Geometry Classes (Rotation2d, Translation2d, etc.) EJML question #7562
-
|
Has any thought been given to having these classes backed up more by the Matrix, Vector classes in Java or are there concerns that are not easily noticeable. Seems like it would be a logical fit. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 7 replies
-
|
Using matrix/vector types instead of native types like double internally would massively hurt performance due to the allocation overhead. Rewriting Quaternion to use doubles instead of Vector in #5525 gave a 100x performance improvement for Rotation3d.rotateBy(). Furthermore, matrix representations of Rotation2d/3d and Pose2d/3d require more FLOPs than what we use now. If you're after interoperability with affine transformation matrices, we added that in #7430. |
Beta Was this translation helpful? Give feedback.
-
|
So in theory (I know you wouldn't want to do this to keep the code bases as close as possible) you would have better performance if you did this in C++. Or if you used some form of mutable setup with Java. |
Beta Was this translation helpful? Give feedback.
-
|
A mutable API would be much harder to use correctly. Math things work better as immutable APIs. |
Beta Was this translation helpful? Give feedback.
-
|
Last thought. Any plans for this on the new controllers or are we still limited? |
Beta Was this translation helpful? Give feedback.
-
|
No, because replacing the internals with affine transformation matrices isn't actually solving a problem or offering some improvement. |
Beta Was this translation helpful? Give feedback.
-
|
I understand. I'm a teacher so I wondered if it would tie in better with teaching fundamental concepts involving Vectors and Matrices, but I do understand not integrating it if it offers no performance (or API structure) improvements. |
Beta Was this translation helpful? Give feedback.
Using matrix/vector types instead of native types like double internally would massively hurt performance due to the allocation overhead. Rewriting Quaternion to use doubles instead of Vector in #5525 gave a 100x performance improvement for Rotation3d.rotateBy().
Furthermore, matrix representations of Rotation2d/3d and Pose2d/3d require more FLOPs than what we use now.
If you're after interoperability with affine transformation matrices, we added that in #7430.