In my testing with converting the aprox* rates to use autodiff, I found a lot of instances where code templated with autodiff::dual would give slightly different values than the same code templated with amrex::Real. It turns out to be due to some optimizations autodiff makes to speed up evaluation, which this PR disables to avoid any weird inconsistencies in the future.
In my testing with converting the
aprox*
rates to use autodiff, I found a lot of instances where code templated withautodiff::dual
would give slightly different values than the same code templated withamrex::Real
. It turns out to be due to some optimizations autodiff makes to speed up evaluation, which this PR disables to avoid any weird inconsistencies in the future.