Parallel Integer Multiplication

[Dennis Clarke]

On 3/2/22 08:25, Samuel VIVIEN wrote:
> Hi,
> 
> 
> On an internship with Paul Zimmermann and Emmanuel Thomé I have written a
> parallel version of mpn_mul_fft_full that achieves satisfying performances.
> You can look athttps://hal.inria.fr/hal-03541726  for more details and
> https://github.com/samsa1/integer-para-mul  for the code.
> 
> 
> For information the code managed to multiply two integers of 10^7 words
> in 0.24s when it required 4.4s on the same computer in single thread
> (the computer had 32 cores), which is a speedup of almost 20.
> 

This looks fascinating and I will give it a test on a system with a pile
of cores and buckets of memory. Sadly we will likely need to use NVidia
GPU type technology to get massively parallel computer speeds but those
things are usually brain damaged at the factory to prevent 64-bit math
operations. Who knows, your code may be vastly superior in speed
compared to the NVidia gpu stuff. I will give it a test today.


-- 
Dennis Clarke
RISC-V/SPARC/PPC/ARM/CISC
UNIX and Linux spoken
GreyBeard and suspenders optional