-
Notifications
You must be signed in to change notification settings - Fork 4
/
stepwise_align.h
1816 lines (1232 loc) · 55.6 KB
/
stepwise_align.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
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
* Copyright (C) 2009-2012 Simon A. Berger
*
* This file is part of papara.
*
* papara is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* papara is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with papara. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef __stepwise_align_h
#define __stepwise_align_h
#include <iostream>
#include <ostream>
#include <iterator>
#include <cassert>
#include "ivymike/aligned_buffer.h"
#include "ivymike/fasta.h"
#include "ivymike/cycle.h"
#include "vec_unit.h"
#include "parsimony.h"
namespace {
//
// general: the freeshift version of this type of aligner differs from the
// implementation used in papara mainly by allowing free gaps also on the reference side
//
// update: the QS-side free-gaps seem to work quite well for normal papara as well, plus
// the positive/negaitve scoring scheme also works better than the original one from papara.
// Because of these factors this is bound to become the standard alignment kernel for 'papara 2.0'
//
template<typename score_t>
struct align_arrays {
ivy_mike::aligned_buffer<score_t> s;
ivy_mike::aligned_buffer<score_t> si;
};
template<typename score_t,bool global>
score_t align_pvec_score( std::vector<uint8_t> &a, std::vector<uint8_t> &a_aux, const std::vector<uint8_t> &b, score_t match_score, score_t match_cgap, score_t gap_open, score_t gap_extend, align_arrays<score_t> &arr ) {
if( arr.s.size() < a.size() ) {
arr.s.resize( a.size() );
arr.si.resize( a.size() );
}
if( global ) {
for( size_t i = 0; i < a.size(); ++i ) {
arr.s[i] = gap_open + (i+1) * gap_extend;
}
} else {
std::fill( arr.s.begin(), arr.s.end(), 0 );
}
const score_t SMALL = -32000;
std::fill( arr.si.begin(), arr.si.end(), SMALL );
score_t max_score = SMALL;
for( size_t ib = 0; ib < b.size(); ib++ ) {
int bc = b[ib];
score_t last_sl = SMALL;
score_t last_sc;// = 0.0;
score_t last_sdiag;// = 0.0;
if( global ) {
if( ib == 0 ) {
last_sdiag = 0;
} else {
last_sdiag = gap_open + ib * gap_extend;
}
last_sc = gap_open + (ib+1) * gap_extend;
} else {
last_sc = 0;
last_sdiag = 0;
}
score_t * __restrict s_iter = arr.s.base();
score_t * __restrict si_iter = arr.si.base();
score_t * __restrict s_end = s_iter + a.size();
bool lastrow = ib == (b.size() - 1);
for( size_t ia = 0; ia < a.size(); ++ia, ++s_iter, ++si_iter ) {
score_t ac = a[ia];
const bool cgap = a_aux[ia] == AUX_CGAP;
// determine match or mis-match according to parsimony bits coming from the tree.
score_t match = ( ac & bc ) != 0 ? match_score : 0;
score_t sm = last_sdiag + match;
last_sdiag = *s_iter;
score_t last_sc_OPEN;
score_t sl_score_stay;
if( cgap ) {
last_sc_OPEN = last_sc;
sl_score_stay = last_sl;
sm += match_cgap;
} else {
last_sc_OPEN = last_sc + gap_open;
sl_score_stay = last_sl + gap_extend;
}
score_t sl;
if( sl_score_stay > last_sc_OPEN ) {
sl = sl_score_stay;
} else {
sl = last_sc_OPEN;
}
last_sl = sl;
score_t su_gap_open = last_sdiag + gap_open;
score_t su_GAP_EXTEND = *si_iter + gap_extend;
score_t su;// = max( su_GAP_EXTEND, );
if( su_GAP_EXTEND > su_gap_open ) {
su = su_GAP_EXTEND;
} else {
su = su_gap_open;
}
*si_iter = su;
score_t sc = std::max( sm, std::max( su, sl ) );
last_sc = sc;
*s_iter = sc;
if( global ) {
max_score = sc;
} else {
if( s_iter == (s_end - 1) || lastrow ) {
if( sc > max_score ) {
max_score = sc;
}
}
}
}
}
return max_score;
}
#if 0
template<typename score_t>
struct align_vec_arrays {
aligned_buffer<score_t> s;
aligned_buffer<score_t> si;
};
template<typename score_t, size_t W, bool global, typename bstate_t>
void align_pvec_score_vec( aligned_buffer<score_t> &a_prof, aligned_buffer<score_t> &a_aux_prof, const std::vector<bstate_t> &b, const score_t match_score_sc, const score_t match_cgap_sc, const score_t gap_open_sc, const score_t gap_extend_sc, aligned_buffer<score_t> &out, align_vec_arrays<score_t> &arr ) {
typedef vector_unit<score_t,W> vu;
typedef typename vu::vec_t vec_t;
size_t av_size = a_prof.size();
// if( arr.s.size() < av_size ) {
arr.s.resize( av_size );
arr.si.resize( av_size );
// }
if( arr.s.size() % W != 0 ) {
throw std::runtime_error( "profile length not multiple of vec-unit width." );
}
if( global ) {
int i = 0;
for( typename aligned_buffer<score_t>::iterator it = arr.s.begin(); it != arr.s.end(); it += W, ++i ) {
std::fill( it, it + W, gap_open_sc + (i+1) * gap_extend_sc );
}
} else {
std::fill( arr.s.begin(), arr.s.end(), 0 );
}
const score_t SMALL = vu::SMALL_VALUE;
std::fill( arr.si.begin(), arr.si.end(), SMALL );
vec_t max_score = vu::set1(SMALL);
const vec_t zero = vu::setzero();
const vec_t gap_extend = vu::set1(gap_extend_sc);
const vec_t gap_open = vu::set1(gap_open_sc);
const vec_t match_cgap = vu::set1( match_cgap_sc );
const vec_t match_score = vu::set1( match_score_sc );
for( size_t ib = 0; ib < b.size(); ++ib ) {
vec_t bc = vu::set1(b[ib]);
vec_t last_sl = vu::set1(SMALL);
vec_t last_sc;// = vu::set1(0);
vec_t last_sdiag; // = vu::set1(0);
if( global ) {
if( ib == 0 ) {
last_sdiag = vu::set1(0);
} else {
last_sdiag = vu::set1(gap_open_sc + ib * gap_extend_sc);
}
last_sc = vu::set1(gap_open_sc + (ib+1) * gap_extend_sc);
} else {
last_sc = vu::set1(0);;
last_sdiag = vu::set1(0);;
}
score_t * __restrict a_prof_iter = a_prof.base();
score_t * __restrict a_aux_prof_iter = a_aux_prof.base();
score_t * __restrict s_iter = arr.s.base();
score_t * __restrict si_iter = arr.si.base();
score_t * __restrict s_end = s_iter + av_size;
bool lastrow = ib == (b.size() - 1);
vec_t row_max_score = vu::set1(SMALL);
for(; s_iter != s_end; a_prof_iter += W, a_aux_prof_iter += W, s_iter += W, si_iter += W ) {
//break_aloop = s_iter == (s_end - W);
const vec_t ac = vu::load( a_prof_iter );
const vec_t cgap = vu::load( a_aux_prof_iter );
const vec_t non_match = vu::cmp_eq( vu::bit_and( ac, bc ), zero );
// match increase: sum of match and match_cgap score/penalty
const vec_t sm_inc = vu::add( vu::bit_andnot(non_match, match_score), vu::bit_and(cgap, match_cgap) );
// match score (last_sdiag is preloaded in the previous iteration)
const vec_t sm = vu::add(last_sdiag, sm_inc );
// last_sdiag now is preloaded with the 'above' value (it will become the 'diag'
// element in the next iteration.)
last_sdiag = vu::load( s_iter );
// open score for gap-from-left (last_sc is the previous main-cell score)
const vec_t sl_open = vu::add( last_sc, vu::bit_andnot( cgap, gap_open) );
// extension score for gap-from-left (last_sl is the previous gap-from-left score)
const vec_t sl_extend = vu::add( last_sl, vu::bit_andnot( cgap, gap_extend) );
const vec_t sl = vu::max( sl_open, sl_extend );
last_sl = sl;
// open score for gap from above (remember, last_sdiag is preloaded with the 'above' value)
const vec_t su_open = vu::add( last_sdiag, gap_open);
// extension score for gap from above (si_iter is the previous gap-from-above score)
const vec_t su_GAP_EXTEND = vu::add( vu::load( si_iter ), gap_extend );
// chose between open/extension for gap-from-above
const vec_t su = vu::max( su_GAP_EXTEND, su_open );// = max( su_GAP_EXTEND, );
vu::store( su, si_iter );
// overall max = new main-cell score
const vec_t sc = vu::max( sm, vu::max( su, sl ) );
last_sc = sc;
vu::store( sc, s_iter );
if( !global ) {
row_max_score = vu::max( row_max_score, sc );
}
}
if( global ) {
max_score = last_sc;
} else {
max_score = vu::max( max_score, last_sc );
if( lastrow ) {
max_score = vu::max( max_score, row_max_score );
}
}
}
vu::store( max_score, out.base() );
}
#else
template<typename score_t>
struct align_vec_arrays {
ivy_mike::aligned_buffer<score_t> s;
ivy_mike::aligned_buffer<score_t> si;
};
template<typename vec_t>
struct ali_score_block_t {
vec_t last_sdiag;
vec_t last_sl;
vec_t last_sc;
};
template<typename score_t, size_t W, typename aiter, typename biter, typename oiter>
inline void align_pvec_score_vec( aiter a_start, aiter a_end, aiter a_aux_start, biter b_start, biter b_end, const score_t match_score_sc, const score_t match_cgap_sc, const score_t gap_open_sc, const score_t gap_extend_sc, oiter out_start, align_vec_arrays<score_t> &arr ) {
typedef vector_unit<score_t,W> vu;
typedef typename vu::vec_t vec_t;
{
// some basic sanity checks for the input arguments
aiter xxx;
assert( sizeof(*xxx) == sizeof(score_t));
vu::assert_alignment( &(*a_start) );
vu::assert_alignment( &(*a_aux_start) );
vu::assert_alignment( &(*out_start) );
}
const size_t av_size = std::distance( a_start, a_end );
const size_t bsize = std::distance( b_start, b_end );
const size_t block_width = 512;
assert( av_size >= block_width * W ); // the code below should handle this case, but is untested
const size_t av_minsize = std::max(av_size, block_width * W);
arr.s.resize( av_minsize );
arr.si.resize( av_minsize );
if( arr.s.size() % W != 0 ) {
throw std::runtime_error( "profile length not multiple of vec-unit width." );
}
const score_t SMALL = vu::SMALL_VALUE;
std::fill( arr.s.begin(), arr.s.end(), 0 );
std::fill( arr.si.begin(), arr.si.end(), SMALL );
vec_t max_score = vu::set1(SMALL);
const vec_t zero = vu::setzero();
const vec_t gap_extend = vu::set1(gap_extend_sc);
const vec_t gap_open = vu::set1(gap_open_sc);
const vec_t match_cgap = vu::set1( match_cgap_sc );
const vec_t match_score = vu::set1( match_score_sc );
bool done = false;
// ali_score_block_t<vec_t> btemp;
//
// btemp.last_sl = vu::set1(SMALL);
// btemp.last_sc = vu::set1(0);;
// btemp.last_sdiag = vu::set1(0);;
//
// std::vector<ali_score_block_t<vec_t> > blocks( bsize, btemp ); // TODO: maybe put this into the persistent state, if sbrk mucks up again.
typedef ivy_mike::aligned_buffer<score_t,4096> block_vec;
block_vec block_sdiag(bsize * W, 0);
block_vec block_sl(bsize * W, SMALL);
block_vec block_sc(bsize * W, 0);
struct ali_ptr_block2_t {
aiter a_prof_iter;
aiter a_aux_prof_iter;
//score_t * __restrict s_iter;
//score_t * __restrict si_iter;
};
ali_ptr_block2_t ptr_block_outer;
ptr_block_outer.a_prof_iter = a_start;
ptr_block_outer.a_aux_prof_iter = a_aux_start;
//ptr_block_outer.s_iter = arr.s.base();
//ptr_block_outer.si_iter = arr.si.base();
while( !done ) {
ali_ptr_block2_t ptr_block = ptr_block_outer;
std::fill( arr.s.begin(), arr.s.begin() + W * block_width, 0 );
std::fill( arr.si.begin(), arr.si.begin() + W * block_width, SMALL );
// typename std::vector<ali_score_block_t<vec_t> >::iterator it_block = blocks.begin();
typename block_vec::iterator block_sl_it = block_sl.begin();
typename block_vec::iterator block_sc_it = block_sc.begin();
typename block_vec::iterator block_sdiag_it = block_sdiag.begin();
biter it_b = b_start;
for( ; it_b != b_end; ++it_b, block_sl_it += W, block_sc_it += W, block_sdiag_it += W ) {
vec_t bc = vu::set1(*it_b);
bool lastrow = it_b == (b_end - 1);
vec_t row_max_score = vu::set1(SMALL);
//ali_score_block_t<vec_t> block = *it_block;
ptr_block = ptr_block_outer;
aiter a_end_this = ptr_block.a_prof_iter + W * block_width;
if( a_end_this > a_end ) {
a_end_this = a_end;
}
score_t * __restrict s_iter = arr.s.base();
score_t * __restrict si_iter = arr.si.base();
_mm_prefetch( &(*ptr_block.a_prof_iter), _MM_HINT_T0 );
_mm_prefetch( &(*ptr_block.a_aux_prof_iter), _MM_HINT_T0 );
_mm_prefetch( s_iter, _MM_HINT_T0 );
_mm_prefetch( si_iter, _MM_HINT_T0 );
vec_t last_sdiag = vu::load( &(*block_sdiag_it));
vec_t last_sl = vu::load( &(*block_sl_it));
vec_t last_sc = vu::load( &(*block_sc_it));
for(; ptr_block.a_prof_iter != a_end_this; ptr_block.a_prof_iter += W, ptr_block.a_aux_prof_iter += W, s_iter += W, si_iter += W ) {
// some 'lessions learned' about instruction ordering when using sse intrinsics:
// 1. assigning values that are read/written only once to (const) variables is ok, to improve
// readability, if the assignment is near the use. (rule: don't force the compiler to waste
// registers, by putting unnecessary stuff in between. why? the compiler should be able to reorder everything...)
// 2. arithmetic operations should be placed relative to load/store instructions, in a way
// which gives the compiler the most freedom for rearrangements (as long as rule 1 is not violated)
// 3. the parameter order of vu::max seems critical (why?)
//
// The code is written in a kind of static single assignment form to make manual analysis easier:
// The basic optimization rule is: place operations so that the number of 'active' values that 'cross'
// the operation is minimal. A value is active between its initialization and it's last read-access.
const vec_t ac = vu::load( &(*ptr_block.a_prof_iter) );
// there is still large optimization potential in pre-generating match profiles based on bc
// there may be combinatorial problem with allowing multi state characters in the QS (allowing it only for
// DNA but not AA will be a mid sized p.i.t.a.)
const vec_t non_match = vu::cmp_eq( vu::bit_and( ac, bc ), zero );
const vec_t cgap = vu::load( &(*ptr_block.a_aux_prof_iter) );
// match increase: sum of match and match_cgap score/penalty
const vec_t sm_inc = vu::add( vu::bit_andnot(non_match, match_score), vu::bit_and(cgap, match_cgap) );
// match score (last_sdiag is preloaded in the previous iteration)
const vec_t s_diag = last_sdiag;
const vec_t sm = vu::add(s_diag, sm_inc );
// open score for gap-from-left (last_sc is the previous main-cell score)
const vec_t sc_left = last_sc;
const vec_t sl_open = vu::add( sc_left, vu::bit_andnot( cgap, gap_open) );
// extension score for gap-from-left (last_sl is the previous gap-from-left score)
const vec_t sl_left = last_sl;
const vec_t sl_extend = vu::add( sl_left, vu::bit_andnot( cgap, gap_extend) );
const vec_t sl = vu::max( sl_open, sl_extend );
last_sl = sl;
const vec_t sc_above = vu::load( s_iter ); // OPT: it seems to be critical to put this load as late as possible
// last_sdiag now is preloaded with the 'above' value (it will become the 'diag'
// element in the next iteration.)
last_sdiag = sc_above;
// open score for gap from above (remember, last_sdiag is preloaded with the 'above' value)
const vec_t su_open = vu::add( sc_above, gap_open);
// extension score for gap from above (si_iter is the previous gap-from-above score)
const vec_t su_extend = vu::add( vu::load( si_iter ), gap_extend );
// chose between open/extension for gap-from-above
const vec_t su = vu::max( su_open, su_extend );// = max( su_GAP_EXTEND, );
vu::store( su, si_iter );
// overall max = new main-cell score
const vec_t sc = vu::max( sm, vu::max( su, sl ) );
last_sc = sc;
row_max_score = vu::max( row_max_score, sc );
vu::store( last_sc, s_iter );
}
done = ptr_block.a_prof_iter == a_end;
if( done ) {
max_score = vu::max( max_score, last_sc );
}
if( lastrow ) {
max_score = vu::max( max_score, row_max_score );
}
//*it_block = block;
vu::store( last_sdiag, &(*block_sdiag_it) );
vu::store( last_sc, &(*block_sc_it) );
vu::store( last_sl, &(*block_sl_it) );
}
ptr_block_outer = ptr_block;
}
vu::store( max_score, &(*out_start) );
}
template<typename score_t, size_t W, bool global, typename bstate_t>
inline void align_pvec_score_vec( ivy_mike::aligned_buffer<score_t> &a_prof, ivy_mike::aligned_buffer<score_t> &a_aux_prof, const std::vector<bstate_t> &b, const score_t match_score_sc, const score_t match_cgap_sc, const score_t gap_open_sc, const score_t gap_extend_sc, ivy_mike::aligned_buffer<score_t> &out, align_vec_arrays<score_t> &arr ) {
assert( !global );
align_pvec_score_vec<score_t, W>( a_prof.begin(), a_prof.end(), a_aux_prof.begin(), b.begin(), b.end(), match_score_sc, match_cgap_sc, gap_open_sc, gap_extend_sc, out.begin(), arr );
}
#endif
//
// rewrite of the vectorized stepwise-style aligner as a class to encasulate profile pre-generation
//
template<typename score_t, size_t W>
class pvec_aligner_vec {
public:
typedef vector_unit<score_t,W> vu;
typedef typename vu::vec_t vec_t;
template<typename mapf>
pvec_aligner_vec( const int *seqptrs[W], const unsigned int *auxptrs[W], size_t reflen, const score_t match_score_sc, const score_t match_cgap_sc, const score_t gap_open_sc, const score_t gap_extend_sc, mapf map, size_t nstates )
: pvec_prof_( W * reflen ),
aux_prof_( W * reflen ),
sm_inc_prof_( W * reflen * nstates ),
num_cstates_(nstates),
ticks_all_(0),
inner_iters_all_(0)
// gap_open_cgap_prof_( W * reflen ),
// gap_extend_cgap_prof_( W * reflen )
{
typename ivy_mike::aligned_buffer<score_t>::iterator it = pvec_prof_.begin();
typename ivy_mike::aligned_buffer<score_t>::iterator ait = aux_prof_.begin();
// typename aligned_buffer<score_t>::iterator oit = gap_open_cgap_prof_.begin();
// typename aligned_buffer<score_t>::iterator eit = gap_extend_cgap_prof_.begin();
for( size_t i = 0; i < reflen; ++i ) {
for( size_t j = 0; j < W; ++j ) {
*it = score_t(seqptrs[j][i]);
*ait = (auxptrs[j][i] == AUX_CGAP) ? score_t(-1) : 0;
// *oit = (auxptrs[j][i] == AUX_CGAP) ? 0 : gap_open_sc;
// *eit = (auxptrs[j][i] == AUX_CGAP) ? 0 : gap_extend_sc;
++it;
++ait;
// ++oit;
// ++eit;
}
}
assert( it == pvec_prof_.end() );
assert( ait == aux_prof_.end() );
assert( match_cgap_sc + match_score_sc < 0 );
// assert( W == 8 );
for( size_t i = 0; i < nstates; i++ ) {
typename ivy_mike::aligned_buffer<score_t>::iterator it = sm_inc_prof_.begin() + i * reflen * W;
typename ivy_mike::aligned_buffer<score_t>::iterator pit = pvec_prof_.begin();
typename ivy_mike::aligned_buffer<score_t>::iterator ait = aux_prof_.begin();
score_t bc = map(i);
for( size_t j = 0; j < reflen * W; ++j, ++it, ++pit, ++ait ) {
bool match = (bc & *pit) != 0;
if( match ) {
*it = match_score_sc;
} else {
*it = 0;
}
if( *ait == score_t(-1) ) {
*it += match_cgap_sc;
}
}
}
}
template<typename biter, typename oiter>
inline void align( biter b_start, biter b_end, const score_t match_score_sc, const score_t match_cgap_sc, const score_t gap_open_sc, const score_t gap_extend_sc, oiter out_start, size_t a_start_idx = -1, size_t a_end_idx = -1 ) {
typedef typename ivy_mike::aligned_buffer<score_t>::iterator aiter;
// aiter a_start, a_end, a_aux_start;
//
// if( a_start_idx == size_t(-1) || a_end_idx == size_t(-1) ) {
// assert( a_start_idx == a_end_idx );
//
// a_start = pvec_prof_.begin();
// a_end = pvec_prof_.end();
// a_aux_start = aux_prof_.begin();
// } else {
// assert( a_start_idx != a_end_idx );
// a_start = pvec_prof_.begin() + W * a_start_idx;
// a_end = pvec_prof_.begin() + W * a_end_idx;
// a_aux_start = aux_prof_.begin() + W * a_start_idx;
//
// }
//
//
//
//
//
// {
// // some basic sanity checks for the input arguments
//
// aiter xxx;
// assert( sizeof(*xxx) == sizeof(score_t));
//
// vu::assert_alignment( &(*a_start) );
// vu::assert_alignment( &(*a_aux_start) );
// vu::assert_alignment( &(*out_start) );
// }
// if a_start_idx and a_end_idx == -1, fall back to whole range (=do full alignment)
if( a_start_idx == size_t(-1) || a_end_idx == size_t(-1) ) {
assert( a_start_idx == a_end_idx );
a_start_idx = 0;
a_end_idx = pvec_prof_.size() / W;
}
// const size_t a_size_bound = av_size_bound / W;
const size_t bsize = std::distance( b_start, b_end );
// overall size of a whole row of vectors (=length of 'a' * vector width)
const size_t av_size_all = pvec_prof_.size();
// const size_t a_size_all = av_size_all / W;
const size_t block_width = 512;
// assert( av_size >= block_width * W ); // the code below should handle this case, but is untested
// actual row length (times vector width) inside the bounded region
const size_t av_size_bound = (a_end_idx - a_start_idx) * W;
const size_t av_minsize = std::max(av_size_bound, block_width * W);
s_.resize( av_minsize );
si_.resize( av_minsize );
if( s_.size() % W != 0 ) {
throw std::runtime_error( "profile length not multiple of vec-unit width." );
}
const score_t SMALL = vu::SMALL_VALUE;
std::fill( s_.begin(), s_.end(), 0 );
std::fill( si_.begin(), si_.end(), SMALL );
// si_[0] = 0;
vec_t max_score = vu::set1(SMALL);
const vec_t zero = vu::setzero();
const vec_t gap_extend = vu::set1(gap_extend_sc);
const vec_t gap_open = vu::set1(gap_open_sc);
// const vec_t match_cgap = vu::set1( match_cgap_sc );
// const vec_t match_score = vu::set1( match_score_sc );
bool done = false;
// ali_score_block_t<vec_t> btemp;
//
// btemp.last_sl = vu::set1(SMALL);
// btemp.last_sc = vu::set1(0);;
// btemp.last_sdiag = vu::set1(0);;
//
// std::vector<ali_score_block_t<vec_t> > blocks( bsize, btemp ); // TODO: maybe put this into the persistent state, if sbrk mucks up again.
typedef ivy_mike::aligned_buffer<score_t,4096> block_vec;
block_vec block_sdiag(bsize * W, 0);
block_vec block_sl(bsize * W, SMALL);
block_vec block_sc(bsize * W, 0);
// ali_ptr_block2_t ptr_block_outer;
// ptr_block_outer.a_prof_iter = a_start;
//ptr_block_outer.a_aux_prof_iter = a_aux_start;
//ptr_block_outer.start = 0;
//size_t block_start_outer;
size_t block_start_outer = a_start_idx;
//ptr_block_outer.s_iter = s_.base();
//ptr_block_outer.si_iter = si_.base();
//std::cerr << "sm_inc: " << &(*sm_inc_prof_.begin()) << std::endl;
ticks ticks1 = getticks();
//size_t inner_iters = 0;
while( !done ) {
// std::cout << "block start outer: " << block_start_outer << "\n";
std::fill( s_.begin(), s_.begin() + W * block_width, 0 );
std::fill( si_.begin(), si_.begin() + W * block_width, SMALL );
// typename std::vector<ali_score_block_t<vec_t> >::iterator it_block = blocks.begin();
typename block_vec::iterator block_sl_it = block_sl.begin();
typename block_vec::iterator block_sc_it = block_sc.begin();
typename block_vec::iterator block_sdiag_it = block_sdiag.begin();
size_t block_end = block_start_outer + block_width;
if( block_end > a_end_idx ) {
block_end = a_end_idx;
}
biter it_b = b_start;
inner_iters_all_ += std::distance(b_start, b_end) * (block_end - block_start_outer);
for( ; it_b != b_end; ++it_b, block_sl_it += W, block_sc_it += W, block_sdiag_it += W ) {
//assert(*it_b >= 0);
assert(*it_b < num_cstates_);
bool lastrow = it_b == (b_end - 1);
vec_t row_max_score = vu::set1(SMALL);
//ali_score_block_t<vec_t> block = *it_block;
//ptr_block = ptr_block_outer;
size_t block_start = block_start_outer;
//aiter a_aux_end_this = ptr_block.a_aux_prof_iter + W * block_width;
score_t * s_iter = s_.base();
score_t * si_iter = si_.base();
// score_t * __restrict a_aux_prof_iter = &(*(a_aux_start + block_start * W));
// score_t * __restrict a_aux_prof_end = &(*(a_aux_start + block_end * W));
score_t * sm_inc_iter = &(*(sm_inc_prof_.begin() + (*it_b) * av_size_all + block_start * W));
score_t * sm_inc_end = &(*(sm_inc_prof_.begin() + (*it_b) * av_size_all + block_end * W));
// _mm_prefetch( (const char *)sm_inc_iter, _MM_HINT_T0 );
vec_t last_sdiag = vu::load( &(*block_sdiag_it));
vec_t last_sl = vu::load( &(*block_sl_it));
vec_t last_sc = vu::load( &(*block_sc_it));
for(; sm_inc_iter != sm_inc_end; sm_inc_iter += W/*, a_aux_prof_iter += W*/, s_iter += W, si_iter += W ) {
// some 'lessions learned' about instruction ordering when using sse intrinsics:
// 1. assigning values that are read/written only once to (const) variables is ok, to improve
// readability, if the assignment is near the use. (rule: don't force the compiler to waste
// registers, by putting unnecessary stuff in between. why? the compiler should be able to reorder everything...)
// 2. arithmetic operations should be placed relative to load/store instructions, in a way
// which gives the compiler the most freedom for rearrangements (as long as rule 1 is not violated)
// 3. the parameter order of vu::max seems critical (why?)
//
// The code is written in a kind of static single assignment form to make manual analysis easier:
// The basic optimization rule is: place operations so that the number of 'active' values that 'cross'
// the operation is minimal. A value is active between its initialization and it's last read-access.
//
// The influence of the compiler is quite large:
// (1) When optimizing for corei7 instead of amdfam10, gcc will order add and andnot instructions differently
// (andnot add andnot add vs. andnot andnot add add). This gives a 5% advantage on nehalem (but not
// sandy bridge or barcelona)
// (2) When optimizing for corei7-avx, gcc will automatically use the new 3-operand versions of the 128bit
// vector instructions, which shortens the inner-loop by around 20% (with corresponding performance increase)
// (3) gcc always uses SIB (scaled index basis) addressing fir the loads/stores. This gains very compact
// code but has a negative influence on AMD cpus (at least) barcelona. Supposingly the SIB addressing
// uses multiple micro-ops on amd, which seem to clog-up the pipeline (the throughput is not much more than
// 2ops/cycle.). Clang (as of late 2011) calculates addresses 'by-hand' and uses non SIP addressing. This
// increses the size of the inner-loop, which in turn runs a bit slower on intel but about 10% faster on AMD.
// (4) MSVC (2011 dev. preview) mixes SIB and manual address calculation (for no apparent reason), and is
// about 20% slower than gcc on sandy-bridge (might be on par with clang).
//const vec_t cgap = vu::load( a_aux_prof_iter );
// match increase: sum of match and match_cgap score/penalty
const vec_t sm_inc = vu::load( sm_inc_iter );
const vec_t cgap = vu::cmp_lt( sm_inc, zero ); // HACK: assume that match_score + match_cgap_penalty < 0
// match score (last_sdiag is preloaded in the previous iteration)
const vec_t s_diag = last_sdiag;
const vec_t sm = vu::add(s_diag, sm_inc );
// open score for gap-from-left (last_sc is the previous main-cell score)
const vec_t sc_left = last_sc;
// const vec_t sl_open = vu::add( sc_left, vu::load( &(*ptr_block.gap_open_cgap_iter )));
const vec_t sl_open = vu::add( sc_left, vu::bit_andnot( cgap, gap_open) );
// extension score for gap-from-left (last_sl is the previous gap-from-left score)
const vec_t sl_left = last_sl;
const vec_t sl_extend = vu::add( sl_left, vu::bit_andnot( cgap, gap_extend) );
//const vec_t sl_extend = vu::add( sl_left, vu::load( &(*ptr_block.gap_extend_cgap_iter )));
const vec_t sl = vu::max( sl_open, sl_extend );
last_sl = sl;
const vec_t sc_above = vu::load( s_iter ); // OPT: it seems to be critical to put this load as late as possible
// last_sdiag now is preloaded with the 'above' value (it will become the 'diag'
// element in the next iteration.)
last_sdiag = sc_above;
// open score for gap from above (remember, last_sdiag is preloaded with the 'above' value)
const vec_t su_open = vu::add( sc_above, gap_open);
// extension score for gap from above (si_iter is the previous gap-from-above score)
const vec_t su_extend = vu::add( vu::load( si_iter ), gap_extend );
// chose between open/extension for gap-from-above
const vec_t su = vu::max( su_open, su_extend );// = max( su_GAP_EXTEND, );
vu::store( su, si_iter );
// overall max = new main-cell score
const vec_t sc = vu::max( sm, vu::max( su, sl ) );
last_sc = sc;
row_max_score = vu::max( row_max_score, sc );
vu::store( last_sc, s_iter );
}
done = block_start == a_end_idx;
if( done ) {
max_score = vu::max( max_score, last_sc );
}
if( lastrow ) {
max_score = vu::max( max_score, row_max_score );
}
//*it_block = block;
vu::store( last_sdiag, &(*block_sdiag_it) );
vu::store( last_sc, &(*block_sc_it) );
vu::store( last_sl, &(*block_sl_it) );
}
block_start_outer = block_end;
}
ticks ticks2 = getticks();
ticks_all_ += uint64_t(elapsed(ticks2, ticks1 ));
//
vu::store( max_score, &(*out_start) );
}
double ticks_per_inner_iter() {
return ticks_all_ / double(inner_iters_all_);
}
uint64_t ticks_all() {
return ticks_all_;
}
uint64_t inner_iters_all() {
return inner_iters_all_;
}
private:
struct ali_score_block_t {
vec_t last_sdiag;
vec_t last_sl;
vec_t last_sc;
};
ivy_mike::aligned_buffer<score_t> s_;
ivy_mike::aligned_buffer<score_t> si_;
ivy_mike::aligned_buffer<score_t> pvec_prof_;