From: Rafaël Carré <rafael.carre_at_gmail.com>
Date: Fri, 17 Sep 2010 13:32:59 +0200

On Fri, 17 Sep 2010 12:08:38 +0100
Dave Hooper <dave_at_beermex.com> wrote:

> If I remember correctly, gcc was not clever enough to maintain the
> ordering constraints needed by the ldmia/stmia calls, so I needed to
> assign the registers manually to prevent generating bad code

Ah right, GCC can not be aware of such a constraint, it doesn't see the
ldmia inside the asm()

> (by the
> way x86 shouldn't use this file at all so compilation issues wouldn't
> show up on your build).

Yeah I just tried x86 because there were some changes in Tremor SVN
(os_types.h was renamed -> misc.h)

> I had thought that register("r10") would
> ensure the code-generation step preserves existing value of r10 - is
> that not the case? If not then it sounds like we need to stack all
> registers at entry/unstack on exist from all functions that use the
> TRANSFORM macros. I'd kinda assumed that gcc would just do this by
> default though. If not, maybe there's another syntax or idiom that
> lets us specify an explicit register AND have gcc ensure its original
> value is preserved on function prolog/epilog.

Hm i just tried on a small file and r10 should be preserved on
prolog/epilog so maybe something's going wrong?

I think the OP should post .S output of gcc and also give
us compiler version and options being used

-- 
Rafaël Carré

Received on 2010-09-17