[ADOL-C] ADOL-C on a GPU

[Giacomo Perantoni]

Hello,

The problem is not in exploiting the sparsity (at least for now),
but just in computing the darivatives of the same function at many different points (more than 500).
What I am interested in is just using the forward mode on a gpu,
and distribute several function calculations on several cores.

A better question might be: 
is the ADOL-C source code compatible with compilers for gpu's?
(I know that some gpu's work only with floats and have some other code compatibility issues)

I hope what I have just said makes sense, please tell me if there are any basic mistakes in the reasoning.

Regards,
Giacomo.

-----Original Message-----
From: Andrea Walther [mailto:andrea.walther at uni-paderborn.de] 
Sent: 02 November 2011 10:18
To: Giacomo Perantoni
Cc: adol-c at list.coin-or.org
Subject: Re: [ADOL-C] ADOL-C on a GPU

Hi,

> I need to evaluate first and second derivatives of a vector valued 
> function at several points, and I suppose each computation would be 
> independent from each other.
> What might need to be shared is the tape if using the reverse mode, 
> but it probably needs to be computed only once.
>
> Is there any way to use ADOL-C on a GPU?

We are currently investigating this but so far just for the forward mode. So, there would be no soon accessable implementation of ADOL-C for GPUs covering forward and reverse mode.

> Or, is there an alternative efficient way to solve my problem?

An alternative would be to compute these independent functions in parallel and exploit the sparsity in this parallel implementation.

Best regards,

Andrea

--
Prof. Dr. Andrea Walther
Lehrstuhl fuer Mathematik und ihre Anwendungen Institut fuer Mathematik Universitaet Paderborn Warburger Str. 100
33098 Paderborn

Email: andrea.walther at uni-paderborn.de
Phone: ++49 5251 602721
Fax:   ++49 5251 603728

**********