
.. _section_omega_releasenotes_252:

Omega TK 2.5.2
================================================================================

New features
--------------------------------------------------------------------------------

* The fragment library is now shared across multiple
  instances of the :ref:`OEOmega <OEConfGen_OEOmega>` objects in a
  thread-safe manner. This can greatly reduce the amount of memory
  needed to run a multi-threaded version of Omega TK. For example, the
  memory consumption of Omega TK running across 8-cores dropped from
  715MB to 258MB for 14,000 lead-like compounds from MDDR. This can
  also lead to a performance improvement since expensive fragment
  generation can sometimes be avoided if it has already been completed
  by another thread.

.. oe:language:: java

   .. note:: 
      
      For Java, the internal fraglib is not shared because the fraglib
      is not shipped as part of the omega2 library binary, but as a
      separate standalone file to allow multiple JAR files to be
      combined together and not duplicate the fragment library across
      them. Fragments generated on the fly will be shared across
      :ref:`OEOmega <OEConfGen_OEOmega>` objects, just not the builtin
      fragments.


.. oe:language:: c++

   * A multi-threaded Omega TK example has been added ``omega_omp`` to
     demonstrate how to use Omega TK with OpenMP.

   * The internal fragment library for the Omega TK static library is
     now stored in a more space and time efficient manner. Binaries
     linked to the Omega TK library should now be smaller, for
     example, the RHEL6 ``simple_omega`` example is now 53MB instead
     of 77MB. The library load time has also been improved. For
     example, running the RHEL6 ``simple_omega`` example on a single
     benzene will now take 0.20 seconds instead of 0.78 seconds. Your
     mileage may vary depending upon operating system and hardware.

.. oe:language:: python
   
   * The internal fragment library for the Omega TK shared library is
     now stored in a more space and time efficient manner. Creating an
     :ref:`OEOmega <OEConfGen_OEOmega>` object and running it on a single
     molecule should now be faster. For example, running the
     ``simple_omega.py`` example on a single benzene now takes 0.25
     seconds instead of 0.83 seconds. Your mileage may vary depending
     upon operating system and hardware.

   * The python global interpreter lock will now be released when entering OMEGA toolkit calls. 


Minor bug fixes
--------------------------------------------------------------------------------

* Now operator() will fail nicely and return false for non-3D input if `SetFromCT`
  is set to false.

* All ring bonds will be left unset by :ref:`OEFlipper <OEConfGen_OEFlipper>` unless forceFlip=true.

