Re: [gomp] Support statically scheduled parallel loops

[Diego Novillo <dnovillo at redhat dot com>]

On 09/20/05 17:42, Andrew Pinski wrote:

> How this interact with C99 VLAs or alloca?

It doesn't.

> Also how does GOMP interact with nested functions?

It doesn't.

>     #pragma omp for shared (a), private (i)
>     for (int i = 0; i < N; i++)
>       a[i] = ...

You don't need the private marker for 'i'.  Loop controlling variables 
are implicitly private.

> Those seems like places where GOMP can go wrong.

It's all designed to work.  The language features that you mention do 
not affect the meaning of the data sharing and copying directives.  The 
implementation is still very rough, so you will find places where we 
fail and shouldn't.  But something like this should work today:

#include <omp.h>

main()
{
 void foo (int n)
   {
     int i;
     int c[n];
     char *x = alloca (10);

     strcpy (x, "xxx");

#pragma omp parallel shared (c,x) private (i)
       {
#pragma omp for
         for (i = 0; i < 80; i++)
           printf ("Thread %d, x = %s, i = %d\n", omp_get_thread_num 
(), x, i);
       }
   }


 foo (80);
}