Skip to content
Snippets Groups Projects
  1. Apr 05, 2022
  2. May 15, 2018
  3. Jul 17, 2015
  4. Feb 05, 2013
  5. Jul 20, 2012
  6. Jun 25, 2012
  7. Mar 17, 2012
    • Dimitris Karkoulis's avatar
      Fixed setup.py · 2158dfda
      Dimitris Karkoulis authored
      Copy external library dependencies of clPPM and
      cuPPM modules to the installation directory
      2158dfda
  8. Mar 16, 2012
  9. Mar 13, 2012
  10. Mar 12, 2012
  11. Mar 05, 2012
  12. Mar 04, 2012
  13. Mar 02, 2012
    • Dimitris Karkoulis's avatar
      3cd870b7
    • Dimitris Karkoulis's avatar
      Fixes in extension setup script · 6bf63a6e
      Dimitris Karkoulis authored
      6bf63a6e
    • Dimitris Karkoulis's avatar
      Created a dedicated setup script for · 3ede3fc4
      Dimitris Karkoulis authored
      cuPPM and clPPM.
      
      Building cuda and OpenCL applications is
      complicated due to many possible
      combinations of paths, environment variables
      and libraries.
      
      The purpose is to declutter setup.py and to be
      able to build cuPPM and clPPM in a separate/child
      process in order to allow PPM to always try to install
      no matter what complication might come up with
      OpenCL or CUDA.
      
      New functionality is added and the script is
      significantly tidied up:
      -Better autoconfiguration- Recognise CUDA and OpenCL paths
      more effectively, as well as take CUDA versions into account (for the OAR
      clusters)
      -OpenCL ICD checks on windows via registry checks
      -Can locate vcvars VC++ dev configuration environemt
      for Windows via registry checks
      
      Signed-off-by: default avatarDimitris Karkoulis <dimitris.karkoulis@esrf.eu>
      3ede3fc4
  14. Mar 01, 2012
  15. Feb 29, 2012
  16. Feb 28, 2012
  17. Feb 27, 2012
  18. Feb 26, 2012
  19. Feb 25, 2012
    • Dimitris Karkoulis's avatar
      Now PPM can recover from a clPPM runtime failure and fallback. · f7e3aa63
      Dimitris Karkoulis authored
      That is most probable to occur from insufficient memory
      when the CPU is used in 32bit systems, or a device is requested manually.
      
      clPPM checks the allocatable memory of the device, however clPPM is
      called from within Python. In 32bit systems if OpenCL buffers + Python
      exceed 2Gb failure will occur and that clPPM cannot check.
      
      Signed-off-by: default avatarDimitris Karkoulis <dimitris.karkoulis@esrf.eu>
      f7e3aa63
    • Dimitris Karkoulis's avatar
      Handle clPPM runtime error in the python level, by · ae7b68d8
      Dimitris Karkoulis authored
      automatically switching to default PPM.
      
      This is necessary because clPPM requires more memory than
      PPM. If clPPM on CPU is used on a machine with not enough
      memory it can fail, while PPM might not.
      
      
      Signed-off-by: default avatarDimitris Karkoulis <dimitris.karkoulis@esrf.eu>
      ae7b68d8
    • Dimitris Karkoulis's avatar
      More fixes to setup.py · 150ab4ff
      Dimitris Karkoulis authored
      Added the 3 options of module creation to clPPM
      The direct approach for the modules of cuPPM and clPPM were
      tested and work now under Windows
      
      In clPPM the query_for_gpu behaviour was modified so that
      when a GPU is found, it is also checked for double precision
      capability. If it fails it returns a CPU device. If it fails to
      to find a CPU however PPM will crash.
      
      In the future it must throw an exception if OpenCL CPU fallback occurs
       so that PPM reverts to the default algorithm.
      
      A function was added in ocl_tools, similar to the existing
      first-find first-get function used for automatic device
      configuration. However, the new one add FP64 capability
      requirement.
      
      Signed-off-by: default avatarDimitris Karkoulis <dimitris.karkoulis@esrf.eu>
      150ab4ff
  20. Feb 24, 2012
  21. Feb 23, 2012
Loading