This PR implements changes made by @oleksandr-pavlyk in #49 . Since the memory branch has not been merged into master, we are including his (still relevant) changes here.
These changes result in relatively modest speedups of a few percent.
It should be noted that a small amount of numerical error is introduced by these changes, somewhat akin to those in #71 . We did confirm that (flux1-flux2)*np.sqrt(ivar) is centered around 0 and its maximum value is 1E-9.
This PR implements changes made by @oleksandr-pavlyk in #49 . Since the memory branch has not been merged into master, we are including his (still relevant) changes here.
These changes result in relatively modest speedups of a few percent.
It should be noted that a small amount of numerical error is introduced by these changes, somewhat akin to those in #71 . We did confirm that (flux1-flux2)*np.sqrt(ivar) is centered around 0 and its maximum value is 1E-9.