ChangeLog 8.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222
  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. StarPU 1.0.0 (svn revision 6306)
  21. ==============================================
  22. The extensions-again release
  23. New features:
  24. * Add SOCL, an OpenCL interface on top of StarPU.
  25. * Add a gcc plugin to extend the C interface with pragmas which allows to
  26. easily define codelets and issue tasks.
  27. * Add reduction mode to starpu_mpi_insert_task.
  28. * A new multi-format interface permits to use different binary formats
  29. on CPUs & GPUs, the conversion functions being provided by the
  30. application and called by StarPU as needed (and as less as
  31. possible).
  32. * Deprecate cost_model, and introduce cost_function, which is provided
  33. with the whole task structure, the target arch and implementation
  34. number.
  35. * Permit the application to provide its own size base for performance
  36. models.
  37. * Applications can provide several implementations of a codelet for the
  38. same architecture.
  39. * Add a StarPU-Top feedback and steering interface.
  40. * Permit to specify MPI tags for more efficient starpu_mpi_insert_task
  41. Changes:
  42. * Fix several memory leaks and race conditions
  43. * Make environment variables take precedence over the configuration
  44. passed to starpu_init()
  45. * Libtool interface versioning has been included in libraries names
  46. (libstarpu-1.0.so, libstarpumpi-1.0.so,
  47. libstarpufft-1.0.so, libsocl-1.0.so)
  48. * Install headers under $includedir/starpu/1.0.
  49. * Make where field for struct starpu_codelet optional. When unset, its
  50. value will be automatically set based on the availability of the
  51. different XXX_funcs fields of the codelet.
  52. * Define access modes for data handles into starpu_codelet and no longer
  53. in starpu_task. Hence mark (struct starpu_task).buffers as
  54. deprecated, and add (struct starpu_task).handles and (struct
  55. starpu_codelet).modes
  56. * Fields xxx_func of struct starpu_codelet are made deprecated. One
  57. should use fields xxx_funcs instead.
  58. * Some types were renamed for consistency. when using pkg-config libstarpu,
  59. starpu_deprecated_api.h is automatically included (after starpu.h) to
  60. keep compatibility with existing software. Other changes are mentioned
  61. below, compatibility is also preserved for them.
  62. To port code to use new names (this is not mandatory), the
  63. tools/dev/rename.sh script can be used, and pkg-config starpu-1.0 should
  64. be used.
  65. * The communication cost in the heft and dmda scheduling strategies now
  66. take into account the contention brought by the number of GPUs. This
  67. changes the meaning of the beta factor, whose default 1.0 value should
  68. now be good enough in most case.
  69. Small features:
  70. * Allow users to disable asynchronous data transfers between CPUs and
  71. GPUs.
  72. * Update OpenCL driver to enable CPU devices (the environment variable
  73. STARPU_OPENCL_ON_CPUS must be set to a positive value when
  74. executing an application)
  75. * struct starpu_data_interface_ops --- operations on a data
  76. interface --- define a new function pointer allocate_new_data
  77. which creates a new data interface of the given type based on
  78. an existing handle
  79. * Add a field named magic to struct starpu_task which is set when
  80. initialising the task. starpu_task_submit will fail if the
  81. field does not have the right value. This will hence avoid
  82. submitting tasks which have not been properly initialised.
  83. * Add a hook function pre_exec_hook in struct starpu_sched_policy.
  84. The function is meant to be called in drivers. Schedulers
  85. can use it to be notified when a task is about being computed.
  86. * Add codelet execution time statistics plot.
  87. * Add bus speed in starpu_machine_display.
  88. * Add a STARPU_DATA_ACQUIRE_CB which permits to inline the code to be
  89. done.
  90. * Add gdb functions.
  91. * Add complex support to LU example.
  92. * Permit to use the same data several times in write mode in the
  93. parameters of the same task.
  94. Small changes:
  95. * Increase default value for STARPU_MAXCPUS -- Maximum number of
  96. CPUs supported -- to 64.
  97. * Add man pages for some of the tools
  98. * Add C++ application example in examples/cpp/
  99. * Add an OpenMP fork-join example.
  100. * Documentation improvement.
  101. StarPU 0.9 (svn revision 3721)
  102. ==============================================
  103. The extensions release
  104. * Provide the STARPU_REDUX data access mode
  105. * Externalize the scheduler API.
  106. * Add theoretical bound computation
  107. * Add the void interface
  108. * Add power consumption optimization
  109. * Add parallel task support
  110. * Add starpu_mpi_insert_task
  111. * Add profiling information interface.
  112. * Add STARPU_LIMIT_GPU_MEM environment variable.
  113. * OpenCL fixes
  114. * MPI fixes
  115. * Improve optimization documentation
  116. * Upgrade to hwloc 1.1 interface
  117. * Add fortran example
  118. * Add mandelbrot OpenCL example
  119. * Add cg example
  120. * Add stencil MPI example
  121. * Initial support for CUDA4
  122. StarPU 0.4 (svn revision 2535)
  123. ==============================================
  124. The API strengthening release
  125. * Major API improvements
  126. - Provide the STARPU_SCRATCH data access mode
  127. - Rework data filter interface
  128. - Rework data interface structure
  129. - A script that automatically renames old functions to accomodate with the new
  130. API is available from https://scm.gforge.inria.fr/svn/starpu/scripts/renaming
  131. (login: anonsvn, password: anonsvn)
  132. * Implement dependencies between task directly (eg. without tags)
  133. * Implicit data-driven task dependencies simplifies the design of
  134. data-parallel algorithms
  135. * Add dynamic profiling capabilities
  136. - Provide per-task feedback
  137. - Provide per-worker feedback
  138. - Provide feedback about memory transfers
  139. * Provide a library to help accelerating MPI applications
  140. * Improve data transfers overhead prediction
  141. - Transparently benchmark buses to generate performance models
  142. - Bind accelerator-controlling threads with respect to NUMA locality
  143. * Improve StarPU's portability
  144. - Add OpenCL support
  145. - Add support for Windows
  146. StarPU 0.2.901 aka 0.3-rc1 (svn revision 1236)
  147. ==============================================
  148. The asynchronous heterogeneous multi-accelerator release
  149. * Many API changes and code cleanups
  150. - Implement starpu_worker_get_id
  151. - Implement starpu_worker_get_name
  152. - Implement starpu_worker_get_type
  153. - Implement starpu_worker_get_count
  154. - Implement starpu_display_codelet_stats
  155. - Implement starpu_data_prefetch_on_node
  156. - Expose the starpu_data_set_wt_mask function
  157. * Support nvidia (heterogeneous) multi-GPU
  158. * Add the data request mechanism
  159. - All data transfers use data requests now
  160. - Implement asynchronous data transfers
  161. - Implement prefetch mechanism
  162. - Chain data requests to support GPU->RAM->GPU transfers
  163. * Make it possible to bypass the scheduler and to assign a task to a specific
  164. worker
  165. * Support restartable tasks to reinstanciate dependencies task graphs
  166. * Improve performance prediction
  167. - Model data transfer overhead
  168. - One model is created for each accelerator
  169. * Support for CUDA's driver API is deprecated
  170. * The STARPU_WORKERS_CUDAID and STARPU_WORKERS_CPUID env. variables make it possible to
  171. specify where to bind the workers
  172. * Use the hwloc library to detect the actual number of cores
  173. StarPU 0.2.0 (svn revision 1013)
  174. ==============================================
  175. The Stabilizing-the-Basics release
  176. * Various API cleanups
  177. * Mac OS X is supported now
  178. * Add dynamic code loading facilities onto Cell's SPUs
  179. * Improve performance analysis/feedback tools
  180. * Application can interact with StarPU tasks
  181. - The application may access/modify data managed by the DSM
  182. - The application may wait for the termination of a (set of) task(s)
  183. * An initial documentation is added
  184. * More examples are supplied
  185. StarPU 0.1.0 (svn revision 794)
  186. ==============================================
  187. First release.
  188. Status:
  189. * Only supports Linux platforms yet
  190. * Supported architectures
  191. - multicore CPUs
  192. - NVIDIA GPUs (with CUDA 2.x)
  193. - experimental Cell/BE support
  194. Changes:
  195. * Scheduling facilities
  196. - run-time selection of the scheduling policy
  197. - basic auto-tuning facilities
  198. * Software-based DSM
  199. - transparent data coherency management
  200. - High-level expressive interface