Ifpack2::ILUT: "fact: ilut level-of-fill" parameter really should be double; documentation is wrong
Created by: mhoemmen
@atoth1 noticed that Ifpack2::ILUT wants the "fact: ilut level-of-fill" parameter to have type int
. (It accepts it as magnitude_type
as a "backwards compatibility" measure, according to the comment.) However, the code immediately casts the value to double
. AztecOO's ILUT implementation does something similar. The code in both packages uses the parameter as a fraction, not as a count of things. Therefore, a double
would be more appropriate. magnitude_type
is not appropriate, because the fraction's required precision should depend on local_ordinal_type
, not on scalar_type
.
Ifpack2's documentation for this parameter is also wrong.
@trilinos/ifpack2