tevador from design.md: "The approximate range of possible group E register values is 1.7E-77 to infinity."

But according to my calculations, group E registers can go as low as 2^-511 ~ 1.49E-154

after an iteration

Start with 2^-255, then 256 FDIV_M where each FDIV_M divides by 1.11111...11b*2^0

thankfully, there's never an underflow, so no need to specify the case of underflow

I noticed that JIT used flush-to-zero mode, but interpreted VM uses whatever is by default

on the other hand, a program consisting of 256 FDIV_M instruction is probably impossible because the probability of that is (6/256)^256 ~ 4.98e-418 and we have only ~1.34e154 possible inputs

no, (4/256)^256 ~ 4.14e-463

actually, even (1/256)^256 ~ 3.09e-617 because it has to be the same register every time

sech1: the value in the docs is not the theoretical minimum; IIRC I ran 100k programs and took the lowest value

yeah, but we need to know theoretical minimum, because flush-to-zero is never mentioned in docs

you are right, it was in one of the previous versions, I'm not sure why it got removed

FTZ and DAZ should be mandatory

but 1e-154 is still not low enough to trigger FTZ

what a waste of hardware