Christophe Rhodes <csr21@...> writes:
> Hence new derive-type
> optimizers, which required factoring things so that I could apply
> combinations of auxiliary functions rather than writing 8
> almost-identical pieces of code :-)
Good! It will be very useful, if some day we will be asked to derive
types of *-MOD32.
> Alexey, does this look right so far?
Yes. You can also strip several lines from definitions of
32BIT-LOGICAL-* and BIGNUM-LOGNOT. Ideally, their definitions may be
shared between backends.
> I confess that I _still_ don't
> understand what a "good" modular function is, so I've ignored that
Neither do I, as you can see from the name "good" :-). The idea is
that for some functions (now AND and IOR) the general variant with the
ub32 VOP works good enough. The only difference is in a situation,
when 32-bit result is produced from bignum or negative source. But I'd
prefer to see such code before thinking how to optimize it properly.
"Alas, the spheres of truth are less transparent than those of
illusion." -- L.E.J. Brouwer