[Beowulf] Stupid MPI programming question

[Jakob Oestergaard]

On Thu, Sep 28, 2006 at 01:47:17AM +0100, Clements, Brent M (SAIC) wrote:
> Hey Guys,
...
>  
> This is what I have in my code(some things have been removed)
>  
> #define MASTER_RANK 0
>  
> char* mystring;

So mystring is a pointer to character(s).

>  
> mystring = "some test";

Now mystring points to a statically allocated string "some test".

You must not overwrite "some test" with other data - the data area is
assumed to be constant and compilers are even free to merge such
strings. For example:

char *a = "some fancy test";
char *b = "fancy test";

This can result in the following memory layout

"some fancy test"
 ^    ^
 a    |
      b

Imagine what would happen to *a if you changed the contents of *b...

I don't know if your compiler does this - but it is just one example of
what *could* go wrong.

My current gcc will place the strings in a read-only part of the program
memory. This means, the following code will segfault:

---
char *foo = "my test"; // 8 bytes in possibly read-only memory
...
foo[0] = ' ';          //    <- blamo! segfault
---


What you probably want to do is:
---
char foo[1024];          // 1KiB on the stack - writable
strncpy(foo, sizeof foo, "my test"); // Assign contents by copying
...
foo[0] = ' ';             // <- fine
---


>  
> MPI_Bcast(&mystring, sizeof(basedir), MPI_CHAR, MASTER_RANK, MPI_COMM_WORLD);

It's really odd to me that MPI_Bcast with MPI_CHAR takes the *address*
of the *pointer* rather than the address of the data that needs to be
sent... But I looked up another example and it does look like that's
really what you need to do. Odd...

So, anyway, I think what you want to do is:

char foo[1024];
strncpy(foo, sizeof foo, "my test");
MPI_Bcast(&foo, sizeof foo, MPI_CHAR, ...);

...
>  
> Am I doing something wrong? Or if someone has some sample code they can point me to on how to send a string to a slave mpi process and have that slave mpi process print out the string..that would be even better. That way..I relearn MPI basics 101

Looks like the problem is with C not MPI  - unless of course I've completely missed something  ;)

-- 

 / jakob