ChangeLog 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367
  1. # StarPU --- Runtime system for heterogeneous multicore architectures.
  2. #
  3. # Copyright (C) 2009-2012 Université de Bordeaux 1
  4. # Copyright (C) 2010, 2011, 2012, 2013 Centre National de la Recherche Scientifique
  5. #
  6. # StarPU is free software; you can redistribute it and/or modify
  7. # it under the terms of the GNU Lesser General Public License as published by
  8. # the Free Software Foundation; either version 2.1 of the License, or (at
  9. # your option) any later version.
  10. #
  11. # StarPU is distributed in the hope that it will be useful, but
  12. # WITHOUT ANY WARRANTY; without even the implied warranty of
  13. # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  14. #
  15. # See the GNU Lesser General Public License in COPYING.LGPL for more details.
  16. StarPU 1.1.0 (svn revision xxxx)
  17. ==============================================
  18. New features:
  19. * OpenGL interoperability support.
  20. * Capability to store compiled OpenCL kernels on the file system
  21. * Capability to load compiled OpenCL kernels
  22. * Performance models measurements can now be provided explicitly by
  23. applications.
  24. * Capability to emit communication statistics when running MPI code
  25. * Add starpu_unregister_submit, starpu_data_acquire_on_node and
  26. starpu_data_invalidate_submit
  27. * New functionnality to wrapper starpu_insert_task to pass a array of
  28. data_handles via the parameter STARPU_DATA_ARRAY
  29. * Enable GPU-GPU direct transfers.
  30. * GCC plug-in
  31. - Add `registered' attribute
  32. - A new pass was added that warns about the use of possibly
  33. unregistered memory buffers.
  34. * SOCL
  35. - Manual mapping of commands on specific devices is now
  36. possible
  37. - SOCL does not require StarPU CPU tasks anymore. CPU workers
  38. are automatically disabled to enhance performance of OpenCL
  39. CPU devices
  40. * New interface: COO matrix.
  41. * Data interfaces: The pack operation of user-defined data interface
  42. defines a new parameter count which should be set to the size of
  43. the buffer created by the packing of the data.
  44. * MPI:
  45. - Communication statistics for MPI can only be enabled at
  46. execution time by defining the environment variable
  47. STARPU_COMM_STATS
  48. - Communication cache mechanism is enabled by default, and can
  49. only be disabled at execution time by setting the
  50. environment variable STARPU_MPI_CACHE to 0.
  51. - Initialisation functions starpu_mpi_initialize_extended()
  52. and starpu_mpi_initialize() have been made deprecated. One
  53. should now use starpu_mpi_init(int *, char ***, int). The
  54. last parameter indicates if MPI should be initialised.
  55. - Collective detached operations have new parameters, a
  56. callback function and a argument. This is to be consistent
  57. with the detached point-to-point communications.
  58. - When exchanging user-defined data interfaces, the size of
  59. the data is the size returned by the pack operation, i.e
  60. data with dynamic size can now be exchanged with StarPU-MPI.
  61. - New functionality starpu_mpi_irecv_probe_detached which
  62. first tests if the message is available before calling MPI_Recv.
  63. * Add experimental simgrid support, to simulate execution with various
  64. number of CPUs, GPUs, amount of memory, etc.
  65. * Add support for OpenCL simulators (which provide simulated execution time)
  66. * Add support for Temanejo, a task graph debugger
  67. * Theoretical bound lp output now includes data transfer time.
  68. * Update OpenCL driver to only enable CPU devices (the environment
  69. variable STARPU_OPENCL_ONLY_ON_CPUS must be set to a positive
  70. value when executing an application)
  71. * Add Scheduling contexts to separate computation resources
  72. - Scheduling policies take into account the set of resources corresponding
  73. to the context it belongs to
  74. - Add support to dynamically change scheduling contexts
  75. (Create and Delete a context, Add Workers to a context, Remove workers from a context)
  76. - Add support to indicate to which contexts the tasks are submitted
  77. * Add the Hypervisor to manage the Scheduling Contexts automatically
  78. - The Contexts can be registered to the Hypervisor
  79. - Only the registered contexts are managed by the Hypervisor
  80. - The Hypervisor can detect the initial distribution of resources of
  81. a context and constructs it consequently (the cost of execution is required)
  82. - Several policies can adapt dynamically the distribution of resources
  83. in contexts if the initial one was not appropriate
  84. - Add a platform to implement new policies of redistribution of resources
  85. Changes:
  86. * Fix the block filter functions.
  87. * Fix StarPU-MPI on Darwin.
  88. * The FxT code can now be used on systems other than Linux.
  89. * Keep only one hashtable implementation common/uthash.h
  90. * The cache of starpu_mpi_insert_task is fixed and thus now enabled by
  91. default.
  92. * Improve starpu_machine_display output.
  93. * Standardize objects name in the performance model API
  94. * SOCL
  95. - Virtual SOCL device has been removed
  96. - Automatic scheduling still available with command queues not
  97. assigned to any device
  98. - Remove modified OpenCL headers. ICD is now the only supported
  99. way to use SOCL.
  100. - SOCL test suite is only run when environment variable
  101. SOCL_OCL_LIB_OPENCL is defined. It should contain the location
  102. of the libOpenCL.so file of the OCL ICD implementation.
  103. * Fix main memory leak on multiple unregister/re-register.
  104. * Improve hwloc detection by configure
  105. * Cell:
  106. - It is no longer possible to enable the cell support via the
  107. gordon driver
  108. * Fix data transfer arrows in paje traces
  109. * The "heft" scheduler no longer exists. Users should now pick "dmda"
  110. instead.
  111. * StarPU can now use poti to generate paje traces.
  112. * Rename scheduling policy "parallel greedy" to "parallel eager"
  113. Small changes:
  114. * STARPU_NCPU should now be used instead of STARPU_NCPUS. STARPU_NCPUS is
  115. still available for compatibility reasons.
  116. * include/starpu.h includes all include/starpu_*.h files, applications
  117. therefore only need to have #include <starpu.h>
  118. * Active task wait is now included in blocked time.
  119. * Fix GCC plugin linking issues starting with GCC 4.7.
  120. * Fix forcing calibration of never-calibrated archs.
  121. * CUDA applications are no longer compiled with the "-arch sm_13"
  122. option. It is specifically added to applications which need it.
  123. StarPU 1.0.3 (svn revision 7379)
  124. ==============================================
  125. Changes:
  126. * Several bug fixes in the build system
  127. * Bug fixes in source code for non-Linux systems
  128. * Fix generating FXT traces bigger than 64MiB.
  129. * Improve ENODEV error detections in StarPU FFT
  130. StarPU 1.0.2 (svn revision xxx)
  131. ==============================================
  132. Changes:
  133. * Add starpu_block_shadow_filter_func_vector and an example.
  134. * Add tag dependency in trace-generated DAG.
  135. * Fix CPU binding for optimized CPU-GPU transfers.
  136. * Fix parallel tasks CPU binding and combined worker generation.
  137. * Fix generating FXT traces bigger than 64MiB.
  138. StarPU 1.0.1 (svn revision 6659)
  139. ==============================================
  140. Changes:
  141. * hwloc support. Warn users when hwloc is not found on the system and
  142. produce error when not explicitely disabled.
  143. * Several bug fixes
  144. * GCC plug-in
  145. - Add `#pragma starpu release'
  146. - Fix bug when using `acquire' pragma with function parameters
  147. - Slightly improve test suite coverage
  148. - Relax the GCC version check
  149. * Update SOCL to use new API
  150. * Documentation improvement.
  151. StarPU 1.0.0 (svn revision 6306)
  152. ==============================================
  153. The extensions-again release
  154. New features:
  155. * Add SOCL, an OpenCL interface on top of StarPU.
  156. * Add a gcc plugin to extend the C interface with pragmas which allows to
  157. easily define codelets and issue tasks.
  158. * Add reduction mode to starpu_mpi_insert_task.
  159. * A new multi-format interface permits to use different binary formats
  160. on CPUs & GPUs, the conversion functions being provided by the
  161. application and called by StarPU as needed (and as less as
  162. possible).
  163. * Deprecate cost_model, and introduce cost_function, which is provided
  164. with the whole task structure, the target arch and implementation
  165. number.
  166. * Permit the application to provide its own size base for performance
  167. models.
  168. * Applications can provide several implementations of a codelet for the
  169. same architecture.
  170. * Add a StarPU-Top feedback and steering interface.
  171. * Permit to specify MPI tags for more efficient starpu_mpi_insert_task
  172. Changes:
  173. * Fix several memory leaks and race conditions
  174. * Make environment variables take precedence over the configuration
  175. passed to starpu_init()
  176. * Libtool interface versioning has been included in libraries names
  177. (libstarpu-1.0.so, libstarpumpi-1.0.so,
  178. libstarpufft-1.0.so, libsocl-1.0.so)
  179. * Install headers under $includedir/starpu/1.0.
  180. * Make where field for struct starpu_codelet optional. When unset, its
  181. value will be automatically set based on the availability of the
  182. different XXX_funcs fields of the codelet.
  183. * Define access modes for data handles into starpu_codelet and no longer
  184. in starpu_task. Hence mark (struct starpu_task).buffers as
  185. deprecated, and add (struct starpu_task).handles and (struct
  186. starpu_codelet).modes
  187. * Fields xxx_func of struct starpu_codelet are made deprecated. One
  188. should use fields xxx_funcs instead.
  189. * Some types were renamed for consistency. when using pkg-config libstarpu,
  190. starpu_deprecated_api.h is automatically included (after starpu.h) to
  191. keep compatibility with existing software. Other changes are mentioned
  192. below, compatibility is also preserved for them.
  193. To port code to use new names (this is not mandatory), the
  194. tools/dev/rename.sh script can be used, and pkg-config starpu-1.0 should
  195. be used.
  196. * The communication cost in the heft and dmda scheduling strategies now
  197. take into account the contention brought by the number of GPUs. This
  198. changes the meaning of the beta factor, whose default 1.0 value should
  199. now be good enough in most case.
  200. Small features:
  201. * Allow users to disable asynchronous data transfers between CPUs and
  202. GPUs.
  203. * Update OpenCL driver to enable CPU devices (the environment variable
  204. STARPU_OPENCL_ON_CPUS must be set to a positive value when
  205. executing an application)
  206. * struct starpu_data_interface_ops --- operations on a data
  207. interface --- define a new function pointer allocate_new_data
  208. which creates a new data interface of the given type based on
  209. an existing handle
  210. * Add a field named magic to struct starpu_task which is set when
  211. initialising the task. starpu_task_submit will fail if the
  212. field does not have the right value. This will hence avoid
  213. submitting tasks which have not been properly initialised.
  214. * Add a hook function pre_exec_hook in struct starpu_sched_policy.
  215. The function is meant to be called in drivers. Schedulers
  216. can use it to be notified when a task is about being computed.
  217. * Add codelet execution time statistics plot.
  218. * Add bus speed in starpu_machine_display.
  219. * Add a STARPU_DATA_ACQUIRE_CB which permits to inline the code to be
  220. done.
  221. * Add gdb functions.
  222. * Add complex support to LU example.
  223. * Permit to use the same data several times in write mode in the
  224. parameters of the same task.
  225. Small changes:
  226. * Increase default value for STARPU_MAXCPUS -- Maximum number of
  227. CPUs supported -- to 64.
  228. * Add man pages for some of the tools
  229. * Add C++ application example in examples/cpp/
  230. * Add an OpenMP fork-join example.
  231. * Documentation improvement.
  232. StarPU 0.9 (svn revision 3721)
  233. ==============================================
  234. The extensions release
  235. * Provide the STARPU_REDUX data access mode
  236. * Externalize the scheduler API.
  237. * Add theoretical bound computation
  238. * Add the void interface
  239. * Add power consumption optimization
  240. * Add parallel task support
  241. * Add starpu_mpi_insert_task
  242. * Add profiling information interface.
  243. * Add STARPU_LIMIT_GPU_MEM environment variable.
  244. * OpenCL fixes
  245. * MPI fixes
  246. * Improve optimization documentation
  247. * Upgrade to hwloc 1.1 interface
  248. * Add fortran example
  249. * Add mandelbrot OpenCL example
  250. * Add cg example
  251. * Add stencil MPI example
  252. * Initial support for CUDA4
  253. StarPU 0.4 (svn revision 2535)
  254. ==============================================
  255. The API strengthening release
  256. * Major API improvements
  257. - Provide the STARPU_SCRATCH data access mode
  258. - Rework data filter interface
  259. - Rework data interface structure
  260. - A script that automatically renames old functions to accomodate with the new
  261. API is available from https://scm.gforge.inria.fr/svn/starpu/scripts/renaming
  262. (login: anonsvn, password: anonsvn)
  263. * Implement dependencies between task directly (eg. without tags)
  264. * Implicit data-driven task dependencies simplifies the design of
  265. data-parallel algorithms
  266. * Add dynamic profiling capabilities
  267. - Provide per-task feedback
  268. - Provide per-worker feedback
  269. - Provide feedback about memory transfers
  270. * Provide a library to help accelerating MPI applications
  271. * Improve data transfers overhead prediction
  272. - Transparently benchmark buses to generate performance models
  273. - Bind accelerator-controlling threads with respect to NUMA locality
  274. * Improve StarPU's portability
  275. - Add OpenCL support
  276. - Add support for Windows
  277. StarPU 0.2.901 aka 0.3-rc1 (svn revision 1236)
  278. ==============================================
  279. The asynchronous heterogeneous multi-accelerator release
  280. * Many API changes and code cleanups
  281. - Implement starpu_worker_get_id
  282. - Implement starpu_worker_get_name
  283. - Implement starpu_worker_get_type
  284. - Implement starpu_worker_get_count
  285. - Implement starpu_display_codelet_stats
  286. - Implement starpu_data_prefetch_on_node
  287. - Expose the starpu_data_set_wt_mask function
  288. * Support nvidia (heterogeneous) multi-GPU
  289. * Add the data request mechanism
  290. - All data transfers use data requests now
  291. - Implement asynchronous data transfers
  292. - Implement prefetch mechanism
  293. - Chain data requests to support GPU->RAM->GPU transfers
  294. * Make it possible to bypass the scheduler and to assign a task to a specific
  295. worker
  296. * Support restartable tasks to reinstanciate dependencies task graphs
  297. * Improve performance prediction
  298. - Model data transfer overhead
  299. - One model is created for each accelerator
  300. * Support for CUDA's driver API is deprecated
  301. * The STARPU_WORKERS_CUDAID and STARPU_WORKERS_CPUID env. variables make it possible to
  302. specify where to bind the workers
  303. * Use the hwloc library to detect the actual number of cores
  304. StarPU 0.2.0 (svn revision 1013)
  305. ==============================================
  306. The Stabilizing-the-Basics release
  307. * Various API cleanups
  308. * Mac OS X is supported now
  309. * Add dynamic code loading facilities onto Cell's SPUs
  310. * Improve performance analysis/feedback tools
  311. * Application can interact with StarPU tasks
  312. - The application may access/modify data managed by the DSM
  313. - The application may wait for the termination of a (set of) task(s)
  314. * An initial documentation is added
  315. * More examples are supplied
  316. StarPU 0.1.0 (svn revision 794)
  317. ==============================================
  318. First release.
  319. Status:
  320. * Only supports Linux platforms yet
  321. * Supported architectures
  322. - multicore CPUs
  323. - NVIDIA GPUs (with CUDA 2.x)
  324. - experimental Cell/BE support
  325. Changes:
  326. * Scheduling facilities
  327. - run-time selection of the scheduling policy
  328. - basic auto-tuning facilities
  329. * Software-based DSM
  330. - transparent data coherency management
  331. - High-level expressive interface
  332. # Local Variables:
  333. # mode: text
  334. # coding: utf-8
  335. # ispell-local-dictionary: "american"
  336. # End: