[Oberon] FP

[Holger Goetz]

Hi,

my 2cts:
This sounds like best thinking for the emulator, but it makes a 
difference for the FPGA, where it is like it's with most embedded 
systems: "size matters". That was done to keep memory requirements low 
(well, even lower).

Thanks,
Holger

On 15/04/14 10:33, Jan Verhoeven wrote:
> Hi all,
>
> Back in 1982 when trying to understand the ZX81 (Z80 based personal
> computer) I ran into the Floating point routines. At first it was a bit
> difficult to understand, but after a few hours it became clear:
>
> An FPnum consisted of three machine words for the mantissa and one
> machine word for the exponent. It nicely fitted inside a Z80 QuadWord.
>
> Our latest 2014 FPGA processor (and hence the emulator) uses 1/4
> machineword for the exponent and 3/4 machineword for the mantissa.
>
> Why don't we use one 32-bit integer for the exponent and one 32 bit word
> for the mantissa?
> It would make things so much more easier. And faster. Now each FPnum
> must be shifted, masked, shifted back, remasked, etc. All of it is
> unnecessary. It is a tribute to the ZX81 Basic.
>
> We could even go one step beyond! Define an FPnum as 32 bits exponent
> and 64 bit mantissa. This results in a 1 billion digit exponent
> (10^10^9) and an 18 digit mantissa.
>
>