Re: [boost] Interest in a GPU computing library

Subject: Re: [boost] Interest in a GPU computing library
From: Christof Donat (cd_at_[hidden])
Date: 2012-09-18 10:43:11

Hi,

> This is a call for interest in a GPU computing library for Boost.

For me this is very interesting.

> * C++ library for general-purpose computing on GPUs/Accelerators
> * Based on OpenCL (Open Computing Language)

Is it possible to have various backends, so that you could use Cuda with 
NVidia GPUs or e.g. something completely different on High Performance 
Clusters that distributes the code and uses e.g. MPI for communication, or 
whatever might come in handy? As far as I have understood, Thrust is designed 
that way and there are a fiew Backends available.

> Furthermore, a lambda expression library was written using Boost.Proto
> which allows for mathematical expressions to be defined at the call
> site of an algorithm and then be executed on the GPU. For example, to
> multiply each element in a vector by the square root of itself and
> then add four:
> 
> transform(v.begin(), v.end(), v.begin(), _1 * sqrt(_1) + 4);

Is there a way to use C++11 Lambdas here? I think that would make the library 
feel native in future as well. On the other hand I hav no idea how that could 
be possible at all.

> // transfer the values to the device
> device_vector = host_vector;

Is there a way to do some kind of streaming here as well? So instead of moving 
all the data to the GPU and getting the result back I think of some kind of 
stream, that I could use to transfer data to the GPU, have it do some 
calculations on it and get the results back from another stream while the GPU 
is working on the next chunk of data?

Christof

-- 
okunah gmbh                                  Software nach Maß
Zugspitzstraße 211                               www.okunah.de
86165 Augsburg                                    cd_at_[hidden]
                                      Registergericht Augsburg
Geschäftsführer                             Augsburg HRB 21896
Christof Donat                           UStID: DE 248 815 055