[ADOL-C] Using Adol-c within OpenCl framework

[RICARDO PUENTE RICO]

Hello,

I am trying to use Adol-c to generate the adjoint code of a OpenCL 
kernel. I have seen the thread regarding GPU computing from November 
2011, and I was wondering if there were any updates on the issue.

Just to narrow down the problem, my objective would be the 
differentiate code from within a openCL kernel, confining all Adol-c 
functions and variables to such kernel. Thus I would not have to deal 
with interfacing issues between the main code and the kernels, and 
neither would I be worrying in principle about sparsity, checkpointing 
or any advanced stuff.

The problems I have right now are that I need to work with single 
precision float variables, which Adol-c apparently does not support, and 
with telling the OpenCL compiler to use Adol-c's functions.

A workaround I can think of is copying locally to my code all the 
Adol-c functions I need, adapting them to work with floats. If you 
believe this is feasible, I would be thankful for any pointers to do 
this as painlessly as possible.


Regards,


-- 
RICARDO PUENTE RICO
Universidad Politecnica de Madrid