ARM linux mailing lists FAQ about floating point support.

ARM linux mailing lists FAQ about floating point support.

[Gilles Chanteperdrix]

Hi,

It seems the FAQ here:
http://www.arm.linux.org.uk/mailinglists/faq.php#f4

is a bit outdated, since it does not mention that EABI solves the issue.

Regards.

-- 
					    Gilles.

ARM linux mailing lists FAQ about floating point support.

[Nicolas Pitre]

On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:

> 
> Hi,
> 
> It seems the FAQ here:
> http://www.arm.linux.org.uk/mailinglists/faq.php#f4
> 
> is a bit outdated, since it does not mention that EABI solves the issue.

And just how would EABI solve the issue?


Nicolas

ARM linux mailing lists FAQ about floating point support.

[Gilles Chanteperdrix]

Nicolas Pitre wrote:
> On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:
> 
>> Hi,
>>
>> It seems the FAQ here:
>> http://www.arm.linux.org.uk/mailinglists/faq.php#f4
>>
>> is a bit outdated, since it does not mention that EABI solves the issue.
> 
> And just how would EABI solve the issue?

By allowing to mix soft-float and hard-float programs with only one libc?

-- 
					    Gilles.

ARM linux mailing lists FAQ about floating point support.

[Nicolas Pitre]

On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:

> Nicolas Pitre wrote:
> > On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:
> > 
> >> Hi,
> >>
> >> It seems the FAQ here:
> >> http://www.arm.linux.org.uk/mailinglists/faq.php#f4
> >>
> >> is a bit outdated, since it does not mention that EABI solves the issue.
> > 
> > And just how would EABI solve the issue?
> 
> By allowing to mix soft-float and hard-float programs with only one libc?

Err... How?


Nicolas

ARM linux mailing lists FAQ about floating point support.

[Gilles Chanteperdrix]

Nicolas Pitre wrote:
> On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:
> 
>> Nicolas Pitre wrote:
>>> On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:
>>>
>>>> Hi,
>>>>
>>>> It seems the FAQ here:
>>>> http://www.arm.linux.org.uk/mailinglists/faq.php#f4
>>>>
>>>> is a bit outdated, since it does not mention that EABI solves the issue.
>>> And just how would EABI solve the issue?
>> By allowing to mix soft-float and hard-float programs with only one libc?
> 
> Err... How?

Well, I read somewhere that EABI allows mixing soft-float and hard-float
code and must have assumed it was true. But you made me doubt.

But since compiling both a hard-float program and a soft-float program
with the same toolchain does not produce any compilation or link-edition
error, I still tend to think it is true.

I just need to run the test on actual hardware to finish checking.

-- 
					    Gilles.

ARM linux mailing lists FAQ about floating point support.

[Nicolas Pitre]

On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:

> Nicolas Pitre wrote:
> > On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:
> > 
> >> Nicolas Pitre wrote:
> >>> On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:
> >>>
> >>>> Hi,
> >>>>
> >>>> It seems the FAQ here:
> >>>> http://www.arm.linux.org.uk/mailinglists/faq.php#f4
> >>>>
> >>>> is a bit outdated, since it does not mention that EABI solves the issue.
> >>> And just how would EABI solve the issue?
> >> By allowing to mix soft-float and hard-float programs with only one libc?
> > 
> > Err... How?
> 
> Well, I read somewhere that EABI allows mixing soft-float and hard-float
> code and must have assumed it was true. But you made me doubt.
> 
> But since compiling both a hard-float program and a soft-float program
> with the same toolchain does not produce any compilation or link-edition
> error, I still tend to think it is true.

Well, actually it seems that gcc can only do -mfloat-abi=softfp and 
surprizingly -mfloat-abi=hard is unimplemented for VFP.  So in that case 
mixing soft and hard float objects is fine.


Nicolas

ARM linux mailing lists FAQ about floating point support.

[Gilles Chanteperdrix]

Nicolas Pitre wrote:
> On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:
> 
>> Nicolas Pitre wrote:
>>> On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:
>>>
>>>> Hi,
>>>>
>>>> It seems the FAQ here:
>>>> http://www.arm.linux.org.uk/mailinglists/faq.php#f4
>>>>
>>>> is a bit outdated, since it does not mention that EABI solves the issue.
>>> And just how would EABI solve the issue?
>> By allowing to mix soft-float and hard-float programs with only one libc?
> 
> Err... How?

Right, it does not work, so, what does:
"Allows the mixing of softfloat and hardfloat code"

is supposed to mean here for instance:
http://www.eurotech-inc.com/embedded-eabi-linux.asp

-- 
					    Gilles.

ARM linux mailing lists FAQ about floating point support.

[Nicolas Pitre]

On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:

> Right, it does not work, so, what does:
> "Allows the mixing of softfloat and hardfloat code"
> 
> is supposed to mean here for instance:
> http://www.eurotech-inc.com/embedded-eabi-linux.asp

As I just posted, it looks like gcc only implemented the soft-float ABI 
when using VFP.  So if your CPU is VFP capable then using 
-mfloat-abi=softfp should produce actual VFP instructions while still 
passing function arguments around using non VFP registers.


Nicolas

ARM linux mailing lists FAQ about floating point support.

[Laurent Desnogues]

On Thu, Sep 10, 2009 at 11:03 PM, Nicolas Pitre <nico@cam.org> wrote:
> On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:
>
>> Right, it does not work, so, what does:
>> "Allows the mixing of softfloat and hardfloat code"
>>
>> is supposed to mean here for instance:
>> http://www.eurotech-inc.com/embedded-eabi-linux.asp
>
> As I just posted, it looks like gcc only implemented the soft-float ABI
> when using VFP. ?So if your CPU is VFP capable then using
> -mfloat-abi=softfp should produce actual VFP instructions while still
> passing function arguments around using non VFP registers.

Didn't latest CodeSourcery ARM toolchain (2009q1) add hardfp?


Laurent

ARM linux mailing lists FAQ about floating point support.

[Nicolas Pitre]

On Thu, 10 Sep 2009, Laurent Desnogues wrote:

> On Thu, Sep 10, 2009 at 11:03 PM, Nicolas Pitre <nico@cam.org> wrote:
> > On Thu, 10 Sep 2009, Gilles Chanteperdrix wrote:
> >
> >> Right, it does not work, so, what does:
> >> "Allows the mixing of softfloat and hardfloat code"
> >>
> >> is supposed to mean here for instance:
> >> http://www.eurotech-inc.com/embedded-eabi-linux.asp
> >
> > As I just posted, it looks like gcc only implemented the soft-float ABI
> > when using VFP. ?So if your CPU is VFP capable then using
> > -mfloat-abi=softfp should produce actual VFP instructions while still
> > passing function arguments around using non VFP registers.
> 
> Didn't latest CodeSourcery ARM toolchain (2009q1) add hardfp?

I don't know.  If so then the FAQ entry about not being able to mix 
hard-float and soft-float would still apply when using that mode.


Nicolas