Statistiques
| Révision :

root / www / errata.html @ 1

Historique | Voir | Annoter | Télécharger (3,94 ko)

1
<HTML>
2
<HEAD>
3
<TITLE>HPL Errata-Bugs</TITLE>
4
</HEAD>
5

    
6
<BODY
7
BGCOLOR     = "WHITE"
8
BACKGROUND  = "WHITE"
9
TEXT        = "#000000"
10
VLINK       = "#000099"
11
ALINK       = "#947153"
12
LINK        = "#0000ff">
13

    
14
<H2>HPL Errata - Bugs</H2>
15

    
16
<H3>Issues fixed in Version 2.0, September 10th, 2008</H3>
17

    
18
Gregory Bauer  found  a  problem  size  corresponding  to the
19
periodicity of the pseudo-random matrix generator used in the
20
HPL  timing  program. This causes  the  LU  factorization  to
21
detect the singularity of the input matrix as it should have.<BR><BR>
22

    
23
A problem size of 2^17 = 131072 causes columns 14 modulo 2^14
24
(i.e. 16384)  (starting from 0)  to be bitwise identical on a
25
homogeneous platform.  Every problem size  being a power of 2
26
and larger than  2^15  will  feature a similar problem if one
27
searches far enough in the columns of the square input matrix.<BR><BR>
28

    
29
The pseudo-random  generator  uses  the  linear  congruential
30
algorithm:  X(n+1) = (a * X(n) + c) mod m as described in the
31
Art of Computer  Programming, Knuth 1973,  Vol. 2. In the HPL
32
case, m is set to 2^31.<BR><BR>
33

    
34
It is very important  to realize that this issue is a problem
35
of  the  testing  part  of the  HPL software.  The  numerical
36
properties  of the  algorithms  used in the factorization and
37
the solve should not be questioned because of this.  In fact,
38
this is just the opposite: the factorization demonstrated the
39
weakness of the testing part of the software by detecting the
40
singularity of the input matrix.<BR><BR>
41

    
42
This issue of  the testing program  is not easy to fix.  This
43
pseudo-random  generator  has  very useful properties despite
44
this.  It is  thus currently recommended to HPL users willing
45
to test matrices of size larger than  2^15  to  not use power
46
twos.<BR><BR>
47

    
48
This  issue  has  been fixed by  changing  the  pseudo-random
49
matrix  generator.   Now the  periodicity of the generator is
50
2^64.<BR><BR>
51

    
52
<H3>Issues fixed in Version 1.0b, December 15th, 2004</H3>
53

    
54
When the matrix size is such that one needs  more  than 16 GB
55
per  MPI  rank,  the  intermediate  calculation  (mat.ld+1) *
56
mat.nq in  HPL_pdtest.c  ends up  overflowing  because  it is
57
done using  32-bit arithmetic.   This issue has been fixed by
58
typecasting to size_t; Thanks to John Baron.<BR><BR>
59

    
60
<H3>Issues fixed in Version 1.0a, January 20th, 2004</H3>
61

    
62
The  MPI  process  grid numbering scheme defaults now to row-
63
major ordering. This option can now be selected at run time.<BR><BR>
64

    
65
The  inlined  assembly  timer  routine  that  was causing the
66
compilation to fail when using gcc version  3.3 and above has
67
been removed from the package.<BR><BR>
68

    
69
Various building problems on the T3E have been fixed;  Thanks
70
to Edward Anderson.<BR><BR>
71

    
72
<H3>Issues fixed in Version 1.0, September 27th, 2000</H3>
73

    
74
Due to a  couple errors  spotted in the  VSIPL  port  of  the
75
software,  the  distribution  contained  in  the  tar file of
76
September 9th, 2000 had been updated on September 27th,  2000
77
with a corrected  distribution.  <STRONG>These  problems were
78
not affecting in any way possible the  BLAS  version  of  the
79
software.</STRONG>  If you are using  the  VSIPL port of HPL,
80
and  want  to  make  sure  you are  indeed  using  the latest
81
corrected version, please  check  the  date  contained in the
82
file HPL.build.log contained in the main directory.<BR><BR>
83

    
84

    
85

    
86

    
87
<HR NOSHADE
88
<CENTER
89
<A  = "index.html">            [Home]</A>
90
<A HREF = "copyright.html">        [Copyright and Licensing Terms]</A>
91
<A HREF = "algorithm.html">        [Algorithm]</A>
92
<A HREF = "scalability.html">      [Scalability]</A>
93
<A HREF = "results.html">          [Performance Results]</A>
94
<A HREF = "documentation.html">    [Documentation]</A>
95
<A HREF = "software.html">         [Software]</A>
96
<A HREF = "faqs.html">             [FAQs]</A>
97
<A HREF = "tuning.html">           [Tuning]</A>
98
<A HREF = "errata.html">           [Errata-Bugs]</A>
99
<A HREF = "references.html">       [References]</A>
100
<A HREF = "links.html">            [Related Links]</A><BR>
101
</CENTER>
102
<HR NOSHADE
103
</BODY
104
</HTML