ChangeLog 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311
  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. * New interface: COO matrix.
  37. Changes:
  38. * Fix the block filter functions.
  39. * Fix StarPU-MPI on Darwin.
  40. * The FxT code can now be used on systems other than Linux.
  41. * Keep only one hashtable implementation common/uthash.h
  42. * The cache of starpu_mpi_insert_task is fixed and thus now enabled by
  43. default.
  44. * Improve starpu_machine_display output.
  45. * Standardize objects name in the performance model API
  46. * SOCL
  47. - Virtual SOCL device has been removed
  48. - Automatic scheduling still available with command queues not
  49. assigned to any device
  50. - Remove modified OpenCL headers. ICD is now the only supported
  51. way to use SOCL.
  52. - SOCL test suite is only run when environment variable
  53. SOCL_OCL_LIB_OPENCL is defined. It should contain the location
  54. of the libOpenCL.so file of the OCL ICD implementation.
  55. * Fix main memory leak on multiple unregister/re-register.
  56. * Improve hwloc detection by configure
  57. Small changes:
  58. * STARPU_NCPU should now be used instead of STARPU_NCPUS. STARPU_NCPUS is
  59. still available for compatibility reasons.
  60. * include/starpu.h includes all include/starpu_*.h files, applications
  61. therefore only need to have #include <starpu.h>
  62. * Active task wait is now included in blocked time.
  63. * Fix GCC plugin linking issues starting with GCC 4.7.
  64. * Fix forcing calibration of never-calibrated archs.
  65. * CUDA applications are no longer compiled with the "-arch sm_13"
  66. option. It is specifically added to applications which need it.
  67. StarPU 1.0.3 (svn revision 7379)
  68. ==============================================
  69. Changes:
  70. * Several bug fixes in the build system
  71. * Bug fixes in source code for non-Linux systems
  72. * Fix generating FXT traces bigger than 64MiB.
  73. * Improve ENODEV error detections in StarPU FFT
  74. StarPU 1.0.2 (svn revision xxx)
  75. ==============================================
  76. Changes:
  77. * Add starpu_block_shadow_filter_func_vector and an example.
  78. * Add tag dependency in trace-generated DAG.
  79. * Fix CPU binding for optimized CPU-GPU transfers.
  80. * Fix parallel tasks CPU binding and combined worker generation.
  81. * Fix generating FXT traces bigger than 64MiB.
  82. StarPU 1.0.1 (svn revision 6659)
  83. ==============================================
  84. Changes:
  85. * hwloc support. Warn users when hwloc is not found on the system and
  86. produce error when not explicitely disabled.
  87. * Several bug fixes
  88. * GCC plug-in
  89. - Add `#pragma starpu release'
  90. - Fix bug when using `acquire' pragma with function parameters
  91. - Slightly improve test suite coverage
  92. - Relax the GCC version check
  93. * Update SOCL to use new API
  94. * Documentation improvement.
  95. StarPU 1.0.0 (svn revision 6306)
  96. ==============================================
  97. The extensions-again release
  98. New features:
  99. * Add SOCL, an OpenCL interface on top of StarPU.
  100. * Add a gcc plugin to extend the C interface with pragmas which allows to
  101. easily define codelets and issue tasks.
  102. * Add reduction mode to starpu_mpi_insert_task.
  103. * A new multi-format interface permits to use different binary formats
  104. on CPUs & GPUs, the conversion functions being provided by the
  105. application and called by StarPU as needed (and as less as
  106. possible).
  107. * Deprecate cost_model, and introduce cost_function, which is provided
  108. with the whole task structure, the target arch and implementation
  109. number.
  110. * Permit the application to provide its own size base for performance
  111. models.
  112. * Applications can provide several implementations of a codelet for the
  113. same architecture.
  114. * Add a StarPU-Top feedback and steering interface.
  115. * Permit to specify MPI tags for more efficient starpu_mpi_insert_task
  116. Changes:
  117. * Fix several memory leaks and race conditions
  118. * Make environment variables take precedence over the configuration
  119. passed to starpu_init()
  120. * Libtool interface versioning has been included in libraries names
  121. (libstarpu-1.0.so, libstarpumpi-1.0.so,
  122. libstarpufft-1.0.so, libsocl-1.0.so)
  123. * Install headers under $includedir/starpu/1.0.
  124. * Make where field for struct starpu_codelet optional. When unset, its
  125. value will be automatically set based on the availability of the
  126. different XXX_funcs fields of the codelet.
  127. * Define access modes for data handles into starpu_codelet and no longer
  128. in starpu_task. Hence mark (struct starpu_task).buffers as
  129. deprecated, and add (struct starpu_task).handles and (struct
  130. starpu_codelet).modes
  131. * Fields xxx_func of struct starpu_codelet are made deprecated. One
  132. should use fields xxx_funcs instead.
  133. * Some types were renamed for consistency. when using pkg-config libstarpu,
  134. starpu_deprecated_api.h is automatically included (after starpu.h) to
  135. keep compatibility with existing software. Other changes are mentioned
  136. below, compatibility is also preserved for them.
  137. To port code to use new names (this is not mandatory), the
  138. tools/dev/rename.sh script can be used, and pkg-config starpu-1.0 should
  139. be used.
  140. * The communication cost in the heft and dmda scheduling strategies now
  141. take into account the contention brought by the number of GPUs. This
  142. changes the meaning of the beta factor, whose default 1.0 value should
  143. now be good enough in most case.
  144. Small features:
  145. * Allow users to disable asynchronous data transfers between CPUs and
  146. GPUs.
  147. * Update OpenCL driver to enable CPU devices (the environment variable
  148. STARPU_OPENCL_ON_CPUS must be set to a positive value when
  149. executing an application)
  150. * struct starpu_data_interface_ops --- operations on a data
  151. interface --- define a new function pointer allocate_new_data
  152. which creates a new data interface of the given type based on
  153. an existing handle
  154. * Add a field named magic to struct starpu_task which is set when
  155. initialising the task. starpu_task_submit will fail if the
  156. field does not have the right value. This will hence avoid
  157. submitting tasks which have not been properly initialised.
  158. * Add a hook function pre_exec_hook in struct starpu_sched_policy.
  159. The function is meant to be called in drivers. Schedulers
  160. can use it to be notified when a task is about being computed.
  161. * Add codelet execution time statistics plot.
  162. * Add bus speed in starpu_machine_display.
  163. * Add a STARPU_DATA_ACQUIRE_CB which permits to inline the code to be
  164. done.
  165. * Add gdb functions.
  166. * Add complex support to LU example.
  167. * Permit to use the same data several times in write mode in the
  168. parameters of the same task.
  169. Small changes:
  170. * Increase default value for STARPU_MAXCPUS -- Maximum number of
  171. CPUs supported -- to 64.
  172. * Add man pages for some of the tools
  173. * Add C++ application example in examples/cpp/
  174. * Add an OpenMP fork-join example.
  175. * Documentation improvement.
  176. StarPU 0.9 (svn revision 3721)
  177. ==============================================
  178. The extensions release
  179. * Provide the STARPU_REDUX data access mode
  180. * Externalize the scheduler API.
  181. * Add theoretical bound computation
  182. * Add the void interface
  183. * Add power consumption optimization
  184. * Add parallel task support
  185. * Add starpu_mpi_insert_task
  186. * Add profiling information interface.
  187. * Add STARPU_LIMIT_GPU_MEM environment variable.
  188. * OpenCL fixes
  189. * MPI fixes
  190. * Improve optimization documentation
  191. * Upgrade to hwloc 1.1 interface
  192. * Add fortran example
  193. * Add mandelbrot OpenCL example
  194. * Add cg example
  195. * Add stencil MPI example
  196. * Initial support for CUDA4
  197. StarPU 0.4 (svn revision 2535)
  198. ==============================================
  199. The API strengthening release
  200. * Major API improvements
  201. - Provide the STARPU_SCRATCH data access mode
  202. - Rework data filter interface
  203. - Rework data interface structure
  204. - A script that automatically renames old functions to accomodate with the new
  205. API is available from https://scm.gforge.inria.fr/svn/starpu/scripts/renaming
  206. (login: anonsvn, password: anonsvn)
  207. * Implement dependencies between task directly (eg. without tags)
  208. * Implicit data-driven task dependencies simplifies the design of
  209. data-parallel algorithms
  210. * Add dynamic profiling capabilities
  211. - Provide per-task feedback
  212. - Provide per-worker feedback
  213. - Provide feedback about memory transfers
  214. * Provide a library to help accelerating MPI applications
  215. * Improve data transfers overhead prediction
  216. - Transparently benchmark buses to generate performance models
  217. - Bind accelerator-controlling threads with respect to NUMA locality
  218. * Improve StarPU's portability
  219. - Add OpenCL support
  220. - Add support for Windows
  221. StarPU 0.2.901 aka 0.3-rc1 (svn revision 1236)
  222. ==============================================
  223. The asynchronous heterogeneous multi-accelerator release
  224. * Many API changes and code cleanups
  225. - Implement starpu_worker_get_id
  226. - Implement starpu_worker_get_name
  227. - Implement starpu_worker_get_type
  228. - Implement starpu_worker_get_count
  229. - Implement starpu_display_codelet_stats
  230. - Implement starpu_data_prefetch_on_node
  231. - Expose the starpu_data_set_wt_mask function
  232. * Support nvidia (heterogeneous) multi-GPU
  233. * Add the data request mechanism
  234. - All data transfers use data requests now
  235. - Implement asynchronous data transfers
  236. - Implement prefetch mechanism
  237. - Chain data requests to support GPU->RAM->GPU transfers
  238. * Make it possible to bypass the scheduler and to assign a task to a specific
  239. worker
  240. * Support restartable tasks to reinstanciate dependencies task graphs
  241. * Improve performance prediction
  242. - Model data transfer overhead
  243. - One model is created for each accelerator
  244. * Support for CUDA's driver API is deprecated
  245. * The STARPU_WORKERS_CUDAID and STARPU_WORKERS_CPUID env. variables make it possible to
  246. specify where to bind the workers
  247. * Use the hwloc library to detect the actual number of cores
  248. StarPU 0.2.0 (svn revision 1013)
  249. ==============================================
  250. The Stabilizing-the-Basics release
  251. * Various API cleanups
  252. * Mac OS X is supported now
  253. * Add dynamic code loading facilities onto Cell's SPUs
  254. * Improve performance analysis/feedback tools
  255. * Application can interact with StarPU tasks
  256. - The application may access/modify data managed by the DSM
  257. - The application may wait for the termination of a (set of) task(s)
  258. * An initial documentation is added
  259. * More examples are supplied
  260. StarPU 0.1.0 (svn revision 794)
  261. ==============================================
  262. First release.
  263. Status:
  264. * Only supports Linux platforms yet
  265. * Supported architectures
  266. - multicore CPUs
  267. - NVIDIA GPUs (with CUDA 2.x)
  268. - experimental Cell/BE support
  269. Changes:
  270. * Scheduling facilities
  271. - run-time selection of the scheduling policy
  272. - basic auto-tuning facilities
  273. * Software-based DSM
  274. - transparent data coherency management
  275. - High-level expressive interface
  276. # Local Variables:
  277. # mode: text
  278. # coding: utf-8
  279. # ispell-local-dictionary: "american"
  280. # End: