HPL_pdlaswp01N - Man Page

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

HPL_pdlaswp01N applies the  NB  row interchanges to  NN columns of the trailing submatrix and broadcast a column panel.
 A "Spread then roll" algorithm performs  the swap :: broadcast  of the row panel U at once,  resulting in a minimal communication volume  and a "very good"  use of the connectivity if available.  With  P  process rows  and  assuming  bi-directional links,  the  running time  of this function can be approximated by:

  (log_2(P)+(P-1)) * lat +   K * NB * LocQ(N) / bdwth
 where  NB  is the number of rows of the row panel U,  N is the global number of columns being updated,  lat and bdwth  are the latency  and bandwidth  of  the  network  for  double  precision real words.  K is a constant in (2,3] that depends on the achieved bandwidth  during  a simultaneous  message exchange  between two processes.  An  empirical optimistic value of K is typically 2.4.

PBCST   (local input/output)    HPL_T_panel *

On entry,  PBCST  points to the data structure containing the panel (to be broadcast) information.

IFLAG   (local input/output)    int *

On entry, IFLAG  indicates  whether or not  the broadcast has already been completed.  If not,  probing will occur, and the outcome will be contained in IFLAG on exit.

PANEL   (local input/output)    HPL_T_panel *

On entry,  PANEL  points to the data structure containing the panel information.

NN      (local input)           const int

On entry, NN specifies  the  local  number  of columns of the trailing  submatrix  to  be swapped and broadcast starting at the current position. NN must be at least zero.

HPL_pdgesv (3), HPL_pdgesvK2 (3), HPL_pdupdateNN (3), HPL_pdupdateTN (3), HPL_pipid (3), HPL_plindx1 (3), HPL_plindx10 (3), HPL_spreadN (3), HPL_equil (3), HPL_rollN (3), HPL_dlaswp00N (3), HPL_dlaswp01N (3), HPL_dlaswp06N (3).

HPL_equil(3), HPL_logsort(3), HPL_pdupdateNN(3), HPL_pdupdateTN(3), HPL_perm(3), HPL_pipid(3), HPL_plindx0(3), HPL_plindx1(3), HPL_plindx10(3), HPL_rollN(3), HPL_spreadN(3).