ChangeLog 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306
  1. # StarPU --- Runtime system for heterogeneous multicore architectures.
  2. #
  3. # Copyright (C) 2009-2012 Université de Bordeaux 1
  4. # Copyright (C) 2010, 2011, 2012 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 possible
  36. Changes:
  37. * Fix the block filter functions.
  38. * Fix StarPU-MPI on Darwin.
  39. * The FxT code can now be used on systems other than Linux.
  40. * Keep only one hashtable implementation common/uthash.h
  41. * The cache of starpu_mpi_insert_task is fixed and thus now enabled by
  42. default.
  43. * Improve starpu_machine_display output.
  44. * Standardize objects name in the performance model API
  45. * SOCL
  46. - Virtual SOCL device has been removed
  47. - Automatic scheduling still available with command queues not
  48. assigned to any device
  49. - Remove modified OpenCL headers. ICD is now the only supported
  50. way to use SOCL.
  51. - SOCL test suite is only run when environment variable
  52. SOCL_OCL_LIB_OPENCL is defined. It should contain the location
  53. of the libOpenCL.so file of the OCL ICD implementation.
  54. * Fix main memory leak on multiple unregister/re-register.
  55. * Improve hwloc detection by configure
  56. Small changes:
  57. * STARPU_NCPU should now be used instead of STARPU_NCPUS. STARPU_NCPUS is
  58. still available for compatibility reasons.
  59. * include/starpu.h includes all include/starpu_*.h files, applications
  60. therefore only need to have #include <starpu.h>
  61. * Active task wait is now included in blocked time.
  62. StarPU 1.0.3 (svn revision 7379)
  63. ==============================================
  64. Changes:
  65. * Several bug fixes in the build system
  66. * Bug fixes in source code for non-Linux systems
  67. * Fix generating FXT traces bigger than 64MiB.
  68. * Improve ENODEV error detections in StarPU FFT
  69. StarPU 1.0.2 (svn revision xxx)
  70. ==============================================
  71. Changes:
  72. * Add starpu_block_shadow_filter_func_vector and an example.
  73. * Add tag dependency in trace-generated DAG.
  74. * Fix CPU binding for optimized CPU-GPU transfers.
  75. * Fix parallel tasks CPU binding and combined worker generation.
  76. * Fix generating FXT traces bigger than 64MiB.
  77. StarPU 1.0.1 (svn revision 6659)
  78. ==============================================
  79. Changes:
  80. * hwloc support. Warn users when hwloc is not found on the system and
  81. produce error when not explicitely disabled.
  82. * Several bug fixes
  83. * GCC plug-in
  84. - Add `#pragma starpu release'
  85. - Fix bug when using `acquire' pragma with function parameters
  86. - Slightly improve test suite coverage
  87. - Relax the GCC version check
  88. * Update SOCL to use new API
  89. * Documentation improvement.
  90. StarPU 1.0.0 (svn revision 6306)
  91. ==============================================
  92. The extensions-again release
  93. New features:
  94. * Add SOCL, an OpenCL interface on top of StarPU.
  95. * Add a gcc plugin to extend the C interface with pragmas which allows to
  96. easily define codelets and issue tasks.
  97. * Add reduction mode to starpu_mpi_insert_task.
  98. * A new multi-format interface permits to use different binary formats
  99. on CPUs & GPUs, the conversion functions being provided by the
  100. application and called by StarPU as needed (and as less as
  101. possible).
  102. * Deprecate cost_model, and introduce cost_function, which is provided
  103. with the whole task structure, the target arch and implementation
  104. number.
  105. * Permit the application to provide its own size base for performance
  106. models.
  107. * Applications can provide several implementations of a codelet for the
  108. same architecture.
  109. * Add a StarPU-Top feedback and steering interface.
  110. * Permit to specify MPI tags for more efficient starpu_mpi_insert_task
  111. Changes:
  112. * Fix several memory leaks and race conditions
  113. * Make environment variables take precedence over the configuration
  114. passed to starpu_init()
  115. * Libtool interface versioning has been included in libraries names
  116. (libstarpu-1.0.so, libstarpumpi-1.0.so,
  117. libstarpufft-1.0.so, libsocl-1.0.so)
  118. * Install headers under $includedir/starpu/1.0.
  119. * Make where field for struct starpu_codelet optional. When unset, its
  120. value will be automatically set based on the availability of the
  121. different XXX_funcs fields of the codelet.
  122. * Define access modes for data handles into starpu_codelet and no longer
  123. in starpu_task. Hence mark (struct starpu_task).buffers as
  124. deprecated, and add (struct starpu_task).handles and (struct
  125. starpu_codelet).modes
  126. * Fields xxx_func of struct starpu_codelet are made deprecated. One
  127. should use fields xxx_funcs instead.
  128. * Some types were renamed for consistency. when using pkg-config libstarpu,
  129. starpu_deprecated_api.h is automatically included (after starpu.h) to
  130. keep compatibility with existing software. Other changes are mentioned
  131. below, compatibility is also preserved for them.
  132. To port code to use new names (this is not mandatory), the
  133. tools/dev/rename.sh script can be used, and pkg-config starpu-1.0 should
  134. be used.
  135. * The communication cost in the heft and dmda scheduling strategies now
  136. take into account the contention brought by the number of GPUs. This
  137. changes the meaning of the beta factor, whose default 1.0 value should
  138. now be good enough in most case.
  139. Small features:
  140. * Allow users to disable asynchronous data transfers between CPUs and
  141. GPUs.
  142. * Update OpenCL driver to enable CPU devices (the environment variable
  143. STARPU_OPENCL_ON_CPUS must be set to a positive value when
  144. executing an application)
  145. * struct starpu_data_interface_ops --- operations on a data
  146. interface --- define a new function pointer allocate_new_data
  147. which creates a new data interface of the given type based on
  148. an existing handle
  149. * Add a field named magic to struct starpu_task which is set when
  150. initialising the task. starpu_task_submit will fail if the
  151. field does not have the right value. This will hence avoid
  152. submitting tasks which have not been properly initialised.
  153. * Add a hook function pre_exec_hook in struct starpu_sched_policy.
  154. The function is meant to be called in drivers. Schedulers
  155. can use it to be notified when a task is about being computed.
  156. * Add codelet execution time statistics plot.
  157. * Add bus speed in starpu_machine_display.
  158. * Add a STARPU_DATA_ACQUIRE_CB which permits to inline the code to be
  159. done.
  160. * Add gdb functions.
  161. * Add complex support to LU example.
  162. * Permit to use the same data several times in write mode in the
  163. parameters of the same task.
  164. Small changes:
  165. * Increase default value for STARPU_MAXCPUS -- Maximum number of
  166. CPUs supported -- to 64.
  167. * Add man pages for some of the tools
  168. * Add C++ application example in examples/cpp/
  169. * Add an OpenMP fork-join example.
  170. * Documentation improvement.
  171. StarPU 0.9 (svn revision 3721)
  172. ==============================================
  173. The extensions release
  174. * Provide the STARPU_REDUX data access mode
  175. * Externalize the scheduler API.
  176. * Add theoretical bound computation
  177. * Add the void interface
  178. * Add power consumption optimization
  179. * Add parallel task support
  180. * Add starpu_mpi_insert_task
  181. * Add profiling information interface.
  182. * Add STARPU_LIMIT_GPU_MEM environment variable.
  183. * OpenCL fixes
  184. * MPI fixes
  185. * Improve optimization documentation
  186. * Upgrade to hwloc 1.1 interface
  187. * Add fortran example
  188. * Add mandelbrot OpenCL example
  189. * Add cg example
  190. * Add stencil MPI example
  191. * Initial support for CUDA4
  192. StarPU 0.4 (svn revision 2535)
  193. ==============================================
  194. The API strengthening release
  195. * Major API improvements
  196. - Provide the STARPU_SCRATCH data access mode
  197. - Rework data filter interface
  198. - Rework data interface structure
  199. - A script that automatically renames old functions to accomodate with the new
  200. API is available from https://scm.gforge.inria.fr/svn/starpu/scripts/renaming
  201. (login: anonsvn, password: anonsvn)
  202. * Implement dependencies between task directly (eg. without tags)
  203. * Implicit data-driven task dependencies simplifies the design of
  204. data-parallel algorithms
  205. * Add dynamic profiling capabilities
  206. - Provide per-task feedback
  207. - Provide per-worker feedback
  208. - Provide feedback about memory transfers
  209. * Provide a library to help accelerating MPI applications
  210. * Improve data transfers overhead prediction
  211. - Transparently benchmark buses to generate performance models
  212. - Bind accelerator-controlling threads with respect to NUMA locality
  213. * Improve StarPU's portability
  214. - Add OpenCL support
  215. - Add support for Windows
  216. StarPU 0.2.901 aka 0.3-rc1 (svn revision 1236)
  217. ==============================================
  218. The asynchronous heterogeneous multi-accelerator release
  219. * Many API changes and code cleanups
  220. - Implement starpu_worker_get_id
  221. - Implement starpu_worker_get_name
  222. - Implement starpu_worker_get_type
  223. - Implement starpu_worker_get_count
  224. - Implement starpu_display_codelet_stats
  225. - Implement starpu_data_prefetch_on_node
  226. - Expose the starpu_data_set_wt_mask function
  227. * Support nvidia (heterogeneous) multi-GPU
  228. * Add the data request mechanism
  229. - All data transfers use data requests now
  230. - Implement asynchronous data transfers
  231. - Implement prefetch mechanism
  232. - Chain data requests to support GPU->RAM->GPU transfers
  233. * Make it possible to bypass the scheduler and to assign a task to a specific
  234. worker
  235. * Support restartable tasks to reinstanciate dependencies task graphs
  236. * Improve performance prediction
  237. - Model data transfer overhead
  238. - One model is created for each accelerator
  239. * Support for CUDA's driver API is deprecated
  240. * The STARPU_WORKERS_CUDAID and STARPU_WORKERS_CPUID env. variables make it possible to
  241. specify where to bind the workers
  242. * Use the hwloc library to detect the actual number of cores
  243. StarPU 0.2.0 (svn revision 1013)
  244. ==============================================
  245. The Stabilizing-the-Basics release
  246. * Various API cleanups
  247. * Mac OS X is supported now
  248. * Add dynamic code loading facilities onto Cell's SPUs
  249. * Improve performance analysis/feedback tools
  250. * Application can interact with StarPU tasks
  251. - The application may access/modify data managed by the DSM
  252. - The application may wait for the termination of a (set of) task(s)
  253. * An initial documentation is added
  254. * More examples are supplied
  255. StarPU 0.1.0 (svn revision 794)
  256. ==============================================
  257. First release.
  258. Status:
  259. * Only supports Linux platforms yet
  260. * Supported architectures
  261. - multicore CPUs
  262. - NVIDIA GPUs (with CUDA 2.x)
  263. - experimental Cell/BE support
  264. Changes:
  265. * Scheduling facilities
  266. - run-time selection of the scheduling policy
  267. - basic auto-tuning facilities
  268. * Software-based DSM
  269. - transparent data coherency management
  270. - High-level expressive interface
  271. # Local Variables:
  272. # mode: text
  273. # coding: utf-8
  274. # ispell-local-dictionary: "american"
  275. # End: