README 4.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140
  1. ++=================++
  2. || I. Introduction ||
  3. ++=================++
  4. +---------------------
  5. | I.a. What is StarPU?
  6. StarPU is a runtime system that offers support for heterogeneous multicore
  7. machines. While many efforts are devoted to design efficient computation kernels
  8. for those architectures (e.g. to implement BLAS kernels on GPUs or on Cell's
  9. SPUs), StarPU not only takes care of offloading such kernels (and implementing
  10. data coherency across the machine), but it also makes sure the kernels are
  11. executed as efficiently as possible.
  12. +------------------------
  13. | I.b. What StarPU is not
  14. StarPU is not a new language, and it does not extends existing languages either.
  15. StarPU does not help to write computation kernels.
  16. +---------------------------------
  17. | I.c. (How) Could StarPU help me?
  18. While StarPU will not make it easier to write computation kernels, it does
  19. simplify their actual offloading as StarPU handle most low level aspects
  20. transparently.
  21. Obviously, it is crucial to have efficient kernels, but it must be noted that
  22. the way those kernels are mapped and scheduled onto the computational resources
  23. also affect the overall performance to a great extent.
  24. StarPU is especially helpful when considering multiple heterogeneous processing
  25. resources: statically mapping and synchronizing tasks in such a heterogeneous
  26. environment is already very difficult, making it in a portable way is virtually
  27. impossible. On the other hand, the scheduling capabilities of StarPU makes it
  28. possible to easily exploit all processors at the same time while taking
  29. advantage of their specificities in a portable fashion.
  30. ++==================++
  31. || II. Requirements ||
  32. ++==================++
  33. * make
  34. * gcc (version >= 4.1)
  35. * if CUDA support is enabled
  36. * CUDA (version >= 2.2)
  37. * CUBLAS (version >= 2.2)
  38. * extra requirements for the svn version
  39. * autoconf (version >= 2.60)
  40. * automake
  41. * Remark: It is strongly recommanded that you also install the hwloc library
  42. before installing StarPU. This permits StarPU to actually map the processing
  43. units according to the machine topology. For more details on hwloc, see
  44. http://www.open-mpi.org/projects/hwloc/ .
  45. ++=====================++
  46. || III. Getting StarPU ||
  47. ++=====================++
  48. StarPU is available on https://gforge.inria.fr/projects/starpu/.
  49. The project's SVN repository can be checked out through anonymous
  50. access with the following command(s).
  51. $ svn checkout svn://scm.gforge.inria.fr/svn/starpu/trunk
  52. $ svn checkout --username anonsvn https://scm.gforge.inria.fr/svn/starpu/trunk
  53. The password is 'anonsvn'
  54. ++=============================++
  55. || IV. Building and Installing ||
  56. ++=============================++
  57. +---------------------------
  58. | IV.a. For svn version only
  59. $ ./autogen.sh
  60. +-----------------------
  61. | IV.b. For all versions
  62. $ ./configure
  63. $ make
  64. $ make install
  65. +---------------------
  66. | IV.c. Windows build:
  67. - c:\cuda\include\host_defines.h has a bogus CUDARTAPI definition which makes
  68. linking fail completely. Replace the first occurence of
  69. #define CUDARTAPI
  70. with
  71. #ifdef _WIN32
  72. #define CUDARTAPI __stdcall
  73. #else
  74. #define CUDARTAPI
  75. #endif
  76. While at it, you can also comment the __cdecl definition to avoid spurious
  77. warnings.
  78. - If you have a non-english version of windows, use
  79. export LANG=C
  80. else libtool has troubles parsing the translated output of the toolchain.
  81. - libtool is not able to find the libraries automatically, you need to make some
  82. copies:
  83. copy c:\cuda\lib\cuda.lib c:\cuda\lib\libcuda.lib
  84. copy c:\cuda\lib\cudart.lib c:\cuda\lib\libcudart.lib
  85. copy c:\cuda\lib\cublas.lib c:\cuda\lib\libcublas.lib
  86. copy c:\cuda\lib\cufft.lib c:\cuda\lib\libcufft.lib
  87. copy c:\cuda\lib\OpenCL.lib c:\cuda\lib\libOpenCL.lib
  88. ++===========++
  89. || V. Trying ||
  90. ++===========++
  91. Some examples ready to run are installed into $prefix/lib/starpu/{examples,mpi}
  92. ++=============++
  93. || VI. Upgrade ||
  94. ++=============++
  95. To upgrade your source code from older version (there were quite a few
  96. renamings), use the tools/rename.sh script
  97. ++==============++
  98. || VII. Contact ||
  99. ++==============++
  100. For any questions regarding StarPU, please contact the starpu-devel
  101. mailing-list at starpu-devel@lists.gforge.inria.fr .