Re: [glob2-devel] gradient improvement

[Kai Antweiler]

> Over the static optimization, we can use the dynamical optimization. It's a
> really cool idea:
> 1- For some time, you try each gradient computation system. Doing so you 
> collect
> speed measurement, in order to find out which is the fastest one, given the 
> type
> of gradient.

How do you intend to measure the speed?
I thought we had a hard time to do this even manually.  Always using
the same saved game, 60s, no mouse movement, no other programs running
and doing things twice because of the cache.

> 2- After some time, you only use the fastest gradient.
> 3- Optionally you can retry another measurement time.
> This way, the gradient computation way will automatically adapt to the CPU and
> map. I have put the structure so it's ready for such an use. I'll maybe try a
> few things with it another day.


I like the idea.

-- 
Kai Antweiler