsInLastRun;
}
// Sort it
if ((ae = m_internalSorter->sort(inStream, curOutputRunStream,
nItemsInThisRun))!= NO_ERROR)
{
TP_LOG_FATAL_ID ("main_mem_operate failed");
return ae;
}
} // For each run in this stream
// All runs created for this stream, clean up
TP_LOG_DEBUG_ID ("Wrote " << runsInStream << " runs and "
<< curOutputRunStream->stream_len() << " items to file " << ii);
check_size+=curOutputRunStream->stream_len();
curOutputRunStream->persist(PERSIST_PERSISTENT);
delete curOutputRunStream;
if (m_indicator) {
m_indicator->step();
}
}//For each output stream
tp_assert(check_size == nInputItems, "item count mismatch");
// Done with partitioning and initial run formation
// free space associated with internal memory sorting
m_internalSorter->deallocate();
if (m_indicator) {
m_indicator->done();
}
if(use2xSpace){
//recall outStream/inStream point to same file in this case
inStream->truncate(0); //free up disk space
inStream->seek(0);
}
return NO_ERROR;
}
template<class T, class I, class M>
err sort_manager<T,I,M>::merge_to_output(void){
// ********************************************************************
// * PHASE 4: Merge *
// * Loop over all levels of the merge tree, reading mrgArity runs *
// * at a time from the streams at the current level and distributing *
// * merged runs over mrgArity output streams one level up, until *
// * a single output stream exists *
// ********************************************************************
// The input streams we from which will read sorted runs
// This Memory allocation accounted for in phase 2:
// mrgArity*sizeof(stream<T>*) + space_overhead()[fixed cost]
stream<T> **mergeInputStreams = new stream<T>*[mrgArity];
TP_LOG_DEBUG_ID("Allocated " << sizeof(stream<T>*)*mrgArity
<< " bytes for " << mrgArity << " merge input stream pointers.\n"
<< "Allocated " << MM_manager.space_overhead() << " bytes"
<< " of overhead on \"new\" call.\n"
<< "Mem. avail. is " << MM_manager.memory_available () );
// the number of iterations the main loop has gone through,
// at most the height of the merge tree log_{M/B}(N/B),
// typically 1 or 2
int mrgHeight = 0;
int treeHeight = 0; //for progress
TPIE_OS_SIZE_T ii; //index vars
TPIE_OS_OFFSET jj; //index vars
// This Memory allocation accounted for in phase 2:
// mrgArity*space_per_merge_item
m_mergeHeap->allocate( mrgArity ); //Allocate mem for mergeheap
// *****************************************************************
// * *
// * The main loop. At the outermost level we are looping over *
// * levels of the merge tree. Typically this will be very small, *
// * e.g. 1-3. The final merge pass is handled outside the loop. *
// * Future extension may want to do something special in the last *
// * merge *
// * *
// *****************************************************************
if (m_indicator) {
//compute merge depth, number of passes over data
treeHeight= static_cast<int>(ceil(log(static_cast<float>(nRuns)) /
log(static_cast<float>(mrgArity))));
}
//nRuns is initially the number of runs we formed in partition_and_sort
//phase. nXtraRuns is initially the number of outputs streams that
//contain one extra run. Runs and nXtraRuns are updated as we
//complete a merge level.
while (nRuns > mrgArity){
if (m_indicator) {
std::string description;
std::stringstream buf;
buf << "Merge pass " << mrgHeight+1 << " of " << treeHeight << " ";
buf >> description;
m_indicator->set_percentage_range(0, nInputItems);
m_indicator->init(description);
}
// We are not yet at the top of the merge tree
// Write merged runs to temporary output streams
TP_LOG_DEBUG ("Intermediate merge. level="<<mrgHeight << "\n");
// The number of output runs we will form after a mrgArity merge
nRuns = (nRuns + mrgArity - 1)/mrgArity;
// Distribute the new nRuns evenly across mrgArity (or fewer)
// output streams
minRunsPerStream = nRuns/mrgArity;
// We may have less mrgArity input runs for the last
// merged output run if the current set of merge streams has
// xtra runs
arity_t mergeRunsInLastOutputRun=(nXtraRuns>0) ? nXtraRuns : mrgArity;
// The number of extra runs or the number of output streams that
// get one additional run. This is less than mrgArity and
// it is OK to downcast to an arity_t.
nXtraRuns = static_cast<arity_t>(nRuns - minRunsPerStream*mrgArity);
tp_assert(nXtraRuns<mrgArity, "Too many extra runs");
// How many Streams we will create at the next level
arity_t nOutputStreams = (minRunsPerStream > 0) ? mrgArity : nXtraRuns;
arity_t nRunsToMerge = mrgArity; // may change for last output run
// is current merge output the last run on this merge level?
bool lastOutputRun = false;
// open the mrgArity Input streams from which to read runs
for(ii = 0; ii < mrgArity; ii++){
// Make the input file name
make_name(working_disk, suffixName[mrgHeight%2], ii, newName);
// Dynamically allocate the stream
// We account for these mmBytesPerStream in phase 2
// (input stream to read from)
mergeInputStreams[ii] = new stream<T>(newName);
mergeInputStreams[ii]->seek(0);
}
TPIE_OS_OFFSET check_size=0;
// For each new output stream, fill with merged runs.
// strange indexing is for the case that there are fewer than mrgArity
// output streams needed, and we use the LAST nOutputStreams. This
// always keeps the one possible short run in the LAST of the
// mrgArity output streams.
TP_LOG_DEBUG("Writing " << nRuns << " runs to " << nOutputStreams
<< " output files.\nEach output file has at least "
<< minRunsPerStream << " runs.\n");
for(ii = mrgArity-nOutputStreams; ii < mrgArity; ii++){
// Make the output file name
make_name(working_disk, suffixName[(mrgHeight+1)%2], ii, newName);
// Dynamically allocate the stream
// We account for these mmBytesPerStream in phase 2
// (temp merge output stream)
curOutputRunStream = new stream<T>(newName);
// How many runs should this stream get?
// extra runs go in the LAST nXtraRuns streams so that
// the one short run is always in the LAST output stream
runsInStream = minRunsPerStream + ((ii >= mrgArity-nXtraRuns)?1:0);
TP_LOG_DEBUG("Writing " << runsInStream << " runs to output "
<< " file " << ii << "\n");
for( jj=0; jj < runsInStream; jj++ ) { // For each run in this stream
// See if this is the last run.
if( (ii==mrgArity-1) && (jj==runsInStream-1)) {
lastOutputRun=true;
nRunsToMerge=mergeRunsInLastOutputRun;
}
// Merge runs to curOutputRunStream
ae = single_merge(mergeInputStreams+mrgArity-nRunsToMerge,
nRunsToMerge, curOutputRunStream,
nItemsPerRun);
if (ae != NO_ERROR) {
TP_LOG_FATAL_ID("AMI_single_merge error"<< ae <<" in deep merge");
return ae;
}
} // For each output run in this stream
// Commit new output stream to disk
TP_LOG_DEBUG("Wrote " << runsInStream << " runs and "
<< curOutputRunStream->stream_len() << " items "
<< "to file " << ii << "\n");
check_size+=curOutputRunStream->stream_len();
curOutputRunStream->persist(PERSIST_PERSISTENT);
delete curOutputRunStream;
} // For each new output stream
tp_assert(check_size==nInputItems, "item count mismatch in merge");
// All output streams created/filled.
// Clean up, go up to next level
// Delete temp input merge streams
for(ii = 0; ii < mrgArity; ii++){
mergeInputStreams[ii]->persist(PERSIST_DELETE);
delete mergeInputStreams[ii];
}
// Update run lengths
nItemsPerRun=mrgArity*nItemsPerRun; //except for maybe last run
mrgHeight++; // moving up a level
} // while (nRuns > mrgArity)
tp_assert( nRuns > 1, "Not enough runs to merge to final output");
tp_assert( nRuns <= mrgArity, "Too many runs to merge to final output");
// We are at the last merge phase, write to specified output stream
// Open up the nRuns final merge streams to merge
// These runs are packed in the LAST nRuns elements of the array
// nRuns is small, so it is safe to downcast.
TP_LOG_DEBUG_ID ("Final merge. level="<<mrgHeight);
TP_LOG_DEBUG("Merge runs left="<<nRuns<<"\n");
for(ii = mrgArity-static_cast<TPIE_OS_SIZE_T>(nRuns); ii < mrgArity; ii++){
// Make the input file name
make_name(working_disk, suffixName[mrgHeight%2], ii, newName);
/* Dynamically allocate the stream
We account for these mmBytesPerStream in phase 2
(input stream to read from)
Put LAST nRuns files in FIRST nRuns spots here
either one of mergeInputStreams loading or the call to
single_merge is a little messy. I put the mess here. (abd) */
TP_LOG_DEBUG ("Putting merge stream "<< ii << " in slot "
<< ii-(mrgArity-static_cast<TPIE_OS_SIZE_T>(nRuns)) << "\n");
mergeInputStreams[ii-(mrgArity-static_cast<TPIE_OS_SIZE_T>(nRuns))] = new stream<T>(newName);
mergeInputStreams[ii-(mrgArity-static_cast<TPIE_OS_SIZE_T>(nRuns))]->seek(0);
}
if (m_indicator) {
m_indicator->set_percentage_range(0, nInputItems);
m_indicator->init("Final merge pass ");
}
// Merge last remaining runs to the output stream.
// mergeInputStreams is address( address (the first input stream) )
// N.B. nRuns is small, so it is safe to downcast.
ae = single_merge (mergeInputStreams, static_cast<arity_t>(nRuns), outStream);
tp_assert(outStream->stream_len() == nInputItems, "item count mismatch");
if (ae != NO_ERROR) {
TP_LOG_FATAL_ID ("AMI_ERROR " << ae << " returned by single_merge "
<< "in final merge phase");
return ae;
}
TP_LOG_DEBUG("merge cleanup\n");
// We are done, except for cleanup. Is anyone still reading this?
// Delete temp input merge streams
for(ii = 0; ii < nRuns; ii++){
mergeInputStreams[ii]->persist(PERSIST_DELETE);
delete mergeInputStreams[ii];
}
// Delete stream ptr arrays
delete [] mergeInputStreams;
// Deallocate the merge heap, free up memory
m_mergeHeap->deallocate();
TP_LOG_DEBUG_ID ("Number of passes incl run formation is " <<
mrgHeight+2 );
TP_LOG_DEBUG("AMI_partition_and_merge END\n");
return NO_ERROR;
}
template<class T, class I, class M>
err sort_manager<T,I,M>::single_merge( stream < T > **inStreams,
arity_t arity, stream < T >*outStream, TPIE_OS_OFFSET cutoff)
{
return merge_sorted_runs(inStreams, arity, outStream, m_mergeHeap,
cutoff, m_indicator);
}
template<class T, class I, class M>
inline void sort_manager<T,I,M>::make_name(
const std::string& prepre, const std::string& pre, int id, std::string& dest)
{
//This buffer must be long enough to hold the
//largest possible stream id (in decimal)
//largest ID is at most mrgArity
std::stringstream buf;
buf << prepre << pre << id;
buf >> dest;
}
} // ami namespace
} // tpie namespace
#endif // _TPIE_AMI_SORT_MANAGER_H