Open osrf-migration opened 9 years ago
Original comment by Jose Luis Rivero (Bitbucket: Jose Luis Rivero, GitHub: j-rivero).
Hey Davide!
The ignition-math code was split from gazebo as a first step to make it a bit mode modular. I agree that makes very little sense to reimplement another math library when we have good open source alternatives already in place. I want to see ign-math as a first step to be able to migrate gazebo maths to one of them, as this issue (#19) is proposing.
Original comment by Steve Peters (Bitbucket: Steven Peters, GitHub: scpeters).
We can also consider Blaze:
Original comment by Davide Faconti (Bitbucket: facontidavide).
I am also toying with boost::simd (not a Boost library yet). It provide a portable way to use SIMD and it is quite easy to use.
I guess I will publish my code just as reference pretty soon. The performance is MUCH faster for floats and considerably better even for doubles.
Original comment by Davide Faconti (Bitbucket: facontidavide).
A nice note about boost::simd is that it support NEON. I believe neither Eigen nor Blaze does.
Original comment by Steve Peters (Bitbucket: Steven Peters, GitHub: scpeters).
We could also look at the bullet linear math library, which has lots of SIMD/ neon optimizations
Original comment by Steve Peters (Bitbucket: Steven Peters, GitHub: scpeters).
Cc @hmazhar who told me about blaze
Original comment by Davide Faconti (Bitbucket: facontidavide).
Hi,
a quick question. Is there any reason to NOT use libraries such as Eigen, Blaze or the lower lever Boost::Simd to speed up this one?
Just asking ...
Original report (archived issue) by Davide Faconti (Bitbucket: facontidavide).
Hi,
just a question/suggestion. A quick inspection of the code reveals that no SSE or similar vectorization was used. Being this library a potential building block of several robotic components (kinematics, graphics, dynamics, etc.) I wonder why you need to build from scratch yet another implementation when Bullet 3 has a vectorized one. Why aren't you using Eigen? This is of course an observation that I would give to anyone writing somethin like Matrix4x4.h... ;)
Kind regards
Davide