Re: [avr-libc-dev] Bug in math library

[Marek Michalkiewicz]

On Sat, Jan 08, 2005 at 05:46:14PM -0400, Colin O'Flynn wrote:

> Anyway the other thing I thought of - if the hidden bit is set when zero 
> detected, it might be easier to just check for 0x01 being present in that 
> register.

Before fp_split: exponent = 0
After  fp_split: exponent = 1, hidden = 0 (denormal)

Before fp_split: exponent = 1
After  fp_split: exponent = 1, hidden = 1 (small, but still "normal")

So you can't just check for 0x01, it would be true in both cases.

> However it is that sort of thing that will probably get broken in the 
> future... to be completely safe I would need to do this:
> 
>     clr     __tmp_reg__
>     OR      __tmp_reg__, rB2 
>     OR      __tmp_reg__, rB1
>     OR      __tmp_reg__, rB0

Be sure to check that __tmp_reg__ (r0) doesn't hold some important
value - fplib uses almost all available registers.

> Then the Z flag would be set if all those registers are zero. It is an 
> increase from 1 to 4 instructions.

Not 4 - replace clr/or with mov.  Depending on registers used, it
could also be possible to use sbiw/cpc - see "tstsi" pattern in GCC.
But some register containing zero is necessary - in fplib you can't
depend on __zero_reg__ (r1), but some other zero register might be
available (like rAE/rBE cleared in fp_split).

It depends how exactly you want to compare against zero - perhaps
check for denormal (test hidden bit with sbrc/sbrs) is sufficient?
Division by denormal can overflow too, just like division by zero.

Marek