Mainframe floating point math implementation.

[Charles Anthony]

On Mon, Mar 28, 2016 at 7:26 AM, Noel Chiappa <jnc at mercury.lcs.mit.edu>
wrote:

>     > From: Charles Anthony
>
>     > the missing piece of the rounding algorithim has been identified:
>     >     Only round if the mantissa was shifted more then 71 bits.
>
> Wow. I'm really impressed that they implemented that in hardware, back
> then!
>
> Then again, they threw so many gates at the Multics CPU, I guess they
> figured
> a few more wouldn't matter... ;-)
>
>
The floating point actually predates Multics (GE-635 I believe); I suspect
that the driving force for sophisticated floating point was the growing
'scientific computing' market, remember this was Cold War era and the
military was spending piles of money on nuclear weapon research. (I don't
have any documentation, just my informal understanding of mainframe
history.)

-- Charles


        Noel
>