some questions of gcc-4.1.0/glibc-2.4 toolchain

Hi,

I got a toolchain with
gcc-4.1.0/glibc-2.4/binutils-2.16.91.0.7 by
using crossgcc. It has a litte difference with the old
version. Here
are some questions.

1. How to support multiple abi?
It can compile application with -mabi=aapcs. This can be
observed with
"arm-iwmmxt-linux-gnueabi-objdum -p a.out | grep
private", the result
is
              private flags = 4000002: [Version4 EABI] [has
entry point]

I suppose the toolchain should support old abi and the
default is
eabi. The test case shows the result with and without
-mabi=aapcs is
the same. But I cannot get an old abi object file. I cannot
use
-mabi=apcs option. I can use -mapcs, but the result is the
same as
-mabi=aapcs. How should I support multiple abi to the
gcc-4.1.0/glibc-2.4 toolchain.

2. Which floating point is used?
I ever used objdump to see private flags in binary file show
the
floating point format. But from item 1 we cannot know it. I
checked
the binutils source file, seems FPA used, but I'm not sure.

3. How to modify gcc specs?
I remembered there is a specs file which can control some
behaviours
of gcc. But following output shows the new toolchain uses
built-in
specs. Can I still modify gcc specs?
$ arm-iwmmxt-linux-gnueabi-gcc -v
Using built-in specs.
Target: arm-iwmmxt-linux-gnueabi
Configured with:
/home/test/gcc41-glibc24-binu216/build/arm-iwmmxt-linux-gnue
abi/gcc-4.1.0-glibc-2.4/gcc-4.1.0/configure
--target=arm-iwmmxt-linux-gnueabi
--host=i686-host_pc-linux-gnu
--prefix=/usr/local/bridge/arm-iwmmxt-linux-gnueabi
--with-cpu=iwmmxt
--with-float=soft --enable-cxx-flags=-msoft-float
--with-headers=/usr/local/bridge/arm-iwmmxt-linux-gnueabi/ar
m-iwmmxt-linux-gnueabi/include
--with-local-prefix=/usr/local/bridge/arm-iwmmxt-linux-gnuea
bi/arm-iwmmxt-linux-gnueabi
--disable-nls --enable-threads=posix --enable-symvers=gnu
--enable-__cxa_atexit --enable-languages=c,c++
--enable-shared
--enable-c99 --enable-long-long
Thread model: posix
gcc version 4.1.0

--
best regards,
-Bridge

--
For unsubscribe information see http://sourcewar
e.org/lists.html#faq

--- Bridge Wu <mingqiao.wugmail.com> wrote:

> Hi,
> 
> I got a toolchain with
gcc-4.1.0/glibc-2.4/binutils-2.16.91.0.7 by
> using crossgcc. It has a litte difference with the old
version. Here
> are some questions.
> 
> 1. How to support multiple abi?
> It can compile application with -mabi=aapcs. This can
be observed with
> "arm-iwmmxt-linux-gnueabi-objdum -p a.out | grep
private", the result
> is
>               private flags = 4000002: [Version4 EABI]
[has entry point]
> 
> I suppose the toolchain should support old abi and the
default is
> eabi. The test case shows the result with and without
-mabi=aapcs is
> the same. But I cannot get an old abi object file. I
cannot use
> -mabi=apcs option. I can use -mapcs, but the result is
the same as
> -mabi=aapcs. How should I support multiple abi to the
> gcc-4.1.0/glibc-2.4 toolchain.
> 
Are you using my patch for arm-iwmmxt-linux-gnueabi?  I
defaults to mcpu=iwmmxt
and mabi=aapcs-linux.  

Be aware that you need an aapcs compatible ABI for iWMMXt
mode.  You will have
to make sure that your -march is set to something
appropriate.

> 2. Which floating point is used?
> I ever used objdump to see private flags in binary file
show the
> floating point format. But from item 1 we cannot know
it. I checked
> the binutils source file, seems FPA used, but I'm not
sure.
>
Again my patch uses VFP since the iWMMXt does it in
hardware.  If you you want
to be backward compatible with OABI you'll have to use FPA
and/or softfloat
since it was a part of the ABI.  You'll need to use
multilib I believe.

> 3. How to modify gcc specs?
> I remembered there is a specs file which can control
some behaviours
> of gcc. But following output shows the new toolchain
uses built-in
> specs. Can I still modify gcc specs?
> $ arm-iwmmxt-linux-gnueabi-gcc -v
> Using built-in specs.
> Target: arm-iwmmxt-linux-gnueabi
> Configured with:
>
/home/test/gcc41-glibc24-binu216/build/arm-iwmmxt-linux-gnue
abi/gcc-4.1.0-glibc-2.4/gcc-4.1.0/configure
> --target=arm-iwmmxt-linux-gnueabi
--host=i686-host_pc-linux-gnu
> --prefix=/usr/local/bridge/arm-iwmmxt-linux-gnueabi
--with-cpu=iwmmxt
> --with-float=soft --enable-cxx-flags=-msoft-float
>
--with-headers=/usr/local/bridge/arm-iwmmxt-linux-gnueabi/ar
m-iwmmxt-linux-gnueabi/include
>
--with-local-prefix=/usr/local/bridge/arm-iwmmxt-linux-gnuea
bi/arm-iwmmxt-linux-gnueabi
> --disable-nls --enable-threads=posix
--enable-symvers=gnu
> --enable-__cxa_atexit --enable-languages=c,c++
--enable-shared
> --enable-c99 --enable-long-long
> Thread model: posix
> gcc version 4.1.0
Ah, you've used softfloat.  That may won't use FPA or VFP
unless you specifiy
-mfloat-hard.

Yes, you should be able to use your own specs in theory!
 Just
make sure the
underlying support has been compiled in for any changes!

> 
> --
> best regards,
> -Bridge

I've only been digging deeply into this stuff for a couple
of months so
although I don't think I've done anything wrong in my
patch; at least nobody
has said anything!  It does seem to work! 


Steve


		
___________________________________________________________ 
Does your mail provider give you FREE antivirus protection? 
Get Yahoo! Mail http://uk.mail.yahoo.com

--
For unsubscribe information see http://sourcewar
e.org/lists.html#faq

On 4/29/06, Steven Newbury <s_j_newburyyahoo.co.uk> wrote:
>
> --- Bridge Wu <mingqiao.wugmail.com> wrote:
>
> > Hi,
> >
> > I got a toolchain with
gcc-4.1.0/glibc-2.4/binutils-2.16.91.0.7 by
> > using crossgcc. It has a litte difference with the
old version. Here
> > are some questions.
> >
> > 1. How to support multiple abi?
> > It can compile application with -mabi=aapcs. This
can be observed with
> > "arm-iwmmxt-linux-gnueabi-objdum -p a.out |
grep private", the result
> > is
> >               private flags = 4000002: [Version4
EABI] [has entry point]
> >
> > I suppose the toolchain should support old abi and
the default is
> > eabi. The test case shows the result with and
without -mabi=aapcs is
> > the same. But I cannot get an old abi object file.
I cannot use
> > -mabi=apcs option. I can use -mapcs, but the
result is the same as
> > -mabi=aapcs. How should I support multiple abi to
the
> > gcc-4.1.0/glibc-2.4 toolchain.
> >
> Are you using my patch for arm-iwmmxt-linux-gnueabi?  I
defaults to mcpu=iwmmxt
> and mabi=aapcs-linux.
>
> Be aware that you need an aapcs compatible ABI for
iWMMXt mode.  You will have
> to make sure that your -march is set to something
appropriate.
>
I did not apply your patch. I searched here and found you
have sent
two iwmmxt patches. It's the second one, right?

> > 2. Which floating point is used?
> > I ever used objdump to see private flags in binary
file show the
> > floating point format. But from item 1 we cannot
know it. I checked
> > the binutils source file, seems FPA used, but I'm
not sure.
> >
> Again my patch uses VFP since the iWMMXt does it in
hardware.  If you you want
> to be backward compatible with OABI you'll have to use
FPA and/or softfloat
> since it was a part of the ABI.  You'll need to use
multilib I believe.
>
> > 3. How to modify gcc specs?
> > I remembered there is a specs file which can
control some behaviours
> > of gcc. But following output shows the new
toolchain uses built-in
> > specs. Can I still modify gcc specs?
> > $ arm-iwmmxt-linux-gnueabi-gcc -v
> > Using built-in specs.
> > Target: arm-iwmmxt-linux-gnueabi
> > Configured with:
> >
>
/home/test/gcc41-glibc24-binu216/build/arm-iwmmxt-linux-gnue
abi/gcc-4.1.0-glibc-2.4/gcc-4.1.0/configure
> > --target=arm-iwmmxt-linux-gnueabi
--host=i686-host_pc-linux-gnu
> >
--prefix=/usr/local/bridge/arm-iwmmxt-linux-gnueabi
--with-cpu=iwmmxt
> > --with-float=soft --enable-cxx-flags=-msoft-float
> >
>
--with-headers=/usr/local/bridge/arm-iwmmxt-linux-gnueabi/ar
m-iwmmxt-linux-gnueabi/include
> >
>
--with-local-prefix=/usr/local/bridge/arm-iwmmxt-linux-gnuea
bi/arm-iwmmxt-linux-gnueabi
> > --disable-nls --enable-threads=posix
--enable-symvers=gnu
> > --enable-__cxa_atexit --enable-languages=c,c++
--enable-shared
> > --enable-c99 --enable-long-long
> > Thread model: posix
> > gcc version 4.1.0
> Ah, you've used softfloat.  That may won't use FPA or
VFP unless you specifiy
> -mfloat-hard.
A little confused. I remember -mhard-float is used to
indicate hard
FPU, -msoft-float indicate soft FPA and none indicate soft
VFP
according to gcc/config/arm/xscale-elf.h. But why gcc-4.1.0
does not
use gcc/config/arm/xscale-elf.h and t-xscale-elf any more? I
tried
these options to compile c code and objdump always show the
same
result (i.e.private flags = 4000002: [Version4 EABI] [has
entry
point]).

The files added in your patch are used as the spec, right?
How are they used?

>
> Yes, you should be able to use your own specs in
theory!  Just make
sure the
> underlying support has been compiled in for any
changes!
I know not much about the specs. Would you like to give more
detail?
How to use our own specs? How to compile underlying support?

>
> I've only been digging deeply into this stuff for a
couple of months so
> although I don't think I've done anything wrong in my
patch; at least nobody
> has said anything!  It does seem to work! 
>
>
> Steve
>
>
>
>
___________________________________________________________
> Does your mail provider give you FREE antivirus
protection?
> Get Yahoo! Mail http://uk.mail.yahoo.com
>


--
best regards,
-Bridge

--
For unsubscribe information see http://sourcewar
e.org/lists.html#faq

--- Bridge Wu <mingqiao.wugmail.com> wrote:

> On 4/29/06, Steven Newbury <s_j_newburyyahoo.co.uk> wrote:
> >
> > --- Bridge Wu <mingqiao.wugmail.com> wrote:
> >
> > > Hi,
> > >
> > > I got a toolchain with
gcc-4.1.0/glibc-2.4/binutils-2.16.91.0.7 by
> > > using crossgcc. It has a litte difference
with the old version. Here
> > > are some questions.
> > >
> > > 1. How to support multiple abi?
> > > It can compile application with -mabi=aapcs.
This can be observed with
> > > "arm-iwmmxt-linux-gnueabi-objdum -p
a.out | grep private", the result
> > > is
> > >               private flags = 4000002:
[Version4 EABI] [has entry point]
> > >
> > > I suppose the toolchain should support old
abi and the default is
> > > eabi. The test case shows the result with and
without -mabi=aapcs is
> > > the same. But I cannot get an old abi object
file. I cannot use
> > > -mabi=apcs option. I can use -mapcs, but the
result is the same as
> > > -mabi=aapcs. How should I support multiple
abi to the
> > > gcc-4.1.0/glibc-2.4 toolchain.
> > >
> > Are you using my patch for
arm-iwmmxt-linux-gnueabi?  I defaults to
> mcpu=iwmmxt
> > and mabi=aapcs-linux.
> >
> > Be aware that you need an aapcs compatible ABI for
iWMMXt mode.  You will
> have
> > to make sure that your -march is set to something
appropriate.
> >
> I did not apply your patch. I searched here and found
you have sent
> two iwmmxt patches. It's the second one, right?
Yes, the second one.  It may not apply cleanly to 4.1.0,
it's actually diff'd
from 4.1.1, it should be easy to fix up though.

> 
> > > 2. Which floating point is used?
> > > I ever used objdump to see private flags in
binary file show the
> > > floating point format. But from item 1 we
cannot know it. I checked
> > > the binutils source file, seems FPA used, but
I'm not sure.
> > >
> > Again my patch uses VFP since the iWMMXt does it
in hardware.  If you you
> want
> > to be backward compatible with OABI you'll have
to use FPA and/or softfloat
> > since it was a part of the ABI.  You'll need to
use multilib I believe.
> >
> > > 3. How to modify gcc specs?
> > > I remembered there is a specs file which can
control some behaviours
> > > of gcc. But following output shows the new
toolchain uses built-in
> > > specs. Can I still modify gcc specs?
> > > $ arm-iwmmxt-linux-gnueabi-gcc -v
> > > Using built-in specs.
> > > Target: arm-iwmmxt-linux-gnueabi
> > > Configured with:
> > >
> >
>
/home/test/gcc41-glibc24-binu216/build/arm-iwmmxt-linux-gnue
abi/gcc-4.1.0-glibc-2.4/gcc-4.1.0/configure
> > > --target=arm-iwmmxt-linux-gnueabi
--host=i686-host_pc-linux-gnu
> > >
--prefix=/usr/local/bridge/arm-iwmmxt-linux-gnueabi
--with-cpu=iwmmxt
> > > --with-float=soft
--enable-cxx-flags=-msoft-float
> > >
> >
>
--with-headers=/usr/local/bridge/arm-iwmmxt-linux-gnueabi/ar
m-iwmmxt-linux-gnueabi/include
> > >
> >
>
--with-local-prefix=/usr/local/bridge/arm-iwmmxt-linux-gnuea
bi/arm-iwmmxt-linux-gnueabi
> > > --disable-nls --enable-threads=posix
--enable-symvers=gnu
> > > --enable-__cxa_atexit
--enable-languages=c,c++ --enable-shared
> > > --enable-c99 --enable-long-long
> > > Thread model: posix
> > > gcc version 4.1.0
> > Ah, you've used softfloat.  That may won't use
FPA or VFP unless you
> specifiy
> > -mfloat-hard.
> A little confused. I remember -mhard-float is used to
indicate hard
> FPU, -msoft-float indicate soft FPA and none indicate
soft VFP
> according to gcc/config/arm/xscale-elf.h. But why
gcc-4.1.0 does not
> use gcc/config/arm/xscale-elf.h and t-xscale-elf any
more? I tried
> these options to compile c code and objdump always show
the same
> result (i.e.private flags = 4000002: [Version4 EABI]
[has entry
> point]).
As I understand the difference between soft FPA and soft VFP
is down to which
standard is implemented in the softfloat lib.  I really
doesn't matter much
unless you expect certain behaviour in you application.

The lack of inclusion of xscale-elf.h and t-xscale-elf are
what motivated me to
make my patch.  The reason I made new configs and didn't
just include those is
both to keep compatiblity for the "obsolete"
xscale-* targets and because the
existing configs were not really suitable for Linux/EABI.
The xscale-* targets
are for a stand-alone toolchain.
> 
> The files added in your patch are used as the spec,
right? How are they used?
It just configures gcc to produce objects for the specified
default.  Otherwise
"generic" ARM is targeted with the default
specs.  The standard
arm-none-linux-gnueabi tool chain will produce code for
iWMMXt if you tell it
to, but it's not the default.
> 
> >
> > Yes, you should be able to use your own specs in
theory!  Just make
sure
> the
> > underlying support has been compiled in for any
changes!
> I know not much about the specs. Would you like to give
more detail?
> How to use our own specs? How to compile underlying
support?
I just mean that you need to have available to you any
options you specifify in
your spec.  If you don't have a softfloat lib available it
won't work to put
softfloat in the specs for example.
> 
> >
> > I've only been digging deeply into this stuff for
a couple of months so
> > although I don't think I've done anything wrong
in my patch; at least
> nobody
> > has said anything!  It does seem to work! 
As I said I'm not really an expert (yet!)..


Steve


	
	
		
___________________________________________________________ 
Win tickets to the 2006 FIFA World Cup Germany with Yahoo!
Messenger. 
http://advision.webevents.yahoo.com/fifaworldcup_uk/

--
For unsubscribe information see http://sourcewar
e.org/lists.html#faq

--- Steven Newbury <s_j_newburyyahoo.co.uk> wrote:

> 
> --- Bridge Wu <mingqiao.wugmail.com> wrote:
> 
> > On 4/29/06, Steven Newbury <s_j_newburyyahoo.co.uk> wrote:
> > >
> > > --- Bridge Wu <mingqiao.wugmail.com> wrote:
> > >
> > > > Hi,
> > > >
> > > > I got a toolchain with
gcc-4.1.0/glibc-2.4/binutils-2.16.91.0.7 by
> > > > using crossgcc. It has a litte
difference with the old version. Here
> > > > are some questions.
> > > >
> > > > 1. How to support multiple abi?
> > > > It can compile application with
-mabi=aapcs. This can be observed with
> > > > "arm-iwmmxt-linux-gnueabi-objdum
-p a.out | grep private", the result
> > > > is
> > > >               private flags = 4000002:
[Version4 EABI] [has entry
> point]
> > > >
> > > > I suppose the toolchain should support
old abi and the default is
> > > > eabi. The test case shows the result
with and without -mabi=aapcs is
> > > > the same. But I cannot get an old abi
object file. I cannot use
> > > > -mabi=apcs option. I can use -mapcs, but
the result is the same as
> > > > -mabi=aapcs. How should I support
multiple abi to the
> > > > gcc-4.1.0/glibc-2.4 toolchain.
> > > >
> > > Are you using my patch for
arm-iwmmxt-linux-gnueabi?  I defaults to
> > mcpu=iwmmxt
> > > and mabi=aapcs-linux.
> > >
> > > Be aware that you need an aapcs compatible
ABI for iWMMXt mode.  You will
> > have
> > > to make sure that your -march is set to
something appropriate.
> > >
> > I did not apply your patch. I searched here and
found you have sent
> > two iwmmxt patches. It's the second one, right?
> Yes, the second one.  It may not apply cleanly to
4.1.0, it's actually diff'd
> from 4.1.1, it should be easy to fix up though.
> 
> > 
> > > > 2. Which floating point is used?
> > > > I ever used objdump to see private flags
in binary file show the
> > > > floating point format. But from item 1
we cannot know it. I checked
> > > > the binutils source file, seems FPA
used, but I'm not sure.
> > > >
> > > Again my patch uses VFP since the iWMMXt does
it in hardware.  If you 
I've managed to confuse myself a bit here; there is no
iWMMXt processor with
hardware FP.  However iWMMXt processors are incompatible
with FPA due to opcode
overlap, while they could have an VFP co-processor in
principal.  That said
iWMMXt instructions should make softfloat fairly quick
anyway.

So the result is if you want use software using the FPA ABI
(oldABI requires
"hard" FPA) support you can't target iWMMXt and
have to use ARMv5t instead
unless the software in question was built with softfloat, it
should work ok on
a iWMMXt targetted system in that case.

If anybody knows better please let me know.



Steve


		
___________________________________________________________ 
Switch an email account to Yahoo! Mail, you could win FIFA
World Cup tickets. http://uk.mail.yahoo.com

--
For unsubscribe information see http://sourcewar
e.org/lists.html#faq

iWMMXt processor uses CP0 and CP1 while FPA uses CP1 and
CP2, which
causes iWMMXt processor not compatible with FPA. So to
support
floating point in iWMMXt, we cannot use FPA, but VFP or
EABI.

My understanding is EABI provides another implementation to
IEEE-754
spec, similar to VFP. Does anyone know what the performance
of VFP and
EABI are?

On 5/6/06, Steven Newbury <s_j_newburyyahoo.co.uk> wrote:
>
> I've managed to confuse myself a bit here; there is no
iWMMXt processor with
> hardware FP.  However iWMMXt processors are
incompatible with FPA due to opcode
> overlap, while they could have an VFP co-processor in
principal.  That said
> iWMMXt instructions should make softfloat fairly quick
anyway.
>
> So the result is if you want use software using the FPA
ABI (oldABI requires
> "hard" FPA) support you can't target
iWMMXt and have to use ARMv5t instead
> unless the software in question was built with
softfloat, it should work ok on
> a iWMMXt targetted system in that case.
>
> If anybody knows better please let me know.
>
>
>
> Steve
>
>
>
>
___________________________________________________________
> Switch an email account to Yahoo! Mail, you could win
FIFA World Cup tickets. http://uk.mail.yahoo.com
>


--
best regards,
-Bridge

--
For unsubscribe information see http://sourcewar
e.org/lists.html#faq