Groups | Search | Server Info | Login | Register

Groups > comp.parallel.mpi > #59

Re: MPI_Allreduce on structure of doubles. Portable?

Show all headers | View raw

Am 05.02.2013, 17:30 Uhr, schrieb <jacolagwa@gmail.com>:

> Hello,
>
>
> Consider something like this:
>
> ---
>
> typedef struct TS {
>     double a,b,c;
> } S;
>
> ...
> S x,y;
> ...
> MPI_Allreduce(&x, &y, 3, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
>
> ---
>
> Is the above code completely portable (without using MPI_Type_struct and  
> all)?
>
> Thanks!
> JS
>

Since I am 5 months late with the answer this is just for the records, I'm  
afraid.

Since any ANSI C compiler must be able to pack doubles tight in arrays it  
will pack doubles tight in structs too. And of course we expect the  
struct's address to be the same as the first field's. If these assumptions  
are not met, then we are not talking about an implementation of ANSI C as  
I understood it.

So under all forseeable and sane conditions the code is portable. If not,  
then it is a fault in the target of the port. ;-)

BUT: you may start all the hassle of defining your MPI datatypes just to  
free your mind when laying out your data structures.

Another way to think about it: when you need binary serialization for you  
data structures to store them in files you may consider abusing this  
serialization for messages.

ciao
Georg

Back to comp.parallel.mpi | Previous | Next — Previous in thread | Find similar

Thread

MPI_Allreduce on structure of doubles. Portable? jacolagwa@gmail.com - 2013-02-05 08:30 -0800
  Re: MPI_Allreduce on structure of doubles. Portable? "Georg Bisseling" <spam@bisseling.de> - 2013-07-07 11:03 +0200

csiph-web