Re: S/360

"Steve Myers" <noone@[EMAIL PROTECTED]
> wrote in message
news:f0e$447e3ede$cf672f7d$3139@[EMAIL PROTECTED]
> I don't think too many people will argue that S/360 floating point was
> "good," particularly single precision floating point.  I think IBM tried
> to get some numeric range at the expense of precision in 32 bit
> hexadecimal floating point.  I think the "4-bit ****ft is faster than the
> 1-bit ****ft" argument IBM laid on us in the 60s was entirely specious.

No it wasn't.
Take a look at some old timings for the SLL and SRL instructions.
Time was 8.16 + 0.48V (V = no. of bits ****fted)
so for a 15 bit ****ft, T = 8.16 + 0.48*15 = 15.36µS
A load took 8.88µS.
AE took 19.20µS
AD took 27.69µS.
With a SLL, there are up to 15 ****fts.
With AE there are up to 6 ****fts (15 for AD).

In the float operations, there are extra steps in the (microcode) loop
(compared to SLL etc), namely, for each ****ft,
the exponent must be incremented/decremented.
    ****fting (and add/subtract of exponent) occurs twice -
once for pre-normalizing and once for post-normalising.

The above times were for a model 50 (not IBM).

For a faster machine, SLL was 0.9 + 0.12T (T = no of 4-bit ****fts)
L was 1.8µS.
while AE was 3.6 and AD was 4.0.