mirror of
https://github.com/lcn2/calc.git
synced 2025-08-19 01:13:27 +03:00
74 lines
2.5 KiB
Plaintext
74 lines
2.5 KiB
Plaintext
NAME
|
|
rcin - encode for REDC algorithms
|
|
|
|
SYNOPSIS
|
|
rcin(x, m)
|
|
|
|
TYPES
|
|
x integer
|
|
m odd positive integer
|
|
|
|
return integer v, 0 <= v < m.
|
|
|
|
DESCRIPTION
|
|
Let B be the base calc uses for representing integers internally
|
|
(B = 2^16 for 32-bit machines, 2^32 for 64-bit machines) and N the
|
|
number of words (base-B digits) in the representation of m. Then
|
|
rcin(x,m) returns the value of B^N * x % m, where the modulus
|
|
operator % here gives the least nonnegative residue.
|
|
|
|
If y = rcin(x,m), x % m may be evaluated by x % m = rcout(y, m).
|
|
|
|
The "encoding" method of using rcmul(), rcsq(), and rcpow() for
|
|
evaluating products, squares and powers modulo m correspond to the
|
|
formulae:
|
|
|
|
rcin(x * y, m) = rcmul(rcin(x,m), rcin(y,m), m);
|
|
|
|
rcin(x^2, m) = rcsq(rcin(x,m), m);
|
|
|
|
rcin(x^k, m) = rcpow(rcin(x,m), k, m).
|
|
|
|
Here k is any nonnegative integer. Using these formulae may be
|
|
faster than direct evaluation of x * y % m, x^2 % m, x^k % m.
|
|
Some encoding and decoding may be bypassed by formulae like:
|
|
|
|
x * y % m = rcin(rcmul(x, y, m), m).
|
|
|
|
If m is a divisor of B^N - h for some integer h, rcin(x,m) may be
|
|
computed by using rcin(x,m) = h * x % m. In particular, if
|
|
m is a divisor of B^N - 1 and 0 <= x < m, then rcin(x,m) = x.
|
|
For example if B = 2^16 or 2^32, this is so for m = (B^N - 1)/d
|
|
for the divisors d = 3, 5, 15, 17, ...
|
|
|
|
RUNTIME
|
|
The first time a particular value for m is used in rcin(x, m),
|
|
the information required for the REDC algorithms is
|
|
calculated and stored for future use in a table covering up to
|
|
5 (i.e. MAXREDC) values of m. The runtime required for this is about
|
|
two that required for multiplying two N-word integers.
|
|
|
|
Two algorithms are available for evaluating rcin(x, m), the one
|
|
which is usually faster for small N is used when N <
|
|
config("pow2"); the other is usually faster for larger N. If
|
|
config("pow2") is set at about 200 and x has both been reduced
|
|
modulo m, the runtime required for rcin(x, m) is at most about f
|
|
times the runtime required for an N-word by N-word multiplication,
|
|
where f increases from about 1.3 for N = 1 to near 2 for N > 200.
|
|
More runtime may be required if x has to be reduced modulo m.
|
|
|
|
EXAMPLE
|
|
Using a 64-bit machine with B = 2^32:
|
|
|
|
> for (i = 0; i < 9; i++) print rcin(x, 9),:; print;
|
|
0 4 8 3 7 2 6 1 5
|
|
|
|
LIMITS
|
|
none
|
|
|
|
LIBRARY
|
|
void zredcencode(REDC *rp, ZVALUE z1, ZVALUE *res)
|
|
|
|
SEE ALSO
|
|
rcout, rcmul, rcsq, rcpow
|