root / man / man3 / HPL_pdlaswp00N.3
Historique | Voir | Annoter | Télécharger (2,13 ko)
| 1 |
.TH HPL_pdlaswp00N 3 "September 10, 2008" "HPL 2.0" "HPL Library Functions" |
|---|---|
| 2 |
.SH NAME |
| 3 |
HPL_pdlaswp00N \- Broadcast a column panel L and swap the row panel U. |
| 4 |
.SH SYNOPSIS |
| 5 |
\fB\&#include "hpl.h"\fR |
| 6 |
|
| 7 |
\fB\&void\fR |
| 8 |
\fB\&HPL_pdlaswp00N(\fR |
| 9 |
\fB\&HPL_T_panel *\fR |
| 10 |
\fI\&PBCST\fR, |
| 11 |
\fB\&int *\fR |
| 12 |
\fI\&IFLAG\fR, |
| 13 |
\fB\&HPL_T_panel *\fR |
| 14 |
\fI\&PANEL\fR, |
| 15 |
\fB\&const int\fR |
| 16 |
\fI\&NN\fR |
| 17 |
\fB\&);\fR |
| 18 |
.SH DESCRIPTION |
| 19 |
\fB\&HPL_pdlaswp00N\fR |
| 20 |
applies the NB row interchanges to NN columns of the |
| 21 |
trailing submatrix and broadcast a column panel. |
| 22 |
|
| 23 |
Bi-directional exchange is used to perform the swap :: broadcast of |
| 24 |
the row panel U at once, resulting in a lower number of messages than |
| 25 |
usual as well as a lower communication volume. With P process rows and |
| 26 |
assuming bi-directional links, the running time of this function can |
| 27 |
be approximated by: |
| 28 |
|
| 29 |
log_2(P) * (lat + NB*LocQ(N) / bdwth) |
| 30 |
|
| 31 |
where NB is the number of rows of the row panel U, N is the global |
| 32 |
number of columns being updated, lat and bdwth are the latency and |
| 33 |
bandwidth of the network for double precision real words. Mono |
| 34 |
directional links will double this communication cost. |
| 35 |
.SH ARGUMENTS |
| 36 |
.TP 8 |
| 37 |
PBCST (local input/output) HPL_T_panel * |
| 38 |
On entry, PBCST points to the data structure containing the |
| 39 |
panel (to be broadcast) information. |
| 40 |
.TP 8 |
| 41 |
IFLAG (local input/output) int * |
| 42 |
On entry, IFLAG indicates whether or not the broadcast has |
| 43 |
already been completed. If not, probing will occur, and the |
| 44 |
outcome will be contained in IFLAG on exit. |
| 45 |
.TP 8 |
| 46 |
PANEL (local input/output) HPL_T_panel * |
| 47 |
On entry, PANEL points to the data structure containing the |
| 48 |
panel (to be broadcast and swapped) information. |
| 49 |
.TP 8 |
| 50 |
NN (local input) const int |
| 51 |
On entry, NN specifies the local number of columns of the |
| 52 |
trailing submatrix to be swapped and broadcast starting at |
| 53 |
the current position. NN must be at least zero. |
| 54 |
.SH SEE ALSO |
| 55 |
.BR HPL_pdgesv \ (3), |
| 56 |
.BR HPL_pdgesvK2 \ (3), |
| 57 |
.BR HPL_pdupdateNN \ (3), |
| 58 |
.BR HPL_pdupdateTN \ (3), |
| 59 |
.BR HPL_pipid \ (3), |
| 60 |
.BR HPL_plindx0 \ (3), |
| 61 |
.BR HPL_dlaswp01N \ (3), |
| 62 |
.BR HPL_dlaswp02N \ (3), |
| 63 |
.BR HPL_dlaswp03N \ (3), |
| 64 |
.BR HPL_dlaswp04N \ (3), |
| 65 |
.BR HPL_dlaswp05N \ (3). |