Re: LuaJIT2 performance for number crunching

[T T <t34www@...>]

On 19 February 2011 16:14, Francesco Abbate <francesco.bbt@gmail.com> wrote:
> 2011/2/19 T T <t34www@googlemail.com>:
>> On 19 February 2011 15:04, Francesco Abbate <francesco.bbt@gmail.com> wrote:
>>> input. The difference is that I don't use vector but everything is
>>> expanded to local variables using a template preprocessor.
>>
>> What will happen for an ODE system of large dimension, say 1000 or 10,000?
>
> In this case your computer may explode... be careful!

Too late, now I'm left with a smoking pile of transistors ;)

> No, I'm kidding... I know about this problem and I was planning to add
> different mechanism with arrays to manage ODE systems with non-small
> dimension.

But then the user supplied function needs a different format depending
on the size of the ode, right?

for small ode:

  dydt_1, dydt_2, ... = f(t, y_1, y_2, ...)

for large ode:

  f(t, y, dydt)

Hmm, not that nice from the UI design POV.

Cheers,

Tomek

PS. Do you have performance numbers for C variant with -O3?  I'm
curious how does this compares to luajit.