| blob | 10112198bf03a9e6af06bbb2fd03e99cf51707d2 |
1 This file describes the most significant changes. For more detail, use
2 'git log' on a clone of the charm repository.
4 ================================================================================
5 What's new in Charm++ 6.8.0
6 ================================================================================
8 Over 900 bug fixes, improvements, and cleanups have been applied
9 across the entire system. Major changes are described below:
11 Charm++ Features
13 - Calls to entry methods taking a single fixed-size parameter can now
14 automatically be aggregated and routed through the TRAM library by
15 marking them with the [aggregate] attribute.
17 - Calls to parameter-marshalled entry methods with large array
18 arguments can ask for asynchronous zero-copy send behavior with an
19 `nocopy' tag in the parameter's declaration.
21 - The runtime system now integrates an OpenMP runtime library so that
22 code using OpenMP parallelism will dispatch work to idle worker
23 threads within the Charm++ process.
25 - Applications can ask the runtime system to perform automatic
26 high-level end-of-run performance analysis by linking with the
27 `-tracemode perfReport' option.
29 - Added a new dynamic remapping/load-balancing strategy,
30 GreedyRefineLB, that offers high result quality and well bounded
31 execution time.
33 - Improved and expanded topology-aware spanning tree generation
34 strategies, including support for runs on a torus with holes, such
35 as Blue Waters and other Cray XE/XK systems.
37 - Charm++ programs can now define their own main() function, rather
38 than using a generated implementation from a mainmodule/mainchare
39 combination. This extends the existing Charm++/MPI interoperation
40 feature.
42 - Improvements to Sections:
44 * Array sections API has been simplified, with array sections being
45 automatically delegated to CkMulticastMgr (the most efficient implementation
46 in Charm++). Changes are reflected in Chapter 14 of the manual.
48 * Group sections can now be delegated to CkMulticastMgr (improved performance
49 compared to default implementation). Note that they have to be manually
50 delegated. Documentation is in Chapter 14 of Charm++ manual.
52 * Group section reductions are now supported for delegated sections
53 via CkMulticastMgr.
55 * Improved performance of section creation in CkMulticastMgr.
57 * CkMulticastMgr uses the improved spanning tree strategies. See above.
59 - GPU manager now creates one instance per OS process and scales the
60 pre-allocated memory pool size according to the GPU memory size and
61 number of GPU manager instances on a physical node.
63 - Several GPU Manager API changes including:
65 * Replaced references to global variables in the GPU manager API with calls to
66 functions.
68 * The user is no longer required to specify a bufferID in dataInfo struct.
70 * Replaced calls to kernelSelect with direct invocation of functions passed
71 via the work request object (allows CUDA to be built with all programs).
73 - Added support for malleable jobs that can dynamically shrink and
74 expand the set of compute nodes hosting Charm++ processes.
76 - Greatly expanded and improved reduction operations:
78 * Added built-in reductions for all logical and bitwise operations
79 on integer and boolean input.
81 * Reductions over groups and chare arrays that apply commutative,
82 associative operations (e.g. MIN, MAX, SUM, AND, OR, XOR) are now
83 processed in a streaming fashion. This reduces the memory footprint of
84 reductions. User-defined reductions can opt into this mode as well.
86 * Added a new `Tuple' reducer that allows combining multiple reductions
87 of different input data and operations from a common set of source
88 objects to a single target callback.
90 * Added a new `Summary Statistics' reducer that provides count, mean,
91 and standard deviation using a numerically-stable streaming algorithm.
93 - Added a `++quiet' option to suppress charmrun and charm++ non-error
94 messages at startup.
96 - Calls to chare array element entry methods with the [inline] tag now
97 avoid copying their arguments when the called method takes its
98 parameters by const&, offering a substantial reduction in overhead in
99 those cases.
101 - Synchronous entry methods that block until completion (marked with
102 the [sync] attribute) can now return any type that defines a PUP
103 method, rather than only message types.
105 - Static (non-generated) header files are now warning-free for
106 gcc -Wall -Wextra -pedantic.
108 - Deprecated setReductionClient and CkSetReductionClient in favor of
109 explicitly passing callbacks to contribute calls.
111 - On C++ standard library implementations with support for
112 std::is_constructible (e.g. GCC libstdc++ >4.5), chare array
113 elements only need to define a constructor taking CkMigrateMessage*
114 if it will actually be migrated.
116 - The PUP serialization framework gained support for some C++11
117 library classes, including unique_ptr and unordered_map, when the
118 underlying types have PUP operators.
120 AMPI Features
122 - More efficient implementations of message matching infrastructure, multiple
123 completion routines, and all varieties of reductions and gathers.
125 - Support for user-defined non-commutative reductions, MPI_BOTTOM, cancelling
126 receive requests, MPI_THREAD_FUNNELED, PSCW synchronization for RMA, and more.
128 - Fixes to AMPI's extensions for load balancing and to Isomalloc on SMP builds.
130 - More robust derived datatype support, optimizations for truly contiguous types.
132 - ROMIO is now built on AMPI and linked in by ampicc by default.
134 - A version of HDF5 v1.10.1 that builds and runs on AMPI with virtualization
135 is now available at https://charm.cs.illinois.edu/gerrit/#/admin/projects/hdf5-ampi
137 - Improved support for performance analysis and visualization with Projections.
139 Platforms and Portability
141 - The runtime system code now requires compiler support for C++11
142 R-value references and move constructors. This is not expected to be
143 incompatible with any currently supported compilers.
145 - The next feature release (anticipated to be 6.9.0 or 7.0) will require
146 full C++11 support from the compiler and standard library.
148 - Added support for IBM POWER8 systems with the PAMI communication API,
149 such as development/test platforms for the upcoming Sierra and Summit
150 supercomputers at LLNL and ORNL. Contributed by Sameer Kumar of IBM.
152 - Mac OS (darwin) builds now default to the modern libc++ standard
153 library instead of the older libstdc++.
155 - Blue Gene/Q build targets have been added for the `bgclang' compiler.
157 - Charm++ can now be built on Cray's CCE 8.5.4+.
159 - Charm++ will now build without custom configuration on Arch Linux
161 - Charmrun can automatically detect rank and node count from
162 Slurm/srun environment variables.
164 - Many obsolete architecture, network, and compiler support files have
165 been removed. These include:
166 * IBM Blue Gene/P
167 * Sony/Toshiba/IBM Cell (including PlayStation 3)
168 * Cray XT
169 * Intel IA-64 (Itanium)
170 * Intel x86-32 for Windows, Mac OS X (darwin), and Solaris
171 * Cygwin for Windows
172 * Older IBM AIX/POWER configurations
173 * GCC 3 and KAI compilers
174 * Sun/Oracle Solaris
176 ================================================================================
177 What's new in Charm++ 6.7.1
178 ================================================================================
180 Changes in this release are primarily bug fixes for 6.7.0. The major exception
181 is AMPI, which has seen changes to its extension APIs and now complies with more
182 of the MPI standard. A brief list of changes follows:
184 Charm++ Bug Fixes
186 - Startup and exit sequences are more robust
188 - Error and warning messages are generally more informative
190 - CkMulticast's set and concat reducers work correctly
192 AMPI Features
194 - AMPI's extensions have been renamed to use the prefix AMPI_ instead of MPI_
195 and to generally follow MPI's naming conventions
197 - AMPI_Migrate(MPI_Info) is now used for dynamic load balancing and all fault
198 tolerance schemes (see the AMPI manual)
200 - AMPI officially supports MPI-2.2, and also implements the non-blocking
201 collectives and neighborhood collectives from MPI-3.1
203 Platforms and Portability
205 - Cray regularpages build target has been fixed
207 - Clang compiler target for BlueGene/Q systems added
209 - Comm. thread tracing for SMP mode added
211 - AMPI's compiler wrappers are easier to use with autoconf and cmake
213 ================================================================================
214 What's new in Charm++ 6.7.0
215 ================================================================================
217 Over 120 bugs fixed, spanning areas across the entire system
219 Charm++ Features
221 - New API for efficient formula-based distributed sparse array creation
223 - CkLoop is now built by default
225 - CBase_Foo::pup need not be called from Foo::pup in user code anymore - runtime
226 code handles this automatically
228 - Error reporting and recovery in .ci files is greatly improved, providing more
229 precise line numbers and often column information
231 - Many data races occurring under shared-memory builds (smp, multicore) were
232 fixed, facilitating use of tools like ThreadSanitizer and Helgrind
234 AMPI Enhancements
236 - Further MPI standard compliance in AMPI allows users to build and run
237 Hypre-2.10.1 on AMPI with virtualization, migration, etc.
239 - Improved AMPI Fortran2003 PUP interface 'apup', similar to C++'s STL PUP
241 Platforms and Portability
243 - Compiling Charm++ now requires support for C++11 variadic templates. In GCC,
244 this became available with version 4.3, released in 2008
246 - New machine target for multicore Linux ARM7: multicore-linux-arm7
248 - Preliminary support for POWER8 processors, in preparation for the upcoming
249 Summit and Sierra supercomputers
251 - The charmrun process launcher is now much more robust in the face of slow
252 or rate-limited connections to compute nodes
254 - PXSHM now auto-detects the node size, so the '+nodesize' is no longer needed
256 - Out-of-tree builds are now supported
258 Deprecations
260 - CommLib has been removed.
262 - CmiBool has been dropped in favor of C++'s bool
265 ================================================================================
266 What's new in Charm++ 6.6.1
267 ================================================================================
269 Changes in this release are primarily bug fixes for 6.6.0. A concise list of
270 affected components follows:
272 - CkIO
274 - Reductions with syncFT
276 - mpicxx based MPI builds
278 - Increased support for macros in CI file
280 - GNI + RDMA related communication
282 - MPI_STATUSES_IGNORE support for AMPIF
284 - Restart on different node count with chkpt
286 - Immediate msgs on multicore builds
288 ================================================================================
289 What's new in Charm++ 6.6.0
290 ================================================================================
292 - Machine target files for Cray XC systems ('gni-crayxc') have been added
294 - Interoperability with MPI code using native communication interfaces on Blue
295 Gene Q (PAMI) and Cray XE/XK/XC (uGNI) systems, in addition to the universal
296 MPI communication interface
298 - Support for partitioned jobs on all machine types, including TCP/IP and IB
299 Verbs networks using 'netlrts' and 'verbs' machine layers
301 - A substantially improved version of our asynchronous library, CkIO, for
302 parallel output of large files
304 - Narrowing the circumstances in which the runtime system will send
305 overhead-inducing ReductionStarting messages
307 - A new fully distributed load balancing strategy, DistributedLB, that produces
308 high quality results with very low latency
310 - An API for applications to feed custom per-object data to specialized load
311 balancing strategies (e.g. physical simulation coordinates)
313 - SMP builds on LRTS-based machine layers (pamilrts, gni, mpi, netlrts, verbs)
314 support tracing messages through communication threads
316 - Thread affinity mapping with +pemap now supports Intel's Hyperthreading more
317 conveniently
319 - After restarting from a checkpoint, thread affinity will use new
320 +pemap/+commap arguments
322 - Queue order randomization options were added to assist in debugging race
323 conditions in application and runtime code
325 - The full runtime code and associated libraries can now compile under the C11
326 and C++11/14 standards.
328 - Numerous bug fixes, performance enhancements, and smaller improvements in the
329 provided runtime facilities
331 - Deprecations
332 * The long-unsupported FEM library has been deprecated in favor of ParFUM
333 * The CmiBool typedefs have been deleted, as C++ bool has long been universal
334 * Future versions of the runtime system and libraries will require some degree
335 of support for C++11 features from compilers
337 ================================================================================
338 What's new in Charm++ 6.5.0
339 ================================================================================
341 - The Charm++ manual has been thoroughly revised to improve its organization,
342 comprehensiveness, and clarity, with many additional example code snippets
343 throughout.
345 - The runtime system now includes the 'Metabalancer', which can provide
346 substantial performance improvements for applications that exhibit dynamic
347 load imbalance. It provides two primary benefits. First, it automatically
348 optimizes the frequency of load balancer invocation, to avoid work stoppage
349 when it will provide too little benefit. Second, calls to AtSync() are made
350 less synchronous, to further reduce overhead when the load balancer doesn't
351 need to run. To activate the Metabalancer, pass the option +MetaLB at
352 runtime. To get the full benefits, calls to AtSync() should be made at every
353 iteration, rather than at some arbitrary longer interval as was previously
354 common.
356 - Many feature additions and usability improvements have been made in the
357 interface translator that generates code from .ci files:
358 * Charmxi now provides much better error reports, including more accurate
359 line numbers and clearer reasons for failure, including some semantic
360 problems that would otherwise appear when compiling the C++ code or even at
361 runtime.
362 * A new SDAG construct 'case' has been added that defines a disjunction over a
363 set of 'when' clauses: only one 'when' out of a set will ever be triggered.
364 * Entry method templates are now supported. An example program can be found
365 in tests/charm++/method_templates/.
366 * SDAG keyword "atomic" has been deprecated in favor of the newly supported
367 keyword "serial". The two are synonymous, but "atomic" is now provided only
368 for backward compatibility.
369 * It is no longer necessary to call __sdag_init() in chares that contain SDAG
370 code - the generated code does this automatically. The function is left as
371 a no-op for compatibility, but may be removed in a future version.
372 * Code generated from .ci files is now primarily in .def.h files, with only
373 declarations in .decl.h. This improves debugging, speeds compilation,
374 provides clearer compiler output, and enables more complete encapsulation,
375 especially in SDAG code.
376 * Mainchare constructors are expected to take CkArgMsg*, and always have
377 been. However, charmxi would allow declarations with no argument, and
378 assume the message. This is now deprecated, and generates a warning.
380 - Projections tracing has been extended and improved in various ways
381 * The trace module can generate a record of network topology of the nodes in
382 a run for certain platforms (including Cray), which Projections can
383 visualize.
384 * If the gzip library (libz) is available when Charm++ is compiled, traces
385 are compressed by default.
386 * If traces were flushed as a results of filled buffers during the run, a
387 warning will be printed at exit to indicate that the user should be wary of
388 interference that may have resulted.
389 * In SMP builds, it is now possible to trace message progression through the
390 communication threads. This is disabled by default to avoid overhead and
391 potential misleading interpretation.
393 - Array elements can be block-mapped at the SMP node level instead of at the
394 per-PE level (option "+useNodeBlkMapping").
396 - AMPI can now privatize global and static variables using TLS. This is
397 supported in C and C++ with __thread markings on the variable declarations
398 and definitions, and in Fortran with a patched version of the gfortran
399 compiler. To activate this feature, append '-tls' to the '-thread' option's
400 argument when you link your AMPI program.
402 - Charm can now be built to only support message priorities of a specific data
403 type. This enables an optimized message queue within the the runtime
404 system. Typical applications with medium sized compute grains may not benefit
405 noticeably when switching to the new scheduler. However, this may permit
406 further optimizations in later releases.
408 The new queue is enabled by specifying the data type of the message
409 priorities while building charm using --with-prio-type=dtype. Here, dtype can
410 be one of char, short, int, long, float, double and bitvec. Specifying bitvec
411 will permit arbitrary-length bitvector priorities, and is the current default
412 mode of operation. However, we may change this in a future release.
414 - Converse now provides a complete set of wrappers for
415 fopen/fread/fwrite/fclose to handle EINTR, which is not uncommon on the
416 increasingly-popular Lustre. They are named CmiF{open,read,write,close}, and
417 are available from C and C++ code.
419 - The utility class 'CkEntryOptions' now permits method chaining for cleaner
420 usage. This applies to all its set methods (setPriority, setQueueing,
421 setGroupDepID). Example usage can be found in examples/charm++/prio/pgm.C.
423 - When creating groups or chare arrays that depend on the previous construction
424 of another such entity on the local PE, it is now possible to declare that
425 dependence to the runtime. Creation messages whose dependence is not yet
426 satisfied will be buffered until it is.
428 - For any given chare class Foo and entry method Bar, the supporting class's
429 member CkIndex_Foo::Bar() is used to lookup/specify the entry method
430 index. This release adds a newer API for such members where the argument is a
431 function pointer of the same signature as the entry method. Those new
432 functions are used like CkIndex_Foo::idx_Bar(&Foo::Bar). This permits entry
433 point index lookup without instantiating temporary variables just to feed the
434 CkIndex_Foo::Bar() methods. In cases where Foo::Bar is overloaded, &Foo::Bar
435 must be cast to the desired type to disambiguate it.
437 - CkReduction::reducerType now have PUP methods defined; and can hence be
438 passed as parameter-marshalled arguments to entry methods.
440 - The runtime option +stacksize for controlling the allocation of user-level
441 threads' stacks now accepts shorthanded annotation such as 1M.
443 - The -optimize flag to the charmc compiler wrapper now passes more aggressive
444 options to the various underlying compilers than the previous '-O'.
446 - The charmc compiler wrapper now provides a flag -use-new-std to enable
447 support for C11 and C++11 where available. To use this in application code,
448 the runtime system must have been built with that flag as well.
450 - When using, CmiMemoryUsage(), the runtime can be instructed not to use the
451 underlying mallinfo() library call, which can be inaccurate in settings where
452 usage exceeds INT_MAX. This is accomplished by setting the environment
453 variable "MEMORYUSAGE_NO_MALLINFO".
455 - Experimental Features
456 * Initial implementation of a fast message-logging protocol. Use option
457 'mlogft' to build it.
458 * Message compression support for persistent message on Gemini machine layer.
459 * Node-level inter-PE loop/task parallelization is now supported through
460 CkLoop
461 * New temperature/CPU frequency aware load balancer
462 * Support interoperation of Charm++ and native MPI code through dynamically
463 switching control between the two
464 * API in centralized load balancers to get and set PE speed
465 * A new scheme for optimization of double in-memory checkpoint/restart.
466 * Message combining library for improved fine-grained communication
467 performance
468 * Support for partitioning of allocated nodes into subsets that run
469 independent Charm++ instances but can interact with each other.
471 Platform-Specific Changes
472 -------------------------
474 - Cray XE/XK
475 * The gemini_gni network layer has been heavily tuned and optimized,
476 providing substantial improvements in performance, scalability, and
477 stability.
478 * The gemini_gni-crayxe machine layer supports a 'hugepages' option at build
479 time, rather than requiring manual configuration file editing.
480 * Persistent message optimizations can be used to reduce latency and
481 overheads
482 * Experimental support for 'urgent' sends, which are sent ahead of any other
483 outgoing messages queued for transmission.
485 - IBM Blue Gene Q: Experimental machine-layer support for the native PAMI
486 interface and MPI, with and without SMP support. This supports many new
487 systems, including LLNL's Sequoia, ALCF's Mira, and FZ Juelich's Juqueen.
489 There are three network-layer implementations for these systems: 'mpi',
490 'pami', and 'pamilrts'. The 'mpi' layer is stable, but its performance and
491 scalability suffers from the additional overhead of using MPI rather than
492 driving the interconnect directly. The 'pami' layer is well tested for NAMD,
493 but has shown instability for other applications. It is likely to be replaced
494 by the 'pamilrts' layer, which is more generally stable and seems to provide
495 the same performance, in the next release.
497 In addition to the common 'smp' option to build the runtime system with
498 shared memory support, there is an 'async' option which sometimes provides
499 better performance on SMP builds. This option passes tests on 'pamilrts', but
500 is still experimental.
502 Note: Applications that have large number of messages may crash in default
503 setup due to overflow in the low-level FIFOs. Environment variables
504 MUSPI_INJFIFOSIZE and PAMI_RGETINJFIFOSIZE can be set to avoid application
505 failures due to large number of small and large messages respectively. The
506 default value of these variable is 65536 which is sufficient for 1000
507 messages in flight.
509 - Infiniband Verbs: Better support for more flavors of ibverbs libraries
511 - MPI Network Layer
512 * Experimental rendezvous protocol for better performance above some MPI
513 implementations.
514 * Some tuning parameters ("+dynCapSend" and "+dynCapRecv") are now
515 configurable at job launch, rather than Charm++ compilation.
517 - PGI C++: Disable automatic 'using namespace std;'
519 - Charm++ now supports ARM, both non-smp and smp.
521 - Mac OS X: Compilation options to build and link correctly on newer versions
524 ================================================================================
525 What's new in Charm++ 6.4.0
526 ================================================================================
528 --------------------------------------------------------------------------------
529 Platform Support
530 --------------------------------------------------------------------------------
532 - Cray XE and XK systems using the Gemini network via either MPI
533 (mpi-crayxe) or the native uGNI (gemini_gni-crayxe)
535 - IBM Blue Gene Q, using MPI (mpi-bluegeneq) or PAMI (pami-bluegeneq)
537 - Clang, Cray, and Fujitsu compilers
539 - MPI-based machine layers can now run on >64k PEs
541 --------------------------------------------------------------------------------
542 General Changes
543 --------------------------------------------------------------------------------
545 - Added a new [reductiontarget] attribute to enable
546 parameter-marshaled recipients of reduction messages
548 - Enabled pipelining of large messages in CkMulticast by default
550 - New load balancers added:
551 * TreeMatch
552 * Zoltan
553 * Scotch graph partitioning based: ScotchLB and Refine and Topo variants
554 * RefineSwap
556 - Load balancing improvements:
558 * Allow reduced load database size using floats instead of doubles
559 * Improved hierarchical balancer
560 * Periodic balancing adapts its interval dynamically
561 * User code can request a callback when migration is complete
562 * More balancers properly consider object migratability and PE
563 availability and speed
564 * Instrumentation records multicasts
566 - Chare arrays support options that can enable some optimizations
568 - New 'completion detection' library for parallel process termination
569 detection, when the need for modularity excludes full quiescence
570 detection
572 - New 'mesh streamer' library for fine-grain many-to-many collectives,
573 handling message bundling and network topology
575 - Memory pooling allocator performance and resource usage improved
576 substantially
578 - AMPI: More routines support MPI_IN_PLACE, and those that don't check
579 for it
581 ================================================================================
582 What's new in Charm++ 6.2.1 (since 6.2.0)
583 ================================================================================
585 --------------------------------------------------------------------------------
586 New Supported Platforms:
587 --------------------------------------------------------------------------------
589 POWER7 with LAPI on Linux
591 Infiniband on PowerPC
593 --------------------------------------------------------------------------------
594 General Changes
595 --------------------------------------------------------------------------------
597 - Better support for multicasts on groups
598 - Topology information gathering has been optimized
599 - Converse (seed) load balancers have many new optimizations applied
600 - CPU affinity can be set more easily using +pemap and +commap options
601 instead of the older +coremap
602 - HybridLB (hierarchical balancing for very large core-count systems)
603 has been substantially improved
604 - Load balancing infrastructure has further optimizations and bug fixes
605 - Object mappings can be read from a file, to allow offline
606 topology-aware placement
607 - Projections logs can be spread across multiple directories, speeding
608 up output when dealing with thousands of cores (+trace-subdirs N
609 will divide log files evenly among N subdirectories of the trace
610 root, named PROGNAME.projdir.K)
611 - AMPI now implements MPI_Issend
612 - AMPI's MPI_Alltoall uses a flooding algorithm more agressively,
613 versus pairwise exchange
614 - Virtualized ARMCI support has been extended to cover the functions
615 needed by CAF
617 --------------------------------------------------------------------------------
618 Architecture-specific changes
619 --------------------------------------------------------------------------------
621 - LAPI SMP has many new optimizations applied
623 - Net builds support the use of clusters' mpiexec systems for job
624 launch, via the ++mpiexec option to charmrun
626 ================================================================================
627 What's new in Charm++ 6.2.0 (since 6.1)
628 ================================================================================
630 --------------------------------------------------------------------------------
631 New Supported Platforms:
632 --------------------------------------------------------------------------------
634 64-bit MIPS, such as SiCortex, using mpi-linux-mips64
636 Windows HPC cluster, using mpi-win32/mpi-win64
638 Mac OSX 10.6, Snow Leopard (32-bit and 64-bit).
640 --------------------------------------------------------------------------------
641 General Changes
642 --------------------------------------------------------------------------------
644 Runtime support
645 - Smarter build/configure scripts
646 - A new interface for model-based load balancing
647 - new CPU topology API
648 - a general implementation of CmiMemoryUsage()
649 - Bug fix: Quiescence detection (QD) works with immediate messages
650 - New reduction functions implemented in Converse
651 - CCS (Converse Client-Server) can deliver message to more than one processor
652 - Added a memory-aware adaptive scheduler, which can be optionally
653 compiled in to charm
654 - Added preliminary support for automatic message prioritization
655 (disabled by default)
657 Charm++
658 - Cross-array and cross-group sections
659 - Structured Dagger (SDAG): Support templated arguments properly
660 - Plain chares support checkpoint/restart (both in-memory and disk-based)
661 - Conditional packing of messages and parameters in SMP scenario
662 - Changes to the CkArrayIndex class hierarchy
663 -- sizeof() all CkArrayIndex* classes is now the same
664 -- Codes using custom array indices have to use placement-new to construct
665 their custom index. Refer example code: examples/charm++/hello/fancyarray/
666 -- *** Backward Incompatibility ***
667 CkArrayIndex[4D/5D/6D]::index are now of type int (instead of short)
668 However the data is stored as shorts. Access by casting
669 CkArrayIndexND::data() appropriately
670 -- *** Deprecated ***
671 The direct use of public data member
672 CkArrayIndexND::index (N=1..6) is deprecated. We reserve the right to
673 change/remove this variable in future releases of Charm++.
674 Instead, please access the indices via member function:
675 int CkArrayIndexND::data()
677 Adaptive MPI (AMPI)
678 - Compilers renamed to avoid collision with host MPI (ampicc, ampiCC,
679 ampif77, ampif90)
680 - Improved MPI standard conformance, and documentation of non-conformance
681 * Bug fixes in: MPI_Ssend, MPI_Cart_shift, MPI_Get_count
682 * Support MPI_IN_PLACE in MPI_(All)Reduce
683 * Define various missing constants
684 - Return the received message's tag in response to a non-blocking
685 wildcard receive, to support SuperLU
686 - Improved tracing for BigSim
688 Multiphase Shared Arrays (MSA)
689 - Typed handles to enforce phases
690 - Split-phase synchronization to enable message-driven execution
691 - 3D arrays
693 TCharm
694 - Automatic tracing of API calls for simulation and analysis
696 Debugging
697 - Wider support for architectures other than net- (in particular MPI layers)
698 - Improved support for large scale debugging (better scalability)
699 - Enhanced record/replay stability to handle various events, and to
700 signal unexpected messages
701 - New detailed record/replay: The full content of messages can be
702 recorded, and a single processor can be re-executed outside of the
703 parallel application
705 Performance analysis
706 - Tracing of nested entry methods
708 Automatic Performance Tuning
709 - Created an automatic tuning framework [still for experimental use only]
711 CkMulticast
712 - Network-topology / node aware spanning trees used internally for and
713 lower bytes on the network and improved performance in multicasts and
714 reductions delegated to this library
716 Comlib
717 - Improved OneTimeMulticastStrategy classes
719 BigSim
720 - Out-of-core support, with prefetching capability
721 - Detailed tracing of MPI calls
722 - Detailed record/replay support at emulation time, capable of
723 replaying any emulated processor after obtained recorded logs.
725 --------------------------------------------------------------------------------
726 Architecture-specific changes
727 --------------------------------------------------------------------------------
729 Net-*
730 - Can run jobs with more than 1024 PEs
732 Net-Linux
733 - New charmrun option ++no-va-randomization to disable address space
734 randomization (ASLR). This is most useful for running AMPI with
735 isomalloc
737 MPI
738 - Default to using ampicxx instead of mpiCC
740 MPI-SMP
741 - The +p option now has the same semantics as in other smp builds
743 Power 7
744 - Support for VSX in SIMD abstraction API
746 Blue Gene/L
747 - Compilers and options have been updated to the latest ones
749 Blue Gene/P
750 - Added routines for measuring performance counters on BG/P.
751 - Updated to support latest DCMF driver version. On ANL's Intrepid, you may
752 need to set BGP_INSTALL=/bgsys/drivers/V1R4M1_460_2009-091110P/ppc in your
753 environment. This is the default on ANL's Surveyor.
755 Cray XT
756 - cputopology information is now available on XT3/4/5
758 Infiniband (ibverbs)
759 - Bug fix: plug memory leaks that caused failures in long runs
760 - Optimized to reduce startup delays
762 LAPI
763 - Support for SMP (experimental)
766 ================================================================================
767 Note that changes from 5.9, 6.0, and 6.1 are not documented here. A partial list
768 can be found on the charm download page, or by reading through version control
769 logs.
771 ================================================================================
772 What's New since Charm++ 5.4 release 1
773 ================================================================================
775 --------------------------------------------------------------------------------
776 New Supported Platforms:
777 --------------------------------------------------------------------------------
778 1. Charm++ ported to IA64 Itanium running Win2K and Linux, Charm++ also support
779 Intel C/C++ compilers;
781 2. Charm++ ported to Power Macintosh powerpc running Darwin;
783 3. Charm++ ported to Myrinet networking with GM API;
785 --------------------------------------------------------------------------------
786 Summary of New Features:
787 --------------------------------------------------------------------------------
788 1. Structure Dagger
789 Structured Dagger is a coordination language built on top of CHARM++.
790 Structured Dagger allows easy expression of dependences among messages and
791 computations and also among computations within the same object using
792 when-blocks and various structured constructs.
794 2. Entry functions support parameter marshalling
795 Now you can declare and invoke remote entry functions using parameter
796 marshalling instead of defining messages.
798 3. Easier running - standalone mode
799 For net-* version running locally, you can now run Charm programs without
800 charmrun. Running a node program directly from command line is now the
801 same as "charmrun +p1 <program>"; for SMP version, you can also specify
802 multiple (local) processors, as in "program +p2".
805 --------------------------------------------------------------------------------
806 Summary of Changes:
807 --------------------------------------------------------------------------------
808 1. "build" changed for compilation of Charm++
809 To build Charm++ from scratch, we now take additional command line options
810 to compile with addon features and using different compilers other than gcc.
811 For example, to build Linux IA64 with Myrinet support, type command:
812 ./build net-linux-ia64 gm
815 ******* Old Change histories *******
818 ================================================================================
819 What's New in Charm++ 5.4 release 1 since 5.0
820 ================================================================================
822 --------------------------------------------------------------------------------
823 New Supported Platforms:
824 --------------------------------------------------------------------------------
826 1. Win9x/2000/NT: with Visual C++ or Cygwin gcc/g++, you can compile and run
827 Charm++ programs on all Win32 platforms.
829 2. Scyld Beowulf: Charm++ has been ported to the Linux-based Scyld Beowulf
830 operating system. For more information on Scyld, see <http://www.scyld.com>
832 3. MPI with VMI: Charm++ has been ported to NCSA's Virtual Machine Interface,
833 which is an efficient messaging library for heterogeneous cluster
834 communication.
837 --------------------------------------------------------------------------------
838 Summary of New Features:
839 --------------------------------------------------------------------------------
840 1. Dynamic Load balancing:
841 Chare migration is supported in the new release. Migration-based dynamic
842 load balancing framework with various load balancing strategies library has
843 been added.
845 2. Chare Array
846 Charm++ array is supported. You can now create an array of Chare objects
847 and use array index to refer the Charm++ array elements. A reduction
848 library on top of Chare array has been implemented and included.
850 3. Projections
851 Projections, a Java application for Charm++ program performance analysis and
852 visualization, has been included and distributed in the new release. Two
853 trace modes are available: trace-projections and trace-summary. Trace-summary
854 is a light-weight trace library compared to trace-projections.
856 4. AMPI
857 AMPI is a load-balancing based library for porting legacy MPI applications
858 to Charm++. With few changes in the original MPI code to AMPI, the new
859 legacy MPI application on Charm++ will gain from Charm++'s adptive
860 load balancing ability.
862 5. Easier invocation
863 "Charmrun" is now available on all platforms, with a uniform command line
864 syntax. You can forget the difference between net-* versions and MPI versions,
865 and run charm++ application with this same charmrun command syntax.
866 ++local option is added in charmrun for net-* version, it provides
867 simple local use of Charm and no longer require the ability to
868 "rsh localhost" or a nodelist file in order to run charm only on the local
869 machine. This is especially attractive when you run Charm++ on Windows.
871 6. New libraries:
872 Many new libraries have been added in this release. They include:
873 1) master-slave library: for writing manager-worker paradigm programs.
874 2) receiver library: provide asynchronous communication mode for chare array.
875 3) f90charm: provides Fortran90 bindings for Charm++ Array.
876 4) BlueGene: a Charm++/Converse emulator for IBM proposed Blue Gene.
878 --------------------------------------------------------------------------------
879 Summary of Changes:
880 --------------------------------------------------------------------------------
881 1. message declaration syntax in .ci file:
882 The message declaration syntax for packed/varsize messages has been changed.
883 The packed/varsize keywords are eliminated, and you can specify the actual
884 actual varsize arrays in the interface file and have the translator generate
885 alloc, pack and unpack.
888 Here is the detailed list of Changes:
890 --------------------------------------------------------------------------------
891 Major Features:
892 --------------------------------------------------------------------------------
894 10/06/1999 rbrunner Added migration-based dynamic load balancing
895 framework.
896 11/15/1999 olawlor Added reduction support foe Charm++ arrays
897 02/06/2000 milind Added AMPI, an implementation of MPI with
898 dynamic load balancing
899 02/18/2000 paranjpy New platforms supported: net-win32, and net-win32-smp
900 04/04/2000 olawlor Added arbitrarily indexed Charm++ arrays.
901 Also, added translator support for new arrays.
902 04/15/2000 olawlor Added "puppers" for packing and unpacking
903 objects.
904 06/14/2000 milind Added the threaded FEM framework.
906 --------------------------------------------------------------------------------
907 Minor Features:
908 --------------------------------------------------------------------------------
910 10/09/1999 rbrunner Added packlib, a library for C and C++ to
911 pack-unpack data to/from Charm++ messages.
912 10/13/1999 gzheng New LB strategy: RefineLB
913 10/13/1999 paranjpy New LB Strategy: Heap
914 10/14/1999 milind New LB Strategy: Metis
915 10/19/1999 olawlor New test program for testing LB strategies.
916 10/21/1999 gzheng New trace mode: trace-summary
917 10/28/1999 milind New supported platform: net-sol-x86
918 10/29/1999 milind Added runtime checks for ChareID assignment.
919 11/10/1999 rbrunner Added Neighborhood base strategy for LB
920 framework.
921 11/15/1999 olawlor conv-host now reads in a startup file
922 ~/.conv-hostrc
923 11/15/1999 olawlor New test program for testing array reductions.
924 11/16/1999 rbrunner Added processor-speed checking functions to
925 LB framework
926 11/19/1999 milind Mapped SIGUSR to a Ccd condtion handler
927 11/22/1999 rbrunner New LB strategy: WSLB
928 11/29/1999 ruiliu Modified Metis LB strategy to deal with
929 different processor speeds
930 12/16/1999 rbrunner New LB strategy: GreedyRef
931 12/16/1999 rbrunner New LB strategy: RandRef
932 12/21/1999 skumar2 New LB strategy: CommLB
933 01/03/2000 rbrunner New LB strategy: RecBisectBfLB
934 01/08/2000 skumar2 New LB strategy: Comm1LB, with varying processor
935 speeds
936 01/18/2000 milind Modified SM library syntax, and added a test
937 program for SM.
938 01/19/2000 gzheng Added irecv, a library to simplify conversion
939 of message-passing programs to Charm++
940 02/20/2000 olawlor Added preliminary broadcast support to Charm++
941 arrays.
942 02/23/2000 paranjpy Added converse-level quiescence detection
943 03/02/2000 milind Added ++server-port option to pre-specify
944 CCS port.
945 03/10/2000 wilmarth Random seed-based load balancer now uses
946 bit-vector for active PEs.
947 03/21/2000 gzheng Added support for marking user-defined events
948 in trace-summary.
949 03/28/2000 wilmarth Added CMK_TRUECRASH. Very helpful for
950 post-mortem debugging of Charm++ programs on
951 net-* versions.
952 03/31/2000 jdesouza Added Fortran90 support to the Charm++
953 interface translator.
954 03/09/2000 milind Added support for -LANG and -rpath options
955 in charmc for Origin2000.
956 04/28/2000 milind Added prioritized converse threads.
957 05/01/2000 milind Added test programs for TeMPO, AMPI and irecv.
958 05/04/2000 milind New supported platform: mpi-sp.
959 05/04/2000 gzheng Added irecv pingpong program.
960 05/17/2000 olawlor Each chare, group and array element now has to
961 have migration constructor.
962 05/24/2000 milind Added Jacobi3D programs for irecv and AMPI both.
963 05/24/2000 milind Made migratable an optional attribute of
964 chares, groups, and nodegroups.
965 Arrays are by default migratable.
966 05/29/2000 paranjpy Added pup methods to arrays, reductions etc
967 internal objects.
968 06/13/2000 milind Made CtvInitialize idempotent. That is, it
969 can be called by any number of threads now,
970 only the first one will actually do
971 CtvInitialize.
972 06/20/2000 milind Added a simple test program for the FEM
973 framework.
974 07/06/2000 milind Imported Metis 4.0 sources in the CVS tree.
975 Also added code to make metis libraries and
976 executables to Makefile.
977 07/07/2000 milind Added more meaningfull error messages using
978 perror in addition to a cryptic error codes in
979 net-* versions.
980 07/10/2000 milind fem and femf are now recognized as "languages"
981 by charmc.
982 07/10/2000 saboo Added the derived datatypes library.
983 07/13/2000 milind Added +idle_timeout functionality. It takes a
984 commandline parameter denoting milliseconds of
985 maximum consecutive idle time allowed per
986 processor.
987 07/14/2000 milind Added group multicast. Added
988 CkSendMsgBranchMulti, CldEnqueueMulti, and
989 translator changes to support it.
990 07/14/2000 milind SUPER_INSTALL now takes "-*" arguments prior
991 to the target, that will be passed to make as
992 "makeflags". This makes it easy to suppress
993 make's output of commands etc (with the -s
994 flag). As a result of this, several Makefiles
995 have been massaged.
996 07/18/2000 milind Added support for using "dbx" on suns as
997 debugger.
998 07/19/2000 milind Added ability to tracemode projections which
999 produces binary trace files. Use flag
1000 +binary-trace on the command line.
1001 07/26/2000 milind Separated AMPI from TeMPO.
1002 07/28/2000 milind Added test programs to test reduce, alltoall
1003 and allreduce functionality of AMPI.
1004 08/02/2000 milind Added an option to let the user specify which
1005 "xterm" to use. For example, on some systems
1006 (CDE), only dtterm is installed. So, by
1007 putting ++xterm dtterm on the conv-host
1008 commandline, one can use dtterm when ++in-xterm
1009 option is specified on conv-host commandline.
1010 08/14/2000 milind FEM Framework: Added capabilities to handle
1011 esoteric meshes to standalone offline programs.
1012 Makefile now produces gmap and fgmap programs,
1013 which are used for this purpose. They convert
1014 the mesh to a graph before partitioning it
1015 using Metis.
1016 08/24/2000 milind Added the 2D crack propagation program as a
1017 test program for FEM framework.
1018 08/25/2000 milind Initial implementation of isomalloc-based
1019 threads. This implementation uses a fixed
1020 stack size for all threads (can be set at
1021 runtime.)
1022 08/26/2000 milind Added a macro CtvAccessOther that lets you
1023 get/set a Ctv variable of any thread. It
1024 should be invoked as CtvAccessOther(thread,
1025 varname); Added CthGetData function to each of
1026 the threads implementation. This function is
1027 used in the CtvAccessOther macro.
1028 08/27/2000 milind FEM Framework: Separated mesh to graph
1029 conversion capability into a separate program.
1030 This way, the generated graph can be partitioned
1031 repeatedly.
1032 09/04/2000 milind Added the class static readonly variables to
1033 ci file syntax.
1034 09/05/2000 milind FEM Framework: A very fast O(n) algorithm for
1035 mesh2graph , uses more memory, but the tradeoff
1036 was worth it. Coded by Karthik Mahesh, minor
1037 optimizations by Milind.
1038 09/05/2000 milind Added a barebones charm kernel scheduling
1039 overhead measurement program.
1040 09/15/2000 milind Added pup support for AMPI and FEM framework.
1041 09/20/2000 olawlor Added capability to have an array of base type
1042 where individual element could be of derived
1043 types.
1044 10/03/2000 gzheng New supported platform: net-linux-axp
1045 10/05/2000 skumar2 Added program littleMD to the test suite.
1046 10/07/2000 skumar2 New job scheduler (Faucets projects).
1047 10/15/2000 milind Improved support for Fortran90 in charmc.
1048 11/04/2000 jdesouza Made the Faucets scheduler multi-threaded.
1049 11/05/2000 olawlor FEM Framework: supports multiple element types,
1050 mesh re-assembly, etc.
1051 11/15/2000 gzheng New platform support: net-cygwin
1052 11/18/2000 gzheng conv-host no longer needs /bin/csh to start
1053 remote program. set
1054 CMK_CONV_HOST_CSH_UNAVAILABLE to 1 to use
1055 /bin/sh instead.
1056 11/25/2000 milind Finished experimental implementation of
1057 converse-threads based on co-operative pthreads.
1058 11/25/2000 milind Added a benchmark suite of all pingpongs in
1059 Charm++.
1060 11/28/2000 milind Removed deletion of _idx at the end of every
1061 send or doneInserting call. Instead now it is
1062 in the destructor of the proxy. This allows us
1063 to cache proxies, when proxy creation becomes
1064 a bottleneck.
1065 11/28/2000 olawlor Added "seek blocks" to puppers. This should
1066 allow out-of-order pup'ing without the ugliness
1067 of getBuf; and in a way that works with all
1068 PUP::ers.
1069 11/29/2000 olawlor Simplified and regularized command-line-argument
1070 handling.
1071 11/29/2000 milind AMPI: Added multiple-communicators capability.
1072 12/05/2000 gzheng Now /bin/sh is default shell to fork node
1073 program on remote machines.
1074 12/13/2000 olawlor Added charmrun wrapper for poe on mpi-sp.
1075 12/14/2000 milind Added bluegene emulator sources and test
1076 programs. Added "bluegene" as a language known
1077 to charmc. Makefile now has a target called
1078 bluegene. Added preliminary bluegene
1079 documentation. (copied from Arun's webpage.)
1080 12/15/2000 gzheng f90charm addition to Makefile and charmc. Also,
1081 added fixed size arrays support to f90charm. A
1082 test program f90charm/hello is checked in.
1083 12/17/2000 milind Added rtest test program. Contributed by jim to
1084 test Converse message transmission.
1085 12/20/2000 olawlor Added charmconfig script. Enables automatic
1086 determination of C++ compiler properties,
1087 replacing the verbose and error-prone
1088 conv-mach.h entries for CMK_BOOL,
1089 CMK_STL_USE_DOT_H, CMK_CPP_CAST_OK, ...
1090 12/20/2000 olawlor Charm++ Arrays optimizations: Key and object
1091 now variable-length fields, instead of pointers.
1092 This extra flexibility lets us save many
1093 dynamic allocations in the array framework.
1094 12/20/2000 olawlor Added PUP::able support-- dynamic type
1095 identification, allocation, and deletion.
1096 Allows you to write: p(objPtr); and
1097 objPointer will be properly identified,
1098 allocated, packed, and deallocated (depending
1099 on the PUP::er). Requires you to register any
1100 such classes with DECLARE_PUPable and
1101 DEFINE_PUPable.
1102 12/20/2000 olawlor Arrays optimizations: Made CkArrayIndex
1103 fixed-size. This significantly improves
1104 messaging speed (7 us instead of 10 us
1105 roundtrip). Move spring cleaning check into a
1106 CcdCallFnAfter, which gains more speed (down to
1107 4 us roundtrip).
1108 12/20/2000 olawlor More optimizations: Minor speed tweaks--
1109 conv-ccs.c uses hashtable for handler lookup;
1110 conv-conds skips timer test until needed;
1111 convcore.c scheduler loop optmizations (no
1112 superfluous EndIdle calls); threads.c
1113 CMK_OPTIMIZE-> no mprotect.
1114 12/20/2000 olawlor More Optimizations: Minor speed tweaks-- ck.C
1115 groups cldEnqueue skip; init.h defines
1116 CkLocalBranch inline; and supporting changes.
1117 12/22/2000 gzheng IA64 support for Converse user level threads.
1118 01/02/2001 olawlor CCS: Minor update-- enabled CcsProbe, cleaned
1119 up superflous debug messages in server, added
1120 Java interface (originally written for
1121 AppSpecter).
1122 01/09/2001 gzheng charmconfig converted to autoconf style, need
1123 to change configure.in and conv-autoconfig.h.in,
1124 and run autoconf to get configure and copy to
1125 charmconfig. added fortran subroutine name
1126 test and get libpthread.a
1127 01/10/2001 milind Added telnet method of getting libpthread.a
1128 from charm webserver.
1129 01/11/2001 olawlor Moved projections files here from
1130 CVSROOT/projections-java. Added fast Java
1131 versions of the .log file input routines in
1132 LogReader, LogLoader, LogAnalyzer, and
1133 UsageCalc. Added "U.java" user interface
1134 utility file, allowing times to be input in
1135 seconds, milliseconds, or microseconds,
1136 instead of just microseconds.
1137 01/15/2001 gzheng add +trace-root to specify the directory to
1138 put log files in. this is need in Scyld cluster
1139 where there is no NFS mounting and no i/o
1140 access to home directory sharing on nodes.
1141 01/15/2001 milind Made AMPI into a f90 module instead of
1142 'ampif.h' inclusion. AMPI f90 bindings are
1143 now more inclusive. Fixed argc,argv handling
1144 bugs in ArgsInfo message. Fixed a bug in pup
1145 that caused thread not to be sized, but was
1146 packed nevertheless. Moved irecv to waitall
1147 instead of at in ampi_start. Made
1148 AMPI_COMM_WORLD to be 0, because it clashed
1149 with wildcard(-1). AMPI_COMM_UNIVERSE is now
1150 handled properly in the AMPI module.
1151 C/C++ data members are NOT visible to
1152 Fortran 90.
1153 01/18/2001 gzheng New supported platform: net-linux-scyld
1154 01/20/2001 olawlor Moved array index field from CMessage_* to the
1155 Ck envelope itself. This is the right thing
1156 to do, because any message may be sent to/from
1157 an array element. To reduce the wasted space
1158 in a message, a union is used to overlay the
1159 fields for the various possible message types.
1160 01/29/2001 olawlor Freed charmrun on net-* version from using
1161 remote shell to fork off processes. One can now
1162 use a daemon provided in the distribution.
1163 02/07/2001 olawlor Added debugging support to puppers.
1164 02/13/2000 gzheng Added ++local option to charmrun to start node
1165 program locally without any daemon; fix the
1166 hang program if you type wrong pgm name in
1167 scyld version, and redirect all output to
1168 /dev/null, otherwise all node program can send
1169 its output to console in scyld. Also implemented ++local in net-win32 version.
1170 02/26/2000 milind Changed the varsize syntax. Now one can specify
1171 actual varsize arrays in the interface file
1172 and have the translator generate alloc, pack
1173 and unpack.
1175 --------------------------------------------------------------------------------
1176 Bug Fixes:
1177 --------------------------------------------------------------------------------
1179 10/29/1999 milind Replaced jmemcpy by memcpy in net versions, as
1180 it was causing a bit to flip (bug reported
1181 by jim.)
1182 10/29/1999 milind Fixed multiline macros in all header files.
1183 02/05/2000 milind Fixed linking errors by getting the order of
1184 libraries right from the charmc command-line.
1185 02/18/2000 paranjpy Fixed Charm++ initialization bug on SMPs.
1186 02/21/2000 milind Fixed a context-switching bug in mipspro version
1187 of QuickThreads.
1188 02/25/2000 milind Charm++ interface translator was segfaulting
1189 on interface file errors. Fixed that. Also,
1190 added linenumbers to error messages.
1191 03/02/2000 milind Made CCS work on SMPs.
1192 03/07/2000 milind Made ConverseInit consistent with the manual on
1193 Origin2000 version.
1194 04/18/2000 milind Fixed a bug in CkWaitFuture, which was caching
1195 a variable locally, while it was changed by
1196 another thread.
1197 05/04/2000 paranjpy Fixed argv deletion bug on net-win32-smp.
1198 06/08/2000 milind sp3 version: changed optimization flags, which
1199 where power2 processor-specific.
1200 06/20/2000 milind mpi-* versions: Fixed ConverseExit since it was
1201 not obeying the following statement in the MPI
1202 standard: The user must ensure that all pending
1203 communications involving a process completes
1204 before the process calls MPI_FINALIZE.
1205 07/05/2000 milind Fixed a nasty bug in charmc in the -cp option.
1206 It used to append the name provided to -o flag
1207 to the directory provided to the -cp flag.
1208 Thus, -o ../pgm -cp ../bin options meant that
1209 the pgm would be copied to ../bin/.., which is
1210 not the expected behavior. This fix correctly
1211 copies pgm to ../bin.
1212 07/07/2000 milind Removed variable arg_myhome, as it was not
1213 being used anywhere, and also, setting it was
1214 causing problems of env var HOME was not set.
1215 07/27/2000 milind thishandle for the arrayelement was not being
1216 correctly set. Bug was reported by Neelam.
1217 08/26/2000 milind Origin2000: Changed the page alignment to
1218 reflect the mmap alignment. The mmap man page
1219 specifically states that it is not the same as
1220 page size.
1221 09/02/2000 milind Fixed a bug in code generated for threaded
1222 (void) entry methods of array elements. The
1223 dummy message that is passed to that method in
1224 a thread has to be deleted before calling the
1225 object method, because upon object method's
1226 return, the thread might have migrated.
1227 09/03/2000 olawlor Minor fix-fixes: 1.) Change to LBObjid hash
1228 function would fail for >4-int object indices.
1229 Replaced with proper function, which also
1230 preserves the 1-int case. 2.) Array element
1231 sends must go via the message queue to prevent
1232 stack build-up for deep single-processor call
1233 chains. These might happen, e.g., in a driver
1234 element calling itself for the main time loop.
1235 Messages are now properly noted as sent, then
1236 wait through the queue for delivery. This
1237 entailed minor reorganization of the message
1238 delivery subsystem.
1239 09/21/2000 olawlor Tiny SMP thread fix-- registrations of a
1240 thread-private variable now reserve space on
1241 calls after the first. This wastes space for
1242 multiple CthInitialize's-- it's a quick hack to
1243 get threads working again on SMP versions.
1244 10/16/2000 olawlor A few CCS fixes: -Added split-phase reply
1245 (delay reply indefinitely) -Cleaned up error
1246 handling -Pass user data as "void *" instead of
1247 "char *"
1248 11/03/2000 wilmarth Removed 0 size array allocation in Charm++
1249 quiescence detection.
1250 11/20/2000 gzheng Rewrote part of Fiber thread, including a bug
1251 fix for a the non thread-safe function, and a
1252 different fiber free strategy.
1253 11/29/2000 gzheng The LB init procedure tried to allocate
1254 65536*160 as initial size, which is 10M memory
1255 for communication table, which is too big.
1256 Cut it down to roughly 1M, and it can expand
1257 in later code.
1258 12/05/2000 gzheng In many cases, conv-host exits without print
1259 out the error message from remote shell. try
1260 to fix it by calling sync to flush the pipe
1261 before exit 1.
1262 12/10/2000 milind net-linux: Made static linking the default
1263 option because dynamic linking runtime causes
1264 isomalloc threads to crash.
1265 12/18/2000 milind Increased portability of isomalloc threads by
1266 removing dependence on alloca.
1267 12/28/2000 milind Fixed ctrl-getone abort bug on SMP.
1268 12/28/2000 milind Made _groupTable a pointer on which a
1269 constructor is explicitly called. Since it
1270 was a Cpv variable, its constructor was not
1271 called by default in case of an SMP version.
1272 12/29/2000 olawlor Prevent infinite copy constructor recursion on
1273 Origin2000.
1274 01/10/2001 olawlor Added "explicit" keyword to remove ambiguity
1275 for KCC, which was confused by the private
1276 PUP::er(int) "cast" constructor and the operator
1277 |(PUP::er &p,T &t) into rejecting all operator|
1278 (int,int) as ambiguous.
1279 2001/01/17 gzheng fix the charmconfig bug on paragon-red: the
1280 failure testing of fortran won't stop the
1281 compilation.
1282 01/20/2001 olawlor Arrays reduction: Fixed bug-- reduction may end
1283 because all contributors migrate away.
1284 01/29/2001 olawlor Fix heap-corrupting bug-- call ->init() on
1285 nodeGroupTable, which sets the "pending"
1286 message queue to NULL. This prevents a nasty
1287 delete-unitialized-data bug later on. Also
1288 delayed queue creation until messages actually
1289 arrive.
1291 --------------------------------------------------------------------------------
1292 Documentation Changes:
1293 --------------------------------------------------------------------------------
1295 01/31/2000 milind Installation manual: Fixed bugs pointed out by
1296 quantum TA
1297 02/28/2000 wilmarth Added a new look Charm++ manual.
1298 06/20/2000 milind Added pdflatex support to generate PDF versions
1299 of manuals from LaTeX sources.
1300 12/05/2000 milind Added Orion's FEM manual. Converted from HTML.
1301 12/10/2000 milind Added pplmanual.sty for all manuals.
1302 12/17/2000 milind Added master-slave library documentation to
1303 convext manual.
1304 12/21/2000 saboo Added DDT documentation.
1305 01/02/2001 olawlor Updated for new CCS version.
1307 --------------------------------------------------------------------------------
1308 Other Changes:
1309 --------------------------------------------------------------------------------
1311 10/24/1999 olawlor charmc is changed to Bourne shell script
1312 instead of csh. All conv-mach.csh are
1313 replaced by conv-mach.sh.
1314 10/25/1999 olawlor SUPER_INSTALL is converted to use bourne shell.
1315 10/28/1999 milind All Makefiles now take OPTS commandline
1316 arguments.
1317 01/16/2000 olawlor Simplified Charm++ interface translator.
1318 02/23/2000 ruiliu Changed rand() calls from all over the codes
1319 to the new Converse random number generator.
1320 02/26/2000 milind Simplified the converse scheduler loop by
1321 combining the maxmsgs and poll modes.
1322 08/31/2000 milind Imported system documentation into the CVS tree.
1323 Also added super_install target for docs with
1324 necessary Makefile modifications.
1325 09/08/2000 olawlor Made soft links use relative pathnames instead
1326 of absolute. This lets you move a charm++
1327 installation without having to recompile
1328 anything.
1329 09/11/2000 olawlor Grouped commonly needed code in the new util
1330 directory. Also, added pup_c a C wrapper for
1331 puppers.
1332 09/11/2000 olawlor Slightly reorganized header structure. Now no
1333 headers should need to be listed twice (once in
1334 ALLHEADERS, again in CKHEADERS). Now headers
1335 are soft-linked instead of copied. This makes
1336 development much easier. Added support for the
1337 new Common/util directory.
1338 09/21/2000 olawlor Major reorganization of net-* codes. Now all
1339 the TCP socket routines are in separate files.
1340 Also combined windoes NT code with unix codes.
1341 09/21/2000 olawlor Major rewrite of CCS-- underlying protocol is
1342 now binary (send/recv binary data everywhere);
1343 conv-host forwards requests to nodes; and
1344 source has been significantly re-arranged.
1345 (especially if NODE_0_IS_CONVHOST).
1346 11/22/2000 milind Removed IDL translator from distribution.
1347 12/01/2000 olawlor Renamed conv-host charmrun; added test for
1348 script conv-host. Also added charmrun for most
1349 other machines.
1350 12/17/2000 milind Moved List related data structures into
1351 cklists.h in util. Removed most of the redundant
1352 list implementations.
1353 12/20/2000 gzheng SUPER_INSTALL: format the output of list of
1354 versions and make the help page fit into one
1355 page of xterm.
1356 12/24/2000 milind Added test-{charm,converse,ampi,fem} targets to
1357 super_install.
1358 12/28/2000 milind net-sol-smp now uses pthreads.
1359 01/29/2001 olawlor Merged windowsNT and unix build procedures by
1360 basing the Windows build on cygwin. Added
1361 scripts to deal with unix and windows
1362 differences.