ARM wmmx instructions from gcc ?

ARM wmmx instructions from gcc ?

[Danny Backx]

Hi,

Does anyone know how well gcc-4.4 works with ARM and wmmx instructions ?

I'm working on cegcc. It currently says :
pavilion: {86} arm-mingw32ce-gcc -mcpu=iwmmxt t.c
t.c:1: error: iwmmxt requires an AAPCS compatible ABI for proper
operation
pavilion: {87} 

It's clear to me where in the source this message is generated.

It is not clear whether this is caused by errors in my port, or whether
gcc has some other problem with this.

Has anyone used this ?

	Danny
-- 
Danny Backx ; danny.backx - at - scarlet.be ; http://danny.backx.info

Re: ARM wmmx instructions from gcc ?

[Dave Korn]

Danny Backx wrote:
> Hi,
> 
> Does anyone know how well gcc-4.4 works with ARM and wmmx instructions ?
> 
> I'm working on cegcc. It currently says :
> pavilion: {86} arm-mingw32ce-gcc -mcpu=iwmmxt t.c
> t.c:1: error: iwmmxt requires an AAPCS compatible ABI for proper
> operation
> pavilion: {87} 
> 
> It's clear to me where in the source this message is generated.
> 
> It is not clear whether this is caused by errors in my port, or whether
> gcc has some other problem with this.

  You should mention that long double alignment macro that the guy on the
cegcc list thought it was connected with.  ARM_DOUBLEWORD_ALIGN.  If I was
following that discussion correctly, the ABI requires it, and you have it
#defined to zero to fix the problem you were getting with float double
function argument passing, but maybe what this means is that it there's some
more fundamental problem in the ABI-related MD macros, that fixing it this way
ended up just papering over.

    cheers,
      DaveK

Re: ARM wmmx instructions from gcc ?

[Danny Backx]

On Sun, 2009-08-30 at 21:39 +0100, Dave Korn wrote:
> Danny Backx wrote:
> > Hi,
> > 
> > Does anyone know how well gcc-4.4 works with ARM and wmmx instructions ?
> > 
> > I'm working on cegcc. It currently says :
> > pavilion: {86} arm-mingw32ce-gcc -mcpu=iwmmxt t.c
> > t.c:1: error: iwmmxt requires an AAPCS compatible ABI for proper
> > operation
> > pavilion: {87} 
> > 
> > It's clear to me where in the source this message is generated.
> > 
> > It is not clear whether this is caused by errors in my port, or whether
> > gcc has some other problem with this.
> 
>   You should mention that long double alignment macro that the guy on the
> cegcc list thought it was connected with.  ARM_DOUBLEWORD_ALIGN.  If I was
> following that discussion correctly, the ABI requires it, and you have it
> #defined to zero to fix the problem you were getting with float double
> function argument passing, but maybe what this means is that it there's some
> more fundamental problem in the ABI-related MD macros, that fixing it this way
> ended up just papering over.

I (naively) kept my message simple, hoping to get a reaction like :
- I use this all the time in gcc-4.4 so the issue must be in your port
or
- nobody's been using that stuff for ages, it's almost certainly broken.

I guess life is not as simple as that :-)

You already extended my message with a great summary, so I'm finding it
hard to add sensible stuff to it.

Here are two references to why ARM_DOUBLEWORD_ALIGN was added there :
- the commit message
http://sourceforge.net/mailarchive/message.php?msg_name=E1Itn8S-0001xI-Nu%40sc8-pr-svn3.sourceforge.net
- the explanation
http://sourceforge.net/mailarchive/message.php?msg_name=473F27A8.1060805%40portugalmail.pt

The end result of all that was this code in gcc/config/arm/wince-pe.h :
#undef  DEFAULT_STRUCTURE_SIZE_BOUNDARY
#define DEFAULT_STRUCTURE_SIZE_BOUNDARY 8

#undef ARM_DOUBLEWORD_ALIGN
#define ARM_DOUBLEWORD_ALIGN 0
#undef BIGGEST_ALIGNMENT
#define BIGGEST_ALIGNMENT 64

So, question time again : could this be in my port, or a more
fundamental issue as Dave hints ?

How can I find out ?

BTW the message on the cegcc list that started all this :
http://sourceforge.net/mailarchive/forum.php?thread_name=21121251665810%
40webmail35.yandex.ru&forum_name=cegcc-devel

	Danny
-- 
Danny Backx ; danny.backx - at - scarlet.be ; http://danny.backx.info

Re: ARM wmmx instructions from gcc ?

[Paul Brook]

> So, question time again : could this be in my port, or a more
> fundamental issue as Dave hints ?

Both. By my reading the ABI you're trying to implement does not provide the 
guarantees required to use iwmmxt instructions. There are ways around this, 
but none of them are simple or pretty. The simplest answer is "don't do that".

Paul

Re: ARM wmmx instructions from gcc ?

[Danny Backx]

On Wed, 2009-09-02 at 13:34 +0100, Paul Brook wrote:
> > So, question time again : could this be in my port, or a more
> > fundamental issue as Dave hints ?
> 
> Both. By my reading the ABI you're trying to implement does not provide the 
> guarantees required to use iwmmxt instructions. There are ways around this, 
> but none of them are simple or pretty. The simplest answer is "don't do that".

Hi Paul,

Thanks for the answer, but I'm not sure what you've told me and why.
Could you expand a little ?

Don't do what ? Use WMMX instructions in GCC ?
or
Don't try to use the current implementation with WMMX ?

Thanks,

	Danny
-- 
Danny Backx ; danny.backx - at - scarlet.be ; http://danny.backx.info

Re: ARM wmmx instructions from gcc ?

[Paul Brook]

> > Both. By my reading the ABI you're trying to implement does not provide
> > the guarantees required to use iwmmxt instructions. There are ways around
> > this, but none of them are simple or pretty. The simplest answer is
> > "don't do that".
>
> Hi Paul,
>
> Thanks for the answer, but I'm not sure what you've told me and why.
> Could you expand a little ?
>
> Don't do what ? Use WMMX instructions in GCC ?
> or
> Don't try to use the current implementation with WMMX ?

iwmmxt instructions require doubleword data alignment . Your target 
[apparently] can't provide that. You loose.

Paul