Closed manodeep closed 6 years ago
While this seems like a good option to implement, all of my attempts have not produced any performance improvements for DD
codes even thought the lines-of-code are much larger. Might be worth investigating in the future, or on a case-by-case basis. Closing for now.
For correlation functions with a large number of particles, as is the case for a lot of
DR
computations, the default should be a loop-blocking structure + calls to the appropriate kernel. Now one issue with a loop-blocking implementation is to catch the case where no further computations are necessary.where
COMPUTE_DONE
is a newenum
indefs.h
.Should be fixed alongside #114