RE: coldfire math when using gcc

RE: coldfire math when using gcc

[David J. Fiddes]

Hi,

>> After some benchmarking I came to the conclusion that the
>hardware divide is
>> not being used. Is there any compiler option to turn this on?
>>
>
>If this is the egcs-1.1.something release then it I suspect it will not
>support the hardware divide or mac instructions.

This is the case. The egcs compiler that comes with uCLinux can quite
happily be swapped with something newer like gcc-2.95.2. You still need to
keep egcs-1.1 around to compiler the kernel unless it has been fixed. Modify
user/arch/coldfire/build.mk.

>Wayne Deeter, David Fiddes and others have been working on the support.
>You can find info about it on :

I have a patch that tidies the assembler options and fixes a few bugs in
binutils. Linus Nordberg submitted MAC and hw divide support for the
ColdFire which will be in binutils-2.10 (due any day now). Peter Jakubek
[pjak@snafu.de] has a patch against gcc-2.95.2 that adds hwdivide support
which gives a significant speedup to integer maths.

The collection of patches can be had from
http://fiddes.net/coldfire/download/gcc-m68k-patches-r7-20000413.tar.gz

Wayne Deeter [wrd@logical-co.com] wrote a siginifcantly improved version of
the gcc floating point library for the ColdFire. It requires quite a bit of
work to get it into shape for submission for gcc. This speeds up things like
double multiplies by a factor of two or so. It does not use or require mac
or hw divide instructions.

>I do not know the status of these patches or if anyone has attempted to
>get them into the gcc and binutils sources.

I have submitted my patches to binutils but I'm dragging my feet on getting
the copyright assignment form in... it will happen.

Dave


------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sourceware.cygnus.com

Re: coldfire math when using gcc

[Chris Johns]

Dwight Schauer wrote:
> 
> Hello,
> I am using m68k-elf-gcc version egcs-2.91.66 compiled for use in uClinux
> coldfire. I'm using uClinux-coldfire-20000119.tar.gz and
> uClinux-coldfire-tools-20000119.tar.gz
> 
> After some benchmarking I came to the conclusion that the hardware divide is
> not being used. Is there any compiler option to turn this on?
>

If this is the egcs-1.1.something release then it I suspect it will not
support the hardware divide or mac instructions.

> 
> I'm trying to evaluate the uClinux floating point support to see it will be
> adequate for a particular application.
> 
> Thanks for any help concerning this, or any direction pointing to the
> appropriate FAQs or HOWTOs.
> 

Wayne Deeter, David Fiddes and others have been working on the support.
You can find info about it on :

  http://fiddes.net/coldfire/

I do not know the status of these patches or if anyone has attempted to
get them into the gcc and binutils sources.

-- 
 Chris Johns, mailto:cjohns@cybertec.com.au mailto:ccj@acm.org

------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sourceware.cygnus.com

coldfire math when using gcc

[Dwight Schauer]

Hello,
I am using m68k-elf-gcc version egcs-2.91.66 compiled for use in uClinux
coldfire. I'm using uClinux-coldfire-20000119.tar.gz and
uClinux-coldfire-tools-20000119.tar.gz

After some benchmarking I came to the conclusion that the hardware divide is
not being used. Is there any compiler option to turn this on?

Also, in the floating point emulation support, how do I turn on support for %f
in printf? Or use functions like sqrt, sin, cos and kin? I know that floating
point emulation is not the speediest, but if that speed is adequate, then it
should be OK, since the majority of the intended application does not rely on
floating point.

When I try to do any floating point other the *,/,+,- I get linker errors like
/home/das/uClinux-coldfire/tools/elf2flt  -o testf testf.elf
ERROR: undefined symbol=__gtsf2
ERROR: undefined symbol=__addsf3
ERROR: undefined symbol=__subsf3
ERROR: undefined symbol=__mulsf3
ERROR: undefined symbol=__subsf3
ERROR: undefined symbol=__mulsf3
ERROR: undefined symbol=__subsf3
ERROR: undefined symbol=__mulsf3
... and many more

I'm trying to evaluate the uClinux floating point support to see it will be
adequate for a particular application.

Thanks for any help concerning this, or any direction pointing to the
appropriate FAQs or HOWTOs.


-- 
Dwight Schauer, R&D Engineer
Vertex RSI, Controls & Structures Div.
1915 Harrison Road                     Phone: (903) 295-1480 ext 236
Longview, TX 75604                     Fax: (903) 295-1479

------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sourceware.cygnus.com