Skip to content

GitLab

Switch branch/tag
  1. 30 Jun, 2014 3 commits
  3. 29 Jun, 2014 3 commits
  5. 28 Jun, 2014 1 commit
  7. 27 Jun, 2014 1 commit
    • Wesley Bland's avatar
      Fix redscat MPI_IN_PLACE checking · be899ea1
      Wesley Bland authored 
      

The MPI_Reduce_scatter(_block) test was incorrectly checking the result
and/or printing out incorrect values when detecting an error. This corrects
that situation. Thanks for the patch from Jeff Hammond.

Fixes #2087
Signed-off-by: Sangmin Seo's avatarSangmin Seo <sseo@anl.gov>
      be899ea1
  9. 26 Jun, 2014 5 commits
  11. 24 Jun, 2014 1 commit
  13. 23 Jun, 2014 1 commit
  15. 20 Jun, 2014 4 commits
    • Rob Latham's avatar
      Lustre used constants w/o header file · 034b37d1
      Rob Latham authored 
      If we are going to use UINT_MAX and USHRT_MAX, something needs to
bring in limits.h.  Martin Audet <Martin.Audet@nrc-cnrc.gc.ca> reported
a build failure because we did not include limits.h.  In addition to his
one-line patch I threw in a configure-time check just in case.
      034b37d1
    • Sangmin Seo's avatar
      Removed a unnecessary MPIU_Assert. · bfdad822
      Sangmin Seo authored and Rob Latham's avatar Rob Latham committed 
      

MPIU_Assert(rreq->dev.iov_offset >= 0) generates warning about
comparison which is always true. This happens because the type of
iov_offset is size_t.
Signed-off-by: Rob Latham's avatarRob Latham <robl@mcs.anl.gov>
      bfdad822
    • Sangmin Seo's avatar
      Replaced -Wno-type-limits with -Wtype-limits. · 2b63c241
      Sangmin Seo authored and Rob Latham's avatar Rob Latham committed 
      

-Wno-type-limits was removed because it masks important failures
(see ticket #2094). However, since Intel compiler currently does
not include -Wtype-limits in -Wextra, -Wtype-limits was added to
handle warnings with the Intel compiler.
Signed-off-by: Rob Latham's avatarRob Latham <robl@mcs.anl.gov>
      2b63c241
    • Huiwei Lu's avatar
      Improve checking for C compiler options on Solaris · f8c8d5aa
      Huiwei Lu authored and Kenneth Raffenetti's avatar Kenneth Raffenetti committed 
      

When using --enable-strict, configure tests for support of certain flags
with the selected compiler. PAC_C_CHECK_COMPILER_OPTION (in
confdb/aclocal_cc.m4) should be able to detect that the
"-D_POSIX_C_SOURCE=200112L" flag is not accepted by the Solaris compiler,
but it does not. The method used (AC_LANG_PROGRAM() in confdb/aclocal_cc.m4:48)
uses an empty parameter, the result of which is an empty main that somehow
passes the detection. Replacing it with a "hello world" test case will
correctly detect the error and remove "-D_POSIX_C_SOURCE=200112L" from the
accepted flags. Configure can then complete successfully.

Note that compilation in this configuration will still freeze due to a
solaris compiler bug. See #2105.
Signed-off-by: Kenneth Raffenetti's avatarKen Raffenetti <raffenet@mcs.anl.gov>
      f8c8d5aa
  17. 19 Jun, 2014 7 commits
  19. 17 Jun, 2014 2 commits
    • Huiwei Lu's avatar
      Clean up MPI_Comm_idup code for '--enable-threads=single' case. · eb8837de
      Huiwei Lu authored and Pavan Balaji's avatar Pavan Balaji committed 
      This patch clean the code for the MPI_Comm_idup single-threaded case to
use the same code of the multi-threaded case. In order to do this, three
variables are made visible to both single-threaded and multi-threaded
case: mask_in_use, lowestContextId and lowestTag.

Also, before this patch, #1935 has been fixed for multi-threaded case
but will still fail for single-threaded case. After this patch, the
single-threaded case will be fixed.

Here is a history of related changes: The patch [05eeccb5] fixed
MPI_Comm_idup for multi-threaded case, but forgot to fix the
single-threaded case. A temporary patch [18de1dbe

] is used to revert
the single-threaded case to use old code before multi-thread patch.
The current patch fixes the second patch.

See also #1935.
Signed-off-by: Pavan Balaji's avatarPavan Balaji <balaji@anl.gov>
      eb8837de
    • Kenneth Raffenetti's avatar
      fix assert in win_shared_noncontig tests · 54e9830b
      Kenneth Raffenetti authored 
      

A correct MPI-3 implementation may allocate more memory than requested
when calling MPI_Win_allocate_shared with alloc_shared_noncontig
specified (MPI-3, p409, l13-16). The testsuite incorrectly asserts
that the size of an allocated segment is exactly as requested. This
is fixed by changing the comparison operators from == to >=.

Discussion of this test error originated on OpenMPI ticket #4721.
Signed-off-by: default avatarJunchao Zhang <jczhang@mcs.anl.gov>
      54e9830b
  21. 16 Jun, 2014 1 commit
    • Wesley Bland's avatar
      Add a CVAR to turn off collective alias checking · d505ece8
      Wesley Bland authored 
      Add a new CVAR (MPIR_CVAR_COLL_ALIAS_CHECK) that allows the user to turn off
alias checking in collective operations. In 5b4bdef9

, alias checking was
expanded to be more correct according to the standard, but will cause some
previously misbehaving applications to fail even when they might not have
displayed that behavior before. This patch lets users turn off checking if
they accept the risk and don't want this specific check to be done. All error
checking can still be turned off at configure time as usual. By default, this
new CVAR will be set to `1` to indicate that checking is on.

Fixes #2117
Signed-off-by: default avatarAntonio J. Pena <apenya@mcs.anl.gov>
      d505ece8
  23. 12 Jun, 2014 2 commits
  25. 11 Jun, 2014 2 commits
    • Antonio J. Pena's avatar
      Switched -D_REENTRANT from CFLAGS to CPPFLAGS · 1d82f922
      Antonio J. Pena authored 
      Hotfix for commit be93caec. Thanks to Dave Goodell to bring this out.

See #2107
      1d82f922
    • Antonio J. Pena's avatar
      Added support for thread-safe errno in Solaris · be93caec
      Antonio J. Pena authored 
      

The _REENTRANT macro is required to be defined in order to get
thread-safe "errno" values. Modern POSIX systems provide this by default
when using -lpthreads, but there are still some (Solaris) that require
an explicit define. Since we weren't compiling with that macro, we were
getting incorrect "errno" values from the threads (mainly 0).

More details: since we use nonblocking sockets, our code was checking
"errno" to not be EINPROGRESS whenever we got an error code from a
connect/accept, in order to determine an actual error (a return value
of -1 and errno of EINPROGRESS does not define an actual error but a
"not yet" when using nonblocking sockets).

This patch automatically gets this macro defined by configure whenever
we are not building in MPI_THREAD_SINGLE mode only.

Fixes #2107
Signed-off-by: Kenneth Raffenetti's avatarKen Raffenetti <raffenet@mcs.anl.gov>
      be93caec
  27. 10 Jun, 2014 4 commits
  29. 06 Jun, 2014 1 commit
    • Michael Blocksome's avatar
      pamid: remove blocking shmem mutex; remove shmem CAS/FOP optimizations · e6ddea13
      Michael Blocksome authored 
      The blocking pthread mutex in the shared memory window causes a deadlock
on bgq - perhaps because the messaging state is not advanced while the
thread is waiting for the mutex to be released.

Removing this blocking mutex resolves the bgq failures:

 - rma/strided_putget_indexed_shared
 - rma/strided_getacc_indexed_shared

With the calls to the blocking mutex removed the CAS and FOP functions
are not atomic. Solution is to remove the shared memory optimization and
instead use the common (network) code path.

Removing these shared memory optimizations from CAS/FOP resolves the
bgq hangs:

 - rma/mutex_bench_shared
      e6ddea13
  31. 05 Jun, 2014 2 commits