(original) (raw)

Hi Albert,

A couple of easy suggestions/ideas:

1) since you don't resize the backing array, you can make the size be a template parameter - compiler can then constant propagate it and save a few instructions.

2) if #1 above is not needed, can also consider making size be a power of 2 so you can replace mod instruction with an & when computing idx.

You did mention that this is the bare minimum implementation to support just what you need, so perhaps above is overkill anyway.

Thanks

Sent from my phone

On Feb 21, 2014 3:06 AM, "Albert" <albert.noll@oracle.com> wrote:

Hi,

could I get reviews for this small patch?

Bug: https://bugs.openjdk.java.net/browse/JDK-8034839

Problem:
The problem is that the patch (7194669) - which was supposed to
speed-up dependency checking

� � � � � � �� causes a performance regression. The reason for the
performance regression is that most dependencies

�� are unique, so we have the overhead of determining if
the dependency is already checked plus the

� � � � � �� overhead of dependency checking. The overhead of
searching is significant, since we perform

�� a linear search on 6000+ items each time.

Solution: Use a hashtable instead of linear search to lookup already
checked dependencies. The new hashtable

�� is very rudimentary. It provides only the required
functionality to solve this bug. However, the functionality

�� can be easily extended as needed.

Testing: jprt, failing test case, nashorn. The failing test case
completes in approx. the same time as before 7194669.

�� For nashorn + Octane, this patch yields the following
times spent for dependency checking:

�� with this patch:� 844s

�� 7194669: 1080s

�� before 7194669: 5223s

webrev: http://cr.openjdk.java.net/~anoll/8034939/webrev.00/

Thanks,

Albert