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
|
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta name="generator" content=
"HTML Tidy for Linux/x86 (vers 1st March 2004), see www.w3.org" />
<meta http-equiv="Content-Type" content=
"text/html; charset=us-ascii" />
<link href="ublas.css" type="text/css" />
<title>uBLAS Overview</title>
</head>
<body>
<h1><img src="../../../../boost.png" align="middle" />
uBLAS Overview</h1>
<h2><a name="rationale" id="rationale">Rationale</h2>
<p><cite>It would be nice if every kind of numeric software could
be written in C++ without loss of efficiency, but unless something
can be found that achieves this without compromising the C++ type
system it may be preferable to rely on Fortran, assembler or
architecture-specific extensions (Bjarne Stroustrup).</cite></p>
<p>This C++ library is directed towards scientific computing on the
level of basic linear algebra constructions with matrices and
vectors and their corresponding abstract operations. The primary
design goals were:</p>
<ul type="Disc">
<li>mathematical notation</li>
<li>efficiency</li>
<li>functionality</li>
<li>compatibility</li>
</ul>
<p>Another intention was to evaluate, if the abstraction penalty
resulting from the use of such matrix and vector classes is
acceptable.</p>
<h2>Resources</h2>
<p>The development of this library was guided by a couple of
similar efforts:</p>
<ul type="Disc">
<li><a href="http://www.netlib.org/blas/index.html">BLAS</a> by
Jack Dongarra et al.</li>
<li><a href="http://www.oonumerics.org/blitz/">Blitz++</a> by Todd
Veldhuizen</li>
<li><a href="http://acts.nersc.gov/pooma/">POOMA</a> by Scott
Haney et al.</li>
<li><a href="http://www.lsc.nd.edu/research/mtl/">MTL</a> by Jeremy
Siek et al.</li>
</ul>
<p>BLAS seems to be the most widely used library for basic linear
algebra constructions, so it could be called a de-facto standard.
Its interface is procedural, the individual functions are somewhat
abstracted from simple linear algebra operations. Due to the fact
that is has been implemented using Fortran and its optimizations,
it also seems to be one of the fastest libraries available. As we
decided to design and implement our library in an object-oriented
way, the technical approaches are distinct. However anyone should
be able to express BLAS abstractions in terms of our library
operators and to compare the efficiency of the implementations.</p>
<p>Blitz++ is an impressive library implemented in C++. Its main
design seems to be oriented towards multidimensional arrays and
their associated operators including tensors. The author of Blitz++
states, that the library achieves performance on par or better than
corresponding Fortran code due to his implementation technique
using expression templates and template metaprograms. However we
see some reasons, to develop an own design and implementation
approach. We do not know whether anybody tries to implement
traditional linear algebra and other numerical algorithms using
Blitz++. We also presume that even today Blitz++ needs the most
advanced C++ compiler technology due to its implementation idioms.
On the other hand, Blitz++ convinced us, that the use of expression
templates is mandatory to reduce the abstraction penalty to an
acceptable limit.</p>
<p>POOMA's design goals seem to parallel Blitz++'s in many parts .
It extends Blitz++'s concepts with classes from the domains of
partial differential equations and theoretical physics. The
implementation supports even parallel architectures.</p>
<p>MTL is another approach supporting basic linear algebra
operations in C++. Its design mainly seems to be influenced by BLAS
and the C++ Standard Template Library. We share the insight that a
linear algebra library has to provide functionality comparable to
BLAS. On the other hand we think, that the concepts of the C++
standard library have not yet been proven to support numerical
computations as needed. As another difference MTL currently does
not seem to use expression templates. This may result in one of two
consequences: a possible loss of expressiveness or a possible loss
of performance.</p>
<h2>Concepts</h2>
<h3>Mathematical Notation</h3>
<p>The usage of mathematical notation may ease the development of
scientific algorithms. So a C++ library implementing basic linear
algebra concepts carefully should overload selected C++ operators
on matrix and vector classes.</p>
<p>We decided to use operator overloading for the following
primitives:</p>
<table border="1" summary="operators">
<tbody>
<tr>
<th align="left">Description</th>
<th align="left">Operator</th>
</tr>
<tr>
<td>Indexing of vectors and matrices</td>
<td><code>vector::operator(size_t i);<br />
matrix::operator(size_t i, size_t j);</code></td>
</tr>
<tr>
<td>Assignment of vectors and matrices</td>
<td><code>vector::operator = (const vector_expression &);<br />
vector::operator += (const vector_expression &);<br />
vector::operator -= (const vector_expression &);<br />
vector::operator *= (const scalar_expression &);<br />
matrix::operator = (const matrix_expression &);<br />
matrix::operator += (const matrix_expression &);<br />
matrix::operator -= (const matrix_expression &);<br />
matrix::operator *= (const scalar_expression &);</code></td>
</tr>
<tr>
<td>Unary operations on vectors and matrices</td>
<td><code>vector_expression operator - (const vector_expression
&);<br />
matrix_expression operator - (const matrix_expression
&);</code></td>
</tr>
<tr>
<td>Binary operations on vectors and matrices</td>
<td><code>vector_expression operator + (const vector_expression
&, const vector_expression &);<br />
vector_expression operator - (const vector_expression &, const
vector_expression &);<br />
matrix_expression operator + (const matrix_expression &, const
matrix_expression &);<br />
matrix_expression operator - (const matrix_expression &, const
matrix_expression &);</code></td>
</tr>
<tr>
<td>Multiplication of vectors and matrices with a scalar</td>
<td><code>vector_expression operator * (const scalar_expression
&, const vector_expression &);<br />
vector_expression operator * (const vector_expression &, const
scalar_expression &);<br />
matrix_expression operator * (const scalar_expression &, const
matrix_expression &);<br />
matrix_expression operator * (const matrix_expression &, const
scalar_expression &);</code></td>
</tr>
</tbody>
</table>
<p>We decided to use no operator overloading for the following
other primitives:</p>
<table border="1" summary="functions">
<tbody>
<tr>
<th align="left">Description</th>
<th align="left">Function</th>
</tr>
<tr>
<td>Left multiplication of vectors with a matrix</td>
<td><code>vector_expression
prod<</code><code><em>vector_type</em></code> <code>> (const
matrix_expression &, const vector_expression &);<br />
vector_expression prod (const matrix_expression &, const
vector_expression &);</code></td>
</tr>
<tr>
<td>Right multiplication of vectors with a matrix</td>
<td><code>vector_expression
prod<</code><code><em>vector_type</em></code> <code>> (const
vector_expression &, const matrix_expression &);<br />
vector_expression prod (const vector_expression &, const
matrix_expression &);<br /></code></td>
</tr>
<tr>
<td>Multiplication of matrices</td>
<td><code>matrix_expression
prod<</code><code><em>matrix_type</em></code> <code>> (const
matrix_expression &, const matrix_expression &);<br />
matrix_expression prod (const matrix_expression &, const
matrix_expression &);</code></td>
</tr>
<tr>
<td>Inner product of vectors</td>
<td><code>scalar_expression inner_prod (const vector_expression
&, const vector_expression &);</code></td>
</tr>
<tr>
<td>Outer product of vectors</td>
<td><code>matrix_expression outer_prod (const vector_expression
&, const vector_expression &);</code></td>
</tr>
<tr>
<td>Transpose of a matrix</td>
<td><code>matrix_expression trans (const matrix_expression
&);</code></td>
</tr>
</tbody>
</table>
<h3>Efficiency</h3>
<p>To achieve the goal of efficiency for numerical computing, one
has to overcome two difficulties in formulating abstractions with
C++, namely temporaries and virtual function calls. Expression
templates solve these problems, but tend to slow down compilation
times.</p>
<h4>Eliminating Temporaries</h4>
<p>Abstract formulas on vectors and matrices normally compose a
couple of unary and binary operations. The conventional way of
evaluating such a formula is first to evaluate every leaf operation
of a composition into a temporary and next to evaluate the
composite resulting in another temporary. This method is expensive
in terms of time especially for small and space especially for
large vectors and matrices. The approach to solve this problem is
to use lazy evaluation as known from modern functional programming
languages. The principle of this approach is to evaluate a complex
expression element wise and to assign it directly to the
target.</p>
<p>Two interesting and dangerous facts result:</p>
<h4>Aliases</h4>
<p>One may get serious side effects using element wise
evaluation on vectors or matrices. Consider the matrix vector
product <em>x = A x</em>. Evaluation of
<em>A</em><sub><em>1</em></sub><em>x</em> and assignment to
<em>x</em><sub><em>1</em></sub> changes the right hand side, so
that the evaluation of <em>A</em><sub><em>2</em></sub><em>x</em>
returns a wrong result. In this case there are <strong>aliases</strong> of the elements
<em>x</em><sub><em>n</em></sub> on both the left and right hand side of the assignment.</p>
<p>Our solution for this problem is to
evaluate the right hand side of an assignment into a temporary and
then to assign this temporary to the left hand side. To allow
further optimizations, we provide a corresponding member function
for every assignment operator and also a
<a href="operations_overview.htm#noalias"> <code>noalias</code> syntax.</a>
By using this syntax a programmer can confirm, that the left and right hand sides of an
assignment are independent, so that element wise evaluation and
direct assignment to the target is safe.</p>
<h4>Complexity</h4>
<p>The computational complexity may be unexpectedly large under certain
cirumstances. Consider the chained matrix vector product <em>A (B
x)</em>. Conventional evaluation of <em>A (B x)</em> is quadratic.
Deferred evaluation of <em>B x</em><sub><em>i</em></sub> is linear.
As every element <em>B x</em><sub><em>i</em></sub> is needed
linearly depending of the size, a completely deferred evaluation of
the chained matrix vector product <em>A (B x)</em> is cubic. In
such cases one needs to reintroduce temporaries in the
expression.</p>
<h4>Eliminating Virtual Function Calls</h4>
<p>Lazy expression evaluation normally leads to the definition of a
class hierarchy of terms. This results in the usage of dynamic
polymorphism to access single elements of vectors and matrices,
which is also known to be expensive in terms of time. A solution
was found a couple of years ago independently by David Vandervoorde
and Todd Veldhuizen and is commonly called expression templates.
Expression templates contain lazy evaluation and replace dynamic
polymorphism with static, i.e. compile time polymorphism.
Expression templates heavily depend on the famous Barton-Nackman
trick, also coined 'curiously defined recursive templates' by Jim
Coplien.</p>
<p>Expression templates form the base of our implementation.</p>
<h4>Compilation times</h4>
<p>It is also a well known fact, that expression templates
challenge currently available compilers. We were able to
significantly reduce the amount of needed expression templates
using the Barton-Nackman trick consequently.</p>
<p>We also decided to support a dual conventional implementation
(i.e. not using expression templates) with extensive bounds and
type checking of vector and matrix operations to support the
development cycle. Switching from debug mode to release mode is
controlled by the <code>NDEBUG</code> preprocessor symbol of
<code><cassert></code>.</p>
<h2><a name="functionality" id="functionality">Functionality</h2>
<p>Every C++ library supporting linear algebra will be measured
against the long-standing Fortran package BLAS. We now describe how
BLAS calls may be mapped onto our classes.</p>
<p>The page <a href="operations_overview.htm">Overview of Matrix and Vector Operations</a>
gives a short summary of the most used operations on vectors and
matrices.</p>
<h4>Blas Level 1</h4>
<table border="1" summary="level 1 blas">
<tbody>
<tr>
<th align="left">BLAS Call</th>
<th align="left">Mapped Library Expression</th>
<th align="left">Mathematical Description</th>
<th align="left">Comment</th>
</tr>
<tr>
<td><code>sasum</code> OR <code>dasum</code></td>
<td><code>norm_1 (x)</code></td>
<td><em>sum x<sub>i</sub></em></td>
<td>Computes the <em>l<sub>1</sub></em> (sum) norm of a real vector.</td>
</tr>
<tr>
<td><code>scasum</code> OR <code>dzasum</code></td>
<td><em><code>real (sum (v)) + imag (sum (v))</code></em></td>
<td><em>sum re(x<sub>i</sub>) + sum im(x<sub>i</sub>)</em></td>
<td>Computes the sum of elements of a complex vector.</td>
</tr>
<tr>
<td><code>_nrm2</code></td>
<td><code>norm_2 (x)</code></td>
<td><em>sqrt (sum
|x</em><sub><em>i</em></sub>|<sup><em>2</em></sup> <em>)</em></td>
<td>Computes the <em>l<sub>inf</sub></em> (euclidean) norm of a vector.</td>
</tr>
<tr>
<td><code>i_amax</code></td>
<td><code>norm_inf (x)<br />
norm_inf_index (x)</code></td>
<td><em>max |x</em><sub><em>i</em></sub><em>|</em></td>
<td>Computes the <em>l<sub>2</sub></em> (maximum) norm of a vector.<br />
BLAS computes the index of the first element having this
value.</td>
</tr>
<tr>
<td><code>_dot<br />
_dotu<br />
_dotc</code></td>
<td><code>inner_prod (x, y)</code>or<code><br />
inner_prod (conj (x), y)</code></td>
<td><em>x</em><sup><em>T</em></sup> <em>y</em> or<br />
<em>x</em><sup><em>H</em></sup> <em>y</em></td>
<td>Computes the inner product of two vectors.<br />
BLAS implements certain loop unrollment.</td>
</tr>
<tr>
<td><code>dsdot<br />
sdsdot</code></td>
<td><code>a + prec_inner_prod (x, y)</code></td>
<td><em>a + x</em><sup><em>T</em></sup> <em>y</em></td>
<td>Computes the inner product in double precision.</td>
</tr>
<tr>
<td><code>_copy</code></td>
<td><code>x = y<br />
y.assign (x)</code></td>
<td><em>x <- y</em></td>
<td>Copies one vector to another.<br />
BLAS implements certain loop unrollment.</td>
</tr>
<tr>
<td><code>_swap</code></td>
<td><code>swap (x, y)</code></td>
<td><em>x <-> y</em></td>
<td>Swaps two vectors.<br />
BLAS implements certain loop unrollment.</td>
</tr>
<tr>
<td><code>_scal<br />
csscal<br />
zdscal</code></td>
<td><code>x *= a</code></td>
<td><em>x <- a x</em></td>
<td>Scales a vector.<br />
BLAS implements certain loop unrollment.</td>
</tr>
<tr>
<td><code>_axpy</code></td>
<td><code>y += a * x</code></td>
<td><em>y <- a x + y</em></td>
<td>Adds a scaled vector.<br />
BLAS implements certain loop unrollment.</td>
</tr>
<tr>
<td><code>_rot<br />
_rotm<br />
csrot<br />
zdrot</code></td>
<td><code>t.assign (a * x + b * y),<br />
y.assign (- b * x + a * y),<br />
x.assign (t)</code></td>
<td><em>(x, y) <- (a x + b y, -b x + a y)</em></td>
<td>Applies a plane rotation.</td>
</tr>
<tr>
<td><code>_rotg<br />
_rotmg</code></td>
<td> </td>
<td><em>(a, b) <-<br />
(? a / sqrt (a</em><sup><em>2</em></sup> +
<em>b</em><sup><em>2</em></sup><em>),<br />
? b / sqrt (a</em><sup><em>2</em></sup> +
<em>b</em><sup><em>2</em></sup><em>))</em> or<em><br />
(1, 0) <- (0, 0)</em></td>
<td>Constructs a plane rotation.</td>
</tr>
</tbody>
</table>
<h4>Blas Level 2</h4>
<table border="1" summary="level 2 blas">
<tbody>
<tr>
<th align="left">BLAS Call</th>
<th align="left">Mapped Library Expression</th>
<th align="left">Mathematical Description</th>
<th align="left">Comment</th>
</tr>
<tr>
<td><code>_t_mv</code></td>
<td><code>x = prod (A, x)</code> or<code><br />
x = prod (trans (A), x)</code> or<code><br />
x = prod (herm (A), x)</code></td>
<td><em>x <- A x</em> or<em><br />
x <- A</em><sup><em>T</em></sup> <em>x</em> or<em><br />
x <- A</em><sup><em>H</em></sup> <em>x</em></td>
<td>Computes the product of a matrix with a vector.</td>
</tr>
<tr>
<td><code>_t_sv</code></td>
<td><code>y = solve (A, x, tag)</code> or<br />
<code>inplace_solve (A, x, tag)</code> or<br />
<code>y = solve (trans (A), x, tag)</code> or<br />
<code>inplace_solve (trans (A), x, tag)</code> or<br />
<code>y = solve (herm (A), x, tag)</code>or<br />
<code>inplace_solve (herm (A), x, tag)</code></td>
<!-- TODO: replace nested sub/sup -->
<td><em>y <- A</em><sup><em>-1</em></sup> <em>x</em>
or<em><br />
x <- A</em><sup><em>-1</em></sup> <em>x</em> or<em><br />
y <-
A</em><sup><em>T</em></sup><sup><sup><em>-1</em></sup></sup>
<em>x</em> or<em><br />
x <-
A</em><sup><em>T</em></sup><sup><sup><em>-1</em></sup></sup>
<em>x</em> or<em><br />
y <-
A</em><sup><em>H</em></sup><sup><sup><em>-1</em></sup></sup>
<em>x</em> or<em><br />
x <-
A</em><sup><em>H</em></sup><sup><sup><em>-1</em></sup></sup>
<em>x</em></td>
<td>Solves a system of linear equations with triangular form, i.e.
<em>A</em> is triangular.</td>
</tr>
<tr>
<td><code>_g_mv<br />
_s_mv<br />
_h_mv</code></td>
<td><code>y = a * prod (A, x) + b * y</code> or<code><br />
y = a * prod (trans (A), x) + b * y</code> or<code><br />
y = a * prod (herm (A), x) + b * y</code></td>
<td><em>y <- a A x + b y</em> or<em><br />
y <- a A</em><sup><em>T</em></sup> <em>x + b y<br />
y <- a A</em><sup><em>H</em></sup> <em>x + b y</em></td>
<td>Adds the scaled product of a matrix with a vector.</td>
</tr>
<tr>
<td><code>_g_r<br />
_g_ru<br />
_g_rc</code></td>
<td><code>A += a * outer_prod (x, y)</code> or<code><br />
A += a * outer_prod (x, conj (y))</code></td>
<td><em>A <- a x y</em><sup><em>T</em></sup> <em>+ A</em>
or<em><br />
A <- a x y</em><sup><em>H</em></sup> <em>+ A</em></td>
<td>Performs a rank <em>1</em> update.</td>
</tr>
<tr>
<td><code>_s_r<br />
_h_r</code></td>
<td><code>A += a * outer_prod (x, x)</code> or<code><br />
A += a * outer_prod (x, conj (x))</code></td>
<td><em>A <- a x x</em><sup><em>T</em></sup> <em>+ A</em>
or<em><br />
A <- a x x</em><sup><em>H</em></sup> <em>+ A</em></td>
<td>Performs a symmetric or hermitian rank <em>1</em> update.</td>
</tr>
<tr>
<td><code>_s_r2<br />
_h_r2</code></td>
<td><code>A += a * outer_prod (x, y) +<br />
a * outer_prod (y, x))</code> or<code><br />
A += a * outer_prod (x, conj (y)) +<br />
conj (a) * outer_prod (y, conj (x)))</code></td>
<td><em>A <- a x y</em><sup><em>T</em></sup> <em>+ a y
x</em><sup><em>T</em></sup> <em>+ A</em> or<em><br />
A <- a x y</em><sup><em>H</em></sup> <em>+
a</em><sup><em>-</em></sup> <em>y x</em><sup><em>H</em></sup> <em>+
A</em></td>
<td>Performs a symmetric or hermitian rank <em>2</em> update.</td>
</tr>
</tbody>
</table>
<h4>Blas Level 3</h4>
<table border="1" summary="level 3 blas">
<tbody>
<tr>
<th align="left">BLAS Call</th>
<th align="left">Mapped Library Expression</th>
<th align="left">Mathematical Description</th>
<th align="left">Comment</th>
</tr>
<tr>
<td><code>_t_mm</code></td>
<td><code>B = a * prod (A, B)</code> or<br />
<code>B = a * prod (trans (A), B)</code> or<br />
<code>B = a * prod (A, trans (B))</code> or<br />
<code>B = a * prod (trans (A), trans (B))</code> or<br />
<code>B = a * prod (herm (A), B)</code> or<br />
<code>B = a * prod (A, herm (B))</code> or<br />
<code>B = a * prod (herm (A), trans (B))</code> or<br />
<code>B = a * prod (trans (A), herm (B))</code> or<br />
<code>B = a * prod (herm (A), herm (B))</code></td>
<td><em>B <- a op (A) op (B)</em> with<br />
<em>op (X) = X</em> or<br />
<em>op (X) = X</em><sup><em>T</em></sup> or<br />
<em>op (X) = X</em><sup><em>H</em></sup></td>
<td>Computes the scaled product of two matrices.</td>
</tr>
<tr>
<td><code>_t_sm</code></td>
<td><code>C = solve (A, B, tag)</code> or<br />
<code>inplace_solve (A, B, tag)</code> or<br />
<code>C = solve (trans (A), B, tag)</code> or<code><br />
inplace_solve (trans (A), B, tag)</code> or<code><br />
C = solve (herm (A), B, tag)</code> or<code><br />
inplace_solve (herm (A), B, tag)</code></td>
<td><em>C <- A</em><sup><em>-1</em></sup> <em>B</em>
or<em><br />
B <- A</em><sup><em>-1</em></sup> <em>B</em> or<em><br />
C <-
A</em><sup><em>T</em></sup><sup><sup><em>-1</em></sup></sup>
<em>B</em> or<em><br />
B <- A</em><sup><em>-1</em></sup> <em>B</em> or<em><br />
C <-
A</em><sup><em>H</em></sup><sup><sup><em>-1</em></sup></sup>
<em>B</em> or<em><br />
B <-
A</em><sup><em>H</em></sup><sup><sup><em>-1</em></sup></sup>
<em>B</em></td>
<td>Solves a system of linear equations with triangular form, i.e.
<em>A</em> is triangular.</td>
</tr>
<tr>
<td><code>_g_mm<br />
_s_mm<br />
_h_mm</code></td>
<td><code>C = a * prod (A, B) + b * C</code> or<br />
<code>C = a * prod (trans (A), B) + b * C</code> or<br />
<code>C = a * prod (A, trans (B)) + b * C</code> or<br />
<code>C = a * prod (trans (A), trans (B)) + b * C</code> or<br />
<code>C = a * prod (herm (A), B) + b * C</code> or<br />
<code>C = a * prod (A, herm (B)) + b * C</code> or<br />
<code>C = a * prod (herm (A), trans (B)) + b * C</code> or<br />
<code>C = a * prod (trans (A), herm (B)) + b * C</code> or<br />
<code>C = a * prod (herm (A), herm (B)) + b * C</code></td>
<td><em>C <- a op (A) op (B) + b C</em> with<br />
<em>op (X) = X</em> or<br />
<em>op (X) = X</em><sup><em>T</em></sup> or<br />
<em>op (X) = X</em><sup><em>H</em></sup></td>
<td>Adds the scaled product of two matrices.</td>
</tr>
<tr>
<td><code>_s_rk<br />
_h_rk</code></td>
<td><code>B = a * prod (A, trans (A)) + b * B</code> or<br />
<code>B = a * prod (trans (A), A) + b * B</code> or<br />
<code>B = a * prod (A, herm (A)) + b * B</code> or<br />
<code>B = a * prod (herm (A), A) + b * B</code></td>
<td><em>B <- a A A</em><sup><em>T</em></sup> <em>+ b B</em>
or<em><br />
B <- a A</em><sup><em>T</em></sup> <em>A + b B</em> or<br />
<em>B <- a A A</em><sup><em>H</em></sup> <em>+ b B</em>
or<em><br />
B <- a A</em><sup><em>H</em></sup> <em>A + b B</em></td>
<td>Performs a symmetric or hermitian rank <em>k</em> update.</td>
</tr>
<tr>
<td><code>_s_r2k<br />
_h_r2k</code></td>
<td><code>C = a * prod (A, trans (B)) +<br />
a * prod (B, trans (A)) + b * C</code> or<br />
<code>C = a * prod (trans (A), B) +<br />
a * prod (trans (B), A) + b * C</code> or<br />
<code>C = a * prod (A, herm (B)) +<br />
conj (a) * prod (B, herm (A)) + b * C</code> or<br />
<code>C = a * prod (herm (A), B) +<br />
conj (a) * prod (herm (B), A) + b * C</code></td>
<td><em>C <- a A B</em><sup><em>T</em></sup> <em>+ a B
A</em><sup><em>T</em></sup> <em>+ b C</em> or<em><br />
C <- a A</em><sup><em>T</em></sup> <em>B + a
B</em><sup><em>T</em></sup> <em>A + b C</em> or<em><br />
C <- a A B</em><sup><em>H</em></sup> <em>+
a</em><sup><em>-</em></sup> <em>B A</em><sup><em>H</em></sup> <em>+
b C</em> or<em><br />
C <- a A</em><sup><em>H</em></sup> <em>B +
a</em><sup><em>-</em></sup> <em>B</em><sup><em>H</em></sup> <em>A +
b C</em></td>
<td>Performs a symmetric or hermitian rank <em>2 k</em>
update.</td>
</tr>
</tbody>
</table>
<h2>Storage Layout</h2>
<p>uBLAS supports may different storage layouts. The full details can be
found at the <a href="types_overview.htm">Overview of Types</a>. Most types like
<code>vector<double></code> and <code>matrix<double></code> are
by default compatible to C arrays, but can also be configured to contain
FORTAN compatible data.
</p>
<h2>Compatibility</h2>
<p>For compatibility reasons we provide array like indexing for vectors and matrices. For some types (hermitian, sparse etc) this can be expensive for matrices due to the needed temporary proxy objects.</p>
<p>uBLAS uses STL compatible allocators for the allocation of the storage required for it's containers.</p>
<h2>Benchmark Results</h2>
<p>The following tables contain results of one of our benchmarks.
This benchmark compares a native C implementation ('C array') and
some library based implementations. The safe variants based on the
library assume aliasing, the fast variants do not use temporaries
and are functionally equivalent to the native C implementation.
Besides the generic vector and matrix classes the benchmark
utilizes special classes <code>c_vector</code> and
<code>c_matrix</code>, which are intended to avoid every overhead
through genericity.</p>
<p>The benchmark program <strong>bench1</strong> was compiled with GCC 4.0 and run on an Athlon 64 3000+. Times are scales for reasonable precision by running <strong>bench1 100</strong>.</p>
<p>First we comment the results for double vectors and matrices of dimension 3 and 3 x 3, respectively.</p>
<table border="1" summary="1st benchmark">
<tbody>
<tr>
<th align="left">Comment</th>
</tr>
<tr>
<td rowspan="3">inner_prod</td>
<td>C array</td>
<td align="right">0.61</td>
<td align="right">782</td>
<td rowspan="3">Some abstraction penalty</td>
</tr>
<tr>
<td>c_vector</td>
<td align="right">0.86</td>
<td align="right">554</td>
</tr>
<tr>
<td>vector<unbounded_array></td>
<td align="right">1.02</td>
<td align="right">467</td>
</tr>
<tr>
<td rowspan="5">vector + vector</td>
<td>C array</td>
<td align="right">0.51</td>
<td align="right">1122</td>
<td rowspan="5">Abstraction penalty: factor 2</td>
</tr>
<tr>
<td>c_vector fast</td>
<td align="right">1.17</td>
<td align="right">489</td>
</tr>
<tr>
<td>vector<unbounded_array> fast</td>
<td align="right">1.32</td>
<td align="right">433</td>
</tr>
<tr>
<td>c_vector safe</td>
<td align="right">2.02</td>
<td align="right">283</td>
</tr>
<tr>
<td>vector<unbounded_array> safe</td>
<td align="right">6.95</td>
<td align="right">82</td>
</tr>
<tr>
<td rowspan="5">outer_prod</td>
<td>C array</td>
<td align="right">0.59</td>
<td align="right">872</td>
<td rowspan="5">Some abstraction penalty</td>
</tr>
<tr>
<td>c_matrix, c_vector fast</td>
<td align="right">0.88</td>
<td align="right">585</td>
</tr>
<tr>
<td>matrix<unbounded_array>, vector<unbounded_array> fast</td>
<td align="right">0.90</td>
<td align="right">572</td>
</tr>
<tr>
<td>c_matrix, c_vector safe</td>
<td align="right">1.66</td>
<td align="right">310</td>
</tr>
<tr>
<td>matrix<unbounded_array>, vector<unbounded_array> safe</td>
<td align="right">2.95</td>
<td align="right">175</td>
</tr>
<tr>
<td rowspan="5">prod (matrix, vector)</td>
<td>C array</td>
<td align="right">0.64</td>
<td align="right">671</td>
<td rowspan="5">No significant abstraction penalty</td>
</tr>
<tr>
<td>c_matrix, c_vector fast</td>
<td align="right">0.70</td>
<td align="right">613</td>
</tr>
<tr>
<td>matrix<unbounded_array>, vector<unbounded_array> fast</td>
<td align="right">0.79</td>
<td align="right">543</td>
</tr>
<tr>
<td>c_matrix, c_vector safe</td>
<td align="right">0.95</td>
<td align="right">452</td>
</tr>
<tr>
<td>matrix<unbounded_array>, vector<unbounded_array> safe</td>
<td align="right">2.61</td>
<td align="right">164</td>
</tr>
<tr>
<td rowspan="5">matrix + matrix</td>
<td>C array</td>
<td align="right">0.75</td>
<td align="right">686</td>
<td rowspan="5">No significant abstraction penalty</td>
</tr>
<tr>
<td>c_matrix fast</td>
<td align="right">0.99</td>
<td align="right">520</td>
</tr>
<tr>
<td>matrix<unbounded_array> fast</td>
<td align="right">1.29</td>
<td align="right">399</td>
</tr>
<tr>
<td>c_matrix safe</td>
<td align="right">1.7</td>
<td align="right">303</td>
</tr>
<tr>
<td>matrix<unbounded_array> safe</td>
<td align="right">3.14</td>
<td align="right">164</td>
</tr>
<tr>
<td rowspan="5">prod (matrix, matrix)</td>
<td>C array</td>
<td align="right">0.94</td>
<td align="right">457</td>
<td rowspan="5">No significant abstraction penalty</td>
</tr>
<tr>
<td>c_matrix fast</td>
<td align="right">1.17</td>
<td align="right">367</td>
</tr>
<tr>
<td>matrix<unbounded_array> fast</td>
<td align="right">1.34</td>
<td align="right">320</td>
</tr>
<tr>
<td>c_matrix safe</td>
<td align="right">1.56</td>
<td align="right">275</td>
</tr>
<tr>
<td>matrix<unbounded_array> safe</td>
<td align="right">2.06</td>
<td align="right">208</td>
</tr>
</tbody>
</table>
<p>We notice a two fold performance loss for small vectors and matrices: first the general abstraction penalty for using classes, and then a small loss when using the generic vector and matrix classes. The difference w.r.t. alias assumptions is also significant.</p>
<p>Next we comment the results for double vectors and matrices of
dimension 100 and 100 x 100, respectively.</p>
<table border="1" summary="2nd benchmark">
<tbody>
<tr>
<th align="left">Operation</th>
<th align="left">Implementation</th>
<th align="left">Elapsed [s]</th>
<th align="left">MFLOP/s</th>
<th align="left">Comment</th>
</tr>
<tr>
<td rowspan="3">inner_prod</td>
<td>C array</td>
<td align="right">0.64</td>
<td align="right">889</td>
<td rowspan="3">No significant abstraction penalty</td>
</tr>
<tr>
<td>c_vector</td>
<td align="right">0.66</td>
<td align="right">862</td>
</tr>
<tr>
<td>vector<unbounded_array></td>
<td align="right">0.66</td>
<td align="right">862</td>
</tr>
<tr>
<td rowspan="5">vector + vector</td>
<td>C array</td>
<td align="right">0.64</td>
<td align="right">894</td>
<td rowspan="5">No significant abstraction penalty</td>
</tr>
<tr>
<td>c_vector fast</td>
<td align="right">0.66</td>
<td align="right">867</td>
</tr>
<tr>
<td>vector<unbounded_array> fast</td>
<td align="right">0.66</td>
<td align="right">867</td>
</tr>
<tr>
<td>c_vector safe</td>
<td align="right">1.14</td>
<td align="right">501</td>
</tr>
<tr>
<td>vector<unbounded_array> safe</td>
<td align="right">1.23</td>
<td align="right">465</td>
</tr>
<tr>
<td rowspan="5">outer_prod</td>
<td>C array</td>
<td align="right">0.50</td>
<td align="right">1144</td>
<td rowspan="5">No significant abstraction penalty</td>
</tr>
<tr>
<td>c_matrix, c_vector fast</td>
<td align="right">0.71</td>
<td align="right">806</td>
</tr>
<tr>
<td>matrix<unbounded_array>, vector<unbounded_array> fast</td>
<td align="right">0.57</td>
<td align="right">1004</td>
</tr>
<tr>
<td>c_matrix, c_vector safe</td>
<td align="right">1.91</td>
<td align="right">300</td>
</tr>
<tr>
<td>matrix<unbounded_array>, vector<unbounded_array> safe</td>
<td align="right">0.89</td>
<td align="right">643</td>
</tr>
<tr>
<td rowspan="5">prod (matrix, vector)</td>
<td>C array</td>
<td align="right">0.65</td>
<td align="right">876</td>
<td rowspan="5">No significant abstraction penalty</td>
</tr>
<tr>
<td>c_matrix, c_vector fast</td>
<td align="right">0.65</td>
<td align="right">876</td>
</tr>
<tr>
<td>matrix<unbounded_array>, vector<unbounded_array>
fast</td>
<td align="right">0.66</td>
<td align="right">863</td>
</tr>
<tr>
<td>c_matrix, c_vector safe</td>
<td align="right">0.66</td>
<td align="right">863</td>
</tr>
<tr>
<td>matrix<unbounded_array>, vector<unbounded_array>
safe</td>
<td align="right">0.66</td>
<td align="right">863</td>
</tr>
<tr>
<td rowspan="5">matrix + matrix</td>
<td>C array</td>
<td align="right">0.96</td>
<td align="right">596</td>
<td rowspan="5">No significant abstraction penalty</td>
</tr>
<tr>
<td>c_matrix fast</td>
<td align="right">1.21</td>
<td align="right">473</td>
</tr>
<tr>
<td>matrix<unbounded_array> fast</td>
<td align="right">1.00</td>
<td align="right">572</td>
</tr>
<tr>
<td>c_matrix safe</td>
<td align="right">2.44</td>
<td align="right">235</td>
</tr>
<tr>
<td>matrix<unbounded_array> safe</td>
<td align="right">1.30</td>
<td align="right">440</td>
</tr>
<tr>
<td rowspan="5">prod (matrix, matrix)</td>
<td>C array</td>
<td align="right">0.70</td>
<td align="right">813</td>
<td rowspan="5">No significant abstraction penalty</td>
</tr>
<tr>
<td>c_matrix fast</td>
<td align="right">0.73</td>
<td align="right">780</td>
</tr>
<tr>
<td>matrix<unbounded_array> fast</td>
<td align="right">0.76</td>
<td align="right">749</td>
</tr>
<tr>
<td>c_matrix safe</td>
<td align="right">0.75</td>
<td align="right">759</td>
</tr>
<tr>
<td>matrix<unbounded_array> safe</td>
<td align="right">0.76</td>
<td align="right">749</td>
</tr>
</tbody>
</table>
<p>For larger vectors and matrices the general abstraction penalty
for using classes seems to decrease, the small loss when using
generic vector and matrix classes seems to remain. The difference
w.r.t. alias assumptions remains visible, too.</p>
<hr />
<p>Copyright (©) 2000-2002 Joerg Walter, Mathias Koch<br />
Permission to copy, use, modify, sell and distribute this document
is granted provided this copyright notice appears in all copies.
This document is provided ``as is'' without express or implied
warranty, and with no claim as to its suitability for any
purpose.</p>
</body>
</html>
|
