Thursday 8th March 2001
> Date: Wed, 7 Mar 2001 14:56:34 -0600
> From: Kurt W Hirchert <[log in to unmask]>
> I have been told the original problem was exactly the opposite. In the
> days when IBM was first implementing FORTRAN for the 360 line, they
> supposedly considered using 64-bit floating point for type REAL but didn't
> because they were not prepared at that time to support 64-bit
> integers. (Remember that at that time, most vendors were still using 6-bit
> character codes and typical floating point word sizes were 36, 48, or 60
and 32 bits.
Also, there was at least one machine that used 64-bit words as standard
precision, using 32-bit mantissa and 32-bit exponent.
Block floating-point was also used, with 32-bit words (30 + 1 bit mantissas,
16-bit exponents).
> bits, so most FORTRAN programs used REAL primarily and DOUBLE PRECISION
> only for special cases.
This was mainly on account of double precision being VERY expensive
in both time and in storage, when storage was a premium.
Double precision division took an especially long time. . . .
> --
> Kurt W Hirchert [log in to unmask]
> UIUC Department of Atmospheric Sciences +1-217-265-0327
|