Charm++ parallel programming system

[Sam White <white67 AT illinois.edu>]

We have fixed a couple small issues in response to bug reports from users already, and have pushed these fixes to the charm-6.9.0 branch and tagged v6.9.0-rc2 now. Please use this tag for further testing.

Thank you,
Sam White
white67 AT illinois.edu

On Wed, Oct 24, 2018 at 11:10 AM Tom Quinn <trq AT astro.washington.edu> wrote:

I see it has been fixed!  Sorry for the false alarm.

Tom Quinn       Astronomy, University of Washington
Internet:       trq AT astro.washington.edu
Phone:          206-685-9009

On Wed, 24 Oct 2018, Tom Quinn wrote:

> Hi Sam,
>
> I wonder if anyone has looked at:
> https://charm.cs.illinois.edu/redmine/issues/2001
>
> I think the machine.C file for gni (and possibly MPI) erroneously declares
> CmiTimerAbsolute() to be "inline".
>
> Tom Quinn     Astronomy, University of Washington
> Internet:     trq AT astro.washington.edu
> Phone:                206-685-9009
>
> On Wed, 24 Oct 2018, Sam White wrote:
>
>>  Hello everyone,
>>
>>  We have just tagged a release candidate for the next stable release,
>>  version
>>  6.9.0, of Charm++. This version has passed all of our automated tests, and
>>  has been tested by various Charm++ users already. If you find a regression
>>  in the RC, please notify us at core-ppl AT cs.illinois.edu. If there are no
>>  major regressions found in it, we will tag the 6.9.0 release next week.
>>  You
>>  can check out the release in the charm git repo on the branch
>>  'charm-6.9.0'
>>  or at the git tag 'v6.9.0-rc1'.
>>
>>  Version 6.9.0 is a feature release for Charm++ and AMPI that includes many
>>  bug fixes, new features, and improvements. Here's a list of some of the
>>  major ones:
>>
>>  Highlights:
>>
>>  - Charm++ now requires C++11 and better supports use of modern C++ in
>>  applications.
>>  - New "Zero Copy" messaging APIs for more efficient communication of large
>>  arrays.
>>  - CharmPy provides a new Python interface to Charm++, without the need for
>>  .ci files.
>>  - AMPI performance, standard compliance, and usability improvements.
>>  - GPU Manager improvements for asynchronous offloading and a new, more
>>  CUDA-like API.
>>
>>  Charm++ Features & Fixes:
>>
>>  - Added new, more intuitive process launching commands based on hwloc
>>  support, such as '++processPer{Host,Socket,Core,PU} <num>' and
>>  '++oneWthPer{Host,Socket,Core,PU}'. Also added a '++autoProvision' option,
>>  which by default uses all hardware resources available to the job.
>>  - Added a new 'zero copy' direct API which allows users to communicate
>>  large
>>  message buffers directly via RDMA on networks that support it, avoiding
>>  any
>>  intermediate buffering of data between the sender and the receiver. The
>>  API
>>  is also optimized for shared memory.
>>  - A new Python interface to Charm++, currently named CharmPy, is now
>>  available for Python users. More documentation on CharmPy can be found
>>  here:
>>  http://charmpy.readthedocs.io
>>  - Charmxi now supports r-value references, std::array, std::tuple, the
>>  'typename' keyword, parameter packs, variadic templates, array indices
>>  with
>>  template parameters, and attributes on explicit instantiations of
>>  templated
>>  entry methods.
>>  - Projections traces of templated entry methods now display demangled
>>  template type names.
>>  - [local] and [inline] entry method attributes now work for templated
>>  entry
>>  methods and now support perfect forwarding of their arguments.
>>  - Added various type traits for generic programming with Charm++ entities
>>  inside charm++_type_traits.h
>>  - Chare array index types are now exposed as 'array_index_t'.
>>  - Support for default arguments to Group entry methods.
>>  - Charm++ now throws a runtime error when a user calls an SDAG entry
>>  method
>>  containing a 'when' clause directly, without calling it via a proxy.
>>  - Users can now pass std::vector's directly to contribute() rather than
>>  passing the size and buffer address separately. Cross-array section
>>  reduction contributions can now take a callback.
>>  - Added a simplified STL-based interface for section creation.
>>  - Added PUP support for C++ enums, for std::deque and std::forward_list,
>>  for
>>  STL containers of objects with abstract base classes, and for avoiding
>>  default construction during unpacking by defining a constructor that takes
>>  a
>>  value of type PUP::reconstruct.
>>  - Improved performance for PUP of STL containers of arithmetic types and
>>  types declared as PUPbytes.
>>  - Allow setting queueing type and priorities on entry methods with no
>>  parameters.
>>  - Enable setting Group and Node Group dependencies on all types of entry
>>  methods and constructors, as well as multiple dependencies.
>>  - Support for model-based runtime load balancing strategy selection via
>>  MetaLB. This can be enabled with +MetaLBModelDir <path-to-model> used
>>  alongside +MetaLB option. A trained model can be found in
>>  charm/src/ck-ldb/rf_model.
>>  - A new lock-free producer-consumer queue implementation has been added as
>>  a
>>  build option '--enable-lockless-queue' for LRTS's within-process messaging
>>  queues in SMP mode.
>>  - CkLoop now supports lambda syntax, uses more dynamic scheduling
>>  policies,
>>  and exposes
>>    an API for users to set whether or not a PE will steal work from others
>>  in
>>  its process.
>>  - Updated our integrated LLVM OpenMP runtime to support more OpenMP
>>  directives.
>>  - Updated f90charm interface for more functionality and usability, and
>>  fixed
>>  all example programs.
>>  - The Infiniband 'verbs' communication layer now automatically selects the
>>  fastest active Infiniband device and port at startup.
>>  - Fixed '-tracemode utilization', tracing of user-level threads, and
>>  nested
>>  local/inline methods.
>>  - Fixed a performance bug introduced in v6.8.0 for dynamic location
>>  management.
>>  - Added support for using Boost's lightweight uFcontext user-level
>>  threads,
>>  now the default ULT implementation on most platforms.
>>  - '++debug' now works using lldb on Mac (Darwin) systems.
>>  - CkAbort() is now marked with the C++ attribute [[noreturn]].
>>  - CkExit() now takes an optional integer argument which is returned from
>>  the
>>  program's exit.
>>  - Improved error checking throughout, and fixes to race conditions during
>>  startup.
>>
>>  AMPI Changes:
>>
>>  - Improved performance of point-to-point message matching and reduced
>>  per-rank memory footprint.
>>  - Fixes to derived datatypes handling, MPI_Sendrecv_replace,
>>  MPI_(I)Alltoall{v,w}, MPI_(I)Scatter(v), MPI_IN_PLACE in gather
>>  collectives,
>>  MPI_Buffer_detach, MPI_Type_free, MPI_Op_free, and MPI_Comm_free.
>>  - Implemented support for generalized requests, MPI_Comm_create_group,
>>  keyval attribute callbacks, the distributed graph virtual topology, large
>>  count routines, matched probe and recv, and MPI_Comm_idup(_with_info)
>>  routines.
>>  - Added support for using -tlsglobals for privatization of global/static
>>  variables in shared objects. Previously -tlsglobals required static
>>  linking.
>>  - '-memory os-isomalloc', which uses the system's malloc underneath, now
>>  works everywhere Isomalloc does. Both versions of Isomalloc now wrap calls
>>  to posix_memalign(), and we removed the need to link with
>>  '-Wl,--allow-multiple-definition' on some systems.
>>  - Updated AMPI_Migrate() with built-in MPI_Info objects, such as
>>  AMPI_INFO_LB_SYNC.
>>  - AMPI now only renames the user's MPI calls from MPI_* to AMPI_* if
>>  Charm++/AMPI is built on top of another MPI implementation for its
>>  communication substrate.
>>  - Support for compiling mpif.h in both fixed form and free form.
>>  - PMPI profiling interface support added.
>>  - Added an ampirun script that wraps charmrun to enable easier integration
>>  with build and test scripts that take mpirun/mpiexec as an option.
>>
>>  GPU Manager Changes:
>>
>>  - Enable concurrent kernel execution by removing the limit imposed by the
>>  internal implementation using only 3 streams.
>>  - New API (Hybrid API, or HAPI) that is similar to the CUDA API.
>>  - Added NVIDIA NVTX support for profiling host-side functions.
>>  - Deprecated the workRequest API. New users are now strongly recommended
>>  to
>>  use the new API, or Hybrid API (HAPI).
>>
>>  Build System Changes:
>>
>>  - Charm++ now requires C++11 support, and as such defaults to using
>>  bgclang
>>  on BGQ. Compilers GCC v4.8+, ICC v15.0+, XLC v13.1+, Cray CC v8.6+, MSVC
>>  v19.00.24+ and Clang v3.3+ are required.
>>  - Building Charm++ from the git repository now requires autoconf and
>>  automake.
>>  - Support for the Flang Fortran compiler added.
>>  - Users can now specify compiler versions to our top-level build script
>>  when
>>  building with gcc or clang.
>>  - Windows users can now build Charm++ with GCC, Clang, or MSVC.
>>  - All of Charm++ and AMPI can now be built as shared objects.
>>  - Added a CMake wrapper for compiling .ci files.
>>  - Charm++ is now available in Spack under the name 'charmpp'.
>>  - Added {pamilrts,mpi,multicore,netlrts}-linux-ppc64le build targets for
>>  new
>>  IBM POWER systems.
>>
>>
>>  Thank you!
>>  Sam White
>>  white67 AT illinois.edu
>>
>>
>