Hi,
On Fri, Mar 20, 2009 at 06:32:41AM +0000, Steve Smith wrote:
> Hi - I guess you mean via OpenCL? Yes this would be nice but is probably
> quite a lot of work, as I don't think any of the low-level libraries that
> we currently use (e.g. newmat linear algebra library) are going to be
> ported to OpenCL in the very short term. As i understand it this stuff is
> not going to be accessible simply via changing some compiler flags,.
> right?
That is right. There is no simple porting at all, since the whole
approach is totally different. Having 800 shader units in a GPU, you
would have to split a problem into 800 pieces to take full advantage of
it. It is already non-trivial to refactor a codebase from single-thread
logic to multi-threading (e.g. two or four parallel threads to harvest
the capabilities of today's CPUs). Moreover, these shaders typically
support single precision data only -- which is usually not the common
datatype for scientific computing.
Moving towards OpenCL/CUDA/... is more of an evolutionary process than a
switch and I have to agree with Steve that the low-level libs have to go
first -- which is probably not going to happen very fast, since there is
not even a _single_ standard, yet -- but the implementations vary from
vendor to vendor.
Cheers,
Michael
--
GPG key: 1024D/3144BE0F Michael Hanke
http://apsy.gse.uni-magdeburg.de/hanke
ICQ: 48230050
|