SoftFloat implementation for MIPS in GCC

SoftFloat implementation for MIPS in GCC

[Rajesh Palani]

[-- Attachment #1: Type: text/plain, Size: 719 bytes --]

Hi,
 
   We are using a gcc 2.96 20000731 (Red Hat Linux 7.1 2.96-99.1) GCC cross-compiler with -msoftfloat to use software floating point routines.
 
   When we profied an application using the Linux Trace Toolkit, we observed that  there were a lot of CpU (Co-processor unusable) exceptions.  Some of the floating point routines ( eg. __floatdidf) expect values to be passed in floating point registers and take FP exceptions even though the application has been built with -msoftfloat.  Is this a general MIPS/GCC issue?  What is the status of softfloat  for MIPS in GCC?
 
   Thanks in advance.
 
   Rajesh


---------------------------------
Do you Yahoo!?
Yahoo! SiteBuilder - Free web site building tool. Try it!

[-- Attachment #2: Type: text/html, Size: 995 bytes --]

Re: SoftFloat implementation for MIPS in GCC

[Daniel Jacobowitz]

On Wed, Jan 28, 2004 at 11:26:36AM -0800, Rajesh Palani wrote:
> Hi,
>  
>    We are using a gcc 2.96 20000731 (Red Hat Linux 7.1 2.96-99.1) GCC cross-compiler with -msoftfloat to use software floating point routines.
>  
>    When we profied an application using the Linux Trace Toolkit, we observed that  there were a lot of CpU (Co-processor unusable) exceptions.  Some of the floating point routines ( eg. __floatdidf) expect values to be passed in floating point registers and take FP exceptions even though the application has been built with -msoftfloat.  Is this a general MIPS/GCC issue?  What is the status of softfloat  for MIPS in GCC?

Try a more recent compiler, that one is ancient.  If you configure
correctly, you should get no references to the floating point registers
at all.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: SoftFloat implementation for MIPS in GCC

[Jun Sun]

On Wed, Jan 28, 2004 at 02:33:55PM -0500, Daniel Jacobowitz wrote:
> On Wed, Jan 28, 2004 at 11:26:36AM -0800, Rajesh Palani wrote:
> > Hi,
> >  
> >    We are using a gcc 2.96 20000731 (Red Hat Linux 7.1 2.96-99.1) GCC cross-compiler with -msoftfloat to use software floating point routines.
> >  
> >    When we profied an application using the Linux Trace Toolkit, we observed that  there were a lot of CpU (Co-processor unusable) exceptions.  Some of the floating point routines ( eg. __floatdidf) expect values to be passed in floating point registers and take FP exceptions even though the application has been built with -msoftfloat.  Is this a general MIPS/GCC issue?  What is the status of softfloat  for MIPS in GCC?
> 
> Try a more recent compiler, that one is ancient.  If you configure
> correctly, you should get no references to the floating point registers
> at all.
> 

If glibc is not compiled with -msoftfloat, I think you will get a few
FPU exceptions from glibc no matter how apps are compiled.  

Actually, will it be a problem if glibc and apps are compiled differently
(such as in longjump, sig handling area)?

Jun

Re: SoftFloat implementation for MIPS in GCC

[David Daney]

Jun Sun wrote:

>On Wed, Jan 28, 2004 at 02:33:55PM -0500, Daniel Jacobowitz wrote:
>  
>
>>On Wed, Jan 28, 2004 at 11:26:36AM -0800, Rajesh Palani wrote:
>>    
>>
>>>Hi,
>>> 
>>>   We are using a gcc 2.96 20000731 (Red Hat Linux 7.1 2.96-99.1) GCC cross-compiler with -msoftfloat to use software floating point routines.
>>> 
>>>   When we profied an application using the Linux Trace Toolkit, we observed that  there were a lot of CpU (Co-processor unusable) exceptions.  Some of the floating point routines ( eg. __floatdidf) expect values to be passed in floating point registers and take FP exceptions even though the application has been built with -msoftfloat.  Is this a general MIPS/GCC issue?  What is the status of softfloat  for MIPS in GCC?
>>>      
>>>
>>Try a more recent compiler, that one is ancient.  If you configure
>>correctly, you should get no references to the floating point registers
>>at all.
>>
>>    
>>
>
>If glibc is not compiled with -msoftfloat, I think you will get a few
>FPU exceptions from glibc no matter how apps are compiled.  
>
>Actually, will it be a problem if glibc and apps are compiled differently
>(such as in longjump, sig handling area)?
>  
>
I have found that some versions of glibc (2.2.5 for example) have hard 
coded floating point code (in longjump handling IIRC) so no matter what 
you do you get the exceptions.

David Daney.

Re: SoftFloat implementation for MIPS in GCC

[Daniel Jacobowitz]

On Wed, Jan 28, 2004 at 11:53:12AM -0800, Jun Sun wrote:
> On Wed, Jan 28, 2004 at 02:33:55PM -0500, Daniel Jacobowitz wrote:
> > On Wed, Jan 28, 2004 at 11:26:36AM -0800, Rajesh Palani wrote:
> > > Hi,
> > >  
> > >    We are using a gcc 2.96 20000731 (Red Hat Linux 7.1 2.96-99.1) GCC cross-compiler with -msoftfloat to use software floating point routines.
> > >  
> > >    When we profied an application using the Linux Trace Toolkit, we observed that  there were a lot of CpU (Co-processor unusable) exceptions.  Some of the floating point routines ( eg. __floatdidf) expect values to be passed in floating point registers and take FP exceptions even though the application has been built with -msoftfloat.  Is this a general MIPS/GCC issue?  What is the status of softfloat  for MIPS in GCC?
> > 
> > Try a more recent compiler, that one is ancient.  If you configure
> > correctly, you should get no references to the floating point registers
> > at all.
> > 
> 
> If glibc is not compiled with -msoftfloat, I think you will get a few
> FPU exceptions from glibc no matter how apps are compiled.  
> 
> Actually, will it be a problem if glibc and apps are compiled differently
> (such as in longjump, sig handling area)?

Yes, that will be a problem.  Nothing that takes or returns a floating
point value will work either.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer