[fpu] fdiv problem

[Jamil Khatib <jamilkhatib75@yahoo.com>]

OK no problem I'll start working on the conversion
block as soon as I finish the fdiv.
Currently I have some problems with the fdiv it does
not produce teh same results as the MS-calculator

for example when I do 8388616/8388610 or
2147485696/2147484160  on a normalized format I get 1
denormalized format but the result from MS-calculator
is 1.0000007152555667744715751477301

I got this result before conversion back from extended
format to single precision

I do not know the source of this problem. 
Anyhow the denormalized numbers produce good results

Jamil Khatib

--- Damjan Lampret <damjanlampret@yahoo.com> wrote:
> 
> > 
> > We need even more than that because the FPU is not
> > just the hardware but also the software library
> and
> > compiler so the testing software must cover
> everything
> > 
> 
> Compiler support is already done. Done !
> 
> Currently it emits all the most common instructions
> that our FPU
> supports. These are fadd, fsub, fmul, fdiv. If you
> add support for
> others in the FPU core, then it is a matter of
> minutes that I add
> support to the GCC. I don't think we need fsqrt and
> log functions as
> much as we need integer<->float conversion.
> 
> So integer<->float is currently done in software
> library (automatically
> generated when building GCC for OpenRISC).
> 
> regards,
> Damjan


__________________________________________________
Do You Yahoo!?
Yahoo! Mail - Free email you can access from anywhere!
http://mail.yahoo.com/