Hal Fulton wrote:
> I can see some value to it.

I use an unsigned 64-bit count of 10-ns intervals
since the (non-existent) year 0 of the Gregorian
calendar. Since it's non-existent, I double up any
value in year zero as meaning "time only, no date"
up to one year duration.

The Windows SystemTime() API returns 100ns units,
and I might have done that instead to avoid the
multiplication by ten (multiplication and division
or 64-bit qtys can be pretty horrendous on Intel
machines), but I can still construct a new C++
"current time" object in 100ns on a P4, so that's
fast enough to use in calculating my timings, which
sometimes schedule things down below 1ms, which
is the resolution of Sleep().