Experimental Physics and
Industrial Control System

"J. Lewis Muir" <[email protected]> · Fri, 18 May 2018 17:20:21 -0500

On 05/18, Mark Rivers wrote:
> This might actually be a good case to use a double.  You get exact
> representation for 52 bits, i.e. 20 bits more than int32, or 1M times
> higher rate.

Interesting.  So that's for the IEEE 754 double-precision binary
floating-point format (64-bit).  Assuming epicsFloat64 is that, then
the maximum integer, such that it and all smaller integers can be
represented exactly, is 2^53 = 9,007,199,254,740,992 according to [1]
and [2].  And I believe the minimum is just the negation of that since
the sign has its own bit.

I had to go read about the 64-bit format [3], and apparently there's
one more bit available for the signficand, sometimes referred to as the
hidden bit, which is where the 53 comes from (i.e., 52 + 1).  The 64-bit
format stores the number in binary normalized form, so the first bit
will always be 1, so it is omitted, thus gaining one extra bit for the
significand.

Anyway, I guess I'll use asynParamFloat64, then, and return an error if
the bit rate is less than -2^53 or greater than 2^53.

Thanks!

Lewis

[1] https://stackoverflow.com/a/1848762
[2] https://stackoverflow.com/a/1848953
[3] https://en.wikipedia.org/wiki/IEEE_754-1985

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System