[prev in list] [next in list] [prev in thread] [next in thread] 

List:       gmp-discuss
Subject:    Re: A question about GMP
From:       Marc Glisse <marc.glisse () inria ! fr>
Date:       2019-10-22 12:05:21
Message-ID: alpine.DEB.2.21.1910221340100.8249 () grove ! saclay ! inria ! fr
[Download RAW message or body]

On Fri, 18 Oct 2019, Ivan Christov wrote:

> Do you know whether GMP library is vectorised?
>
> I want to utilize both SIMD level and thread level of parallelism with 
> the INTEL(R) Knights Landing Processors.

Hello,

the question in unclear in that vectorization can be used in 2 ways (in 
both cases the answer will be "no" though). You can use vector 
instructions to speed up the operations on a single big number. In 
particular I see some people playing with AVX512-IFMA. Or you can use 
vector instructions to operate on several numbers at the same time. The 
second option is usually more efficient, but its use is much more 
constrained, you need to have several numbers of the same size on which 
you want to perform the same operation at the same time.

-- 
Marc Glisse
_______________________________________________
gmp-discuss mailing list
gmp-discuss@gmplib.org
https://gmplib.org/mailman/listinfo/gmp-discuss
[prev in list] [next in list] [prev in thread] [next in thread]