We now have a multiplyHigh intrinsic, but it is signed. Unsigned
multiplyHigh is in general a more useful primitive for crypto than
signed, and I wonder if we need an intrinsic for that as well. I've
looked at cooking up an unsigned multiplyHigh in Java, and I think the
fastest way is this:
private static final long unsignedMultiplyHigh(long a, long b) {
long result = Math.multiplyHigh(a, b);
if (a < 0) result += b;
if (b < 0) result += a;
// Can also be written as:
// result += (a >> 63) & b;
// result += (b >> 63) & a;
return result;
}
It's still about 50% slower than the signed multiplyHigh, though.
Thoughts?
--
Andrew Haley
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671
