On Fri, 2025-08-22 at 16:56 -0700, Eduard Zingerman wrote: > On Sat, 2025-08-23 at 05:18 +0530, Nandakumar Edamana wrote: > > [...] > > > I personally don't think `best(a*b, b*a)` is ugly. What about > > `best(oldprod, newprod)`, where oldprod and newprod are each found > > like this, using the old tnum_mul and the new tnum_mul respectively? > > Hm, given that both are correct if we go for a hybrid approach we can > peek known bits from both. Thinking it over the weekend, I tend to agree with Harishankar. Few percent improvement does not merit complications with best part selection or maintaining two algorithms for multiplication. I'd stick with the new algorithm as in the current patch.
