diep at xs4all.nl
Mon Oct 11 10:54:04 CEST 2010
>> Greetings. Are there currently any active projects to port GMP to
>> hardware libraries, such as CUDA or CTM?
>> Not any that I am aware of.
>> (It is very difficult to make good use of SIMD processors in GMP, for
>> reasons that have been discussed here in the past.)
> I have an idea to approach SIMD differently for this kind of parallel
> processing (using algorithms which are not inherently geared toward a
> native port to SIMD). Is there someone on this list who would be
> willing to discuss this possibility with me?
Of course i'm very interested in all this.
> I tried to post this question only to the developers on the developers
> list, but the list admin denied my post.
This is a better list to ask anyway.
Parallellization using vector cores (not so much SIMD) is very complicated.
The dominant hardwqare is AMD currently. It delivers 4.6+ or more tflop
(single precision) versus nvidia 1+ Tflop.
Most projects are CUDA, for unknown reason to me.
> Rick C. Hodgin
> gmp-discuss mailing list
> gmp-discuss at gmplib.org
More information about the gmp-discuss