[Buildroot] ARM softfloat patch for gcc 4.2

[Buildroot] ARM softfloat patch for gcc 4.2

[Andre]

Tested with gcc 4.2 20061205 snapshot + binutils 2.17

Seems to work with one minor issue: although the newly created
arm-linux-uclibc-gcc defaults to -mfloat-abi=soft, binutils does not.

Therefore objects created by arm-linux-uclibc-as can't be linked with
objects created by arm-linux-uclibc-gcc unless -mfloat-abi=soft is
explicitly given on the arm-linux-uclibc-as commandline.

How can I force arm-linux-uclibc-as to generate softfloat compatible
object files by default ??

Andre
--



 
____________________________________________________________________________________
Yahoo! Music Unlimited
Access over 1 million songs.
http://music.yahoo.com/unlimited
-------------- next part --------------
A non-text attachment was scrubbed...
Name: patch_buildroot_gcc420_arm_softfloat
Type: application/octet-stream
Size: 1908 bytes
Desc: 4023813560-patch_buildroot_gcc420_arm_softfloat
Url : http://busybox.net/lists/buildroot/attachments/20061208/5777e715/attachment.obj 

[Buildroot] ARM softfloat patch for gcc 4.2

[Bernhard Fischer]

On Fri, Dec 08, 2006 at 07:21:14PM -0800, Andre wrote:
>Tested with gcc 4.2 20061205 snapshot + binutils 2.17
>
>Seems to work with one minor issue: although the newly created
>arm-linux-uclibc-gcc defaults to -mfloat-abi=soft, binutils does not.

There are (at least) two upstream bugs about arm softfloat:
gcc.gnu.org/PR16314
gcc.gnu.org/PR14352

Richard is the arm arch maintainer, so please discuss the problems there
and also fix it there.
thanks,

[Buildroot] ARM softfloat patch for gcc 4.2

[Andre]

Bernhard Fischer <rep.nop@aon.at> wrote:
> On Fri, Dec 08, 2006 at 07:21:14PM -0800, Andre wrote:
> >Tested with gcc 4.2 20061205 snapshot + binutils 2.17
> >
> >Seems to work with one minor issue: although the newly created
> >arm-linux-uclibc-gcc defaults to -mfloat-abi=soft, binutils does
> not.
> 
> There are (at least) two upstream bugs about arm softfloat:
> gcc.gnu.org/PR16314
> gcc.gnu.org/PR14352
> 
> Richard is the arm arch maintainer, so please discuss the problems
> there and also fix it there. thanks,
> 

Not quite sure I follow you... the issue with binutils seems likely
to be in binutils or in the way buildroot configures and builds
binutils, or not ??

The gcc bugs you reference seem to be unrelated...

Anyway, does the rest of the patch look OK ?? (There should be
nothing too controversial in there, I just updated an existing
buildroot patch for gcc so that it applies to the latest gcc 4.2
snapshot).

Andre
--



 
____________________________________________________________________________________
Do you Yahoo!?
Everyone is raving about the all-new Yahoo! Mail beta.
http://new.mail.yahoo.com

[Buildroot] ARM softfloat patch for gcc 4.2

[Bernhard Fischer]

On Mon, Dec 11, 2006 at 11:32:53PM -0800, Andre wrote:
>Bernhard Fischer <rep.nop@aon.at> wrote:
>> On Fri, Dec 08, 2006 at 07:21:14PM -0800, Andre wrote:
>> >Tested with gcc 4.2 20061205 snapshot + binutils 2.17
>> >
>> >Seems to work with one minor issue: although the newly created
>> >arm-linux-uclibc-gcc defaults to -mfloat-abi=soft, binutils does
>> not.
>> 
>> There are (at least) two upstream bugs about arm softfloat:
>> gcc.gnu.org/PR16314
>> gcc.gnu.org/PR14352
>> 
>> Richard is the arm arch maintainer, so please discuss the problems
>> there and also fix it there. thanks,
>> 
>
>Not quite sure I follow you... the issue with binutils seems likely
>to be in binutils or in the way buildroot configures and builds
>binutils, or not ??
>
>The gcc bugs you reference seem to be unrelated...

Did you read the PRs mentioned above? Especially the thread mentioned in
PR16314, comment #1 ?

If binutils defaults to not creating softfloat code in a toolchain that
was ment to be softfloat-(only?), then that sounds like a glitch, yes.
Does that work as expected with 2.17.50.0.8?

[Buildroot] ARM softfloat patch for gcc 4.2

[Andre]

Bernhard Fischer <rep.nop@aon.at> wrote:
> 
> Did you read the PRs mentioned above? Especially the thread
> mentioned in PR16314, comment #1 ?

Yes, I did. It refers to problems associated with creating softfloat
toolchains for ARM cores supporting "Maverick Crunch" floating point
hardware, which are _very_much_ in the minority (only the Cirrus
Logic EP9312 as far as I know, which isn't even a supported 'Target
Architecture Variant' from buildroot's point of view...).

Certainly doesn't seem like a good reason to hold back on softfloat
for the rest of the ARM community...

> If binutils defaults to not creating softfloat code in a toolchain
> that was ment to be softfloat-(only?), then that sounds like a
> glitch, yes. Does that work as expected with 2.17.50.0.8?

No, binutils 2.17.50.0.8 has the same problem.

Although buildroot tries to configure binutils with
'--with-float=soft' binutils seems to silently ignore that option...

Andre
--



 
____________________________________________________________________________________
Do you Yahoo!?
Everyone is raving about the all-new Yahoo! Mail beta.
http://new.mail.yahoo.com