Re: [Boost-users] math.quaternion design issues

Subject: Re: [Boost-users] math.quaternion design issues
From: John Maddock (jz.maddock_at_[hidden])
Date: 2015-07-29 14:02:29

On 29/07/2015 11:52, Matwey V. Kornilov wrote:
> Hello,
>
> I am facing the following performance issues due to internals of quaternion.
>
> abs(q) function is quite slow, and most of the time (according to
> callgrind) is taken to create valarray which is easily unnecessarily
> here. valarray requires doing malloc and free, because it is dynamic
> container, at the same time the size of quaternion is 4 and it is fixed.
>
> So, was was the purpose to use std::valarray internally?
As the person who has probably inherited maintenance of quaternion (but 
who's also been ignoring it!)  I feel I should comment - and confess I 
don't know the answer.  But I guess the idea was to take advantage of 
valarray's vector operations to simplify the implementation.  
Unfortunately as you rightly point out, this is a dis-optimization in 
this case, particularly as valarray has no allocator parameter, so a 
scoped storage allocator is out of the question :(

To be honest, quaternion and octonion should probably be rewritten to 
bring them into the post C++ 11 age.  There's a lot of cunningly crafted 
code in there though so it's not necessarily a trivial job. Can you 
please file a bug report though so this doesn't get lost/forgotten?

Thanks, John.