Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

From: Brian Barrett (brbarret_at_[hidden])
Date: 2005-09-12 13:34:37


Ok, I see what's happening, although I'm not sure the two problems
are actually related. The first is that the component to provide
high resolution timer support on Linux is disabling itself because:

   1) it doesn't know how to figure out the clock rate of the CPU
   2) there's no assembly for reading a CPU counter on SPARC chips

The only reliable way to get CPU frequency is reading /proc/cpuinfo,
and for Linux, each architecture seems to have a different format.
So that part's covered with the information provided below. Now I
just need to figure out how to get cycle counts out of a SPARC. So
much easier on Solaris ;).

Brian

On Sep 12, 2005, at 1:05 PM, Ferris McCormick wrote:

> On Mon, 2005-09-12 at 11:14 -0500, Brian Barrett wrote:
>
>> Thanks for the heads up. We are not seeing this on other platforms,
>> so it might be a Sparc-specific issue. Any chance you could compile
>> with debugging symbols and generate a backtrace? Also, could you
>> send the contents of /proc/cpuinfo (long story...)?
>>
>> Thanks!
>>
>> Brian
>>
>>
> Here's /proc/cpuinfo from the SB1000:
> =====================================
> fmccor_at_polylepis AGT [93]% cat /proc/cpuinfo
> cpu : TI UltraSparc III (Cheetah)
> fpu : UltraSparc III integrated FPU
> promlib : Version 3 Revision 13
> prom : 4.13.0
> type : sun4u
> ncpus probed : 2
> ncpus active : 2
> Cpu0Bogo : 598.01
> Cpu0ClkTck : 0000000035a4e900
> Cpu1Bogo : 598.01
> Cpu1ClkTck : 0000000035a4e900
> MMU Type : Cheetah
> State:
> CPU0: online
> CPU1: online
> ========================================
>
> And here's a back-trace from ompi_info:
> ==========================================
> Program received signal SIGSEGV, Segmentation fault.
> opal_output_close (output_id=1880710872) at opal_object.h:409
> 409 for (i = 0; i < cls->cls_depth; i++) {
> Current language: auto; currently c
> (gdb) bt
> #0 opal_output_close (output_id=1880710872) at opal_object.h:4
> #1 0x700d8e00 in mca_topo_base_close () at topo_base_close.c:46
> #2 0x00016aa4 in close_components () at components.cc:254
> #3 0x00018bbc in main (argc=1, argv=0xefa253f4) at ompi_info.cc:251
> =============================================
> HOWEVER: If I configure with --enable-debug, two things happen:
> 1. I have to build ompi/mca/rcache/rb by hand because of incorrect
> CFLAGS;
> 2. The SegFault disappears.
>
> (The line# in #0 above is incorrect; by accident I edited the email
> as I
> was writing it and erased too much. I can rebuild with '-g' but not
> with --enable-debug if necessary.)
> ====================================
> Other failing system:
> fmccor_at_lacewing openmpi-1.0a1r7305 [96]% cat /proc/cpuinfo
> cpu : TI UltraSparc II (BlackBird)
> fpu : UltraSparc II integrated FPU
> promlib : Version 3 Revision 19
> prom : 3.19.0
> type : sun4u
> ncpus probed : 2
> ncpus active : 2
> Cpu0Bogo : 799.53
> Cpu0ClkTck : 0000000017d746a8
> Cpu1Bogo : 799.53
> Cpu1ClkTck : 0000000017d746a8
> MMU Type : Spitfire
> State:
> CPU0: online
> CPU1: online
> ===========================
> Regards,
>
>
> --
> Ferris McCormick (P44646, MI) <fmccor_at_[hidden]>
> Developer, Gentoo Linux (Sparc, Devrel)
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel