WIP: fix #21825 part c ; low-level code for reduce-plattice
The regression (mentioned in #21825) between 737f4eea and 06d9470b (well, actually, between 737f4eea and 41ba6139) is actually caused by be3db721; while that commit is buggy, the bug it had introduced got fixed shortly afterwards by 41ba6139. But ditching the low-level asm code for reduce_plattice
is not, after all, totally innocuous.
I'm not re-enabling the asm code for reduce_plattice
now, however, because:
- this would obviously break our freshly implemented support of fb primes up to 2^32
- this asm code is terribly fragile and not tested
- I would very much prefer if we had a separate thorough validate/bench utility à la
torture-redc
ortest-smallsieve
.