Suspect the splits to be the culprit for the factor 4 slowdown of propagation that sometimes occurs.
TODO: find way to record the “split” event in metadata:
2014-11-20 19:47:44,927 INFO env.geant4.geometry.collada.g4daeview.daechromacontext:66 using seed 0
2014-11-20 19:47:44,927 INFO env.geant4.geometry.collada.g4daeview.g4daechroma:73 ***** post G4DAEChroma ctor
2014-11-20 19:47:44,927 INFO env.geant4.geometry.collada.g4daeview.g4daechroma:50 start polling responder: DAEDirectResponder connect tcp://127.0.0.1:5002
2014-11-20 19:47:44,927 INFO env.geant4.geometry.collada.g4daeview.g4daechroma:54 polling 0
2014-11-20 19:47:52,657 INFO env.geant4.geometry.collada.g4daeview.daedirectresponder:47 DAEDirectResponder request (4165, 4, 4)
2014-11-20 19:47:52,657 INFO env.geant4.geometry.collada.g4daeview.g4daechroma:43 handler got obj (cpl or npl)
2014-11-20 19:47:52,657 INFO env.geant4.geometry.collada.g4daeview.daedirectpropagator:54 ctrl {u'reset_rng_states': 1, u'nthreads_per_block': 64, u'seed': 0, u'max_blocks': 1024, u'max_steps': 30, u'COLUMNS': u'max_blocks:i,max_steps:i,nthreads_per_block:i,reset_rng_states:i,seed:i'}
2014-11-20 19:47:52,657 WARNING env.geant4.geometry.collada.g4daeview.daedirectpropagator:62 reset_rng_states
2014-11-20 19:47:52,657 INFO env.geant4.geometry.collada.g4daeview.daechromacontext:122 _set_rng_states
2014-11-20 19:47:53,005 INFO chroma.gpu.geometry :171 Splitting BVH between GPU and CPU memory at node 100
2014-11-20 19:47:53,030 INFO chroma.gpu.geometry :201 device usage:
----------
nodes 100.0 1.6K
total 1.6K
----------
device total 2.1G
device used 1.9G
device free 292.1M
2014-11-20 19:47:53,033 INFO env.geant4.geometry.collada.g4daeview.daechromacontext:117 _get_rng_states
2014-11-20 19:47:53,033 INFO env.geant4.geometry.collada.g4daeview.daechromacontext:72 setup_rng_states using seed 0