Re: [bgl-discuss] NAN for mpi_wtime() ?

[Anthony Chan <chan@xxxxxxxxxxx>]

I think this timer bug is there ever since BGL arrives, a simple
java timing code on the front-end node has similar problem.

clog2TOslog2 -o tmp.slog2 tmp.clog2
.....
timeElapsed between 1 & 2 = 105 msec
timeElapsed between 2 & 3 = -1130612245368 msec

It has something to do with how long the 2nd timer runs, if it is
short enough.  It will work fine.

But the MPI_Wtime() on the compute nodes seems to work fine,
at least from MPE's respective.

A.Chan

On Sat, 29 Oct 2005, Paul Fischer wrote:

>
> Does anyone have a clue why I should get NAN as a return
> result for mpi_wtime ?
>
> I have the following code fragment:
>
>       t0 = mpi_wtime()
>       gs_handle = fgs_init(glo_num,ntot,np)
>       t1 = mpi_wtime()
>       et = t1-t0
>       if (nid.eq.0) then
>          write(6,1) et,t0,t1,nx,ntot,ngv,np,gs_handle
>     1    format('fgs_init A:',1p3e11.4,' seconds ',i4,4i10)
>       endif
>
> which produces:
>
>  fgs_init A:   -NAN 1.1306E+09   -NAN seconds    2       960   16320 128 2
>  fgs_init A:   -NAN 1.1306E+09   -NAN seconds    4      7680  423360 128 3
>  fgs_init A:   -NAN 1.1306E+09   -NAN seconds    6     25920 1944000 128 4
>  fgs_init A:   -NAN 1.1306E+09   -NAN seconds    8     61440 5315520 128 5
>
> There is no problem w/ the first mpi_wtime() call, only w/ the second??
>
> Thanks -- Paul
>
>
> - --------------------------------------------------------------------
> To add or remove yourself from this mailing list, use the 'notifyme'
> command on any BGL machine. To remove: notifyme -n, to add: notifyme -y.
>
>

- --------------------------------------------------------------------
To add or remove yourself from this mailing list, use the 'notifyme'
command on any BGL machine. To remove: notifyme -n, to add: notifyme -y.