using.texi 4.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115
  1. @c -*-texinfo-*-
  2. @c This file is part of the StarPU Handbook.
  3. @c Copyright (C) 2009--2011 Universit@'e de Bordeaux 1
  4. @c Copyright (C) 2010, 2011, 2012 Centre National de la Recherche Scientifique
  5. @c Copyright (C) 2011 Institut National de Recherche en Informatique et Automatique
  6. @c See the file starpu.texi for copying conditions.
  7. @menu
  8. * Setting flags for compiling and linking applications::
  9. * Running a basic StarPU application::
  10. * Kernel threads started by StarPU::
  11. * Enabling OpenCL::
  12. @end menu
  13. @node Setting flags for compiling and linking applications
  14. @section Setting flags for compiling and linking applications
  15. Compiling and linking an application against StarPU may require to use
  16. specific flags or libraries (for instance @code{CUDA} or @code{libspe2}).
  17. To this end, it is possible to use the @code{pkg-config} tool.
  18. If StarPU was not installed at some standard location, the path of StarPU's
  19. library must be specified in the @code{PKG_CONFIG_PATH} environment variable so
  20. that @code{pkg-config} can find it. For example if StarPU was installed in
  21. @code{$prefix_dir}:
  22. @example
  23. % PKG_CONFIG_PATH=$PKG_CONFIG_PATH:$prefix_dir/lib/pkgconfig
  24. @end example
  25. The flags required to compile or link against StarPU are then
  26. accessible with the following commands@footnote{It is still possible to use the API
  27. provided in the version 0.9 of StarPU by calling @code{pkg-config}
  28. with the @code{libstarpu} package. Similar packages are provided for
  29. @code{libstarpumpi} and @code{libstarpufft}.}:
  30. @example
  31. % pkg-config --cflags starpu-1.0 # options for the compiler
  32. % pkg-config --libs starpu-1.0 # options for the linker
  33. @end example
  34. Also pass the @code{--static} option if the application is to be
  35. linked statically.
  36. @node Running a basic StarPU application
  37. @section Running a basic StarPU application
  38. Basic examples using StarPU are built in the directory
  39. @code{examples/basic_examples/} (and installed in
  40. @code{$prefix_dir/lib/starpu/examples/}). You can for example run the example
  41. @code{vector_scal}.
  42. @example
  43. % ./examples/basic_examples/vector_scal
  44. BEFORE: First element was 1.000000
  45. AFTER: First element is 3.140000
  46. %
  47. @end example
  48. When StarPU is used for the first time, the directory
  49. @code{$STARPU_HOME/.starpu/} is created, performance models will be stored in
  50. that directory (@code{STARPU_HOME} defaults to @code{$HOME}, or @code{$USERPROFILE
  51. } in windows environments).
  52. Please note that buses are benchmarked when StarPU is launched for the
  53. first time. This may take a few minutes, or less if @code{hwloc} is
  54. installed. This step is done only once per user and per machine.
  55. @node Kernel threads started by StarPU
  56. @section Kernel threads started by StarPU
  57. StarPU automatically binds one thread per CPU core. It does not use
  58. SMT/hyperthreading because kernels are usually already optimized for using a
  59. full core, and using hyperthreading would make kernel calibration rather random.
  60. Since driving GPUs is a CPU-consuming task, StarPU dedicates one core per GPU
  61. While StarPU tasks are executing, the application is not supposed to do
  62. computations in the threads it starts itself, tasks should be used instead.
  63. TODO: add a StarPU function to bind an application thread (e.g. the main thread)
  64. to a dedicated core (and thus disable the corresponding StarPU CPU worker).
  65. @node Enabling OpenCL
  66. @section Enabling OpenCL
  67. When both CUDA and OpenCL drivers are enabled, StarPU will launch an
  68. OpenCL worker for NVIDIA GPUs only if CUDA is not already running on them.
  69. This design choice was necessary as OpenCL and CUDA can not run at the
  70. same time on the same NVIDIA GPU, as there is currently no interoperability
  71. between them.
  72. To enable OpenCL, you need either to disable CUDA when configuring StarPU:
  73. @example
  74. % ./configure --disable-cuda
  75. @end example
  76. or when running applications:
  77. @example
  78. % STARPU_NCUDA=0 ./application
  79. @end example
  80. OpenCL will automatically be started on any device not yet used by
  81. CUDA. So on a machine running 4 GPUS, it is therefore possible to
  82. enable CUDA on 2 devices, and OpenCL on the 2 other devices by doing
  83. so:
  84. @example
  85. % STARPU_NCUDA=2 ./application
  86. @end example