Subject: printf() using floating point ...
To: None <firstname.lastname@example.org>
From: Gordon W. Ross <email@example.com>
Date: 06/15/1995 09:28:15
> Date: Thu, 15 Jun 1995 12:04:08 +1000 (EST)
> From: Andrew Cagney <firstname.lastname@example.org>
> Excerpts from mail: 14-Jun-95 Re: printf() using floating.. Chris G
> Demetriou@lagavu (935)
> > I'd definitely like to see such numbers, for a couple of different
> > processors, before this was jumped into.
Agreed. As for the following "rash assumptions," beware:
> (Using the back of a spare envelope ...)
> Lets make some rash assumptions:
> o all the data has to come from/to memory
> o the memory path is (8bytes @ 70ns or 70*10^-9)
The m68k (and most others?) are still 32-bit memory data path,
so the above would be off by a factor of two.
> o we ignore things like writeback caches stalls etc
> o 32 double precision fp registers.
The amount of data to move is often much more than just the FP
registers due to other saved state. (>300 bytes on m68k...)
In summary, one would probably need to just measure it.
We'll probably never know, given the lack of interest in
(or time to implement) lazy FP saves...