1HPL_pdlaswp01N(3)            HPL Library Functions           HPL_pdlaswp01N(3)
2
3
4

NAME

6       HPL_pdlaswp01N - Broadcast a column panel L and swap the row panel U.
7

SYNOPSIS

9       #include "hpl.h"
10
11       void  HPL_pdlaswp01N(  HPL_T_panel  * PBCST, int * IFLAG, HPL_T_panel *
12       PANEL, const int NN );
13

DESCRIPTION

15       HPL_pdlaswp01N applies the  NB  row interchanges to  NN columns of  the
16       trailing submatrix and broadcast a column panel.
17
18       A  "Spread then roll" algorithm performs  the swap :: broadcast  of the
19       row panel U at once,  resulting in a minimal communication volume   and
20       a  "very good"  use of the connectivity if available.  With  P  process
21       rows  and  assuming  bi-directional links,  the  running time  of  this
22       function can be approximated by:
23
24          (log_2(P)+(P-1)) * lat +   K * NB * LocQ(N) / bdwth
25
26       where   NB   is the number of rows of the row panel U,  N is the global
27       number of columns being updated,  lat and bdwth  are the  latency   and
28       bandwidth  of  the  network  for  double  precision real words.  K is a
29       constant in (2,3] that depends on the  achieved  bandwidth   during   a
30       simultaneous   message  exchange  between two processes.  An  empirical
31       optimistic value of K is typically 2.4.
32

ARGUMENTS

34       PBCST   (local input/output)    HPL_T_panel *
35               On entry,  PBCST  points to the data structure  containing  the
36               panel (to be broadcast) information.
37
38       IFLAG   (local input/output)    int *
39               On  entry,  IFLAG  indicates  whether or not  the broadcast has
40               already been completed.  If not,  probing will occur,  and  the
41               outcome will be contained in IFLAG on exit.
42
43       PANEL   (local input/output)    HPL_T_panel *
44               On  entry,   PANEL  points to the data structure containing the
45               panel information.
46
47       NN      (local input)           const int
48               On entry, NN specifies  the  local  number  of columns  of  the
49               trailing   submatrix   to  be swapped and broadcast starting at
50               the current position. NN must be at least zero.
51

SEE ALSO

53       HPL_pdgesv (3),   HPL_pdgesvK2 (3),    HPL_pdupdateNN (3),    HPL_pdup‐
54       dateTN (3),     HPL_pipid (3),    HPL_plindx1 (3),    HPL_plindx10 (3),
55       HPL_spreadN (3),   HPL_equil (3),   HPL_rollN (3),   HPL_dlaswp00N (3),
56       HPL_dlaswp01N (3), HPL_dlaswp06N (3).
57
58
59
60HPL 2.1                        October 26, 2012              HPL_pdlaswp01N(3)
Impressum