for large radii, the number of buckets that have to be looked at
when doing a HasWithin starts to dominate query performance. To
work around these cases we now do the following two things:

  -  store the bounding box for the buckets and intersect the cube
     of buckets which have to be traversed with the bounding box.
  -  when the number of buckets which have to be traversed gets
     larger than the total number of buckets in the organizer,
     loop over all buckets instead of looping over the cube.

they added very little benefits and were in general very slow.
The call sites of AtomsBegin have been replaced by a triple
for loop, ResiduesBegin by a double for loop. While that adds
some typing overhead, it gives a speed up compared to the old
version. It also happens to fix a problem noticed by Niklaus.

the superposition code has also been refactored to reduce
redundancy and make it easier to read.

thanks to Ansgar

Marco Biasini authored 12 years ago and Bienchen committed 12 years ago

The code becomes much simpler, cleaner and fixes as weird crash
with Qt 4.8

Valerio Mariani authored 12 years ago and Bienchen committed 12 years ago

Commited patched file by mistake

Valerio Mariani authored 12 years ago and Bienchen committed 12 years ago

Thanks Tobias for diagnosing the problem

Valerio Mariani authored 12 years ago and Bienchen committed 12 years ago

distance to check

Up until Python 2.6, MacOS X systems were detected as 32 bit
platforms. Now, they are detected as 64bit systems. This
caused sys.path to be wrong and the ost module could not
be imported

Commited patched file by mistake

Thanks Tobias for diagnosing the problem

On some old versions of Boost.Python, std::out_of_range
is not properly converted to IndexError. Instead, a
RuntimeError is raised. For these versions, we overwrite
the IndexError type with the type of the exception
raised by geom.Vec2()[3].

The code becomes much simpler, cleaner and fixes as weird crash
with Qt 4.8