Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: [OMPI devel] 1.4.5rc2 linux/ppc/xlc-{7,8,9} [3 PASS, 3 FAIL]
From: Paul H. Hargrove (PHHargrove_at_[hidden])
Date: 2012-01-30 00:28:59

While I previously reported success with xlc-11.1 on Linux for both
ppc32 and ppc64 ABIs, I find that all is not perfect with earlier compilers.

   xlc-9.0 / xlf-11.1
       PASS on ILP32 ABI
       FAIL on LP64 ABI (details below)
   xlc-8.0 / xlf-10.1
       PASS on ILP32 ABI
       PASS on LP64 ABI
   xlc-7.0 / xlf-9.1
       FAIL on ILP32 ABI
       FAIL on LP64 ABI

I don't seem to have any access to an xlc-10.x release.

For the xlc-9.0 compiler, "make check" is failing one of the atomics
tests, but only for ppc64 ABI (ppc32 is fine):
> --> Testing atomic_spinlock
> /home/hargrove/OMPI/openmpi-1.4.5rc2-linux-ppc64-xlc-9.0//openmpi-1.4.5rc2/test/asm/run_tests:
> line 8: 27655 Segmentation fault $* $threads
> - 1 threads: Failed
> /home/hargrove/OMPI/openmpi-1.4.5rc2-linux-ppc64-xlc-9.0//openmpi-1.4.5rc2/test/asm/run_tests:
> line 8: 27657 Segmentation fault $* $threads
> - 2 threads: Failed
> /home/hargrove/OMPI/openmpi-1.4.5rc2-linux-ppc64-xlc-9.0//openmpi-1.4.5rc2/test/asm/run_tests:
> line 8: 27659 Segmentation fault $* $threads
> - 4 threads: Failed
> /home/hargrove/OMPI/openmpi-1.4.5rc2-linux-ppc64-xlc-9.0//openmpi-1.4.5rc2/test/asm/run_tests:
> line 8: 27661 Segmentation fault $* $threads
> - 5 threads: Failed
> /home/hargrove/OMPI/openmpi-1.4.5rc2-linux-ppc64-xlc-9.0//openmpi-1.4.5rc2/test/asm/run_tests:
> line 8: 27663 Segmentation fault $* $threads
> - 8 threads: Failed
> FAIL: atomic_spinlock

The xlc-7.0 release I have access to could not compile asm.c (says
"memory" and "cc" are unknown registers) and complained that
atomic-asm.S had an invalid suffix. So, this compiler is pretty much

Since the 11.1 compilers work, I don't see the failure w/ xlc-9.0/LP64
as a serious issue.
However, it might be worth noting the in the compiler-specific section
of the README.
The total uselessness of the xlc-7.0 compiler is not likely to impact
too many users, but could still be noted.

Though I doubt it is related to the xlc-9/ppc64 failure, I should
mention for completeness that the xlc tests were on 3 different CPUs:
    11.1 POWER7 (IBM pSeries 8233-E8B) [reported in a previous email]
     9.0 PPC970MP (IBM pSeries 8844-AC1)
     8.0 POWER6 (IBM pSeries 9124-720)
     7.0 POWER6 (IBM pSeries 9124-720)

I also had to disable the opal_path_nfs test again on the POWER6
machine, even w/ the linux->__linux__ change.
I will report on that when/if I can determine the cause.


On 1/19/2012 9:55 AM, Jeff Squyres wrote:
> Please test:

Paul H. HargrovePHHargrove_at_[hidden]
Future Technologies Group
HPC Research Department                   Tel: +1-510-495-2352
Lawrence Berkeley National Laboratory     Fax: +1-510-486-6900