On Thu, 16 Dec 2021, Henry Rich wrote:
it doesn't use any of the fancy 256-bit instructions that integers use
In fairness--I think the code I posted demonstrates that attempting to parallelize extended-precision arithmetic is fraught :)
(Parallel computation on multiple bignums is promising; but I think that it requires either avx512 gather/scatter, or for the numbers to all have the same number of limbs.)
Wouldn't you like to replace J's extendeds with GMP or something similar?
I might. I have thought about it before, and recently had call for extended-precision floating point numbers. But life has also been rather busy lately.
If time, energy, and interest align, I will post a patch; but do not let my hesitant interest turn anyone else away
-E ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
