Hardware Locality Development Mailing List Archives

Subject: Re: [hwloc-devel] [hwloc-svn] svn:hwloc r1940
From: Bert Wesarg (bert.wesarg_at_[hidden])
Date: 2010-04-06 12:59:29

• Next message: Samuel Thibault: "Re: [hwloc-devel] Embedding: is it worth it?"
• Previous message: Jeff Squyres: "Re: [hwloc-devel] rc1?"

On Tue, Apr 6, 2010 at 17:36, <bgoglin_at_[hidden]> wrote:
> Author: bgoglin
> Date: 2010-04-06 11:36:17 EDT (Tue, 06 Apr 2010)
> New Revision: 1940
> URL: https://svn.open-mpi.org/trac/hwloc/changeset/1940
>
> Log:
> Stop using HWLOC_NBMAXCPUS in Linux hwloc_linux_set/get_tid_cpubind
> Text files modified:
>   branches/dyncpusets/src/topology-linux.c |    31 ++++++++++++++++++++++---------
>   1 files changed, 22 insertions(+), 9 deletions(-)
>
> Modified: branches/dyncpusets/src/topology-linux.c
> ==============================================================================
> --- branches/dyncpusets/src/topology-linux.c    (original)
> +++ branches/dyncpusets/src/topology-linux.c    2010-04-06 11:36:17 EDT (Tue, 06 Apr 2010)
> @@ -192,8 +192,6 @@
>  #endif
>  }
>
> -#define HWLOC_NBMAXCPUS 1024 /* FIXME: drop */
> -
>  int
>  hwloc_linux_set_tid_cpubind(hwloc_topology_t topology __hwloc_attribute_unused, pid_t tid, hwloc_const_cpuset_t hwloc_set)
>  {
> @@ -205,13 +203,21 @@
>
>   /* The resulting binding is always strict */
>
> -#if defined(HWLOC_HAVE_CPU_SET_S) && !defined(HWLOC_HAVE_OLD_SCHED_SETAFFINITY) && CPU_SETSIZE < HWLOC_NBMAXCPUS
> +#if defined(HWLOC_HAVE_CPU_SET_S) && !defined(HWLOC_HAVE_OLD_SCHED_SETAFFINITY)
>   cpu_set_t *plinux_set;
>   unsigned cpu;
> -  size_t setsize = CPU_ALLOC_SIZE(HWLOC_NBMAXCPUS);
> +  int last;
> +  size_t setsize;
>   int err;
>
> -  plinux_set = CPU_ALLOC(HWLOC_NBMAXCPUS);
> +  last = hwloc_cpuset_last(hwloc_set);
> +  if (last == -1) {
> +    errno = -EINVAL;
> +    return -1;
> +  }
> +
> +  setsize = CPU_ALLOC_SIZE(last+1);
> +  plinux_set = CPU_ALLOC(last+1);
>
>   CPU_ZERO_S(setsize, plinux_set);
>   hwloc_cpuset_foreach_begin(cpu, hwloc_set)
> @@ -253,12 +259,17 @@
>   int err;
>   /* TODO Kerrighed */
>
> -#if defined(HWLOC_HAVE_CPU_SET_S) && !defined(HWLOC_HAVE_OLD_SCHED_SETAFFINITY) && CPU_SETSIZE < HWLOC_NBMAXCPUS
> +#if defined(HWLOC_HAVE_CPU_SET_S) && !defined(HWLOC_HAVE_OLD_SCHED_SETAFFINITY)
>   cpu_set_t *plinux_set;
>   unsigned cpu;
> -  size_t setsize = CPU_ALLOC_SIZE(HWLOC_NBMAXCPUS);
> +  int last;
> +  size_t setsize;
> +
> +  last = hwloc_cpuset_last(topology->levels[0][0]->complete_cpuset);
> +  assert(last != -1);
>
> -  plinux_set = CPU_ALLOC(HWLOC_NBMAXCPUS);
> +  setsize = CPU_ALLOC_SIZE(last+1);
> +  plinux_set = CPU_ALLOC(last+1);
>
>   err = sched_getaffinity(tid, setsize, plinux_set);
>
> @@ -268,7 +279,7 @@
>   }
>
>   hwloc_cpuset_zero(hwloc_set);
> -  for(cpu=0; cpu<HWLOC_NBMAXCPUS; cpu++)
> +  for(cpu=0; cpu<last; cpu++)

Shouldn't this be "cpu<=last"?

Bert

• Next message: Samuel Thibault: "Re: [hwloc-devel] Embedding: is it worth it?"
• Previous message: Jeff Squyres: "Re: [hwloc-devel] rc1?"