RE: [f-cpu] Re: Floating-Point?

["Ekkehard Morgenstern" <emorgens@prs-gmbh.de>]

> Was talking to Yann lately if it won't be possible
> to design fast 32 bit floating point instructions
> that keep the computed value as one part of a 64 bit
> value pair, the other being the imprecise error of
> this value due to the limitations of precision.

This might be a good idea, but why not use 128 bits internally
(i.e. by making the registers that wide), and design FMOV
instructions that cut off / expand mantissa and exponent when
loading/storing floating-point data to different bit widths?

When the FPU internally uses one bit-depth with high precision,
rounding and representation errors can be minimized.

Whether execution speed suffers from 128 bits, is a question of
FPU design, I think.

For example, Intel's Pentium processor line uses lots of tables
to maximize the speed of multiplication and division operations.




*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/