commons-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Gilles (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (MATH-1248) BigFraction optimizations around object creation for ZERO values and reduce()
Date Sat, 11 Jul 2015 13:02:04 GMT

     [ https://issues.apache.org/jira/browse/MATH-1248?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Gilles updated MATH-1248:
-------------------------
    Affects Version/s: 3.5
        Fix Version/s: 3.6

Leaving open in order to not forget to backport to 3.6.

> BigFraction optimizations around object creation for ZERO values and reduce()
> -----------------------------------------------------------------------------
>
>                 Key: MATH-1248
>                 URL: https://issues.apache.org/jira/browse/MATH-1248
>             Project: Commons Math
>          Issue Type: Improvement
>    Affects Versions: 3.5
>            Reporter: Chris Popp
>            Priority: Minor
>             Fix For: 3.6
>
>         Attachments: MINIMIZEALLOCATION-MATH-1248.patch
>
>
> In many cases a BigFraction object may represent a fraction in its most reduced form.
 Despite this, the reduce() method doesn't bother checking whether the BigFraction needs to
be reduced before returning a new object.
> This is problematic because equals() calls reduce() on both operands.  This means that
doing a simple comparison like BigFraction.ONE.equals(BigFraction.TWO) will result in two
object allocations.
> Additionally, while there is a very minimal attempt at optimizing around operations that
take in a parameter of BigFraction.ZERO, it is somewhat inconsistent and incomplete.  For
example, BigFraction add(final BigFraction fraction) checks to see whether the incoming fraction
is zero and if so, returns *this*.  Why doesn't it also check whether *this* is zero and return
the incoming fraction if so, avoiding an allocation?  
> This is one example, but it exists in other locations where we not only end up with allocations
but expensive arithmetic operations as well (multiplication or exponentiation).
> At minimum I think the reduce() method should just return *this* if gcd <= 1.
> I attached a patch with the change for reduce() and some ideas for avoiding object allocations,
attempting to follow the patterns in BigInteger.  I verified all of the tests in the commons
math pass after the changes.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message