[net.arch] Data General Eclipse Floating Point

wallach@convexs.UUCP (12/27/85)

the data general and ibm formats and interpretation are IDENTICAL.

there is one difference.  on the data general machine this is an optional
floating point round mode, UNBIASED ROUNDING with two guard digits (hex)
and a sticky bit.  this is equivalent to the ieee rounding mode extended to
hex exponent machines.  thus on the dg machine, using the unbiased round
mode, more accurate answers will be obtained as contrasted to the equivalent
program executed on a ibm machine.

meissner@dg_rtp.UUCP (Michael Meissner) (12/30/85)

In article <3000001@convexs> wallach@convexs.UUCP writes:
>
>the data general and ibm formats and interpretation are IDENTICAL.
>
>there is one difference.  on the data general machine this is an optional
>floating point round mode, UNBIASED ROUNDING with two guard digits (hex)
>and a sticky bit.  this is equivalent to the ieee rounding mode extended to
>hex exponent machines.  thus on the dg machine, using the unbiased round
>mode, more accurate answers will be obtained as contrasted to the equivalent
>program executed on a ibm machine.

What can I say, except that I stand corrected.  I orignally thought they were
the same too, but a knowlable DG system's engineer steered me wrong.  In any
case, since Steve Wallach was the architect for the MV/8000, he should know.

	Michael Meissner