root / www / HPL_pdlaswp01T.html
Historique | Voir | Annoter | Télécharger (2,92 ko)
| 1 |
<HTML>
|
|---|---|
| 2 |
<HEAD>
|
| 3 |
<TITLE>HPL_pdlaswp01T HPL 2.0 Library Functions September 10, 2008</TITLE> |
| 4 |
</HEAD>
|
| 5 |
|
| 6 |
<BODY BGCOLOR="WHITE" TEXT = "#000000" LINK = "#0000ff" VLINK = "#000099" |
| 7 |
ALINK = "#ffff00"> |
| 8 |
|
| 9 |
<H1>Name</H1> |
| 10 |
<B>HPL_pdlaswp01T</B> Broadcast a column panel L and swap the row panel U. |
| 11 |
|
| 12 |
<H1>Synopsis</H1> |
| 13 |
<CODE>#include "hpl.h"</CODE><BR><BR> |
| 14 |
<CODE>void</CODE> |
| 15 |
<CODE>HPL_pdlaswp01T(</CODE> |
| 16 |
<CODE>HPL_T_panel *</CODE> |
| 17 |
<CODE>PBCST</CODE>, |
| 18 |
<CODE>int *</CODE> |
| 19 |
<CODE>IFLAG</CODE>, |
| 20 |
<CODE>HPL_T_panel *</CODE> |
| 21 |
<CODE>PANEL</CODE>, |
| 22 |
<CODE>const int</CODE> |
| 23 |
<CODE>NN</CODE> |
| 24 |
<CODE>);</CODE> |
| 25 |
|
| 26 |
<H1>Description</H1> |
| 27 |
<B>HPL_pdlaswp01T</B> |
| 28 |
applies the NB row interchanges to NN columns of the |
| 29 |
trailing submatrix and broadcast a column panel. |
| 30 |
|
| 31 |
A "Spread then roll" algorithm performs the swap :: broadcast of the |
| 32 |
row panel U at once, resulting in a minimal communication volume and |
| 33 |
a "very good" use of the connectivity if available. With P process |
| 34 |
rows and assuming bi-directional links, the running time of this |
| 35 |
function can be approximated by: |
| 36 |
|
| 37 |
(log_2(P)+(P-1)) * lat + K * NB * LocQ(N) / bdwth |
| 38 |
|
| 39 |
where NB is the number of rows of the row panel U, N is the global |
| 40 |
number of columns being updated, lat and bdwth are the latency and |
| 41 |
bandwidth of the network for double precision real words. K is |
| 42 |
a constant in (2,3] that depends on the achieved bandwidth during a |
| 43 |
simultaneous message exchange between two processes. An empirical |
| 44 |
optimistic value of K is typically 2.4. |
| 45 |
|
| 46 |
<H1>Arguments</H1> |
| 47 |
<PRE>
|
| 48 |
PBCST (local input/output) HPL_T_panel * |
| 49 |
On entry, PBCST points to the data structure containing the |
| 50 |
panel (to be broadcast) information. |
| 51 |
</PRE>
|
| 52 |
<PRE>
|
| 53 |
IFLAG (local input/output) int * |
| 54 |
On entry, IFLAG indicates whether or not the broadcast has |
| 55 |
already been completed. If not, probing will occur, and the |
| 56 |
outcome will be contained in IFLAG on exit. |
| 57 |
</PRE>
|
| 58 |
<PRE>
|
| 59 |
PANEL (local input/output) HPL_T_panel * |
| 60 |
On entry, PANEL points to the data structure containing the |
| 61 |
panel information. |
| 62 |
</PRE>
|
| 63 |
<PRE>
|
| 64 |
NN (local input) const int |
| 65 |
On entry, NN specifies the local number of columns of the |
| 66 |
trailing submatrix to be swapped and broadcast starting at |
| 67 |
the current position. NN must be at least zero. |
| 68 |
</PRE>
|
| 69 |
|
| 70 |
<H1>See Also</H1> |
| 71 |
<A HREF="HPL_pdgesv.html">HPL_pdgesv</A>, |
| 72 |
<A HREF="HPL_pdgesvK2.html">HPL_pdgesvK2</A>, |
| 73 |
<A HREF="HPL_pdupdateNT.html">HPL_pdupdateNT</A>, |
| 74 |
<A HREF="HPL_pdupdateTT.html">HPL_pdupdateTT</A>, |
| 75 |
<A HREF="HPL_pipid.html">HPL_pipid</A>, |
| 76 |
<A HREF="HPL_plindx1.html">HPL_plindx1</A>, |
| 77 |
<A HREF="HPL_plindx10.html">HPL_plindx10</A>, |
| 78 |
<A HREF="HPL_spreadT.html">HPL_spreadT</A>, |
| 79 |
<A HREF="HPL_equil.html">HPL_equil</A>, |
| 80 |
<A HREF="HPL_rollT.html">HPL_rollT</A>, |
| 81 |
<A HREF="HPL_dlaswp10N.html">HPL_dlaswp10N</A>, |
| 82 |
<A HREF="HPL_dlaswp01T.html">HPL_dlaswp01T</A>, |
| 83 |
<A HREF="HPL_dlaswp06T.html">HPL_dlaswp06T</A>. |
| 84 |
|
| 85 |
</BODY>
|
| 86 |
</HTML>
|