(forw) [Re: [Openzaurus-users] Soft float (was Opera on OZ 3.5.2)]

[Chris L. <ke...@ha...>]

Heh, hit the wrong reply button.

----- Forwarded message from Chris Larson <ke...@ha...> -----

Date: Sat, 26 Feb 2005 10:54:38 -0600
From: Chris Larson <ke...@ha...>
To: Michael Rozdoba <mr...@ca...>
User-Agent: Mutt/1.3.28i
Subject: Re: [Openzaurus-users] Soft float (was Opera on OZ 3.5.2)

* Michael Rozdoba (mr...@ca...) wrote:
> Robert Longbottom wrote:
> >On Thu, February 24, 2005 7:18 pm, Michael Rozdoba said:
>=20
> >>Does soft float cause problems with Opera? Is there a list
> >>of apps that will work via oz-compat but only if soft float is not in u=
se?
> >
> >
> >I'm not really sure about this, but I'd quite like to know too.  I think=
 I
> >used to have Opera running fine using soft float.  You only run into
> >problems when you try to do floating point calculations - do you need to
> >do this to render a web page???  I'm not really sure.
>=20
> I used to write C/Assembler apps for desktop machines built with ARM=20
> hence generally without hardware floating point. Thus there was good=20
> reason to not use floats unnecessarily; nevertheless it was very hard to=
=20
> avoid some use at times, without going to great lengths (which would=20
> only be justified if it is clear low precision is sufficient, or code is=
=20
> extremely computation intensive, such as with mp3 generation, which is a=
=20
> whole other ballgame). In short, most big apps would have quite a lot of=
=20
> float usage, even though the performance impact wasn't that great as a=20
> result of care.
>=20
> Does the OE soft float usage really break every external app which makes=
=20
> /any/ use of floating point ops?
>=20
> I was wondering if perhaps it only affected apps which either input or=20
> output (ie to the user via the gui) floats? If that was the case, I can=
=20
> see how most stuff wouldn't be affected, whilst obvious apps such as=20
> spreadsheets would be well & truly stuffed.

The use of software floating point is actually a different _ABI_.  It
isn't just a matter of not using floating point instructions and having
the function calls instead.

> I imagine if the result is that one format gets treated as though it=20
> were the other, the consequences would be very odd.

Soft float is generally VFP.  Hard float is generally FPA, and the
kernel fpu emulation is FPA.

> Can anyone in passing comment on the format soft float uses? I was=20
> wondering if it's a fast floating point emulation or perhaps just a=20
> fixed point implementation.

It's the arm asm software floating point functions coded by Nicolas
Pitre.  Search the linux-arm and linux-arm-kernel mailing list archives
for details.
--
Chris Larson - kergoth at handhelds dot org
Linux Software Systems Engineer - clarson at ti dot com
Core Developer/Architect - BitBake, OpenEmbedded, OpenZaurus

----- End forwarded message -----
--
Chris Larson - kergoth at handhelds dot org
Linux Software Systems Engineer - clarson at ti dot com
Core Developer/Architect - BitBake, OpenEmbedded, OpenZaurus