By the way, I recently saw results for OpenMPI performance that were really
bad on infiniband, at least on one setup.
Regards - Bob
----- Original Message -----
From: "David A. Case" <case.scripps.edu>
To: <amber.scripps.edu>
Sent: Wednesday, July 02, 2008 6:58 PM
Subject: Re: AMBER: Inconsistent types (INTEGER(4)/REAL(8)) - Is it serious
?
> On Wed, Jul 02, 2008, David A. Case wrote:
>
>> >
>> > ============================================================
>> > In file nmr_calls.f90:285
>> >
>> > call mpi_bcast(nmr_iwork, intreq, mpi_integer, &
>> > 1
>> > In file nmr_calls.f90:273
>> >
>> > call mpi_bcast(rwell, nmr_dat_dbl_cnt, mpi_double_precision, 0, &
>> > 2
>> > Warning (155): Inconsistent types (INTEGER(4)/REAL(8)) in actual
>> > argument lists at (1) and (2)
>> >
>
> Here's my reasons I said not to worry: the mpi_bcast() function is the
> same
> call, no matter whether integers or reals are being broadcast. The third
> argument (mpi_integer or mpi_double_precision above) gives the
> size and type of the data being transferred.
>
> In the C language binding, the first argument of mpi_bcast is void*, which
> essentially says "don't worry about what type of data this is."
>
> There is no simple equivalent in Fortran. If you want to broadcast both
> reals and integers in the same subroutine, you have to do it as written
> above,
> and a picky compiler will issue warnings (only!) like the ones you report.
>
> In my view, this has nothing to do with the fact that this is in NMR, nor
> is
> there any real way to avoid it. G95 puts out warnings like this all over
> the
> place. Until shown otherwise, I will stand by my original advice to
> ignore
> this sort of warning.
>
> ...dac
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber.scripps.edu
> To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
> to majordomo.scripps.edu
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
to majordomo.scripps.edu
Received on Sun Jul 06 2008 - 06:07:21 PDT
