In practice, this is not ideal in the case of WarpX, because we have regression tests, that check that results are unchanged (compared to previous commits) to machine precision. As a consequence, for the regression tests, it would be great for use to have the option to use FFTW_ESTIMATE.
This PR allow PICSAR users to do this by setting the corresponding module variable fftw_plan_measure to .FALSE. (the default is True)
@atmyers recently pointed out that FFTW_MEASURE can result in different results (at machine precision) for identical runs: see http://www.fftw.org/faq/section3.html#nondeterministic
In practice, this is not ideal in the case of WarpX, because we have regression tests, that check that results are unchanged (compared to previous commits) to machine precision. As a consequence, for the regression tests, it would be great for use to have the option to use FFTW_ESTIMATE.
This PR allow PICSAR users to do this by setting the corresponding module variable
fftw_plan_measure
to.FALSE.
(the default is True)