Pavan Balaji wrote:
> This is the latest version of the PGI compiler. If this is indeed
> broken, then it'll be ideal if the configure can detect it and basically
> disable everything. That is, hwloc_topology_init can return a failure,
> so packages relying on hwloc can do something else.
>
We don't have pgcc, so we can't really really/test such a detection
code. Maybe AC_COMPILE_IFELSE or AC_LINK_IFELSE with a code like below
would be enough:
#define _GNU_SOURCE 1
#include <sched.h>
void foo(cpu_set_t *set) {
CPU_SET(0, set);
}
If such a basic code doesn't compile fine with pgcc, it's worth a
bug-report to pgi...
> FWIW, I don't have this problem with plpa.
>
PLPA seems to be redefining its own PLPA_CPU_* so it's much easier to
avoid wrong redefinitions :)
Brice
