Hexagon has updated software/libraries. Users should recompile their application to gain new features and updates.
MPI
xt-mpt 5.1.2 -> 5.1.3
Compilers and debugging
xt-asyncpe 4.5 -> 4.6
Intel compiler 11.1.073 -> 12.0.084
Chapel 1.2.0 -> 1.2.1
ATP 1.0.3 -> 1.1.0
STAT 1.1.1 -> 1.1.2
Performance tools and Math libs
xt-craypat/apprentice2 5.1.2 -> 5.1.3
xt-libsci 10.4.9 -> 10.5.0
NOTES:
xt-mpt
The following features were added to MPT 5.1.3:
- Improvements to MPI-IO collective buffering.
ATP
Abnormal Termination Processing (ATP) is a system that monitors Cray XT
System user applications, and should an application take a system trap,
ATP performs analysis on the dying application. With release 1.1 all of
the stack backtraces of the application processes are gathered into a
merged stack backtrace tree and written to disk as the file
"atpMergedBT.dot". The stack backtrace for the first process to die is
sent to stderr as is the number of the signal that caused the death.
atpMergedBT.dot can be viewed with 'statview', a component of the STAT
package (module load stat). The merged stack backtrace tree provides
a concise, yet comprehensive, view of what the application was doing at
the time of its death.
Further information on ATP can found in the intro_atp man page.
Release notes for release 1.1.0
--------------------------------
1.1.0:
- ATP is now automatically linked in to user applications
and automatically initialized. That is, users do not need
to modify their source code nor their link line (and, in
fact, should not). One must use the Cray compiler drivers
(cc, CC, ftn) to achieve this.
In order for this to occur one must do all of the following:
o have the module atp/1.1.0 or greater loaded (which
is automatically done by the latest PrgEnv modules)
o use the Cray compiler drivers when linking
o relink your application
- It is now necessary to overtly define the environment variable
'ATP_ENABLED' so that the running of an application gets ATP
processing.
- ATP will now perform its analysis in the event of the
queuing system aborting the job due to the wall clock
expiring. Note that the amount of time between when the
queuing system signals the that wall clock has expired
and when the queuing system SIGKILLs the job is
something that sites can customize. If sufficient time
is not configured, this feature may not be able to
complete its task. Thirty seconds is typically more than
generous.
- The environment variable ATP_HOLD_TIME can be used to
define the number of minutes that ATP should hold a dying
application in stasis so that it can be attached via
a debugger.
- ATP is now willing to collect data, even if some nodes
have stopped responding. Since such a system is clearly
sick in some manner, this may not always be successful.
- Fixed a memory corruption bug that could cause various,
unpredictable
symptoms.
- ATP no longer needs be installed on a cross mounted location
xt-libsci
The following features were added to LibSci 10.5.0:
* Now includes CASE, a collection of simplified interfaces into high
performance LAPACK and ScaLAPACK style routines that find the
eigenvalues
and eigenvectors of a symmetric or hermitian matrix. CASE is
written in
Fortran but has interfaces for C users as well. CASE is provided for
serial or parallel problems, with real or complex data types and
single
or double precision. It has generic interfaces callable from
Fortran and
specific interfaces callable from Fortran or C. See the
'intro_case' man
page for more information.
The LibSci 10.5.0 release adds new interfaces for CRAFFT providing
users
an option to use CRAFFT Serial and Distributed Routines in C
applications.
CRAFFT offers a simpler interface for FFT routines to improve
application
developer productivity. In some cases the performance of the CRAFFT
distributed transforms is 10-50% better than FFTW2 MPI transforms.
Users
requiring more information on usage should see the intro_crafft
man page.
STAT
Eliminate the need to have the STAT daemon installed on the lustre
file system.
Perftools (craypat/apprentice2)
IMPORTANT NOTE: The perftools modulefile needs to be loaded,
otherwise the
following error will occur if a user attempts to load craypat or
apprentice2.
ERROR: xt-craypat and apprentice2 have been merged into one module
called perftools.
Please run the following to load perftools:
module unload xt-craypat apprentice2
module load perftools
General
* UPC and CAF require CCE version 7.2.7 or later (see bug 763219)
pat_build
* update the following trace groups: adios, dmapp, hdf5, netcdf,
petsc, pgas, pthreads, upc
* allow tracing of functions defined as WEAK (bug 764102)
* remove PAT_BUILD_ADDSYM and PAT_BUILD_TRACE_ARCHIVE environment
variables
* add new directives to control addsym utility features
* improve tracing functions that have aggregates as formal parameters
(bug 764058)
pat_report
* now shows inclusive loop times from CCE -hprofile_generate option